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 static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
307 struct fd f = fdget_pos(fd);
312 if (whence <= SEEK_MAX) {
313 loff_t res = vfs_llseek(f.file, offset, whence);
315 if (res != (loff_t)retval)
316 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
324 return ksys_lseek(fd, offset, whence);
328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
330 return ksys_lseek(fd, offset, whence);
334 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
335 defined(__ARCH_WANT_SYS_LLSEEK)
336 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
337 unsigned long, offset_low, loff_t __user *, result,
338 unsigned int, whence)
341 struct fd f = fdget_pos(fd);
348 if (whence > SEEK_MAX)
351 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
354 retval = (int)offset;
357 if (!copy_to_user(result, &offset, sizeof(offset)))
366 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
369 int retval = -EINVAL;
371 inode = file_inode(file);
372 if (unlikely((ssize_t) count < 0))
376 * ranged mandatory locking does not apply to streams - it makes sense
377 * only for files where position has a meaning.
382 if (unlikely(pos < 0)) {
383 if (!unsigned_offsets(file))
385 if (count >= -pos) /* both values are in 0..LLONG_MAX */
387 } else if (unlikely((loff_t) (pos + count) < 0)) {
388 if (!unsigned_offsets(file))
392 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
393 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
394 read_write == READ ? F_RDLCK : F_WRLCK);
400 return security_file_permission(file,
401 read_write == READ ? MAY_READ : MAY_WRITE);
404 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
406 struct iovec iov = { .iov_base = buf, .iov_len = len };
408 struct iov_iter iter;
411 init_sync_kiocb(&kiocb, filp);
412 kiocb.ki_pos = (ppos ? *ppos : 0);
413 iov_iter_init(&iter, READ, &iov, 1, len);
415 ret = call_read_iter(filp, &kiocb, &iter);
416 BUG_ON(ret == -EIOCBQUEUED);
418 *ppos = kiocb.ki_pos;
422 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
424 mm_segment_t old_fs = get_fs();
427 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
429 if (!(file->f_mode & FMODE_CAN_READ))
432 if (count > MAX_RW_COUNT)
433 count = MAX_RW_COUNT;
435 if (file->f_op->read)
436 ret = file->f_op->read(file, (void __user *)buf, count, pos);
437 else if (file->f_op->read_iter)
438 ret = new_sync_read(file, (void __user *)buf, count, pos);
443 fsnotify_access(file);
444 add_rchar(current, ret);
450 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
454 ret = rw_verify_area(READ, file, pos, count);
457 return __kernel_read(file, buf, count, pos);
459 EXPORT_SYMBOL(kernel_read);
461 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
465 if (!(file->f_mode & FMODE_READ))
467 if (!(file->f_mode & FMODE_CAN_READ))
469 if (unlikely(!access_ok(buf, count)))
472 ret = rw_verify_area(READ, file, pos, count);
475 if (count > MAX_RW_COUNT)
476 count = MAX_RW_COUNT;
478 if (file->f_op->read)
479 ret = file->f_op->read(file, buf, count, pos);
480 else if (file->f_op->read_iter)
481 ret = new_sync_read(file, buf, count, pos);
485 fsnotify_access(file);
486 add_rchar(current, ret);
492 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
494 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
496 struct iov_iter iter;
499 init_sync_kiocb(&kiocb, filp);
500 kiocb.ki_pos = (ppos ? *ppos : 0);
501 iov_iter_init(&iter, WRITE, &iov, 1, len);
503 ret = call_write_iter(filp, &kiocb, &iter);
504 BUG_ON(ret == -EIOCBQUEUED);
506 *ppos = kiocb.ki_pos;
510 /* caller is responsible for file_start_write/file_end_write */
511 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
514 const char __user *p;
517 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
519 if (!(file->f_mode & FMODE_CAN_WRITE))
524 p = (__force const char __user *)buf;
525 if (count > MAX_RW_COUNT)
526 count = MAX_RW_COUNT;
527 if (file->f_op->write)
528 ret = file->f_op->write(file, p, count, pos);
529 else if (file->f_op->write_iter)
530 ret = new_sync_write(file, p, count, pos);
535 fsnotify_modify(file);
536 add_wchar(current, ret);
542 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
543 * but autofs is one of the few internal kernel users that actually
544 * wants this _and_ can be built as a module. So we need to export
545 * this symbol for autofs, even though it really isn't appropriate
546 * for any other kernel modules.
