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/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include <linux/compat.h>
19 #include <linux/mount.h>
23 #include <asm/uaccess.h>
24 #include <asm/unistd.h>
26 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
29 const struct file_operations generic_ro_fops = {
30 .llseek = generic_file_llseek,
31 .read_iter = generic_file_read_iter,
32 .mmap = generic_file_readonly_mmap,
33 .splice_read = generic_file_splice_read,
36 EXPORT_SYMBOL(generic_ro_fops);
38 static inline int unsigned_offsets(struct file *file)
40 return file->f_mode & FMODE_UNSIGNED_OFFSET;
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
53 * Return the specified offset on success and -EINVAL on invalid offset.
55 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
57 if (offset < 0 && !unsigned_offsets(file))
62 if (offset != file->f_pos) {
68 EXPORT_SYMBOL(vfs_setpos);
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
87 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
88 loff_t maxsize, loff_t eof)
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
108 spin_lock(&file->f_lock);
109 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
110 spin_unlock(&file->f_lock);
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
117 if ((unsigned long long)offset >= eof)
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
125 if ((unsigned long long)offset >= eof)
131 return vfs_setpos(file, offset, maxsize);
133 EXPORT_SYMBOL(generic_file_llseek_size);
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
145 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
147 struct inode *inode = file->f_mapping->host;
149 return generic_file_llseek_size(file, offset, whence,
150 inode->i_sb->s_maxbytes,
153 EXPORT_SYMBOL(generic_file_llseek);
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
163 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
166 case SEEK_SET: case SEEK_CUR: case SEEK_END:
167 return generic_file_llseek_size(file, offset, whence,
173 EXPORT_SYMBOL(fixed_size_llseek);
176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
182 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
185 case SEEK_SET: case SEEK_CUR:
186 return generic_file_llseek_size(file, offset, whence,
192 EXPORT_SYMBOL(no_seek_end_llseek);
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
202 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
205 case SEEK_SET: case SEEK_CUR:
206 return generic_file_llseek_size(file, offset, whence,
212 EXPORT_SYMBOL(no_seek_end_llseek_size);
215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
218 * @whence: type of seek
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
225 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
229 EXPORT_SYMBOL(noop_llseek);
231 loff_t no_llseek(struct file *file, loff_t offset, int whence)
235 EXPORT_SYMBOL(no_llseek);
237 loff_t default_llseek(struct file *file, loff_t offset, int whence)
239 struct inode *inode = file_inode(file);
245 offset += i_size_read(inode);
249 retval = file->f_pos;
252 offset += file->f_pos;
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
260 if (offset >= inode->i_size) {
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
271 if (offset >= inode->i_size) {
275 offset = inode->i_size;
279 if (offset >= 0 || unsigned_offsets(file)) {
280 if (offset != file->f_pos) {
281 file->f_pos = offset;
290 EXPORT_SYMBOL(default_llseek);
292 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
294 loff_t (*fn)(struct file *, loff_t, int);
297 if (file->f_mode & FMODE_LSEEK) {
298 if (file->f_op->llseek)
299 fn = file->f_op->llseek;
301 return fn(file, offset, whence);
303 EXPORT_SYMBOL(vfs_llseek);
305 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
308 struct fd f = fdget_pos(fd);
313 if (whence <= SEEK_MAX) {
314 loff_t res = vfs_llseek(f.file, offset, whence);
316 if (res != (loff_t)retval)
317 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
324 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
326 return sys_lseek(fd, offset, whence);
