1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/read_write.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/stat.h>
10 #include <linux/sched/xacct.h>
11 #include <linux/fcntl.h>
12 #include <linux/file.h>
13 #include <linux/uio.h>
14 #include <linux/fsnotify.h>
15 #include <linux/security.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/pagemap.h>
19 #include <linux/splice.h>
20 #include <linux/compat.h>
21 #include <linux/mount.h>
25 #include <linux/uaccess.h>
26 #include <asm/unistd.h>
28 const struct file_operations generic_ro_fops = {
29 .llseek = generic_file_llseek,
30 .read_iter = generic_file_read_iter,
31 .mmap = generic_file_readonly_mmap,
32 .splice_read = generic_file_splice_read,
35 EXPORT_SYMBOL(generic_ro_fops);
37 static inline bool unsigned_offsets(struct file *file)
39 return file->f_mode & FMODE_UNSIGNED_OFFSET;
43 * vfs_setpos - update the file offset for lseek
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
48 * This is a low-level filesystem helper for updating the file offset to
49 * the value specified by @offset if the given offset is valid and it is
50 * not equal to the current file offset.
52 * Return the specified offset on success and -EINVAL on invalid offset.
54 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
56 if (offset < 0 && !unsigned_offsets(file))
61 if (offset != file->f_pos) {
67 EXPORT_SYMBOL(vfs_setpos);
70 * generic_file_llseek_size - generic llseek implementation for regular files
71 * @file: file structure to seek on
72 * @offset: file offset to seek to
73 * @whence: type of seek
74 * @size: max size of this file in file system
75 * @eof: offset used for SEEK_END position
77 * This is a variant of generic_file_llseek that allows passing in a custom
78 * maximum file size and a custom EOF position, for e.g. hashed directories
81 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
82 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83 * read/writes behave like SEEK_SET against seeks.
86 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
87 loff_t maxsize, loff_t eof)
95 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 * position-querying operation. Avoid rewriting the "same"
97 * f_pos value back to the file because a concurrent read(),
98 * write() or lseek() might have altered it
103 * f_lock protects against read/modify/write race with other
104 * SEEK_CURs. Note that parallel writes and reads behave
107 spin_lock(&file->f_lock);
108 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
109 spin_unlock(&file->f_lock);
113 * In the generic case the entire file is data, so as long as
114 * offset isn't at the end of the file then the offset is data.
116 if ((unsigned long long)offset >= eof)
121 * There is a virtual hole at the end of the file, so as long as
122 * offset isn't i_size or larger, return i_size.
124 if ((unsigned long long)offset >= eof)
130 return vfs_setpos(file, offset, maxsize);
132 EXPORT_SYMBOL(generic_file_llseek_size);
135 * generic_file_llseek - generic llseek implementation for regular files
136 * @file: file structure to seek on
137 * @offset: file offset to seek to
138 * @whence: type of seek
140 * This is a generic implemenation of ->llseek useable for all normal local
141 * filesystems. It just updates the file offset to the value specified by
142 * @offset and @whence.
144 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
146 struct inode *inode = file->f_mapping->host;
148 return generic_file_llseek_size(file, offset, whence,
149 inode->i_sb->s_maxbytes,
152 EXPORT_SYMBOL(generic_file_llseek);
155 * fixed_size_llseek - llseek implementation for fixed-sized devices
156 * @file: file structure to seek on
157 * @offset: file offset to seek to
158 * @whence: type of seek
159 * @size: size of the file
162 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
165 case SEEK_SET: case SEEK_CUR: case SEEK_END:
166 return generic_file_llseek_size(file, offset, whence,
172 EXPORT_SYMBOL(fixed_size_llseek);
175 * no_seek_end_llseek - llseek implementation for fixed-sized devices
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
181 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
184 case SEEK_SET: case SEEK_CUR:
185 return generic_file_llseek_size(file, offset, whence,
191 EXPORT_SYMBOL(no_seek_end_llseek);
194 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195 * @file: file structure to seek on
196 * @offset: file offset to seek to
197 * @whence: type of seek
198 * @size: maximal offset allowed
201 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
204 case SEEK_SET: case SEEK_CUR:
205 return generic_file_llseek_size(file, offset, whence,
211 EXPORT_SYMBOL(no_seek_end_llseek_size);
214 * noop_llseek - No Operation Performed llseek implementation
215 * @file: file structure to seek on
216 * @offset: file offset to seek to
217 * @whence: type of seek
219 * This is an implementation of ->llseek useable for the rare special case when
220 * userspace expects the seek to succeed but the (device) file is actually not
221 * able to perform the seek. In this case you use noop_llseek() instead of
222 * falling back to the default implementation of ->llseek.
