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 bool 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.
115 if ((unsigned long long)offset >= eof)
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.
123 if ((unsigned long long)offset >= eof)
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);
417 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
424 /* The cast to a user pointer is valid due to the set_fs() */
425 result = vfs_read(file, (void __user *)buf, count, pos);
429 EXPORT_SYMBOL(kernel_read);
431 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
435 if (!(file->f_mode & FMODE_READ))
437 if (!(file->f_mode & FMODE_CAN_READ))
439 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
442 ret = rw_verify_area(READ, file, pos, count);
444 if (count > MAX_RW_COUNT)
445 count = MAX_RW_COUNT;
446 ret = __vfs_read(file, buf, count, pos);
448 fsnotify_access(file);
449 add_rchar(current, ret);
457 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
459 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
461 struct iov_iter iter;
464 init_sync_kiocb(&kiocb, filp);
465 kiocb.ki_pos = *ppos;
466 iov_iter_init(&iter, WRITE, &iov, 1, len);
468 ret = call_write_iter(filp, &kiocb, &iter);
469 BUG_ON(ret == -EIOCBQUEUED);
471 *ppos = kiocb.ki_pos;
475 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
478 if (file->f_op->write)
479 return file->f_op->write(file, p, count, pos);
480 else if (file->f_op->write_iter)
481 return new_sync_write(file, p, count, pos);
486 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
489 const char __user *p;
492 if (!(file->f_mode & FMODE_CAN_WRITE))
497 p = (__force const char __user *)buf;
498 if (count > MAX_RW_COUNT)
499 count = MAX_RW_COUNT;
500 ret = __vfs_write(file, p, count, pos);
503 fsnotify_modify(file);
504 add_wchar(current, ret);
509 EXPORT_SYMBOL(__kernel_write);
511 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
519 /* The cast to a user pointer is valid due to the set_fs() */
520 res = vfs_write(file, (__force const char __user *)buf, count, pos);
525 EXPORT_SYMBOL(kernel_write);
527 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
531 if (!(file->f_mode & FMODE_WRITE))
533 if (!(file->f_mode & FMODE_CAN_WRITE))
535 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
538 ret = rw_verify_area(WRITE, file, pos, count);
540 if (count > MAX_RW_COUNT)
541 count = MAX_RW_COUNT;
542 file_start_write(file);
543 ret = __vfs_write(file, buf, count, pos);
545 fsnotify_modify(file);
546 add_wchar(current, ret);
549 file_end_write(file);
555 static inline loff_t file_pos_read(struct file *file)
560 static inline void file_pos_write(struct file *file, loff_t pos)
565 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
567 struct fd f = fdget_pos(fd);
568 ssize_t ret = -EBADF;
571 loff_t pos = file_pos_read(f.file);
572 ret = vfs_read(f.file, buf, count, &pos);
574 file_pos_write(f.file, pos);
580 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
583 struct fd f = fdget_pos(fd);
584 ssize_t ret = -EBADF;
587 loff_t pos = file_pos_read(f.file);
588 ret = vfs_write(f.file, buf, count, &pos);
590 file_pos_write(f.file, pos);
597 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
598 size_t, count, loff_t, pos)
601 ssize_t ret = -EBADF;
609 if (f.file->f_mode & FMODE_PREAD)
610 ret = vfs_read(f.file, buf, count, &pos);
617 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
618 size_t, count, loff_t, pos)
621 ssize_t ret = -EBADF;
629 if (f.file->f_mode & FMODE_PWRITE)
630 ret = vfs_write(f.file, buf, count, &pos);
638 * Reduce an iovec's length in-place. Return the resulting number of segments
640 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
642 unsigned long seg = 0;
645 while (seg < nr_segs) {
647 if (len + iov->iov_len >= to) {
648 iov->iov_len = to - len;
656 EXPORT_SYMBOL(iov_shorten);
658 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
659 loff_t *ppos, int type, rwf_t flags)
664 init_sync_kiocb(&kiocb, filp);
665 ret = kiocb_set_rw_flags(&kiocb, flags);
668 kiocb.ki_pos = *ppos;
671 ret = call_read_iter(filp, &kiocb, iter);
673 ret = call_write_iter(filp, &kiocb, iter);
674 BUG_ON(ret == -EIOCBQUEUED);
675 *ppos = kiocb.ki_pos;
679 /* Do it by hand, with file-ops */
680 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
681 loff_t *ppos, int type, rwf_t flags)
685 if (flags & ~RWF_HIPRI)
688 while (iov_iter_count(iter)) {
689 struct iovec iovec = iov_iter_iovec(iter);
693 nr = filp->f_op->read(filp, iovec.iov_base,
694 iovec.iov_len, ppos);
696 nr = filp->f_op->write(filp, iovec.iov_base,
697 iovec.iov_len, ppos);
706 if (nr != iovec.iov_len)
708 iov_iter_advance(iter, nr);
714 /* A write operation does a read from user space and vice versa */
715 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
718 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
719 * into the kernel and check that it is valid.
