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 = filemap_splice_read,
35 EXPORT_SYMBOL(generic_ro_fops);
37 static inline bool unsigned_offsets(struct file *file)
39 return file->f_op->fop_flags & FOP_UNSIGNED_OFFSET;
43 * vfs_setpos_cookie - update the file offset for lseek and reset cookie
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
47 * @cookie: cookie to reset
49 * Update the file offset to the value specified by @offset if the given
50 * offset is valid and it is not equal to the current file offset and
51 * reset the specified cookie to indicate that a seek happened.
53 * Return the specified offset on success and -EINVAL on invalid offset.
55 static loff_t vfs_setpos_cookie(struct file *file, loff_t offset,
56 loff_t maxsize, u64 *cookie)
58 if (offset < 0 && !unsigned_offsets(file))
63 if (offset != file->f_pos) {
72 * vfs_setpos - update the file offset for lseek
73 * @file: file structure in question
74 * @offset: file offset to seek to
75 * @maxsize: maximum file size
77 * This is a low-level filesystem helper for updating the file offset to
78 * the value specified by @offset if the given offset is valid and it is
79 * not equal to the current file offset.
81 * Return the specified offset on success and -EINVAL on invalid offset.
83 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
85 return vfs_setpos_cookie(file, offset, maxsize, NULL);
87 EXPORT_SYMBOL(vfs_setpos);
90 * must_set_pos - check whether f_pos has to be updated
91 * @file: file to seek on
92 * @offset: offset to use
93 * @whence: type of seek operation
96 * Check whether f_pos needs to be updated and update @offset according
99 * Return: 0 if f_pos doesn't need to be updated, 1 if f_pos has to be
100 * updated, and negative error code on failure.
102 static int must_set_pos(struct file *file, loff_t *offset, int whence, loff_t eof)
110 * Here we special-case the lseek(fd, 0, SEEK_CUR)
111 * position-querying operation. Avoid rewriting the "same"
112 * f_pos value back to the file because a concurrent read(),
113 * write() or lseek() might have altered it
116 *offset = file->f_pos;
122 * In the generic case the entire file is data, so as long as
123 * offset isn't at the end of the file then the offset is data.
125 if ((unsigned long long)*offset >= eof)
130 * There is a virtual hole at the end of the file, so as long as
131 * offset isn't i_size or larger, return i_size.
133 if ((unsigned long long)*offset >= eof)
143 * generic_file_llseek_size - generic llseek implementation for regular files
144 * @file: file structure to seek on
145 * @offset: file offset to seek to
146 * @whence: type of seek
147 * @maxsize: max size of this file in file system
148 * @eof: offset used for SEEK_END position
150 * This is a variant of generic_file_llseek that allows passing in a custom
151 * maximum file size and a custom EOF position, for e.g. hashed directories
154 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
155 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
156 * read/writes behave like SEEK_SET against seeks.
159 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
160 loff_t maxsize, loff_t eof)
164 ret = must_set_pos(file, &offset, whence, eof);
170 if (whence == SEEK_CUR) {
172 * f_lock protects against read/modify/write race with
173 * other SEEK_CURs. Note that parallel writes and reads
174 * behave like SEEK_SET.
176 guard(spinlock)(&file->f_lock);
177 return vfs_setpos(file, file->f_pos + offset, maxsize);
180 return vfs_setpos(file, offset, maxsize);
182 EXPORT_SYMBOL(generic_file_llseek_size);
185 * generic_llseek_cookie - versioned llseek implementation
186 * @file: file structure to seek on
187 * @offset: file offset to seek to
188 * @whence: type of seek
189 * @cookie: cookie to update
191 * See generic_file_llseek for a general description and locking assumptions.
193 * In contrast to generic_file_llseek, this function also resets a
194 * specified cookie to indicate a seek took place.
196 loff_t generic_llseek_cookie(struct file *file, loff_t offset, int whence,
199 struct inode *inode = file->f_mapping->host;
200 loff_t maxsize = inode->i_sb->s_maxbytes;
201 loff_t eof = i_size_read(inode);
204 if (WARN_ON_ONCE(!cookie))
208 * Require that this is only used for directories that guarantee
209 * synchronization between readdir and seek so that an update to
210 * @cookie is correctly synchronized with concurrent readdir.
