blob: 02ce7e7bbdf5ba4da0db50e50d8d4cab49cde4a3 [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>
Theodore Ts'obc0b0d62009-06-13 10:09:48 -040023#include <linux/mount.h>
24#include <linux/path.h>
Matthew Wilcoxc94c2ac2015-09-08 14:58:40 -070025#include <linux/dax.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
Jan Kara364443c2016-11-20 17:36:06 -050034#ifdef CONFIG_FS_DAX
35static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
36{
37 struct inode *inode = file_inode(iocb->ki_filp);
38 ssize_t ret;
39
40 inode_lock_shared(inode);
41 /*
42 * Recheck under inode lock - at this point we are sure it cannot
43 * change anymore
44 */
45 if (!IS_DAX(inode)) {
46 inode_unlock_shared(inode);
47 /* Fallback to buffered IO in case we cannot support DAX */
48 return generic_file_read_iter(iocb, to);
49 }
50 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
51 inode_unlock_shared(inode);
52
53 file_accessed(iocb->ki_filp);
54 return ret;
55}
56#endif
57
58static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
59{
Theodore Ts'o0db1ff22017-02-05 01:28:48 -050060 if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
61 return -EIO;
62
Jan Kara364443c2016-11-20 17:36:06 -050063 if (!iov_iter_count(to))
64 return 0; /* skip atime */
65
66#ifdef CONFIG_FS_DAX
67 if (IS_DAX(file_inode(iocb->ki_filp)))
68 return ext4_dax_read_iter(iocb, to);
69#endif
70 return generic_file_read_iter(iocb, to);
71}
72
Dave Kleikampac27a0e2006-10-11 01:20:50 -070073/*
74 * Called when an inode is released. Note that this is different
Mingming Cao617ba132006-10-11 01:20:53 -070075 * from ext4_file_open: open gets called at every open, but release
Dave Kleikampac27a0e2006-10-11 01:20:50 -070076 * gets called only when /all/ the files are closed.
77 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -040078static int ext4_release_file(struct inode *inode, struct file *filp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070079{
Theodore Ts'o19f5fb72010-01-24 14:34:07 -050080 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -050081 ext4_alloc_da_blocks(inode);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -050082 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -050083 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 /* if we are the last writer on the inode, drop the block reservation */
85 if ((filp->f_mode & FMODE_WRITE) &&
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -040086 (atomic_read(&inode->i_writecount) == 1) &&
87 !EXT4_I(inode)->i_reserved_data_blocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070088 {
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -050089 down_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -040090 ext4_discard_preallocations(inode);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -050091 up_write(&EXT4_I(inode)->i_data_sem);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070092 }
93 if (is_dx(inode) && filp->private_data)
Mingming Cao617ba132006-10-11 01:20:53 -070094 ext4_htree_free_dir_info(filp->private_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070095
96 return 0;
97}
98
Stephen Hemmingerc1978552014-05-12 10:50:23 -040099static void ext4_unwritten_wait(struct inode *inode)
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500100{
101 wait_queue_head_t *wq = ext4_ioend_wq(inode);
102
Dmitry Monakhove27f41e2012-09-28 23:24:52 -0400103 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500104}
105
106/*
107 * This tests whether the IO in question is block-aligned or not.
108 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
109 * are converted to written only after the IO is complete. Until they are
110 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
111 * it needs to zero out portions of the start and/or end block. If 2 AIO
112 * threads are at work on the same unwritten block, they must be synchronized
113 * or one thread will zero the other's data, causing corruption.
114 */
115static int
Al Viro9b884162014-04-17 16:09:22 -0400116ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500117{
118 struct super_block *sb = inode->i_sb;
119 int blockmask = sb->s_blocksize - 1;
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500120
Theodore Ts'o6e6358f2014-04-12 12:45:25 -0400121 if (pos >= i_size_read(inode))
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500122 return 0;
123
Al Viro9b884162014-04-17 16:09:22 -0400124 if ((pos | iov_iter_alignment(from)) & blockmask)
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500125 return 1;
126
127 return 0;
128}
129
Jan Kara213bcd92016-11-20 17:29:51 -0500130/* Is IO overwriting allocated and initialized blocks? */
131static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
132{
133 struct ext4_map_blocks map;
134 unsigned int blkbits = inode->i_blkbits;
135 int err, blklen;
136
137 if (pos + len > i_size_read(inode))
138 return false;
139
140 map.m_lblk = pos >> blkbits;
141 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
142 blklen = map.m_len;
143
144 err = ext4_map_blocks(NULL, inode, &map, 0);
145 /*
146 * 'err==len' means that all of the blocks have been preallocated,
147 * regardless of whether they have been initialized or not. To exclude
148 * unwritten extents, we need to check m_flags.