721 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
722 * @uvector: Pointer to the userspace array.
723 * @nr_segs: Number of elements in userspace array.
724 * @fast_segs: Number of elements in @fast_pointer.
725 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
726 * @ret_pointer: (output parameter) Pointer to a variable that will point to
727 * either @fast_pointer, a newly allocated kernel array, or NULL,
728 * depending on which array was used.
730 * This function copies an array of &struct iovec of @nr_segs from
731 * userspace into the kernel and checks that each element is valid (e.g.
732 * it does not point to a kernel address or cause overflow by being too
735 * As an optimization, the caller may provide a pointer to a small
736 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
737 * (the size of this array, or 0 if unused, should be given in @fast_segs).
739 * @ret_pointer will always point to the array that was used, so the
740 * caller must take care not to call kfree() on it e.g. in case the
741 * @fast_pointer array was used and it was allocated on the stack.
743 * Return: The total number of bytes covered by the iovec array on success
744 * or a negative error code on error.
746 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
747 unsigned long nr_segs, unsigned long fast_segs,
748 struct iovec *fast_pointer,
749 struct iovec **ret_pointer)
753 struct iovec *iov = fast_pointer;
756 * SuS says "The readv() function *may* fail if the iovcnt argument
757 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
758 * traditionally returned zero for zero segments, so...
766 * First get the "struct iovec" from user memory and
767 * verify all the pointers
769 if (nr_segs > UIO_MAXIOV) {
773 if (nr_segs > fast_segs) {
774 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
780 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
786 * According to the Single Unix Specification we should return EINVAL
787 * if an element length is < 0 when cast to ssize_t or if the
788 * total length would overflow the ssize_t return value of the
791 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
795 for (seg = 0; seg < nr_segs; seg++) {
796 void __user *buf = iov[seg].iov_base;
797 ssize_t len = (ssize_t)iov[seg].iov_len;
799 /* see if we we're about to use an invalid len or if
800 * it's about to overflow ssize_t */
806 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
810 if (len > MAX_RW_COUNT - ret) {
811 len = MAX_RW_COUNT - ret;
812 iov[seg].iov_len = len;
822 ssize_t compat_rw_copy_check_uvector(int type,
823 const struct compat_iovec __user *uvector, unsigned long nr_segs,
824 unsigned long fast_segs, struct iovec *fast_pointer,
825 struct iovec **ret_pointer)
827 compat_ssize_t tot_len;
828 struct iovec *iov = *ret_pointer = fast_pointer;
833 * SuS says "The readv() function *may* fail if the iovcnt argument
834 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
835 * traditionally returned zero for zero segments, so...
841 if (nr_segs > UIO_MAXIOV)
843 if (nr_segs > fast_segs) {
845 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
852 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
856 * Single unix specification:
857 * We should -EINVAL if an element length is not >= 0 and fitting an
860 * In Linux, the total length is limited to MAX_RW_COUNT, there is
861 * no overflow possibility.
