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>
22 #include <asm/uaccess.h>
23 #include <asm/unistd.h>
25 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
26 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
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 int 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.
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.
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 * noop_llseek - No Operation Performed llseek implementation
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
180 * This is an implementation of ->llseek useable for the rare special case when
181 * userspace expects the seek to succeed but the (device) file is actually not
182 * able to perform the seek. In this case you use noop_llseek() instead of
183 * falling back to the default implementation of ->llseek.
185 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
189 EXPORT_SYMBOL(noop_llseek);
191 loff_t no_llseek(struct file *file, loff_t offset, int whence)
195 EXPORT_SYMBOL(no_llseek);
197 loff_t default_llseek(struct file *file, loff_t offset, int whence)
199 struct inode *inode = file_inode(file);
202 mutex_lock(&inode->i_mutex);
205 offset += i_size_read(inode);
209 retval = file->f_pos;
212 offset += file->f_pos;
216 * In the generic case the entire file is data, so as
217 * long as offset isn't at the end of the file then the
220 if (offset >= inode->i_size) {
227 * There is a virtual hole at the end of the file, so
228 * as long as offset isn't i_size or larger, return
231 if (offset >= inode->i_size) {
235 offset = inode->i_size;
239 if (offset >= 0 || unsigned_offsets(file)) {
240 if (offset != file->f_pos) {
241 file->f_pos = offset;
247 mutex_unlock(&inode->i_mutex);
250 EXPORT_SYMBOL(default_llseek);
252 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
254 loff_t (*fn)(struct file *, loff_t, int);
257 if (file->f_mode & FMODE_LSEEK) {
258 if (file->f_op->llseek)
259 fn = file->f_op->llseek;
261 return fn(file, offset, whence);
263 EXPORT_SYMBOL(vfs_llseek);
265 static inline struct fd fdget_pos(int fd)
267 return __to_fd(__fdget_pos(fd));
270 static inline void fdput_pos(struct fd f)
272 if (f.flags & FDPUT_POS_UNLOCK)
273 mutex_unlock(&f.file->f_pos_lock);
277 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
280 struct fd f = fdget_pos(fd);
285 if (whence <= SEEK_MAX) {
286 loff_t res = vfs_llseek(f.file, offset, whence);
288 if (res != (loff_t)retval)
289 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
296 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
298 return sys_lseek(fd, offset, whence);
302 #ifdef __ARCH_WANT_SYS_LLSEEK
303 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
304 unsigned long, offset_low, loff_t __user *, result,
305 unsigned int, whence)
308 struct fd f = fdget_pos(fd);
315 if (whence > SEEK_MAX)
318 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
321 retval = (int)offset;
324 if (!copy_to_user(result, &offset, sizeof(offset)))
333 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
338 if (!file->f_op->read_iter)
341 init_sync_kiocb(&kiocb, file);
342 kiocb.ki_pos = *ppos;
345 ret = file->f_op->read_iter(&kiocb, iter);
346 BUG_ON(ret == -EIOCBQUEUED);
348 *ppos = kiocb.ki_pos;
351 EXPORT_SYMBOL(vfs_iter_read);
353 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
358 if (!file->f_op->write_iter)
361 init_sync_kiocb(&kiocb, file);
362 kiocb.ki_pos = *ppos;
365 ret = file->f_op->write_iter(&kiocb, iter);
366 BUG_ON(ret == -EIOCBQUEUED);
368 *ppos = kiocb.ki_pos;
371 EXPORT_SYMBOL(vfs_iter_write);
374 * rw_verify_area doesn't like huge counts. We limit
375 * them to something that fits in "int" so that others
376 * won't have to do range checks all the time.