548 EXPORT_SYMBOL_GPL(__kernel_write);
550 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
555 ret = rw_verify_area(WRITE, file, pos, count);
559 file_start_write(file);
560 ret = __kernel_write(file, buf, count, pos);
561 file_end_write(file);
564 EXPORT_SYMBOL(kernel_write);
566 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
570 if (!(file->f_mode & FMODE_WRITE))
572 if (!(file->f_mode & FMODE_CAN_WRITE))
574 if (unlikely(!access_ok(buf, count)))
577 ret = rw_verify_area(WRITE, file, pos, count);
580 if (count > MAX_RW_COUNT)
581 count = MAX_RW_COUNT;
582 file_start_write(file);
583 if (file->f_op->write)
584 ret = file->f_op->write(file, buf, count, pos);
585 else if (file->f_op->write_iter)
586 ret = new_sync_write(file, buf, count, pos);
590 fsnotify_modify(file);
591 add_wchar(current, ret);
594 file_end_write(file);
598 /* file_ppos returns &file->f_pos or NULL if file is stream */
599 static inline loff_t *file_ppos(struct file *file)
601 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
604 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
606 struct fd f = fdget_pos(fd);
607 ssize_t ret = -EBADF;
610 loff_t pos, *ppos = file_ppos(f.file);
615 ret = vfs_read(f.file, buf, count, ppos);
616 if (ret >= 0 && ppos)
623 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
625 return ksys_read(fd, buf, count);
628 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
630 struct fd f = fdget_pos(fd);
631 ssize_t ret = -EBADF;
634 loff_t pos, *ppos = file_ppos(f.file);
639 ret = vfs_write(f.file, buf, count, ppos);
640 if (ret >= 0 && ppos)
648 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
651 return ksys_write(fd, buf, count);
654 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
658 ssize_t ret = -EBADF;
666 if (f.file->f_mode & FMODE_PREAD)
667 ret = vfs_read(f.file, buf, count, &pos);
674 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
675 size_t, count, loff_t, pos)
677 return ksys_pread64(fd, buf, count, pos);
680 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
681 size_t count, loff_t pos)
684 ssize_t ret = -EBADF;
692 if (f.file->f_mode & FMODE_PWRITE)
693 ret = vfs_write(f.file, buf, count, &pos);
700 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
701 size_t, count, loff_t, pos)
703 return ksys_pwrite64(fd, buf, count, pos);
706 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
707 loff_t *ppos, int type, rwf_t flags)
712 init_sync_kiocb(&kiocb, filp);
713 ret = kiocb_set_rw_flags(&kiocb, flags);
716 kiocb.ki_pos = (ppos ? *ppos : 0);
719 ret = call_read_iter(filp, &kiocb, iter);
721 ret = call_write_iter(filp, &kiocb, iter);
722 BUG_ON(ret == -EIOCBQUEUED);
724 *ppos = kiocb.ki_pos;
728 /* Do it by hand, with file-ops */
729 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
730 loff_t *ppos, int type, rwf_t flags)
734 if (flags & ~RWF_HIPRI)
737 while (iov_iter_count(iter)) {
738 struct iovec iovec = iov_iter_iovec(iter);
742 nr = filp->f_op->read(filp, iovec.iov_base,
743 iovec.iov_len, ppos);
745 nr = filp->f_op->write(filp, iovec.iov_base,
746 iovec.iov_len, ppos);
755 if (nr != iovec.iov_len)
757 iov_iter_advance(iter, nr);
763 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
764 loff_t *pos, rwf_t flags)
769 if (!(file->f_mode & FMODE_READ))
771 if (!(file->f_mode & FMODE_CAN_READ))
774 tot_len = iov_iter_count(iter);
777 ret = rw_verify_area(READ, file, pos, tot_len);
781 if (file->f_op->read_iter)
782 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
784 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
787 fsnotify_access(file);
791 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
792 struct iov_iter *iter)
797 if (!file->f_op->read_iter)
799 if (!(file->f_mode & FMODE_READ))
801 if (!(file->f_mode & FMODE_CAN_READ))
804 tot_len = iov_iter_count(iter);
807 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
811 ret = call_read_iter(file, iocb, iter);
814 fsnotify_access(file);
817 EXPORT_SYMBOL(vfs_iocb_iter_read);
819 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
822 if (!file->f_op->read_iter)
824 return do_iter_read(file, iter, ppos, flags);
826 EXPORT_SYMBOL(vfs_iter_read);
828 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
829 loff_t *pos, rwf_t flags)
834 if (!(file->f_mode & FMODE_WRITE))
