2 * linux/fs/read_write.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/sched/xacct.h>
10 #include <linux/fcntl.h>
11 #include <linux/file.h>
12 #include <linux/uio.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/export.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include <linux/compat.h>
20 #include <linux/mount.h>
24 #include <linux/uaccess.h>
25 #include <asm/unistd.h>
27 const struct file_operations generic_ro_fops = {
28 .llseek = generic_file_llseek,
29 .read_iter = generic_file_read_iter,
30 .mmap = generic_file_readonly_mmap,
31 .splice_read = generic_file_splice_read,
34 EXPORT_SYMBOL(generic_ro_fops);
36 static inline int unsigned_offsets(struct file *file)
38 return file->f_mode & FMODE_UNSIGNED_OFFSET;
42 * vfs_setpos - update the file offset for lseek
43 * @file: file structure in question
44 * @offset: file offset to seek to
45 * @maxsize: maximum file size
47 * This is a low-level filesystem helper for updating the file offset to
48 * the value specified by @offset if the given offset is valid and it is
49 * not equal to the current file offset.
51 * Return the specified offset on success and -EINVAL on invalid offset.
53 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
55 if (offset < 0 && !unsigned_offsets(file))
60 if (offset != file->f_pos) {
66 EXPORT_SYMBOL(vfs_setpos);
69 * generic_file_llseek_size - generic llseek implementation for regular files
70 * @file: file structure to seek on
71 * @offset: file offset to seek to
72 * @whence: type of seek
73 * @size: max size of this file in file system
74 * @eof: offset used for SEEK_END position
76 * This is a variant of generic_file_llseek that allows passing in a custom
77 * maximum file size and a custom EOF position, for e.g. hashed directories
80 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
81 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
82 * read/writes behave like SEEK_SET against seeks.
85 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
86 loff_t maxsize, loff_t eof)
94 * Here we special-case the lseek(fd, 0, SEEK_CUR)
95 * position-querying operation. Avoid rewriting the "same"
96 * f_pos value back to the file because a concurrent read(),
97 * write() or lseek() might have altered it
102 * f_lock protects against read/modify/write race with other
103 * SEEK_CURs. Note that parallel writes and reads behave
106 spin_lock(&file->f_lock);
107 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
108 spin_unlock(&file->f_lock);
112 * In the generic case the entire file is data, so as long as
113 * offset isn't at the end of the file then the offset is data.
120 * There is a virtual hole at the end of the file, so as long as
121 * offset isn't i_size or larger, return i_size.
129 return vfs_setpos(file, offset, maxsize);
131 EXPORT_SYMBOL(generic_file_llseek_size);
134 * generic_file_llseek - generic llseek implementation for regular files
135 * @file: file structure to seek on
136 * @offset: file offset to seek to
137 * @whence: type of seek
139 * This is a generic implemenation of ->llseek useable for all normal local
140 * filesystems. It just updates the file offset to the value specified by
141 * @offset and @whence.
143 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
145 struct inode *inode = file->f_mapping->host;
147 return generic_file_llseek_size(file, offset, whence,
148 inode->i_sb->s_maxbytes,
151 EXPORT_SYMBOL(generic_file_llseek);
154 * fixed_size_llseek - llseek implementation for fixed-sized devices
155 * @file: file structure to seek on
156 * @offset: file offset to seek to
157 * @whence: type of seek
158 * @size: size of the file
161 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
164 case SEEK_SET: case SEEK_CUR: case SEEK_END:
165 return generic_file_llseek_size(file, offset, whence,
171 EXPORT_SYMBOL(fixed_size_llseek);
174 * no_seek_end_llseek - llseek implementation for fixed-sized devices
175 * @file: file structure to seek on
176 * @offset: file offset to seek to
177 * @whence: type of seek
180 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
183 case SEEK_SET: case SEEK_CUR:
184 return generic_file_llseek_size(file, offset, whence,
190 EXPORT_SYMBOL(no_seek_end_llseek);
193 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
194 * @file: file structure to seek on
195 * @offset: file offset to seek to
196 * @whence: type of seek
197 * @size: maximal offset allowed
200 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
203 case SEEK_SET: case SEEK_CUR:
204 return generic_file_llseek_size(file, offset, whence,
210 EXPORT_SYMBOL(no_seek_end_llseek_size);
213 * noop_llseek - No Operation Performed llseek implementation
214 * @file: file structure to seek on
215 * @offset: file offset to seek to
216 * @whence: type of seek
218 * This is an implementation of ->llseek useable for the rare special case when
219 * userspace expects the seek to succeed but the (device) file is actually not
220 * able to perform the seek. In this case you use noop_llseek() instead of
221 * falling back to the default implementation of ->llseek.
