blob: 598abbbe678619c347dbe53d8dfe2c8cd6c86c65 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/file.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/file.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
Mingming Cao617ba132006-10-11 01:20:53 -070015 * ext4 fs regular file handling primitives
Dave Kleikampac27a0e2006-10-11 01:20:50 -070016 *
17 * 64-bit file support on 64-bit platforms by Jakub Jelinek
18 * (jj@sunsite.ms.mff.cuni.cz)
19 */
20
21#include <linux/time.h>
22#include <linux/fs.h>
Mingming Caodab291a2006-10-11 01:21:01 -070023#include <linux/jbd2.h>
Theodore Ts'obc0b0d62009-06-13 10:09:48 -040024#include <linux/mount.h>
25#include <linux/path.h>
Christoph Hellwig871a2932010-03-03 09:05:07 -050026#include <linux/quotaops.h>
Zheng Liuc8c0df22012-11-08 21:57:40 -050027#include <linux/pagevec.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080028#include <linux/uio.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040029#include "ext4.h"
30#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070031#include "xattr.h"
32#include "acl.h"
33
34/*
35 * Called when an inode is released. Note that this is different
Mingming Cao617ba132006-10-11 01:20:53 -070036 * from ext4_file_open: open gets called at every open, but release
Dave Kleikampac27a0e2006-10-11 01:20:50 -070037 * gets called only when /all/ the files are closed.
38 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -040039static int ext4_release_file(struct inode *inode, struct file *filp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070040{
Theodore Ts'o19f5fb72010-01-24 14:34:07 -050041 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -050042 ext4_alloc_da_blocks(inode);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -050043 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -050044 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -070045 /* if we are the last writer on the inode, drop the block reservation */
46 if ((filp->f_mode & FMODE_WRITE) &&
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -040047 (atomic_read(&inode->i_writecount) == 1) &&
48 !EXT4_I(inode)->i_reserved_data_blocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070049 {
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -050050 down_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -040051 ext4_discard_preallocations(inode);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -050052 up_write(&EXT4_I(inode)->i_data_sem);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070053 }
54 if (is_dx(inode) && filp->private_data)
Mingming Cao617ba132006-10-11 01:20:53 -070055 ext4_htree_free_dir_info(filp->private_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070056
57 return 0;
58}
59
Stephen Hemmingerc1978552014-05-12 10:50:23 -040060static void ext4_unwritten_wait(struct inode *inode)
Eric Sandeene9e3bce2011-02-12 08:17:34 -050061{
62 wait_queue_head_t *wq = ext4_ioend_wq(inode);
63
Dmitry Monakhove27f41e2012-09-28 23:24:52 -040064 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
Eric Sandeene9e3bce2011-02-12 08:17:34 -050065}
66
67/*
68 * This tests whether the IO in question is block-aligned or not.
69 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
70 * are converted to written only after the IO is complete. Until they are
71 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
72 * it needs to zero out portions of the start and/or end block. If 2 AIO
73 * threads are at work on the same unwritten block, they must be synchronized
74 * or one thread will zero the other's data, causing corruption.