224 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
228 EXPORT_SYMBOL(noop_llseek);
230 loff_t no_llseek(struct file *file, loff_t offset, int whence)
234 EXPORT_SYMBOL(no_llseek);
236 loff_t default_llseek(struct file *file, loff_t offset, int whence)
238 struct inode *inode = file_inode(file);
244 offset += i_size_read(inode);
248 retval = file->f_pos;
251 offset += file->f_pos;
255 * In the generic case the entire file is data, so as
256 * long as offset isn't at the end of the file then the
259 if (offset >= inode->i_size) {
266 * There is a virtual hole at the end of the file, so
267 * as long as offset isn't i_size or larger, return
270 if (offset >= inode->i_size) {
274 offset = inode->i_size;
278 if (offset >= 0 || unsigned_offsets(file)) {
279 if (offset != file->f_pos) {
280 file->f_pos = offset;
289 EXPORT_SYMBOL(default_llseek);
291 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
293 loff_t (*fn)(struct file *, loff_t, int);
296 if (file->f_mode & FMODE_LSEEK) {
297 if (file->f_op->llseek)
298 fn = file->f_op->llseek;
300 return fn(file, offset, whence);
302 EXPORT_SYMBOL(vfs_llseek);
304 off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
307 struct fd f = fdget_pos(fd);
312 if (whence <= SEEK_MAX) {
313 loff_t res = vfs_llseek(f.file, offset, whence);
315 if (res != (loff_t)retval)
316 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
324 return ksys_lseek(fd, offset, whence);
328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
330 return ksys_lseek(fd, offset, whence);
334 #ifdef __ARCH_WANT_SYS_LLSEEK
335 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
336 unsigned long, offset_low, loff_t __user *, result,
337 unsigned int, whence)
340 struct fd f = fdget_pos(fd);
347 if (whence > SEEK_MAX)
350 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
353 retval = (int)offset;
356 if (!copy_to_user(result, &offset, sizeof(offset)))
365 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
369 int retval = -EINVAL;
371 inode = file_inode(file);
372 if (unlikely((ssize_t) count < 0))
375 if (unlikely(pos < 0)) {
376 if (!unsigned_offsets(file))
378 if (count >= -pos) /* both values are in 0..LLONG_MAX */
380 } else if (unlikely((loff_t) (pos + count) < 0)) {
381 if (!unsigned_offsets(file))
385 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
386 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
387 read_write == READ ? F_RDLCK : F_WRLCK);
391 return security_file_permission(file,
392 read_write == READ ? MAY_READ : MAY_WRITE);
395 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
397 struct iovec iov = { .iov_base = buf, .iov_len = len };
399 struct iov_iter iter;
402 init_sync_kiocb(&kiocb, filp);
403 kiocb.ki_pos = *ppos;
404 iov_iter_init(&iter, READ, &iov, 1, len);
406 ret = call_read_iter(filp, &kiocb, &iter);
407 BUG_ON(ret == -EIOCBQUEUED);
408 *ppos = kiocb.ki_pos;
412 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
415 if (file->f_op->read)
416 return file->f_op->read(file, buf, count, pos);
417 else if (file->f_op->read_iter)
418 return new_sync_read(file, buf, count, pos);
423 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
430 /* The cast to a user pointer is valid due to the set_fs() */
431 result = vfs_read(file, (void __user *)buf, count, pos);
435 EXPORT_SYMBOL(kernel_read);
437 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
441 if (!(file->f_mode & FMODE_READ))
443 if (!(file->f_mode & FMODE_CAN_READ))
445 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
448 ret = rw_verify_area(READ, file, pos, count);
450 if (count > MAX_RW_COUNT)
451 count = MAX_RW_COUNT;
452 ret = __vfs_read(file, buf, count, pos);
454 fsnotify_access(file);
455 add_rchar(current, ret);
463 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
465 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
467 struct iov_iter iter;
470 init_sync_kiocb(&kiocb, filp);
471 kiocb.ki_pos = *ppos;
472 iov_iter_init(&iter, WRITE, &iov, 1, len);
474 ret = call_write_iter(filp, &kiocb, &iter);
475 BUG_ON(ret == -EIOCBQUEUED);
477 *ppos = kiocb.ki_pos;
481 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
