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.
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.
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 static inline struct fd fdget_pos(int fd)
307 return __to_fd(__fdget_pos(fd));
310 static inline void fdput_pos(struct fd f)
312 if (f.flags & FDPUT_POS_UNLOCK)
313 mutex_unlock(&f.file->f_pos_lock);
317 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
320 struct fd f = fdget_pos(fd);
325 if (whence <= SEEK_MAX) {
326 loff_t res = vfs_llseek(f.file, offset, whence);
328 if (res != (loff_t)retval)
329 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
336 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
338 return sys_lseek(fd, offset, whence);
342 #ifdef __ARCH_WANT_SYS_LLSEEK
343 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
344 unsigned long, offset_low, loff_t __user *, result,
345 unsigned int, whence)
348 struct fd f = fdget_pos(fd);
355 if (whence > SEEK_MAX)
358 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
361 retval = (int)offset;
364 if (!copy_to_user(result, &offset, sizeof(offset)))
373 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
378 if (!file->f_op->read_iter)
381 init_sync_kiocb(&kiocb, file);
382 kiocb.ki_pos = *ppos;
385 ret = file->f_op->read_iter(&kiocb, iter);
386 BUG_ON(ret == -EIOCBQUEUED);
388 *ppos = kiocb.ki_pos;
391 EXPORT_SYMBOL(vfs_iter_read);
393 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
398 if (!file->f_op->write_iter)
401 init_sync_kiocb(&kiocb, file);
402 kiocb.ki_pos = *ppos;
405 ret = file->f_op->write_iter(&kiocb, iter);
406 BUG_ON(ret == -EIOCBQUEUED);
408 *ppos = kiocb.ki_pos;
411 EXPORT_SYMBOL(vfs_iter_write);
414 * rw_verify_area doesn't like huge counts. We limit
415 * them to something that fits in "int" so that others
416 * won't have to do range checks all the time.
418 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
422 int retval = -EINVAL;
424 inode = file_inode(file);
425 if (unlikely((ssize_t) count < 0))
428 if (unlikely(pos < 0)) {
429 if (!unsigned_offsets(file))
431 if (count >= -pos) /* both values are in 0..LLONG_MAX */
433 } else if (unlikely((loff_t) (pos + count) < 0)) {
434 if (!unsigned_offsets(file))
438 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
439 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
440 read_write == READ ? F_RDLCK : F_WRLCK);
444 retval = security_file_permission(file,
445 read_write == READ ? MAY_READ : MAY_WRITE);
448 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
451 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
453 struct iovec iov = { .iov_base = buf, .iov_len = len };
455 struct iov_iter iter;
458 init_sync_kiocb(&kiocb, filp);
459 kiocb.ki_pos = *ppos;
460 iov_iter_init(&iter, READ, &iov, 1, len);
462 ret = filp->f_op->read_iter(&kiocb, &iter);
463 BUG_ON(ret == -EIOCBQUEUED);
464 *ppos = kiocb.ki_pos;
468 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
471 if (file->f_op->read)
472 return file->f_op->read(file, buf, count, pos);
473 else if (file->f_op->read_iter)
474 return new_sync_read(file, buf, count, pos);
478 EXPORT_SYMBOL(__vfs_read);
480 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
484 if (!(file->f_mode & FMODE_READ))
486 if (!(file->f_mode & FMODE_CAN_READ))
488 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
491 ret = rw_verify_area(READ, file, pos, count);
494 ret = __vfs_read(file, buf, count, pos);
496 fsnotify_access(file);
497 add_rchar(current, ret);
505 EXPORT_SYMBOL(vfs_read);
507 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
509 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
511 struct iov_iter iter;
