blob: 0d24ebcd7c9e535dc3126d5c10b18c745aef8a2e [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'o7ed07ba2014-04-21 14:36:30 -040096 struct mutex *aio_mutex = NULL;
Theodore Ts'o8ad28502014-04-21 14:26:57 -040097 struct blk_plug plug;
Al Viro2ba48ce2015-04-09 13:52:01 -040098 int o_direct = iocb->ki_flags & IOCB_DIRECT;
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
Eric Sandeene2b46572008-01-28 23:58:27 -0500102 /*
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400103 * Unaligned direct AIO must be serialized; see comment above
104 * In the case of O_APPEND, assume that we must always serialize
105 */
106 if (o_direct &&
107 ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
108 !is_sync_kiocb(iocb) &&
Al Viro2ba48ce2015-04-09 13:52:01 -0400109 (iocb->ki_flags & IOCB_APPEND ||
Al Viroe768d7f2015-04-07 14:48:22 -0400110 ext4_unaligned_aio(inode, from, iocb->ki_pos))) {
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400111 aio_mutex = ext4_aio_mutex(inode);
112 mutex_lock(aio_mutex);
113 ext4_unwritten_wait(inode);
114 }
115
116 mutex_lock(&inode->i_mutex);
Al Viro3309dd02015-04-09 12:55:47 -0400117 ret = generic_write_checks(iocb, from);
118 if (ret <= 0)
Al Viroe768d7f2015-04-07 14:48:22 -0400119 goto out;
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400120
121 /*
Eric Sandeene2b46572008-01-28 23:58:27 -0500122 * If we have encountered a bitmap-format file, the size limit
123 * is smaller than s_maxbytes, which is for extent-mapped files.
124 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400125 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -0500126 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Eric Sandeene2b46572008-01-28 23:58:27 -0500127
Al Viro3309dd02015-04-09 12:55:47 -0400128 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400129 ret = -EFBIG;
Al Viroe768d7f2015-04-07 14:48:22 -0400130 goto out;
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400131 }
Al Viro3309dd02015-04-09 12:55:47 -0400132 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
Eric Sandeene2b46572008-01-28 23:58:27 -0500133 }
134
Dmitry Monakhova41537e2014-10-30 10:53:16 -0400135 iocb->private = &overwrite;
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400136 if (o_direct) {
Al Viro3309dd02015-04-09 12:55:47 -0400137 size_t length = iov_iter_count(from);
138 loff_t pos = iocb->ki_pos;
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400139 blk_start_plug(&plug);
140
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400141 /* check whether we do a DIO overwrite or not */
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400142 if (ext4_should_dioread_nolock(inode) && !aio_mutex &&
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400143 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
144 struct ext4_map_blocks map;
145 unsigned int blkbits = inode->i_blkbits;
146 int err, len;
147
148 map.m_lblk = pos >> blkbits;
149 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
150 - map.m_lblk;
151 len = map.m_len;
152
153 err = ext4_map_blocks(NULL, inode, &map, 0);
154 /*
155 * 'err==len' means that all of blocks has
156 * been preallocated no matter they are
157 * initialized or not. For excluding
158 * unwritten extents, we need to check
159 * m_flags. There are two conditions that
160 * indicate for initialized extents. 1) If we
161 * hit extent cache, EXT4_MAP_MAPPED flag is
162 * returned; 2) If we do a real lookup,
163 * non-flags are returned. So we should check
164 * these two conditions.
