1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/read_write.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/stat.h>
10 #include <linux/sched/xacct.h>
11 #include <linux/fcntl.h>
12 #include <linux/file.h>
13 #include <linux/uio.h>
14 #include <linux/fsnotify.h>
15 #include <linux/security.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/pagemap.h>
19 #include <linux/splice.h>
20 #include <linux/compat.h>
21 #include <linux/mount.h>
25 #include <linux/uaccess.h>
26 #include <asm/unistd.h>
28 const struct file_operations generic_ro_fops = {
29 .llseek = generic_file_llseek,
30 .read_iter = generic_file_read_iter,
31 .mmap = generic_file_readonly_mmap,
32 .splice_read = generic_file_splice_read,
35 EXPORT_SYMBOL(generic_ro_fops);
37 static inline bool unsigned_offsets(struct file *file)
39 return file->f_mode & FMODE_UNSIGNED_OFFSET;
43 * vfs_setpos - update the file offset for lseek
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
48 * This is a low-level filesystem helper for updating the file offset to
49 * the value specified by @offset if the given offset is valid and it is
50 * not equal to the current file offset.
52 * Return the specified offset on success and -EINVAL on invalid offset.
54 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
56 if (offset < 0 && !unsigned_offsets(file))
61 if (offset != file->f_pos) {
67 EXPORT_SYMBOL(vfs_setpos);
70 * generic_file_llseek_size - generic llseek implementation for regular files
71 * @file: file structure to seek on
72 * @offset: file offset to seek to
73 * @whence: type of seek
74 * @size: max size of this file in file system
75 * @eof: offset used for SEEK_END position
77 * This is a variant of generic_file_llseek that allows passing in a custom
78 * maximum file size and a custom EOF position, for e.g. hashed directories
81 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
82 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83 * read/writes behave like SEEK_SET against seeks.
86 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
87 loff_t maxsize, loff_t eof)
95 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 * position-querying operation. Avoid rewriting the "same"
97 * f_pos value back to the file because a concurrent read(),
98 * write() or lseek() might have altered it
103 * f_lock protects against read/modify/write race with other
104 * SEEK_CURs. Note that parallel writes and reads behave
107 spin_lock(&file->f_lock);
108 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
109 spin_unlock(&file->f_lock);
113 * In the generic case the entire file is data, so as long as
114 * offset isn't at the end of the file then the offset is data.
116 if ((unsigned long long)offset >= eof)
121 * There is a virtual hole at the end of the file, so as long as
122 * offset isn't i_size or larger, return i_size.
124 if ((unsigned long long)offset >= eof)
130 return vfs_setpos(file, offset, maxsize);
132 EXPORT_SYMBOL(generic_file_llseek_size);
135 * generic_file_llseek - generic llseek implementation for regular files
136 * @file: file structure to seek on
137 * @offset: file offset to seek to
138 * @whence: type of seek
140 * This is a generic implemenation of ->llseek useable for all normal local
141 * filesystems. It just updates the file offset to the value specified by
142 * @offset and @whence.
144 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
146 struct inode *inode = file->f_mapping->host;
148 return generic_file_llseek_size(file, offset, whence,
149 inode->i_sb->s_maxbytes,
152 EXPORT_SYMBOL(generic_file_llseek);
155 * fixed_size_llseek - llseek implementation for fixed-sized devices
156 * @file: file structure to seek on
157 * @offset: file offset to seek to
158 * @whence: type of seek
159 * @size: size of the file
162 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
165 case SEEK_SET: case SEEK_CUR: case SEEK_END:
166 return generic_file_llseek_size(file, offset, whence,
172 EXPORT_SYMBOL(fixed_size_llseek);
175 * no_seek_end_llseek - llseek implementation for fixed-sized devices
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
181 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
184 case SEEK_SET: case SEEK_CUR:
185 return generic_file_llseek_size(file, offset, whence,
191 EXPORT_SYMBOL(no_seek_end_llseek);
194 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195 * @file: file structure to seek on
196 * @offset: file offset to seek to
197 * @whence: type of seek
198 * @size: maximal offset allowed
201 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
204 case SEEK_SET: case SEEK_CUR:
205 return generic_file_llseek_size(file, offset, whence,
211 EXPORT_SYMBOL(no_seek_end_llseek_size);
214 * noop_llseek - No Operation Performed llseek implementation
215 * @file: file structure to seek on
216 * @offset: file offset to seek to
217 * @whence: type of seek
219 * This is an implementation of ->llseek useable for the rare special case when
220 * userspace expects the seek to succeed but the (device) file is actually not
221 * able to perform the seek. In this case you use noop_llseek() instead of
222 * falling back to the default implementation of ->llseek.
