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 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
363 int retval = -EINVAL;
365 inode = file_inode(file);
366 if (unlikely((ssize_t) count < 0))
369 if (unlikely(pos < 0)) {
370 if (!unsigned_offsets(file))
372 if (count >= -pos) /* both values are in 0..LLONG_MAX */
374 } else if (unlikely((loff_t) (pos + count) < 0)) {
375 if (!unsigned_offsets(file))
379 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
380 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
381 read_write == READ ? F_RDLCK : F_WRLCK);
385 return security_file_permission(file,
386 read_write == READ ? MAY_READ : MAY_WRITE);
389 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
391 struct iovec iov = { .iov_base = buf, .iov_len = len };
393 struct iov_iter iter;
396 init_sync_kiocb(&kiocb, filp);
397 kiocb.ki_pos = *ppos;
398 iov_iter_init(&iter, READ, &iov, 1, len);
400 ret = call_read_iter(filp, &kiocb, &iter);
401 BUG_ON(ret == -EIOCBQUEUED);
402 *ppos = kiocb.ki_pos;
406 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
409 if (file->f_op->read)
410 return file->f_op->read(file, buf, count, pos);
411 else if (file->f_op->read_iter)
412 return new_sync_read(file, buf, count, pos);
416 EXPORT_SYMBOL(__vfs_read);
418 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
422 if (!(file->f_mode & FMODE_READ))
424 if (!(file->f_mode & FMODE_CAN_READ))
426 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
429 ret = rw_verify_area(READ, file, pos, count);
431 if (count > MAX_RW_COUNT)
432 count = MAX_RW_COUNT;
433 ret = __vfs_read(file, buf, count, pos);
435 fsnotify_access(file);
436 add_rchar(current, ret);
444 EXPORT_SYMBOL(vfs_read);
446 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
448 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
450 struct iov_iter iter;
453 init_sync_kiocb(&kiocb, filp);
454 kiocb.ki_pos = *ppos;
455 iov_iter_init(&iter, WRITE, &iov, 1, len);
457 ret = call_write_iter(filp, &kiocb, &iter);
458 BUG_ON(ret == -EIOCBQUEUED);
460 *ppos = kiocb.ki_pos;
464 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
467 if (file->f_op->write)
468 return file->f_op->write(file, p, count, pos);
469 else if (file->f_op->write_iter)
470 return new_sync_write(file, p, count, pos);
474 EXPORT_SYMBOL(__vfs_write);
476 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
479 const char __user *p;
482 if (!(file->f_mode & FMODE_CAN_WRITE))
487 p = (__force const char __user *)buf;
488 if (count > MAX_RW_COUNT)
489 count = MAX_RW_COUNT;
490 ret = __vfs_write(file, p, count, pos);
493 fsnotify_modify(file);
494 add_wchar(current, ret);
500 EXPORT_SYMBOL(__kernel_write);
502 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
506 if (!(file->f_mode & FMODE_WRITE))
508 if (!(file->f_mode & FMODE_CAN_WRITE))
510 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
513 ret = rw_verify_area(WRITE, file, pos, count);
515 if (count > MAX_RW_COUNT)
516 count = MAX_RW_COUNT;
517 file_start_write(file);
518 ret = __vfs_write(file, buf, count, pos);
520 fsnotify_modify(file);
521 add_wchar(current, ret);
524 file_end_write(file);
530 EXPORT_SYMBOL(vfs_write);
532 static inline loff_t file_pos_read(struct file *file)
537 static inline void file_pos_write(struct file *file, loff_t pos)
542 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
544 struct fd f = fdget_pos(fd);
545 ssize_t ret = -EBADF;
548 loff_t pos = file_pos_read(f.file);
549 ret = vfs_read(f.file, buf, count, &pos);
551 file_pos_write(f.file, pos);
557 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
560 struct fd f = fdget_pos(fd);
561 ssize_t ret = -EBADF;
564 loff_t pos = file_pos_read(f.file);
565 ret = vfs_write(f.file, buf, count, &pos);
567 file_pos_write(f.file, pos);
574 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
575 size_t, count, loff_t, pos)
578 ssize_t ret = -EBADF;
586 if (f.file->f_mode & FMODE_PREAD)
587 ret = vfs_read(f.file, buf, count, &pos);
594 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
595 size_t, count, loff_t, pos)
598 ssize_t ret = -EBADF;
606 if (f.file->f_mode & FMODE_PWRITE)
607 ret = vfs_write(f.file, buf, count, &pos);
615 * Reduce an iovec's length in-place. Return the resulting number of segments
