blob: 261ac3734c580a9b555f0e8f5e24fe56ce7ec047 [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
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'o8ad28502014-04-21 14:26:57 -040096 struct blk_plug plug;
Al Viro2ba48ce2015-04-09 13:52:01 -040097 int o_direct = iocb->ki_flags & IOCB_DIRECT;
Jan Karae142d052016-03-08 22:44:50 -050098 int unaligned_aio = 0;
Theodore Ts'o8ad28502014-04-21 14:26:57 -040099 int overwrite = 0;
Zheng Liu85630002012-05-28 18:06:51 -0400100 ssize_t ret;
Theodore Ts'o7608e612014-04-21 14:26:28 -0400101
Al Viro59551022016-01-22 15:40:57 -0500102 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -0400103 ret = generic_write_checks(iocb, from);
104 if (ret <= 0)
Al Viroe768d7f2015-04-07 14:48:22 -0400105 goto out;
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400106
107 /*
Jan Karae142d052016-03-08 22:44:50 -0500108 * Unaligned direct AIO must be serialized among each other as zeroing
109 * of partial blocks of two competing unaligned AIOs can result in data
110 * corruption.
111 */
112 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
113 !is_sync_kiocb(iocb) &&
114 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
115 unaligned_aio = 1;
116 ext4_unwritten_wait(inode);
117 }
118
119 /*
Eric Sandeene2b46572008-01-28 23:58:27 -0500120 * If we have encountered a bitmap-format file, the size limit
121 * is smaller than s_maxbytes, which is for extent-mapped files.
122 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400123 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -0500124 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Eric Sandeene2b46572008-01-28 23:58:27 -0500125
Al Viro3309dd02015-04-09 12:55:47 -0400126 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400127 ret = -EFBIG;
Al Viroe768d7f2015-04-07 14:48:22 -0400128 goto out;
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400129 }
Al Viro3309dd02015-04-09 12:55:47 -0400130 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
Eric Sandeene2b46572008-01-28 23:58:27 -0500131 }
132
Dmitry Monakhova41537e2014-10-30 10:53:16 -0400133 iocb->private = &overwrite;
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400134 if (o_direct) {
Al Viro3309dd02015-04-09 12:55:47 -0400135 size_t length = iov_iter_count(from);
136 loff_t pos = iocb->ki_pos;
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400137 blk_start_plug(&plug);
138
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400139 /* check whether we do a DIO overwrite or not */
Jan Karae142d052016-03-08 22:44:50 -0500140 if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400141 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
142 struct ext4_map_blocks map;
143 unsigned int blkbits = inode->i_blkbits;
144 int err, len;
145
146 map.m_lblk = pos >> blkbits;
147 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
148 - map.m_lblk;
149 len = map.m_len;
150
151 err = ext4_map_blocks(NULL, inode, &map, 0);
152 /*
153 * 'err==len' means that all of blocks has
154 * been preallocated no matter they are
155 * initialized or not. For excluding
156 * unwritten extents, we need to check
157 * m_flags. There are two conditions that
158 * indicate for initialized extents. 1) If we
159 * hit extent cache, EXT4_MAP_MAPPED flag is
160 * returned; 2) If we do a real lookup,
161 * non-flags are returned. So we should check
162 * these two conditions.
