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 SYSCALL_DEFINE3(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 */
323 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
325 return sys_lseek(fd, offset, whence);
329 #ifdef __ARCH_WANT_SYS_LLSEEK
330 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
331 unsigned long, offset_low, loff_t __user *, result,
332 unsigned int, whence)
335 struct fd f = fdget_pos(fd);
342 if (whence > SEEK_MAX)
345 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
348 retval = (int)offset;
351 if (!copy_to_user(result, &offset, sizeof(offset)))
360 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
364 int retval = -EINVAL;
366 inode = file_inode(file);
367 if (unlikely((ssize_t) count < 0))
370 if (unlikely(pos < 0)) {
371 if (!unsigned_offsets(file))
373 if (count >= -pos) /* both values are in 0..LLONG_MAX */
375 } else if (unlikely((loff_t) (pos + count) < 0)) {
376 if (!unsigned_offsets(file))
380 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
381 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
382 read_write == READ ? F_RDLCK : F_WRLCK);
386 return security_file_permission(file,
387 read_write == READ ? MAY_READ : MAY_WRITE);
390 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
392 struct iovec iov = { .iov_base = buf, .iov_len = len };
394 struct iov_iter iter;
397 init_sync_kiocb(&kiocb, filp);
398 kiocb.ki_pos = *ppos;
399 iov_iter_init(&iter, READ, &iov, 1, len);
401 ret = call_read_iter(filp, &kiocb, &iter);
402 BUG_ON(ret == -EIOCBQUEUED);
403 *ppos = kiocb.ki_pos;
407 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
410 if (file->f_op->read)
411 return file->f_op->read(file, buf, count, pos);
412 else if (file->f_op->read_iter)
413 return new_sync_read(file, buf, count, pos);
418 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
425 /* The cast to a user pointer is valid due to the set_fs() */
426 result = vfs_read(file, (void __user *)buf, count, pos);
430 EXPORT_SYMBOL(kernel_read);
432 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
436 if (!(file->f_mode & FMODE_READ))
438 if (!(file->f_mode & FMODE_CAN_READ))
440 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
443 ret = rw_verify_area(READ, file, pos, count);
445 if (count > MAX_RW_COUNT)
446 count = MAX_RW_COUNT;
447 ret = __vfs_read(file, buf, count, pos);
449 fsnotify_access(file);
450 add_rchar(current, ret);
458 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
460 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
462 struct iov_iter iter;
465 init_sync_kiocb(&kiocb, filp);
466 kiocb.ki_pos = *ppos;
467 iov_iter_init(&iter, WRITE, &iov, 1, len);
469 ret = call_write_iter(filp, &kiocb, &iter);
470 BUG_ON(ret == -EIOCBQUEUED);
472 *ppos = kiocb.ki_pos;
476 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
479 if (file->f_op->write)
480 return file->f_op->write(file, p, count, pos);
481 else if (file->f_op->write_iter)
482 return new_sync_write(file, p, count, pos);
487 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
490 const char __user *p;
493 if (!(file->f_mode & FMODE_CAN_WRITE))
498 p = (__force const char __user *)buf;
499 if (count > MAX_RW_COUNT)
500 count = MAX_RW_COUNT;
501 ret = __vfs_write(file, p, count, pos);
504 fsnotify_modify(file);
505 add_wchar(current, ret);
510 EXPORT_SYMBOL(__kernel_write);
512 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
520 /* The cast to a user pointer is valid due to the set_fs() */
521 res = vfs_write(file, (__force const char __user *)buf, count, pos);
526 EXPORT_SYMBOL(kernel_write);
528 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
532 if (!(file->f_mode & FMODE_WRITE))
534 if (!(file->f_mode & FMODE_CAN_WRITE))
536 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
539 ret = rw_verify_area(WRITE, file, pos, count);
541 if (count > MAX_RW_COUNT)