836 if (!(file->f_mode & FMODE_CAN_WRITE))
839 tot_len = iov_iter_count(iter);
842 ret = rw_verify_area(WRITE, file, pos, tot_len);
846 if (file->f_op->write_iter)
847 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
849 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
851 fsnotify_modify(file);
855 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
856 struct iov_iter *iter)
861 if (!file->f_op->write_iter)
863 if (!(file->f_mode & FMODE_WRITE))
865 if (!(file->f_mode & FMODE_CAN_WRITE))
868 tot_len = iov_iter_count(iter);
871 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
875 ret = call_write_iter(file, iocb, iter);
877 fsnotify_modify(file);
881 EXPORT_SYMBOL(vfs_iocb_iter_write);
883 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
886 if (!file->f_op->write_iter)
888 return do_iter_write(file, iter, ppos, flags);
890 EXPORT_SYMBOL(vfs_iter_write);
892 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
893 unsigned long vlen, loff_t *pos, rwf_t flags)
895 struct iovec iovstack[UIO_FASTIOV];
896 struct iovec *iov = iovstack;
897 struct iov_iter iter;
900 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
902 ret = do_iter_read(file, &iter, pos, flags);
909 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
910 unsigned long vlen, loff_t *pos, rwf_t flags)
912 struct iovec iovstack[UIO_FASTIOV];
913 struct iovec *iov = iovstack;
914 struct iov_iter iter;
917 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
919 file_start_write(file);
920 ret = do_iter_write(file, &iter, pos, flags);
921 file_end_write(file);
927 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
928 unsigned long vlen, rwf_t flags)
930 struct fd f = fdget_pos(fd);
931 ssize_t ret = -EBADF;
934 loff_t pos, *ppos = file_ppos(f.file);
939 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
940 if (ret >= 0 && ppos)
946 add_rchar(current, ret);
951 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
952 unsigned long vlen, rwf_t flags)
954 struct fd f = fdget_pos(fd);
955 ssize_t ret = -EBADF;
958 loff_t pos, *ppos = file_ppos(f.file);
963 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
964 if (ret >= 0 && ppos)
970 add_wchar(current, ret);
975 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
977 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
978 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
981 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
982 unsigned long vlen, loff_t pos, rwf_t flags)
985 ssize_t ret = -EBADF;
993 if (f.file->f_mode & FMODE_PREAD)
994 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
999 add_rchar(current, ret);
1004 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1005 unsigned long vlen, loff_t pos, rwf_t flags)
1008 ssize_t ret = -EBADF;
1016 if (f.file->f_mode & FMODE_PWRITE)
1017 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1022 add_wchar(current, ret);
1027 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1028 unsigned long, vlen)
1030 return do_readv(fd, vec, vlen, 0);
1033 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1034 unsigned long, vlen)
1036 return do_writev(fd, vec, vlen, 0);
1039 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1040 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1042 loff_t pos = pos_from_hilo(pos_h, pos_l);
1044 return do_preadv(fd, vec, vlen, pos, 0);
1047 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1048 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1051 loff_t pos = pos_from_hilo(pos_h, pos_l);
1054 return do_readv(fd, vec, vlen, flags);
1056 return do_preadv(fd, vec, vlen, pos, flags);
1059 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1060 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1062 loff_t pos = pos_from_hilo(pos_h, pos_l);
1064 return do_pwritev(fd, vec, vlen, pos, 0);
1067 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1068 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1071 loff_t pos = pos_from_hilo(pos_h, pos_l);
1074 return do_writev(fd, vec, vlen, flags);
1076 return do_pwritev(fd, vec, vlen, pos, flags);
1080 * Various compat syscalls. Note that they all pretend to take a native
1081 * iovec - import_iovec will properly treat those as compat_iovecs based on
1082 * in_compat_syscall().