223 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
227 EXPORT_SYMBOL(noop_llseek);
229 loff_t no_llseek(struct file *file, loff_t offset, int whence)
233 EXPORT_SYMBOL(no_llseek);
235 loff_t default_llseek(struct file *file, loff_t offset, int whence)
237 struct inode *inode = file_inode(file);
243 offset += i_size_read(inode);
247 retval = file->f_pos;
250 offset += file->f_pos;
254 * In the generic case the entire file is data, so as
255 * long as offset isn't at the end of the file then the
258 if (offset >= inode->i_size) {
265 * There is a virtual hole at the end of the file, so
266 * as long as offset isn't i_size or larger, return
269 if (offset >= inode->i_size) {
273 offset = inode->i_size;
277 if (offset >= 0 || unsigned_offsets(file)) {
278 if (offset != file->f_pos) {
279 file->f_pos = offset;
288 EXPORT_SYMBOL(default_llseek);
290 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
292 loff_t (*fn)(struct file *, loff_t, int);
295 if (file->f_mode & FMODE_LSEEK) {
296 if (file->f_op->llseek)
297 fn = file->f_op->llseek;
299 return fn(file, offset, whence);
301 EXPORT_SYMBOL(vfs_llseek);
303 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
306 struct fd f = fdget_pos(fd);
311 if (whence <= SEEK_MAX) {
312 loff_t res = vfs_llseek(f.file, offset, whence);
314 if (res != (loff_t)retval)
315 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
322 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
324 return sys_lseek(fd, offset, whence);
328 #ifdef __ARCH_WANT_SYS_LLSEEK
329 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
330 unsigned long, offset_low, loff_t __user *, result,
331 unsigned int, whence)
334 struct fd f = fdget_pos(fd);
341 if (whence > SEEK_MAX)
344 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
347 retval = (int)offset;
350 if (!copy_to_user(result, &offset, sizeof(offset)))
359 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
364 if (!file->f_op->read_iter)
367 init_sync_kiocb(&kiocb, file);
368 kiocb.ki_pos = *ppos;
371 ret = call_read_iter(file, &kiocb, iter);
372 BUG_ON(ret == -EIOCBQUEUED);
374 *ppos = kiocb.ki_pos;
377 EXPORT_SYMBOL(vfs_iter_read);
379 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
384 if (!file->f_op->write_iter)
387 init_sync_kiocb(&kiocb, file);
388 kiocb.ki_pos = *ppos;
391 ret = call_write_iter(file, &kiocb, iter);
392 BUG_ON(ret == -EIOCBQUEUED);
394 *ppos = kiocb.ki_pos;
397 EXPORT_SYMBOL(vfs_iter_write);
399 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
403 int retval = -EINVAL;
405 inode = file_inode(file);
406 if (unlikely((ssize_t) count < 0))
409 if (unlikely(pos < 0)) {
410 if (!unsigned_offsets(file))
412 if (count >= -pos) /* both values are in 0..LLONG_MAX */
414 } else if (unlikely((loff_t) (pos + count) < 0)) {
415 if (!unsigned_offsets(file))
419 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
420 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
421 read_write == READ ? F_RDLCK : F_WRLCK);
425 return security_file_permission(file,
426 read_write == READ ? MAY_READ : MAY_WRITE);
429 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
431 struct iovec iov = { .iov_base = buf, .iov_len = len };
433 struct iov_iter iter;
436 init_sync_kiocb(&kiocb, filp);
437 kiocb.ki_pos = *ppos;
438 iov_iter_init(&iter, READ, &iov, 1, len);
440 ret = call_read_iter(filp, &kiocb, &iter);
441 BUG_ON(ret == -EIOCBQUEUED);
442 *ppos = kiocb.ki_pos;
446 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
449 if (file->f_op->read)
450 return file->f_op->read(file, buf, count, pos);
451 else if (file->f_op->read_iter)
452 return new_sync_read(file, buf, count, pos);
456 EXPORT_SYMBOL(__vfs_read);
458 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
462 if (!(file->f_mode & FMODE_READ))
464 if (!(file->f_mode & FMODE_CAN_READ))
466 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
469 ret = rw_verify_area(READ, file, pos, count);
471 if (count > MAX_RW_COUNT)