330 #ifdef __ARCH_WANT_SYS_LLSEEK
331 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
332 unsigned long, offset_low, loff_t __user *, result,
333 unsigned int, whence)
336 struct fd f = fdget_pos(fd);
343 if (whence > SEEK_MAX)
346 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
349 retval = (int)offset;
352 if (!copy_to_user(result, &offset, sizeof(offset)))
361 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
366 if (!file->f_op->read_iter)
369 init_sync_kiocb(&kiocb, file);
370 kiocb.ki_pos = *ppos;
373 ret = file->f_op->read_iter(&kiocb, iter);
374 BUG_ON(ret == -EIOCBQUEUED);
376 *ppos = kiocb.ki_pos;
379 EXPORT_SYMBOL(vfs_iter_read);
381 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
386 if (!file->f_op->write_iter)
389 init_sync_kiocb(&kiocb, file);
390 kiocb.ki_pos = *ppos;
393 ret = file->f_op->write_iter(&kiocb, iter);
394 BUG_ON(ret == -EIOCBQUEUED);
396 *ppos = kiocb.ki_pos;
397 fsnotify_modify(file);
401 EXPORT_SYMBOL(vfs_iter_write);
403 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
407 int retval = -EINVAL;
409 inode = file_inode(file);
410 if (unlikely((ssize_t) count < 0))
413 if (unlikely(pos < 0)) {
414 if (!unsigned_offsets(file))
416 if (count >= -pos) /* both values are in 0..LLONG_MAX */
418 } else if (unlikely((loff_t) (pos + count) < 0)) {
419 if (!unsigned_offsets(file))
423 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
424 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
425 read_write == READ ? F_RDLCK : F_WRLCK);
429 return security_file_permission(file,
430 read_write == READ ? MAY_READ : MAY_WRITE);
433 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
435 struct iovec iov = { .iov_base = buf, .iov_len = len };
437 struct iov_iter iter;
440 init_sync_kiocb(&kiocb, filp);
441 kiocb.ki_pos = *ppos;
442 iov_iter_init(&iter, READ, &iov, 1, len);
444 ret = filp->f_op->read_iter(&kiocb, &iter);
445 BUG_ON(ret == -EIOCBQUEUED);
446 *ppos = kiocb.ki_pos;
450 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
453 if (file->f_op->read)
454 return file->f_op->read(file, buf, count, pos);
455 else if (file->f_op->read_iter)
456 return new_sync_read(file, buf, count, pos);
460 EXPORT_SYMBOL(__vfs_read);
462 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
466 if (!(file->f_mode & FMODE_READ))
468 if (!(file->f_mode & FMODE_CAN_READ))
470 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
473 ret = rw_verify_area(READ, file, pos, count);
475 if (count > MAX_RW_COUNT)
476 count = MAX_RW_COUNT;
477 ret = __vfs_read(file, buf, count, pos);
479 fsnotify_access(file);
480 add_rchar(current, ret);
488 EXPORT_SYMBOL(vfs_read);
490 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
492 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
494 struct iov_iter iter;
497 init_sync_kiocb(&kiocb, filp);
498 kiocb.ki_pos = *ppos;
499 iov_iter_init(&iter, WRITE, &iov, 1, len);
501 ret = filp->f_op->write_iter(&kiocb, &iter);
502 BUG_ON(ret == -EIOCBQUEUED);
504 *ppos = kiocb.ki_pos;
508 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
511 if (file->f_op->write)
512 return file->f_op->write(file, p, count, pos);
513 else if (file->f_op->write_iter)
514 return new_sync_write(file, p, count, pos);
518 EXPORT_SYMBOL(__vfs_write);
520 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
523 const char __user *p;
526 if (!(file->f_mode & FMODE_CAN_WRITE))
531 p = (__force const char __user *)buf;
532 if (count > MAX_RW_COUNT)
533 count = MAX_RW_COUNT;
534 ret = __vfs_write(file, p, count, pos);
537 fsnotify_modify(file);
538 add_wchar(current, ret);
544 EXPORT_SYMBOL(__kernel_write);
546 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
550 if (!(file->f_mode & FMODE_WRITE))
552 if (!(file->f_mode & FMODE_CAN_WRITE))
554 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
557 ret = rw_verify_area(WRITE, file, pos, count);
559 if (count > MAX_RW_COUNT)
560 count = MAX_RW_COUNT;
561 file_start_write(file);
562 ret = __vfs_write(file, buf, count, pos);
564 fsnotify_modify(file);
565 add_wchar(current, ret);