149 */
150 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
151}
152
153static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
154{
155 struct inode *inode = file_inode(iocb->ki_filp);
156 ssize_t ret;
157
158 ret = generic_write_checks(iocb, from);
159 if (ret <= 0)
160 return ret;
161 /*
162 * If we have encountered a bitmap-format file, the size limit
163 * is smaller than s_maxbytes, which is for extent-mapped files.
164 */
165 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
166 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
167
168 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
169 return -EFBIG;
170 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
171 }
172 return iov_iter_count(from);
173}
174
Jan Kara776722e2016-11-20 18:09:11 -0500175#ifdef CONFIG_FS_DAX
176static ssize_t
177ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
178{
179 struct inode *inode = file_inode(iocb->ki_filp);
180 ssize_t ret;
Jan Kara776722e2016-11-20 18:09:11 -0500181
182 inode_lock(inode);
183 ret = ext4_write_checks(iocb, from);
184 if (ret <= 0)
185 goto out;
186 ret = file_remove_privs(iocb->ki_filp);
187 if (ret)
188 goto out;
189 ret = file_update_time(iocb->ki_filp);
190 if (ret)
191 goto out;
192
Jan Kara776722e2016-11-20 18:09:11 -0500193 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
194out:
Christoph Hellwigff5462e2017-02-08 14:39:27 -0500195 inode_unlock(inode);
Jan Kara776722e2016-11-20 18:09:11 -0500196 if (ret > 0)
197 ret = generic_write_sync(iocb, ret);
198 return ret;
199}
200#endif
201
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700202static ssize_t
Al Viro9b884162014-04-17 16:09:22 -0400203ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700204{
Al Viro496ad9a2013-01-23 17:07:38 -0500205 struct inode *inode = file_inode(iocb->ki_filp);
Al Viro2ba48ce2015-04-09 13:52:01 -0400206 int o_direct = iocb->ki_flags & IOCB_DIRECT;
Jan Karae142d052016-03-08 22:44:50 -0500207 int unaligned_aio = 0;
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400208 int overwrite = 0;
Zheng Liu85630002012-05-28 18:06:51 -0400209 ssize_t ret;
Theodore Ts'o7608e612014-04-21 14:26:28 -0400210
Theodore Ts'o0db1ff22017-02-05 01:28:48 -0500211 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
212 return -EIO;
213
Jan Kara776722e2016-11-20 18:09:11 -0500214#ifdef CONFIG_FS_DAX
215 if (IS_DAX(inode))
216 return ext4_dax_write_iter(iocb, from);
217#endif
218
Al Viro59551022016-01-22 15:40:57 -0500219 inode_lock(inode);
Jan Kara213bcd92016-11-20 17:29:51 -0500220 ret = ext4_write_checks(iocb, from);
Al Viro3309dd02015-04-09 12:55:47 -0400221 if (ret <= 0)
Al Viroe768d7f2015-04-07 14:48:22 -0400222 goto out;
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400223
224 /*
Jan Karae142d052016-03-08 22:44:50 -0500225 * Unaligned direct AIO must be serialized among each other as zeroing
226 * of partial blocks of two competing unaligned AIOs can result in data
227 * corruption.