1084 #ifdef CONFIG_COMPAT
1085 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1086 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1087 const struct iovec __user *, vec,
1088 unsigned long, vlen, loff_t, pos)
1090 return do_preadv(fd, vec, vlen, pos, 0);
1094 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1095 const struct iovec __user *, vec,
1096 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1098 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1100 return do_preadv(fd, vec, vlen, pos, 0);
1103 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1104 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1105 const struct iovec __user *, vec,
1106 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1109 return do_readv(fd, vec, vlen, flags);
1110 return do_preadv(fd, vec, vlen, pos, flags);
1114 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1115 const struct iovec __user *, vec,
1116 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1119 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1122 return do_readv(fd, vec, vlen, flags);
1123 return do_preadv(fd, vec, vlen, pos, flags);
1126 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1127 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1128 const struct iovec __user *, vec,
1129 unsigned long, vlen, loff_t, pos)
1131 return do_pwritev(fd, vec, vlen, pos, 0);
1135 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1136 const struct iovec __user *,vec,
1137 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1139 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1141 return do_pwritev(fd, vec, vlen, pos, 0);
1144 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1145 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1146 const struct iovec __user *, vec,
1147 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1150 return do_writev(fd, vec, vlen, flags);
1151 return do_pwritev(fd, vec, vlen, pos, flags);
1155 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1156 const struct iovec __user *,vec,
1157 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1159 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1162 return do_writev(fd, vec, vlen, flags);
1163 return do_pwritev(fd, vec, vlen, pos, flags);
1165 #endif /* CONFIG_COMPAT */
1167 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1168 size_t count, loff_t max)
1171 struct inode *in_inode, *out_inode;
1178 * Get input file, and verify that it is ok..
1184 if (!(in.file->f_mode & FMODE_READ))
1188 pos = in.file->f_pos;
1191 if (!(in.file->f_mode & FMODE_PREAD))
1194 retval = rw_verify_area(READ, in.file, &pos, count);
1197 if (count > MAX_RW_COUNT)
1198 count = MAX_RW_COUNT;
1201 * Get output file, and verify that it is ok..
1204 out = fdget(out_fd);
1207 if (!(out.file->f_mode & FMODE_WRITE))
1209 in_inode = file_inode(in.file);
1210 out_inode = file_inode(out.file);
1211 out_pos = out.file->f_pos;
1212 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1217 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1219 if (unlikely(pos + count > max)) {
1220 retval = -EOVERFLOW;
1229 * We need to debate whether we can enable this or not. The
1230 * man page documents EAGAIN return for the output at least,
1231 * and the application is arguably buggy if it doesn't expect
1232 * EAGAIN on a non-blocking file descriptor.