212 if (WARN_ON_ONCE(!(file->f_mode & FMODE_ATOMIC_POS)))
215 ret = must_set_pos(file, &offset, whence, eof);
221 /* No need to hold f_lock because we know that f_pos_lock is held. */
222 if (whence == SEEK_CUR)
223 return vfs_setpos_cookie(file, file->f_pos + offset, maxsize, cookie);
225 return vfs_setpos_cookie(file, offset, maxsize, cookie);
227 EXPORT_SYMBOL(generic_llseek_cookie);
230 * generic_file_llseek - generic llseek implementation for regular files
231 * @file: file structure to seek on
232 * @offset: file offset to seek to
233 * @whence: type of seek
235 * This is a generic implemenation of ->llseek useable for all normal local
236 * filesystems. It just updates the file offset to the value specified by
237 * @offset and @whence.
239 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
241 struct inode *inode = file->f_mapping->host;
243 return generic_file_llseek_size(file, offset, whence,
244 inode->i_sb->s_maxbytes,
247 EXPORT_SYMBOL(generic_file_llseek);
250 * fixed_size_llseek - llseek implementation for fixed-sized devices
251 * @file: file structure to seek on
252 * @offset: file offset to seek to
253 * @whence: type of seek
254 * @size: size of the file
257 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
260 case SEEK_SET: case SEEK_CUR: case SEEK_END:
261 return generic_file_llseek_size(file, offset, whence,
267 EXPORT_SYMBOL(fixed_size_llseek);
270 * no_seek_end_llseek - llseek implementation for fixed-sized devices
271 * @file: file structure to seek on
272 * @offset: file offset to seek to
273 * @whence: type of seek
276 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
279 case SEEK_SET: case SEEK_CUR:
280 return generic_file_llseek_size(file, offset, whence,
286 EXPORT_SYMBOL(no_seek_end_llseek);
289 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
290 * @file: file structure to seek on
291 * @offset: file offset to seek to
292 * @whence: type of seek
293 * @size: maximal offset allowed
296 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
299 case SEEK_SET: case SEEK_CUR:
300 return generic_file_llseek_size(file, offset, whence,
306 EXPORT_SYMBOL(no_seek_end_llseek_size);
309 * noop_llseek - No Operation Performed llseek implementation
310 * @file: file structure to seek on
311 * @offset: file offset to seek to
312 * @whence: type of seek
314 * This is an implementation of ->llseek useable for the rare special case when
315 * userspace expects the seek to succeed but the (device) file is actually not
316 * able to perform the seek. In this case you use noop_llseek() instead of
317 * falling back to the default implementation of ->llseek.
319 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
323 EXPORT_SYMBOL(noop_llseek);
325 loff_t default_llseek(struct file *file, loff_t offset, int whence)
327 struct inode *inode = file_inode(file);
333 offset += i_size_read(inode);
337 retval = file->f_pos;
340 offset += file->f_pos;
344 * In the generic case the entire file is data, so as
345 * long as offset isn't at the end of the file then the
348 if (offset >= inode->i_size) {
355 * There is a virtual hole at the end of the file, so
356 * as long as offset isn't i_size or larger, return
359 if (offset >= inode->i_size) {
363 offset = inode->i_size;
367 if (offset >= 0 || unsigned_offsets(file)) {
368 if (offset != file->f_pos)
369 file->f_pos = offset;
376 EXPORT_SYMBOL(default_llseek);
378 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
380 if (!(file->f_mode & FMODE_LSEEK))
382 return file->f_op->llseek(file, offset, whence);
384 EXPORT_SYMBOL(vfs_llseek);
386 static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
389 struct fd f = fdget_pos(fd);
394 if (whence <= SEEK_MAX) {
395 loff_t res = vfs_llseek(fd_file(f), offset, whence);
397 if (res != (loff_t)retval)
398 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
404 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
406 return ksys_lseek(fd, offset, whence);
410 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
412 return ksys_lseek(fd, offset, whence);
416 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
417 defined(__ARCH_WANT_SYS_LLSEEK)
418 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
419 unsigned long, offset_low, loff_t __user *, result,
420 unsigned int, whence)
423 struct fd f = fdget_pos(fd);
430 if (whence > SEEK_MAX)
433 offset = vfs_llseek(fd_file(f), ((loff_t) offset_high << 32) | offset_low,
436 retval = (int)offset;
439 if (!copy_to_user(result, &offset, sizeof(offset)))
448 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
450 int mask = read_write == READ ? MAY_READ : MAY_WRITE;
453 if (unlikely((ssize_t) count < 0))
459 if (unlikely(pos < 0)) {
460 if (!unsigned_offsets(file))
462 if (count >= -pos) /* both values are in 0..LLONG_MAX */
464 } else if (unlikely((loff_t) (pos + count) < 0)) {
465 if (!unsigned_offsets(file))
470 ret = security_file_permission(file, mask);
474 return fsnotify_file_area_perm(file, mask, ppos, count);
476 EXPORT_SYMBOL(rw_verify_area);
478 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
481 struct iov_iter iter;
484 init_sync_kiocb(&kiocb, filp);
485 kiocb.ki_pos = (ppos ? *ppos : 0);
486 iov_iter_ubuf(&iter, ITER_DEST, buf, len);
488 ret = filp->f_op->read_iter(&kiocb, &iter);
489 BUG_ON(ret == -EIOCBQUEUED);
491 *ppos = kiocb.ki_pos;
495 static int warn_unsupported(struct file *file, const char *op)
498 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
499 op, file, current->pid, current->comm);
503 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
507 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
510 struct iov_iter iter;
513 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
515 if (!(file->f_mode & FMODE_CAN_READ))
518 * Also fail if ->read_iter and ->read are both wired up as that
519 * implies very convoluted semantics.