617 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
619 unsigned long seg = 0;
622 while (seg < nr_segs) {
624 if (len + iov->iov_len >= to) {
625 iov->iov_len = to - len;
633 EXPORT_SYMBOL(iov_shorten);
635 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
636 loff_t *ppos, int type, int flags)
641 init_sync_kiocb(&kiocb, filp);
642 ret = kiocb_set_rw_flags(&kiocb, flags);
645 kiocb.ki_pos = *ppos;
648 ret = call_read_iter(filp, &kiocb, iter);
650 ret = call_write_iter(filp, &kiocb, iter);
651 BUG_ON(ret == -EIOCBQUEUED);
652 *ppos = kiocb.ki_pos;
656 /* Do it by hand, with file-ops */
657 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
658 loff_t *ppos, int type, int flags)
662 if (flags & ~RWF_HIPRI)
665 while (iov_iter_count(iter)) {
666 struct iovec iovec = iov_iter_iovec(iter);
670 nr = filp->f_op->read(filp, iovec.iov_base,
671 iovec.iov_len, ppos);
673 nr = filp->f_op->write(filp, iovec.iov_base,
674 iovec.iov_len, ppos);
683 if (nr != iovec.iov_len)
685 iov_iter_advance(iter, nr);
691 /* A write operation does a read from user space and vice versa */
692 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
695 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
696 * into the kernel and check that it is valid.
698 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
699 * @uvector: Pointer to the userspace array.
700 * @nr_segs: Number of elements in userspace array.
701 * @fast_segs: Number of elements in @fast_pointer.
702 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
703 * @ret_pointer: (output parameter) Pointer to a variable that will point to
704 * either @fast_pointer, a newly allocated kernel array, or NULL,
705 * depending on which array was used.
707 * This function copies an array of &struct iovec of @nr_segs from
708 * userspace into the kernel and checks that each element is valid (e.g.
709 * it does not point to a kernel address or cause overflow by being too
712 * As an optimization, the caller may provide a pointer to a small
713 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
714 * (the size of this array, or 0 if unused, should be given in @fast_segs).
716 * @ret_pointer will always point to the array that was used, so the
717 * caller must take care not to call kfree() on it e.g. in case the
718 * @fast_pointer array was used and it was allocated on the stack.
720 * Return: The total number of bytes covered by the iovec array on success
721 * or a negative error code on error.
723 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
724 unsigned long nr_segs, unsigned long fast_segs,
725 struct iovec *fast_pointer,
726 struct iovec **ret_pointer)
730 struct iovec *iov = fast_pointer;
733 * SuS says "The readv() function *may* fail if the iovcnt argument
734 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
735 * traditionally returned zero for zero segments, so...
743 * First get the "struct iovec" from user memory and
744 * verify all the pointers
746 if (nr_segs > UIO_MAXIOV) {
750 if (nr_segs > fast_segs) {
751 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
757 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
763 * According to the Single Unix Specification we should return EINVAL
764 * if an element length is < 0 when cast to ssize_t or if the
765 * total length would overflow the ssize_t return value of the
768 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
772 for (seg = 0; seg < nr_segs; seg++) {
773 void __user *buf = iov[seg].iov_base;
774 ssize_t len = (ssize_t)iov[seg].iov_len;
776 /* see if we we're about to use an invalid len or if
777 * it's about to overflow ssize_t */
783 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
787 if (len > MAX_RW_COUNT - ret) {
788 len = MAX_RW_COUNT - ret;
789 iov[seg].iov_len = len;
799 ssize_t compat_rw_copy_check_uvector(int type,
800 const struct compat_iovec __user *uvector, unsigned long nr_segs,
801 unsigned long fast_segs, struct iovec *fast_pointer,
802 struct iovec **ret_pointer)
804 compat_ssize_t tot_len;
805 struct iovec *iov = *ret_pointer = fast_pointer;
810 * SuS says "The readv() function *may* fail if the iovcnt argument
811 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
812 * traditionally returned zero for zero segments, so...