865 for (seg = 0; seg < nr_segs; seg++) {
869 if (__get_user(len, &uvector->iov_len) ||
870 __get_user(buf, &uvector->iov_base)) {
874 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
877 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
881 if (len > MAX_RW_COUNT - tot_len)
882 len = MAX_RW_COUNT - tot_len;
884 iov->iov_base = compat_ptr(buf);
885 iov->iov_len = (compat_size_t) len;
896 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
897 loff_t *pos, rwf_t flags)
902 if (!(file->f_mode & FMODE_READ))
904 if (!(file->f_mode & FMODE_CAN_READ))
907 tot_len = iov_iter_count(iter);
910 ret = rw_verify_area(READ, file, pos, tot_len);
914 if (file->f_op->read_iter)
915 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
917 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
920 fsnotify_access(file);
924 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
927 if (!file->f_op->read_iter)
929 return do_iter_read(file, iter, ppos, flags);
931 EXPORT_SYMBOL(vfs_iter_read);
933 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
934 loff_t *pos, rwf_t flags)
939 if (!(file->f_mode & FMODE_WRITE))
941 if (!(file->f_mode & FMODE_CAN_WRITE))
944 tot_len = iov_iter_count(iter);
947 ret = rw_verify_area(WRITE, file, pos, tot_len);
951 if (file->f_op->write_iter)
952 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
954 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
956 fsnotify_modify(file);
960 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
963 if (!file->f_op->write_iter)
965 return do_iter_write(file, iter, ppos, flags);
967 EXPORT_SYMBOL(vfs_iter_write);
969 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
970 unsigned long vlen, loff_t *pos, rwf_t flags)
972 struct iovec iovstack[UIO_FASTIOV];
973 struct iovec *iov = iovstack;
974 struct iov_iter iter;
977 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
979 ret = do_iter_read(file, &iter, pos, flags);
986 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
987 unsigned long vlen, loff_t *pos, rwf_t flags)
989 struct iovec iovstack[UIO_FASTIOV];
990 struct iovec *iov = iovstack;
991 struct iov_iter iter;
994 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
996 file_start_write(file);
997 ret = do_iter_write(file, &iter, pos, flags);
998 file_end_write(file);
1004 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1005 unsigned long vlen, rwf_t flags)
1007 struct fd f = fdget_pos(fd);
1008 ssize_t ret = -EBADF;
1011 loff_t pos = file_pos_read(f.file);
1012 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1014 file_pos_write(f.file, pos);
1019 add_rchar(current, ret);
1024 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1025 unsigned long vlen, rwf_t flags)
1027 struct fd f = fdget_pos(fd);
1028 ssize_t ret = -EBADF;
1031 loff_t pos = file_pos_read(f.file);
1032 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1034 file_pos_write(f.file, pos);
1039 add_wchar(current, ret);
1044 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1046 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1047 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1050 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1051 unsigned long vlen, loff_t pos, rwf_t flags)
1054 ssize_t ret = -EBADF;
1062 if (f.file->f_mode & FMODE_PREAD)
1063 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1068 add_rchar(current, ret);
1073 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1074 unsigned long vlen, loff_t pos, rwf_t flags)
1077 ssize_t ret = -EBADF;
1085 if (f.file->f_mode & FMODE_PWRITE)
1086 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1091 add_wchar(current, ret);
1096 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1097 unsigned long, vlen)
1099 return do_readv(fd, vec, vlen, 0);
1102 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1103 unsigned long, vlen)
1105 return do_writev(fd, vec, vlen, 0);
1108 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1109 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1111 loff_t pos = pos_from_hilo(pos_h, pos_l);
1113 return do_preadv(fd, vec, vlen, pos, 0);
1116 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1117 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1120 loff_t pos = pos_from_hilo(pos_h, pos_l);
1123 return do_readv(fd, vec, vlen, flags);
1125 return do_preadv(fd, vec, vlen, pos, flags);
1128 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1129 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1131 loff_t pos = pos_from_hilo(pos_h, pos_l);
1133 return do_pwritev(fd, vec, vlen, pos, 0);
1136 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1137 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1140 loff_t pos = pos_from_hilo(pos_h, pos_l);
1143 return do_writev(fd, vec, vlen, flags);
1145 return do_pwritev(fd, vec, vlen, pos, flags);
1148 #ifdef CONFIG_COMPAT
1149 static size_t compat_readv(struct file *file,
1150 const struct compat_iovec __user *vec,
1151 unsigned long vlen, loff_t *pos, rwf_t flags)
1153 struct iovec iovstack[UIO_FASTIOV];
1154 struct iovec *iov = iovstack;
1155 struct iov_iter iter;
1158 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1160 ret = do_iter_read(file, &iter, pos, flags);
1164 add_rchar(current, ret);
1169 static size_t do_compat_readv(compat_ulong_t fd,
1170 const struct compat_iovec __user *vec,
1171 compat_ulong_t vlen, rwf_t flags)
1173 struct fd f = fdget_pos(fd);
1179 pos = f.file->f_pos;
1180 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1182 f.file->f_pos = pos;
1188 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1189 const struct compat_iovec __user *,vec,
1190 compat_ulong_t, vlen)
1192 return do_compat_readv(fd, vec, vlen, 0);
1195 static long do_compat_preadv64(unsigned long fd,
1196 const struct compat_iovec __user *vec,
1197 unsigned long vlen, loff_t pos, rwf_t flags)
1208 if (f.file->f_mode & FMODE_PREAD)
1209 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1214 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1215 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1216 const struct compat_iovec __user *,vec,
1217 unsigned long, vlen, loff_t, pos)