514 init_sync_kiocb(&kiocb, filp);
515 kiocb.ki_pos = *ppos;
516 iov_iter_init(&iter, WRITE, &iov, 1, len);
518 ret = filp->f_op->write_iter(&kiocb, &iter);
519 BUG_ON(ret == -EIOCBQUEUED);
521 *ppos = kiocb.ki_pos;
525 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
528 if (file->f_op->write)
529 return file->f_op->write(file, p, count, pos);
530 else if (file->f_op->write_iter)
531 return new_sync_write(file, p, count, pos);
535 EXPORT_SYMBOL(__vfs_write);
537 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
540 const char __user *p;
543 if (!(file->f_mode & FMODE_CAN_WRITE))
548 p = (__force const char __user *)buf;
549 if (count > MAX_RW_COUNT)
550 count = MAX_RW_COUNT;
551 ret = __vfs_write(file, p, count, pos);
554 fsnotify_modify(file);
555 add_wchar(current, ret);
561 EXPORT_SYMBOL(__kernel_write);
563 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
567 if (!(file->f_mode & FMODE_WRITE))
569 if (!(file->f_mode & FMODE_CAN_WRITE))
571 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
574 ret = rw_verify_area(WRITE, file, pos, count);
577 file_start_write(file);
578 ret = __vfs_write(file, buf, count, pos);
580 fsnotify_modify(file);
581 add_wchar(current, ret);
584 file_end_write(file);
590 EXPORT_SYMBOL(vfs_write);
592 static inline loff_t file_pos_read(struct file *file)
597 static inline void file_pos_write(struct file *file, loff_t pos)
602 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
604 struct fd f = fdget_pos(fd);
605 ssize_t ret = -EBADF;
608 loff_t pos = file_pos_read(f.file);
609 ret = vfs_read(f.file, buf, count, &pos);
611 file_pos_write(f.file, pos);
617 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
620 struct fd f = fdget_pos(fd);
621 ssize_t ret = -EBADF;
624 loff_t pos = file_pos_read(f.file);
625 ret = vfs_write(f.file, buf, count, &pos);
627 file_pos_write(f.file, pos);
634 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
635 size_t, count, loff_t, pos)
638 ssize_t ret = -EBADF;
646 if (f.file->f_mode & FMODE_PREAD)
647 ret = vfs_read(f.file, buf, count, &pos);
654 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
655 size_t, count, loff_t, pos)
658 ssize_t ret = -EBADF;
666 if (f.file->f_mode & FMODE_PWRITE)
667 ret = vfs_write(f.file, buf, count, &pos);
675 * Reduce an iovec's length in-place. Return the resulting number of segments
677 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
679 unsigned long seg = 0;
682 while (seg < nr_segs) {
684 if (len + iov->iov_len >= to) {
685 iov->iov_len = to - len;
693 EXPORT_SYMBOL(iov_shorten);
695 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
696 loff_t *ppos, iter_fn_t fn)
701 init_sync_kiocb(&kiocb, filp);
702 kiocb.ki_pos = *ppos;
704 ret = fn(&kiocb, iter);
705 BUG_ON(ret == -EIOCBQUEUED);
706 *ppos = kiocb.ki_pos;
710 /* Do it by hand, with file-ops */
711 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
712 loff_t *ppos, io_fn_t fn)
716 while (iov_iter_count(iter)) {
717 struct iovec iovec = iov_iter_iovec(iter);
720 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
728 if (nr != iovec.iov_len)
730 iov_iter_advance(iter, nr);
736 /* A write operation does a read from user space and vice versa */
737 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
739 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
740 unsigned long nr_segs, unsigned long fast_segs,
741 struct iovec *fast_pointer,
742 struct iovec **ret_pointer)
746 struct iovec *iov = fast_pointer;
749 * SuS says "The readv() function *may* fail if the iovcnt argument
750 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
751 * traditionally returned zero for zero segments, so...