568 file_end_write(file);
574 EXPORT_SYMBOL(vfs_write);
576 static inline loff_t file_pos_read(struct file *file)
578 return file->f_mode & FMODE_STREAM ? 0 : file->f_pos;
581 static inline void file_pos_write(struct file *file, loff_t pos)
583 if ((file->f_mode & FMODE_STREAM) == 0)
587 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
589 struct fd f = fdget_pos(fd);
590 ssize_t ret = -EBADF;
593 loff_t pos = file_pos_read(f.file);
594 ret = vfs_read(f.file, buf, count, &pos);
596 file_pos_write(f.file, pos);
602 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
605 struct fd f = fdget_pos(fd);
606 ssize_t ret = -EBADF;
609 loff_t pos = file_pos_read(f.file);
610 ret = vfs_write(f.file, buf, count, &pos);
612 file_pos_write(f.file, pos);
619 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
620 size_t, count, loff_t, pos)
623 ssize_t ret = -EBADF;
631 if (f.file->f_mode & FMODE_PREAD)
632 ret = vfs_read(f.file, buf, count, &pos);
639 SYSCALL_DEFINE4(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);
660 * Reduce an iovec's length in-place. Return the resulting number of segments
662 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
664 unsigned long seg = 0;
667 while (seg < nr_segs) {
669 if (len + iov->iov_len >= to) {
670 iov->iov_len = to - len;
678 EXPORT_SYMBOL(iov_shorten);
680 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
681 loff_t *ppos, iter_fn_t fn, int flags)
686 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
689 init_sync_kiocb(&kiocb, filp);
690 if (flags & RWF_HIPRI)
691 kiocb.ki_flags |= IOCB_HIPRI;
692 if (flags & RWF_DSYNC)
693 kiocb.ki_flags |= IOCB_DSYNC;
694 if (flags & RWF_SYNC)
695 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
696 kiocb.ki_pos = *ppos;
698 ret = fn(&kiocb, iter);
699 BUG_ON(ret == -EIOCBQUEUED);
700 *ppos = kiocb.ki_pos;
704 /* Do it by hand, with file-ops */
705 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
706 loff_t *ppos, io_fn_t fn, int flags)
710 if (flags & ~RWF_HIPRI)
713 while (iov_iter_count(iter)) {
714 struct iovec iovec = iov_iter_iovec(iter);
717 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
725 if (nr != iovec.iov_len)
727 iov_iter_advance(iter, nr);
733 /* A write operation does a read from user space and vice versa */
734 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
737 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
738 * into the kernel and check that it is valid.
740 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
741 * @uvector: Pointer to the userspace array.
742 * @nr_segs: Number of elements in userspace array.
743 * @fast_segs: Number of elements in @fast_pointer.
744 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
745 * @ret_pointer: (output parameter) Pointer to a variable that will point to
746 * either @fast_pointer, a newly allocated kernel array, or NULL,
747 * depending on which array was used.
749 * This function copies an array of &struct iovec of @nr_segs from
750 * userspace into the kernel and checks that each element is valid (e.g.
751 * it does not point to a kernel address or cause overflow by being too
754 * As an optimization, the caller may provide a pointer to a small
755 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
756 * (the size of this array, or 0 if unused, should be given in @fast_segs).
758 * @ret_pointer will always point to the array that was used, so the
759 * caller must take care not to call kfree() on it e.g. in case the
760 * @fast_pointer array was used and it was allocated on the stack.
762 * Return: The total number of bytes covered by the iovec array on success
763 * or a negative error code on error.
765 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
766 unsigned long nr_segs, unsigned long fast_segs,
767 struct iovec *fast_pointer,
768 struct iovec **ret_pointer)
772 struct iovec *iov = fast_pointer;
775 * SuS says "The readv() function *may* fail if the iovcnt argument
776 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
777 * traditionally returned zero for zero segments, so...