75 */
76static int
Al Viro9b884162014-04-17 16:09:22 -040077ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
Eric Sandeene9e3bce2011-02-12 08:17:34 -050078{
79 struct super_block *sb = inode->i_sb;
80 int blockmask = sb->s_blocksize - 1;
Eric Sandeene9e3bce2011-02-12 08:17:34 -050081
Theodore Ts'o6e6358f2014-04-12 12:45:25 -040082 if (pos >= i_size_read(inode))
Eric Sandeene9e3bce2011-02-12 08:17:34 -050083 return 0;
84
Al Viro9b884162014-04-17 16:09:22 -040085 if ((pos | iov_iter_alignment(from)) & blockmask)
Eric Sandeene9e3bce2011-02-12 08:17:34 -050086 return 1;
87
88 return 0;
89}
90
Dave Kleikampac27a0e2006-10-11 01:20:50 -070091static ssize_t
Al Viro9b884162014-04-17 16:09:22 -040092ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070093{
Theodore Ts'o7608e612014-04-21 14:26:28 -040094 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -050095 struct inode *inode = file_inode(iocb->ki_filp);
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -040096 struct mutex *aio_mutex = NULL;
Theodore Ts'o8ad28502014-04-21 14:26:57 -040097 struct blk_plug plug;
Ross Zwisler923ae0f2015-02-16 15:59:38 -080098 int o_direct = io_is_direct(file);
Theodore Ts'o8ad28502014-04-21 14:26:57 -040099 int overwrite = 0;
Al Viro9b884162014-04-17 16:09:22 -0400100 size_t length = iov_iter_count(from);
Zheng Liu85630002012-05-28 18:06:51 -0400101 ssize_t ret;
Al Viro9b884162014-04-17 16:09:22 -0400102 loff_t pos = iocb->ki_pos;
Theodore Ts'o7608e612014-04-21 14:26:28 -0400103
Eric Sandeene2b46572008-01-28 23:58:27 -0500104 /*
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400105 * Unaligned direct AIO must be serialized; see comment above
106 * In the case of O_APPEND, assume that we must always serialize
107 */
108 if (o_direct &&
109 ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
110 !is_sync_kiocb(iocb) &&
111 (file->f_flags & O_APPEND ||
Al Viro9b884162014-04-17 16:09:22 -0400112 ext4_unaligned_aio(inode, from, pos))) {
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400113 aio_mutex = ext4_aio_mutex(inode);
114 mutex_lock(aio_mutex);
115 ext4_unwritten_wait(inode);
116 }
117
118 mutex_lock(&inode->i_mutex);
119 if (file->f_flags & O_APPEND)
120 iocb->ki_pos = pos = i_size_read(inode);
121
122 /*
Eric Sandeene2b46572008-01-28 23:58:27 -0500123 * If we have encountered a bitmap-format file, the size limit
124 * is smaller than s_maxbytes, which is for extent-mapped files.
125 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400126 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -0500127 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Eric Sandeene2b46572008-01-28 23:58:27 -0500128
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400129 if ((pos > sbi->s_bitmap_maxbytes) ||
130 (pos == sbi->s_bitmap_maxbytes && length > 0)) {
131 mutex_unlock(&inode->i_mutex);
132 ret = -EFBIG;
133 goto errout;
134 }
Eric Sandeene2b46572008-01-28 23:58:27 -0500135
Al Viro9b884162014-04-17 16:09:22 -0400136 if (pos + length > sbi->s_bitmap_maxbytes)
137 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - pos);
Eric Sandeene2b46572008-01-28 23:58:27 -0500138 }
139
Dmitry Monakhova41537e2014-10-30 10:53:16 -0400140 iocb->private = &overwrite;
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400141 if (o_direct) {
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400142 blk_start_plug(&plug);
143
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400144
145 /* check whether we do a DIO overwrite or not */
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400146 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400147 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
148 struct ext4_map_blocks map;
149 unsigned int blkbits = inode->i_blkbits;
150 int err, len;
151
152 map.m_lblk = pos >> blkbits;
153 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
154 - map.m_lblk;
155 len = map.m_len;
156
157 err = ext4_map_blocks(NULL, inode, &map, 0);
158 /*
159 * 'err==len' means that all of blocks has
160 * been preallocated no matter they are
161 * initialized or not. For excluding
162 * unwritten extents, we need to check
163 * m_flags. There are two conditions that
164 * indicate for initialized extents. 1) If we
165 * hit extent cache, EXT4_MAP_MAPPED flag is
166 * returned; 2) If we do a real lookup,
167 * non-flags are returned. So we should check
168 * these two conditions.