818 if (nr_segs > UIO_MAXIOV)
820 if (nr_segs > fast_segs) {
822 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
829 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
833 * Single unix specification:
834 * We should -EINVAL if an element length is not >= 0 and fitting an
837 * In Linux, the total length is limited to MAX_RW_COUNT, there is
838 * no overflow possibility.
842 for (seg = 0; seg < nr_segs; seg++) {
846 if (__get_user(len, &uvector->iov_len) ||
847 __get_user(buf, &uvector->iov_base)) {
851 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
854 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
858 if (len > MAX_RW_COUNT - tot_len)
859 len = MAX_RW_COUNT - tot_len;
861 iov->iov_base = compat_ptr(buf);
862 iov->iov_len = (compat_size_t) len;
873 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
874 loff_t *pos, int flags)
879 if (!(file->f_mode & FMODE_READ))
881 if (!(file->f_mode & FMODE_CAN_READ))
884 tot_len = iov_iter_count(iter);
887 ret = rw_verify_area(READ, file, pos, tot_len);
891 if (file->f_op->read_iter)
892 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
894 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
897 fsnotify_access(file);
901 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
904 if (!file->f_op->read_iter)
906 return do_iter_read(file, iter, ppos, flags);
908 EXPORT_SYMBOL(vfs_iter_read);
910 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
911 loff_t *pos, int flags)
916 if (!(file->f_mode & FMODE_WRITE))
918 if (!(file->f_mode & FMODE_CAN_WRITE))
921 tot_len = iov_iter_count(iter);
924 ret = rw_verify_area(WRITE, file, pos, tot_len);
928 if (file->f_op->write_iter)
929 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
931 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
933 fsnotify_modify(file);
937 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
940 if (!file->f_op->write_iter)
942 return do_iter_write(file, iter, ppos, flags);
944 EXPORT_SYMBOL(vfs_iter_write);
946 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
947 unsigned long vlen, loff_t *pos, int flags)
949 struct iovec iovstack[UIO_FASTIOV];
950 struct iovec *iov = iovstack;
951 struct iov_iter iter;
954 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
956 ret = do_iter_read(file, &iter, pos, flags);
962 EXPORT_SYMBOL(vfs_readv);
964 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
965 unsigned long vlen, loff_t *pos, int flags)
967 struct iovec iovstack[UIO_FASTIOV];
968 struct iovec *iov = iovstack;
969 struct iov_iter iter;
972 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
974 file_start_write(file);
975 ret = do_iter_write(file, &iter, pos, flags);
976 file_end_write(file);
981 EXPORT_SYMBOL(vfs_writev);
983 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
984 unsigned long vlen, int flags)
986 struct fd f = fdget_pos(fd);
987 ssize_t ret = -EBADF;
990 loff_t pos = file_pos_read(f.file);
991 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
993 file_pos_write(f.file, pos);
998 add_rchar(current, ret);
1003 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1004 unsigned long vlen, int flags)
1006 struct fd f = fdget_pos(fd);
1007 ssize_t ret = -EBADF;
1010 loff_t pos = file_pos_read(f.file);
1011 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1013 file_pos_write(f.file, pos);
1018 add_wchar(current, ret);
1023 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1025 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1026 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1029 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1030 unsigned long vlen, loff_t pos, int flags)
1033 ssize_t ret = -EBADF;
1041 if (f.file->f_mode & FMODE_PREAD)
1042 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1047 add_rchar(current, ret);
1052 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1053 unsigned long vlen, loff_t pos, int flags)
1056 ssize_t ret = -EBADF;
1064 if (f.file->f_mode & FMODE_PWRITE)
1065 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1070 add_wchar(current, ret);
1075 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1076 unsigned long, vlen)
1078 return do_readv(fd, vec, vlen, 0);
1081 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1082 unsigned long, vlen)
1084 return do_writev(fd, vec, vlen, 0);
1087 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1088 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1090 loff_t pos = pos_from_hilo(pos_h, pos_l);
1092 return do_preadv(fd, vec, vlen, pos, 0);
1095 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1096 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1099 loff_t pos = pos_from_hilo(pos_h, pos_l);
1102 return do_readv(fd, vec, vlen, flags);
1104 return do_preadv(fd, vec, vlen, pos, flags);
1107 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1108 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1110 loff_t pos = pos_from_hilo(pos_h, pos_l);
1112 return do_pwritev(fd, vec, vlen, pos, 0);