521 if (unlikely(!file->f_op->read_iter || file->f_op->read))
522 return warn_unsupported(file, "read");
524 init_sync_kiocb(&kiocb, file);
525 kiocb.ki_pos = pos ? *pos : 0;
526 iov_iter_kvec(&iter, ITER_DEST, &iov, 1, iov.iov_len);
527 ret = file->f_op->read_iter(&kiocb, &iter);
531 fsnotify_access(file);
532 add_rchar(current, ret);
538 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
542 ret = rw_verify_area(READ, file, pos, count);
545 return __kernel_read(file, buf, count, pos);
547 EXPORT_SYMBOL(kernel_read);
549 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
553 if (!(file->f_mode & FMODE_READ))
555 if (!(file->f_mode & FMODE_CAN_READ))
557 if (unlikely(!access_ok(buf, count)))
560 ret = rw_verify_area(READ, file, pos, count);
563 if (count > MAX_RW_COUNT)
564 count = MAX_RW_COUNT;
566 if (file->f_op->read)
567 ret = file->f_op->read(file, buf, count, pos);
568 else if (file->f_op->read_iter)
569 ret = new_sync_read(file, buf, count, pos);
573 fsnotify_access(file);
574 add_rchar(current, ret);
580 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
583 struct iov_iter iter;
586 init_sync_kiocb(&kiocb, filp);
587 kiocb.ki_pos = (ppos ? *ppos : 0);
588 iov_iter_ubuf(&iter, ITER_SOURCE, (void __user *)buf, len);
590 ret = filp->f_op->write_iter(&kiocb, &iter);
591 BUG_ON(ret == -EIOCBQUEUED);
593 *ppos = kiocb.ki_pos;
597 /* caller is responsible for file_start_write/file_end_write */
598 ssize_t __kernel_write_iter(struct file *file, struct iov_iter *from, loff_t *pos)
603 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
605 if (!(file->f_mode & FMODE_CAN_WRITE))
608 * Also fail if ->write_iter and ->write are both wired up as that
609 * implies very convoluted semantics.
611 if (unlikely(!file->f_op->write_iter || file->f_op->write))
612 return warn_unsupported(file, "write");
614 init_sync_kiocb(&kiocb, file);
615 kiocb.ki_pos = pos ? *pos : 0;
616 ret = file->f_op->write_iter(&kiocb, from);
620 fsnotify_modify(file);
621 add_wchar(current, ret);
627 /* caller is responsible for file_start_write/file_end_write */
628 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
631 .iov_base = (void *)buf,
632 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
634 struct iov_iter iter;
635 iov_iter_kvec(&iter, ITER_SOURCE, &iov, 1, iov.iov_len);
636 return __kernel_write_iter(file, &iter, pos);
639 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
640 * but autofs is one of the few internal kernel users that actually
641 * wants this _and_ can be built as a module. So we need to export
642 * this symbol for autofs, even though it really isn't appropriate
643 * for any other kernel modules.