542 count = MAX_RW_COUNT;
543 file_start_write(file);
544 ret = __vfs_write(file, buf, count, pos);
546 fsnotify_modify(file);
547 add_wchar(current, ret);
550 file_end_write(file);
556 static inline loff_t file_pos_read(struct file *file)
561 static inline void file_pos_write(struct file *file, loff_t pos)
566 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
568 struct fd f = fdget_pos(fd);
569 ssize_t ret = -EBADF;
572 loff_t pos = file_pos_read(f.file);
573 ret = vfs_read(f.file, buf, count, &pos);
575 file_pos_write(f.file, pos);
581 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
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_DEFINE3(write, unsigned int, fd, const char __user *, buf,
600 return ksys_write(fd, buf, count);
603 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
604 size_t, count, loff_t, pos)
607 ssize_t ret = -EBADF;
615 if (f.file->f_mode & FMODE_PREAD)
616 ret = vfs_read(f.file, buf, count, &pos);
623 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
624 size_t, count, loff_t, pos)
627 ssize_t ret = -EBADF;
635 if (f.file->f_mode & FMODE_PWRITE)
636 ret = vfs_write(f.file, buf, count, &pos);
643 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
644 loff_t *ppos, int type, rwf_t flags)
649 init_sync_kiocb(&kiocb, filp);
650 ret = kiocb_set_rw_flags(&kiocb, flags);
653 kiocb.ki_pos = *ppos;
656 ret = call_read_iter(filp, &kiocb, iter);
658 ret = call_write_iter(filp, &kiocb, iter);
659 BUG_ON(ret == -EIOCBQUEUED);
660 *ppos = kiocb.ki_pos;
664 /* Do it by hand, with file-ops */
665 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
666 loff_t *ppos, int type, rwf_t flags)
670 if (flags & ~RWF_HIPRI)
673 while (iov_iter_count(iter)) {
674 struct iovec iovec = iov_iter_iovec(iter);
678 nr = filp->f_op->read(filp, iovec.iov_base,
679 iovec.iov_len, ppos);
681 nr = filp->f_op->write(filp, iovec.iov_base,
682 iovec.iov_len, ppos);
691 if (nr != iovec.iov_len)
693 iov_iter_advance(iter, nr);
699 /* A write operation does a read from user space and vice versa */
700 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
703 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
704 * into the kernel and check that it is valid.
706 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
707 * @uvector: Pointer to the userspace array.
708 * @nr_segs: Number of elements in userspace array.
709 * @fast_segs: Number of elements in @fast_pointer.
710 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
711 * @ret_pointer: (output parameter) Pointer to a variable that will point to
712 * either @fast_pointer, a newly allocated kernel array, or NULL,
713 * depending on which array was used.
715 * This function copies an array of &struct iovec of @nr_segs from
716 * userspace into the kernel and checks that each element is valid (e.g.
717 * it does not point to a kernel address or cause overflow by being too
720 * As an optimization, the caller may provide a pointer to a small
721 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
722 * (the size of this array, or 0 if unused, should be given in @fast_segs).
724 * @ret_pointer will always point to the array that was used, so the
725 * caller must take care not to call kfree() on it e.g. in case the
726 * @fast_pointer array was used and it was allocated on the stack.
728 * Return: The total number of bytes covered by the iovec array on success
729 * or a negative error code on error.
731 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
732 unsigned long nr_segs, unsigned long fast_segs,
733 struct iovec *fast_pointer,
734 struct iovec **ret_pointer)
738 struct iovec *iov = fast_pointer;
741 * SuS says "The readv() function *may* fail if the iovcnt argument
742 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
743 * traditionally returned zero for zero segments, so...