224 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
228 EXPORT_SYMBOL(noop_llseek);
230 loff_t no_llseek(struct file *file, loff_t offset, int whence)
234 EXPORT_SYMBOL(no_llseek);
236 loff_t default_llseek(struct file *file, loff_t offset, int whence)
238 struct inode *inode = file_inode(file);
244 offset += i_size_read(inode);
248 retval = file->f_pos;
251 offset += file->f_pos;
255 * In the generic case the entire file is data, so as
256 * long as offset isn't at the end of the file then the
259 if (offset >= inode->i_size) {
266 * There is a virtual hole at the end of the file, so
267 * as long as offset isn't i_size or larger, return
270 if (offset >= inode->i_size) {
274 offset = inode->i_size;
278 if (offset >= 0 || unsigned_offsets(file)) {
279 if (offset != file->f_pos) {
280 file->f_pos = offset;
289 EXPORT_SYMBOL(default_llseek);
291 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
293 loff_t (*fn)(struct file *, loff_t, int);
296 if (file->f_mode & FMODE_LSEEK) {
297 if (file->f_op->llseek)
298 fn = file->f_op->llseek;
300 return fn(file, offset, whence);
302 EXPORT_SYMBOL(vfs_llseek);
304 off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
307 struct fd f = fdget_pos(fd);
312 if (whence <= SEEK_MAX) {
313 loff_t res = vfs_llseek(f.file, offset, whence);
315 if (res != (loff_t)retval)
316 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
324 return ksys_lseek(fd, offset, whence);
328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
330 return ksys_lseek(fd, offset, whence);
334 #ifdef __ARCH_WANT_SYS_LLSEEK
335 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
336 unsigned long, offset_low, loff_t __user *, result,
337 unsigned int, whence)
340 struct fd f = fdget_pos(fd);
347 if (whence > SEEK_MAX)
350 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
353 retval = (int)offset;
356 if (!copy_to_user(result, &offset, sizeof(offset)))
365 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
369 int retval = -EINVAL;
371 inode = file_inode(file);
372 if (unlikely((ssize_t) count < 0))
375 if (unlikely(pos < 0)) {
376 if (!unsigned_offsets(file))
378 if (count >= -pos) /* both values are in 0..LLONG_MAX */
380 } else if (unlikely((loff_t) (pos + count) < 0)) {
381 if (!unsigned_offsets(file))
385 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
386 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
387 read_write == READ ? F_RDLCK : F_WRLCK);
391 return security_file_permission(file,
392 read_write == READ ? MAY_READ : MAY_WRITE);
395 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
397 struct iovec iov = { .iov_base = buf, .iov_len = len };
399 struct iov_iter iter;
402 init_sync_kiocb(&kiocb, filp);
403 kiocb.ki_pos = *ppos;
404 iov_iter_init(&iter, READ, &iov, 1, len);
406 ret = call_read_iter(filp, &kiocb, &iter);
407 BUG_ON(ret == -EIOCBQUEUED);
408 *ppos = kiocb.ki_pos;
412 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
415 if (file->f_op->read)
416 return file->f_op->read(file, buf, count, pos);
417 else if (file->f_op->read_iter)
418 return new_sync_read(file, buf, count, pos);
423 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
430 /* The cast to a user pointer is valid due to the set_fs() */
431 result = vfs_read(file, (void __user *)buf, count, pos);
435 EXPORT_SYMBOL(kernel_read);
437 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
441 if (!(file->f_mode & FMODE_READ))
443 if (!(file->f_mode & FMODE_CAN_READ))
445 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
448 ret = rw_verify_area(READ, file, pos, count);
450 if (count > MAX_RW_COUNT)
451 count = MAX_RW_COUNT;
452 ret = __vfs_read(file, buf, count, pos);
454 fsnotify_access(file);
455 add_rchar(current, ret);
463 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
465 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
467 struct iov_iter iter;
470 init_sync_kiocb(&kiocb, filp);
471 kiocb.ki_pos = *ppos;
472 iov_iter_init(&iter, WRITE, &iov, 1, len);
474 ret = call_write_iter(filp, &kiocb, &iter);