645 EXPORT_SYMBOL_GPL(__kernel_write);
647 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
652 ret = rw_verify_area(WRITE, file, pos, count);
656 file_start_write(file);
657 ret = __kernel_write(file, buf, count, pos);
658 file_end_write(file);
661 EXPORT_SYMBOL(kernel_write);
663 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
667 if (!(file->f_mode & FMODE_WRITE))
669 if (!(file->f_mode & FMODE_CAN_WRITE))
671 if (unlikely(!access_ok(buf, count)))
674 ret = rw_verify_area(WRITE, file, pos, count);
677 if (count > MAX_RW_COUNT)
678 count = MAX_RW_COUNT;
679 file_start_write(file);
680 if (file->f_op->write)
681 ret = file->f_op->write(file, buf, count, pos);
682 else if (file->f_op->write_iter)
683 ret = new_sync_write(file, buf, count, pos);
687 fsnotify_modify(file);
688 add_wchar(current, ret);
691 file_end_write(file);
695 /* file_ppos returns &file->f_pos or NULL if file is stream */
696 static inline loff_t *file_ppos(struct file *file)
698 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
701 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
703 struct fd f = fdget_pos(fd);
704 ssize_t ret = -EBADF;
707 loff_t pos, *ppos = file_ppos(fd_file(f));
712 ret = vfs_read(fd_file(f), buf, count, ppos);
713 if (ret >= 0 && ppos)
714 fd_file(f)->f_pos = pos;
720 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
722 return ksys_read(fd, buf, count);
725 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
727 struct fd f = fdget_pos(fd);
728 ssize_t ret = -EBADF;
731 loff_t pos, *ppos = file_ppos(fd_file(f));
736 ret = vfs_write(fd_file(f), buf, count, ppos);
737 if (ret >= 0 && ppos)
738 fd_file(f)->f_pos = pos;
745 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
748 return ksys_write(fd, buf, count);
751 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
755 ssize_t ret = -EBADF;
763 if (fd_file(f)->f_mode & FMODE_PREAD)
764 ret = vfs_read(fd_file(f), buf, count, &pos);
771 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
772 size_t, count, loff_t, pos)
774 return ksys_pread64(fd, buf, count, pos);
777 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
778 COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,
779 size_t, count, compat_arg_u64_dual(pos))
781 return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));
785 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
786 size_t count, loff_t pos)
789 ssize_t ret = -EBADF;
797 if (fd_file(f)->f_mode & FMODE_PWRITE)
798 ret = vfs_write(fd_file(f), buf, count, &pos);
805 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
806 size_t, count, loff_t, pos)
808 return ksys_pwrite64(fd, buf, count, pos);
811 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
812 COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,
813 size_t, count, compat_arg_u64_dual(pos))
815 return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));
819 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
820 loff_t *ppos, int type, rwf_t flags)
825 init_sync_kiocb(&kiocb, filp);
826 ret = kiocb_set_rw_flags(&kiocb, flags, type);
829 kiocb.ki_pos = (ppos ? *ppos : 0);
832 ret = filp->f_op->read_iter(&kiocb, iter);
834 ret = filp->f_op->write_iter(&kiocb, iter);
835 BUG_ON(ret == -EIOCBQUEUED);
837 *ppos = kiocb.ki_pos;
841 /* Do it by hand, with file-ops */
842 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
843 loff_t *ppos, int type, rwf_t flags)
847 if (flags & ~RWF_HIPRI)
850 while (iov_iter_count(iter)) {
854 nr = filp->f_op->read(filp, iter_iov_addr(iter),
855 iter_iov_len(iter), ppos);
857 nr = filp->f_op->write(filp, iter_iov_addr(iter),
858 iter_iov_len(iter), ppos);
867 if (nr != iter_iov_len(iter))
869 iov_iter_advance(iter, nr);
875 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
876 struct iov_iter *iter)
881 if (!file->f_op->read_iter)
883 if (!(file->f_mode & FMODE_READ))
885 if (!(file->f_mode & FMODE_CAN_READ))
888 tot_len = iov_iter_count(iter);
891 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
895 ret = file->f_op->read_iter(iocb, iter);
898 fsnotify_access(file);
901 EXPORT_SYMBOL(vfs_iocb_iter_read);
903 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
909 if (!file->f_op->read_iter)
911 if (!(file->f_mode & FMODE_READ))
913 if (!(file->f_mode & FMODE_CAN_READ))
916 tot_len = iov_iter_count(iter);
919 ret = rw_verify_area(READ, file, ppos, tot_len);
923 ret = do_iter_readv_writev(file, iter, ppos, READ, flags);
926 fsnotify_access(file);
929 EXPORT_SYMBOL(vfs_iter_read);
932 * Caller is responsible for calling kiocb_end_write() on completion
933 * if async iocb was queued.