751 * First get the "struct iovec" from user memory and
752 * verify all the pointers
754 if (nr_segs > UIO_MAXIOV) {
758 if (nr_segs > fast_segs) {
759 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
765 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
771 * According to the Single Unix Specification we should return EINVAL
772 * if an element length is < 0 when cast to ssize_t or if the
773 * total length would overflow the ssize_t return value of the
776 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
780 for (seg = 0; seg < nr_segs; seg++) {
781 void __user *buf = iov[seg].iov_base;
782 ssize_t len = (ssize_t)iov[seg].iov_len;
784 /* see if we we're about to use an invalid len or if
785 * it's about to overflow ssize_t */
791 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
795 if (len > MAX_RW_COUNT - ret) {
796 len = MAX_RW_COUNT - ret;
797 iov[seg].iov_len = len;
807 ssize_t compat_rw_copy_check_uvector(int type,
808 const struct compat_iovec __user *uvector, unsigned long nr_segs,
809 unsigned long fast_segs, struct iovec *fast_pointer,
810 struct iovec **ret_pointer)
812 compat_ssize_t tot_len;
813 struct iovec *iov = *ret_pointer = fast_pointer;
818 * SuS says "The readv() function *may* fail if the iovcnt argument
819 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
820 * traditionally returned zero for zero segments, so...
826 if (nr_segs > UIO_MAXIOV)
828 if (nr_segs > fast_segs) {
830 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
837 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
841 * Single unix specification:
842 * We should -EINVAL if an element length is not >= 0 and fitting an
845 * In Linux, the total length is limited to MAX_RW_COUNT, there is
846 * no overflow possibility.
850 for (seg = 0; seg < nr_segs; seg++) {
854 if (__get_user(len, &uvector->iov_len) ||
855 __get_user(buf, &uvector->iov_base)) {
859 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
862 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
866 if (len > MAX_RW_COUNT - tot_len)
867 len = MAX_RW_COUNT - tot_len;
869 iov->iov_base = compat_ptr(buf);
870 iov->iov_len = (compat_size_t) len;
881 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
882 loff_t *pos, rwf_t flags)
887 if (!(file->f_mode & FMODE_READ))
889 if (!(file->f_mode & FMODE_CAN_READ))
892 tot_len = iov_iter_count(iter);
895 ret = rw_verify_area(READ, file, pos, tot_len);
899 if (file->f_op->read_iter)
900 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
902 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
905 fsnotify_access(file);
909 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
912 if (!file->f_op->read_iter)
914 return do_iter_read(file, iter, ppos, flags);
916 EXPORT_SYMBOL(vfs_iter_read);
918 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
919 loff_t *pos, rwf_t flags)
924 if (!(file->f_mode & FMODE_WRITE))
926 if (!(file->f_mode & FMODE_CAN_WRITE))
929 tot_len = iov_iter_count(iter);
932 ret = rw_verify_area(WRITE, file, pos, tot_len);
936 if (file->f_op->write_iter)
937 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
939 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
941 fsnotify_modify(file);
945 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
948 if (!file->f_op->write_iter)
950 return do_iter_write(file, iter, ppos, flags);
952 EXPORT_SYMBOL(vfs_iter_write);
954 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
955 unsigned long vlen, loff_t *pos, rwf_t flags)
957 struct iovec iovstack[UIO_FASTIOV];
958 struct iovec *iov = iovstack;
959 struct iov_iter iter;
962 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
964 ret = do_iter_read(file, &iter, pos, flags);
971 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
972 unsigned long vlen, loff_t *pos, rwf_t flags)
974 struct iovec iovstack[UIO_FASTIOV];
975 struct iovec *iov = iovstack;
976 struct iov_iter iter;
979 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
981 file_start_write(file);
982 ret = do_iter_write(file, &iter, pos, flags);
983 file_end_write(file);
989 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
990 unsigned long vlen, rwf_t flags)
992 struct fd f = fdget_pos(fd);
993 ssize_t ret = -EBADF;
996 loff_t pos = file_pos_read(f.file);
997 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
999 file_pos_write(f.file, pos);
1004 add_rchar(current, ret);
1009 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1010 unsigned long vlen, rwf_t flags)
1012 struct fd f = fdget_pos(fd);
1013 ssize_t ret = -EBADF;
1016 loff_t pos = file_pos_read(f.file);
1017 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1019 file_pos_write(f.file, pos);
1024 add_wchar(current, ret);
1029 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1031 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1032 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1035 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1036 unsigned long vlen, loff_t pos, rwf_t flags)
1039 ssize_t ret = -EBADF;
1047 if (f.file->f_mode & FMODE_PREAD)
1048 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1053 add_rchar(current, ret);
1058 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1059 unsigned long vlen, loff_t pos, rwf_t flags)
1062 ssize_t ret = -EBADF;
1070 if (f.file->f_mode & FMODE_PWRITE)
1071 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1076 add_wchar(current, ret);
1081 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1082 unsigned long, vlen)
1084 return do_readv(fd, vec, vlen, 0);
1087 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1088 unsigned long, vlen)
1090 return do_writev(fd, vec, vlen, 0);