475 BUG_ON(ret == -EIOCBQUEUED);
477 *ppos = kiocb.ki_pos;
481 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
484 if (file->f_op->write)
485 return file->f_op->write(file, p, count, pos);
486 else if (file->f_op->write_iter)
487 return new_sync_write(file, p, count, pos);
492 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
495 const char __user *p;
498 if (!(file->f_mode & FMODE_CAN_WRITE))
503 p = (__force const char __user *)buf;
504 if (count > MAX_RW_COUNT)
505 count = MAX_RW_COUNT;
506 ret = __vfs_write(file, p, count, pos);
509 fsnotify_modify(file);
510 add_wchar(current, ret);
515 EXPORT_SYMBOL(__kernel_write);
517 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
525 /* The cast to a user pointer is valid due to the set_fs() */
526 res = vfs_write(file, (__force const char __user *)buf, count, pos);
531 EXPORT_SYMBOL(kernel_write);
533 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
537 if (!(file->f_mode & FMODE_WRITE))
539 if (!(file->f_mode & FMODE_CAN_WRITE))
541 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
544 ret = rw_verify_area(WRITE, file, pos, count);
546 if (count > MAX_RW_COUNT)
547 count = MAX_RW_COUNT;
548 file_start_write(file);
549 ret = __vfs_write(file, buf, count, pos);
551 fsnotify_modify(file);
552 add_wchar(current, ret);
555 file_end_write(file);
561 static inline loff_t file_pos_read(struct file *file)
566 static inline void file_pos_write(struct file *file, loff_t pos)
571 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
573 struct fd f = fdget_pos(fd);
574 ssize_t ret = -EBADF;
577 loff_t pos = file_pos_read(f.file);
578 ret = vfs_read(f.file, buf, count, &pos);
580 file_pos_write(f.file, pos);
586 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
588 return ksys_read(fd, buf, count);
591 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
593 struct fd f = fdget_pos(fd);
594 ssize_t ret = -EBADF;
597 loff_t pos = file_pos_read(f.file);
598 ret = vfs_write(f.file, buf, count, &pos);
600 file_pos_write(f.file, pos);
607 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
610 return ksys_write(fd, buf, count);
613 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
614 size_t, count, loff_t, pos)
617 ssize_t ret = -EBADF;
625 if (f.file->f_mode & FMODE_PREAD)
626 ret = vfs_read(f.file, buf, count, &pos);
633 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
634 size_t, count, loff_t, pos)
637 ssize_t ret = -EBADF;
645 if (f.file->f_mode & FMODE_PWRITE)
646 ret = vfs_write(f.file, buf, count, &pos);
653 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
654 loff_t *ppos, int type, rwf_t flags)
659 init_sync_kiocb(&kiocb, filp);
660 ret = kiocb_set_rw_flags(&kiocb, flags);
663 kiocb.ki_pos = *ppos;
666 ret = call_read_iter(filp, &kiocb, iter);
668 ret = call_write_iter(filp, &kiocb, iter);
669 BUG_ON(ret == -EIOCBQUEUED);
670 *ppos = kiocb.ki_pos;
674 /* Do it by hand, with file-ops */
675 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
676 loff_t *ppos, int type, rwf_t flags)
680 if (flags & ~RWF_HIPRI)
683 while (iov_iter_count(iter)) {
684 struct iovec iovec = iov_iter_iovec(iter);
688 nr = filp->f_op->read(filp, iovec.iov_base,
689 iovec.iov_len, ppos);
691 nr = filp->f_op->write(filp, iovec.iov_base,
692 iovec.iov_len, ppos);
701 if (nr != iovec.iov_len)
703 iov_iter_advance(iter, nr);
709 /* A write operation does a read from user space and vice versa */
710 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
713 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
714 * into the kernel and check that it is valid.
716 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
717 * @uvector: Pointer to the userspace array.
718 * @nr_segs: Number of elements in userspace array.
719 * @fast_segs: Number of elements in @fast_pointer.
720 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
721 * @ret_pointer: (output parameter) Pointer to a variable that will point to
722 * either @fast_pointer, a newly allocated kernel array, or NULL,
723 * depending on which array was used.