935 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
936 struct iov_iter *iter)
941 if (!file->f_op->write_iter)
943 if (!(file->f_mode & FMODE_WRITE))
945 if (!(file->f_mode & FMODE_CAN_WRITE))
948 tot_len = iov_iter_count(iter);
951 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
955 kiocb_start_write(iocb);
956 ret = file->f_op->write_iter(iocb, iter);
957 if (ret != -EIOCBQUEUED)
958 kiocb_end_write(iocb);
960 fsnotify_modify(file);
964 EXPORT_SYMBOL(vfs_iocb_iter_write);
966 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
972 if (!(file->f_mode & FMODE_WRITE))
974 if (!(file->f_mode & FMODE_CAN_WRITE))
976 if (!file->f_op->write_iter)
979 tot_len = iov_iter_count(iter);
983 ret = rw_verify_area(WRITE, file, ppos, tot_len);
987 file_start_write(file);
988 ret = do_iter_readv_writev(file, iter, ppos, WRITE, flags);
990 fsnotify_modify(file);
991 file_end_write(file);
995 EXPORT_SYMBOL(vfs_iter_write);
997 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
998 unsigned long vlen, loff_t *pos, rwf_t flags)
1000 struct iovec iovstack[UIO_FASTIOV];
1001 struct iovec *iov = iovstack;
1002 struct iov_iter iter;
1006 if (!(file->f_mode & FMODE_READ))
1008 if (!(file->f_mode & FMODE_CAN_READ))
1011 ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov,
1016 tot_len = iov_iter_count(&iter);
1020 ret = rw_verify_area(READ, file, pos, tot_len);
1024 if (file->f_op->read_iter)
1025 ret = do_iter_readv_writev(file, &iter, pos, READ, flags);
1027 ret = do_loop_readv_writev(file, &iter, pos, READ, flags);
1030 fsnotify_access(file);
1035 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
1036 unsigned long vlen, loff_t *pos, rwf_t flags)
1038 struct iovec iovstack[UIO_FASTIOV];
1039 struct iovec *iov = iovstack;
1040 struct iov_iter iter;
1044 if (!(file->f_mode & FMODE_WRITE))
1046 if (!(file->f_mode & FMODE_CAN_WRITE))
1049 ret = import_iovec(ITER_SOURCE, vec, vlen, ARRAY_SIZE(iovstack), &iov,
1054 tot_len = iov_iter_count(&iter);
1058 ret = rw_verify_area(WRITE, file, pos, tot_len);
1062 file_start_write(file);
1063 if (file->f_op->write_iter)
1064 ret = do_iter_readv_writev(file, &iter, pos, WRITE, flags);
1066 ret = do_loop_readv_writev(file, &iter, pos, WRITE, flags);
1068 fsnotify_modify(file);
1069 file_end_write(file);
1075 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1076 unsigned long vlen, rwf_t flags)
1078 struct fd f = fdget_pos(fd);
1079 ssize_t ret = -EBADF;
1082 loff_t pos, *ppos = file_ppos(fd_file(f));
1087 ret = vfs_readv(fd_file(f), vec, vlen, ppos, flags);
1088 if (ret >= 0 && ppos)
1089 fd_file(f)->f_pos = pos;
1094 add_rchar(current, ret);
1099 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1100 unsigned long vlen, rwf_t flags)
1102 struct fd f = fdget_pos(fd);
1103 ssize_t ret = -EBADF;
1106 loff_t pos, *ppos = file_ppos(fd_file(f));
1111 ret = vfs_writev(fd_file(f), vec, vlen, ppos, flags);
1112 if (ret >= 0 && ppos)
1113 fd_file(f)->f_pos = pos;
1118 add_wchar(current, ret);
1123 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1125 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1126 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1129 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1130 unsigned long vlen, loff_t pos, rwf_t flags)
1133 ssize_t ret = -EBADF;
1141 if (fd_file(f)->f_mode & FMODE_PREAD)
1142 ret = vfs_readv(fd_file(f), vec, vlen, &pos, flags);
1147 add_rchar(current, ret);
1152 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1153 unsigned long vlen, loff_t pos, rwf_t flags)
1156 ssize_t ret = -EBADF;
1164 if (fd_file(f)->f_mode & FMODE_PWRITE)
1165 ret = vfs_writev(fd_file(f), vec, vlen, &pos, flags);
1170 add_wchar(current, ret);
1175 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1176 unsigned long, vlen)
1178 return do_readv(fd, vec, vlen, 0);
1181 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1182 unsigned long, vlen)
1184 return do_writev(fd, vec, vlen, 0);
1187 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1188 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1190 loff_t pos = pos_from_hilo(pos_h, pos_l);
1192 return do_preadv(fd, vec, vlen, pos, 0);
1195 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1196 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1199 loff_t pos = pos_from_hilo(pos_h, pos_l);
1202 return do_readv(fd, vec, vlen, flags);
1204 return do_preadv(fd, vec, vlen, pos, flags);
1207 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1208 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1210 loff_t pos = pos_from_hilo(pos_h, pos_l);
1212 return do_pwritev(fd, vec, vlen, pos, 0);
1215 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1216 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1219 loff_t pos = pos_from_hilo(pos_h, pos_l);
1222 return do_writev(fd, vec, vlen, flags);
1224 return do_pwritev(fd, vec, vlen, pos, flags);
1228 * Various compat syscalls. Note that they all pretend to take a native
1229 * iovec - import_iovec will properly treat those as compat_iovecs based on
1230 * in_compat_syscall().