165 */
166 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
167 overwrite = 1;
168 }
Theodore Ts'of5ccfe12014-04-21 14:37:52 -0400169 }
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400170
Al Viro9b884162014-04-17 16:09:22 -0400171 ret = __generic_file_write_iter(iocb, from);
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400172 mutex_unlock(&inode->i_mutex);
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400173
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400174 if (ret > 0) {
175 ssize_t err;
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400176
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400177 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
178 if (err < 0)
179 ret = err;
180 }
181 if (o_direct)
Theodore Ts'o8ad28502014-04-21 14:26:57 -0400182 blk_finish_plug(&plug);
183
Al Viroe768d7f2015-04-07 14:48:22 -0400184 if (aio_mutex)
185 mutex_unlock(aio_mutex);
186 return ret;
187
188out:
189 mutex_unlock(&inode->i_mutex);
Theodore Ts'o7ed07ba2014-04-21 14:36:30 -0400190 if (aio_mutex)
191 mutex_unlock(aio_mutex);
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500192 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700193}
194
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800195#ifdef CONFIG_FS_DAX
Dave Chinnere842f292015-06-04 09:18:18 +1000196static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate)
197{
198 struct inode *inode = bh->b_assoc_map->host;
Matthew Wilcoxe676a4c2015-09-08 14:59:11 -0700199 /* XXX: breaks on 32-bit > 16TB. Is that even supported? */
Dave Chinnere842f292015-06-04 09:18:18 +1000200 loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits;
201 int err;
202 if (!uptodate)
203 return;
204 WARN_ON(!buffer_unwritten(bh));
205 err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size);
206}
207
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800208static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
209{
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700210 int result;
211 handle_t *handle = NULL;
Jan Karaea3d7202015-12-07 14:28:03 -0500212 struct inode *inode = file_inode(vma->vm_file);
213 struct super_block *sb = inode->i_sb;
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700214 bool write = vmf->flags & FAULT_FLAG_WRITE;
215
216 if (write) {
217 sb_start_pagefault(sb);
218 file_update_time(vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500219 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700220 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
221 EXT4_DATA_TRANS_BLOCKS(sb));
Jan Karaea3d7202015-12-07 14:28:03 -0500222 } else
223 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700224
225 if (IS_ERR(handle))
226 result = VM_FAULT_SIGBUS;
227 else
228 result = __dax_fault(vma, vmf, ext4_get_block_dax,
229 ext4_end_io_unwritten);
230
231 if (write) {
232 if (!IS_ERR(handle))
233 ext4_journal_stop(handle);
Jan Karaea3d7202015-12-07 14:28:03 -0500234 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700235 sb_end_pagefault(sb);
Jan Karaea3d7202015-12-07 14:28:03 -0500236 } else
237 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700238
239 return result;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800240}
241
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700242static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
243 pmd_t *pmd, unsigned int flags)
244{
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700245 int result;
246 handle_t *handle = NULL;
247 struct inode *inode = file_inode(vma->vm_file);
248 struct super_block *sb = inode->i_sb;
249 bool write = flags & FAULT_FLAG_WRITE;
250
251 if (write) {
252 sb_start_pagefault(sb);
253 file_update_time(vma->vm_file);
Jan Karaea3d7202015-12-07 14:28:03 -0500254 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700255 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
256 ext4_chunk_trans_blocks(inode,
257 PMD_SIZE / PAGE_SIZE));
Jan Karaea3d7202015-12-07 14:28:03 -0500258 } else
259 down_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700260
261 if (IS_ERR(handle))
262 result = VM_FAULT_SIGBUS;
263 else
264 result = __dax_pmd_fault(vma, addr, pmd, flags,
265 ext4_get_block_dax, ext4_end_io_unwritten);
266
267 if (write) {
268 if (!IS_ERR(handle))
269 ext4_journal_stop(handle);
Jan Karaea3d7202015-12-07 14:28:03 -0500270 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700271 sb_end_pagefault(sb);
Jan Karaea3d7202015-12-07 14:28:03 -0500272 } else
273 up_read(&EXT4_I(inode)->i_mmap_sem);
Matthew Wilcox01a33b42015-09-08 14:59:22 -0700274
275 return result;
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700276}
277
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800278static int ext4_dax_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
279{
Jan Karaea3d7202015-12-07 14:28:03 -0500280 int err;
281 struct inode *inode = file_inode(vma->vm_file);
282
283 sb_start_pagefault(inode->i_sb);
284 file_update_time(vma->vm_file);
285 down_read(&EXT4_I(inode)->i_mmap_sem);
286 err = __dax_mkwrite(vma, vmf, ext4_get_block_dax,
287 ext4_end_io_unwritten);
288 up_read(&EXT4_I(inode)->i_mmap_sem);
289 sb_end_pagefault(inode->i_sb);
290
291 return err;
292}
293
294/*
295 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite()
296 * handler we check for races agaist truncate. Note that since we cycle through
297 * i_mmap_sem, we are sure that also any hole punching that began before we
298 * were called is finished by now and so if it included part of the file we
299 * are working on, our pte will get unmapped and the check for pte_same() in
300 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
301 * desired.