228 */
229 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
230 !is_sync_kiocb(iocb) &&
231 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
232 unaligned_aio = 1;
233 ext4_unwritten_wait(inode);
234 }
235
Dmitry Monakhova41537e2014-10-30 10:53:16 -0400236 iocb->private = &overwrite;
Jan Kara213bcd92016-11-20 17:29:51 -0500237 /* Check whether we do a DIO overwrite or not */
238 if (o_direct && ext4_should_dioread_nolock(inode) && !unaligned_aio &&
239 ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from)))
240 overwrite = 1;
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400241
Al Viro9b884162014-04-17 16:09:22 -0400242 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -0500243 inode_unlock(inode);
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400244
Christoph Hellwige2592212016-04-07 08:52:01 -0700245 if (ret > 0)
246 ret = generic_write_sync(iocb, ret);
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400247
Al Viroe768d7f2015-04-07 14:48:22 -0400248 return ret;
249
250out:
Al Viro59551022016-01-22 15:40:57 -0500251 inode_unlock(inode);
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500252 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700253}
254
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800255#ifdef CONFIG_FS_DAX
Dave Jiangc791ace2017-02-24 14:57:08 -0800256static int ext4_dax_huge_fault(struct vm_fault *vmf,
257 enum page_entry_size pe_size)
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800258{
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700259 int result;
Jan Karafb26a1c2017-05-12 15:46:54 -0700260 handle_t *handle = NULL;
Dave Jiang11bac802017-02-24 14:56:41 -0800261 struct inode *inode = file_inode(vmf->vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500262 struct super_block *sb = inode->i_sb;
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700263 bool write = vmf->flags & FAULT_FLAG_WRITE;
264
265 if (write) {
266 sb_start_pagefault(sb);
Dave Jiang11bac802017-02-24 14:56:41 -0800267 file_update_time(vmf->vma->vm_file);
Jan Karafb26a1c2017-05-12 15:46:54 -0700268 down_read(&EXT4_I(inode)->i_mmap_sem);
269 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
270 EXT4_DATA_TRANS_BLOCKS(sb));
271 } else {
272 down_read(&EXT4_I(inode)->i_mmap_sem);
Jan Kara1db17542016-10-21 11:33:49 +0200273 }
Jan Karafb26a1c2017-05-12 15:46:54 -0700274 if (!IS_ERR(handle))
275 result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
276 else
277 result = VM_FAULT_SIGBUS;
278 if (write) {
279 if (!IS_ERR(handle))
280 ext4_journal_stop(handle);
281 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700282 sb_end_pagefault(sb);
Jan Karafb26a1c2017-05-12 15:46:54 -0700283 } else {
284 up_read(&EXT4_I(inode)->i_mmap_sem);
285 }
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700286
287 return result;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800288}
289
Dave Jiangc791ace2017-02-24 14:57:08 -0800290static int ext4_dax_fault(struct vm_fault *vmf)
291{
292 return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
293}
294
Jan Karaea3d7202015-12-07 14:28:03 -0500295/*
Ross Zwisler1e9d1802016-02-27 14:01:13 -0500296 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
Jan Karaea3d7202015-12-07 14:28:03 -0500297 * handler we check for races agaist truncate. Note that since we cycle through
298 * i_mmap_sem, we are sure that also any hole punching that began before we
299 * were called is finished by now and so if it included part of the file we
300 * are working on, our pte will get unmapped and the check for pte_same() in
301 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
302 * desired.