472 count = MAX_RW_COUNT;
473 ret = __vfs_read(file, buf, count, pos);
475 fsnotify_access(file);
476 add_rchar(current, ret);
484 EXPORT_SYMBOL(vfs_read);
486 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
488 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
490 struct iov_iter iter;
493 init_sync_kiocb(&kiocb, filp);
494 kiocb.ki_pos = *ppos;
495 iov_iter_init(&iter, WRITE, &iov, 1, len);
497 ret = call_write_iter(filp, &kiocb, &iter);
498 BUG_ON(ret == -EIOCBQUEUED);
500 *ppos = kiocb.ki_pos;
504 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
507 if (file->f_op->write)
508 return file->f_op->write(file, p, count, pos);
509 else if (file->f_op->write_iter)
510 return new_sync_write(file, p, count, pos);
514 EXPORT_SYMBOL(__vfs_write);
516 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
519 const char __user *p;
522 if (!(file->f_mode & FMODE_CAN_WRITE))
527 p = (__force const char __user *)buf;
528 if (count > MAX_RW_COUNT)
529 count = MAX_RW_COUNT;
530 ret = __vfs_write(file, p, count, pos);
533 fsnotify_modify(file);
534 add_wchar(current, ret);
540 EXPORT_SYMBOL(__kernel_write);
542 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
546 if (!(file->f_mode & FMODE_WRITE))
548 if (!(file->f_mode & FMODE_CAN_WRITE))
550 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
553 ret = rw_verify_area(WRITE, file, pos, count);
555 if (count > MAX_RW_COUNT)
556 count = MAX_RW_COUNT;
557 file_start_write(file);
558 ret = __vfs_write(file, buf, count, pos);
560 fsnotify_modify(file);
561 add_wchar(current, ret);
564 file_end_write(file);
570 EXPORT_SYMBOL(vfs_write);
572 static inline loff_t file_pos_read(struct file *file)
577 static inline void file_pos_write(struct file *file, loff_t pos)
582 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
584 struct fd f = fdget_pos(fd);
585 ssize_t ret = -EBADF;
588 loff_t pos = file_pos_read(f.file);
589 ret = vfs_read(f.file, buf, count, &pos);
591 file_pos_write(f.file, pos);
597 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
600 struct fd f = fdget_pos(fd);
601 ssize_t ret = -EBADF;
604 loff_t pos = file_pos_read(f.file);
605 ret = vfs_write(f.file, buf, count, &pos);
607 file_pos_write(f.file, pos);
614 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
615 size_t, count, loff_t, pos)
618 ssize_t ret = -EBADF;
626 if (f.file->f_mode & FMODE_PREAD)
627 ret = vfs_read(f.file, buf, count, &pos);
634 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
635 size_t, count, loff_t, pos)
638 ssize_t ret = -EBADF;
646 if (f.file->f_mode & FMODE_PWRITE)
647 ret = vfs_write(f.file, buf, count, &pos);
655 * Reduce an iovec's length in-place. Return the resulting number of segments
657 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
659 unsigned long seg = 0;
662 while (seg < nr_segs) {
664 if (len + iov->iov_len >= to) {
665 iov->iov_len = to - len;
673 EXPORT_SYMBOL(iov_shorten);
675 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
676 loff_t *ppos, int type, int flags)
681 init_sync_kiocb(&kiocb, filp);
682 ret = kiocb_set_rw_flags(&kiocb, flags);
685 kiocb.ki_pos = *ppos;
688 ret = call_read_iter(filp, &kiocb, iter);
690 ret = call_write_iter(filp, &kiocb, iter);
691 BUG_ON(ret == -EIOCBQUEUED);
692 *ppos = kiocb.ki_pos;
696 /* Do it by hand, with file-ops */
697 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
698 loff_t *ppos, int type, int flags)
702 if (flags & ~RWF_HIPRI)
705 while (iov_iter_count(iter)) {
706 struct iovec iovec = iov_iter_iovec(iter);
710 nr = filp->f_op->read(filp, iovec.iov_base,
711 iovec.iov_len, ppos);
713 nr = filp->f_op->write(filp, iovec.iov_base,
714 iovec.iov_len, ppos);
723 if (nr != iovec.iov_len)
725 iov_iter_advance(iter, nr);
731 /* A write operation does a read from user space and vice versa */
732 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
735 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
736 * into the kernel and check that it is valid.