169 */
170 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
171 overwrite = 1;
172 }
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400173 }
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400174
Al Viro9b884162014-04-17 16:09:22 -0400175 ret = __generic_file_write_iter(iocb, from);
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400176 mutex_unlock(&inode->i_mutex);
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400177
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400178 if (ret > 0) {
179 ssize_t err;
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400180
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400181 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
182 if (err < 0)
183 ret = err;
184 }
185 if (o_direct)
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400186 blk_finish_plug(&plug);
187
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400188errout:
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400189 if (aio_mutex)
190 mutex_unlock(aio_mutex);
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500191 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700192}
193
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800194#ifdef CONFIG_FS_DAX
195static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
196{
197 return dax_fault(vma, vmf, ext4_get_block);
198 /* Is this the right get_block? */
199}
200
201static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
202{
203 return dax_mkwrite(vma, vmf, ext4_get_block);
204}
205
206static const struct vm_operations_struct ext4_dax_vm_ops = {
207 .fault = ext4_dax_fault,
208 .page_mkwrite = ext4_dax_mkwrite,
209};
210#else
211#define ext4_dax_vm_ops ext4_file_vm_ops
212#endif
213
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400214static const struct vm_operations_struct ext4_file_vm_ops = {
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400215 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700216 .map_pages = filemap_map_pages,
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400217 .page_mkwrite = ext4_page_mkwrite,
218};
219
220static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
221{
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400222 file_accessed(file);
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800223 if (IS_DAX(file_inode(file))) {
224 vma->vm_ops = &ext4_dax_vm_ops;
225 vma->vm_flags |= VM_MIXEDMAP;
226 } else {
227 vma->vm_ops = &ext4_file_vm_ops;
228 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400229 return 0;
230}
231
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400232static int ext4_file_open(struct inode * inode, struct file * filp)
233{
234 struct super_block *sb = inode->i_sb;
235 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
236 struct vfsmount *mnt = filp->f_path.mnt;
237 struct path path;
238 char buf[64], *cp;
239
240 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
241 !(sb->s_flags & MS_RDONLY))) {
242 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
243 /*
244 * Sample where the filesystem has been mounted and
245 * store it in the superblock for sysadmin convenience
246 * when trying to sort through large numbers of block
247 * devices or filesystem images.
248 */
249 memset(buf, 0, sizeof(buf));
Al Viro38991672010-01-23 20:10:29 -0500250 path.mnt = mnt;
251 path.dentry = mnt->mnt_root;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400252 cp = d_path(&path, buf, sizeof(buf));
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400253 if (!IS_ERR(cp)) {
Jan Kara044ce472012-07-22 20:31:31 -0400254 handle_t *handle;
255 int err;
256
Theodore Ts'o9924a922013-02-08 21:59:22 -0500257 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Jan Kara044ce472012-07-22 20:31:31 -0400258 if (IS_ERR(handle))
259 return PTR_ERR(handle);
liang xie5d601252014-05-12 22:06:43 -0400260 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Jan Kara044ce472012-07-22 20:31:31 -0400261 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
262 if (err) {
263 ext4_journal_stop(handle);
264 return err;
265 }
Darrick J. Wongcf803902011-10-25 09:18:41 -0400266 strlcpy(sbi->s_es->s_last_mounted, cp,
267 sizeof(sbi->s_es->s_last_mounted));
Jan Kara044ce472012-07-22 20:31:31 -0400268 ext4_handle_dirty_super(handle, sb);
269 ext4_journal_stop(handle);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400270 }
271 }
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500272 /*
273 * Set up the jbd2_inode if we are opening the inode for
274 * writing and the journal is present
275 */
Jan Karaa3612932013-08-16 21:19:41 -0400276 if (filp->f_mode & FMODE_WRITE) {
277 int ret = ext4_inode_attach_jinode(inode);
278 if (ret < 0)
279 return ret;
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500280 }
Christoph Hellwig907f4552010-03-03 09:05:06 -0500281 return dquot_file_open(inode, filp);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400282}
283
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400284/*
Zheng Liuc8c0df22012-11-08 21:57:40 -0500285 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
286 * file rather than ext4_ext_walk_space() because we can introduce
287 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
288 * function. When extent status tree has been fully implemented, it will
289 * track all extent status for a file and we can directly use it to
290 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
291 */
292
293/*
294 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
295 * lookup page cache to check whether or not there has some data between
296 * [startoff, endoff] because, if this range contains an unwritten extent,
297 * we determine this extent as a data or a hole according to whether the
298 * page cache has data or not.