163 */
164 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
165 overwrite = 1;
166 }
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400167 }
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400168
Al Viro9b884162014-04-17 16:09:22 -0400169 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -0500170 inode_unlock(inode);
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400171
Christoph Hellwige2592212016-04-07 08:52:01 -0700172 if (ret > 0)
173 ret = generic_write_sync(iocb, ret);
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400174 if (o_direct)
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400175 blk_finish_plug(&plug);
176
Al Viroe768d7f2015-04-07 14:48:22 -0400177 return ret;
178
179out:
Al Viro59551022016-01-22 15:40:57 -0500180 inode_unlock(inode);
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500181 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700182}
183
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800184#ifdef CONFIG_FS_DAX
185static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
186{
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700187 int result;
188 handle_t *handle = NULL;
Jan Karaea3d7202015-12-07 14:28:03 -0500189 struct inode *inode = file_inode(vma->vm_file);
190 struct super_block *sb = inode->i_sb;
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700191 bool write = vmf->flags & FAULT_FLAG_WRITE;
192
193 if (write) {
194 sb_start_pagefault(sb);
195 file_update_time(vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500196 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700197 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
198 EXT4_DATA_TRANS_BLOCKS(sb));
Jan Karaea3d7202015-12-07 14:28:03 -0500199 } else
200 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700201
202 if (IS_ERR(handle))
203 result = VM_FAULT_SIGBUS;
204 else
Ross Zwisler6b524992016-07-26 15:21:05 -0700205 result = dax_fault(vma, vmf, ext4_dax_get_block);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700206
207 if (write) {
208 if (!IS_ERR(handle))
209 ext4_journal_stop(handle);
Jan Karaea3d7202015-12-07 14:28:03 -0500210 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700211 sb_end_pagefault(sb);
Jan Karaea3d7202015-12-07 14:28:03 -0500212 } else
213 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700214
215 return result;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800216}
217
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700218static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
219 pmd_t *pmd, unsigned int flags)
220{
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700221 int result;
222 handle_t *handle = NULL;
223 struct inode *inode = file_inode(vma->vm_file);
224 struct super_block *sb = inode->i_sb;
225 bool write = flags & FAULT_FLAG_WRITE;
226
227 if (write) {
228 sb_start_pagefault(sb);
229 file_update_time(vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500230 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700231 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
232 ext4_chunk_trans_blocks(inode,
233 PMD_SIZE / PAGE_SIZE));
Jan Karaea3d7202015-12-07 14:28:03 -0500234 } else
235 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700236
237 if (IS_ERR(handle))
238 result = VM_FAULT_SIGBUS;
239 else
Ross Zwisler6b524992016-07-26 15:21:05 -0700240 result = dax_pmd_fault(vma, addr, pmd, flags,
Jan Kara02fbd132016-05-11 11:58:48 +0200241 ext4_dax_get_block);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700242
243 if (write) {
244 if (!IS_ERR(handle))
245 ext4_journal_stop(handle);
Jan Karaea3d7202015-12-07 14:28:03 -0500246 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700247 sb_end_pagefault(sb);
Jan Karaea3d7202015-12-07 14:28:03 -0500248 } else
249 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700250
251 return result;
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700252}
253
Jan Karaea3d7202015-12-07 14:28:03 -0500254/*
Ross Zwisler1e9d1802016-02-27 14:01:13 -0500255 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
Jan Karaea3d7202015-12-07 14:28:03 -0500256 * handler we check for races agaist truncate. Note that since we cycle through
257 * i_mmap_sem, we are sure that also any hole punching that began before we
258 * were called is finished by now and so if it included part of the file we
259 * are working on, our pte will get unmapped and the check for pte_same() in
260 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
261 * desired.