484 if (file->f_op->write)
485 return file->f_op->write(file, p, count, pos);
486 else if (file->f_op->write_iter)
487 return new_sync_write(file, p, count, pos);
492 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
495 const char __user *p;
498 if (!(file->f_mode & FMODE_CAN_WRITE))
503 p = (__force const char __user *)buf;
504 if (count > MAX_RW_COUNT)
505 count = MAX_RW_COUNT;
506 ret = __vfs_write(file, p, count, pos);
509 fsnotify_modify(file);
510 add_wchar(current, ret);
515 EXPORT_SYMBOL(__kernel_write);
517 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
525 /* The cast to a user pointer is valid due to the set_fs() */
526 res = vfs_write(file, (__force const char __user *)buf, count, pos);
531 EXPORT_SYMBOL(kernel_write);
533 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
537 if (!(file->f_mode & FMODE_WRITE))
539 if (!(file->f_mode & FMODE_CAN_WRITE))
541 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
544 ret = rw_verify_area(WRITE, file, pos, count);
546 if (count > MAX_RW_COUNT)
547 count = MAX_RW_COUNT;
548 file_start_write(file);
549 ret = __vfs_write(file, buf, count, pos);
551 fsnotify_modify(file);
552 add_wchar(current, ret);
555 file_end_write(file);
561 static inline loff_t file_pos_read(struct file *file)
566 static inline void file_pos_write(struct file *file, loff_t pos)
571 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
573 struct fd f = fdget_pos(fd);
574 ssize_t ret = -EBADF;
577 loff_t pos = file_pos_read(f.file);
578 ret = vfs_read(f.file, buf, count, &pos);
580 file_pos_write(f.file, pos);
586 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
588 return ksys_read(fd, buf, count);
591 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
593 struct fd f = fdget_pos(fd);
594 ssize_t ret = -EBADF;
597 loff_t pos = file_pos_read(f.file);
598 ret = vfs_write(f.file, buf, count, &pos);
600 file_pos_write(f.file, pos);
607 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
610 return ksys_write(fd, buf, count);
613 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
617 ssize_t ret = -EBADF;
625 if (f.file->f_mode & FMODE_PREAD)
626 ret = vfs_read(f.file, buf, count, &pos);
633 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
634 size_t, count, loff_t, pos)
636 return ksys_pread64(fd, buf, count, pos);
639 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
640 size_t count, loff_t pos)
643 ssize_t ret = -EBADF;
651 if (f.file->f_mode & FMODE_PWRITE)
652 ret = vfs_write(f.file, buf, count, &pos);
659 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
660 size_t, count, loff_t, pos)
662 return ksys_pwrite64(fd, buf, count, pos);
665 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
666 loff_t *ppos, int type, rwf_t flags)
671 init_sync_kiocb(&kiocb, filp);
672 ret = kiocb_set_rw_flags(&kiocb, flags);
675 kiocb.ki_pos = *ppos;
678 ret = call_read_iter(filp, &kiocb, iter);
680 ret = call_write_iter(filp, &kiocb, iter);
681 BUG_ON(ret == -EIOCBQUEUED);
682 *ppos = kiocb.ki_pos;
686 /* Do it by hand, with file-ops */
687 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
688 loff_t *ppos, int type, rwf_t flags)
692 if (flags & ~RWF_HIPRI)
695 while (iov_iter_count(iter)) {
696 struct iovec iovec = iov_iter_iovec(iter);
700 nr = filp->f_op->read(filp, iovec.iov_base,
701 iovec.iov_len, ppos);
703 nr = filp->f_op->write(filp, iovec.iov_base,
704 iovec.iov_len, ppos);
713 if (nr != iovec.iov_len)
715 iov_iter_advance(iter, nr);
721 /* A write operation does a read from user space and vice versa */
722 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
725 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
726 * into the kernel and check that it is valid.
728 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
729 * @uvector: Pointer to the userspace array.
730 * @nr_segs: Number of elements in userspace array.
731 * @fast_segs: Number of elements in @fast_pointer.
732 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
733 * @ret_pointer: (output parameter) Pointer to a variable that will point to
734 * either @fast_pointer, a newly allocated kernel array, or NULL,
735 * depending on which array was used.