759 * First get the "struct iovec" from user memory and
760 * verify all the pointers
762 if (nr_segs > UIO_MAXIOV) {
766 if (nr_segs > fast_segs) {
767 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
773 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
779 * According to the Single Unix Specification we should return EINVAL
780 * if an element length is < 0 when cast to ssize_t or if the
781 * total length would overflow the ssize_t return value of the
784 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
788 for (seg = 0; seg < nr_segs; seg++) {
789 void __user *buf = iov[seg].iov_base;
790 ssize_t len = (ssize_t)iov[seg].iov_len;
792 /* see if we we're about to use an invalid len or if
793 * it's about to overflow ssize_t */
799 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
803 if (len > MAX_RW_COUNT - ret) {
804 len = MAX_RW_COUNT - ret;
805 iov[seg].iov_len = len;
814 static ssize_t do_readv_writev(int type, struct file *file,
815 const struct iovec __user * uvector,
816 unsigned long nr_segs, loff_t *pos)
819 struct iovec iovstack[UIO_FASTIOV];
820 struct iovec *iov = iovstack;
821 struct iov_iter iter;
826 ret = import_iovec(type, uvector, nr_segs,
827 ARRAY_SIZE(iovstack), &iov, &iter);
831 tot_len = iov_iter_count(&iter);
834 ret = rw_verify_area(type, file, pos, tot_len);
839 fn = file->f_op->read;
840 iter_fn = file->f_op->read_iter;
842 fn = (io_fn_t)file->f_op->write;
843 iter_fn = file->f_op->write_iter;
844 file_start_write(file);
848 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
850 ret = do_loop_readv_writev(file, &iter, pos, fn);
853 file_end_write(file);
857 if ((ret + (type == READ)) > 0) {
859 fsnotify_access(file);
861 fsnotify_modify(file);
866 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
867 unsigned long vlen, loff_t *pos)
869 if (!(file->f_mode & FMODE_READ))
871 if (!(file->f_mode & FMODE_CAN_READ))
874 return do_readv_writev(READ, file, vec, vlen, pos);
877 EXPORT_SYMBOL(vfs_readv);
879 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
880 unsigned long vlen, loff_t *pos)
882 if (!(file->f_mode & FMODE_WRITE))
884 if (!(file->f_mode & FMODE_CAN_WRITE))
887 return do_readv_writev(WRITE, file, vec, vlen, pos);
890 EXPORT_SYMBOL(vfs_writev);
892 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
895 struct fd f = fdget_pos(fd);
896 ssize_t ret = -EBADF;
899 loff_t pos = file_pos_read(f.file);
900 ret = vfs_readv(f.file, vec, vlen, &pos);
902 file_pos_write(f.file, pos);
907 add_rchar(current, ret);
912 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
915 struct fd f = fdget_pos(fd);
916 ssize_t ret = -EBADF;
919 loff_t pos = file_pos_read(f.file);
920 ret = vfs_writev(f.file, vec, vlen, &pos);
922 file_pos_write(f.file, pos);
927 add_wchar(current, ret);
932 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
934 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
935 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
938 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
939 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
941 loff_t pos = pos_from_hilo(pos_h, pos_l);
943 ssize_t ret = -EBADF;
951 if (f.file->f_mode & FMODE_PREAD)
952 ret = vfs_readv(f.file, vec, vlen, &pos);
957 add_rchar(current, ret);
962 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
963 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
965 loff_t pos = pos_from_hilo(pos_h, pos_l);
967 ssize_t ret = -EBADF;
975 if (f.file->f_mode & FMODE_PWRITE)
976 ret = vfs_writev(f.file, vec, vlen, &pos);
981 add_wchar(current, ret);
988 static ssize_t compat_do_readv_writev(int type, struct file *file,
989 const struct compat_iovec __user *uvector,
990 unsigned long nr_segs, loff_t *pos)
992 compat_ssize_t tot_len;
993 struct iovec iovstack[UIO_FASTIOV];
994 struct iovec *iov = iovstack;
995 struct iov_iter iter;
1000 ret = compat_import_iovec(type, uvector, nr_segs,
1001 UIO_FASTIOV, &iov, &iter);
1005 tot_len = iov_iter_count(&iter);
1008 ret = rw_verify_area(type, file, pos, tot_len);
1013 fn = file->f_op->read;
1014 iter_fn = file->f_op->read_iter;
1016 fn = (io_fn_t)file->f_op->write;
1017 iter_fn = file->f_op->write_iter;
1018 file_start_write(file);
1022 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
1024 ret = do_loop_readv_writev(file, &iter, pos, fn);
1027 file_end_write(file);
1031 if ((ret + (type == READ)) > 0) {
1033 fsnotify_access(file);
1035 fsnotify_modify(file);
1040 static size_t compat_readv(struct file *file,
1041 const struct compat_iovec __user *vec,
1042 unsigned long vlen, loff_t *pos)
1044 ssize_t ret = -EBADF;
1046 if (!(file->f_mode & FMODE_READ))
1050 if (!(file->f_mode & FMODE_CAN_READ))
1053 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1057 add_rchar(current, ret);
1062 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1063 const struct compat_iovec __user *,vec,
1064 compat_ulong_t, vlen)
1066 struct fd f = fdget_pos(fd);
1072 pos = f.file->f_pos;
1073 ret = compat_readv(f.file, vec, vlen, &pos);