738 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
739 * @uvector: Pointer to the userspace array.
740 * @nr_segs: Number of elements in userspace array.
741 * @fast_segs: Number of elements in @fast_pointer.
742 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
743 * @ret_pointer: (output parameter) Pointer to a variable that will point to
744 * either @fast_pointer, a newly allocated kernel array, or NULL,
745 * depending on which array was used.
747 * This function copies an array of &struct iovec of @nr_segs from
748 * userspace into the kernel and checks that each element is valid (e.g.
749 * it does not point to a kernel address or cause overflow by being too
752 * As an optimization, the caller may provide a pointer to a small
753 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
754 * (the size of this array, or 0 if unused, should be given in @fast_segs).
756 * @ret_pointer will always point to the array that was used, so the
757 * caller must take care not to call kfree() on it e.g. in case the
758 * @fast_pointer array was used and it was allocated on the stack.
760 * Return: The total number of bytes covered by the iovec array on success
761 * or a negative error code on error.
763 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
764 unsigned long nr_segs, unsigned long fast_segs,
765 struct iovec *fast_pointer,
766 struct iovec **ret_pointer)
770 struct iovec *iov = fast_pointer;
773 * SuS says "The readv() function *may* fail if the iovcnt argument
774 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
775 * traditionally returned zero for zero segments, so...
783 * First get the "struct iovec" from user memory and
784 * verify all the pointers
786 if (nr_segs > UIO_MAXIOV) {
790 if (nr_segs > fast_segs) {
791 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
797 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
803 * According to the Single Unix Specification we should return EINVAL
804 * if an element length is < 0 when cast to ssize_t or if the
805 * total length would overflow the ssize_t return value of the
808 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
812 for (seg = 0; seg < nr_segs; seg++) {
813 void __user *buf = iov[seg].iov_base;
814 ssize_t len = (ssize_t)iov[seg].iov_len;
816 /* see if we we're about to use an invalid len or if
817 * it's about to overflow ssize_t */
823 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
827 if (len > MAX_RW_COUNT - ret) {
828 len = MAX_RW_COUNT - ret;
829 iov[seg].iov_len = len;
839 ssize_t compat_rw_copy_check_uvector(int type,
840 const struct compat_iovec __user *uvector, unsigned long nr_segs,
841 unsigned long fast_segs, struct iovec *fast_pointer,
842 struct iovec **ret_pointer)
844 compat_ssize_t tot_len;
845 struct iovec *iov = *ret_pointer = fast_pointer;
850 * SuS says "The readv() function *may* fail if the iovcnt argument
851 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
852 * traditionally returned zero for zero segments, so...
858 if (nr_segs > UIO_MAXIOV)
860 if (nr_segs > fast_segs) {
862 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
869 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
873 * Single unix specification:
874 * We should -EINVAL if an element length is not >= 0 and fitting an
877 * In Linux, the total length is limited to MAX_RW_COUNT, there is
878 * no overflow possibility.
882 for (seg = 0; seg < nr_segs; seg++) {
886 if (__get_user(len, &uvector->iov_len) ||
887 __get_user(buf, &uvector->iov_base)) {
891 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
894 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
898 if (len > MAX_RW_COUNT - tot_len)
899 len = MAX_RW_COUNT - tot_len;
901 iov->iov_base = compat_ptr(buf);
902 iov->iov_len = (compat_size_t) len;
913 static ssize_t __do_readv_writev(int type, struct file *file,
914 struct iov_iter *iter, loff_t *pos, int flags)
919 tot_len = iov_iter_count(iter);
922 ret = rw_verify_area(type, file, pos, tot_len);
927 file_start_write(file);
929 if ((type == READ && file->f_op->read_iter) ||
930 (type == WRITE && file->f_op->write_iter))
931 ret = do_iter_readv_writev(file, iter, pos, type, flags);
933 ret = do_loop_readv_writev(file, iter, pos, type, flags);
936 file_end_write(file);
939 if ((ret + (type == READ)) > 0) {
941 fsnotify_access(file);
943 fsnotify_modify(file);
948 static ssize_t do_readv_writev(int type, struct file *file,
949 const struct iovec __user *uvector,
950 unsigned long nr_segs, loff_t *pos,
953 struct iovec iovstack[UIO_FASTIOV];
954 struct iovec *iov = iovstack;
955 struct iov_iter iter;
958 ret = import_iovec(type, uvector, nr_segs,
959 ARRAY_SIZE(iovstack), &iov, &iter);
963 ret = __do_readv_writev(type, file, &iter, pos, flags);
969 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
970 unsigned long vlen, loff_t *pos, int flags)
972 if (!(file->f_mode & FMODE_READ))
974 if (!(file->f_mode & FMODE_CAN_READ))
977 return do_readv_writev(READ, file, vec, vlen, pos, flags);
980 EXPORT_SYMBOL(vfs_readv);
982 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
983 unsigned long vlen, loff_t *pos, int flags)
985 if (!(file->f_mode & FMODE_WRITE))
