blob: 58294c9a7e1df0028f267e4eaad67d7b0d2a910b [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
Goldwyn Rodrigues728fbc02017-06-20 07:05:47 -050040 if (!inode_trylock_shared(inode)) {
41 if (iocb->ki_flags & IOCB_NOWAIT)
42 return -EAGAIN;
43 inode_lock_shared(inode);
44 }
Jan Kara364443c2016-11-20 17:36:06 -050045 /*
46 * Recheck under inode lock - at this point we are sure it cannot
47 * change anymore
48 */
49 if (!IS_DAX(inode)) {
50 inode_unlock_shared(inode);
51 /* Fallback to buffered IO in case we cannot support DAX */
52 return generic_file_read_iter(iocb, to);
53 }
54 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
55 inode_unlock_shared(inode);
56
57 file_accessed(iocb->ki_filp);
58 return ret;
59}
60#endif
61
62static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
63{
Theodore Ts'o0db1ff22017-02-05 01:28:48 -050064 if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
65 return -EIO;
66
Jan Kara364443c2016-11-20 17:36:06 -050067 if (!iov_iter_count(to))
68 return 0; /* skip atime */
69
70#ifdef CONFIG_FS_DAX
71 if (IS_DAX(file_inode(iocb->ki_filp)))
72 return ext4_dax_read_iter(iocb, to);
73#endif
74 return generic_file_read_iter(iocb, to);
75}
76
Dave Kleikampac27a0e2006-10-11 01:20:50 -070077/*
78 * Called when an inode is released. Note that this is different
Mingming Cao617ba132006-10-11 01:20:53 -070079 * from ext4_file_open: open gets called at every open, but release
Dave Kleikampac27a0e2006-10-11 01:20:50 -070080 * gets called only when /all/ the files are closed.
81 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -040082static int ext4_release_file(struct inode *inode, struct file *filp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070083{
Theodore Ts'o19f5fb72010-01-24 14:34:07 -050084 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -050085 ext4_alloc_da_blocks(inode);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -050086 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -050087 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -070088 /* if we are the last writer on the inode, drop the block reservation */
89 if ((filp->f_mode & FMODE_WRITE) &&
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -040090 (atomic_read(&inode->i_writecount) == 1) &&
91 !EXT4_I(inode)->i_reserved_data_blocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070092 {
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -050093 down_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -040094 ext4_discard_preallocations(inode);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -050095 up_write(&EXT4_I(inode)->i_data_sem);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 }
97 if (is_dx(inode) && filp->private_data)
Mingming Cao617ba132006-10-11 01:20:53 -070098 ext4_htree_free_dir_info(filp->private_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070099
100 return 0;
101}
102
Stephen Hemmingerc1978552014-05-12 10:50:23 -0400103static void ext4_unwritten_wait(struct inode *inode)
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500104{
105 wait_queue_head_t *wq = ext4_ioend_wq(inode);
106
Dmitry Monakhove27f41e2012-09-28 23:24:52 -0400107 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500108}
109
110/*
111 * This tests whether the IO in question is block-aligned or not.
112 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
113 * are converted to written only after the IO is complete. Until they are
114 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
115 * it needs to zero out portions of the start and/or end block. If 2 AIO
116 * threads are at work on the same unwritten block, they must be synchronized
117 * or one thread will zero the other's data, causing corruption.
118 */
119static int
Al Viro9b884162014-04-17 16:09:22 -0400120ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500121{
122 struct super_block *sb = inode->i_sb;
123 int blockmask = sb->s_blocksize - 1;
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500124
Theodore Ts'o6e6358f2014-04-12 12:45:25 -0400125 if (pos >= i_size_read(inode))
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500126 return 0;
127
Al Viro9b884162014-04-17 16:09:22 -0400128 if ((pos | iov_iter_alignment(from)) & blockmask)
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500129 return 1;
130
131 return 0;
132}
133
Jan Kara213bcd92016-11-20 17:29:51 -0500134/* Is IO overwriting allocated and initialized blocks? */
135static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
136{
137 struct ext4_map_blocks map;
138 unsigned int blkbits = inode->i_blkbits;
139 int err, blklen;
140
141 if (pos + len > i_size_read(inode))
142 return false;
143
144 map.m_lblk = pos >> blkbits;
145 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
146 blklen = map.m_len;
147
148 err = ext4_map_blocks(NULL, inode, &map, 0);
149 /*
150 * 'err==len' means that all of the blocks have been preallocated,
151 * regardless of whether they have been initialized or not. To exclude
152 * unwritten extents, we need to check m_flags.