1075 f.file->f_pos = pos;
1080 static long __compat_sys_preadv64(unsigned long fd,
1081 const struct compat_iovec __user *vec,
1082 unsigned long vlen, loff_t pos)
1093 if (f.file->f_mode & FMODE_PREAD)
1094 ret = compat_readv(f.file, vec, vlen, &pos);
1099 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1100 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1101 const struct compat_iovec __user *,vec,
1102 unsigned long, vlen, loff_t, pos)
1104 return __compat_sys_preadv64(fd, vec, vlen, pos);
1108 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1109 const struct compat_iovec __user *,vec,
1110 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1112 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1114 return __compat_sys_preadv64(fd, vec, vlen, pos);
1117 static size_t compat_writev(struct file *file,
1118 const struct compat_iovec __user *vec,
1119 unsigned long vlen, loff_t *pos)
1121 ssize_t ret = -EBADF;
1123 if (!(file->f_mode & FMODE_WRITE))
1127 if (!(file->f_mode & FMODE_CAN_WRITE))
1130 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1134 add_wchar(current, ret);
1139 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1140 const struct compat_iovec __user *, vec,
1141 compat_ulong_t, vlen)
1143 struct fd f = fdget_pos(fd);
1149 pos = f.file->f_pos;
1150 ret = compat_writev(f.file, vec, vlen, &pos);
1152 f.file->f_pos = pos;
1157 static long __compat_sys_pwritev64(unsigned long fd,
1158 const struct compat_iovec __user *vec,
1159 unsigned long vlen, loff_t pos)
1170 if (f.file->f_mode & FMODE_PWRITE)
1171 ret = compat_writev(f.file, vec, vlen, &pos);
1176 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1177 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1178 const struct compat_iovec __user *,vec,
1179 unsigned long, vlen, loff_t, pos)
1181 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1185 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1186 const struct compat_iovec __user *,vec,
1187 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1189 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1191 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1195 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1196 size_t count, loff_t max)
1199 struct inode *in_inode, *out_inode;
1206 * Get input file, and verify that it is ok..
1212 if (!(in.file->f_mode & FMODE_READ))
1216 pos = in.file->f_pos;
1219 if (!(in.file->f_mode & FMODE_PREAD))
1222 retval = rw_verify_area(READ, in.file, &pos, count);
1228 * Get output file, and verify that it is ok..
1231 out = fdget(out_fd);
1234 if (!(out.file->f_mode & FMODE_WRITE))
1237 in_inode = file_inode(in.file);
1238 out_inode = file_inode(out.file);
1239 out_pos = out.file->f_pos;
1240 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1246 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1248 if (unlikely(pos + count > max)) {
1249 retval = -EOVERFLOW;
1258 * We need to debate whether we can enable this or not. The
1259 * man page documents EAGAIN return for the output at least,
1260 * and the application is arguably buggy if it doesn't expect
1261 * EAGAIN on a non-blocking file descriptor.
1263 if (in.file->f_flags & O_NONBLOCK)
1264 fl = SPLICE_F_NONBLOCK;
1266 file_start_write(out.file);
1267 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1268 file_end_write(out.file);
1271 add_rchar(current, retval);
1272 add_wchar(current, retval);
1273 fsnotify_access(in.file);
1274 fsnotify_modify(out.file);
1275 out.file->f_pos = out_pos;
1279 in.file->f_pos = pos;
1285 retval = -EOVERFLOW;
1295 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1302 if (unlikely(get_user(off, offset)))
1305 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1306 if (unlikely(put_user(pos, offset)))
1311 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1314 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1320 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1322 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1323 if (unlikely(put_user(pos, offset)))
1328 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1331 #ifdef CONFIG_COMPAT
1332 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1333 compat_off_t __user *, offset, compat_size_t, count)
1340 if (unlikely(get_user(off, offset)))
1343 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1344 if (unlikely(put_user(pos, offset)))
1349 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1352 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1353 compat_loff_t __user *, offset, compat_size_t, count)
1359 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1361 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1362 if (unlikely(put_user(pos, offset)))
1367 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1372 * copy_file_range() differs from regular file read and write in that it
1373 * specifically allows return partial success. When it does so is up to
1374 * the copy_file_range method.