303 */
Dave Jiang11bac802017-02-24 14:56:41 -0800304static int ext4_dax_pfn_mkwrite(struct vm_fault *vmf)
Jan Karaea3d7202015-12-07 14:28:03 -0500305{
Dave Jiang11bac802017-02-24 14:56:41 -0800306 struct inode *inode = file_inode(vmf->vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500307 struct super_block *sb = inode->i_sb;
Jan Karaea3d7202015-12-07 14:28:03 -0500308 loff_t size;
Ross Zwislerd5be7a02016-01-22 15:10:53 -0800309 int ret;
Jan Karaea3d7202015-12-07 14:28:03 -0500310
311 sb_start_pagefault(sb);
Dave Jiang11bac802017-02-24 14:56:41 -0800312 file_update_time(vmf->vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500313 down_read(&EXT4_I(inode)->i_mmap_sem);
314 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
315 if (vmf->pgoff >= size)
316 ret = VM_FAULT_SIGBUS;
Ross Zwislerd5be7a02016-01-22 15:10:53 -0800317 else
Dave Jiang11bac802017-02-24 14:56:41 -0800318 ret = dax_pfn_mkwrite(vmf);
Jan Karaea3d7202015-12-07 14:28:03 -0500319 up_read(&EXT4_I(inode)->i_mmap_sem);
320 sb_end_pagefault(sb);
321
322 return ret;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800323}
324
325static const struct vm_operations_struct ext4_dax_vm_ops = {
326 .fault = ext4_dax_fault,
Dave Jiangc791ace2017-02-24 14:57:08 -0800327 .huge_fault = ext4_dax_huge_fault,
Ross Zwisler1e9d1802016-02-27 14:01:13 -0500328 .page_mkwrite = ext4_dax_fault,
Jan Karaea3d7202015-12-07 14:28:03 -0500329 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800330};
331#else
332#define ext4_dax_vm_ops ext4_file_vm_ops
333#endif
334
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400335static const struct vm_operations_struct ext4_file_vm_ops = {
Jan Karaea3d7202015-12-07 14:28:03 -0500336 .fault = ext4_filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700337 .map_pages = filemap_map_pages,
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400338 .page_mkwrite = ext4_page_mkwrite,
339};
340
341static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
342{
Michael Halcrowc9c74292015-04-12 00:56:10 -0400343 struct inode *inode = file->f_mapping->host;
344
Theodore Ts'o0db1ff22017-02-05 01:28:48 -0500345 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
346 return -EIO;
347
Michael Halcrowc9c74292015-04-12 00:56:10 -0400348 if (ext4_encrypted_inode(inode)) {
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400349 int err = fscrypt_get_encryption_info(inode);
Michael Halcrowc9c74292015-04-12 00:56:10 -0400350 if (err)
351 return 0;
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400352 if (!fscrypt_has_encryption_key(inode))
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400353 return -ENOKEY;
Michael Halcrowc9c74292015-04-12 00:56:10 -0400354 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400355 file_accessed(file);
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800356 if (IS_DAX(file_inode(file))) {
357 vma->vm_ops = &ext4_dax_vm_ops;
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700358 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800359 } else {
360 vma->vm_ops = &ext4_file_vm_ops;
361 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400362 return 0;
363}
364
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400365static int ext4_file_open(struct inode * inode, struct file * filp)
366{
367 struct super_block *sb = inode->i_sb;
368 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
369 struct vfsmount *mnt = filp->f_path.mnt;
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400370 struct dentry *dir;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400371 struct path path;
372 char buf[64], *cp;
Michael Halcrowc9c74292015-04-12 00:56:10 -0400373 int ret;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400374
Theodore Ts'o0db1ff22017-02-05 01:28:48 -0500375 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
376 return -EIO;
377
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400378 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
379 !(sb->s_flags & MS_RDONLY))) {
380 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
381 /*
382 * Sample where the filesystem has been mounted and
383 * store it in the superblock for sysadmin convenience
384 * when trying to sort through large numbers of block
385 * devices or filesystem images.
386 */
387 memset(buf, 0, sizeof(buf));
Al Viro38991672010-01-23 20:10:29 -0500388 path.mnt = mnt;
389 path.dentry = mnt->mnt_root;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400390 cp = d_path(&path, buf, sizeof(buf));
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400391 if (!IS_ERR(cp)) {
Jan Kara044ce472012-07-22 20:31:31 -0400392 handle_t *handle;
393 int err;
394
Theodore Ts'o9924a922013-02-08 21:59:22 -0500395 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Jan Kara044ce472012-07-22 20:31:31 -0400396 if (IS_ERR(handle))
397 return PTR_ERR(handle);
liang xie5d601252014-05-12 22:06:43 -0400398 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Jan Kara044ce472012-07-22 20:31:31 -0400399 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
400 if (err) {
401 ext4_journal_stop(handle);
402 return err;
403 }