1219 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1223 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1224 const struct compat_iovec __user *,vec,
1225 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1227 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1229 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1232 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1233 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1234 const struct compat_iovec __user *,vec,
1235 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1237 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1241 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1242 const struct compat_iovec __user *,vec,
1243 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1246 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1249 return do_compat_readv(fd, vec, vlen, flags);
1251 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1254 static size_t compat_writev(struct file *file,
1255 const struct compat_iovec __user *vec,
1256 unsigned long vlen, loff_t *pos, rwf_t flags)
1258 struct iovec iovstack[UIO_FASTIOV];
1259 struct iovec *iov = iovstack;
1260 struct iov_iter iter;
1263 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1265 file_start_write(file);
1266 ret = do_iter_write(file, &iter, pos, flags);
1267 file_end_write(file);
1271 add_wchar(current, ret);
1276 static size_t do_compat_writev(compat_ulong_t fd,
1277 const struct compat_iovec __user* vec,
1278 compat_ulong_t vlen, rwf_t flags)
1280 struct fd f = fdget_pos(fd);
1286 pos = f.file->f_pos;
1287 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1289 f.file->f_pos = pos;
1294 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1295 const struct compat_iovec __user *, vec,
1296 compat_ulong_t, vlen)
1298 return do_compat_writev(fd, vec, vlen, 0);
1301 static long do_compat_pwritev64(unsigned long fd,
1302 const struct compat_iovec __user *vec,
1303 unsigned long vlen, loff_t pos, rwf_t flags)
1314 if (f.file->f_mode & FMODE_PWRITE)
1315 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1320 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1321 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1322 const struct compat_iovec __user *,vec,
1323 unsigned long, vlen, loff_t, pos)
1325 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1329 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1330 const struct compat_iovec __user *,vec,
1331 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1333 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1335 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1338 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1339 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1340 const struct compat_iovec __user *,vec,
1341 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1343 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1347 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1348 const struct compat_iovec __user *,vec,
1349 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1351 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1354 return do_compat_writev(fd, vec, vlen, flags);
1356 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1361 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1362 size_t count, loff_t max)
1365 struct inode *in_inode, *out_inode;
1372 * Get input file, and verify that it is ok..
1378 if (!(in.file->f_mode & FMODE_READ))
1382 pos = in.file->f_pos;
1385 if (!(in.file->f_mode & FMODE_PREAD))
1388 retval = rw_verify_area(READ, in.file, &pos, count);
1391 if (count > MAX_RW_COUNT)
1392 count = MAX_RW_COUNT;
1395 * Get output file, and verify that it is ok..
1398 out = fdget(out_fd);
1401 if (!(out.file->f_mode & FMODE_WRITE))
1404 in_inode = file_inode(in.file);
1405 out_inode = file_inode(out.file);
1406 out_pos = out.file->f_pos;
1407 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1412 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1414 if (unlikely(pos + count > max)) {
1415 retval = -EOVERFLOW;
1424 * We need to debate whether we can enable this or not. The
1425 * man page documents EAGAIN return for the output at least,
1426 * and the application is arguably buggy if it doesn't expect
1427 * EAGAIN on a non-blocking file descriptor.
1429 if (in.file->f_flags & O_NONBLOCK)
1430 fl = SPLICE_F_NONBLOCK;
1432 file_start_write(out.file);
1433 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1434 file_end_write(out.file);
1437 add_rchar(current, retval);
1438 add_wchar(current, retval);
1439 fsnotify_access(in.file);
1440 fsnotify_modify(out.file);
1441 out.file->f_pos = out_pos;
1445 in.file->f_pos = pos;
1451 retval = -EOVERFLOW;
1461 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1468 if (unlikely(get_user(off, offset)))
1471 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1472 if (unlikely(put_user(pos, offset)))
1477 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1480 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1486 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1488 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1489 if (unlikely(put_user(pos, offset)))
1494 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1497 #ifdef CONFIG_COMPAT
1498 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1499 compat_off_t __user *, offset, compat_size_t, count)
1506 if (unlikely(get_user(off, offset)))
1509 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1510 if (unlikely(put_user(pos, offset)))
1515 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1518 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1519 compat_loff_t __user *, offset, compat_size_t, count)
1525 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1527 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1528 if (unlikely(put_user(pos, offset)))
1533 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1538 * copy_file_range() differs from regular file read and write in that it
1539 * specifically allows return partial success. When it does so is up to
1540 * the copy_file_range method.