1093 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1094 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1096 loff_t pos = pos_from_hilo(pos_h, pos_l);
1098 return do_preadv(fd, vec, vlen, pos, 0);
1101 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1102 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1105 loff_t pos = pos_from_hilo(pos_h, pos_l);
1108 return do_readv(fd, vec, vlen, flags);
1110 return do_preadv(fd, vec, vlen, pos, flags);
1113 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1114 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1116 loff_t pos = pos_from_hilo(pos_h, pos_l);
1118 return do_pwritev(fd, vec, vlen, pos, 0);
1121 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1122 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1125 loff_t pos = pos_from_hilo(pos_h, pos_l);
1128 return do_writev(fd, vec, vlen, flags);
1130 return do_pwritev(fd, vec, vlen, pos, flags);
1133 #ifdef CONFIG_COMPAT
1134 static size_t compat_readv(struct file *file,
1135 const struct compat_iovec __user *vec,
1136 unsigned long vlen, loff_t *pos, rwf_t flags)
1138 struct iovec iovstack[UIO_FASTIOV];
1139 struct iovec *iov = iovstack;
1140 struct iov_iter iter;
1143 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1145 ret = do_iter_read(file, &iter, pos, flags);
1149 add_rchar(current, ret);
1154 static size_t do_compat_readv(compat_ulong_t fd,
1155 const struct compat_iovec __user *vec,
1156 compat_ulong_t vlen, rwf_t flags)
1158 struct fd f = fdget_pos(fd);
1164 pos = f.file->f_pos;
1165 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1167 f.file->f_pos = pos;
1173 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1174 const struct compat_iovec __user *,vec,
1175 compat_ulong_t, vlen)
1177 return do_compat_readv(fd, vec, vlen, 0);
1180 static long do_compat_preadv64(unsigned long fd,
1181 const struct compat_iovec __user *vec,
1182 unsigned long vlen, loff_t pos, rwf_t flags)
1193 if (f.file->f_mode & FMODE_PREAD)
1194 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1199 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1200 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1201 const struct compat_iovec __user *,vec,
1202 unsigned long, vlen, loff_t, pos)
1204 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1208 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1209 const struct compat_iovec __user *,vec,
1210 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1212 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1214 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1217 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1218 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1219 const struct compat_iovec __user *,vec,
1220 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1222 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1226 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1227 const struct compat_iovec __user *,vec,
1228 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1231 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1234 return do_compat_readv(fd, vec, vlen, flags);
1236 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1239 static size_t compat_writev(struct file *file,
1240 const struct compat_iovec __user *vec,
1241 unsigned long vlen, loff_t *pos, rwf_t flags)
1243 struct iovec iovstack[UIO_FASTIOV];
1244 struct iovec *iov = iovstack;
1245 struct iov_iter iter;
1248 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1250 file_start_write(file);
1251 ret = do_iter_write(file, &iter, pos, flags);
1252 file_end_write(file);
1256 add_wchar(current, ret);
1261 static size_t do_compat_writev(compat_ulong_t fd,
1262 const struct compat_iovec __user* vec,
1263 compat_ulong_t vlen, rwf_t flags)
1265 struct fd f = fdget_pos(fd);
1271 pos = f.file->f_pos;
1272 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1274 f.file->f_pos = pos;
1279 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1280 const struct compat_iovec __user *, vec,
1281 compat_ulong_t, vlen)
1283 return do_compat_writev(fd, vec, vlen, 0);
1286 static long do_compat_pwritev64(unsigned long fd,
1287 const struct compat_iovec __user *vec,
1288 unsigned long vlen, loff_t pos, rwf_t flags)
1299 if (f.file->f_mode & FMODE_PWRITE)
1300 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1305 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1306 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1307 const struct compat_iovec __user *,vec,
1308 unsigned long, vlen, loff_t, pos)
1310 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1314 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1315 const struct compat_iovec __user *,vec,
1316 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1318 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1320 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1323 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1324 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1325 const struct compat_iovec __user *,vec,
1326 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1328 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1332 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1333 const struct compat_iovec __user *,vec,
1334 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1336 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1339 return do_compat_writev(fd, vec, vlen, flags);
1341 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1346 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1347 size_t count, loff_t max)
1350 struct inode *in_inode, *out_inode;
1357 * Get input file, and verify that it is ok..