378 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
382 int retval = -EINVAL;
384 inode = file_inode(file);
385 if (unlikely((ssize_t) count < 0))
388 if (unlikely(pos < 0)) {
389 if (!unsigned_offsets(file))
391 if (count >= -pos) /* both values are in 0..LLONG_MAX */
393 } else if (unlikely((loff_t) (pos + count) < 0)) {
394 if (!unsigned_offsets(file))
398 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
399 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
400 read_write == READ ? F_RDLCK : F_WRLCK);
404 retval = security_file_permission(file,
405 read_write == READ ? MAY_READ : MAY_WRITE);
408 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
411 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
413 struct iovec iov = { .iov_base = buf, .iov_len = len };
415 struct iov_iter iter;
418 init_sync_kiocb(&kiocb, filp);
419 kiocb.ki_pos = *ppos;
420 iov_iter_init(&iter, READ, &iov, 1, len);
422 ret = filp->f_op->read_iter(&kiocb, &iter);
423 BUG_ON(ret == -EIOCBQUEUED);
424 *ppos = kiocb.ki_pos;
428 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
431 if (file->f_op->read)
432 return file->f_op->read(file, buf, count, pos);
433 else if (file->f_op->read_iter)
434 return new_sync_read(file, buf, count, pos);
438 EXPORT_SYMBOL(__vfs_read);
440 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
444 if (!(file->f_mode & FMODE_READ))
446 if (!(file->f_mode & FMODE_CAN_READ))
448 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
451 ret = rw_verify_area(READ, file, pos, count);
454 ret = __vfs_read(file, buf, count, pos);
456 fsnotify_access(file);
457 add_rchar(current, ret);
465 EXPORT_SYMBOL(vfs_read);
467 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
469 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
471 struct iov_iter iter;
474 init_sync_kiocb(&kiocb, filp);
475 kiocb.ki_pos = *ppos;
476 iov_iter_init(&iter, WRITE, &iov, 1, len);
478 ret = filp->f_op->write_iter(&kiocb, &iter);
479 BUG_ON(ret == -EIOCBQUEUED);
481 *ppos = kiocb.ki_pos;
485 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
488 if (file->f_op->write)
489 return file->f_op->write(file, p, count, pos);
490 else if (file->f_op->write_iter)
491 return new_sync_write(file, p, count, pos);
495 EXPORT_SYMBOL(__vfs_write);
497 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
500 const char __user *p;
503 if (!(file->f_mode & FMODE_CAN_WRITE))
508 p = (__force const char __user *)buf;
509 if (count > MAX_RW_COUNT)
510 count = MAX_RW_COUNT;
511 ret = __vfs_write(file, p, count, pos);
514 fsnotify_modify(file);
515 add_wchar(current, ret);
521 EXPORT_SYMBOL(__kernel_write);
523 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
527 if (!(file->f_mode & FMODE_WRITE))
529 if (!(file->f_mode & FMODE_CAN_WRITE))
531 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
534 ret = rw_verify_area(WRITE, file, pos, count);
537 file_start_write(file);
538 ret = __vfs_write(file, buf, count, pos);
540 fsnotify_modify(file);
541 add_wchar(current, ret);
544 file_end_write(file);
550 EXPORT_SYMBOL(vfs_write);
552 static inline loff_t file_pos_read(struct file *file)
557 static inline void file_pos_write(struct file *file, loff_t pos)
562 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
564 struct fd f = fdget_pos(fd);
565 ssize_t ret = -EBADF;
568 loff_t pos = file_pos_read(f.file);
569 ret = vfs_read(f.file, buf, count, &pos);
571 file_pos_write(f.file, pos);
577 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
580 struct fd f = fdget_pos(fd);
581 ssize_t ret = -EBADF;
584 loff_t pos = file_pos_read(f.file);
585 ret = vfs_write(f.file, buf, count, &pos);
587 file_pos_write(f.file, pos);
594 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
595 size_t, count, loff_t, pos)
598 ssize_t ret = -EBADF;
606 if (f.file->f_mode & FMODE_PREAD)
607 ret = vfs_read(f.file, buf, count, &pos);
614 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
615 size_t, count, loff_t, pos)
618 ssize_t ret = -EBADF;
626 if (f.file->f_mode & FMODE_PWRITE)
627 ret = vfs_write(f.file, buf, count, &pos);
635 * Reduce an iovec's length in-place. Return the resulting number of segments
637 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
639 unsigned long seg = 0;
642 while (seg < nr_segs) {
644 if (len + iov->iov_len >= to) {
645 iov->iov_len = to - len;
653 EXPORT_SYMBOL(iov_shorten);
655 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
656 loff_t *ppos, iter_fn_t fn)
661 init_sync_kiocb(&kiocb, filp);
662 kiocb.ki_pos = *ppos;
664 ret = fn(&kiocb, iter);
665 BUG_ON(ret == -EIOCBQUEUED);
666 *ppos = kiocb.ki_pos;
670 /* Do it by hand, with file-ops */
671 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
672 loff_t *ppos, io_fn_t fn)
676 while (iov_iter_count(iter)) {
677 struct iovec iovec = iov_iter_iovec(iter);
680 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
688 if (nr != iovec.iov_len)
690 iov_iter_advance(iter, nr);
696 /* A write operation does a read from user space and vice versa */
697 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
699 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
700 unsigned long nr_segs, unsigned long fast_segs,
701 struct iovec *fast_pointer,
702 struct iovec **ret_pointer)
706 struct iovec *iov = fast_pointer;
709 * SuS says "The readv() function *may* fail if the iovcnt argument
710 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
711 * traditionally returned zero for zero segments, so...