725 * This function copies an array of &struct iovec of @nr_segs from
726 * userspace into the kernel and checks that each element is valid (e.g.
727 * it does not point to a kernel address or cause overflow by being too
730 * As an optimization, the caller may provide a pointer to a small
731 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
732 * (the size of this array, or 0 if unused, should be given in @fast_segs).
734 * @ret_pointer will always point to the array that was used, so the
735 * caller must take care not to call kfree() on it e.g. in case the
736 * @fast_pointer array was used and it was allocated on the stack.
738 * Return: The total number of bytes covered by the iovec array on success
739 * or a negative error code on error.
741 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
742 unsigned long nr_segs, unsigned long fast_segs,
743 struct iovec *fast_pointer,
744 struct iovec **ret_pointer)
748 struct iovec *iov = fast_pointer;
751 * SuS says "The readv() function *may* fail if the iovcnt argument
752 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
753 * traditionally returned zero for zero segments, so...
761 * First get the "struct iovec" from user memory and
762 * verify all the pointers
764 if (nr_segs > UIO_MAXIOV) {
768 if (nr_segs > fast_segs) {
769 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
775 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
781 * According to the Single Unix Specification we should return EINVAL
782 * if an element length is < 0 when cast to ssize_t or if the
783 * total length would overflow the ssize_t return value of the
786 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
790 for (seg = 0; seg < nr_segs; seg++) {
791 void __user *buf = iov[seg].iov_base;
792 ssize_t len = (ssize_t)iov[seg].iov_len;
794 /* see if we we're about to use an invalid len or if
795 * it's about to overflow ssize_t */
801 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
805 if (len > MAX_RW_COUNT - ret) {
806 len = MAX_RW_COUNT - ret;
807 iov[seg].iov_len = len;
817 ssize_t compat_rw_copy_check_uvector(int type,
818 const struct compat_iovec __user *uvector, unsigned long nr_segs,
819 unsigned long fast_segs, struct iovec *fast_pointer,
820 struct iovec **ret_pointer)
822 compat_ssize_t tot_len;
823 struct iovec *iov = *ret_pointer = fast_pointer;
828 * SuS says "The readv() function *may* fail if the iovcnt argument
829 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
830 * traditionally returned zero for zero segments, so...
836 if (nr_segs > UIO_MAXIOV)
838 if (nr_segs > fast_segs) {
840 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
847 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
851 * Single unix specification:
852 * We should -EINVAL if an element length is not >= 0 and fitting an
855 * In Linux, the total length is limited to MAX_RW_COUNT, there is
856 * no overflow possibility.
860 for (seg = 0; seg < nr_segs; seg++) {
864 if (__get_user(len, &uvector->iov_len) ||
865 __get_user(buf, &uvector->iov_base)) {
869 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
872 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
876 if (len > MAX_RW_COUNT - tot_len)
877 len = MAX_RW_COUNT - tot_len;
879 iov->iov_base = compat_ptr(buf);
880 iov->iov_len = (compat_size_t) len;
891 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
892 loff_t *pos, rwf_t flags)
897 if (!(file->f_mode & FMODE_READ))
899 if (!(file->f_mode & FMODE_CAN_READ))
902 tot_len = iov_iter_count(iter);
905 ret = rw_verify_area(READ, file, pos, tot_len);
909 if (file->f_op->read_iter)
910 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
912 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
915 fsnotify_access(file);
919 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
922 if (!file->f_op->read_iter)
924 return do_iter_read(file, iter, ppos, flags);
926 EXPORT_SYMBOL(vfs_iter_read);
928 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
929 loff_t *pos, rwf_t flags)
934 if (!(file->f_mode & FMODE_WRITE))
936 if (!(file->f_mode & FMODE_CAN_WRITE))
939 tot_len = iov_iter_count(iter);
942 ret = rw_verify_area(WRITE, file, pos, tot_len);
946 if (file->f_op->write_iter)
947 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
949 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
951 fsnotify_modify(file);
955 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
958 if (!file->f_op->write_iter)
960 return do_iter_write(file, iter, ppos, flags);
962 EXPORT_SYMBOL(vfs_iter_write);
964 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
965 unsigned long vlen, loff_t *pos, rwf_t flags)
967 struct iovec iovstack[UIO_FASTIOV];
968 struct iovec *iov = iovstack;
969 struct iov_iter iter;
972 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
974 ret = do_iter_read(file, &iter, pos, flags);
981 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
982 unsigned long vlen, loff_t *pos, rwf_t flags)
984 struct iovec iovstack[UIO_FASTIOV];
985 struct iovec *iov = iovstack;
986 struct iov_iter iter;