1542 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1543 struct file *file_out, loff_t pos_out,
1544 size_t len, unsigned int flags)
1546 struct inode *inode_in = file_inode(file_in);
1547 struct inode *inode_out = file_inode(file_out);
1553 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1555 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1558 ret = rw_verify_area(READ, file_in, &pos_in, len);
1562 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1566 if (!(file_in->f_mode & FMODE_READ) ||
1567 !(file_out->f_mode & FMODE_WRITE) ||
1568 (file_out->f_flags & O_APPEND))
1571 /* this could be relaxed once a method supports cross-fs copies */
1572 if (inode_in->i_sb != inode_out->i_sb)
1578 file_start_write(file_out);
1581 * Try cloning first, this is supported by more file systems, and
1582 * more efficient if both clone and copy are supported (e.g. NFS).
1584 if (file_in->f_op->clone_file_range) {
1585 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1586 file_out, pos_out, len);
1593 if (file_out->f_op->copy_file_range) {
1594 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1595 pos_out, len, flags);
1596 if (ret != -EOPNOTSUPP)
1600 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1601 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1605 fsnotify_access(file_in);
1606 add_rchar(current, ret);
1607 fsnotify_modify(file_out);
1608 add_wchar(current, ret);
1614 file_end_write(file_out);
1618 EXPORT_SYMBOL(vfs_copy_file_range);
1620 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1621 int, fd_out, loff_t __user *, off_out,
1622 size_t, len, unsigned int, flags)
1628 ssize_t ret = -EBADF;
1630 f_in = fdget(fd_in);
1634 f_out = fdget(fd_out);
1640 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1643 pos_in = f_in.file->f_pos;
1647 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1650 pos_out = f_out.file->f_pos;
1653 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1660 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1663 f_in.file->f_pos = pos_in;
1667 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1670 f_out.file->f_pos = pos_out;
1682 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1684 struct inode *inode = file_inode(file);
1686 if (unlikely(pos < 0))
1689 if (unlikely((loff_t) (pos + len) < 0))
1692 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1693 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1696 retval = locks_mandatory_area(inode, file, pos, end,
1697 write ? F_WRLCK : F_RDLCK);
1702 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1706 * Check that the two inodes are eligible for cloning, the ranges make
1707 * sense, and then flush all dirty data. Caller must ensure that the
1708 * inodes have been locked against any other modifications.
1710 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1711 * the usual negative error code.
1713 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1714 struct inode *inode_out, loff_t pos_out,
1715 u64 *len, bool is_dedupe)
1717 loff_t bs = inode_out->i_sb->s_blocksize;
1720 bool same_inode = (inode_in == inode_out);
1723 /* Don't touch certain kinds of inodes */
1724 if (IS_IMMUTABLE(inode_out))
1727 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1730 /* Don't reflink dirs, pipes, sockets... */
1731 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1733 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1736 /* Are we going all the way to the end? */
1737 isize = i_size_read(inode_in);
1741 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1743 if (is_dedupe || pos_in == isize)
1747 *len = isize - pos_in;
1750 /* Ensure offsets don't wrap and the input is inside i_size */
1751 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1752 pos_in + *len > isize)
1755 /* Don't allow dedupe past EOF in the dest file */
1759 disize = i_size_read(inode_out);
1760 if (pos_out >= disize || pos_out + *len > disize)
1764 /* If we're linking to EOF, continue to the block boundary. */
1765 if (pos_in + *len == isize)
1766 blen = ALIGN(isize, bs) - pos_in;
1770 /* Only reflink if we're aligned to block boundaries */
1771 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1772 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1775 /* Don't allow overlapped reflink within the same file */
1777 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1781 /* Wait for the completion of any pending IOs on both files */
1782 inode_dio_wait(inode_in);
1784 inode_dio_wait(inode_out);
1786 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1787 pos_in, pos_in + *len - 1);
1791 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1792 pos_out, pos_out + *len - 1);
1797 * Check that the extents are the same.