1234 if (in.file->f_flags & O_NONBLOCK)
1235 fl = SPLICE_F_NONBLOCK;
1237 file_start_write(out.file);
1238 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1239 file_end_write(out.file);
1242 add_rchar(current, retval);
1243 add_wchar(current, retval);
1244 fsnotify_access(in.file);
1245 fsnotify_modify(out.file);
1246 out.file->f_pos = out_pos;
1250 in.file->f_pos = pos;
1256 retval = -EOVERFLOW;
1266 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1273 if (unlikely(get_user(off, offset)))
1276 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1277 if (unlikely(put_user(pos, offset)))
1282 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1285 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1291 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1293 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1294 if (unlikely(put_user(pos, offset)))
1299 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1302 #ifdef CONFIG_COMPAT
1303 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1304 compat_off_t __user *, offset, compat_size_t, count)
1311 if (unlikely(get_user(off, offset)))
1314 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1315 if (unlikely(put_user(pos, offset)))
1320 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1323 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1324 compat_loff_t __user *, offset, compat_size_t, count)
1330 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1332 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1333 if (unlikely(put_user(pos, offset)))
1338 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1343 * generic_copy_file_range - copy data between two files
1344 * @file_in: file structure to read from
1345 * @pos_in: file offset to read from
1346 * @file_out: file structure to write data to
1347 * @pos_out: file offset to write data to
1348 * @len: amount of data to copy
1349 * @flags: copy flags
1351 * This is a generic filesystem helper to copy data from one file to another.
1352 * It has no constraints on the source or destination file owners - the files
1353 * can belong to different superblocks and different filesystem types. Short
1354 * copies are allowed.
1356 * This should be called from the @file_out filesystem, as per the
1357 * ->copy_file_range() method.
1359 * Returns the number of bytes copied or a negative error indicating the
1363 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1364 struct file *file_out, loff_t pos_out,
1365 size_t len, unsigned int flags)
1367 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1368 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1370 EXPORT_SYMBOL(generic_copy_file_range);
1372 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1373 struct file *file_out, loff_t pos_out,
1374 size_t len, unsigned int flags)
1377 * Although we now allow filesystems to handle cross sb copy, passing
1378 * a file of the wrong filesystem type to filesystem driver can result
1379 * in an attempt to dereference the wrong type of ->private_data, so
1380 * avoid doing that until we really have a good reason. NFS defines
1381 * several different file_system_type structures, but they all end up
1382 * using the same ->copy_file_range() function pointer.
1384 if (file_out->f_op->copy_file_range &&
1385 file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1386 return file_out->f_op->copy_file_range(file_in, pos_in,
1390 return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1395 * copy_file_range() differs from regular file read and write in that it
1396 * specifically allows return partial success. When it does so is up to
1397 * the copy_file_range method.
1399 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1400 struct file *file_out, loff_t pos_out,
1401 size_t len, unsigned int flags)
1408 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1413 ret = rw_verify_area(READ, file_in, &pos_in, len);
1417 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1424 file_start_write(file_out);
1427 * Try cloning first, this is supported by more file systems, and
1428 * more efficient if both clone and copy are supported (e.g. NFS).
1430 if (file_in->f_op->remap_file_range &&
1431 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1434 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1436 min_t(loff_t, MAX_RW_COUNT, len),
1437 REMAP_FILE_CAN_SHORTEN);
1444 ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1446 WARN_ON_ONCE(ret == -EOPNOTSUPP);
1449 fsnotify_access(file_in);
1450 add_rchar(current, ret);
1451 fsnotify_modify(file_out);
1452 add_wchar(current, ret);
1458 file_end_write(file_out);
1462 EXPORT_SYMBOL(vfs_copy_file_range);
1464 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1465 int, fd_out, loff_t __user *, off_out,
1466 size_t, len, unsigned int, flags)
1472 ssize_t ret = -EBADF;
1474 f_in = fdget(fd_in);
1478 f_out = fdget(fd_out);
1484 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1487 pos_in = f_in.file->f_pos;
1491 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1494 pos_out = f_out.file->f_pos;
1497 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1504 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1507 f_in.file->f_pos = pos_in;
1511 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1514 f_out.file->f_pos = pos_out;
1526 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
1529 struct inode *inode = file_inode(file);
1531 if (unlikely(pos < 0 || len < 0))
1534 if (unlikely((loff_t) (pos + len) < 0))
1537 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1538 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1541 retval = locks_mandatory_area(inode, file, pos, end,
1542 write ? F_WRLCK : F_RDLCK);
1547 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1550 * Ensure that we don't remap a partial EOF block in the middle of something
1551 * else. Assume that the offsets have already been checked for block
1554 * For clone we only link a partial EOF block above or at the destination file's
1555 * EOF. For deduplication we accept a partial EOF block only if it ends at the
1556 * destination file's EOF (can not link it into the middle of a file).