987 if (!(file->f_mode & FMODE_CAN_WRITE))
990 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
993 EXPORT_SYMBOL(vfs_writev);
995 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
996 unsigned long vlen, int flags)
998 struct fd f = fdget_pos(fd);
999 ssize_t ret = -EBADF;
1002 loff_t pos = file_pos_read(f.file);
1003 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1005 file_pos_write(f.file, pos);
1010 add_rchar(current, ret);
1015 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1016 unsigned long vlen, int flags)
1018 struct fd f = fdget_pos(fd);
1019 ssize_t ret = -EBADF;
1022 loff_t pos = file_pos_read(f.file);
1023 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1025 file_pos_write(f.file, pos);
1030 add_wchar(current, ret);
1035 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1037 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1038 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1041 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1042 unsigned long vlen, loff_t pos, int flags)
1045 ssize_t ret = -EBADF;
1053 if (f.file->f_mode & FMODE_PREAD)
1054 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1059 add_rchar(current, ret);
1064 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1065 unsigned long vlen, loff_t pos, int flags)
1068 ssize_t ret = -EBADF;
1076 if (f.file->f_mode & FMODE_PWRITE)
1077 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1082 add_wchar(current, ret);
1087 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1088 unsigned long, vlen)
1090 return do_readv(fd, vec, vlen, 0);
1093 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1094 unsigned long, vlen)
1096 return do_writev(fd, vec, vlen, 0);
1099 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1100 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1102 loff_t pos = pos_from_hilo(pos_h, pos_l);
1104 return do_preadv(fd, vec, vlen, pos, 0);
1107 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1108 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1111 loff_t pos = pos_from_hilo(pos_h, pos_l);
1114 return do_readv(fd, vec, vlen, flags);
1116 return do_preadv(fd, vec, vlen, pos, flags);
1119 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1120 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1122 loff_t pos = pos_from_hilo(pos_h, pos_l);
1124 return do_pwritev(fd, vec, vlen, pos, 0);
1127 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1128 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1131 loff_t pos = pos_from_hilo(pos_h, pos_l);
1134 return do_writev(fd, vec, vlen, flags);
1136 return do_pwritev(fd, vec, vlen, pos, flags);
1139 #ifdef CONFIG_COMPAT
1141 static ssize_t compat_do_readv_writev(int type, struct file *file,
1142 const struct compat_iovec __user *uvector,
1143 unsigned long nr_segs, loff_t *pos,
1146 struct iovec iovstack[UIO_FASTIOV];
1147 struct iovec *iov = iovstack;
1148 struct iov_iter iter;
1151 ret = compat_import_iovec(type, uvector, nr_segs,
1152 UIO_FASTIOV, &iov, &iter);
1156 ret = __do_readv_writev(type, file, &iter, pos, flags);
1162 static size_t compat_readv(struct file *file,
1163 const struct compat_iovec __user *vec,
1164 unsigned long vlen, loff_t *pos, int flags)
1166 ssize_t ret = -EBADF;
1168 if (!(file->f_mode & FMODE_READ))
1172 if (!(file->f_mode & FMODE_CAN_READ))
1175 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1179 add_rchar(current, ret);
1184 static size_t do_compat_readv(compat_ulong_t fd,
1185 const struct compat_iovec __user *vec,
1186 compat_ulong_t vlen, int flags)
1188 struct fd f = fdget_pos(fd);
1194 pos = f.file->f_pos;
1195 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1197 f.file->f_pos = pos;
1203 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1204 const struct compat_iovec __user *,vec,
1205 compat_ulong_t, vlen)
1207 return do_compat_readv(fd, vec, vlen, 0);
1210 static long do_compat_preadv64(unsigned long fd,
1211 const struct compat_iovec __user *vec,
1212 unsigned long vlen, loff_t pos, int flags)
1223 if (f.file->f_mode & FMODE_PREAD)
1224 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1229 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1230 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1231 const struct compat_iovec __user *,vec,
1232 unsigned long, vlen, loff_t, pos)
1234 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1238 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1239 const struct compat_iovec __user *,vec,
1240 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1242 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1244 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1247 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1248 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1249 const struct compat_iovec __user *,vec,
1250 unsigned long, vlen, loff_t, pos, int, flags)
1252 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1256 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1257 const struct compat_iovec __user *,vec,
1258 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1261 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1264 return do_compat_readv(fd, vec, vlen, flags);