262 */
263static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
264 struct vm_fault *vmf)
265{
266 struct inode *inode = file_inode(vma->vm_file);
267 struct super_block *sb = inode->i_sb;
Jan Karaea3d7202015-12-07 14:28:03 -0500268 loff_t size;
Ross Zwislerd5be7a02016-01-22 15:10:53 -0800269 int ret;
Jan Karaea3d7202015-12-07 14:28:03 -0500270
271 sb_start_pagefault(sb);
272 file_update_time(vma->vm_file);
273 down_read(&EXT4_I(inode)->i_mmap_sem);
274 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
275 if (vmf->pgoff >= size)
276 ret = VM_FAULT_SIGBUS;
Ross Zwislerd5be7a02016-01-22 15:10:53 -0800277 else
278 ret = dax_pfn_mkwrite(vma, vmf);
Jan Karaea3d7202015-12-07 14:28:03 -0500279 up_read(&EXT4_I(inode)->i_mmap_sem);
280 sb_end_pagefault(sb);
281
282 return ret;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800283}
284
285static const struct vm_operations_struct ext4_dax_vm_ops = {
286 .fault = ext4_dax_fault,
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700287 .pmd_fault = ext4_dax_pmd_fault,
Ross Zwisler1e9d1802016-02-27 14:01:13 -0500288 .page_mkwrite = ext4_dax_fault,
Jan Karaea3d7202015-12-07 14:28:03 -0500289 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800290};
291#else
292#define ext4_dax_vm_ops ext4_file_vm_ops
293#endif
294
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400295static const struct vm_operations_struct ext4_file_vm_ops = {
Jan Karaea3d7202015-12-07 14:28:03 -0500296 .fault = ext4_filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700297 .map_pages = filemap_map_pages,
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400298 .page_mkwrite = ext4_page_mkwrite,
299};
300
301static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
302{
Michael Halcrowc9c74292015-04-12 00:56:10 -0400303 struct inode *inode = file->f_mapping->host;
304
305 if (ext4_encrypted_inode(inode)) {
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400306 int err = fscrypt_get_encryption_info(inode);
Michael Halcrowc9c74292015-04-12 00:56:10 -0400307 if (err)
308 return 0;
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400309 if (!fscrypt_has_encryption_key(inode))
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400310 return -ENOKEY;
Michael Halcrowc9c74292015-04-12 00:56:10 -0400311 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400312 file_accessed(file);
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800313 if (IS_DAX(file_inode(file))) {
314 vma->vm_ops = &ext4_dax_vm_ops;
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700315 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800316 } else {
317 vma->vm_ops = &ext4_file_vm_ops;
318 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400319 return 0;
320}
321
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400322static int ext4_file_open(struct inode * inode, struct file * filp)
323{
324 struct super_block *sb = inode->i_sb;
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct vfsmount *mnt = filp->f_path.mnt;
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400327 struct dentry *dir;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400328 struct path path;
329 char buf[64], *cp;
Michael Halcrowc9c74292015-04-12 00:56:10 -0400330 int ret;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400331
332 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
333 !(sb->s_flags & MS_RDONLY))) {
334 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
335 /*
336 * Sample where the filesystem has been mounted and
337 * store it in the superblock for sysadmin convenience
338 * when trying to sort through large numbers of block
339 * devices or filesystem images.
340 */
341 memset(buf, 0, sizeof(buf));
Al Viro38991672010-01-23 20:10:29 -0500342 path.mnt = mnt;
343 path.dentry = mnt->mnt_root;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400344 cp = d_path(&path, buf, sizeof(buf));
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400345 if (!IS_ERR(cp)) {
Jan Kara044ce472012-07-22 20:31:31 -0400346 handle_t *handle;
347 int err;
348
Theodore Ts'o9924a922013-02-08 21:59:22 -0500349 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Jan Kara044ce472012-07-22 20:31:31 -0400350 if (IS_ERR(handle))
351 return PTR_ERR(handle);
liang xie5d601252014-05-12 22:06:43 -0400352 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Jan Kara044ce472012-07-22 20:31:31 -0400353 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
354 if (err) {
355 ext4_journal_stop(handle);
356 return err;
357 }
Darrick J. Wongcf803902011-10-25 09:18:41 -0400358 strlcpy(sbi->s_es->s_last_mounted, cp,
359 sizeof(sbi->s_es->s_last_mounted));
Jan Kara044ce472012-07-22 20:31:31 -0400360 ext4_handle_dirty_super(handle, sb);
361 ext4_journal_stop(handle);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400362 }
363 }
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400364 if (ext4_encrypted_inode(inode)) {
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400365 ret = fscrypt_get_encryption_info(inode);
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400366 if (ret)
367 return -EACCES;