989 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
991 file_start_write(file);
992 ret = do_iter_write(file, &iter, pos, flags);
993 file_end_write(file);
999 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1000 unsigned long vlen, rwf_t flags)
1002 struct fd f = fdget_pos(fd);
1003 ssize_t ret = -EBADF;
1006 loff_t pos = file_pos_read(f.file);
1007 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1009 file_pos_write(f.file, pos);
1014 add_rchar(current, ret);
1019 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1020 unsigned long vlen, rwf_t flags)
1022 struct fd f = fdget_pos(fd);
1023 ssize_t ret = -EBADF;
1026 loff_t pos = file_pos_read(f.file);
1027 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1029 file_pos_write(f.file, pos);
1034 add_wchar(current, ret);
1039 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1041 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1042 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1045 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1046 unsigned long vlen, loff_t pos, rwf_t flags)
1049 ssize_t ret = -EBADF;
1057 if (f.file->f_mode & FMODE_PREAD)
1058 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1063 add_rchar(current, ret);
1068 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1069 unsigned long vlen, loff_t pos, rwf_t flags)
1072 ssize_t ret = -EBADF;
1080 if (f.file->f_mode & FMODE_PWRITE)
1081 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1086 add_wchar(current, ret);
1091 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1092 unsigned long, vlen)
1094 return do_readv(fd, vec, vlen, 0);
1097 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1098 unsigned long, vlen)
1100 return do_writev(fd, vec, vlen, 0);
1103 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1104 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1106 loff_t pos = pos_from_hilo(pos_h, pos_l);
1108 return do_preadv(fd, vec, vlen, pos, 0);
1111 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1112 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1115 loff_t pos = pos_from_hilo(pos_h, pos_l);
1118 return do_readv(fd, vec, vlen, flags);
1120 return do_preadv(fd, vec, vlen, pos, flags);
1123 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1124 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1126 loff_t pos = pos_from_hilo(pos_h, pos_l);
1128 return do_pwritev(fd, vec, vlen, pos, 0);
1131 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1132 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1135 loff_t pos = pos_from_hilo(pos_h, pos_l);
1138 return do_writev(fd, vec, vlen, flags);
1140 return do_pwritev(fd, vec, vlen, pos, flags);
1143 #ifdef CONFIG_COMPAT
1144 static size_t compat_readv(struct file *file,
1145 const struct compat_iovec __user *vec,
1146 unsigned long vlen, loff_t *pos, rwf_t flags)
1148 struct iovec iovstack[UIO_FASTIOV];
1149 struct iovec *iov = iovstack;
1150 struct iov_iter iter;
1153 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1155 ret = do_iter_read(file, &iter, pos, flags);
1159 add_rchar(current, ret);
1164 static size_t do_compat_readv(compat_ulong_t fd,
1165 const struct compat_iovec __user *vec,
1166 compat_ulong_t vlen, rwf_t flags)
1168 struct fd f = fdget_pos(fd);
1174 pos = f.file->f_pos;
1175 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1177 f.file->f_pos = pos;
1183 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1184 const struct compat_iovec __user *,vec,
1185 compat_ulong_t, vlen)
1187 return do_compat_readv(fd, vec, vlen, 0);
1190 static long do_compat_preadv64(unsigned long fd,
1191 const struct compat_iovec __user *vec,
1192 unsigned long vlen, loff_t pos, rwf_t flags)
1203 if (f.file->f_mode & FMODE_PREAD)
1204 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1209 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1210 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1211 const struct compat_iovec __user *,vec,
1212 unsigned long, vlen, loff_t, pos)
1214 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1218 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1219 const struct compat_iovec __user *,vec,
1220 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1222 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1224 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1227 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1228 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1229 const struct compat_iovec __user *,vec,
1230 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1232 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1236 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1237 const struct compat_iovec __user *,vec,
1238 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1241 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1244 return do_compat_readv(fd, vec, vlen, flags);
1246 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1249 static size_t compat_writev(struct file *file,
1250 const struct compat_iovec __user *vec,