1115 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1116 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1119 loff_t pos = pos_from_hilo(pos_h, pos_l);
1122 return do_writev(fd, vec, vlen, flags);
1124 return do_pwritev(fd, vec, vlen, pos, flags);
1127 #ifdef CONFIG_COMPAT
1128 static size_t compat_readv(struct file *file,
1129 const struct compat_iovec __user *vec,
1130 unsigned long vlen, loff_t *pos, int flags)
1132 struct iovec iovstack[UIO_FASTIOV];
1133 struct iovec *iov = iovstack;
1134 struct iov_iter iter;
1137 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1139 ret = do_iter_read(file, &iter, pos, flags);
1143 add_rchar(current, ret);
1148 static size_t do_compat_readv(compat_ulong_t fd,
1149 const struct compat_iovec __user *vec,
1150 compat_ulong_t vlen, int flags)
1152 struct fd f = fdget_pos(fd);
1158 pos = f.file->f_pos;
1159 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1161 f.file->f_pos = pos;
1167 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1168 const struct compat_iovec __user *,vec,
1169 compat_ulong_t, vlen)
1171 return do_compat_readv(fd, vec, vlen, 0);
1174 static long do_compat_preadv64(unsigned long fd,
1175 const struct compat_iovec __user *vec,
1176 unsigned long vlen, loff_t pos, int flags)
1187 if (f.file->f_mode & FMODE_PREAD)
1188 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1193 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1194 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1195 const struct compat_iovec __user *,vec,
1196 unsigned long, vlen, loff_t, pos)
1198 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1202 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1203 const struct compat_iovec __user *,vec,
1204 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1206 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1208 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1211 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1212 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1213 const struct compat_iovec __user *,vec,
1214 unsigned long, vlen, loff_t, pos, int, flags)
1216 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1220 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1221 const struct compat_iovec __user *,vec,
1222 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1225 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1228 return do_compat_readv(fd, vec, vlen, flags);
1230 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1233 static size_t compat_writev(struct file *file,
1234 const struct compat_iovec __user *vec,
1235 unsigned long vlen, loff_t *pos, int flags)
1237 struct iovec iovstack[UIO_FASTIOV];
1238 struct iovec *iov = iovstack;
1239 struct iov_iter iter;
1242 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1244 file_start_write(file);
1245 ret = do_iter_write(file, &iter, pos, flags);
1246 file_end_write(file);
1250 add_wchar(current, ret);
1255 static size_t do_compat_writev(compat_ulong_t fd,
1256 const struct compat_iovec __user* vec,
1257 compat_ulong_t vlen, int flags)
1259 struct fd f = fdget_pos(fd);
1265 pos = f.file->f_pos;
1266 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1268 f.file->f_pos = pos;
1273 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1274 const struct compat_iovec __user *, vec,
1275 compat_ulong_t, vlen)
1277 return do_compat_writev(fd, vec, vlen, 0);
1280 static long do_compat_pwritev64(unsigned long fd,
1281 const struct compat_iovec __user *vec,
1282 unsigned long vlen, loff_t pos, int flags)
1293 if (f.file->f_mode & FMODE_PWRITE)
1294 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1299 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1300 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1301 const struct compat_iovec __user *,vec,
1302 unsigned long, vlen, loff_t, pos)
1304 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1308 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1309 const struct compat_iovec __user *,vec,
1310 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1312 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1314 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1317 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1318 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1319 const struct compat_iovec __user *,vec,
1320 unsigned long, vlen, loff_t, pos, int, flags)
1322 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1326 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1327 const struct compat_iovec __user *,vec,
1328 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1330 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1333 return do_compat_writev(fd, vec, vlen, flags);
1335 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1340 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1341 size_t count, loff_t max)
1344 struct inode *in_inode, *out_inode;
1351 * Get input file, and verify that it is ok..