1266 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1269 static size_t compat_writev(struct file *file,
1270 const struct compat_iovec __user *vec,
1271 unsigned long vlen, loff_t *pos, int flags)
1273 ssize_t ret = -EBADF;
1275 if (!(file->f_mode & FMODE_WRITE))
1279 if (!(file->f_mode & FMODE_CAN_WRITE))
1282 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, flags);
1286 add_wchar(current, ret);
1291 static size_t do_compat_writev(compat_ulong_t fd,
1292 const struct compat_iovec __user* vec,
1293 compat_ulong_t vlen, int flags)
1295 struct fd f = fdget_pos(fd);
1301 pos = f.file->f_pos;
1302 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1304 f.file->f_pos = pos;
1309 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1310 const struct compat_iovec __user *, vec,
1311 compat_ulong_t, vlen)
1313 return do_compat_writev(fd, vec, vlen, 0);
1316 static long do_compat_pwritev64(unsigned long fd,
1317 const struct compat_iovec __user *vec,
1318 unsigned long vlen, loff_t pos, int flags)
1329 if (f.file->f_mode & FMODE_PWRITE)
1330 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1335 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1336 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1337 const struct compat_iovec __user *,vec,
1338 unsigned long, vlen, loff_t, pos)
1340 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1344 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1345 const struct compat_iovec __user *,vec,
1346 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1348 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1350 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1353 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1354 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1355 const struct compat_iovec __user *,vec,
1356 unsigned long, vlen, loff_t, pos, int, flags)
1358 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1362 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1363 const struct compat_iovec __user *,vec,
1364 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1366 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1369 return do_compat_writev(fd, vec, vlen, flags);
1371 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1376 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1377 size_t count, loff_t max)
1380 struct inode *in_inode, *out_inode;
1387 * Get input file, and verify that it is ok..
1393 if (!(in.file->f_mode & FMODE_READ))
1397 pos = in.file->f_pos;
1400 if (!(in.file->f_mode & FMODE_PREAD))
1403 retval = rw_verify_area(READ, in.file, &pos, count);
1406 if (count > MAX_RW_COUNT)
1407 count = MAX_RW_COUNT;
1410 * Get output file, and verify that it is ok..
1413 out = fdget(out_fd);
1416 if (!(out.file->f_mode & FMODE_WRITE))
1419 in_inode = file_inode(in.file);
1420 out_inode = file_inode(out.file);
1421 out_pos = out.file->f_pos;
1422 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1427 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1429 if (unlikely(pos + count > max)) {
1430 retval = -EOVERFLOW;
1439 * We need to debate whether we can enable this or not. The
1440 * man page documents EAGAIN return for the output at least,
1441 * and the application is arguably buggy if it doesn't expect
1442 * EAGAIN on a non-blocking file descriptor.
1444 if (in.file->f_flags & O_NONBLOCK)
1445 fl = SPLICE_F_NONBLOCK;
1447 file_start_write(out.file);
1448 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1449 file_end_write(out.file);
1452 add_rchar(current, retval);
1453 add_wchar(current, retval);
1454 fsnotify_access(in.file);
1455 fsnotify_modify(out.file);
1456 out.file->f_pos = out_pos;
1460 in.file->f_pos = pos;
1466 retval = -EOVERFLOW;
1476 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1483 if (unlikely(get_user(off, offset)))
1486 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1487 if (unlikely(put_user(pos, offset)))
1492 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1495 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1501 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1503 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1504 if (unlikely(put_user(pos, offset)))
1509 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1512 #ifdef CONFIG_COMPAT
1513 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1514 compat_off_t __user *, offset, compat_size_t, count)
1521 if (unlikely(get_user(off, offset)))
1524 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1525 if (unlikely(put_user(pos, offset)))
1530 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1533 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1534 compat_loff_t __user *, offset, compat_size_t, count)
1540 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1542 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1543 if (unlikely(put_user(pos, offset)))
1548 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1553 * copy_file_range() differs from regular file read and write in that it
1554 * specifically allows return partial success. When it does so is up to
1555 * the copy_file_range method.