302 */
303static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
304 struct vm_fault *vmf)
305{
306 struct inode *inode = file_inode(vma->vm_file);
307 struct super_block *sb = inode->i_sb;
308 int ret = VM_FAULT_NOPAGE;
309 loff_t size;
310
311 sb_start_pagefault(sb);
312 file_update_time(vma->vm_file);
313 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;
317 up_read(&EXT4_I(inode)->i_mmap_sem);
318 sb_end_pagefault(sb);
319
320 return ret;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800321}
322
323static const struct vm_operations_struct ext4_dax_vm_ops = {
324 .fault = ext4_dax_fault,
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700325 .pmd_fault = ext4_dax_pmd_fault,
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800326 .page_mkwrite = ext4_dax_mkwrite,
Jan Karaea3d7202015-12-07 14:28:03 -0500327 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800328};
329#else
330#define ext4_dax_vm_ops ext4_file_vm_ops
331#endif
332
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400333static const struct vm_operations_struct ext4_file_vm_ops = {
Jan Karaea3d7202015-12-07 14:28:03 -0500334 .fault = ext4_filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700335 .map_pages = filemap_map_pages,
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400336 .page_mkwrite = ext4_page_mkwrite,
337};
338
339static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
340{
Michael Halcrowc9c74292015-04-12 00:56:10 -0400341 struct inode *inode = file->f_mapping->host;
342
343 if (ext4_encrypted_inode(inode)) {
Theodore Ts'ob7236e22015-05-18 13:17:47 -0400344 int err = ext4_get_encryption_info(inode);
Michael Halcrowc9c74292015-04-12 00:56:10 -0400345 if (err)
346 return 0;
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400347 if (ext4_encryption_info(inode) == NULL)
348 return -ENOKEY;
Michael Halcrowc9c74292015-04-12 00:56:10 -0400349 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400350 file_accessed(file);
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800351 if (IS_DAX(file_inode(file))) {
352 vma->vm_ops = &ext4_dax_vm_ops;
Matthew Wilcox11bd1a92015-09-08 14:59:03 -0700353 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800354 } else {
355 vma->vm_ops = &ext4_file_vm_ops;
356 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400357 return 0;
358}
359
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400360static int ext4_file_open(struct inode * inode, struct file * filp)
361{
362 struct super_block *sb = inode->i_sb;
363 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
364 struct vfsmount *mnt = filp->f_path.mnt;
365 struct path path;
366 char buf[64], *cp;
Michael Halcrowc9c74292015-04-12 00:56:10 -0400367 int ret;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400368
369 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
370 !(sb->s_flags & MS_RDONLY))) {
371 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
372 /*
373 * Sample where the filesystem has been mounted and
374 * store it in the superblock for sysadmin convenience
375 * when trying to sort through large numbers of block
376 * devices or filesystem images.
