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 CLASS(fd_pos, f)(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 */
403 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
405 return ksys_lseek(fd, offset, whence);
409 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
411 return ksys_lseek(fd, offset, whence);
415 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
416 defined(__ARCH_WANT_SYS_LLSEEK)
417 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
418 unsigned long, offset_low, loff_t __user *, result,
419 unsigned int, whence)
422 CLASS(fd_pos, f)(fd);
428 if (whence > SEEK_MAX)
431 offset = vfs_llseek(fd_file(f), ((loff_t) offset_high << 32) | offset_low,
434 retval = (int)offset;
437 if (!copy_to_user(result, &offset, sizeof(offset)))
444 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
446 int mask = read_write == READ ? MAY_READ : MAY_WRITE;
449 if (unlikely((ssize_t) count < 0))
455 if (unlikely(pos < 0)) {
456 if (!unsigned_offsets(file))
458 if (count >= -pos) /* both values are in 0..LLONG_MAX */
460 } else if (unlikely((loff_t) (pos + count) < 0)) {
461 if (!unsigned_offsets(file))
466 ret = security_file_permission(file, mask);
470 return fsnotify_file_area_perm(file, mask, ppos, count);
472 EXPORT_SYMBOL(rw_verify_area);
474 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
477 struct iov_iter iter;
480 init_sync_kiocb(&kiocb, filp);
481 kiocb.ki_pos = (ppos ? *ppos : 0);
482 iov_iter_ubuf(&iter, ITER_DEST, buf, len);
484 ret = filp->f_op->read_iter(&kiocb, &iter);
485 BUG_ON(ret == -EIOCBQUEUED);
487 *ppos = kiocb.ki_pos;
491 static int warn_unsupported(struct file *file, const char *op)
494 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
495 op, file, current->pid, current->comm);
499 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
503 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
506 struct iov_iter iter;
509 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
511 if (!(file->f_mode & FMODE_CAN_READ))
514 * Also fail if ->read_iter and ->read are both wired up as that
515 * implies very convoluted semantics.
517 if (unlikely(!file->f_op->read_iter || file->f_op->read))
518 return warn_unsupported(file, "read");
520 init_sync_kiocb(&kiocb, file);
521 kiocb.ki_pos = pos ? *pos : 0;
522 iov_iter_kvec(&iter, ITER_DEST, &iov, 1, iov.iov_len);
523 ret = file->f_op->read_iter(&kiocb, &iter);
527 fsnotify_access(file);
528 add_rchar(current, ret);
534 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
538 ret = rw_verify_area(READ, file, pos, count);
541 return __kernel_read(file, buf, count, pos);
543 EXPORT_SYMBOL(kernel_read);
545 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
549 if (!(file->f_mode & FMODE_READ))
551 if (!(file->f_mode & FMODE_CAN_READ))
553 if (unlikely(!access_ok(buf, count)))
556 ret = rw_verify_area(READ, file, pos, count);
559 if (count > MAX_RW_COUNT)
560 count = MAX_RW_COUNT;
562 if (file->f_op->read)
563 ret = file->f_op->read(file, buf, count, pos);
564 else if (file->f_op->read_iter)
565 ret = new_sync_read(file, buf, count, pos);
569 fsnotify_access(file);
570 add_rchar(current, ret);
576 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
579 struct iov_iter iter;
582 init_sync_kiocb(&kiocb, filp);
583 kiocb.ki_pos = (ppos ? *ppos : 0);
584 iov_iter_ubuf(&iter, ITER_SOURCE, (void __user *)buf, len);
586 ret = filp->f_op->write_iter(&kiocb, &iter);
587 BUG_ON(ret == -EIOCBQUEUED);
589 *ppos = kiocb.ki_pos;
593 /* caller is responsible for file_start_write/file_end_write */
594 ssize_t __kernel_write_iter(struct file *file, struct iov_iter *from, loff_t *pos)
599 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
601 if (!(file->f_mode & FMODE_CAN_WRITE))
604 * Also fail if ->write_iter and ->write are both wired up as that
605 * implies very convoluted semantics.
607 if (unlikely(!file->f_op->write_iter || file->f_op->write))
608 return warn_unsupported(file, "write");
610 init_sync_kiocb(&kiocb, file);
611 kiocb.ki_pos = pos ? *pos : 0;
612 ret = file->f_op->write_iter(&kiocb, from);
616 fsnotify_modify(file);
617 add_wchar(current, ret);
623 /* caller is responsible for file_start_write/file_end_write */
624 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
627 .iov_base = (void *)buf,
628 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
630 struct iov_iter iter;
631 iov_iter_kvec(&iter, ITER_SOURCE, &iov, 1, iov.iov_len);
632 return __kernel_write_iter(file, &iter, pos);
635 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
636 * but autofs is one of the few internal kernel users that actually
637 * wants this _and_ can be built as a module. So we need to export
638 * this symbol for autofs, even though it really isn't appropriate
639 * for any other kernel modules.
