]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/file.c |
ac27a0ec DK |
3 | * |
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 | * | |
617ba13b | 15 | * ext4 fs regular file handling primitives |
ac27a0ec DK |
16 | * |
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> | |
bc0b0d6d TT |
23 | #include <linux/mount.h> |
24 | #include <linux/path.h> | |
c94c2acf | 25 | #include <linux/dax.h> |
871a2931 | 26 | #include <linux/quotaops.h> |
c8c0df24 | 27 | #include <linux/pagevec.h> |
e2e40f2c | 28 | #include <linux/uio.h> |
3dcf5451 CH |
29 | #include "ext4.h" |
30 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
31 | #include "xattr.h" |
32 | #include "acl.h" | |
33 | ||
364443cb JK |
34 | #ifdef CONFIG_FS_DAX |
35 | static 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 | ||
728fbc0e GR |
40 | if (!inode_trylock_shared(inode)) { |
41 | if (iocb->ki_flags & IOCB_NOWAIT) | |
42 | return -EAGAIN; | |
43 | inode_lock_shared(inode); | |
44 | } | |
364443cb JK |
45 | /* |
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 | ||
62 | static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
63 | { | |
0db1ff22 TT |
64 | if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb)))) |
65 | return -EIO; | |
66 | ||
364443cb JK |
67 | 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 | ||
ac27a0ec DK |
77 | /* |
78 | * Called when an inode is released. Note that this is different | |
617ba13b | 79 | * from ext4_file_open: open gets called at every open, but release |
ac27a0ec DK |
80 | * gets called only when /all/ the files are closed. |
81 | */ | |
af5bc92d | 82 | static int ext4_release_file(struct inode *inode, struct file *filp) |
ac27a0ec | 83 | { |
19f5fb7a | 84 | if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) { |
7d8f9f7d | 85 | ext4_alloc_da_blocks(inode); |
19f5fb7a | 86 | ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 87 | } |
ac27a0ec DK |
88 | /* if we are the last writer on the inode, drop the block reservation */ |
89 | if ((filp->f_mode & FMODE_WRITE) && | |
d6014301 AK |
90 | (atomic_read(&inode->i_writecount) == 1) && |
91 | !EXT4_I(inode)->i_reserved_data_blocks) | |
ac27a0ec | 92 | { |
0e855ac8 | 93 | down_write(&EXT4_I(inode)->i_data_sem); |
c2ea3fde | 94 | ext4_discard_preallocations(inode); |
0e855ac8 | 95 | up_write(&EXT4_I(inode)->i_data_sem); |
ac27a0ec DK |
96 | } |
97 | if (is_dx(inode) && filp->private_data) | |
617ba13b | 98 | ext4_htree_free_dir_info(filp->private_data); |
ac27a0ec DK |
99 | |
100 | return 0; | |
101 | } | |
102 | ||
c197855e | 103 | static void ext4_unwritten_wait(struct inode *inode) |
e9e3bcec ES |
104 | { |
105 | wait_queue_head_t *wq = ext4_ioend_wq(inode); | |
106 | ||
e27f41e1 | 107 | wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0)); |
e9e3bcec ES |
108 | } |
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 | */ | |
119 | static int | |
9b884164 | 120 | ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos) |
e9e3bcec ES |
121 | { |
122 | struct super_block *sb = inode->i_sb; | |
123 | int blockmask = sb->s_blocksize - 1; | |
e9e3bcec | 124 | |
6e6358fc | 125 | if (pos >= i_size_read(inode)) |
e9e3bcec ES |
126 | return 0; |
127 | ||
9b884164 | 128 | if ((pos | iov_iter_alignment(from)) & blockmask) |
e9e3bcec ES |
129 | return 1; |
130 | ||
131 | return 0; | |
132 | } | |
133 | ||
213bcd9c JK |
134 | /* Is IO overwriting allocated and initialized blocks? */ |