1232 #ifdef CONFIG_COMPAT
1233 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1234 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1235 const struct iovec __user *, vec,
1236 unsigned long, vlen, loff_t, pos)
1238 return do_preadv(fd, vec, vlen, pos, 0);
1242 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1243 const struct iovec __user *, vec,
1244 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1246 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1248 return do_preadv(fd, vec, vlen, pos, 0);
1251 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1252 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1253 const struct iovec __user *, vec,
1254 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1257 return do_readv(fd, vec, vlen, flags);
1258 return do_preadv(fd, vec, vlen, pos, flags);
1262 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1263 const struct iovec __user *, vec,
1264 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1267 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1270 return do_readv(fd, vec, vlen, flags);
1271 return do_preadv(fd, vec, vlen, pos, flags);
1274 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1275 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1276 const struct iovec __user *, vec,
1277 unsigned long, vlen, loff_t, pos)
1279 return do_pwritev(fd, vec, vlen, pos, 0);
1283 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1284 const struct iovec __user *,vec,
1285 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1287 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1289 return do_pwritev(fd, vec, vlen, pos, 0);
1292 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1293 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1294 const struct iovec __user *, vec,
1295 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1298 return do_writev(fd, vec, vlen, flags);
1299 return do_pwritev(fd, vec, vlen, pos, flags);
1303 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1304 const struct iovec __user *,vec,
1305 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1307 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1310 return do_writev(fd, vec, vlen, flags);
1311 return do_pwritev(fd, vec, vlen, pos, flags);
1313 #endif /* CONFIG_COMPAT */
1315 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1316 size_t count, loff_t max)
1319 struct inode *in_inode, *out_inode;
1320 struct pipe_inode_info *opipe;
1327 * Get input file, and verify that it is ok..
1333 if (!(fd_file(in)->f_mode & FMODE_READ))
1337 pos = fd_file(in)->f_pos;
1340 if (!(fd_file(in)->f_mode & FMODE_PREAD))
1343 retval = rw_verify_area(READ, fd_file(in), &pos, count);
1346 if (count > MAX_RW_COUNT)
1347 count = MAX_RW_COUNT;
1350 * Get output file, and verify that it is ok..
1353 out = fdget(out_fd);
1356 if (!(fd_file(out)->f_mode & FMODE_WRITE))
1358 in_inode = file_inode(fd_file(in));
1359 out_inode = file_inode(fd_file(out));
1360 out_pos = fd_file(out)->f_pos;
1363 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1365 if (unlikely(pos + count > max)) {
1366 retval = -EOVERFLOW;
1375 * We need to debate whether we can enable this or not. The
1376 * man page documents EAGAIN return for the output at least,
1377 * and the application is arguably buggy if it doesn't expect
1378 * EAGAIN on a non-blocking file descriptor.