153 */
154 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
155}
156
157static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
158{
159 struct inode *inode = file_inode(iocb->ki_filp);
160 ssize_t ret;
161
162 ret = generic_write_checks(iocb, from);
163 if (ret <= 0)
164 return ret;
165 /*
166 * If we have encountered a bitmap-format file, the size limit
167 * is smaller than s_maxbytes, which is for extent-mapped files.
168 */
169 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
170 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
171
172 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
173 return -EFBIG;
174 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
175 }
176 return iov_iter_count(from);
177}
178
Jan Kara776722e2016-11-20 18:09:11 -0500179#ifdef CONFIG_FS_DAX
180static ssize_t
181ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
182{
183 struct inode *inode = file_inode(iocb->ki_filp);
184 ssize_t ret;
Jan Kara776722e2016-11-20 18:09:11 -0500185
Goldwyn Rodrigues728fbc02017-06-20 07:05:47 -0500186 if (!inode_trylock(inode)) {
187 if (iocb->ki_flags & IOCB_NOWAIT)
188 return -EAGAIN;
189 inode_lock(inode);
190 }
Jan Kara776722e2016-11-20 18:09:11 -0500191 ret = ext4_write_checks(iocb, from);
192 if (ret <= 0)
193 goto out;
194 ret = file_remove_privs(iocb->ki_filp);
195 if (ret)
196 goto out;
197 ret = file_update_time(iocb->ki_filp);
198 if (ret)
199 goto out;
200
Jan Kara776722e2016-11-20 18:09:11 -0500201 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
202out:
Christoph Hellwigff5462e2017-02-08 14:39:27 -0500203 inode_unlock(inode);
Jan Kara776722e2016-11-20 18:09:11 -0500204 if (ret > 0)
205 ret = generic_write_sync(iocb, ret);
206 return ret;
207}
208#endif
209
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210static ssize_t
Al Viro9b884162014-04-17 16:09:22 -0400211ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700212{
Al Viro496ad9a2013-01-23 17:07:38 -0500213 struct inode *inode = file_inode(iocb->ki_filp);
Al Viro2ba48ce2015-04-09 13:52:01 -0400214 int o_direct = iocb->ki_flags & IOCB_DIRECT;
Jan Karae142d052016-03-08 22:44:50 -0500215 int unaligned_aio = 0;
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400216 int overwrite = 0;
Zheng Liu85630002012-05-28 18:06:51 -0400217 ssize_t ret;
Theodore Ts'o7608e612014-04-21 14:26:28 -0400218
Theodore Ts'o0db1ff22017-02-05 01:28:48 -0500219 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
220 return -EIO;
221
Jan Kara776722e2016-11-20 18:09:11 -0500222#ifdef CONFIG_FS_DAX
223 if (IS_DAX(inode))
224 return ext4_dax_write_iter(iocb, from);
225#endif
226
Goldwyn Rodrigues728fbc02017-06-20 07:05:47 -0500227 if (!inode_trylock(inode)) {
228 if (iocb->ki_flags & IOCB_NOWAIT)
229 return -EAGAIN;
230 inode_lock(inode);
231 }
232
Jan Kara213bcd92016-11-20 17:29:51 -0500233 ret = ext4_write_checks(iocb, from);
Al Viro3309dd02015-04-09 12:55:47 -0400234 if (ret <= 0)
Al Viroe768d7f2015-04-07 14:48:22 -0400235 goto out;
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400236
237 /*
Jan Karae142d052016-03-08 22:44:50 -0500238 * Unaligned direct AIO must be serialized among each other as zeroing
239 * of partial blocks of two competing unaligned AIOs can result in data
240 * corruption.