135 | static 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 | ||
157 | static 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 | ||
776722e8 JK |
179 | #ifdef CONFIG_FS_DAX |
180 | static ssize_t | |
181 | ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
182 | { | |
183 | struct inode *inode = file_inode(iocb->ki_filp); | |
184 | ssize_t ret; | |
776722e8 | 185 | |
728fbc0e GR |
186 | if (!inode_trylock(inode)) { |
187 | if (iocb->ki_flags & IOCB_NOWAIT) | |
188 | return -EAGAIN; | |
189 | inode_lock(inode); | |
190 | } | |
776722e8 JK |
191 | 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 | ||
776722e8 JK |
201 | ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops); |
202 | out: | |
ff5462e3 | 203 | inode_unlock(inode); |
776722e8 JK |
204 | if (ret > 0) |
205 | ret = generic_write_sync(iocb, ret); | |
206 | return ret; | |
207 | } | |
208 | #endif | |
209 | ||
ac27a0ec | 210 | static ssize_t |
9b884164 | 211 | ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ac27a0ec | 212 | { |
8ad2850f | 213 | struct inode *inode = file_inode(iocb->ki_filp); |
2ba48ce5 | 214 | int o_direct = iocb->ki_flags & IOCB_DIRECT; |
e142d052 | 215 | int unaligned_aio = 0; |
4bd809db | 216 | int overwrite = 0; |
8563000d | 217 | ssize_t ret; |
7608e610 | 218 | |
0db1ff22 TT |
219 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
220 | return -EIO; | |
221 | ||
776722e8 JK |
222 | #ifdef CONFIG_FS_DAX |
223 | if (IS_DAX(inode)) | |
224 | return ext4_dax_write_iter(iocb, from); | |
225 | #endif | |
226 | ||
728fbc0e GR |
227 | if (!inode_trylock(inode)) { |
228 | if (iocb->ki_flags & IOCB_NOWAIT) | |
229 | return -EAGAIN; | |
230 | inode_lock(inode); | |
231 | } | |
232 | ||
213bcd9c | 233 | ret = ext4_write_checks(iocb, from); |
e142d052 JK |
234 | if (ret <= 0) |
235 | goto out; | |
236 | ||
f5ccfe1d | 237 | /* |
e142d052 JK |
238 | * 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. | |
f5ccfe1d | 241 | */ |
e142d052 | 242 | if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) && |
f5ccfe1d | 243 | !is_sync_kiocb(iocb) && |
e142d052 JK |
244 | ext4_unaligned_aio(inode, from, iocb->ki_pos)) { |
245 | unaligned_aio = 1; | |
f5ccfe1d TT |
246 | ext4_unwritten_wait(inode); |
247 | } | |
248 | ||
a41537e6 | 249 | iocb->private = &overwrite; |
213bcd9c | 250 | /* Check whether we do a DIO overwrite or not */ |
728fbc0e GR |
251 | 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 | } | |
7608e610 | 260 | |
9b884164 | 261 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 262 | inode_unlock(inode); |
7608e610 | 263 | |
e2592217 CH |
264 | if (ret > 0) |
265 | ret = generic_write_sync(iocb, ret); | |
e9e3bcec | 266 | |
e768d7ff AV |
267 | return ret; |
268 | ||
269 | out: | |
5955102c | 270 | inode_unlock(inode); |
e9e3bcec | 271 | return ret; |
ac27a0ec DK |
272 | } |
273 | ||
923ae0ff | 274 | #ifdef CONFIG_FS_DAX |
c791ace1 DJ |
275 | static int ext4_dax_huge_fault(struct vm_fault *vmf, |
276 | enum page_entry_size pe_size) | |
923ae0ff | 277 | { |
01a33b4a | 278 | int result; |
fb26a1cb | 279 | handle_t *handle = NULL; |
11bac800 | 280 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ea3d7209 | 281 | struct super_block *sb = inode->i_sb; |
01a33b4a MW |
282 | bool write = vmf->flags & FAULT_FLAG_WRITE; |
283 | ||
284 | if (write) { | |
285 | sb_start_pagefault(sb); | |
11bac800 | 286 | file_update_time(vmf->vma->vm_file); |