1251 unsigned long vlen, loff_t *pos, rwf_t flags)
1253 struct iovec iovstack[UIO_FASTIOV];
1254 struct iovec *iov = iovstack;
1255 struct iov_iter iter;
1258 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1260 file_start_write(file);
1261 ret = do_iter_write(file, &iter, pos, flags);
1262 file_end_write(file);
1266 add_wchar(current, ret);
1271 static size_t do_compat_writev(compat_ulong_t fd,
1272 const struct compat_iovec __user* vec,
1273 compat_ulong_t vlen, rwf_t flags)
1275 struct fd f = fdget_pos(fd);
1281 pos = f.file->f_pos;
1282 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1284 f.file->f_pos = pos;
1289 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1290 const struct compat_iovec __user *, vec,
1291 compat_ulong_t, vlen)
1293 return do_compat_writev(fd, vec, vlen, 0);
1296 static long do_compat_pwritev64(unsigned long fd,
1297 const struct compat_iovec __user *vec,
1298 unsigned long vlen, loff_t pos, rwf_t flags)
1309 if (f.file->f_mode & FMODE_PWRITE)
1310 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1315 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1316 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1317 const struct compat_iovec __user *,vec,
1318 unsigned long, vlen, loff_t, pos)
1320 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1324 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1325 const struct compat_iovec __user *,vec,
1326 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1328 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1330 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1333 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1334 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1335 const struct compat_iovec __user *,vec,
1336 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1338 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1342 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1343 const struct compat_iovec __user *,vec,
1344 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1346 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1349 return do_compat_writev(fd, vec, vlen, flags);
1351 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1356 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1357 size_t count, loff_t max)
1360 struct inode *in_inode, *out_inode;
1367 * Get input file, and verify that it is ok..
1373 if (!(in.file->f_mode & FMODE_READ))
1377 pos = in.file->f_pos;
1380 if (!(in.file->f_mode & FMODE_PREAD))
1383 retval = rw_verify_area(READ, in.file, &pos, count);
1386 if (count > MAX_RW_COUNT)
1387 count = MAX_RW_COUNT;
1390 * Get output file, and verify that it is ok..
1393 out = fdget(out_fd);
1396 if (!(out.file->f_mode & FMODE_WRITE))
1399 in_inode = file_inode(in.file);
1400 out_inode = file_inode(out.file);
1401 out_pos = out.file->f_pos;
1402 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1407 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1409 if (unlikely(pos + count > max)) {
1410 retval = -EOVERFLOW;
1419 * We need to debate whether we can enable this or not. The
1420 * man page documents EAGAIN return for the output at least,
1421 * and the application is arguably buggy if it doesn't expect
1422 * EAGAIN on a non-blocking file descriptor.
1424 if (in.file->f_flags & O_NONBLOCK)
1425 fl = SPLICE_F_NONBLOCK;
1427 file_start_write(out.file);
1428 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1429 file_end_write(out.file);
1432 add_rchar(current, retval);
1433 add_wchar(current, retval);
1434 fsnotify_access(in.file);
1435 fsnotify_modify(out.file);
1436 out.file->f_pos = out_pos;
1440 in.file->f_pos = pos;
1446 retval = -EOVERFLOW;
1456 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1463 if (unlikely(get_user(off, offset)))
1466 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1467 if (unlikely(put_user(pos, offset)))
1472 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1475 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1481 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1483 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1484 if (unlikely(put_user(pos, offset)))
1489 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1492 #ifdef CONFIG_COMPAT
1493 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1494 compat_off_t __user *, offset, compat_size_t, count)
1501 if (unlikely(get_user(off, offset)))
1504 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1505 if (unlikely(put_user(pos, offset)))
1510 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1513 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1514 compat_loff_t __user *, offset, compat_size_t, count)
1520 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1522 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1523 if (unlikely(put_user(pos, offset)))
1528 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1533 * copy_file_range() differs from regular file read and write in that it
1534 * specifically allows return partial success. When it does so is up to
1535 * the copy_file_range method.