785 * First get the "struct iovec" from user memory and
786 * verify all the pointers
788 if (nr_segs > UIO_MAXIOV) {
792 if (nr_segs > fast_segs) {
793 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
799 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
805 * According to the Single Unix Specification we should return EINVAL
806 * if an element length is < 0 when cast to ssize_t or if the
807 * total length would overflow the ssize_t return value of the
810 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
814 for (seg = 0; seg < nr_segs; seg++) {
815 void __user *buf = iov[seg].iov_base;
816 ssize_t len = (ssize_t)iov[seg].iov_len;
818 /* see if we we're about to use an invalid len or if
819 * it's about to overflow ssize_t */
825 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
829 if (len > MAX_RW_COUNT - ret) {
830 len = MAX_RW_COUNT - ret;
831 iov[seg].iov_len = len;
840 static ssize_t do_readv_writev(int type, struct file *file,
841 const struct iovec __user * uvector,
842 unsigned long nr_segs, loff_t *pos,
846 struct iovec iovstack[UIO_FASTIOV];
847 struct iovec *iov = iovstack;
848 struct iov_iter iter;
853 ret = import_iovec(type, uvector, nr_segs,
854 ARRAY_SIZE(iovstack), &iov, &iter);
858 tot_len = iov_iter_count(&iter);
861 ret = rw_verify_area(type, file, pos, tot_len);
866 fn = file->f_op->read;
867 iter_fn = file->f_op->read_iter;
869 fn = (io_fn_t)file->f_op->write;
870 iter_fn = file->f_op->write_iter;
871 file_start_write(file);
875 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
877 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
880 file_end_write(file);
884 if ((ret + (type == READ)) > 0) {
886 fsnotify_access(file);
888 fsnotify_modify(file);
893 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
894 unsigned long vlen, loff_t *pos, int flags)
896 if (!(file->f_mode & FMODE_READ))
898 if (!(file->f_mode & FMODE_CAN_READ))
901 return do_readv_writev(READ, file, vec, vlen, pos, flags);
904 EXPORT_SYMBOL(vfs_readv);
906 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
907 unsigned long vlen, loff_t *pos, int flags)
909 if (!(file->f_mode & FMODE_WRITE))
911 if (!(file->f_mode & FMODE_CAN_WRITE))
914 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
917 EXPORT_SYMBOL(vfs_writev);
919 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
920 unsigned long vlen, int flags)
922 struct fd f = fdget_pos(fd);
923 ssize_t ret = -EBADF;
926 loff_t pos = file_pos_read(f.file);
927 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
929 file_pos_write(f.file, pos);
934 add_rchar(current, ret);
939 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
940 unsigned long vlen, int flags)
942 struct fd f = fdget_pos(fd);
943 ssize_t ret = -EBADF;
946 loff_t pos = file_pos_read(f.file);
947 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
949 file_pos_write(f.file, pos);
954 add_wchar(current, ret);
959 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
961 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
962 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
965 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
966 unsigned long vlen, loff_t pos, int flags)
969 ssize_t ret = -EBADF;
977 if (f.file->f_mode & FMODE_PREAD)
978 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
983 add_rchar(current, ret);
988 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
989 unsigned long vlen, loff_t pos, int flags)
992 ssize_t ret = -EBADF;
1000 if (f.file->f_mode & FMODE_PWRITE)
1001 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1006 add_wchar(current, ret);
1011 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1012 unsigned long, vlen)
1014 return do_readv(fd, vec, vlen, 0);
1017 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1018 unsigned long, vlen)
1020 return do_writev(fd, vec, vlen, 0);
1023 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1024 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1026 loff_t pos = pos_from_hilo(pos_h, pos_l);
1028 return do_preadv(fd, vec, vlen, pos, 0);
1031 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1032 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1035 loff_t pos = pos_from_hilo(pos_h, pos_l);
1038 return do_readv(fd, vec, vlen, flags);
1040 return do_preadv(fd, vec, vlen, pos, flags);
1043 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1044 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1046 loff_t pos = pos_from_hilo(pos_h, pos_l);
1048 return do_pwritev(fd, vec, vlen, pos, 0);
1051 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1052 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1055 loff_t pos = pos_from_hilo(pos_h, pos_l);
1058 return do_writev(fd, vec, vlen, flags);
1060 return do_pwritev(fd, vec, vlen, pos, flags);
1063 #ifdef CONFIG_COMPAT
1065 static ssize_t compat_do_readv_writev(int type, struct file *file,
1066 const struct compat_iovec __user *uvector,
1067 unsigned long nr_segs, loff_t *pos,
1070 compat_ssize_t tot_len;
1071 struct iovec iovstack[UIO_FASTIOV];
1072 struct iovec *iov = iovstack;
1073 struct iov_iter iter;
1078 ret = compat_import_iovec(type, uvector, nr_segs,