fb26a1cb JK |
287 | 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); | |
1db17542 | 292 | } |
fb26a1cb JK |
293 | 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); | |
01a33b4a | 301 | sb_end_pagefault(sb); |
fb26a1cb JK |
302 | } else { |
303 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
304 | } | |
01a33b4a MW |
305 | |
306 | return result; | |
923ae0ff RZ |
307 | } |
308 | ||
c791ace1 DJ |
309 | static int ext4_dax_fault(struct vm_fault *vmf) |
310 | { | |
311 | return ext4_dax_huge_fault(vmf, PE_SIZE_PTE); | |
312 | } | |
313 | ||
ea3d7209 | 314 | /* |
1e9d180b | 315 | * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault() |
ea3d7209 JK |
316 | * 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 | */ | |
11bac800 | 323 | static int ext4_dax_pfn_mkwrite(struct vm_fault *vmf) |
ea3d7209 | 324 | { |
11bac800 | 325 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ea3d7209 | 326 | struct super_block *sb = inode->i_sb; |
ea3d7209 | 327 | loff_t size; |
d5be7a03 | 328 | int ret; |
ea3d7209 JK |
329 | |
330 | sb_start_pagefault(sb); | |
11bac800 | 331 | file_update_time(vmf->vma->vm_file); |
ea3d7209 JK |
332 | 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; | |
d5be7a03 | 336 | else |
11bac800 | 337 | ret = dax_pfn_mkwrite(vmf); |
ea3d7209 JK |
338 | up_read(&EXT4_I(inode)->i_mmap_sem); |
339 | sb_end_pagefault(sb); | |
340 | ||
341 | return ret; | |
923ae0ff RZ |
342 | } |
343 | ||
344 | static const struct vm_operations_struct ext4_dax_vm_ops = { | |
345 | .fault = ext4_dax_fault, | |
c791ace1 | 346 | .huge_fault = ext4_dax_huge_fault, |
1e9d180b | 347 | .page_mkwrite = ext4_dax_fault, |
ea3d7209 | 348 | .pfn_mkwrite = ext4_dax_pfn_mkwrite, |
923ae0ff RZ |
349 | }; |
350 | #else | |
351 | #define ext4_dax_vm_ops ext4_file_vm_ops | |
352 | #endif | |
353 | ||
f0f37e2f | 354 | static const struct vm_operations_struct ext4_file_vm_ops = { |
ea3d7209 | 355 | .fault = ext4_filemap_fault, |
f1820361 | 356 | .map_pages = filemap_map_pages, |
2e9ee850 AK |
357 | .page_mkwrite = ext4_page_mkwrite, |
358 | }; | |
359 | ||
360 | static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma) | |
361 | { | |
c9c7429c MH |
362 | struct inode *inode = file->f_mapping->host; |
363 | ||
0db1ff22 TT |
364 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
365 | return -EIO; | |
366 | ||
2e9ee850 | 367 | file_accessed(file); |
923ae0ff RZ |
368 | if (IS_DAX(file_inode(file))) { |
369 | vma->vm_ops = &ext4_dax_vm_ops; | |
11bd1a9e | 370 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
923ae0ff RZ |
371 | } else { |
372 | vma->vm_ops = &ext4_file_vm_ops; | |
373 | } | |
2e9ee850 AK |
374 | return 0; |
375 | } | |
376 | ||
bc0b0d6d TT |
377 | static 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; | |
9dd78d8c | 382 | struct dentry *dir; |
bc0b0d6d TT |
383 | struct path path; |
384 | char buf[64], *cp; | |
c9c7429c | 385 | int ret; |
bc0b0d6d | 386 | |
0db1ff22 TT |
387 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
388 | return -EIO; | |
389 | ||
bc0b0d6d | 390 | if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) && |
bc98a42c | 391 | !sb_rdonly(sb))) { |
bc0b0d6d TT |
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)); | |
3899167d AV |
400 | path.mnt = mnt; |
401 | path.dentry = mnt->mnt_root; | |
bc0b0d6d | 402 | cp = d_path(&path, buf, sizeof(buf)); |
bc0b0d6d | 403 | if (!IS_ERR(cp)) { |
044ce47f JK |