Darrick J. Wongcf803902011-10-25 09:18:41 -0400404 strlcpy(sbi->s_es->s_last_mounted, cp,
405 sizeof(sbi->s_es->s_last_mounted));
Jan Kara044ce472012-07-22 20:31:31 -0400406 ext4_handle_dirty_super(handle, sb);
407 ext4_journal_stop(handle);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400408 }
409 }
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400410 if (ext4_encrypted_inode(inode)) {
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400411 ret = fscrypt_get_encryption_info(inode);
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400412 if (ret)
413 return -EACCES;
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400414 if (!fscrypt_has_encryption_key(inode))
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400415 return -ENOKEY;
416 }
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400417
Miklos Szeredic0a37d4872016-03-26 16:14:42 -0400418 dir = dget_parent(file_dentry(filp));
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400419 if (ext4_encrypted_inode(d_inode(dir)) &&
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400420 !fscrypt_has_permitted_context(d_inode(dir), inode)) {
Theodore Ts'off978b02016-02-08 00:54:26 -0500421 ext4_warning(inode->i_sb,
Jakub Wilk8d2ae1c2016-04-27 01:11:21 -0400422 "Inconsistent encryption contexts: %lu/%lu",
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400423 (unsigned long) d_inode(dir)->i_ino,
Theodore Ts'off978b02016-02-08 00:54:26 -0500424 (unsigned long) inode->i_ino);
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400425 dput(dir);
Theodore Ts'off978b02016-02-08 00:54:26 -0500426 return -EPERM;
427 }
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400428 dput(dir);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500429 /*
430 * Set up the jbd2_inode if we are opening the inode for
431 * writing and the journal is present
432 */
Jan Karaa3612932013-08-16 21:19:41 -0400433 if (filp->f_mode & FMODE_WRITE) {
Michael Halcrowc9c74292015-04-12 00:56:10 -0400434 ret = ext4_inode_attach_jinode(inode);
Jan Karaa3612932013-08-16 21:19:41 -0400435 if (ret < 0)
436 return ret;
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500437 }
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400438 return dquot_file_open(inode, filp);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400439}
440
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400441/*
Zheng Liuc8c0df22012-11-08 21:57:40 -0500442 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
443 * file rather than ext4_ext_walk_space() because we can introduce
444 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
445 * function. When extent status tree has been fully implemented, it will
446 * track all extent status for a file and we can directly use it to
447 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
448 */
449
450/*
451 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
452 * lookup page cache to check whether or not there has some data between
453 * [startoff, endoff] because, if this range contains an unwritten extent,
454 * we determine this extent as a data or a hole according to whether the
455 * page cache has data or not.
456 */
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500457static int ext4_find_unwritten_pgoff(struct inode *inode,
458 int whence,
Jan Kara2d90c162016-03-09 23:11:13 -0500459 ext4_lblk_t end_blk,
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500460 loff_t *offset)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500461{
462 struct pagevec pvec;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500463 unsigned int blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500464 pgoff_t index;
465 pgoff_t end;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500466 loff_t endoff;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500467 loff_t startoff;
468 loff_t lastoff;
469 int found = 0;
470
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500471 blkbits = inode->i_sb->s_blocksize_bits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500472 startoff = *offset;
473 lastoff = startoff;
Jan Kara2d90c162016-03-09 23:11:13 -0500474 endoff = (loff_t)end_blk << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500475
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300476 index = startoff >> PAGE_SHIFT;
Jan Kara3f1d5ba2017-05-21 22:34:23 -0400477 end = (endoff - 1) >> PAGE_SHIFT;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500478
479 pagevec_init(&pvec, 0);
480 do {
481 int i, num;
482 unsigned long nr_pages;
483
Eryu Guan624327f2017-05-24 18:02:20 -0400484 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE - 1) + 1;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500485 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
486 (pgoff_t)num);
Jan Kara7d95edd2017-05-21 22:33:23 -0400487 if (nr_pages == 0)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500488 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500489
490 for (i = 0; i < nr_pages; i++) {
491 struct page *page = pvec.pages[i];
492 struct buffer_head *bh, *head;
493
494 /*
Jan Kara7d95edd2017-05-21 22:33:23 -0400495 * If current offset is smaller than the page offset,
496 * there is a hole at this offset.