1079 UIO_FASTIOV, &iov, &iter);
1083 tot_len = iov_iter_count(&iter);
1086 ret = rw_verify_area(type, file, pos, tot_len);
1091 fn = file->f_op->read;
1092 iter_fn = file->f_op->read_iter;
1094 fn = (io_fn_t)file->f_op->write;
1095 iter_fn = file->f_op->write_iter;
1096 file_start_write(file);
1100 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1102 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1105 file_end_write(file);
1109 if ((ret + (type == READ)) > 0) {
1111 fsnotify_access(file);
1113 fsnotify_modify(file);
1118 static size_t compat_readv(struct file *file,
1119 const struct compat_iovec __user *vec,
1120 unsigned long vlen, loff_t *pos, int flags)
1122 ssize_t ret = -EBADF;
1124 if (!(file->f_mode & FMODE_READ))
1128 if (!(file->f_mode & FMODE_CAN_READ))
1131 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1135 add_rchar(current, ret);
1140 static size_t do_compat_readv(compat_ulong_t fd,
1141 const struct compat_iovec __user *vec,
1142 compat_ulong_t vlen, int flags)
1144 struct fd f = fdget_pos(fd);
1150 pos = f.file->f_pos;
1151 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1153 f.file->f_pos = pos;
1159 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1160 const struct compat_iovec __user *,vec,
1161 compat_ulong_t, vlen)
1163 return do_compat_readv(fd, vec, vlen, 0);
1166 static long do_compat_preadv64(unsigned long fd,
1167 const struct compat_iovec __user *vec,
1168 unsigned long vlen, loff_t pos, int flags)
1179 if (f.file->f_mode & FMODE_PREAD)
1180 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1185 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1186 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1187 const struct compat_iovec __user *,vec,
1188 unsigned long, vlen, loff_t, pos)
1190 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1194 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1195 const struct compat_iovec __user *,vec,
1196 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1198 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1200 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1203 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1204 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1205 const struct compat_iovec __user *,vec,
1206 unsigned long, vlen, loff_t, pos, int, flags)
1209 return do_compat_readv(fd, vec, vlen, flags);
1211 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1215 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1216 const struct compat_iovec __user *,vec,
1217 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1220 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1223 return do_compat_readv(fd, vec, vlen, flags);
1225 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1228 static size_t compat_writev(struct file *file,
1229 const struct compat_iovec __user *vec,
1230 unsigned long vlen, loff_t *pos, int flags)
1232 ssize_t ret = -EBADF;
1234 if (!(file->f_mode & FMODE_WRITE))
1238 if (!(file->f_mode & FMODE_CAN_WRITE))
1241 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, flags);
1245 add_wchar(current, ret);
1250 static size_t do_compat_writev(compat_ulong_t fd,
1251 const struct compat_iovec __user* vec,
1252 compat_ulong_t vlen, int flags)
1254 struct fd f = fdget_pos(fd);
1260 pos = f.file->f_pos;
1261 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1263 f.file->f_pos = pos;
1268 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1269 const struct compat_iovec __user *, vec,
1270 compat_ulong_t, vlen)
1272 return do_compat_writev(fd, vec, vlen, 0);
1275 static long do_compat_pwritev64(unsigned long fd,
1276 const struct compat_iovec __user *vec,
1277 unsigned long vlen, loff_t pos, int flags)
1288 if (f.file->f_mode & FMODE_PWRITE)
1289 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1294 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1295 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1296 const struct compat_iovec __user *,vec,
1297 unsigned long, vlen, loff_t, pos)
1299 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1303 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1304 const struct compat_iovec __user *,vec,
1305 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1307 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1309 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1312 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1313 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1314 const struct compat_iovec __user *,vec,
1315 unsigned long, vlen, loff_t, pos, int, flags)
1318 return do_compat_writev(fd, vec, vlen, flags);
1320 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1324 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1325 const struct compat_iovec __user *,vec,
1326 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1328 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1331 return do_compat_writev(fd, vec, vlen, flags);
1333 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1338 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1339 size_t count, loff_t max)
1342 struct inode *in_inode, *out_inode;
1349 * Get input file, and verify that it is ok..