737 * This function copies an array of &struct iovec of @nr_segs from
738 * userspace into the kernel and checks that each element is valid (e.g.
739 * it does not point to a kernel address or cause overflow by being too
742 * As an optimization, the caller may provide a pointer to a small
743 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
744 * (the size of this array, or 0 if unused, should be given in @fast_segs).
746 * @ret_pointer will always point to the array that was used, so the
747 * caller must take care not to call kfree() on it e.g. in case the
748 * @fast_pointer array was used and it was allocated on the stack.
750 * Return: The total number of bytes covered by the iovec array on success
751 * or a negative error code on error.
753 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
754 unsigned long nr_segs, unsigned long fast_segs,
755 struct iovec *fast_pointer,
756 struct iovec **ret_pointer)
760 struct iovec *iov = fast_pointer;
763 * SuS says "The readv() function *may* fail if the iovcnt argument
764 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
765 * traditionally returned zero for zero segments, so...
773 * First get the "struct iovec" from user memory and
774 * verify all the pointers
776 if (nr_segs > UIO_MAXIOV) {
780 if (nr_segs > fast_segs) {
781 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
787 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
793 * According to the Single Unix Specification we should return EINVAL
794 * if an element length is < 0 when cast to ssize_t or if the
795 * total length would overflow the ssize_t return value of the
798 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
802 for (seg = 0; seg < nr_segs; seg++) {
803 void __user *buf = iov[seg].iov_base;
804 ssize_t len = (ssize_t)iov[seg].iov_len;
806 /* see if we we're about to use an invalid len or if
807 * it's about to overflow ssize_t */
813 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
817 if (len > MAX_RW_COUNT - ret) {
818 len = MAX_RW_COUNT - ret;
819 iov[seg].iov_len = len;
829 ssize_t compat_rw_copy_check_uvector(int type,
830 const struct compat_iovec __user *uvector, unsigned long nr_segs,
831 unsigned long fast_segs, struct iovec *fast_pointer,
832 struct iovec **ret_pointer)
834 compat_ssize_t tot_len;
835 struct iovec *iov = *ret_pointer = fast_pointer;
840 * SuS says "The readv() function *may* fail if the iovcnt argument
841 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
842 * traditionally returned zero for zero segments, so...
848 if (nr_segs > UIO_MAXIOV)
850 if (nr_segs > fast_segs) {
852 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
859 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
863 * Single unix specification:
864 * We should -EINVAL if an element length is not >= 0 and fitting an
867 * In Linux, the total length is limited to MAX_RW_COUNT, there is
868 * no overflow possibility.
872 for (seg = 0; seg < nr_segs; seg++) {
876 if (__get_user(len, &uvector->iov_len) ||
877 __get_user(buf, &uvector->iov_base)) {
881 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
884 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
888 if (len > MAX_RW_COUNT - tot_len)
889 len = MAX_RW_COUNT - tot_len;
891 iov->iov_base = compat_ptr(buf);
892 iov->iov_len = (compat_size_t) len;
903 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
904 loff_t *pos, rwf_t flags)
909 if (!(file->f_mode & FMODE_READ))
911 if (!(file->f_mode & FMODE_CAN_READ))
914 tot_len = iov_iter_count(iter);
917 ret = rw_verify_area(READ, file, pos, tot_len);
921 if (file->f_op->read_iter)
922 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
924 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
927 fsnotify_access(file);
931 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
934 if (!file->f_op->read_iter)
936 return do_iter_read(file, iter, ppos, flags);
938 EXPORT_SYMBOL(vfs_iter_read);
940 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
941 loff_t *pos, rwf_t flags)
946 if (!(file->f_mode & FMODE_WRITE))
948 if (!(file->f_mode & FMODE_CAN_WRITE))
951 tot_len = iov_iter_count(iter);
954 ret = rw_verify_area(WRITE, file, pos, tot_len);
958 if (file->f_op->write_iter)
959 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
961 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
963 fsnotify_modify(file);
967 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
970 if (!file->f_op->write_iter)
972 return do_iter_write(file, iter, ppos, flags);
974 EXPORT_SYMBOL(vfs_iter_write);
976 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
977 unsigned long vlen, loff_t *pos, rwf_t flags)
979 struct iovec iovstack[UIO_FASTIOV];
980 struct iovec *iov = iovstack;
981 struct iov_iter iter;
984 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
986 ret = do_iter_read(file, &iter, pos, flags);
993 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
994 unsigned long vlen, loff_t *pos, rwf_t flags)
996 struct iovec iovstack[UIO_FASTIOV];
997 struct iovec *iov = iovstack;
998 struct iov_iter iter;