1537 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1538 struct file *file_out, loff_t pos_out,
1539 size_t len, unsigned int flags)
1541 struct inode *inode_in = file_inode(file_in);
1542 struct inode *inode_out = file_inode(file_out);
1548 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1550 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1553 ret = rw_verify_area(READ, file_in, &pos_in, len);
1557 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1561 if (!(file_in->f_mode & FMODE_READ) ||
1562 !(file_out->f_mode & FMODE_WRITE) ||
1563 (file_out->f_flags & O_APPEND))
1566 /* this could be relaxed once a method supports cross-fs copies */
1567 if (inode_in->i_sb != inode_out->i_sb)
1573 file_start_write(file_out);
1576 * Try cloning first, this is supported by more file systems, and
1577 * more efficient if both clone and copy are supported (e.g. NFS).
1579 if (file_in->f_op->clone_file_range) {
1580 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1581 file_out, pos_out, len);
1588 if (file_out->f_op->copy_file_range) {
1589 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1590 pos_out, len, flags);
1591 if (ret != -EOPNOTSUPP)
1595 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1596 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1600 fsnotify_access(file_in);
1601 add_rchar(current, ret);
1602 fsnotify_modify(file_out);
1603 add_wchar(current, ret);
1609 file_end_write(file_out);
1613 EXPORT_SYMBOL(vfs_copy_file_range);
1615 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1616 int, fd_out, loff_t __user *, off_out,
1617 size_t, len, unsigned int, flags)
1623 ssize_t ret = -EBADF;
1625 f_in = fdget(fd_in);
1629 f_out = fdget(fd_out);
1635 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1638 pos_in = f_in.file->f_pos;
1642 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1645 pos_out = f_out.file->f_pos;
1648 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1655 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1658 f_in.file->f_pos = pos_in;
1662 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1665 f_out.file->f_pos = pos_out;
1677 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1679 struct inode *inode = file_inode(file);
1681 if (unlikely(pos < 0))
1684 if (unlikely((loff_t) (pos + len) < 0))
1687 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1688 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1691 retval = locks_mandatory_area(inode, file, pos, end,
1692 write ? F_WRLCK : F_RDLCK);
1697 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1701 * Check that the two inodes are eligible for cloning, the ranges make
1702 * sense, and then flush all dirty data. Caller must ensure that the
1703 * inodes have been locked against any other modifications.
1705 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1706 * the usual negative error code.
1708 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1709 struct inode *inode_out, loff_t pos_out,
1710 u64 *len, bool is_dedupe)
1712 loff_t bs = inode_out->i_sb->s_blocksize;
1715 bool same_inode = (inode_in == inode_out);
1718 /* Don't touch certain kinds of inodes */
1719 if (IS_IMMUTABLE(inode_out))
1722 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1725 /* Don't reflink dirs, pipes, sockets... */
1726 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1728 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1731 /* Are we going all the way to the end? */
1732 isize = i_size_read(inode_in);
1736 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1738 if (is_dedupe || pos_in == isize)
1742 *len = isize - pos_in;
1745 /* Ensure offsets don't wrap and the input is inside i_size */
1746 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1747 pos_in + *len > isize)
1750 /* Don't allow dedupe past EOF in the dest file */
1754 disize = i_size_read(inode_out);
1755 if (pos_out >= disize || pos_out + *len > disize)
1759 /* If we're linking to EOF, continue to the block boundary. */
1760 if (pos_in + *len == isize)
1761 blen = ALIGN(isize, bs) - pos_in;
1765 /* Only reflink if we're aligned to block boundaries */
1766 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1767 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1770 /* Don't allow overlapped reflink within the same file */
1772 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1776 /* Wait for the completion of any pending IOs on both files */
1777 inode_dio_wait(inode_in);
1779 inode_dio_wait(inode_out);
1781 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1782 pos_in, pos_in + *len - 1);
1786 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1787 pos_out, pos_out + *len - 1);
1792 * Check that the extents are the same.