1355 if (!(in.file->f_mode & FMODE_READ))
1359 pos = in.file->f_pos;
1362 if (!(in.file->f_mode & FMODE_PREAD))
1365 retval = rw_verify_area(READ, in.file, &pos, count);
1368 if (count > MAX_RW_COUNT)
1369 count = MAX_RW_COUNT;
1372 * Get output file, and verify that it is ok..
1375 out = fdget(out_fd);
1378 if (!(out.file->f_mode & FMODE_WRITE))
1381 in_inode = file_inode(in.file);
1382 out_inode = file_inode(out.file);
1383 out_pos = out.file->f_pos;
1384 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1389 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1391 if (unlikely(pos + count > max)) {
1392 retval = -EOVERFLOW;
1401 * We need to debate whether we can enable this or not. The
1402 * man page documents EAGAIN return for the output at least,
1403 * and the application is arguably buggy if it doesn't expect
1404 * EAGAIN on a non-blocking file descriptor.
1406 if (in.file->f_flags & O_NONBLOCK)
1407 fl = SPLICE_F_NONBLOCK;
1409 file_start_write(out.file);
1410 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1411 file_end_write(out.file);
1414 add_rchar(current, retval);
1415 add_wchar(current, retval);
1416 fsnotify_access(in.file);
1417 fsnotify_modify(out.file);
1418 out.file->f_pos = out_pos;
1422 in.file->f_pos = pos;
1428 retval = -EOVERFLOW;
1438 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1445 if (unlikely(get_user(off, offset)))
1448 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1449 if (unlikely(put_user(pos, offset)))
1454 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1457 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1463 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1465 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1466 if (unlikely(put_user(pos, offset)))
1471 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1474 #ifdef CONFIG_COMPAT
1475 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1476 compat_off_t __user *, offset, compat_size_t, count)
1483 if (unlikely(get_user(off, offset)))
1486 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1487 if (unlikely(put_user(pos, offset)))
1492 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1495 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1496 compat_loff_t __user *, offset, compat_size_t, count)
1502 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1504 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1505 if (unlikely(put_user(pos, offset)))
1510 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1515 * copy_file_range() differs from regular file read and write in that it
1516 * specifically allows return partial success. When it does so is up to
1517 * the copy_file_range method.