377 */
378 memset(buf, 0, sizeof(buf));
Al Viro38991672010-01-23 20:10:29 -0500379 path.mnt = mnt;
380 path.dentry = mnt->mnt_root;
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400381 cp = d_path(&path, buf, sizeof(buf));
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400382 if (!IS_ERR(cp)) {
Jan Kara044ce472012-07-22 20:31:31 -0400383 handle_t *handle;
384 int err;
385
Theodore Ts'o9924a922013-02-08 21:59:22 -0500386 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
Jan Kara044ce472012-07-22 20:31:31 -0400387 if (IS_ERR(handle))
388 return PTR_ERR(handle);
liang xie5d601252014-05-12 22:06:43 -0400389 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Jan Kara044ce472012-07-22 20:31:31 -0400390 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
391 if (err) {
392 ext4_journal_stop(handle);
393 return err;
394 }
Darrick J. Wongcf803902011-10-25 09:18:41 -0400395 strlcpy(sbi->s_es->s_last_mounted, cp,
396 sizeof(sbi->s_es->s_last_mounted));
Jan Kara044ce472012-07-22 20:31:31 -0400397 ext4_handle_dirty_super(handle, sb);
398 ext4_journal_stop(handle);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400399 }
400 }
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400401 if (ext4_encrypted_inode(inode)) {
402 ret = ext4_get_encryption_info(inode);
403 if (ret)
404 return -EACCES;
405 if (ext4_encryption_info(inode) == NULL)
406 return -ENOKEY;
407 }
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500408 /*
409 * Set up the jbd2_inode if we are opening the inode for
410 * writing and the journal is present
411 */
Jan Karaa3612932013-08-16 21:19:41 -0400412 if (filp->f_mode & FMODE_WRITE) {
Michael Halcrowc9c74292015-04-12 00:56:10 -0400413 ret = ext4_inode_attach_jinode(inode);
Jan Karaa3612932013-08-16 21:19:41 -0400414 if (ret < 0)
415 return ret;
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500416 }
Theodore Ts'oabdd4382015-05-31 13:35:39 -0400417 return dquot_file_open(inode, filp);
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400418}
419
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400420/*
Zheng Liuc8c0df22012-11-08 21:57:40 -0500421 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
422 * file rather than ext4_ext_walk_space() because we can introduce
423 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
424 * function. When extent status tree has been fully implemented, it will
425 * track all extent status for a file and we can directly use it to
426 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
427 */
428
429/*
430 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
431 * lookup page cache to check whether or not there has some data between
432 * [startoff, endoff] because, if this range contains an unwritten extent,
433 * we determine this extent as a data or a hole according to whether the
434 * page cache has data or not.
435 */
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500436static int ext4_find_unwritten_pgoff(struct inode *inode,
437 int whence,
438 struct ext4_map_blocks *map,
439 loff_t *offset)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500440{
441 struct pagevec pvec;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500442 unsigned int blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500443 pgoff_t index;
444 pgoff_t end;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500445 loff_t endoff;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500446 loff_t startoff;
447 loff_t lastoff;
448 int found = 0;
449
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500450 blkbits = inode->i_sb->s_blocksize_bits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500451 startoff = *offset;
452 lastoff = startoff;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500453 endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500454
455 index = startoff >> PAGE_CACHE_SHIFT;
456 end = endoff >> PAGE_CACHE_SHIFT;
457
458 pagevec_init(&pvec, 0);
459 do {
460 int i, num;
461 unsigned long nr_pages;
462
463 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
464 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
465 (pgoff_t)num);
466 if (nr_pages == 0) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800467 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500468 break;
469
Andrew Morton965c8e52012-12-17 15:59:39 -0800470 BUG_ON(whence != SEEK_HOLE);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500471 /*
472 * If this is the first time to go into the loop and
473 * offset is not beyond the end offset, it will be a
474 * hole at this offset
475 */
476 if (lastoff == startoff || lastoff < endoff)
477 found = 1;
478 break;
479 }
480
481 /*
482 * If this is the first time to go into the loop and
483 * offset is smaller than the first page offset, it will be a
484 * hole at this offset.
485 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800486 if (lastoff == startoff && whence == SEEK_HOLE &&
Zheng Liuc8c0df22012-11-08 21:57:40 -0500487 lastoff < page_offset(pvec.pages[0])) {
488 found = 1;
489 break;
490 }
491
492 for (i = 0; i < nr_pages; i++) {
493 struct page *page = pvec.pages[i];
494 struct buffer_head *bh, *head;
495
496 /*
497 * If the current offset is not beyond the end of given
498 * range, it will be a hole.
499 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800500 if (lastoff < endoff && whence == SEEK_HOLE &&
Zheng Liuc8c0df22012-11-08 21:57:40 -0500501 page->index > end) {
502 found = 1;
503 *offset = lastoff;
504 goto out;
505 }
506
507 lock_page(page);
508
509 if (unlikely(page->mapping != inode->i_mapping)) {
510 unlock_page(page);
511 continue;
512 }
513
514 if (!page_has_buffers(page)) {
515 unlock_page(page);
516 continue;
517 }
518
519 if (page_has_buffers(page)) {
520 lastoff = page_offset(page);
521 bh = head = page_buffers(page);
522 do {
523 if (buffer_uptodate(bh) ||
524 buffer_unwritten(bh)) {
Andrew Morton965c8e52012-12-17 15:59:39 -0800525 if (whence == SEEK_DATA)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500526 found = 1;
527 } else {
Andrew Morton965c8e52012-12-17 15:59:39 -0800528 if (whence == SEEK_HOLE)
Zheng Liuc8c0df22012-11-08 21:57:40 -0500529 found = 1;
530 }
531 if (found) {
532 *offset = max_t(loff_t,
533 startoff, lastoff);
534 unlock_page(page);
535 goto out;
536 }
537 lastoff += bh->b_size;
538 bh = bh->b_this_page;
539 } while (bh != head);
540 }
541
542 lastoff = page_offset(page) + PAGE_SIZE;
543 unlock_page(page);
544 }
545
546 /*
547 * The no. of pages is less than our desired, that would be a
548 * hole in there.