299 */
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500300static int ext4_find_unwritten_pgoff(struct inode *inode,
301 int whence,
302 struct ext4_map_blocks *map,
303 loff_t *offset)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500304{
305 struct pagevec pvec;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500306 unsigned int blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500307 pgoff_t index;
308 pgoff_t end;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500309 loff_t endoff;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500310 loff_t startoff;
311 loff_t lastoff;
312 int found = 0;
313
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500314 blkbits = inode->i_sb->s_blocksize_bits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500315 startoff = *offset;
316 lastoff = startoff;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500317 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500318
319 index = startoff >> PAGE_CACHE_SHIFT;
320 end = endoff >> PAGE_CACHE_SHIFT;
321
322 pagevec_init(&pvec, 0);
323 do {
324 int i, num;
325 unsigned long nr_pages;
326
327 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
328 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
329 (pgoff_t)num);
330 if (nr_pages == 0) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800331 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500332 break;
333
Andrew Morton965c8e52012-12-17 15:59:39 -0800334 BUG_ON(whence != SEEK_HOLE);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500335 /*
336 * If this is the first time to go into the loop and
337 * offset is not beyond the end offset, it will be a
338 * hole at this offset
339 */
340 if (lastoff == startoff || lastoff < endoff)
341 found = 1;
342 break;
343 }
344
345 /*
346 * If this is the first time to go into the loop and
347 * offset is smaller than the first page offset, it will be a
348 * hole at this offset.
349 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800350 if (lastoff == startoff && whence == SEEK_HOLE &&
Zheng Liuc8c0df22012-11-08 21:57:40 -0500351 lastoff < page_offset(pvec.pages[0])) {
352 found = 1;
353 break;
354 }
355
356 for (i = 0; i < nr_pages; i++) {
357 struct page *page = pvec.pages[i];
358 struct buffer_head *bh, *head;
359
360 /*
361 * If the current offset is not beyond the end of given
362 * range, it will be a hole.
363 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800364 if (lastoff < endoff && whence == SEEK_HOLE &&
Zheng Liuc8c0df22012-11-08 21:57:40 -0500365 page->index > end) {
366 found = 1;
367 *offset = lastoff;
368 goto out;
369 }
370
371 lock_page(page);
372
373 if (unlikely(page->mapping != inode->i_mapping)) {
374 unlock_page(page);
375 continue;
376 }
377
378 if (!page_has_buffers(page)) {
379 unlock_page(page);
380 continue;
381 }
382
383 if (page_has_buffers(page)) {
384 lastoff = page_offset(page);
385 bh = head = page_buffers(page);
386 do {
387 if (buffer_uptodate(bh) ||
388 buffer_unwritten(bh)) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800389 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500390 found = 1;
391 } else {
Andrew Morton965c8e52012-12-17 15:59:39 -0800392 if (whence == SEEK_HOLE)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500393 found = 1;
394 }
395 if (found) {
396 *offset = max_t(loff_t,
397 startoff, lastoff);
398 unlock_page(page);
399 goto out;
400 }
401 lastoff += bh->b_size;
402 bh = bh->b_this_page;
403 } while (bh != head);
404 }
405
406 lastoff = page_offset(page) + PAGE_SIZE;
407 unlock_page(page);
408 }
409
410 /*
411 * The no. of pages is less than our desired, that would be a
412 * hole in there.