1357 if (!(in.file->f_mode & FMODE_READ))
1361 pos = in.file->f_pos;
1364 if (!(in.file->f_mode & FMODE_PREAD))
1367 retval = rw_verify_area(READ, in.file, &pos, count);
1370 if (count > MAX_RW_COUNT)
1371 count = MAX_RW_COUNT;
1374 * Get output file, and verify that it is ok..
1377 out = fdget(out_fd);
1380 if (!(out.file->f_mode & FMODE_WRITE))
1383 in_inode = file_inode(in.file);
1384 out_inode = file_inode(out.file);
1385 out_pos = out.file->f_pos;
1386 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1391 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1393 if (unlikely(pos + count > max)) {
1394 retval = -EOVERFLOW;
1403 * We need to debate whether we can enable this or not. The
1404 * man page documents EAGAIN return for the output at least,
1405 * and the application is arguably buggy if it doesn't expect
1406 * EAGAIN on a non-blocking file descriptor.
1408 if (in.file->f_flags & O_NONBLOCK)
1409 fl = SPLICE_F_NONBLOCK;
1411 file_start_write(out.file);
1412 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1413 file_end_write(out.file);
1416 add_rchar(current, retval);
1417 add_wchar(current, retval);
1418 fsnotify_access(in.file);
1419 fsnotify_modify(out.file);
1420 out.file->f_pos = out_pos;
1424 in.file->f_pos = pos;
1430 retval = -EOVERFLOW;
1440 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1447 if (unlikely(get_user(off, offset)))
1450 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1451 if (unlikely(put_user(pos, offset)))
1456 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1459 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1465 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1467 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1468 if (unlikely(put_user(pos, offset)))
1473 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1476 #ifdef CONFIG_COMPAT
1477 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1478 compat_off_t __user *, offset, compat_size_t, count)
1485 if (unlikely(get_user(off, offset)))
1488 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1489 if (unlikely(put_user(pos, offset)))
1494 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1497 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1498 compat_loff_t __user *, offset, compat_size_t, count)
1504 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1506 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1507 if (unlikely(put_user(pos, offset)))
1512 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1517 * copy_file_range() differs from regular file read and write in that it
1518 * specifically allows return partial success. When it does so is up to
1519 * the copy_file_range method.
1521 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1522 struct file *file_out, loff_t pos_out,
1523 size_t len, unsigned int flags)
1525 struct inode *inode_in = file_inode(file_in);
1526 struct inode *inode_out = file_inode(file_out);
1532 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1534 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1537 ret = rw_verify_area(READ, file_in, &pos_in, len);
1541 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1545 if (!(file_in->f_mode & FMODE_READ) ||
1546 !(file_out->f_mode & FMODE_WRITE) ||
1547 (file_out->f_flags & O_APPEND))
1550 /* this could be relaxed once a method supports cross-fs copies */
1551 if (inode_in->i_sb != inode_out->i_sb)
1557 file_start_write(file_out);
1560 * Try cloning first, this is supported by more file systems, and
1561 * more efficient if both clone and copy are supported (e.g. NFS).