719 * First get the "struct iovec" from user memory and
720 * verify all the pointers
722 if (nr_segs > UIO_MAXIOV) {
726 if (nr_segs > fast_segs) {
727 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
733 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
739 * According to the Single Unix Specification we should return EINVAL
740 * if an element length is < 0 when cast to ssize_t or if the
741 * total length would overflow the ssize_t return value of the
744 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
748 for (seg = 0; seg < nr_segs; seg++) {
749 void __user *buf = iov[seg].iov_base;
750 ssize_t len = (ssize_t)iov[seg].iov_len;
752 /* see if we we're about to use an invalid len or if
753 * it's about to overflow ssize_t */
759 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
763 if (len > MAX_RW_COUNT - ret) {
764 len = MAX_RW_COUNT - ret;
765 iov[seg].iov_len = len;
774 static ssize_t do_readv_writev(int type, struct file *file,
775 const struct iovec __user * uvector,
776 unsigned long nr_segs, loff_t *pos)
779 struct iovec iovstack[UIO_FASTIOV];
780 struct iovec *iov = iovstack;
781 struct iov_iter iter;
786 ret = import_iovec(type, uvector, nr_segs,
787 ARRAY_SIZE(iovstack), &iov, &iter);
791 tot_len = iov_iter_count(&iter);
794 ret = rw_verify_area(type, file, pos, tot_len);
799 fn = file->f_op->read;
800 iter_fn = file->f_op->read_iter;
802 fn = (io_fn_t)file->f_op->write;
803 iter_fn = file->f_op->write_iter;
804 file_start_write(file);
808 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
810 ret = do_loop_readv_writev(file, &iter, pos, fn);
813 file_end_write(file);
817 if ((ret + (type == READ)) > 0) {
819 fsnotify_access(file);
821 fsnotify_modify(file);
826 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
827 unsigned long vlen, loff_t *pos)
829 if (!(file->f_mode & FMODE_READ))
831 if (!(file->f_mode & FMODE_CAN_READ))
834 return do_readv_writev(READ, file, vec, vlen, pos);
837 EXPORT_SYMBOL(vfs_readv);
839 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
840 unsigned long vlen, loff_t *pos)
842 if (!(file->f_mode & FMODE_WRITE))
844 if (!(file->f_mode & FMODE_CAN_WRITE))
847 return do_readv_writev(WRITE, file, vec, vlen, pos);
850 EXPORT_SYMBOL(vfs_writev);
852 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
855 struct fd f = fdget_pos(fd);
856 ssize_t ret = -EBADF;
859 loff_t pos = file_pos_read(f.file);
860 ret = vfs_readv(f.file, vec, vlen, &pos);
862 file_pos_write(f.file, pos);
867 add_rchar(current, ret);
872 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
875 struct fd f = fdget_pos(fd);
876 ssize_t ret = -EBADF;
879 loff_t pos = file_pos_read(f.file);
880 ret = vfs_writev(f.file, vec, vlen, &pos);
882 file_pos_write(f.file, pos);
887 add_wchar(current, ret);
892 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
894 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
895 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
898 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
899 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
901 loff_t pos = pos_from_hilo(pos_h, pos_l);
903 ssize_t ret = -EBADF;
911 if (f.file->f_mode & FMODE_PREAD)
912 ret = vfs_readv(f.file, vec, vlen, &pos);
917 add_rchar(current, ret);
922 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
923 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
925 loff_t pos = pos_from_hilo(pos_h, pos_l);
927 ssize_t ret = -EBADF;
935 if (f.file->f_mode & FMODE_PWRITE)
936 ret = vfs_writev(f.file, vec, vlen, &pos);
941 add_wchar(current, ret);
948 static ssize_t compat_do_readv_writev(int type, struct file *file,
949 const struct compat_iovec __user *uvector,
950 unsigned long nr_segs, loff_t *pos)
952 compat_ssize_t tot_len;
953 struct iovec iovstack[UIO_FASTIOV];
954 struct iovec *iov = iovstack;
955 struct iov_iter iter;
960 ret = compat_import_iovec(type, uvector, nr_segs,
961 UIO_FASTIOV, &iov, &iter);
965 tot_len = iov_iter_count(&iter);
968 ret = rw_verify_area(type, file, pos, tot_len);
973 fn = file->f_op->read;
974 iter_fn = file->f_op->read_iter;