1557 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1558 struct file *file_out, loff_t pos_out,
1559 size_t len, unsigned int flags)
1561 struct inode *inode_in = file_inode(file_in);
1562 struct inode *inode_out = file_inode(file_out);
1568 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1570 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1573 ret = rw_verify_area(READ, file_in, &pos_in, len);
1577 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1581 if (!(file_in->f_mode & FMODE_READ) ||
1582 !(file_out->f_mode & FMODE_WRITE) ||
1583 (file_out->f_flags & O_APPEND))
1586 /* this could be relaxed once a method supports cross-fs copies */
1587 if (inode_in->i_sb != inode_out->i_sb)
1593 file_start_write(file_out);
1596 * Try cloning first, this is supported by more file systems, and
1597 * more efficient if both clone and copy are supported (e.g. NFS).
1599 if (file_in->f_op->clone_file_range) {
1600 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1601 file_out, pos_out, len);
1608 if (file_out->f_op->copy_file_range) {
1609 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1610 pos_out, len, flags);
1611 if (ret != -EOPNOTSUPP)
1615 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1616 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1620 fsnotify_access(file_in);
1621 add_rchar(current, ret);
1622 fsnotify_modify(file_out);
1623 add_wchar(current, ret);
1629 file_end_write(file_out);
1633 EXPORT_SYMBOL(vfs_copy_file_range);
1635 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1636 int, fd_out, loff_t __user *, off_out,
1637 size_t, len, unsigned int, flags)
1643 ssize_t ret = -EBADF;
1645 f_in = fdget(fd_in);
1649 f_out = fdget(fd_out);
1655 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1658 pos_in = f_in.file->f_pos;
1662 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1665 pos_out = f_out.file->f_pos;
1668 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1675 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1678 f_in.file->f_pos = pos_in;
1682 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1685 f_out.file->f_pos = pos_out;
1697 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1699 struct inode *inode = file_inode(file);
1701 if (unlikely(pos < 0))
1704 if (unlikely((loff_t) (pos + len) < 0))
1707 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1708 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1711 retval = locks_mandatory_area(inode, file, pos, end,
1712 write ? F_WRLCK : F_RDLCK);
1717 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1721 * Check that the two inodes are eligible for cloning, the ranges make
1722 * sense, and then flush all dirty data. Caller must ensure that the
1723 * inodes have been locked against any other modifications.
1725 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1726 * the usual negative error code.
1728 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1729 struct inode *inode_out, loff_t pos_out,
1730 u64 *len, bool is_dedupe)
1732 loff_t bs = inode_out->i_sb->s_blocksize;
1735 bool same_inode = (inode_in == inode_out);
1738 /* Don't touch certain kinds of inodes */
1739 if (IS_IMMUTABLE(inode_out))
1742 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1745 /* Don't reflink dirs, pipes, sockets... */
1746 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1748 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1751 /* Are we going all the way to the end? */
1752 isize = i_size_read(inode_in);
1756 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1758 if (is_dedupe || pos_in == isize)
1762 *len = isize - pos_in;
1765 /* Ensure offsets don't wrap and the input is inside i_size */
1766 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1767 pos_in + *len > isize)
1770 /* Don't allow dedupe past EOF in the dest file */
1774 disize = i_size_read(inode_out);
1775 if (pos_out >= disize || pos_out + *len > disize)
1779 /* If we're linking to EOF, continue to the block boundary. */
1780 if (pos_in + *len == isize)
1781 blen = ALIGN(isize, bs) - pos_in;
1785 /* Only reflink if we're aligned to block boundaries */
1786 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1787 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1790 /* Don't allow overlapped reflink within the same file */
1792 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1796 /* Wait for the completion of any pending IOs on both files */
1797 inode_dio_wait(inode_in);
1799 inode_dio_wait(inode_out);
1801 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1802 pos_in, pos_in + *len - 1);
1806 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1807 pos_out, pos_out + *len - 1);
1812 * Check that the extents are the same.