549 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800550 if (nr_pages < num && whence == SEEK_HOLE) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500551 found = 1;
552 *offset = lastoff;
553 break;
554 }
555
556 index = pvec.pages[i - 1]->index + 1;
557 pagevec_release(&pvec);
558 } while (index <= end);
559
560out:
561 pagevec_release(&pvec);
562 return found;
563}
564
565/*
566 * ext4_seek_data() retrieves the offset for SEEK_DATA.
567 */
568static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
569{
570 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500571 struct ext4_map_blocks map;
572 struct extent_status es;
573 ext4_lblk_t start, last, end;
574 loff_t dataoff, isize;
575 int blkbits;
576 int ret = 0;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500577
578 mutex_lock(&inode->i_mutex);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500579
580 isize = i_size_read(inode);
581 if (offset >= isize) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500582 mutex_unlock(&inode->i_mutex);
583 return -ENXIO;
584 }
585
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500586 blkbits = inode->i_sb->s_blocksize_bits;
587 start = offset >> blkbits;
588 last = start;
589 end = isize >> blkbits;
590 dataoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500591
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500592 do {
593 map.m_lblk = last;
594 map.m_len = end - last + 1;
595 ret = ext4_map_blocks(NULL, inode, &map, 0);
596 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
597 if (last != start)
598 dataoff = (loff_t)last << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500599 break;
600 }
601
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500602 /*
603 * If there is a delay extent at this offset,
604 * it will be as a data.
605 */
606 ext4_es_find_delayed_extent_range(inode, last, last, &es);
607 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
608 if (last != start)
609 dataoff = (loff_t)last << blkbits;
610 break;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500611 }
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500612
613 /*
614 * If there is a unwritten extent at this offset,
615 * it will be as a data or a hole according to page
616 * cache that has data or not.
617 */
618 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
619 int unwritten;
620 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
621 &map, &dataoff);
622 if (unwritten)
623 break;
624 }
625
626 last++;
627 dataoff = (loff_t)last << blkbits;
628 } while (last <= end);
629
Zheng Liuc8c0df22012-11-08 21:57:40 -0500630 mutex_unlock(&inode->i_mutex);
631
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500632 if (dataoff > isize)
633 return -ENXIO;
634
635 return vfs_setpos(file, dataoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500636}
637
638/*
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500639 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
Zheng Liuc8c0df22012-11-08 21:57:40 -0500640 */
641static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
642{
643 struct inode *inode = file->f_mapping->host;
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500644 struct ext4_map_blocks map;
645 struct extent_status es;
646 ext4_lblk_t start, last, end;
647 loff_t holeoff, isize;
648 int blkbits;
649 int ret = 0;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500650
651 mutex_lock(&inode->i_mutex);
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500652
653 isize = i_size_read(inode);
654 if (offset >= isize) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500655 mutex_unlock(&inode->i_mutex);
656 return -ENXIO;
657 }
658
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500659 blkbits = inode->i_sb->s_blocksize_bits;
660 start = offset >> blkbits;
661 last = start;
662 end = isize >> blkbits;
663 holeoff = offset;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500664
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500665 do {
666 map.m_lblk = last;
667 map.m_len = end - last + 1;
668 ret = ext4_map_blocks(NULL, inode, &map, 0);
669 if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
670 last += ret;
671 holeoff = (loff_t)last << blkbits;
672 continue;
673 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500674
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500675 /*
676 * If there is a delay extent at this offset,
677 * we will skip this extent.