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400368 if (!fscrypt_has_encryption_key(inode))
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400369 return -ENOKEY;
370 }
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400371
Miklos Szeredic0a37d4872016-03-26 16:14:42 -0400372 dir = dget_parent(file_dentry(filp));
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400373 if (ext4_encrypted_inode(d_inode(dir)) &&
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400374 !fscrypt_has_permitted_context(d_inode(dir), inode)) {
Theodore Ts'off978b02016-02-08 00:54:26 -0500375 ext4_warning(inode->i_sb,
Jakub Wilk8d2ae1c2016-04-27 01:11:21 -0400376 "Inconsistent encryption contexts: %lu/%lu",
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400377 (unsigned long) d_inode(dir)->i_ino,
Theodore Ts'off978b02016-02-08 00:54:26 -0500378 (unsigned long) inode->i_ino);
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400379 dput(dir);
Theodore Ts'off978b02016-02-08 00:54:26 -0500380 return -EPERM;
381 }
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400382 dput(dir);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500383 /*
384 * Set up the jbd2_inode if we are opening the inode for
385 * writing and the journal is present
386 */
Jan Karaa3612932013-08-16 21:19:41 -0400387 if (filp->f_mode & FMODE_WRITE) {
Michael Halcrowc9c74292015-04-12 00:56:10 -0400388 ret = ext4_inode_attach_jinode(inode);
Jan Karaa3612932013-08-16 21:19:41 -0400389 if (ret < 0)
390 return ret;
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500391 }
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400392 return dquot_file_open(inode, filp);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400393}
394
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400395/*
Zheng Liuc8c0df22012-11-08 21:57:40 -0500396 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
397 * file rather than ext4_ext_walk_space() because we can introduce
398 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
399 * function. When extent status tree has been fully implemented, it will
400 * track all extent status for a file and we can directly use it to
401 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
402 */
403
404/*
405 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
406 * lookup page cache to check whether or not there has some data between
407 * [startoff, endoff] because, if this range contains an unwritten extent,
408 * we determine this extent as a data or a hole according to whether the
409 * page cache has data or not.
410 */
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500411static int ext4_find_unwritten_pgoff(struct inode *inode,
412 int whence,
Jan Kara2d90c162016-03-09 23:11:13 -0500413 ext4_lblk_t end_blk,
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500414 loff_t *offset)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500415{
416 struct pagevec pvec;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500417 unsigned int blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500418 pgoff_t index;
419 pgoff_t end;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500420 loff_t endoff;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500421 loff_t startoff;
422 loff_t lastoff;
423 int found = 0;
424
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500425 blkbits = inode->i_sb->s_blocksize_bits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500426 startoff = *offset;
427 lastoff = startoff;
Jan Kara2d90c162016-03-09 23:11:13 -0500428 endoff = (loff_t)end_blk << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500429
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300430 index = startoff >> PAGE_SHIFT;
431 end = endoff >> PAGE_SHIFT;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500432
433 pagevec_init(&pvec, 0);
434 do {
435 int i, num;
436 unsigned long nr_pages;
437
438 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
439 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
440 (pgoff_t)num);
441 if (nr_pages == 0) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800442 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500443 break;
444
Andrew Morton965c8e52012-12-17 15:59:39 -0800445 BUG_ON(whence != SEEK_HOLE);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500446 /*
447 * If this is the first time to go into the loop and
448 * offset is not beyond the end offset, it will be a
449 * hole at this offset
450 */
451 if (lastoff == startoff || lastoff < endoff)
452 found = 1;
453 break;
454 }
455
456 /*
457 * If this is the first time to go into the loop and
458 * offset is smaller than the first page offset, it will be a
459 * hole at this offset.
460 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800461 if (lastoff == startoff && whence == SEEK_HOLE &&
Zheng Liuc8c0df22012-11-08 21:57:40 -0500462 lastoff < page_offset(pvec.pages[0])) {
463 found = 1;
464 break;
465 }
466
467 for (i = 0; i < nr_pages; i++) {
468 struct page *page = pvec.pages[i];
469 struct buffer_head *bh, *head;
470
471 /*
472 * If the current offset is not beyond the end of given
473 * range, it will be a hole.