976 fn = (io_fn_t)file->f_op->write;
977 iter_fn = file->f_op->write_iter;
978 file_start_write(file);
982 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
984 ret = do_loop_readv_writev(file, &iter, pos, fn);
987 file_end_write(file);
991 if ((ret + (type == READ)) > 0) {
993 fsnotify_access(file);
995 fsnotify_modify(file);
1000 static size_t compat_readv(struct file *file,
1001 const struct compat_iovec __user *vec,
1002 unsigned long vlen, loff_t *pos)
1004 ssize_t ret = -EBADF;
1006 if (!(file->f_mode & FMODE_READ))
1010 if (!(file->f_mode & FMODE_CAN_READ))
1013 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1017 add_rchar(current, ret);
1022 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1023 const struct compat_iovec __user *,vec,
1024 compat_ulong_t, vlen)
1026 struct fd f = fdget_pos(fd);
1032 pos = f.file->f_pos;
1033 ret = compat_readv(f.file, vec, vlen, &pos);
1035 f.file->f_pos = pos;
1040 static long __compat_sys_preadv64(unsigned long fd,
1041 const struct compat_iovec __user *vec,
1042 unsigned long vlen, loff_t pos)
1053 if (f.file->f_mode & FMODE_PREAD)
1054 ret = compat_readv(f.file, vec, vlen, &pos);
1059 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1060 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1061 const struct compat_iovec __user *,vec,
1062 unsigned long, vlen, loff_t, pos)
1064 return __compat_sys_preadv64(fd, vec, vlen, pos);
1068 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1069 const struct compat_iovec __user *,vec,
1070 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1072 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1074 return __compat_sys_preadv64(fd, vec, vlen, pos);
1077 static size_t compat_writev(struct file *file,
1078 const struct compat_iovec __user *vec,
1079 unsigned long vlen, loff_t *pos)
1081 ssize_t ret = -EBADF;
1083 if (!(file->f_mode & FMODE_WRITE))
1087 if (!(file->f_mode & FMODE_CAN_WRITE))
1090 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1094 add_wchar(current, ret);
1099 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1100 const struct compat_iovec __user *, vec,
1101 compat_ulong_t, vlen)
1103 struct fd f = fdget_pos(fd);
1109 pos = f.file->f_pos;
1110 ret = compat_writev(f.file, vec, vlen, &pos);
1112 f.file->f_pos = pos;
1117 static long __compat_sys_pwritev64(unsigned long fd,
1118 const struct compat_iovec __user *vec,
1119 unsigned long vlen, loff_t pos)
1130 if (f.file->f_mode & FMODE_PWRITE)
1131 ret = compat_writev(f.file, vec, vlen, &pos);
1136 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1137 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1138 const struct compat_iovec __user *,vec,
1139 unsigned long, vlen, loff_t, pos)
1141 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1145 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1146 const struct compat_iovec __user *,vec,
1147 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1149 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1151 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1155 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1156 size_t count, loff_t max)
1159 struct inode *in_inode, *out_inode;
1166 * Get input file, and verify that it is ok..
1172 if (!(in.file->f_mode & FMODE_READ))
1176 pos = in.file->f_pos;
1179 if (!(in.file->f_mode & FMODE_PREAD))
1182 retval = rw_verify_area(READ, in.file, &pos, count);
1188 * Get output file, and verify that it is ok..
1191 out = fdget(out_fd);
1194 if (!(out.file->f_mode & FMODE_WRITE))
1197 in_inode = file_inode(in.file);
1198 out_inode = file_inode(out.file);
1199 out_pos = out.file->f_pos;
1200 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1206 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1208 if (unlikely(pos + count > max)) {
1209 retval = -EOVERFLOW;
1218 * We need to debate whether we can enable this or not. The
1219 * man page documents EAGAIN return for the output at least,
1220 * and the application is arguably buggy if it doesn't expect
1221 * EAGAIN on a non-blocking file descriptor.