241 */
242 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
243 !is_sync_kiocb(iocb) &&
244 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
245 unaligned_aio = 1;
246 ext4_unwritten_wait(inode);
247 }
248
Dmitry Monakhova41537e2014-10-30 10:53:16 -0400249 iocb->private = &overwrite;
Jan Kara213bcd92016-11-20 17:29:51 -0500250 /* Check whether we do a DIO overwrite or not */
Goldwyn Rodrigues728fbc02017-06-20 07:05:47 -0500251 if (o_direct && !unaligned_aio) {
252 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
253 if (ext4_should_dioread_nolock(inode))
254 overwrite = 1;
255 } else if (iocb->ki_flags & IOCB_NOWAIT) {
256 ret = -EAGAIN;
257 goto out;
258 }
259 }
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400260
Al Viro9b884162014-04-17 16:09:22 -0400261 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -0500262 inode_unlock(inode);
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400263
Christoph Hellwige2592212016-04-07 08:52:01 -0700264 if (ret > 0)
265 ret = generic_write_sync(iocb, ret);
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400266
Al Viroe768d7f2015-04-07 14:48:22 -0400267 return ret;
268
269out:
Al Viro59551022016-01-22 15:40:57 -0500270 inode_unlock(inode);
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500271 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700272}
273
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800274#ifdef CONFIG_FS_DAX
Dave Jiangc791ace2017-02-24 14:57:08 -0800275static int ext4_dax_huge_fault(struct vm_fault *vmf,
276 enum page_entry_size pe_size)
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800277{
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700278 int result;
Jan Karafb26a1c2017-05-12 15:46:54 -0700279 handle_t *handle = NULL;
Dave Jiang11bac802017-02-24 14:56:41 -0800280 struct inode *inode = file_inode(vmf->vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500281 struct super_block *sb = inode->i_sb;
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700282 bool write = vmf->flags & FAULT_FLAG_WRITE;
283
284 if (write) {
285 sb_start_pagefault(sb);
Dave Jiang11bac802017-02-24 14:56:41 -0800286 file_update_time(vmf->vma->vm_file);
Jan Karafb26a1c2017-05-12 15:46:54 -0700287 down_read(&EXT4_I(inode)->i_mmap_sem);
288 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
289 EXT4_DATA_TRANS_BLOCKS(sb));
290 } else {
291 down_read(&EXT4_I(inode)->i_mmap_sem);
Jan Kara1db17542016-10-21 11:33:49 +0200292 }
Jan Karafb26a1c2017-05-12 15:46:54 -0700293 if (!IS_ERR(handle))
294 result = dax_iomap_fault(vmf, pe_size, &ext4_iomap_ops);
295 else
296 result = VM_FAULT_SIGBUS;
297 if (write) {
298 if (!IS_ERR(handle))
299 ext4_journal_stop(handle);
300 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700301 sb_end_pagefault(sb);
Jan Karafb26a1c2017-05-12 15:46:54 -0700302 } else {
303 up_read(&EXT4_I(inode)->i_mmap_sem);
304 }
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700305
306 return result;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800307}
308
Dave Jiangc791ace2017-02-24 14:57:08 -0800309static int ext4_dax_fault(struct vm_fault *vmf)
310{
311 return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
312}
313
Jan Karaea3d7202015-12-07 14:28:03 -0500314/*
Ross Zwisler1e9d1802016-02-27 14:01:13 -0500315 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
Jan Karaea3d7202015-12-07 14:28:03 -0500316 * handler we check for races agaist truncate. Note that since we cycle through
317 * i_mmap_sem, we are sure that also any hole punching that began before we
318 * were called is finished by now and so if it included part of the file we
319 * are working on, our pte will get unmapped and the check for pte_same() in
320 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
321 * desired.