404 | handle_t *handle; |
405 | int err; | |
406 | ||
9924a92a | 407 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
044ce47f JK |
408 | if (IS_ERR(handle)) |
409 | return PTR_ERR(handle); | |
5d601255 | 410 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
044ce47f JK |
411 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
412 | if (err) { | |
413 | ext4_journal_stop(handle); | |
414 | return err; | |
415 | } | |
cf803903 DW |
416 | strlcpy(sbi->s_es->s_last_mounted, cp, |
417 | sizeof(sbi->s_es->s_last_mounted)); | |
044ce47f JK |
418 | ext4_handle_dirty_super(handle, sb); |
419 | ext4_journal_stop(handle); | |
bc0b0d6d TT |
420 | } |
421 | } | |
abdd438b | 422 | if (ext4_encrypted_inode(inode)) { |
a7550b30 | 423 | ret = fscrypt_get_encryption_info(inode); |
abdd438b TT |
424 | if (ret) |
425 | return -EACCES; | |
a7550b30 | 426 | if (!fscrypt_has_encryption_key(inode)) |
abdd438b TT |
427 | return -ENOKEY; |
428 | } | |
9dd78d8c | 429 | |
c0a37d48 | 430 | dir = dget_parent(file_dentry(filp)); |
9dd78d8c | 431 | if (ext4_encrypted_inode(d_inode(dir)) && |
a7550b30 | 432 | !fscrypt_has_permitted_context(d_inode(dir), inode)) { |
ff978b09 | 433 | ext4_warning(inode->i_sb, |
8d2ae1cb | 434 | "Inconsistent encryption contexts: %lu/%lu", |
9dd78d8c | 435 | (unsigned long) d_inode(dir)->i_ino, |
ff978b09 | 436 | (unsigned long) inode->i_ino); |
9dd78d8c | 437 | dput(dir); |
ff978b09 TT |
438 | return -EPERM; |
439 | } | |
9dd78d8c | 440 | dput(dir); |
8aefcd55 TT |
441 | /* |
442 | * Set up the jbd2_inode if we are opening the inode for | |
443 | * writing and the journal is present | |
444 | */ | |
a361293f | 445 | if (filp->f_mode & FMODE_WRITE) { |
c9c7429c | 446 | ret = ext4_inode_attach_jinode(inode); |
a361293f JK |
447 | if (ret < 0) |
448 | return ret; | |
8aefcd55 | 449 | } |
728fbc0e GR |
450 | |
451 | /* Set the flags to support nowait AIO */ | |
452 | filp->f_mode |= FMODE_AIO_NOWAIT; | |
453 | ||
abdd438b | 454 | return dquot_file_open(inode, filp); |
bc0b0d6d TT |
455 | } |
456 | ||
c8c0df24 ZL |
457 | /* |
458 | * 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 | */ | |
ad7fefb1 TT |
473 | static int ext4_find_unwritten_pgoff(struct inode *inode, |
474 | int whence, | |
2d90c160 | 475 | ext4_lblk_t end_blk, |
ad7fefb1 | 476 | loff_t *offset) |
c8c0df24 ZL |
477 | { |
478 | struct pagevec pvec; | |
ad7fefb1 | 479 | unsigned int blkbits; |
c8c0df24 ZL |
480 | pgoff_t index; |
481 | pgoff_t end; | |
ad7fefb1 | 482 | loff_t endoff; |
c8c0df24 ZL |
483 | loff_t startoff; |
484 | loff_t lastoff; | |
485 | int found = 0; | |
486 | ||
ad7fefb1 | 487 | blkbits = inode->i_sb->s_blocksize_bits; |
c8c0df24 ZL |
488 | startoff = *offset; |
489 | lastoff = startoff; | |
2d90c160 | 490 | endoff = (loff_t)end_blk << blkbits; |
c8c0df24 | 491 | |
09cbfeaf | 492 | index = startoff >> PAGE_SHIFT; |
3f1d5bad | 493 | end = (endoff - 1) >> PAGE_SHIFT; |
c8c0df24 ZL |
494 | |
495 | pagevec_init(&pvec, 0); | |
496 | do { | |
497 | int i, num; | |
498 | unsigned long nr_pages; | |
499 | ||
624327f8 | 500 | num = min_t(pgoff_t, end - index, PAGEVEC_SIZE - 1) + 1; |
c8c0df24 ZL |
501 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, |
502 | (pgoff_t)num); | |
7d95eddf | 503 | if (nr_pages == 0) |
c8c0df24 | 504 | break; |
c8c0df24 ZL |
505 | |
506 | for (i = 0; i < nr_pages; i++) { | |
507 | struct page *page = pvec.pages[i]; | |
508 | struct buffer_head *bh, *head; | |