1563 if (file_in->f_op->clone_file_range) {
1564 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1565 file_out, pos_out, len);
1572 if (file_out->f_op->copy_file_range) {
1573 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1574 pos_out, len, flags);
1575 if (ret != -EOPNOTSUPP)
1579 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1580 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1584 fsnotify_access(file_in);
1585 add_rchar(current, ret);
1586 fsnotify_modify(file_out);
1587 add_wchar(current, ret);
1593 file_end_write(file_out);
1597 EXPORT_SYMBOL(vfs_copy_file_range);
1599 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1600 int, fd_out, loff_t __user *, off_out,
1601 size_t, len, unsigned int, flags)
1607 ssize_t ret = -EBADF;
1609 f_in = fdget(fd_in);
1613 f_out = fdget(fd_out);
1619 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1622 pos_in = f_in.file->f_pos;
1626 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1629 pos_out = f_out.file->f_pos;
1632 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1639 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1642 f_in.file->f_pos = pos_in;
1646 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1649 f_out.file->f_pos = pos_out;
1661 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1663 struct inode *inode = file_inode(file);
1665 if (unlikely(pos < 0))
1668 if (unlikely((loff_t) (pos + len) < 0))
1671 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1672 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1675 retval = locks_mandatory_area(inode, file, pos, end,
1676 write ? F_WRLCK : F_RDLCK);
1681 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1685 * Check that the two inodes are eligible for cloning, the ranges make
1686 * sense, and then flush all dirty data. Caller must ensure that the
1687 * inodes have been locked against any other modifications.
1689 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1690 * the usual negative error code.
1692 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1693 struct inode *inode_out, loff_t pos_out,
1694 u64 *len, bool is_dedupe)
1696 loff_t bs = inode_out->i_sb->s_blocksize;
1699 bool same_inode = (inode_in == inode_out);
1702 /* Don't touch certain kinds of inodes */
1703 if (IS_IMMUTABLE(inode_out))
1706 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1709 /* Don't reflink dirs, pipes, sockets... */
1710 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1712 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1715 /* Are we going all the way to the end? */
1716 isize = i_size_read(inode_in);
1720 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1722 if (is_dedupe || pos_in == isize)
1726 *len = isize - pos_in;
1729 /* Ensure offsets don't wrap and the input is inside i_size */
1730 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1731 pos_in + *len > isize)
1734 /* Don't allow dedupe past EOF in the dest file */
1738 disize = i_size_read(inode_out);
1739 if (pos_out >= disize || pos_out + *len > disize)
1743 /* If we're linking to EOF, continue to the block boundary. */
1744 if (pos_in + *len == isize)
1745 blen = ALIGN(isize, bs) - pos_in;
1749 /* Only reflink if we're aligned to block boundaries */
1750 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1751 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1754 /* Don't allow overlapped reflink within the same file */
1756 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1760 /* Wait for the completion of any pending IOs on both files */
1761 inode_dio_wait(inode_in);
1763 inode_dio_wait(inode_out);
1765 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1766 pos_in, pos_in + *len - 1);
1770 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1771 pos_out, pos_out + *len - 1);
1776 * Check that the extents are the same.
1779 bool is_same = false;
1781 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1782 inode_out, pos_out, *len, &is_same);
1791 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1793 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1794 struct file *file_out, loff_t pos_out, u64 len)
1796 struct inode *inode_in = file_inode(file_in);
1797 struct inode *inode_out = file_inode(file_out);
1800 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1802 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1806 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1807 * the same mount. Practically, they only need to be on the same file
1810 if (inode_in->i_sb != inode_out->i_sb)
1813 if (!(file_in->f_mode & FMODE_READ) ||
1814 !(file_out->f_mode & FMODE_WRITE) ||
1815 (file_out->f_flags & O_APPEND))
1818 if (!file_in->f_op->clone_file_range)
1821 ret = clone_verify_area(file_in, pos_in, len, false);
1825 ret = clone_verify_area(file_out, pos_out, len, true);
1829 if (pos_in + len > i_size_read(inode_in))
1832 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1833 file_out, pos_out, len);
1835 fsnotify_access(file_in);
1836 fsnotify_modify(file_out);
1841 EXPORT_SYMBOL(vfs_clone_file_range);
1844 * Read a page's worth of file data into the page cache. Return the page
1847 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1849 struct address_space *mapping;
1853 n = offset >> PAGE_SHIFT;
1854 mapping = inode->i_mapping;
1855 page = read_mapping_page(mapping, n, NULL);
1858 if (!PageUptodate(page)) {
1860 return ERR_PTR(-EIO);
1867 * Compare extents of two files to see if they are the same.