1800 bool is_same = false;
1802 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1803 inode_out, pos_out, *len, &is_same);
1812 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1814 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1815 struct file *file_out, loff_t pos_out, u64 len)
1817 struct inode *inode_in = file_inode(file_in);
1818 struct inode *inode_out = file_inode(file_out);
1821 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1823 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1827 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1828 * the same mount. Practically, they only need to be on the same file
1831 if (inode_in->i_sb != inode_out->i_sb)
1834 if (!(file_in->f_mode & FMODE_READ) ||
1835 !(file_out->f_mode & FMODE_WRITE) ||
1836 (file_out->f_flags & O_APPEND))
1839 if (!file_in->f_op->clone_file_range)
1842 ret = clone_verify_area(file_in, pos_in, len, false);
1846 ret = clone_verify_area(file_out, pos_out, len, true);
1850 if (pos_in + len > i_size_read(inode_in))
1853 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1854 file_out, pos_out, len);
1856 fsnotify_access(file_in);
1857 fsnotify_modify(file_out);
1862 EXPORT_SYMBOL(vfs_clone_file_range);
1865 * Read a page's worth of file data into the page cache. Return the page
1868 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1870 struct address_space *mapping;
1874 n = offset >> PAGE_SHIFT;
1875 mapping = inode->i_mapping;
1876 page = read_mapping_page(mapping, n, NULL);
1879 if (!PageUptodate(page)) {
1881 return ERR_PTR(-EIO);
1888 * Compare extents of two files to see if they are the same.
1889 * Caller must have locked both inodes to prevent write races.
1891 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1892 struct inode *dest, loff_t destoff,
1893 loff_t len, bool *is_same)
1899 struct page *src_page;
1900 struct page *dest_page;
1908 src_poff = srcoff & (PAGE_SIZE - 1);
1909 dest_poff = destoff & (PAGE_SIZE - 1);
1910 cmp_len = min(PAGE_SIZE - src_poff,
1911 PAGE_SIZE - dest_poff);
1912 cmp_len = min(cmp_len, len);
1916 src_page = vfs_dedupe_get_page(src, srcoff);
1917 if (IS_ERR(src_page)) {
1918 error = PTR_ERR(src_page);
1921 dest_page = vfs_dedupe_get_page(dest, destoff);
1922 if (IS_ERR(dest_page)) {
1923 error = PTR_ERR(dest_page);
1924 unlock_page(src_page);
1928 src_addr = kmap_atomic(src_page);
1929 dest_addr = kmap_atomic(dest_page);
1931 flush_dcache_page(src_page);
1932 flush_dcache_page(dest_page);
1934 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1937 kunmap_atomic(dest_addr);
1938 kunmap_atomic(src_addr);
1939 unlock_page(dest_page);
1940 unlock_page(src_page);
1941 put_page(dest_page);
1958 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1960 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1962 struct file_dedupe_range_info *info;
1963 struct inode *src = file_inode(file);
1968 bool is_admin = capable(CAP_SYS_ADMIN);
1969 u16 count = same->dest_count;
1970 struct file *dst_file;
1974 if (!(file->f_mode & FMODE_READ))
1977 if (same->reserved1 || same->reserved2)
1980 off = same->src_offset;
1981 len = same->src_length;
1984 if (S_ISDIR(src->i_mode))
1988 if (!S_ISREG(src->i_mode))
1991 ret = clone_verify_area(file, off, len, false);
1996 if (off + len > i_size_read(src))
1999 /* pre-format output fields to sane values */
2000 for (i = 0; i < count; i++) {
2001 same->info[i].bytes_deduped = 0ULL;
2002 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2005 for (i = 0, info = same->info; i < count; i++, info++) {
2007 struct fd dst_fd = fdget(info->dest_fd);
2009 dst_file = dst_fd.file;
2011 info->status = -EBADF;
2014 dst = file_inode(dst_file);
2016 ret = mnt_want_write_file(dst_file);
2022 dst_off = info->dest_offset;
2023 ret = clone_verify_area(dst_file, dst_off, len, true);
2030 if (info->reserved) {
2031 info->status = -EINVAL;
2032 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2033 info->status = -EINVAL;
2034 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2035 info->status = -EXDEV;
2036 } else if (S_ISDIR(dst->i_mode)) {
2037 info->status = -EISDIR;
2038 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2039 info->status = -EINVAL;
2041 deduped = dst_file->f_op->dedupe_file_range(file, off,
2044 if (deduped == -EBADE)
2045 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2046 else if (deduped < 0)
2047 info->status = deduped;
2049 info->bytes_deduped += deduped;
2053 mnt_drop_write_file(dst_file);
2057 if (fatal_signal_pending(current))
2064 EXPORT_SYMBOL(vfs_dedupe_file_range);