1363 if (!(in.file->f_mode & FMODE_READ))
1367 pos = in.file->f_pos;
1370 if (!(in.file->f_mode & FMODE_PREAD))
1373 retval = rw_verify_area(READ, in.file, &pos, count);
1376 if (count > MAX_RW_COUNT)
1377 count = MAX_RW_COUNT;
1380 * Get output file, and verify that it is ok..
1383 out = fdget(out_fd);
1386 if (!(out.file->f_mode & FMODE_WRITE))
1389 in_inode = file_inode(in.file);
1390 out_inode = file_inode(out.file);
1391 out_pos = out.file->f_pos;
1392 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1397 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1399 if (unlikely(pos + count > max)) {
1400 retval = -EOVERFLOW;
1409 * We need to debate whether we can enable this or not. The
1410 * man page documents EAGAIN return for the output at least,
1411 * and the application is arguably buggy if it doesn't expect
1412 * EAGAIN on a non-blocking file descriptor.
1414 if (in.file->f_flags & O_NONBLOCK)
1415 fl = SPLICE_F_NONBLOCK;
1417 file_start_write(out.file);
1418 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1419 file_end_write(out.file);
1422 add_rchar(current, retval);
1423 add_wchar(current, retval);
1424 fsnotify_access(in.file);
1425 fsnotify_modify(out.file);
1426 out.file->f_pos = out_pos;
1430 in.file->f_pos = pos;
1436 retval = -EOVERFLOW;
1446 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1453 if (unlikely(get_user(off, offset)))
1456 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1457 if (unlikely(put_user(pos, offset)))
1462 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1465 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1471 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1473 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1474 if (unlikely(put_user(pos, offset)))
1479 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1482 #ifdef CONFIG_COMPAT
1483 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1484 compat_off_t __user *, offset, compat_size_t, count)
1491 if (unlikely(get_user(off, offset)))
1494 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1495 if (unlikely(put_user(pos, offset)))
1500 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1503 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1504 compat_loff_t __user *, offset, compat_size_t, count)
1510 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1512 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1513 if (unlikely(put_user(pos, offset)))
1518 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1523 * copy_file_range() differs from regular file read and write in that it
1524 * specifically allows return partial success. When it does so is up to
1525 * the copy_file_range method.
1527 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1528 struct file *file_out, loff_t pos_out,
1529 size_t len, unsigned int flags)
1531 struct inode *inode_in = file_inode(file_in);
1532 struct inode *inode_out = file_inode(file_out);
1538 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1540 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1543 ret = rw_verify_area(READ, file_in, &pos_in, len);
1547 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1551 if (!(file_in->f_mode & FMODE_READ) ||
1552 !(file_out->f_mode & FMODE_WRITE) ||
1553 (file_out->f_flags & O_APPEND))
1556 /* this could be relaxed once a method supports cross-fs copies */
1557 if (inode_in->i_sb != inode_out->i_sb)
1563 file_start_write(file_out);
1566 * Try cloning first, this is supported by more file systems, and
1567 * more efficient if both clone and copy are supported (e.g. NFS).