1815 bool is_same = false;
1817 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1818 inode_out, pos_out, *len, &is_same);
1827 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1829 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1830 struct file *file_out, loff_t pos_out, u64 len)
1832 struct inode *inode_in = file_inode(file_in);
1833 struct inode *inode_out = file_inode(file_out);
1836 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1838 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1842 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1843 * the same mount. Practically, they only need to be on the same file
1846 if (inode_in->i_sb != inode_out->i_sb)
1849 if (!(file_in->f_mode & FMODE_READ) ||
1850 !(file_out->f_mode & FMODE_WRITE) ||
1851 (file_out->f_flags & O_APPEND))
1854 if (!file_in->f_op->clone_file_range)
1857 ret = clone_verify_area(file_in, pos_in, len, false);
1861 ret = clone_verify_area(file_out, pos_out, len, true);
1865 if (pos_in + len > i_size_read(inode_in))
1868 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1869 file_out, pos_out, len);
1871 fsnotify_access(file_in);
1872 fsnotify_modify(file_out);
1877 EXPORT_SYMBOL(vfs_clone_file_range);
1880 * Read a page's worth of file data into the page cache. Return the page
1883 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1885 struct address_space *mapping;
1889 n = offset >> PAGE_SHIFT;
1890 mapping = inode->i_mapping;
1891 page = read_mapping_page(mapping, n, NULL);
1894 if (!PageUptodate(page)) {
1896 return ERR_PTR(-EIO);
1903 * Compare extents of two files to see if they are the same.
1904 * Caller must have locked both inodes to prevent write races.
1906 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1907 struct inode *dest, loff_t destoff,
1908 loff_t len, bool *is_same)
1914 struct page *src_page;
1915 struct page *dest_page;
1923 src_poff = srcoff & (PAGE_SIZE - 1);
1924 dest_poff = destoff & (PAGE_SIZE - 1);
1925 cmp_len = min(PAGE_SIZE - src_poff,
1926 PAGE_SIZE - dest_poff);
1927 cmp_len = min(cmp_len, len);
1931 src_page = vfs_dedupe_get_page(src, srcoff);
1932 if (IS_ERR(src_page)) {
1933 error = PTR_ERR(src_page);
1936 dest_page = vfs_dedupe_get_page(dest, destoff);
1937 if (IS_ERR(dest_page)) {
1938 error = PTR_ERR(dest_page);
1939 unlock_page(src_page);
1943 src_addr = kmap_atomic(src_page);
1944 dest_addr = kmap_atomic(dest_page);
1946 flush_dcache_page(src_page);
1947 flush_dcache_page(dest_page);
1949 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1952 kunmap_atomic(dest_addr);
1953 kunmap_atomic(src_addr);
1954 unlock_page(dest_page);
1955 unlock_page(src_page);
1956 put_page(dest_page);
1973 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1975 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1977 struct file_dedupe_range_info *info;
1978 struct inode *src = file_inode(file);
1983 bool is_admin = capable(CAP_SYS_ADMIN);
1984 u16 count = same->dest_count;
1985 struct file *dst_file;
1989 if (!(file->f_mode & FMODE_READ))
1992 if (same->reserved1 || same->reserved2)
1995 off = same->src_offset;
1996 len = same->src_length;
1999 if (S_ISDIR(src->i_mode))
2003 if (!S_ISREG(src->i_mode))
2006 ret = clone_verify_area(file, off, len, false);
2011 if (off + len > i_size_read(src))
2014 /* pre-format output fields to sane values */
2015 for (i = 0; i < count; i++) {
2016 same->info[i].bytes_deduped = 0ULL;
2017 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2020 for (i = 0, info = same->info; i < count; i++, info++) {
2022 struct fd dst_fd = fdget(info->dest_fd);
2024 dst_file = dst_fd.file;
2026 info->status = -EBADF;
2029 dst = file_inode(dst_file);
2031 ret = mnt_want_write_file(dst_file);
2037 dst_off = info->dest_offset;
2038 ret = clone_verify_area(dst_file, dst_off, len, true);
2045 if (info->reserved) {
2046 info->status = -EINVAL;
2047 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2048 info->status = -EINVAL;
2049 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2050 info->status = -EXDEV;
2051 } else if (S_ISDIR(dst->i_mode)) {
2052 info->status = -EISDIR;
2053 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2054 info->status = -EINVAL;
2056 deduped = dst_file->f_op->dedupe_file_range(file, off,
2059 if (deduped == -EBADE)
2060 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2061 else if (deduped < 0)
2062 info->status = deduped;
2064 info->bytes_deduped += deduped;
2068 mnt_drop_write_file(dst_file);
2072 if (fatal_signal_pending(current))
2079 EXPORT_SYMBOL(vfs_dedupe_file_range);