1380 if (fd_file(in)->f_flags & O_NONBLOCK)
1381 fl = SPLICE_F_NONBLOCK;
1383 opipe = get_pipe_info(fd_file(out), true);
1385 retval = rw_verify_area(WRITE, fd_file(out), &out_pos, count);
1388 retval = do_splice_direct(fd_file(in), &pos, fd_file(out), &out_pos,
1391 if (fd_file(out)->f_flags & O_NONBLOCK)
1392 fl |= SPLICE_F_NONBLOCK;
1394 retval = splice_file_to_pipe(fd_file(in), opipe, &pos, count, fl);
1398 add_rchar(current, retval);
1399 add_wchar(current, retval);
1400 fsnotify_access(fd_file(in));
1401 fsnotify_modify(fd_file(out));
1402 fd_file(out)->f_pos = out_pos;
1406 fd_file(in)->f_pos = pos;
1412 retval = -EOVERFLOW;
1422 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1429 if (unlikely(get_user(off, offset)))
1432 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1433 if (unlikely(put_user(pos, offset)))
1438 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1441 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1447 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1449 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1450 if (unlikely(put_user(pos, offset)))
1455 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1458 #ifdef CONFIG_COMPAT
1459 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1460 compat_off_t __user *, offset, compat_size_t, count)
1467 if (unlikely(get_user(off, offset)))
1470 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1471 if (unlikely(put_user(pos, offset)))
1476 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1479 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1480 compat_loff_t __user *, offset, compat_size_t, count)
1486 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1488 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1489 if (unlikely(put_user(pos, offset)))
1494 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1499 * Performs necessary checks before doing a file copy
1501 * Can adjust amount of bytes to copy via @req_count argument.
1502 * Returns appropriate error code that caller should return or
1503 * zero in case the copy should be allowed.
1505 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1506 struct file *file_out, loff_t pos_out,
1507 size_t *req_count, unsigned int flags)
1509 struct inode *inode_in = file_inode(file_in);
1510 struct inode *inode_out = file_inode(file_out);
1511 uint64_t count = *req_count;
1515 ret = generic_file_rw_checks(file_in, file_out);
1520 * We allow some filesystems to handle cross sb copy, but passing
1521 * a file of the wrong filesystem type to filesystem driver can result
1522 * in an attempt to dereference the wrong type of ->private_data, so
1523 * avoid doing that until we really have a good reason.
1525 * nfs and cifs define several different file_system_type structures
1526 * and several different sets of file_operations, but they all end up
1527 * using the same ->copy_file_range() function pointer.
1529 if (flags & COPY_FILE_SPLICE) {
1530 /* cross sb splice is allowed */
1531 } else if (file_out->f_op->copy_file_range) {
1532 if (file_in->f_op->copy_file_range !=
1533 file_out->f_op->copy_file_range)
1535 } else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {
1539 /* Don't touch certain kinds of inodes */
1540 if (IS_IMMUTABLE(inode_out))
1543 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1546 /* Ensure offsets don't wrap. */
1547 if (pos_in + count < pos_in || pos_out + count < pos_out)
1550 /* Shorten the copy to EOF */
1551 size_in = i_size_read(inode_in);
1552 if (pos_in >= size_in)
1555 count = min(count, size_in - (uint64_t)pos_in);
1557 ret = generic_write_check_limits(file_out, pos_out, &count);
1561 /* Don't allow overlapped copying within the same file. */
1562 if (inode_in == inode_out &&
1563 pos_out + count > pos_in &&
1564 pos_out < pos_in + count)
1572 * copy_file_range() differs from regular file read and write in that it
1573 * specifically allows return partial success. When it does so is up to
1574 * the copy_file_range method.
1576 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1577 struct file *file_out, loff_t pos_out,
1578 size_t len, unsigned int flags)
1581 bool splice = flags & COPY_FILE_SPLICE;
1582 bool samesb = file_inode(file_in)->i_sb == file_inode(file_out)->i_sb;
1584 if (flags & ~COPY_FILE_SPLICE)
1587 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1592 ret = rw_verify_area(READ, file_in, &pos_in, len);
1596 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1603 file_start_write(file_out);
1606 * Cloning is supported by more file systems, so we implement copy on
1607 * same sb using clone, but for filesystems where both clone and copy
1608 * are supported (e.g. nfs,cifs), we only call the copy method.
1610 if (!splice && file_out->f_op->copy_file_range) {
1611 ret = file_out->f_op->copy_file_range(file_in, pos_in,
1614 } else if (!splice && file_in->f_op->remap_file_range && samesb) {
1615 ret = file_in->f_op->remap_file_range(file_in, pos_in,
1617 min_t(loff_t, MAX_RW_COUNT, len),
1618 REMAP_FILE_CAN_SHORTEN);
1619 /* fallback to splice */
1622 } else if (samesb) {
1623 /* Fallback to splice for same sb copy for backward compat */
1627 file_end_write(file_out);
1633 * We can get here for same sb copy of filesystems that do not implement
1634 * ->copy_file_range() in case filesystem does not support clone or in
1635 * case filesystem supports clone but rejected the clone request (e.g.