1558 * Shorten the request if possible.
1560 static int generic_remap_check_len(struct inode *inode_in,
1561 struct inode *inode_out,
1564 unsigned int remap_flags)
1566 u64 blkmask = i_blocksize(inode_in) - 1;
1567 loff_t new_len = *len;
1569 if ((*len & blkmask) == 0)
1572 if (pos_out + *len < i_size_read(inode_out))
1573 new_len &= ~blkmask;
1575 if (new_len == *len)
1578 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
1583 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
1586 /* Read a page's worth of file data into the page cache. */
1587 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1591 page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
1594 if (!PageUptodate(page)) {
1596 return ERR_PTR(-EIO);
1602 * Lock two pages, ensuring that we lock in offset order if the pages are from
1605 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1607 /* Always lock in order of increasing index. */
1608 if (page1->index > page2->index)
1616 /* Unlock two pages, being careful not to unlock the same page twice. */
1617 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1625 * Compare extents of two files to see if they are the same.
1626 * Caller must have locked both inodes to prevent write races.
1628 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1629 struct inode *dest, loff_t destoff,
1630 loff_t len, bool *is_same)
1636 struct page *src_page;
1637 struct page *dest_page;
1645 src_poff = srcoff & (PAGE_SIZE - 1);
1646 dest_poff = destoff & (PAGE_SIZE - 1);
1647 cmp_len = min(PAGE_SIZE - src_poff,
1648 PAGE_SIZE - dest_poff);
1649 cmp_len = min(cmp_len, len);
1653 src_page = vfs_dedupe_get_page(src, srcoff);
1654 if (IS_ERR(src_page)) {
1655 error = PTR_ERR(src_page);
1658 dest_page = vfs_dedupe_get_page(dest, destoff);
1659 if (IS_ERR(dest_page)) {
1660 error = PTR_ERR(dest_page);
1665 vfs_lock_two_pages(src_page, dest_page);
1668 * Now that we've locked both pages, make sure they're still
1669 * mapped to the file data we're interested in. If not,
1670 * someone is invalidating pages on us and we lose.
1672 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
1673 src_page->mapping != src->i_mapping ||
1674 dest_page->mapping != dest->i_mapping) {
1679 src_addr = kmap_atomic(src_page);
1680 dest_addr = kmap_atomic(dest_page);
1682 flush_dcache_page(src_page);
1683 flush_dcache_page(dest_page);
1685 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1688 kunmap_atomic(dest_addr);
1689 kunmap_atomic(src_addr);
1691 vfs_unlock_two_pages(src_page, dest_page);
1692 put_page(dest_page);
1711 * Check that the two inodes are eligible for cloning, the ranges make
1712 * sense, and then flush all dirty data. Caller must ensure that the
1713 * inodes have been locked against any other modifications.
1715 * If there's an error, then the usual negative error code is returned.
1716 * Otherwise returns 0 with *len set to the request length.
1718 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
1719 struct file *file_out, loff_t pos_out,
1720 loff_t *len, unsigned int remap_flags)
1722 struct inode *inode_in = file_inode(file_in);
1723 struct inode *inode_out = file_inode(file_out);
1724 bool same_inode = (inode_in == inode_out);
1727 /* Don't touch certain kinds of inodes */
1728 if (IS_IMMUTABLE(inode_out))
1731 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1734 /* Don't reflink dirs, pipes, sockets... */
1735 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1737 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1740 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1742 loff_t isize = i_size_read(inode_in);
1744 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
1748 *len = isize - pos_in;
1753 /* Check that we don't violate system file offset limits. */
1754 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
1759 /* Wait for the completion of any pending IOs on both files */
1760 inode_dio_wait(inode_in);
1762 inode_dio_wait(inode_out);
1764 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1765 pos_in, pos_in + *len - 1);
1769 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1770 pos_out, pos_out + *len - 1);
1775 * Check that the extents are the same.