1795 bool is_same = false;
1797 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1798 inode_out, pos_out, *len, &is_same);
1807 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1809 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1810 struct file *file_out, loff_t pos_out, u64 len)
1812 struct inode *inode_in = file_inode(file_in);
1813 struct inode *inode_out = file_inode(file_out);
1816 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1818 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1822 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1823 * the same mount. Practically, they only need to be on the same file
1826 if (inode_in->i_sb != inode_out->i_sb)
1829 if (!(file_in->f_mode & FMODE_READ) ||
1830 !(file_out->f_mode & FMODE_WRITE) ||
1831 (file_out->f_flags & O_APPEND))
1834 if (!file_in->f_op->clone_file_range)
1837 ret = clone_verify_area(file_in, pos_in, len, false);
1841 ret = clone_verify_area(file_out, pos_out, len, true);
1845 if (pos_in + len > i_size_read(inode_in))
1848 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1849 file_out, pos_out, len);
1851 fsnotify_access(file_in);
1852 fsnotify_modify(file_out);
1857 EXPORT_SYMBOL(vfs_clone_file_range);
1860 * Read a page's worth of file data into the page cache. Return the page
1863 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1865 struct address_space *mapping;
1869 n = offset >> PAGE_SHIFT;
1870 mapping = inode->i_mapping;
1871 page = read_mapping_page(mapping, n, NULL);
1874 if (!PageUptodate(page)) {
1876 return ERR_PTR(-EIO);
1883 * Compare extents of two files to see if they are the same.
1884 * Caller must have locked both inodes to prevent write races.
1886 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1887 struct inode *dest, loff_t destoff,
1888 loff_t len, bool *is_same)
1894 struct page *src_page;
1895 struct page *dest_page;
1903 src_poff = srcoff & (PAGE_SIZE - 1);
1904 dest_poff = destoff & (PAGE_SIZE - 1);
1905 cmp_len = min(PAGE_SIZE - src_poff,
1906 PAGE_SIZE - dest_poff);
1907 cmp_len = min(cmp_len, len);
1911 src_page = vfs_dedupe_get_page(src, srcoff);
1912 if (IS_ERR(src_page)) {
1913 error = PTR_ERR(src_page);
1916 dest_page = vfs_dedupe_get_page(dest, destoff);
1917 if (IS_ERR(dest_page)) {
1918 error = PTR_ERR(dest_page);
1919 unlock_page(src_page);
1923 src_addr = kmap_atomic(src_page);
1924 dest_addr = kmap_atomic(dest_page);
1926 flush_dcache_page(src_page);
1927 flush_dcache_page(dest_page);
1929 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1932 kunmap_atomic(dest_addr);
1933 kunmap_atomic(src_addr);
1934 unlock_page(dest_page);
1935 unlock_page(src_page);
1936 put_page(dest_page);
1953 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1955 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1957 struct file_dedupe_range_info *info;
1958 struct inode *src = file_inode(file);
1963 bool is_admin = capable(CAP_SYS_ADMIN);
1964 u16 count = same->dest_count;
1965 struct file *dst_file;
1969 if (!(file->f_mode & FMODE_READ))
1972 if (same->reserved1 || same->reserved2)
1975 off = same->src_offset;
1976 len = same->src_length;
1979 if (S_ISDIR(src->i_mode))
1983 if (!S_ISREG(src->i_mode))
1986 ret = clone_verify_area(file, off, len, false);
1991 if (off + len > i_size_read(src))
1994 /* pre-format output fields to sane values */
1995 for (i = 0; i < count; i++) {
1996 same->info[i].bytes_deduped = 0ULL;
1997 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2000 for (i = 0, info = same->info; i < count; i++, info++) {
2002 struct fd dst_fd = fdget(info->dest_fd);
2004 dst_file = dst_fd.file;
2006 info->status = -EBADF;
2009 dst = file_inode(dst_file);
2011 ret = mnt_want_write_file(dst_file);
2017 dst_off = info->dest_offset;
2018 ret = clone_verify_area(dst_file, dst_off, len, true);
2025 if (info->reserved) {
2026 info->status = -EINVAL;
2027 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2028 info->status = -EINVAL;
2029 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2030 info->status = -EXDEV;
2031 } else if (S_ISDIR(dst->i_mode)) {
2032 info->status = -EISDIR;
2033 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2034 info->status = -EINVAL;
2036 deduped = dst_file->f_op->dedupe_file_range(file, off,
2039 if (deduped == -EBADE)
2040 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2041 else if (deduped < 0)
2042 info->status = deduped;
2044 info->bytes_deduped += deduped;
2048 mnt_drop_write_file(dst_file);
2052 if (fatal_signal_pending(current))
2059 EXPORT_SYMBOL(vfs_dedupe_file_range);