1376 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1377 struct file *file_out, loff_t pos_out,
1378 size_t len, unsigned int flags)
1380 struct inode *inode_in = file_inode(file_in);
1381 struct inode *inode_out = file_inode(file_out);
1387 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1388 ret = rw_verify_area(READ, file_in, &pos_in, len);
1390 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1394 if (!(file_in->f_mode & FMODE_READ) ||
1395 !(file_out->f_mode & FMODE_WRITE) ||
1396 (file_out->f_flags & O_APPEND))
1399 /* this could be relaxed once a method supports cross-fs copies */
1400 if (inode_in->i_sb != inode_out->i_sb)
1406 ret = mnt_want_write_file(file_out);
1411 if (file_out->f_op->copy_file_range)
1412 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1413 pos_out, len, flags);
1414 if (ret == -EOPNOTSUPP)
1415 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1416 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1419 fsnotify_access(file_in);
1420 add_rchar(current, ret);
1421 fsnotify_modify(file_out);
1422 add_wchar(current, ret);
1427 mnt_drop_write_file(file_out);
1431 EXPORT_SYMBOL(vfs_copy_file_range);
1433 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1434 int, fd_out, loff_t __user *, off_out,
1435 size_t, len, unsigned int, flags)
1441 ssize_t ret = -EBADF;
1443 f_in = fdget(fd_in);
1447 f_out = fdget(fd_out);
1453 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1456 pos_in = f_in.file->f_pos;
1460 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1463 pos_out = f_out.file->f_pos;
1466 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1473 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1476 f_in.file->f_pos = pos_in;
1480 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1483 f_out.file->f_pos = pos_out;
1495 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1497 struct inode *inode = file_inode(file);
1499 if (unlikely(pos < 0))
1502 if (unlikely((loff_t) (pos + len) < 0))
1505 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1506 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1509 retval = locks_mandatory_area(inode, file, pos, end,
1510 write ? F_WRLCK : F_RDLCK);
1515 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1518 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1519 struct file *file_out, loff_t pos_out, u64 len)
1521 struct inode *inode_in = file_inode(file_in);
1522 struct inode *inode_out = file_inode(file_out);
1525 if (inode_in->i_sb != inode_out->i_sb ||
1526 file_in->f_path.mnt != file_out->f_path.mnt)
1529 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1531 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1534 if (!(file_in->f_mode & FMODE_READ) ||
1535 !(file_out->f_mode & FMODE_WRITE) ||
1536 (file_out->f_flags & O_APPEND))
1539 if (!file_in->f_op->clone_file_range)
1542 ret = clone_verify_area(file_in, pos_in, len, false);
1546 ret = clone_verify_area(file_out, pos_out, len, true);
1550 if (pos_in + len > i_size_read(inode_in))
1553 ret = mnt_want_write_file(file_out);
1557 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1558 file_out, pos_out, len);
1560 fsnotify_access(file_in);
1561 fsnotify_modify(file_out);
1564 mnt_drop_write_file(file_out);
1567 EXPORT_SYMBOL(vfs_clone_file_range);
1569 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1571 struct file_dedupe_range_info *info;
1572 struct inode *src = file_inode(file);
1577 bool is_admin = capable(CAP_SYS_ADMIN);
1578 u16 count = same->dest_count;
1579 struct file *dst_file;
1583 if (!(file->f_mode & FMODE_READ))
1586 if (same->reserved1 || same->reserved2)
1589 off = same->src_offset;
1590 len = same->src_length;
1593 if (S_ISDIR(src->i_mode))
1597 if (!S_ISREG(src->i_mode))
1600 ret = clone_verify_area(file, off, len, false);
1605 /* pre-format output fields to sane values */
1606 for (i = 0; i < count; i++) {
1607 same->info[i].bytes_deduped = 0ULL;
1608 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1611 for (i = 0, info = same->info; i < count; i++, info++) {
1613 struct fd dst_fd = fdget(info->dest_fd);
1615 dst_file = dst_fd.file;
1617 info->status = -EBADF;
1620 dst = file_inode(dst_file);
1622 ret = mnt_want_write_file(dst_file);
1628 dst_off = info->dest_offset;
1629 ret = clone_verify_area(dst_file, dst_off, len, true);
1636 if (info->reserved) {
1637 info->status = -EINVAL;
1638 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1639 info->status = -EINVAL;
1640 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1641 info->status = -EXDEV;
1642 } else if (S_ISDIR(dst->i_mode)) {
1643 info->status = -EISDIR;
1644 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1645 info->status = -EINVAL;
1647 deduped = dst_file->f_op->dedupe_file_range(file, off,
1650 if (deduped == -EBADE)
1651 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1652 else if (deduped < 0)
1653 info->status = deduped;
1655 info->bytes_deduped += deduped;
1659 mnt_drop_write_file(dst_file);
1663 if (fatal_signal_pending(current))
1670 EXPORT_SYMBOL(vfs_dedupe_file_range);