474 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800475 if (lastoff < endoff && whence == SEEK_HOLE &&
Zheng Liuc8c0df22012-11-08 21:57:40 -0500476 page->index > end) {
477 found = 1;
478 *offset = lastoff;
479 goto out;
480 }
481
482 lock_page(page);
483
484 if (unlikely(page->mapping != inode->i_mapping)) {
485 unlock_page(page);
486 continue;
487 }
488
489 if (!page_has_buffers(page)) {
490 unlock_page(page);
491 continue;
492 }
493
494 if (page_has_buffers(page)) {
495 lastoff = page_offset(page);
496 bh = head = page_buffers(page);
497 do {
498 if (buffer_uptodate(bh) ||
499 buffer_unwritten(bh)) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800500 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500501 found = 1;
502 } else {
Andrew Morton965c8e52012-12-17 15:59:39 -0800503 if (whence == SEEK_HOLE)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500504 found = 1;
505 }
506 if (found) {
507 *offset = max_t(loff_t,
508 startoff, lastoff);
509 unlock_page(page);
510 goto out;
511 }
512 lastoff += bh->b_size;
513 bh = bh->b_this_page;
514 } while (bh != head);
515 }
516
517 lastoff = page_offset(page) + PAGE_SIZE;
518 unlock_page(page);
519 }
520
521 /*
522 * The no. of pages is less than our desired, that would be a
523 * hole in there.
524 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800525 if (nr_pages < num && whence == SEEK_HOLE) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500526 found = 1;
527 *offset = lastoff;
528 break;
529 }
530
531 index = pvec.pages[i - 1]->index + 1;
532 pagevec_release(&pvec);
533 } while (index <= end);
534
535out:
536 pagevec_release(&pvec);
537 return found;
538}
539
540/*
541 * ext4_seek_data() retrieves the offset for SEEK_DATA.
542 */
543static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
544{
545 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500546 struct extent_status es;
547 ext4_lblk_t start, last, end;
548 loff_t dataoff, isize;
549 int blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500550 int ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500551
Al Viro59551022016-01-22 15:40:57 -0500552 inode_lock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500553
554 isize = i_size_read(inode);
555 if (offset >= isize) {
Al Viro59551022016-01-22 15:40:57 -0500556 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500557 return -ENXIO;
558 }
559
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500560 blkbits = inode->i_sb->s_blocksize_bits;
561 start = offset >> blkbits;
562 last = start;
563 end = isize >> blkbits;
564 dataoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500565
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500566 do {
Jan Kara2d90c162016-03-09 23:11:13 -0500567 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
568 if (ret <= 0) {
569 /* No extent found -> no data */
570 if (ret == 0)
571 ret = -ENXIO;
572 inode_unlock(inode);
573 return ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500574 }
575
Jan Kara2d90c162016-03-09 23:11:13 -0500576 last = es.es_lblk;
577 if (last != start)
578 dataoff = (loff_t)last << blkbits;
579 if (!ext4_es_is_unwritten(&es))
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500580 break;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500581
582 /*
583 * If there is a unwritten extent at this offset,
584 * it will be as a data or a hole according to page
585 * cache that has data or not.
586 */
Jan Kara2d90c162016-03-09 23:11:13 -0500587 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
588 es.es_lblk + es.es_len, &dataoff))
589 break;
590 last += es.es_len;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500591 dataoff = (loff_t)last << blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500592 cond_resched();
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500593 } while (last <= end);
594
Al Viro59551022016-01-22 15:40:57 -0500595 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500596
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500597 if (dataoff > isize)
598 return -ENXIO;
599
600 return vfs_setpos(file, dataoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500601}
602
603/*
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500604 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
Zheng Liuc8c0df22012-11-08 21:57:40 -0500605 */
606static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
607{
608 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500609 struct extent_status es;
610 ext4_lblk_t start, last, end;
611 loff_t holeoff, isize;
612 int blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500613 int ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500614
Al Viro59551022016-01-22 15:40:57 -0500615 inode_lock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500616
617 isize = i_size_read(inode);
618 if (offset >= isize) {
Al Viro59551022016-01-22 15:40:57 -0500619 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500620 return -ENXIO;
621 }
622
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500623 blkbits = inode->i_sb->s_blocksize_bits;
624 start = offset >> blkbits;
625 last = start;
626 end = isize >> blkbits;
627 holeoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500628
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500629 do {
Jan Kara2d90c162016-03-09 23:11:13 -0500630 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
631 if (ret < 0) {
632 inode_unlock(inode);
633 return ret;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500634 }
Jan Kara2d90c162016-03-09 23:11:13 -0500635 /* Found a hole? */
636 if (ret == 0 || es.es_lblk > last) {
637 if (last != start)
638 holeoff = (loff_t)last << blkbits;
639 break;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500640 }
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500641 /*
642 * If there is a unwritten extent at this offset,
643 * it will be as a data or a hole according to page
644 * cache that has data or not.