1777 if (remap_flags & REMAP_FILE_DEDUP) {
1778 bool is_same = false;
1780 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1781 inode_out, pos_out, *len, &is_same);
1788 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
1793 /* If can't alter the file contents, we're done. */
1794 if (!(remap_flags & REMAP_FILE_DEDUP))
1795 ret = file_modified(file_out);
1799 EXPORT_SYMBOL(generic_remap_file_range_prep);
1801 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
1802 struct file *file_out, loff_t pos_out,
1803 loff_t len, unsigned int remap_flags)
1807 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
1810 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1811 * the same mount. Practically, they only need to be on the same file
1814 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
1817 ret = generic_file_rw_checks(file_in, file_out);
1821 if (!file_in->f_op->remap_file_range)
1824 ret = remap_verify_area(file_in, pos_in, len, false);
1828 ret = remap_verify_area(file_out, pos_out, len, true);
1832 ret = file_in->f_op->remap_file_range(file_in, pos_in,
1833 file_out, pos_out, len, remap_flags);
1837 fsnotify_access(file_in);
1838 fsnotify_modify(file_out);
1841 EXPORT_SYMBOL(do_clone_file_range);
1843 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1844 struct file *file_out, loff_t pos_out,
1845 loff_t len, unsigned int remap_flags)
1849 file_start_write(file_out);
1850 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
1852 file_end_write(file_out);
1856 EXPORT_SYMBOL(vfs_clone_file_range);
1858 /* Check whether we are allowed to dedupe the destination file */
1859 static bool allow_file_dedupe(struct file *file)
1861 if (capable(CAP_SYS_ADMIN))
1863 if (file->f_mode & FMODE_WRITE)
1865 if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
1867 if (!inode_permission(file_inode(file), MAY_WRITE))
1872 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
1873 struct file *dst_file, loff_t dst_pos,
1874 loff_t len, unsigned int remap_flags)
1878 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
1879 REMAP_FILE_CAN_SHORTEN));
1881 ret = mnt_want_write_file(dst_file);
1885 ret = remap_verify_area(dst_file, dst_pos, len, true);
1887 goto out_drop_write;
1890 if (!allow_file_dedupe(dst_file))
1891 goto out_drop_write;
1894 if (src_file->f_path.mnt != dst_file->f_path.mnt)
1895 goto out_drop_write;
1898 if (S_ISDIR(file_inode(dst_file)->i_mode))
1899 goto out_drop_write;
1902 if (!dst_file->f_op->remap_file_range)
1903 goto out_drop_write;
1907 goto out_drop_write;
1910 ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
1911 dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
1913 mnt_drop_write_file(dst_file);
1917 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
1919 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1921 struct file_dedupe_range_info *info;
1922 struct inode *src = file_inode(file);
1927 u16 count = same->dest_count;
1930 if (!(file->f_mode & FMODE_READ))
1933 if (same->reserved1 || same->reserved2)
1936 off = same->src_offset;
1937 len = same->src_length;
1939 if (S_ISDIR(src->i_mode))
1942 if (!S_ISREG(src->i_mode))
1945 if (!file->f_op->remap_file_range)
1948 ret = remap_verify_area(file, off, len, false);
1953 if (off + len > i_size_read(src))
1956 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
1957 len = min_t(u64, len, 1 << 30);
1959 /* pre-format output fields to sane values */
1960 for (i = 0; i < count; i++) {
1961 same->info[i].bytes_deduped = 0ULL;
1962 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1965 for (i = 0, info = same->info; i < count; i++, info++) {
1966 struct fd dst_fd = fdget(info->dest_fd);
1967 struct file *dst_file = dst_fd.file;
1970 info->status = -EBADF;
1974 if (info->reserved) {
1975 info->status = -EINVAL;
1979 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
1980 info->dest_offset, len,
1981 REMAP_FILE_CAN_SHORTEN);
1982 if (deduped == -EBADE)
1983 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1984 else if (deduped < 0)
1985 info->status = deduped;
1987 info->bytes_deduped = len;
1992 if (fatal_signal_pending(current))
1997 EXPORT_SYMBOL(vfs_dedupe_file_range);