Zheng Liuc8c0df22012-11-08 21:57:40 -0500497 */
Jan Kara7d95edd2017-05-21 22:33:23 -0400498 if (whence == SEEK_HOLE && lastoff < endoff &&
499 lastoff < page_offset(pvec.pages[i])) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500500 found = 1;
501 *offset = lastoff;
502 goto out;
503 }
504
Jan Kara7d95edd2017-05-21 22:33:23 -0400505 if (page->index > end)
506 goto out;
507
Zheng Liuc8c0df22012-11-08 21:57:40 -0500508 lock_page(page);
509
510 if (unlikely(page->mapping != inode->i_mapping)) {
511 unlock_page(page);
512 continue;
513 }
514
515 if (!page_has_buffers(page)) {
516 unlock_page(page);
517 continue;
518 }
519
520 if (page_has_buffers(page)) {
521 lastoff = page_offset(page);
522 bh = head = page_buffers(page);
523 do {
524 if (buffer_uptodate(bh) ||
525 buffer_unwritten(bh)) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800526 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500527 found = 1;
528 } else {
Andrew Morton965c8e52012-12-17 15:59:39 -0800529 if (whence == SEEK_HOLE)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500530 found = 1;
531 }
532 if (found) {
533 *offset = max_t(loff_t,
534 startoff, lastoff);
535 unlock_page(page);
536 goto out;
537 }
538 lastoff += bh->b_size;
539 bh = bh->b_this_page;
540 } while (bh != head);
541 }
542
543 lastoff = page_offset(page) + PAGE_SIZE;
544 unlock_page(page);
545 }
546
Jan Kara7d95edd2017-05-21 22:33:23 -0400547 /* The no. of pages is less than our desired, we are done. */
548 if (nr_pages < num)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500549 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500550
551 index = pvec.pages[i - 1]->index + 1;
552 pagevec_release(&pvec);
553 } while (index <= end);
554
Jan Kara7d95edd2017-05-21 22:33:23 -0400555 if (whence == SEEK_HOLE && lastoff < endoff) {
556 found = 1;
557 *offset = lastoff;
558 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500559out:
560 pagevec_release(&pvec);
561 return found;
562}
563
564/*
565 * ext4_seek_data() retrieves the offset for SEEK_DATA.
566 */
567static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
568{
569 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500570 struct extent_status es;
571 ext4_lblk_t start, last, end;
572 loff_t dataoff, isize;
573 int blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500574 int ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500575
Al Viro59551022016-01-22 15:40:57 -0500576 inode_lock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500577
578 isize = i_size_read(inode);
579 if (offset >= isize) {
Al Viro59551022016-01-22 15:40:57 -0500580 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500581 return -ENXIO;
582 }
583
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500584 blkbits = inode->i_sb->s_blocksize_bits;
585 start = offset >> blkbits;
586 last = start;
587 end = isize >> blkbits;
588 dataoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500589
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500590 do {
Jan Kara2d90c162016-03-09 23:11:13 -0500591 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
592 if (ret <= 0) {
593 /* No extent found -> no data */
594 if (ret == 0)
595 ret = -ENXIO;
596 inode_unlock(inode);
597 return ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500598 }
599
Jan Kara2d90c162016-03-09 23:11:13 -0500600 last = es.es_lblk;
601 if (last != start)
602 dataoff = (loff_t)last << blkbits;
603 if (!ext4_es_is_unwritten(&es))
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500604 break;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500605
606 /*
607 * If there is a unwritten extent at this offset,
608 * it will be as a data or a hole according to page
609 * cache that has data or not.
610 */
Jan Kara2d90c162016-03-09 23:11:13 -0500611 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
612 es.es_lblk + es.es_len, &dataoff))
613 break;
614 last += es.es_len;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500615 dataoff = (loff_t)last << blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500616 cond_resched();
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500617 } while (last <= end);
618
Al Viro59551022016-01-22 15:40:57 -0500619 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500620
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500621 if (dataoff > isize)
622 return -ENXIO;
623
624 return vfs_setpos(file, dataoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500625}
626
627/*
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500628 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
Zheng Liuc8c0df22012-11-08 21:57:40 -0500629 */
630static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
631{
632 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500633 struct extent_status es;
634 ext4_lblk_t start, last, end;
635 loff_t holeoff, isize;
636 int blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500637 int ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500638
Al Viro59551022016-01-22 15:40:57 -0500639 inode_lock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500640
641 isize = i_size_read(inode);
642 if (offset >= isize) {
Al Viro59551022016-01-22 15:40:57 -0500643 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500644 return -ENXIO;
645 }
646
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500647 blkbits = inode->i_sb->s_blocksize_bits;
648 start = offset >> blkbits;
649 last = start;
650 end = isize >> blkbits;
651 holeoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500652
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500653 do {
Jan Kara2d90c162016-03-09 23:11:13 -0500654 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
655 if (ret < 0) {
656 inode_unlock(inode);
657 return ret;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500658 }
Jan Kara2d90c162016-03-09 23:11:13 -0500659 /* Found a hole? */
660 if (ret == 0 || es.es_lblk > last) {
661 if (last != start)
662 holeoff = (loff_t)last << blkbits;
663 break;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500664 }
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500665 /*
666 * If there is a unwritten extent at this offset,
667 * it will be as a data or a hole according to page
668 * cache that has data or not.