645 */
Jan Kara2d90c162016-03-09 23:11:13 -0500646 if (ext4_es_is_unwritten(&es) &&
647 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
648 last + es.es_len, &holeoff))
649 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500650
Jan Kara2d90c162016-03-09 23:11:13 -0500651 last += es.es_len;
652 holeoff = (loff_t)last << blkbits;
653 cond_resched();
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500654 } while (last <= end);
655
Al Viro59551022016-01-22 15:40:57 -0500656 inode_unlock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500657
658 if (holeoff > isize)
659 holeoff = isize;
660
661 return vfs_setpos(file, holeoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500662}
663
664/*
Eric Sandeenec7268c2012-04-30 13:14:03 -0500665 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
666 * by calling generic_file_llseek_size() with the appropriate maxbytes
667 * value for each.
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400668 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800669loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400670{
671 struct inode *inode = file->f_mapping->host;
672 loff_t maxbytes;
673
674 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
675 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
676 else
677 maxbytes = inode->i_sb->s_maxbytes;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400678
Andrew Morton965c8e52012-12-17 15:59:39 -0800679 switch (whence) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500680 case SEEK_SET:
681 case SEEK_CUR:
682 case SEEK_END:
Andrew Morton965c8e52012-12-17 15:59:39 -0800683 return generic_file_llseek_size(file, offset, whence,
Zheng Liuc8c0df22012-11-08 21:57:40 -0500684 maxbytes, i_size_read(inode));
685 case SEEK_DATA:
686 return ext4_seek_data(file, offset, maxbytes);
687 case SEEK_HOLE:
688 return ext4_seek_hole(file, offset, maxbytes);
689 }
690
691 return -EINVAL;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400692}
693
Mingming Cao617ba132006-10-11 01:20:53 -0700694const struct file_operations ext4_file_operations = {
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400695 .llseek = ext4_llseek,
Al Viroaad4f8b2014-04-02 14:33:16 -0400696 .read_iter = generic_file_read_iter,
Al Viro9b884162014-04-17 16:09:22 -0400697 .write_iter = ext4_file_write_iter,
Andi Kleen5cdd7b22008-04-29 22:03:54 -0400698 .unlocked_ioctl = ext4_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700699#ifdef CONFIG_COMPAT
Mingming Cao617ba132006-10-11 01:20:53 -0700700 .compat_ioctl = ext4_compat_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700701#endif
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400702 .mmap = ext4_file_mmap,
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400703 .open = ext4_file_open,
Mingming Cao617ba132006-10-11 01:20:53 -0700704 .release = ext4_release_file,
705 .fsync = ext4_sync_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700706 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -0400707 .splice_write = iter_file_splice_write,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100708 .fallocate = ext4_fallocate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700709};
710
Arjan van de Ven754661f2007-02-12 00:55:38 -0800711const struct inode_operations ext4_file_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -0700712 .setattr = ext4_setattr,
Mingming Cao3e3398a2008-07-11 19:27:31 -0400713 .getattr = ext4_getattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700714 .setxattr = generic_setxattr,
715 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -0700716 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700717 .removexattr = generic_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200718 .get_acl = ext4_get_acl,
Christoph Hellwig64e178a2013-12-20 05:16:44 -0800719 .set_acl = ext4_set_acl,
Eric Sandeen6873fa02008-10-07 00:46:36 -0400720 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700721};
722