413 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800414 if (nr_pages < num && whence == SEEK_HOLE) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500415 found = 1;
416 *offset = lastoff;
417 break;
418 }
419
420 index = pvec.pages[i - 1]->index + 1;
421 pagevec_release(&pvec);
422 } while (index <= end);
423
424out:
425 pagevec_release(&pvec);
426 return found;
427}
428
429/*
430 * ext4_seek_data() retrieves the offset for SEEK_DATA.
431 */
432static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
433{
434 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500435 struct ext4_map_blocks map;
436 struct extent_status es;
437 ext4_lblk_t start, last, end;
438 loff_t dataoff, isize;
439 int blkbits;
440 int ret = 0;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500441
442 mutex_lock(&inode->i_mutex);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500443
444 isize = i_size_read(inode);
445 if (offset >= isize) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500446 mutex_unlock(&inode->i_mutex);
447 return -ENXIO;
448 }
449
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500450 blkbits = inode->i_sb->s_blocksize_bits;
451 start = offset >> blkbits;
452 last = start;
453 end = isize >> blkbits;
454 dataoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500455
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500456 do {
457 map.m_lblk = last;
458 map.m_len = end - last + 1;
459 ret = ext4_map_blocks(NULL, inode, &map, 0);
460 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
461 if (last != start)
462 dataoff = (loff_t)last << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500463 break;
464 }
465
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500466 /*
467 * If there is a delay extent at this offset,
468 * it will be as a data.
469 */
470 ext4_es_find_delayed_extent_range(inode, last, last, &es);
471 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
472 if (last != start)
473 dataoff = (loff_t)last << blkbits;
474 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500475 }
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500476
477 /*
478 * If there is a unwritten extent at this offset,
479 * it will be as a data or a hole according to page
480 * cache that has data or not.
481 */
482 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
483 int unwritten;
484 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
485 &map, &dataoff);
486 if (unwritten)
487 break;
488 }
489
490 last++;
491 dataoff = (loff_t)last << blkbits;
492 } while (last <= end);
493
Zheng Liuc8c0df22012-11-08 21:57:40 -0500494 mutex_unlock(&inode->i_mutex);
495
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500496 if (dataoff > isize)
497 return -ENXIO;
498
499 return vfs_setpos(file, dataoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500500}
501
502/*
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500503 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
Zheng Liuc8c0df22012-11-08 21:57:40 -0500504 */
505static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
506{
507 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500508 struct ext4_map_blocks map;
509 struct extent_status es;
510 ext4_lblk_t start, last, end;
511 loff_t holeoff, isize;
512 int blkbits;
513 int ret = 0;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500514
515 mutex_lock(&inode->i_mutex);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500516
517 isize = i_size_read(inode);
518 if (offset >= isize) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500519 mutex_unlock(&inode->i_mutex);
520 return -ENXIO;
521 }
522
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500523 blkbits = inode->i_sb->s_blocksize_bits;
524 start = offset >> blkbits;
525 last = start;
526 end = isize >> blkbits;
527 holeoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500528
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500529 do {
530 map.m_lblk = last;
531 map.m_len = end - last + 1;
532 ret = ext4_map_blocks(NULL, inode, &map, 0);
533 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
534 last += ret;
535 holeoff = (loff_t)last << blkbits;
536 continue;
537 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500538
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500539 /*
540 * If there is a delay extent at this offset,
541 * we will skip this extent.
542 */
543 ext4_es_find_delayed_extent_range(inode, last, last, &es);
544 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
545 last = es.es_lblk + es.es_len;
546 holeoff = (loff_t)last << blkbits;
547 continue;
548 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500549
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500550 /*
551 * If there is a unwritten extent at this offset,
552 * it will be as a data or a hole according to page
553 * cache that has data or not.