322 */
Dave Jiang11bac802017-02-24 14:56:41 -0800323static int ext4_dax_pfn_mkwrite(struct vm_fault *vmf)
Jan Karaea3d7202015-12-07 14:28:03 -0500324{
Dave Jiang11bac802017-02-24 14:56:41 -0800325 struct inode *inode = file_inode(vmf->vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500326 struct super_block *sb = inode->i_sb;
Jan Karaea3d7202015-12-07 14:28:03 -0500327 loff_t size;
Ross Zwislerd5be7a02016-01-22 15:10:53 -0800328 int ret;
Jan Karaea3d7202015-12-07 14:28:03 -0500329
330 sb_start_pagefault(sb);
Dave Jiang11bac802017-02-24 14:56:41 -0800331 file_update_time(vmf->vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500332 down_read(&EXT4_I(inode)->i_mmap_sem);
333 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 if (vmf->pgoff >= size)
335 ret = VM_FAULT_SIGBUS;
Ross Zwislerd5be7a02016-01-22 15:10:53 -0800336 else
Dave Jiang11bac802017-02-24 14:56:41 -0800337 ret = dax_pfn_mkwrite(vmf);
Jan Karaea3d7202015-12-07 14:28:03 -0500338 up_read(&EXT4_I(inode)->i_mmap_sem);
339 sb_end_pagefault(sb);
340
341 return ret;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800342}
343
344static const struct vm_operations_struct ext4_dax_vm_ops = {
345 .fault = ext4_dax_fault,
Dave Jiangc791ace2017-02-24 14:57:08 -0800346 .huge_fault = ext4_dax_huge_fault,
Ross Zwisler1e9d1802016-02-27 14:01:13 -0500347 .page_mkwrite = ext4_dax_fault,
Jan Karaea3d7202015-12-07 14:28:03 -0500348 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800349};
350#else
351#define ext4_dax_vm_ops ext4_file_vm_ops
352#endif
353
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400354static const struct vm_operations_struct ext4_file_vm_ops = {
Jan Karaea3d7202015-12-07 14:28:03 -0500355 .fault = ext4_filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700356 .map_pages = filemap_map_pages,
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400357 .page_mkwrite = ext4_page_mkwrite,
358};
359
360static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
361{
Michael Halcrowc9c74292015-04-12 00:56:10 -0400362 struct inode *inode = file->f_mapping->host;
363
Theodore Ts'o0db1ff22017-02-05 01:28:48 -0500364 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
365 return -EIO;
366
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400367 file_accessed(file);
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800368 if (IS_DAX(file_inode(file))) {
369 vma->vm_ops = &ext4_dax_vm_ops;
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700370 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800371 } else {
372 vma->vm_ops = &ext4_file_vm_ops;
373 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400374 return 0;
375}
376
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400377static int ext4_file_open(struct inode * inode, struct file * filp)
378{
379 struct super_block *sb = inode->i_sb;
380 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
381 struct vfsmount *mnt = filp->f_path.mnt;
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400382 struct dentry *dir;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400383 struct path path;
384 char buf[64], *cp;
Michael Halcrowc9c74292015-04-12 00:56:10 -0400385 int ret;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400386
Theodore Ts'o0db1ff22017-02-05 01:28:48 -0500387 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
388 return -EIO;
389
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400390 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
391 !(sb->s_flags & MS_RDONLY))) {
392 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
393 /*
394 * Sample where the filesystem has been mounted and
395 * store it in the superblock for sysadmin convenience
396 * when trying to sort through large numbers of block
397 * devices or filesystem images.
398 */
399 memset(buf, 0, sizeof(buf));
Al Viro38991672010-01-23 20:10:29 -0500400 path.mnt = mnt;
401 path.dentry = mnt->mnt_root;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400402 cp = d_path(&path, buf, sizeof(buf));
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400403 if (!IS_ERR(cp)) {
Jan Kara044ce472012-07-22 20:31:31 -0400404 handle_t *handle;
405 int err;
406
Theodore Ts'o9924a922013-02-08 21:59:22 -0500407 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Jan Kara044ce472012-07-22 20:31:31 -0400408 if (IS_ERR(handle))
409 return PTR_ERR(handle);
liang xie5d601252014-05-12 22:06:43 -0400410 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Jan Kara044ce472012-07-22 20:31:31 -0400411 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
412 if (err) {
413 ext4_journal_stop(handle);
414 return err;
415 }
Darrick J. Wongcf803902011-10-25 09:18:41 -0400416 strlcpy(sbi->s_es->s_last_mounted, cp,
417 sizeof(sbi->s_es->s_last_mounted));
Jan Kara044ce472012-07-22 20:31:31 -0400418 ext4_handle_dirty_super(handle, sb);
419 ext4_journal_stop(handle);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400420 }
421 }