1001 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1003 file_start_write(file);
1004 ret = do_iter_write(file, &iter, pos, flags);
1005 file_end_write(file);
1011 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1012 unsigned long vlen, rwf_t flags)
1014 struct fd f = fdget_pos(fd);
1015 ssize_t ret = -EBADF;
1018 loff_t pos = file_pos_read(f.file);
1019 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1021 file_pos_write(f.file, pos);
1026 add_rchar(current, ret);
1031 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1032 unsigned long vlen, rwf_t flags)
1034 struct fd f = fdget_pos(fd);
1035 ssize_t ret = -EBADF;
1038 loff_t pos = file_pos_read(f.file);
1039 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1041 file_pos_write(f.file, pos);
1046 add_wchar(current, ret);
1051 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1053 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1054 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1057 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1058 unsigned long vlen, loff_t pos, rwf_t flags)
1061 ssize_t ret = -EBADF;
1069 if (f.file->f_mode & FMODE_PREAD)
1070 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1075 add_rchar(current, ret);
1080 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1081 unsigned long vlen, loff_t pos, rwf_t flags)
1084 ssize_t ret = -EBADF;
1092 if (f.file->f_mode & FMODE_PWRITE)
1093 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1098 add_wchar(current, ret);
1103 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1104 unsigned long, vlen)
1106 return do_readv(fd, vec, vlen, 0);
1109 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1110 unsigned long, vlen)
1112 return do_writev(fd, vec, vlen, 0);
1115 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1116 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1118 loff_t pos = pos_from_hilo(pos_h, pos_l);
1120 return do_preadv(fd, vec, vlen, pos, 0);
1123 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1124 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1127 loff_t pos = pos_from_hilo(pos_h, pos_l);
1130 return do_readv(fd, vec, vlen, flags);
1132 return do_preadv(fd, vec, vlen, pos, flags);
1135 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1136 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1138 loff_t pos = pos_from_hilo(pos_h, pos_l);
1140 return do_pwritev(fd, vec, vlen, pos, 0);
1143 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1144 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1147 loff_t pos = pos_from_hilo(pos_h, pos_l);
1150 return do_writev(fd, vec, vlen, flags);
1152 return do_pwritev(fd, vec, vlen, pos, flags);
1155 #ifdef CONFIG_COMPAT
1156 static size_t compat_readv(struct file *file,
1157 const struct compat_iovec __user *vec,
1158 unsigned long vlen, loff_t *pos, rwf_t flags)
1160 struct iovec iovstack[UIO_FASTIOV];
1161 struct iovec *iov = iovstack;
1162 struct iov_iter iter;
1165 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1167 ret = do_iter_read(file, &iter, pos, flags);
1171 add_rchar(current, ret);
1176 static size_t do_compat_readv(compat_ulong_t fd,
1177 const struct compat_iovec __user *vec,
1178 compat_ulong_t vlen, rwf_t flags)
1180 struct fd f = fdget_pos(fd);
1186 pos = f.file->f_pos;
1187 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1189 f.file->f_pos = pos;
1195 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1196 const struct compat_iovec __user *,vec,
1197 compat_ulong_t, vlen)
1199 return do_compat_readv(fd, vec, vlen, 0);
1202 static long do_compat_preadv64(unsigned long fd,
1203 const struct compat_iovec __user *vec,
1204 unsigned long vlen, loff_t pos, rwf_t flags)
1215 if (f.file->f_mode & FMODE_PREAD)
1216 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1221 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1222 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1223 const struct compat_iovec __user *,vec,
1224 unsigned long, vlen, loff_t, pos)
1226 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1230 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1231 const struct compat_iovec __user *,vec,
1232 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1234 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1236 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1239 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1240 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1241 const struct compat_iovec __user *,vec,
1242 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1244 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1248 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1249 const struct compat_iovec __user *,vec,
1250 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1253 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1256 return do_compat_readv(fd, vec, vlen, flags);
1258 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1261 static size_t compat_writev(struct file *file,
1262 const struct compat_iovec __user *vec,