554 */
555 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
556 int unwritten;
557 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
558 &map, &holeoff);
559 if (!unwritten) {
560 last += ret;
561 holeoff = (loff_t)last << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500562 continue;
563 }
564 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500565
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500566 /* find a hole */
567 break;
568 } while (last <= end);
569
570 mutex_unlock(&inode->i_mutex);
571
572 if (holeoff > isize)
573 holeoff = isize;
574
575 return vfs_setpos(file, holeoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500576}
577
578/*
Eric Sandeenec7268c2012-04-30 13:14:03 -0500579 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
580 * by calling generic_file_llseek_size() with the appropriate maxbytes
581 * value for each.
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400582 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800583loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400584{
585 struct inode *inode = file->f_mapping->host;
586 loff_t maxbytes;
587
588 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
589 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
590 else
591 maxbytes = inode->i_sb->s_maxbytes;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400592
Andrew Morton965c8e52012-12-17 15:59:39 -0800593 switch (whence) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500594 case SEEK_SET:
595 case SEEK_CUR:
596 case SEEK_END:
Andrew Morton965c8e52012-12-17 15:59:39 -0800597 return generic_file_llseek_size(file, offset, whence,
Zheng Liuc8c0df22012-11-08 21:57:40 -0500598 maxbytes, i_size_read(inode));
599 case SEEK_DATA:
600 return ext4_seek_data(file, offset, maxbytes);
601 case SEEK_HOLE:
602 return ext4_seek_hole(file, offset, maxbytes);
603 }
604
605 return -EINVAL;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400606}
607
Mingming Cao617ba132006-10-11 01:20:53 -0700608const struct file_operations ext4_file_operations = {
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400609 .llseek = ext4_llseek,
Al Viroaad4f8b2014-04-02 14:33:16 -0400610 .read = new_sync_read,
Al Viro9b884162014-04-17 16:09:22 -0400611 .write = new_sync_write,
Al Viroaad4f8b2014-04-02 14:33:16 -0400612 .read_iter = generic_file_read_iter,
Al Viro9b884162014-04-17 16:09:22 -0400613 .write_iter = ext4_file_write_iter,
Andi Kleen5cdd7b22008-04-29 22:03:54 -0400614 .unlocked_ioctl = ext4_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700615#ifdef CONFIG_COMPAT
Mingming Cao617ba132006-10-11 01:20:53 -0700616 .compat_ioctl = ext4_compat_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700617#endif
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400618 .mmap = ext4_file_mmap,
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400619 .open = ext4_file_open,
Mingming Cao617ba132006-10-11 01:20:53 -0700620 .release = ext4_release_file,
621 .fsync = ext4_sync_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -0400623 .splice_write = iter_file_splice_write,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100624 .fallocate = ext4_fallocate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700625};
626
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800627#ifdef CONFIG_FS_DAX
628const struct file_operations ext4_dax_file_operations = {
629 .llseek = ext4_llseek,
630 .read = new_sync_read,
631 .write = new_sync_write,
632 .read_iter = generic_file_read_iter,
633 .write_iter = ext4_file_write_iter,
634 .unlocked_ioctl = ext4_ioctl,
635#ifdef CONFIG_COMPAT
636 .compat_ioctl = ext4_compat_ioctl,
637#endif
638 .mmap = ext4_file_mmap,
639 .open = ext4_file_open,
640 .release = ext4_release_file,
641 .fsync = ext4_sync_file,
642 /* Splice not yet supported with DAX */
643 .fallocate = ext4_fallocate,
644};
645#endif
646
Arjan van de Ven754661f2007-02-12 00:55:38 -0800647const struct inode_operations ext4_file_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -0700648 .setattr = ext4_setattr,
Mingming Cao3e3398a2008-07-11 19:27:31 -0400649 .getattr = ext4_getattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700650 .setxattr = generic_setxattr,
651 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -0700652 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700653 .removexattr = generic_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200654 .get_acl = ext4_get_acl,
Christoph Hellwig64e178a2013-12-20 05:16:44 -0800655 .set_acl = ext4_set_acl,
Eric Sandeen6873fa02008-10-07 00:46:36 -0400656 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700657};
658