1868 * Caller must have locked both inodes to prevent write races.
1870 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1871 struct inode *dest, loff_t destoff,
1872 loff_t len, bool *is_same)
1878 struct page *src_page;
1879 struct page *dest_page;
1887 src_poff = srcoff & (PAGE_SIZE - 1);
1888 dest_poff = destoff & (PAGE_SIZE - 1);
1889 cmp_len = min(PAGE_SIZE - src_poff,
1890 PAGE_SIZE - dest_poff);
1891 cmp_len = min(cmp_len, len);
1895 src_page = vfs_dedupe_get_page(src, srcoff);
1896 if (IS_ERR(src_page)) {
1897 error = PTR_ERR(src_page);
1900 dest_page = vfs_dedupe_get_page(dest, destoff);
1901 if (IS_ERR(dest_page)) {
1902 error = PTR_ERR(dest_page);
1903 unlock_page(src_page);
1907 src_addr = kmap_atomic(src_page);
1908 dest_addr = kmap_atomic(dest_page);
1910 flush_dcache_page(src_page);
1911 flush_dcache_page(dest_page);
1913 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1916 kunmap_atomic(dest_addr);
1917 kunmap_atomic(src_addr);
1918 unlock_page(dest_page);
1919 unlock_page(src_page);
1920 put_page(dest_page);
1937 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1939 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1941 struct file_dedupe_range_info *info;
1942 struct inode *src = file_inode(file);
1947 bool is_admin = capable(CAP_SYS_ADMIN);
1948 u16 count = same->dest_count;
1949 struct file *dst_file;
1953 if (!(file->f_mode & FMODE_READ))
1956 if (same->reserved1 || same->reserved2)
1959 off = same->src_offset;
1960 len = same->src_length;
1963 if (S_ISDIR(src->i_mode))
1967 if (!S_ISREG(src->i_mode))
1970 ret = clone_verify_area(file, off, len, false);
1975 if (off + len > i_size_read(src))
1978 /* pre-format output fields to sane values */
1979 for (i = 0; i < count; i++) {
1980 same->info[i].bytes_deduped = 0ULL;
1981 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1984 for (i = 0, info = same->info; i < count; i++, info++) {
1986 struct fd dst_fd = fdget(info->dest_fd);
1988 dst_file = dst_fd.file;
1990 info->status = -EBADF;
1993 dst = file_inode(dst_file);
1995 ret = mnt_want_write_file(dst_file);
2001 dst_off = info->dest_offset;
2002 ret = clone_verify_area(dst_file, dst_off, len, true);
2009 if (info->reserved) {
2010 info->status = -EINVAL;
2011 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2012 info->status = -EINVAL;
2013 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2014 info->status = -EXDEV;
2015 } else if (S_ISDIR(dst->i_mode)) {
2016 info->status = -EISDIR;
2017 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2018 info->status = -EINVAL;
2020 deduped = dst_file->f_op->dedupe_file_range(file, off,
2023 if (deduped == -EBADE)
2024 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2025 else if (deduped < 0)
2026 info->status = deduped;
2028 info->bytes_deduped += deduped;
2032 mnt_drop_write_file(dst_file);
2036 if (fatal_signal_pending(current))
2043 EXPORT_SYMBOL(vfs_dedupe_file_range);