509 | ||
510 | /* | |
7d95eddf JK |
511 | * If current offset is smaller than the page offset, |
512 | * there is a hole at this offset. | |
c8c0df24 | 513 | */ |
7d95eddf JK |
514 | if (whence == SEEK_HOLE && lastoff < endoff && |
515 | lastoff < page_offset(pvec.pages[i])) { | |
c8c0df24 ZL |
516 | found = 1; |
517 | *offset = lastoff; | |
518 | goto out; | |
519 | } | |
520 | ||
7d95eddf JK |
521 | if (page->index > end) |
522 | goto out; | |
523 | ||
c8c0df24 ZL |
524 | 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)) { | |
965c8e59 | 542 | if (whence == SEEK_DATA) |
c8c0df24 ZL |
543 | found = 1; |
544 | } else { | |
965c8e59 | 545 | if (whence == SEEK_HOLE) |
c8c0df24 ZL |
546 | 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 | ||
7d95eddf JK |
563 | /* The no. of pages is less than our desired, we are done. */ |
564 | if (nr_pages < num) | |
c8c0df24 | 565 | break; |
c8c0df24 ZL |
566 | |
567 | index = pvec.pages[i - 1]->index + 1; | |
568 | pagevec_release(&pvec); | |
569 | } while (index <= end); | |
570 | ||
7d95eddf JK |
571 | if (whence == SEEK_HOLE && lastoff < endoff) { |
572 | found = 1; | |
573 | *offset = lastoff; | |
574 | } | |
c8c0df24 ZL |
575 | out: |
576 | pagevec_release(&pvec); | |
577 | return found; | |
578 | } | |
579 | ||
580 | /* | |
581 | * ext4_seek_data() retrieves the offset for SEEK_DATA. | |
582 | */ | |
583 | static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize) | |
584 | { | |
585 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
586 | struct extent_status es; |
587 | ext4_lblk_t start, last, end; | |
588 | loff_t dataoff, isize; | |
589 | int blkbits; | |
2d90c160 | 590 | int ret; |
c8c0df24 | 591 | |
5955102c | 592 | inode_lock(inode); |
ad7fefb1 TT |
593 | |
594 | isize = i_size_read(inode); | |
595 | if (offset >= isize) { | |
5955102c | 596 | inode_unlock(inode); |
c8c0df24 ZL |
597 | return -ENXIO; |
598 | } | |
ad7fefb1 TT |
599 | |
600 | blkbits = inode->i_sb->s_blocksize_bits; | |
601 | start = offset >> blkbits; | |
602 | last = start; | |
603 | end = isize >> blkbits; | |
604 | dataoff = offset; | |
605 | ||
606 | do { | |
2d90c160 JK |
607 | 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; | |
ad7fefb1 | 614 | } |
c8c0df24 | 615 | |
2d90c160 JK |
616 | last = es.es_lblk; |
617 | if (last != start) | |
618 | dataoff = (loff_t)last << blkbits; | |
619 | if (!ext4_es_is_unwritten(&es)) | |
c8c0df24 | 620 | break; |
c8c0df24 | 621 | |
ad7fefb1 TT |
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 | */ | |
2d90c160 JK |
627 | if (ext4_find_unwritten_pgoff(inode, SEEK_DATA, |
628 | es.es_lblk + es.es_len, &dataoff)) | |
629 | break; | |
630 | last += es.es_len; | |
ad7fefb1 | 631 | dataoff = (loff_t)last << blkbits; |
2d90c160 | 632 | cond_resched(); |
ad7fefb1 | 633 | } while (last <= end); |
c8c0df24 | 634 | |
5955102c | 635 | inode_unlock(inode); |
c8c0df24 | 636 | |
ad7fefb1 TT |
637 | if (dataoff > isize) |
638 | return -ENXIO; | |
639 | ||
640 | return vfs_setpos(file, dataoff, maxsize); | |
c8c0df24 ZL |
641 | } |
642 | ||
643 | /* | |
ad7fefb1 | 644 | * ext4_seek_hole() retrieves the offset for SEEK_HOLE. |
c8c0df24 ZL |
645 | */ |
646 | static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize) | |
647 | { | |
648 | struct inode *inode = file->f_mapping->host; | |
ad7fefb1 TT |
649 | struct extent_status es; |
650 | ext4_lblk_t start, last, end; | |