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400422 if (ext4_encrypted_inode(inode)) {
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400423 ret = fscrypt_get_encryption_info(inode);
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400424 if (ret)
425 return -EACCES;
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400426 if (!fscrypt_has_encryption_key(inode))
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400427 return -ENOKEY;
428 }
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400429
Miklos Szeredic0a37d4872016-03-26 16:14:42 -0400430 dir = dget_parent(file_dentry(filp));
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400431 if (ext4_encrypted_inode(d_inode(dir)) &&
Jaegeuk Kima7550b32016-07-10 14:01:03 -0400432 !fscrypt_has_permitted_context(d_inode(dir), inode)) {
Theodore Ts'off978b02016-02-08 00:54:26 -0500433 ext4_warning(inode->i_sb,
Jakub Wilk8d2ae1c2016-04-27 01:11:21 -0400434 "Inconsistent encryption contexts: %lu/%lu",
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400435 (unsigned long) d_inode(dir)->i_ino,
Theodore Ts'off978b02016-02-08 00:54:26 -0500436 (unsigned long) inode->i_ino);
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400437 dput(dir);
Theodore Ts'off978b02016-02-08 00:54:26 -0500438 return -EPERM;
439 }
Miklos Szeredi9dd78d82016-03-26 16:14:41 -0400440 dput(dir);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500441 /*
442 * Set up the jbd2_inode if we are opening the inode for
443 * writing and the journal is present
444 */
Jan Karaa3612932013-08-16 21:19:41 -0400445 if (filp->f_mode & FMODE_WRITE) {
Michael Halcrowc9c74292015-04-12 00:56:10 -0400446 ret = ext4_inode_attach_jinode(inode);
Jan Karaa3612932013-08-16 21:19:41 -0400447 if (ret < 0)
448 return ret;
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500449 }
Goldwyn Rodrigues728fbc02017-06-20 07:05:47 -0500450
451 /* Set the flags to support nowait AIO */
452 filp->f_mode |= FMODE_AIO_NOWAIT;
453
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400454 return dquot_file_open(inode, filp);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400455}
456
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400457/*
Zheng Liuc8c0df22012-11-08 21:57:40 -0500458 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
459 * file rather than ext4_ext_walk_space() because we can introduce
460 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
461 * function. When extent status tree has been fully implemented, it will
462 * track all extent status for a file and we can directly use it to
463 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
464 */
465
466/*
467 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
468 * lookup page cache to check whether or not there has some data between
469 * [startoff, endoff] because, if this range contains an unwritten extent,
470 * we determine this extent as a data or a hole according to whether the
471 * page cache has data or not.
472 */
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500473static int ext4_find_unwritten_pgoff(struct inode *inode,
474 int whence,
Jan Kara2d90c162016-03-09 23:11:13 -0500475 ext4_lblk_t end_blk,
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500476 loff_t *offset)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500477{
478 struct pagevec pvec;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500479 unsigned int blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500480 pgoff_t index;
481 pgoff_t end;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500482 loff_t endoff;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500483 loff_t startoff;
484 loff_t lastoff;
485 int found = 0;
486
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500487 blkbits = inode->i_sb->s_blocksize_bits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500488 startoff = *offset;
489 lastoff = startoff;
Jan Kara2d90c162016-03-09 23:11:13 -0500490 endoff = (loff_t)end_blk << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500491
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300492 index = startoff >> PAGE_SHIFT;
Jan Kara3f1d5ba2017-05-21 22:34:23 -0400493 end = (endoff - 1) >> PAGE_SHIFT;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500494
495 pagevec_init(&pvec, 0);
496 do {
497 int i, num;
498 unsigned long nr_pages;
499
Eryu Guan624327f2017-05-24 18:02:20 -0400500 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE - 1) + 1;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500501 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
502 (pgoff_t)num);
Jan Kara7d95edd2017-05-21 22:33:23 -0400503 if (nr_pages == 0)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500504 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500505
506 for (i = 0; i < nr_pages; i++) {
507 struct page *page = pvec.pages[i];
508 struct buffer_head *bh, *head;
509
510 /*
Jan Kara7d95edd2017-05-21 22:33:23 -0400511 * If current offset is smaller than the page offset,
512 * there is a hole at this offset.