1263 unsigned long vlen, loff_t *pos, rwf_t flags)
1265 struct iovec iovstack[UIO_FASTIOV];
1266 struct iovec *iov = iovstack;
1267 struct iov_iter iter;
1270 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1272 file_start_write(file);
1273 ret = do_iter_write(file, &iter, pos, flags);
1274 file_end_write(file);
1278 add_wchar(current, ret);
1283 static size_t do_compat_writev(compat_ulong_t fd,
1284 const struct compat_iovec __user* vec,
1285 compat_ulong_t vlen, rwf_t flags)
1287 struct fd f = fdget_pos(fd);
1293 pos = f.file->f_pos;
1294 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1296 f.file->f_pos = pos;
1301 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1302 const struct compat_iovec __user *, vec,
1303 compat_ulong_t, vlen)
1305 return do_compat_writev(fd, vec, vlen, 0);
1308 static long do_compat_pwritev64(unsigned long fd,
1309 const struct compat_iovec __user *vec,
1310 unsigned long vlen, loff_t pos, rwf_t flags)
1321 if (f.file->f_mode & FMODE_PWRITE)
1322 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1327 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1328 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1329 const struct compat_iovec __user *,vec,
1330 unsigned long, vlen, loff_t, pos)
1332 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1336 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1337 const struct compat_iovec __user *,vec,
1338 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1340 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1342 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1345 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1346 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1347 const struct compat_iovec __user *,vec,
1348 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1350 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1354 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1355 const struct compat_iovec __user *,vec,
1356 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1358 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1361 return do_compat_writev(fd, vec, vlen, flags);
1363 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1368 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1369 size_t count, loff_t max)
1372 struct inode *in_inode, *out_inode;
1379 * Get input file, and verify that it is ok..
1385 if (!(in.file->f_mode & FMODE_READ))
1389 pos = in.file->f_pos;
1392 if (!(in.file->f_mode & FMODE_PREAD))
1395 retval = rw_verify_area(READ, in.file, &pos, count);
1398 if (count > MAX_RW_COUNT)
1399 count = MAX_RW_COUNT;
1402 * Get output file, and verify that it is ok..
1405 out = fdget(out_fd);
1408 if (!(out.file->f_mode & FMODE_WRITE))
1411 in_inode = file_inode(in.file);
1412 out_inode = file_inode(out.file);
1413 out_pos = out.file->f_pos;
1414 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1419 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1421 if (unlikely(pos + count > max)) {
1422 retval = -EOVERFLOW;
1431 * We need to debate whether we can enable this or not. The
1432 * man page documents EAGAIN return for the output at least,
1433 * and the application is arguably buggy if it doesn't expect
1434 * EAGAIN on a non-blocking file descriptor.
1436 if (in.file->f_flags & O_NONBLOCK)
1437 fl = SPLICE_F_NONBLOCK;
1439 file_start_write(out.file);
1440 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1441 file_end_write(out.file);
1444 add_rchar(current, retval);
1445 add_wchar(current, retval);
1446 fsnotify_access(in.file);
1447 fsnotify_modify(out.file);
1448 out.file->f_pos = out_pos;
1452 in.file->f_pos = pos;
1458 retval = -EOVERFLOW;
1468 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1475 if (unlikely(get_user(off, offset)))
1478 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1479 if (unlikely(put_user(pos, offset)))
1484 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1487 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1493 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1495 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1496 if (unlikely(put_user(pos, offset)))
1501 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1504 #ifdef CONFIG_COMPAT
1505 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1506 compat_off_t __user *, offset, compat_size_t, count)
1513 if (unlikely(get_user(off, offset)))
1516 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1517 if (unlikely(put_user(pos, offset)))
1522 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1525 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1526 compat_loff_t __user *, offset, compat_size_t, count)
1532 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1534 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1535 if (unlikely(put_user(pos, offset)))
1540 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1545 * copy_file_range() differs from regular file read and write in that it
1546 * specifically allows return partial success. When it does so is up to
1547 * the copy_file_range method.