1223 if (in.file->f_flags & O_NONBLOCK)
1224 fl = SPLICE_F_NONBLOCK;
1226 file_start_write(out.file);
1227 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1228 file_end_write(out.file);
1231 add_rchar(current, retval);
1232 add_wchar(current, retval);
1233 fsnotify_access(in.file);
1234 fsnotify_modify(out.file);
1235 out.file->f_pos = out_pos;
1239 in.file->f_pos = pos;
1245 retval = -EOVERFLOW;
1255 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1262 if (unlikely(get_user(off, offset)))
1265 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1266 if (unlikely(put_user(pos, offset)))
1271 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1274 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1280 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1282 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1283 if (unlikely(put_user(pos, offset)))
1288 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1291 #ifdef CONFIG_COMPAT
1292 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1293 compat_off_t __user *, offset, compat_size_t, count)
1300 if (unlikely(get_user(off, offset)))
1303 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1304 if (unlikely(put_user(pos, offset)))
1309 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1312 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1313 compat_loff_t __user *, offset, compat_size_t, count)
1319 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1321 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1322 if (unlikely(put_user(pos, offset)))
1327 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1332 * copy_file_range() differs from regular file read and write in that it
1333 * specifically allows return partial success. When it does so is up to
1334 * the copy_file_range method.
1336 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1337 struct file *file_out, loff_t pos_out,
1338 size_t len, unsigned int flags)
1340 struct inode *inode_in = file_inode(file_in);
1341 struct inode *inode_out = file_inode(file_out);
1347 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1348 ret = rw_verify_area(READ, file_in, &pos_in, len);
1350 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1354 if (!(file_in->f_mode & FMODE_READ) ||
1355 !(file_out->f_mode & FMODE_WRITE) ||
1356 (file_out->f_flags & O_APPEND))
1359 /* this could be relaxed once a method supports cross-fs copies */
1360 if (inode_in->i_sb != inode_out->i_sb)
1366 ret = mnt_want_write_file(file_out);
1371 if (file_out->f_op->copy_file_range)
1372 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1373 pos_out, len, flags);
1374 if (ret == -EOPNOTSUPP)
1375 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1376 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1379 fsnotify_access(file_in);
1380 add_rchar(current, ret);
1381 fsnotify_modify(file_out);
1382 add_wchar(current, ret);
1387 mnt_drop_write_file(file_out);
1391 EXPORT_SYMBOL(vfs_copy_file_range);
1393 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1394 int, fd_out, loff_t __user *, off_out,
1395 size_t, len, unsigned int, flags)
1401 ssize_t ret = -EBADF;
1403 f_in = fdget(fd_in);
1407 f_out = fdget(fd_out);
1413 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1416 pos_in = f_in.file->f_pos;
1420 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1423 pos_out = f_out.file->f_pos;
1426 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1433 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1436 f_in.file->f_pos = pos_in;
1440 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1443 f_out.file->f_pos = pos_out;
1455 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1457 struct inode *inode = file_inode(file);
1459 if (unlikely(pos < 0))
1462 if (unlikely((loff_t) (pos + len) < 0))
1465 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1466 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1469 retval = locks_mandatory_area(inode, file, pos, end,
1470 write ? F_WRLCK : F_RDLCK);
1475 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1478 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1479 struct file *file_out, loff_t pos_out, u64 len)
1481 struct inode *inode_in = file_inode(file_in);
1482 struct inode *inode_out = file_inode(file_out);
1485 if (inode_in->i_sb != inode_out->i_sb ||
1486 file_in->f_path.mnt != file_out->f_path.mnt)
1489 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1491 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1494 if (!(file_in->f_mode & FMODE_READ) ||
1495 !(file_out->f_mode & FMODE_WRITE) ||
1496 (file_out->f_flags & O_APPEND) ||
1497 !file_in->f_op->clone_file_range)
1500 ret = clone_verify_area(file_in, pos_in, len, false);
1504 ret = clone_verify_area(file_out, pos_out, len, true);
1508 if (pos_in + len > i_size_read(inode_in))
1511 ret = mnt_want_write_file(file_out);
1515 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1516 file_out, pos_out, len);
1518 fsnotify_access(file_in);
1519 fsnotify_modify(file_out);
1522 mnt_drop_write_file(file_out);
1525 EXPORT_SYMBOL(vfs_clone_file_range);