Zheng Liuc8c0df22012-11-08 21:57:40 -0500513 */
Jan Kara7d95edd2017-05-21 22:33:23 -0400514 if (whence == SEEK_HOLE && lastoff < endoff &&
515 lastoff < page_offset(pvec.pages[i])) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500516 found = 1;
517 *offset = lastoff;
518 goto out;
519 }
520
Jan Kara7d95edd2017-05-21 22:33:23 -0400521 if (page->index > end)
522 goto out;
523
Zheng Liuc8c0df22012-11-08 21:57:40 -0500524 lock_page(page);
525
526 if (unlikely(page->mapping != inode->i_mapping)) {
527 unlock_page(page);
528 continue;
529 }
530
531 if (!page_has_buffers(page)) {
532 unlock_page(page);
533 continue;
534 }
535
536 if (page_has_buffers(page)) {
537 lastoff = page_offset(page);
538 bh = head = page_buffers(page);
539 do {
540 if (buffer_uptodate(bh) ||
541 buffer_unwritten(bh)) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800542 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500543 found = 1;
544 } else {
Andrew Morton965c8e52012-12-17 15:59:39 -0800545 if (whence == SEEK_HOLE)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500546 found = 1;
547 }
548 if (found) {
549 *offset = max_t(loff_t,
550 startoff, lastoff);
551 unlock_page(page);
552 goto out;
553 }
554 lastoff += bh->b_size;
555 bh = bh->b_this_page;
556 } while (bh != head);
557 }
558
559 lastoff = page_offset(page) + PAGE_SIZE;
560 unlock_page(page);
561 }
562
Jan Kara7d95edd2017-05-21 22:33:23 -0400563 /* The no. of pages is less than our desired, we are done. */
564 if (nr_pages < num)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500565 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500566
567 index = pvec.pages[i - 1]->index + 1;
568 pagevec_release(&pvec);
569 } while (index <= end);
570
Jan Kara7d95edd2017-05-21 22:33:23 -0400571 if (whence == SEEK_HOLE && lastoff < endoff) {
572 found = 1;
573 *offset = lastoff;
574 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500575out:
576 pagevec_release(&pvec);
577 return found;
578}
579
580/*
581 * ext4_seek_data() retrieves the offset for SEEK_DATA.
582 */
583static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
584{
585 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500586 struct extent_status es;
587 ext4_lblk_t start, last, end;
588 loff_t dataoff, isize;
589 int blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500590 int ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500591
Al Viro59551022016-01-22 15:40:57 -0500592 inode_lock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500593
594 isize = i_size_read(inode);
595 if (offset >= isize) {
Al Viro59551022016-01-22 15:40:57 -0500596 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500597 return -ENXIO;
598 }
599
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500600 blkbits = inode->i_sb->s_blocksize_bits;
601 start = offset >> blkbits;
602 last = start;
603 end = isize >> blkbits;
604 dataoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500605
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500606 do {
Jan Kara2d90c162016-03-09 23:11:13 -0500607 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
608 if (ret <= 0) {
609 /* No extent found -> no data */
610 if (ret == 0)
611 ret = -ENXIO;
612 inode_unlock(inode);
613 return ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500614 }
615
Jan Kara2d90c162016-03-09 23:11:13 -0500616 last = es.es_lblk;
617 if (last != start)
618 dataoff = (loff_t)last << blkbits;
619 if (!ext4_es_is_unwritten(&es))
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500620 break;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500621
622 /*
623 * If there is a unwritten extent at this offset,
624 * it will be as a data or a hole according to page
625 * cache that has data or not.
626 */
Jan Kara2d90c162016-03-09 23:11:13 -0500627 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
628 es.es_lblk + es.es_len, &dataoff))
629 break;
630 last += es.es_len;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500631 dataoff = (loff_t)last << blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500632 cond_resched();
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500633 } while (last <= end);
634
Al Viro59551022016-01-22 15:40:57 -0500635 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500636
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500637 if (dataoff > isize)
638 return -ENXIO;
639
640 return vfs_setpos(file, dataoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500641}
642
643/*
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500644 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
Zheng Liuc8c0df22012-11-08 21:57:40 -0500645 */
646static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
647{
648 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500649 struct extent_status es;
650 ext4_lblk_t start, last, end;
651 loff_t holeoff, isize;
652 int blkbits;
Jan Kara2d90c162016-03-09 23:11:13 -0500653 int ret;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500654
Al Viro59551022016-01-22 15:40:57 -0500655 inode_lock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500656
657 isize = i_size_read(inode);
658 if (offset >= isize) {
Al Viro59551022016-01-22 15:40:57 -0500659 inode_unlock(inode);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500660 return -ENXIO;
661 }
662
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500663 blkbits = inode->i_sb->s_blocksize_bits;
664 start = offset >> blkbits;
665 last = start;
666 end = isize >> blkbits;
667 holeoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500668
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500669 do {
Jan Kara2d90c162016-03-09 23:11:13 -0500670 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
671 if (ret < 0) {
672 inode_unlock(inode);
673 return ret;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500674 }
Jan Kara2d90c162016-03-09 23:11:13 -0500675 /* Found a hole? */
676 if (ret == 0 || es.es_lblk > last) {
677 if (last != start)
678 holeoff = (loff_t)last << blkbits;
679 break;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500680 }
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500681 /*
682 * If there is a unwritten extent at this offset,
683 * it will be as a data or a hole according to page
684 * cache that has data or not.