641 EXPORT_SYMBOL_GPL(__kernel_write);
643 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
648 ret = rw_verify_area(WRITE, file, pos, count);
652 file_start_write(file);
653 ret = __kernel_write(file, buf, count, pos);
654 file_end_write(file);
657 EXPORT_SYMBOL(kernel_write);
659 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
663 if (!(file->f_mode & FMODE_WRITE))
665 if (!(file->f_mode & FMODE_CAN_WRITE))
667 if (unlikely(!access_ok(buf, count)))
670 ret = rw_verify_area(WRITE, file, pos, count);
673 if (count > MAX_RW_COUNT)
674 count = MAX_RW_COUNT;
675 file_start_write(file);
676 if (file->f_op->write)
677 ret = file->f_op->write(file, buf, count, pos);
678 else if (file->f_op->write_iter)
679 ret = new_sync_write(file, buf, count, pos);
683 fsnotify_modify(file);
684 add_wchar(current, ret);
687 file_end_write(file);
691 /* file_ppos returns &file->f_pos or NULL if file is stream */
692 static inline loff_t *file_ppos(struct file *file)
694 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
697 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
699 CLASS(fd_pos, f)(fd);
700 ssize_t ret = -EBADF;
703 loff_t pos, *ppos = file_ppos(fd_file(f));
708 ret = vfs_read(fd_file(f), buf, count, ppos);
709 if (ret >= 0 && ppos)
710 fd_file(f)->f_pos = pos;
715 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
717 return ksys_read(fd, buf, count);
720 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
722 CLASS(fd_pos, f)(fd);
723 ssize_t ret = -EBADF;
726 loff_t pos, *ppos = file_ppos(fd_file(f));
731 ret = vfs_write(fd_file(f), buf, count, ppos);
732 if (ret >= 0 && ppos)
733 fd_file(f)->f_pos = pos;
739 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
742 return ksys_write(fd, buf, count);
745 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
755 if (fd_file(f)->f_mode & FMODE_PREAD)
756 return vfs_read(fd_file(f), buf, count, &pos);
761 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
762 size_t, count, loff_t, pos)
764 return ksys_pread64(fd, buf, count, pos);
767 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
768 COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,
769 size_t, count, compat_arg_u64_dual(pos))
771 return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));
775 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
776 size_t count, loff_t pos)
785 if (fd_file(f)->f_mode & FMODE_PWRITE)
786 return vfs_write(fd_file(f), buf, count, &pos);
791 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
792 size_t, count, loff_t, pos)
794 return ksys_pwrite64(fd, buf, count, pos);
797 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
798 COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,
799 size_t, count, compat_arg_u64_dual(pos))
801 return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));
805 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
806 loff_t *ppos, int type, rwf_t flags)
811 init_sync_kiocb(&kiocb, filp);
812 ret = kiocb_set_rw_flags(&kiocb, flags, type);
815 kiocb.ki_pos = (ppos ? *ppos : 0);
818 ret = filp->f_op->read_iter(&kiocb, iter);
820 ret = filp->f_op->write_iter(&kiocb, iter);
821 BUG_ON(ret == -EIOCBQUEUED);
823 *ppos = kiocb.ki_pos;
827 /* Do it by hand, with file-ops */
828 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
829 loff_t *ppos, int type, rwf_t flags)
833 if (flags & ~RWF_HIPRI)
836 while (iov_iter_count(iter)) {
840 nr = filp->f_op->read(filp, iter_iov_addr(iter),
841 iter_iov_len(iter), ppos);
843 nr = filp->f_op->write(filp, iter_iov_addr(iter),
844 iter_iov_len(iter), ppos);
853 if (nr != iter_iov_len(iter))
855 iov_iter_advance(iter, nr);
861 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
862 struct iov_iter *iter)
867 if (!file->f_op->read_iter)
869 if (!(file->f_mode & FMODE_READ))
871 if (!(file->f_mode & FMODE_CAN_READ))
874 tot_len = iov_iter_count(iter);
877 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
881 ret = file->f_op->read_iter(iocb, iter);
884 fsnotify_access(file);
887 EXPORT_SYMBOL(vfs_iocb_iter_read);
889 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
895 if (!file->f_op->read_iter)
897 if (!(file->f_mode & FMODE_READ))
899 if (!(file->f_mode & FMODE_CAN_READ))
902 tot_len = iov_iter_count(iter);
905 ret = rw_verify_area(READ, file, ppos, tot_len);
909 ret = do_iter_readv_writev(file, iter, ppos, READ, flags);
912 fsnotify_access(file);
915 EXPORT_SYMBOL(vfs_iter_read);
918 * Caller is responsible for calling kiocb_end_write() on completion
919 * if async iocb was queued.