678 */
679 ext4_es_find_delayed_extent_range(inode, last, last, &es);
680 if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
681 last = es.es_lblk + es.es_len;
682 holeoff = (loff_t)last << blkbits;
683 continue;
684 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500685
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500686 /*
687 * If there is a unwritten extent at this offset,
688 * it will be as a data or a hole according to page
689 * cache that has data or not.
690 */
691 if (map.m_flags & EXT4_MAP_UNWRITTEN) {
692 int unwritten;
693 unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
694 &map, &holeoff);
695 if (!unwritten) {
696 last += ret;
697 holeoff = (loff_t)last << blkbits;
Zheng Liuc8c0df22012-11-08 21:57:40 -0500698 continue;
699 }
700 }
Zheng Liuc8c0df22012-11-08 21:57:40 -0500701
Theodore Ts'oad7fefb2015-01-02 15:16:00 -0500702 /* find a hole */
703 break;
704 } while (last <= end);
705
706 mutex_unlock(&inode->i_mutex);
707
708 if (holeoff > isize)
709 holeoff = isize;
710
711 return vfs_setpos(file, holeoff, maxsize);
Zheng Liuc8c0df22012-11-08 21:57:40 -0500712}
713
714/*
Eric Sandeenec7268c2012-04-30 13:14:03 -0500715 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
716 * by calling generic_file_llseek_size() with the appropriate maxbytes
717 * value for each.
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400718 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800719loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400720{
721 struct inode *inode = file->f_mapping->host;
722 loff_t maxbytes;
723
724 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
725 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
726 else
727 maxbytes = inode->i_sb->s_maxbytes;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400728
Andrew Morton965c8e52012-12-17 15:59:39 -0800729 switch (whence) {
Zheng Liuc8c0df22012-11-08 21:57:40 -0500730 case SEEK_SET:
731 case SEEK_CUR:
732 case SEEK_END:
Andrew Morton965c8e52012-12-17 15:59:39 -0800733 return generic_file_llseek_size(file, offset, whence,
Zheng Liuc8c0df22012-11-08 21:57:40 -0500734 maxbytes, i_size_read(inode));
735 case SEEK_DATA:
736 return ext4_seek_data(file, offset, maxbytes);
737 case SEEK_HOLE:
738 return ext4_seek_hole(file, offset, maxbytes);
739 }
740
741 return -EINVAL;
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400742}
743
Mingming Cao617ba132006-10-11 01:20:53 -0700744const struct file_operations ext4_file_operations = {
Toshiyuki Okajimae0d10bf2010-10-27 21:30:06 -0400745 .llseek = ext4_llseek,
Al Viroaad4f8b2014-04-02 14:33:16 -0400746 .read_iter = generic_file_read_iter,
Al Viro9b884162014-04-17 16:09:22 -0400747 .write_iter = ext4_file_write_iter,
Andi Kleen5cdd7b22008-04-29 22:03:54 -0400748 .unlocked_ioctl = ext4_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700749#ifdef CONFIG_COMPAT
Mingming Cao617ba132006-10-11 01:20:53 -0700750 .compat_ioctl = ext4_compat_ioctl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700751#endif
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -0400752 .mmap = ext4_file_mmap,
Theodore Ts'obc0b0d62009-06-13 10:09:48 -0400753 .open = ext4_file_open,
Mingming Cao617ba132006-10-11 01:20:53 -0700754 .release = ext4_release_file,
755 .fsync = ext4_sync_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700756 .splice_read = generic_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -0400757 .splice_write = iter_file_splice_write,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100758 .fallocate = ext4_fallocate,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700759};
760
Arjan van de Ven754661f2007-02-12 00:55:38 -0800761const struct inode_operations ext4_file_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -0700762 .setattr = ext4_setattr,
Mingming Cao3e3398a2008-07-11 19:27:31 -0400763 .getattr = ext4_getattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700764 .setxattr = generic_setxattr,
765 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -0700766 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700767 .removexattr = generic_removexattr,
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200768 .get_acl = ext4_get_acl,
Christoph Hellwig64e178a2013-12-20 05:16:44 -0800769 .set_acl = ext4_set_acl,
Eric Sandeen6873fa02008-10-07 00:46:36 -0400770 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700771};
772