1549 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1550 struct file *file_out, loff_t pos_out,
1551 size_t len, unsigned int flags)
1553 struct inode *inode_in = file_inode(file_in);
1554 struct inode *inode_out = file_inode(file_out);
1560 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1562 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1565 ret = rw_verify_area(READ, file_in, &pos_in, len);
1569 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1573 if (!(file_in->f_mode & FMODE_READ) ||
1574 !(file_out->f_mode & FMODE_WRITE) ||
1575 (file_out->f_flags & O_APPEND))
1578 /* this could be relaxed once a method supports cross-fs copies */
1579 if (inode_in->i_sb != inode_out->i_sb)
1585 file_start_write(file_out);
1588 * Try cloning first, this is supported by more file systems, and
1589 * more efficient if both clone and copy are supported (e.g. NFS).
1591 if (file_in->f_op->clone_file_range) {
1592 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1593 file_out, pos_out, len);
1600 if (file_out->f_op->copy_file_range) {
1601 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1602 pos_out, len, flags);
1603 if (ret != -EOPNOTSUPP)
1607 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1608 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1612 fsnotify_access(file_in);
1613 add_rchar(current, ret);
1614 fsnotify_modify(file_out);
1615 add_wchar(current, ret);
1621 file_end_write(file_out);
1625 EXPORT_SYMBOL(vfs_copy_file_range);
1627 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1628 int, fd_out, loff_t __user *, off_out,
1629 size_t, len, unsigned int, flags)
1635 ssize_t ret = -EBADF;
1637 f_in = fdget(fd_in);
1641 f_out = fdget(fd_out);
1647 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1650 pos_in = f_in.file->f_pos;
1654 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1657 pos_out = f_out.file->f_pos;
1660 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1667 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1670 f_in.file->f_pos = pos_in;
1674 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1677 f_out.file->f_pos = pos_out;
1689 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1691 struct inode *inode = file_inode(file);
1693 if (unlikely(pos < 0))
1696 if (unlikely((loff_t) (pos + len) < 0))
1699 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1700 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1703 retval = locks_mandatory_area(inode, file, pos, end,
1704 write ? F_WRLCK : F_RDLCK);
1709 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1713 * Check that the two inodes are eligible for cloning, the ranges make
1714 * sense, and then flush all dirty data. Caller must ensure that the
1715 * inodes have been locked against any other modifications.
1717 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1718 * the usual negative error code.
1720 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1721 struct inode *inode_out, loff_t pos_out,
1722 u64 *len, bool is_dedupe)
1724 loff_t bs = inode_out->i_sb->s_blocksize;
1727 bool same_inode = (inode_in == inode_out);
1730 /* Don't touch certain kinds of inodes */
1731 if (IS_IMMUTABLE(inode_out))
1734 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1737 /* Don't reflink dirs, pipes, sockets... */
1738 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1740 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1743 /* Are we going all the way to the end? */
1744 isize = i_size_read(inode_in);
1748 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1750 if (is_dedupe || pos_in == isize)
1754 *len = isize - pos_in;
1757 /* Ensure offsets don't wrap and the input is inside i_size */
1758 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1759 pos_in + *len > isize)
1762 /* Don't allow dedupe past EOF in the dest file */
1766 disize = i_size_read(inode_out);
1767 if (pos_out >= disize || pos_out + *len > disize)
1771 /* If we're linking to EOF, continue to the block boundary. */
1772 if (pos_in + *len == isize)
1773 blen = ALIGN(isize, bs) - pos_in;
1777 /* Only reflink if we're aligned to block boundaries */
1778 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1779 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1782 /* Don't allow overlapped reflink within the same file */
1784 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1788 /* Wait for the completion of any pending IOs on both files */
1789 inode_dio_wait(inode_in);
1791 inode_dio_wait(inode_out);
1793 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1794 pos_in, pos_in + *len - 1);
1798 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1799 pos_out, pos_out + *len - 1);
1804 * Check that the extents are the same.
1807 bool is_same = false;
1809 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1810 inode_out, pos_out, *len, &is_same);
1819 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1821 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1822 struct file *file_out, loff_t pos_out, u64 len)
1824 struct inode *inode_in = file_inode(file_in);
1825 struct inode *inode_out = file_inode(file_out);
1828 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1830 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1834 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1835 * the same mount. Practically, they only need to be on the same file
1838 if (inode_in->i_sb != inode_out->i_sb)
1841 if (!(file_in->f_mode & FMODE_READ) ||
1842 !(file_out->f_mode & FMODE_WRITE) ||
1843 (file_out->f_flags & O_APPEND))
1846 if (!file_in->f_op->clone_file_range)
1849 ret = clone_verify_area(file_in, pos_in, len, false);
1853 ret = clone_verify_area(file_out, pos_out, len, true);
1857 if (pos_in + len > i_size_read(inode_in))
1860 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1861 file_out, pos_out, len);
1863 fsnotify_access(file_in);
1864 fsnotify_modify(file_out);
1869 EXPORT_SYMBOL(vfs_clone_file_range);
1872 * Read a page's worth of file data into the page cache. Return the page
1875 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1877 struct address_space *mapping;
1881 n = offset >> PAGE_SHIFT;
1882 mapping = inode->i_mapping;
1883 page = read_mapping_page(mapping, n, NULL);
1886 if (!PageUptodate(page)) {
1888 return ERR_PTR(-EIO);
1895 * Compare extents of two files to see if they are the same.