1636 * because it was not block aligned).
1638 * In both cases, fall back to kernel copy so we are able to maintain a
1639 * consistent story about which filesystems support copy_file_range()
1640 * and which filesystems do not, that will allow userspace tools to
1641 * make consistent desicions w.r.t using copy_file_range().
1643 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE
1644 * for server-side-copy between any two sb.
1646 * In any case, we call do_splice_direct() and not splice_file_range(),
1647 * without file_start_write() held, to avoid possible deadlocks related
1648 * to splicing from input file, while file_start_write() is held on
1649 * the output file on a different sb.
1651 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1652 min_t(size_t, len, MAX_RW_COUNT), 0);
1655 fsnotify_access(file_in);
1656 add_rchar(current, ret);
1657 fsnotify_modify(file_out);
1658 add_wchar(current, ret);
1666 EXPORT_SYMBOL(vfs_copy_file_range);
1668 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1669 int, fd_out, loff_t __user *, off_out,
1670 size_t, len, unsigned int, flags)
1676 ssize_t ret = -EBADF;
1678 f_in = fdget(fd_in);
1682 f_out = fdget(fd_out);
1683 if (!fd_file(f_out))
1688 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1691 pos_in = fd_file(f_in)->f_pos;
1695 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1698 pos_out = fd_file(f_out)->f_pos;
1705 ret = vfs_copy_file_range(fd_file(f_in), pos_in, fd_file(f_out), pos_out, len,
1712 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1715 fd_file(f_in)->f_pos = pos_in;
1719 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1722 fd_file(f_out)->f_pos = pos_out;
1735 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1736 * LFS limits. If pos is under the limit it becomes a short access. If it
1737 * exceeds the limit we return -EFBIG.
1739 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1741 struct inode *inode = file->f_mapping->host;
1742 loff_t max_size = inode->i_sb->s_maxbytes;
1743 loff_t limit = rlimit(RLIMIT_FSIZE);
1745 if (limit != RLIM_INFINITY) {
1747 send_sig(SIGXFSZ, current, 0);
1750 *count = min(*count, limit - pos);
1753 if (!(file->f_flags & O_LARGEFILE))
1754 max_size = MAX_NON_LFS;
1756 if (unlikely(pos >= max_size))
1759 *count = min(*count, max_size - pos);
1763 EXPORT_SYMBOL_GPL(generic_write_check_limits);
1765 /* Like generic_write_checks(), but takes size of write instead of iter. */
1766 int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1768 struct file *file = iocb->ki_filp;
1769 struct inode *inode = file->f_mapping->host;
1771 if (IS_SWAPFILE(inode))
1777 if (iocb->ki_flags & IOCB_APPEND)
1778 iocb->ki_pos = i_size_read(inode);
1780 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1781 !((iocb->ki_flags & IOCB_DIRECT) ||
1782 (file->f_op->fop_flags & FOP_BUFFER_WASYNC)))
1785 return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1787 EXPORT_SYMBOL(generic_write_checks_count);
1790 * Performs necessary checks before doing a write
1792 * Can adjust writing position or amount of bytes to write.
1793 * Returns appropriate error code that caller should return or
1794 * zero in case that write should be allowed.
1796 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1798 loff_t count = iov_iter_count(from);
1801 ret = generic_write_checks_count(iocb, &count);
1805 iov_iter_truncate(from, count);
1806 return iov_iter_count(from);
1808 EXPORT_SYMBOL(generic_write_checks);
1811 * Performs common checks before doing a file copy/clone
1812 * from @file_in to @file_out.
1814 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1816 struct inode *inode_in = file_inode(file_in);
1817 struct inode *inode_out = file_inode(file_out);
1819 /* Don't copy dirs, pipes, sockets... */
1820 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1822 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1825 if (!(file_in->f_mode & FMODE_READ) ||
1826 !(file_out->f_mode & FMODE_WRITE) ||
1827 (file_out->f_flags & O_APPEND))
1833 bool generic_atomic_write_valid(struct iov_iter *iter, loff_t pos)
1835 size_t len = iov_iter_count(iter);
1837 if (!iter_is_ubuf(iter))
1840 if (!is_power_of_2(len))
1843 if (!IS_ALIGNED(pos, len))