651 | loff_t holeoff, isize; | |
652 | int blkbits; | |
2d90c160 | 653 | int ret; |
c8c0df24 | 654 | |
5955102c | 655 | inode_lock(inode); |
ad7fefb1 TT |
656 | |
657 | isize = i_size_read(inode); | |
658 | if (offset >= isize) { | |
5955102c | 659 | inode_unlock(inode); |
c8c0df24 ZL |
660 | return -ENXIO; |
661 | } | |
662 | ||
ad7fefb1 TT |
663 | blkbits = inode->i_sb->s_blocksize_bits; |
664 | start = offset >> blkbits; | |
665 | last = start; | |
666 | end = isize >> blkbits; | |
667 | holeoff = offset; | |
c8c0df24 | 668 | |
ad7fefb1 | 669 | do { |
2d90c160 JK |
670 | ret = ext4_get_next_extent(inode, last, end - last + 1, &es); |
671 | if (ret < 0) { | |
672 | inode_unlock(inode); | |
673 | return ret; | |
ad7fefb1 | 674 | } |
2d90c160 JK |
675 | /* Found a hole? */ |
676 | if (ret == 0 || es.es_lblk > last) { | |
677 | if (last != start) | |
678 | holeoff = (loff_t)last << blkbits; | |
679 | break; | |
ad7fefb1 | 680 | } |
ad7fefb1 TT |
681 | /* |
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 | */ | |
2d90c160 JK |
686 | if (ext4_es_is_unwritten(&es) && |
687 | ext4_find_unwritten_pgoff(inode, SEEK_HOLE, | |
688 | last + es.es_len, &holeoff)) | |
689 | break; | |
ad7fefb1 | 690 | |
2d90c160 JK |
691 | last += es.es_len; |
692 | holeoff = (loff_t)last << blkbits; | |
693 | cond_resched(); | |
ad7fefb1 TT |
694 | } while (last <= end); |
695 | ||
5955102c | 696 | inode_unlock(inode); |
c8c0df24 | 697 | |
ad7fefb1 TT |
698 | if (holeoff > isize) |
699 | holeoff = isize; | |
700 | ||
701 | return vfs_setpos(file, holeoff, maxsize); | |
c8c0df24 ZL |
702 | } |
703 | ||
e0d10bfa | 704 | /* |
ec7268ce ES |
705 | * 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. | |
e0d10bfa | 708 | */ |
965c8e59 | 709 | loff_t ext4_llseek(struct file *file, loff_t offset, int whence) |
e0d10bfa TO |
710 | { |
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; | |
e0d10bfa | 718 | |
965c8e59 | 719 | switch (whence) { |
c8c0df24 ZL |
720 | case SEEK_SET: |
721 | case SEEK_CUR: | |
722 | case SEEK_END: | |
965c8e59 | 723 | return generic_file_llseek_size(file, offset, whence, |
c8c0df24 ZL |
724 | 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; | |
e0d10bfa TO |
732 | } |
733 | ||
617ba13b | 734 | const struct file_operations ext4_file_operations = { |
e0d10bfa | 735 | .llseek = ext4_llseek, |
364443cb | 736 | .read_iter = ext4_file_read_iter, |
9b884164 | 737 | .write_iter = ext4_file_write_iter, |
5cdd7b2d | 738 | .unlocked_ioctl = ext4_ioctl, |
ac27a0ec | 739 | #ifdef CONFIG_COMPAT |
617ba13b | 740 | .compat_ioctl = ext4_compat_ioctl, |
ac27a0ec | 741 | #endif |
2e9ee850 | 742 | .mmap = ext4_file_mmap, |
bc0b0d6d | 743 | .open = ext4_file_open, |
617ba13b MC |
744 | .release = ext4_release_file, |
745 | .fsync = ext4_sync_file, | |
dbe6ec81 | 746 | .get_unmapped_area = thp_get_unmapped_area, |
ac27a0ec | 747 | .splice_read = generic_file_splice_read, |
8d020765 | 748 | .splice_write = iter_file_splice_write, |
2fe17c10 | 749 | .fallocate = ext4_fallocate, |
ac27a0ec DK |
750 | }; |
751 | ||
754661f1 | 752 | const struct inode_operations ext4_file_inode_operations = { |
617ba13b | 753 | .setattr = ext4_setattr, |
99652ea5 | 754 | .getattr = ext4_file_getattr, |
617ba13b | 755 | .listxattr = ext4_listxattr, |
4e34e719 | 756 | .get_acl = ext4_get_acl, |
64e178a7 | 757 | .set_acl = ext4_set_acl, |
6873fa0d | 758 | .fiemap = ext4_fiemap, |
ac27a0ec DK |
759 | }; |
760 |