1896 * Caller must have locked both inodes to prevent write races.
1898 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1899 struct inode *dest, loff_t destoff,
1900 loff_t len, bool *is_same)
1906 struct page *src_page;
1907 struct page *dest_page;
1915 src_poff = srcoff & (PAGE_SIZE - 1);
1916 dest_poff = destoff & (PAGE_SIZE - 1);
1917 cmp_len = min(PAGE_SIZE - src_poff,
1918 PAGE_SIZE - dest_poff);
1919 cmp_len = min(cmp_len, len);
1923 src_page = vfs_dedupe_get_page(src, srcoff);
1924 if (IS_ERR(src_page)) {
1925 error = PTR_ERR(src_page);
1928 dest_page = vfs_dedupe_get_page(dest, destoff);
1929 if (IS_ERR(dest_page)) {
1930 error = PTR_ERR(dest_page);
1931 unlock_page(src_page);
1935 src_addr = kmap_atomic(src_page);
1936 dest_addr = kmap_atomic(dest_page);
1938 flush_dcache_page(src_page);
1939 flush_dcache_page(dest_page);
1941 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1944 kunmap_atomic(dest_addr);
1945 kunmap_atomic(src_addr);
1946 unlock_page(dest_page);
1947 unlock_page(src_page);
1948 put_page(dest_page);
1965 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1967 int vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
1968 struct file *dst_file, loff_t dst_pos, u64 len)
1972 ret = mnt_want_write_file(dst_file);
1976 ret = clone_verify_area(dst_file, dst_pos, len, true);
1978 goto out_drop_write;
1981 if (!(capable(CAP_SYS_ADMIN) || (dst_file->f_mode & FMODE_WRITE)))
1982 goto out_drop_write;
1985 if (src_file->f_path.mnt != dst_file->f_path.mnt)
1986 goto out_drop_write;
1989 if (S_ISDIR(file_inode(dst_file)->i_mode))
1990 goto out_drop_write;
1993 if (!dst_file->f_op->dedupe_file_range)
1994 goto out_drop_write;
1996 ret = dst_file->f_op->dedupe_file_range(src_file, src_pos,
1997 dst_file, dst_pos, len);
1999 mnt_drop_write_file(dst_file);
2003 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
2005 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
2007 struct file_dedupe_range_info *info;
2008 struct inode *src = file_inode(file);
2013 u16 count = same->dest_count;
2016 if (!(file->f_mode & FMODE_READ))
2019 if (same->reserved1 || same->reserved2)
2022 off = same->src_offset;
2023 len = same->src_length;
2026 if (S_ISDIR(src->i_mode))
2030 if (!S_ISREG(src->i_mode))
2033 ret = clone_verify_area(file, off, len, false);
2038 if (off + len > i_size_read(src))
2041 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
2042 len = min_t(u64, len, 1 << 30);
2044 /* pre-format output fields to sane values */
2045 for (i = 0; i < count; i++) {
2046 same->info[i].bytes_deduped = 0ULL;
2047 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2050 for (i = 0, info = same->info; i < count; i++, info++) {
2051 struct fd dst_fd = fdget(info->dest_fd);
2052 struct file *dst_file = dst_fd.file;
2055 info->status = -EBADF;
2059 if (info->reserved) {
2060 info->status = -EINVAL;
2064 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
2065 info->dest_offset, len);
2066 if (deduped == -EBADE)
2067 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2068 else if (deduped < 0)
2069 info->status = deduped;
2071 info->bytes_deduped = len;
2076 if (fatal_signal_pending(current))
2083 EXPORT_SYMBOL(vfs_dedupe_file_range);