1569 if (file_in->f_op->clone_file_range) {
1570 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1571 file_out, pos_out, len);
1578 if (file_out->f_op->copy_file_range) {
1579 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1580 pos_out, len, flags);
1581 if (ret != -EOPNOTSUPP)
1585 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1586 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1590 fsnotify_access(file_in);
1591 add_rchar(current, ret);
1592 fsnotify_modify(file_out);
1593 add_wchar(current, ret);
1599 file_end_write(file_out);
1603 EXPORT_SYMBOL(vfs_copy_file_range);
1605 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1606 int, fd_out, loff_t __user *, off_out,
1607 size_t, len, unsigned int, flags)
1613 ssize_t ret = -EBADF;
1615 f_in = fdget(fd_in);
1619 f_out = fdget(fd_out);
1625 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1628 pos_in = f_in.file->f_pos;
1632 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1635 pos_out = f_out.file->f_pos;
1638 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1645 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1648 f_in.file->f_pos = pos_in;
1652 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1655 f_out.file->f_pos = pos_out;
1667 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1669 struct inode *inode = file_inode(file);
1671 if (unlikely(pos < 0))
1674 if (unlikely((loff_t) (pos + len) < 0))
1677 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1678 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1681 retval = locks_mandatory_area(inode, file, pos, end,
1682 write ? F_WRLCK : F_RDLCK);
1687 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1691 * Check that the two inodes are eligible for cloning, the ranges make
1692 * sense, and then flush all dirty data. Caller must ensure that the
1693 * inodes have been locked against any other modifications.
1695 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1696 * the usual negative error code.
1698 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1699 struct inode *inode_out, loff_t pos_out,
1700 u64 *len, bool is_dedupe)
1702 loff_t bs = inode_out->i_sb->s_blocksize;
1705 bool same_inode = (inode_in == inode_out);
1708 /* Don't touch certain kinds of inodes */
1709 if (IS_IMMUTABLE(inode_out))
1712 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1715 /* Don't reflink dirs, pipes, sockets... */
1716 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1718 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1721 /* Are we going all the way to the end? */
1722 isize = i_size_read(inode_in);
1726 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1728 if (is_dedupe || pos_in == isize)
1732 *len = isize - pos_in;
1735 /* Ensure offsets don't wrap and the input is inside i_size */
1736 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1737 pos_in + *len > isize)
1740 /* Don't allow dedupe past EOF in the dest file */
1744 disize = i_size_read(inode_out);
1745 if (pos_out >= disize || pos_out + *len > disize)
1749 /* If we're linking to EOF, continue to the block boundary. */
1750 if (pos_in + *len == isize)
1751 blen = ALIGN(isize, bs) - pos_in;
1755 /* Only reflink if we're aligned to block boundaries */
1756 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1757 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1760 /* Don't allow overlapped reflink within the same file */
1762 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1766 /* Wait for the completion of any pending IOs on both files */
1767 inode_dio_wait(inode_in);
1769 inode_dio_wait(inode_out);
1771 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1772 pos_in, pos_in + *len - 1);
1776 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1777 pos_out, pos_out + *len - 1);
1782 * Check that the extents are the same.
1785 bool is_same = false;
1787 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1788 inode_out, pos_out, *len, &is_same);
1797 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1799 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1800 struct file *file_out, loff_t pos_out, u64 len)
1802 struct inode *inode_in = file_inode(file_in);
1803 struct inode *inode_out = file_inode(file_out);
1806 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1808 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1812 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1813 * the same mount. Practically, they only need to be on the same file
1816 if (inode_in->i_sb != inode_out->i_sb)
1819 if (!(file_in->f_mode & FMODE_READ) ||
1820 !(file_out->f_mode & FMODE_WRITE) ||
1821 (file_out->f_flags & O_APPEND))
1824 if (!file_in->f_op->clone_file_range)
1827 ret = clone_verify_area(file_in, pos_in, len, false);
1831 ret = clone_verify_area(file_out, pos_out, len, true);
1835 if (pos_in + len > i_size_read(inode_in))
1838 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1839 file_out, pos_out, len);
1841 fsnotify_access(file_in);
1842 fsnotify_modify(file_out);
1847 EXPORT_SYMBOL(vfs_clone_file_range);
1850 * Read a page's worth of file data into the page cache. Return the page
1853 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1855 struct address_space *mapping;
1859 n = offset >> PAGE_SHIFT;
1860 mapping = inode->i_mapping;
1861 page = read_mapping_page(mapping, n, NULL);
1864 if (!PageUptodate(page)) {
1866 return ERR_PTR(-EIO);
1873 * Compare extents of two files to see if they are the same.