685 */
Jan Kara2d90c162016-03-09 23:11:13 -0500686 if (ext4_es_is_unwritten(&es) &&
687 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
688 last + es.es_len, &holeoff))
689 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500690
Jan Kara2d90c162016-03-09 23:11:13 -0500691 last += es.es_len;
692 holeoff = (loff_t)last << blkbits;
693 cond_resched();
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500694 } while (last <= end);
695
Al Viro59551022016-01-22 15:40:57 -0500696 inode_unlock(inode);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500697
698 if (holeoff > isize)
699 holeoff = isize;
700
701 return vfs_setpos(file, holeoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500702}
703
704/*
Eric Sandeenec7268c2012-04-30 13:14:03 -0500705 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
706 * by calling generic_file_llseek_size() with the appropriate maxbytes
707 * value for each.
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400708 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800709loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400710{
711 struct inode *inode = file->f_mapping->host;
712 loff_t maxbytes;
713
714 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
715 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
716 else
717 maxbytes = inode->i_sb->s_maxbytes;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400718
Andrew Morton965c8e52012-12-17 15:59:39 -0800719 switch (whence) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500720 case SEEK_SET:
721 case SEEK_CUR:
722 case SEEK_END:
Andrew Morton965c8e52012-12-17 15:59:39 -0800723 return generic_file_llseek_size(file, offset, whence,
Zheng Liuc8c0df22012-11-08 21:57:40 -0500724 maxbytes, i_size_read(inode));
725 case SEEK_DATA:
726 return ext4_seek_data(file, offset, maxbytes);
727 case SEEK_HOLE:
728 return ext4_seek_hole(file, offset, maxbytes);
729 }
730
731 return -EINVAL;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400732}
733
Mingming Cao617ba132006-10-11 01:20:53 -0700734const struct file_operations ext4_file_operations = {
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400735 .llseek = ext4_llseek,
Jan Kara364443c2016-11-20 17:36:06 -0500736 .read_iter = ext4_file_read_iter,
Al Viro9b884162014-04-17 16:09:22 -0400737 .write_iter = ext4_file_write_iter,
Andi Kleen5cdd7b22008-04-29 22:03:54 -0400738 .unlocked_ioctl = ext4_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700739#ifdef CONFIG_COMPAT
Mingming Cao617ba132006-10-11 01:20:53 -0700740 .compat_ioctl = ext4_compat_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700741#endif
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400742 .mmap = ext4_file_mmap,
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400743 .open = ext4_file_open,
Mingming Cao617ba132006-10-11 01:20:53 -0700744 .release = ext4_release_file,
745 .fsync = ext4_sync_file,
Toshi Kanidbe6ec82016-10-07 16:59:59 -0700746 .get_unmapped_area = thp_get_unmapped_area,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700747 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -0400748 .splice_write = iter_file_splice_write,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100749 .fallocate = ext4_fallocate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700750};
751
Arjan van de Ven754661f2007-02-12 00:55:38 -0800752const struct inode_operations ext4_file_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -0700753 .setattr = ext4_setattr,
David Howells99652ea2017-03-31 18:31:56 +0100754 .getattr = ext4_file_getattr,
Mingming Cao617ba132006-10-11 01:20:53 -0700755 .listxattr = ext4_listxattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200756 .get_acl = ext4_get_acl,
Christoph Hellwig64e178a2013-12-20 05:16:44 -0800757 .set_acl = ext4_set_acl,
Eric Sandeen6873fa02008-10-07 00:46:36 -0400758 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700759};
760