1519 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1520 struct file *file_out, loff_t pos_out,
1521 size_t len, unsigned int flags)
1523 struct inode *inode_in = file_inode(file_in);
1524 struct inode *inode_out = file_inode(file_out);
1530 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1532 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1535 ret = rw_verify_area(READ, file_in, &pos_in, len);
1539 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1543 if (!(file_in->f_mode & FMODE_READ) ||
1544 !(file_out->f_mode & FMODE_WRITE) ||
1545 (file_out->f_flags & O_APPEND))
1548 /* this could be relaxed once a method supports cross-fs copies */
1549 if (inode_in->i_sb != inode_out->i_sb)
1555 ret = mnt_want_write_file(file_out);
1560 if (file_out->f_op->copy_file_range)
1561 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1562 pos_out, len, flags);
1563 if (ret == -EOPNOTSUPP)
1564 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1565 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1568 fsnotify_access(file_in);
1569 add_rchar(current, ret);
1570 fsnotify_modify(file_out);
1571 add_wchar(current, ret);
1576 mnt_drop_write_file(file_out);
1580 EXPORT_SYMBOL(vfs_copy_file_range);
1582 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1583 int, fd_out, loff_t __user *, off_out,
1584 size_t, len, unsigned int, flags)
1590 ssize_t ret = -EBADF;
1592 f_in = fdget(fd_in);
1596 f_out = fdget(fd_out);
1602 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1605 pos_in = f_in.file->f_pos;
1609 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1612 pos_out = f_out.file->f_pos;
1615 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1622 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1625 f_in.file->f_pos = pos_in;
1629 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1632 f_out.file->f_pos = pos_out;
1644 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1646 struct inode *inode = file_inode(file);
1648 if (unlikely(pos < 0))
1651 if (unlikely((loff_t) (pos + len) < 0))
1654 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1655 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1658 retval = locks_mandatory_area(inode, file, pos, end,
1659 write ? F_WRLCK : F_RDLCK);
1664 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1667 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1668 struct file *file_out, loff_t pos_out, u64 len)
1670 struct inode *inode_in = file_inode(file_in);
1671 struct inode *inode_out = file_inode(file_out);
1674 if (inode_in->i_sb != inode_out->i_sb ||
1675 file_in->f_path.mnt != file_out->f_path.mnt)
1678 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1680 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1683 if (!(file_in->f_mode & FMODE_READ) ||
1684 !(file_out->f_mode & FMODE_WRITE) ||
1685 (file_out->f_flags & O_APPEND))
1688 if (!file_in->f_op->clone_file_range)
1691 ret = clone_verify_area(file_in, pos_in, len, false);
1695 ret = clone_verify_area(file_out, pos_out, len, true);
1699 if (pos_in + len > i_size_read(inode_in))
1702 ret = mnt_want_write_file(file_out);
1706 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1707 file_out, pos_out, len);
1709 fsnotify_access(file_in);
1710 fsnotify_modify(file_out);
1713 mnt_drop_write_file(file_out);
1716 EXPORT_SYMBOL(vfs_clone_file_range);
1718 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1720 struct file_dedupe_range_info *info;
1721 struct inode *src = file_inode(file);
1726 bool is_admin = capable(CAP_SYS_ADMIN);
1727 u16 count = same->dest_count;
1728 struct file *dst_file;
1732 if (!(file->f_mode & FMODE_READ))
1735 if (same->reserved1 || same->reserved2)
1738 off = same->src_offset;
1739 len = same->src_length;
1742 if (S_ISDIR(src->i_mode))
1746 if (!S_ISREG(src->i_mode))
1749 ret = clone_verify_area(file, off, len, false);
1754 /* pre-format output fields to sane values */
1755 for (i = 0; i < count; i++) {
1756 same->info[i].bytes_deduped = 0ULL;
1757 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1760 for (i = 0, info = same->info; i < count; i++, info++) {
1762 struct fd dst_fd = fdget(info->dest_fd);
1764 dst_file = dst_fd.file;
1766 info->status = -EBADF;
1769 dst = file_inode(dst_file);
1771 ret = mnt_want_write_file(dst_file);
1777 dst_off = info->dest_offset;
1778 ret = clone_verify_area(dst_file, dst_off, len, true);
1785 if (info->reserved) {
1786 info->status = -EINVAL;
1787 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1788 info->status = -EINVAL;
1789 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1790 info->status = -EXDEV;
1791 } else if (S_ISDIR(dst->i_mode)) {
1792 info->status = -EISDIR;
1793 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1794 info->status = -EINVAL;
1796 deduped = dst_file->f_op->dedupe_file_range(file, off,
1799 if (deduped == -EBADE)
1800 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1801 else if (deduped < 0)
1802 info->status = deduped;
1804 info->bytes_deduped += deduped;
1808 mnt_drop_write_file(dst_file);
1812 if (fatal_signal_pending(current))
1819 EXPORT_SYMBOL(vfs_dedupe_file_range);