669 */
Jan Kara2d90c162016-03-09 23:11:13 -0500670 if (ext4_es_is_unwritten(&es) &&
671 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
672 last + es.es_len, &holeoff))
673 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500674
Jan Kara2d90c162016-03-09 23:11:13 -0500675 last += es.es_len;
676 holeoff = (loff_t)last << blkbits;
677 cond_resched();
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500678 } while (last <= end);
679
Al Viro59551022016-01-22 15:40:57 -0500680 inode_unlock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500681
682 if (holeoff > isize)
683 holeoff = isize;
684
685 return vfs_setpos(file, holeoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500686}
687
688/*
Eric Sandeenec7268c2012-04-30 13:14:03 -0500689 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
690 * by calling generic_file_llseek_size() with the appropriate maxbytes
691 * value for each.
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400692 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800693loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400694{
695 struct inode *inode = file->f_mapping->host;
696 loff_t maxbytes;
697
698 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
699 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
700 else
701 maxbytes = inode->i_sb->s_maxbytes;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400702
Andrew Morton965c8e52012-12-17 15:59:39 -0800703 switch (whence) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500704 case SEEK_SET:
705 case SEEK_CUR:
706 case SEEK_END:
Andrew Morton965c8e52012-12-17 15:59:39 -0800707 return generic_file_llseek_size(file, offset, whence,
Zheng Liuc8c0df22012-11-08 21:57:40 -0500708 maxbytes, i_size_read(inode));
709 case SEEK_DATA:
710 return ext4_seek_data(file, offset, maxbytes);
711 case SEEK_HOLE:
712 return ext4_seek_hole(file, offset, maxbytes);
713 }
714
715 return -EINVAL;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400716}
717
Mingming Cao617ba132006-10-11 01:20:53 -0700718const struct file_operations ext4_file_operations = {
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400719 .llseek = ext4_llseek,
Jan Kara364443c2016-11-20 17:36:06 -0500720 .read_iter = ext4_file_read_iter,
Al Viro9b884162014-04-17 16:09:22 -0400721 .write_iter = ext4_file_write_iter,
Andi Kleen5cdd7b22008-04-29 22:03:54 -0400722 .unlocked_ioctl = ext4_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723#ifdef CONFIG_COMPAT
Mingming Cao617ba132006-10-11 01:20:53 -0700724 .compat_ioctl = ext4_compat_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700725#endif
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400726 .mmap = ext4_file_mmap,
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400727 .open = ext4_file_open,
Mingming Cao617ba132006-10-11 01:20:53 -0700728 .release = ext4_release_file,
729 .fsync = ext4_sync_file,
Toshi Kanidbe6ec82016-10-07 16:59:59 -0700730 .get_unmapped_area = thp_get_unmapped_area,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700731 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -0400732 .splice_write = iter_file_splice_write,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100733 .fallocate = ext4_fallocate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700734};
735
Arjan van de Ven754661f2007-02-12 00:55:38 -0800736const struct inode_operations ext4_file_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -0700737 .setattr = ext4_setattr,
David Howells99652ea2017-03-31 18:31:56 +0100738 .getattr = ext4_file_getattr,
Mingming Cao617ba132006-10-11 01:20:53 -0700739 .listxattr = ext4_listxattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200740 .get_acl = ext4_get_acl,
Christoph Hellwig64e178a2013-12-20 05:16:44 -0800741 .set_acl = ext4_set_acl,
Eric Sandeen6873fa02008-10-07 00:46:36 -0400742 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700743};
744