1874 * Caller must have locked both inodes to prevent write races.
1876 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1877 struct inode *dest, loff_t destoff,
1878 loff_t len, bool *is_same)
1884 struct page *src_page;
1885 struct page *dest_page;
1893 src_poff = srcoff & (PAGE_SIZE - 1);
1894 dest_poff = destoff & (PAGE_SIZE - 1);
1895 cmp_len = min(PAGE_SIZE - src_poff,
1896 PAGE_SIZE - dest_poff);
1897 cmp_len = min(cmp_len, len);
1901 src_page = vfs_dedupe_get_page(src, srcoff);
1902 if (IS_ERR(src_page)) {
1903 error = PTR_ERR(src_page);
1906 dest_page = vfs_dedupe_get_page(dest, destoff);
1907 if (IS_ERR(dest_page)) {
1908 error = PTR_ERR(dest_page);
1909 unlock_page(src_page);
1913 src_addr = kmap_atomic(src_page);
1914 dest_addr = kmap_atomic(dest_page);
1916 flush_dcache_page(src_page);
1917 flush_dcache_page(dest_page);
1919 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1922 kunmap_atomic(dest_addr);
1923 kunmap_atomic(src_addr);
1924 unlock_page(dest_page);
1925 unlock_page(src_page);
1926 put_page(dest_page);
1943 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1945 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1947 struct file_dedupe_range_info *info;
1948 struct inode *src = file_inode(file);
1953 bool is_admin = capable(CAP_SYS_ADMIN);
1954 u16 count = same->dest_count;
1955 struct file *dst_file;
1959 if (!(file->f_mode & FMODE_READ))
1962 if (same->reserved1 || same->reserved2)
1965 off = same->src_offset;
1966 len = same->src_length;
1969 if (S_ISDIR(src->i_mode))
1973 if (!S_ISREG(src->i_mode))
1976 ret = clone_verify_area(file, off, len, false);
1981 if (off + len > i_size_read(src))
1984 /* pre-format output fields to sane values */
1985 for (i = 0; i < count; i++) {
1986 same->info[i].bytes_deduped = 0ULL;
1987 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1990 for (i = 0, info = same->info; i < count; i++, info++) {
1992 struct fd dst_fd = fdget(info->dest_fd);
1994 dst_file = dst_fd.file;
1996 info->status = -EBADF;
1999 dst = file_inode(dst_file);
2001 ret = mnt_want_write_file(dst_file);
2007 dst_off = info->dest_offset;
2008 ret = clone_verify_area(dst_file, dst_off, len, true);
2015 if (info->reserved) {
2016 info->status = -EINVAL;
2017 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2018 info->status = -EINVAL;
2019 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2020 info->status = -EXDEV;
2021 } else if (S_ISDIR(dst->i_mode)) {
2022 info->status = -EISDIR;
2023 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2024 info->status = -EINVAL;
2026 deduped = dst_file->f_op->dedupe_file_range(file, off,
2029 if (deduped == -EBADE)
2030 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2031 else if (deduped < 0)
2032 info->status = deduped;
2034 info->bytes_deduped += deduped;
2038 mnt_drop_write_file(dst_file);
2042 if (fatal_signal_pending(current))
2049 EXPORT_SYMBOL(vfs_dedupe_file_range);