921 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
922 struct iov_iter *iter)
927 if (!file->f_op->write_iter)
929 if (!(file->f_mode & FMODE_WRITE))
931 if (!(file->f_mode & FMODE_CAN_WRITE))
934 tot_len = iov_iter_count(iter);
937 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
941 kiocb_start_write(iocb);
942 ret = file->f_op->write_iter(iocb, iter);
943 if (ret != -EIOCBQUEUED)
944 kiocb_end_write(iocb);
946 fsnotify_modify(file);
950 EXPORT_SYMBOL(vfs_iocb_iter_write);
952 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
958 if (!(file->f_mode & FMODE_WRITE))
960 if (!(file->f_mode & FMODE_CAN_WRITE))
962 if (!file->f_op->write_iter)
965 tot_len = iov_iter_count(iter);
969 ret = rw_verify_area(WRITE, file, ppos, tot_len);
973 file_start_write(file);
974 ret = do_iter_readv_writev(file, iter, ppos, WRITE, flags);
976 fsnotify_modify(file);
977 file_end_write(file);
981 EXPORT_SYMBOL(vfs_iter_write);
983 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
984 unsigned long vlen, loff_t *pos, rwf_t flags)
986 struct iovec iovstack[UIO_FASTIOV];
987 struct iovec *iov = iovstack;
988 struct iov_iter iter;
992 if (!(file->f_mode & FMODE_READ))
994 if (!(file->f_mode & FMODE_CAN_READ))
997 ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov,
1002 tot_len = iov_iter_count(&iter);
1006 ret = rw_verify_area(READ, file, pos, tot_len);
1010 if (file->f_op->read_iter)
1011 ret = do_iter_readv_writev(file, &iter, pos, READ, flags);
1013 ret = do_loop_readv_writev(file, &iter, pos, READ, flags);
1016 fsnotify_access(file);
1021 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
1022 unsigned long vlen, loff_t *pos, rwf_t flags)
1024 struct iovec iovstack[UIO_FASTIOV];
1025 struct iovec *iov = iovstack;
1026 struct iov_iter iter;
1030 if (!(file->f_mode & FMODE_WRITE))
1032 if (!(file->f_mode & FMODE_CAN_WRITE))
1035 ret = import_iovec(ITER_SOURCE, vec, vlen, ARRAY_SIZE(iovstack), &iov,
1040 tot_len = iov_iter_count(&iter);
1044 ret = rw_verify_area(WRITE, file, pos, tot_len);
1048 file_start_write(file);
1049 if (file->f_op->write_iter)
1050 ret = do_iter_readv_writev(file, &iter, pos, WRITE, flags);
1052 ret = do_loop_readv_writev(file, &iter, pos, WRITE, flags);
1054 fsnotify_modify(file);
1055 file_end_write(file);
1061 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1062 unsigned long vlen, rwf_t flags)
1064 CLASS(fd_pos, f)(fd);
1065 ssize_t ret = -EBADF;
1068 loff_t pos, *ppos = file_ppos(fd_file(f));
1073 ret = vfs_readv(fd_file(f), vec, vlen, ppos, flags);
1074 if (ret >= 0 && ppos)
1075 fd_file(f)->f_pos = pos;
1079 add_rchar(current, ret);
1084 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1085 unsigned long vlen, rwf_t flags)
1087 CLASS(fd_pos, f)(fd);
1088 ssize_t ret = -EBADF;
1091 loff_t pos, *ppos = file_ppos(fd_file(f));
1096 ret = vfs_writev(fd_file(f), vec, vlen, ppos, flags);
1097 if (ret >= 0 && ppos)
1098 fd_file(f)->f_pos = pos;
1102 add_wchar(current, ret);
1107 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1109 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1110 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1113 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1114 unsigned long vlen, loff_t pos, rwf_t flags)
1116 ssize_t ret = -EBADF;
1124 if (fd_file(f)->f_mode & FMODE_PREAD)
1125 ret = vfs_readv(fd_file(f), vec, vlen, &pos, flags);
1129 add_rchar(current, ret);
1134 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1135 unsigned long vlen, loff_t pos, rwf_t flags)
1137 ssize_t ret = -EBADF;
1145 if (fd_file(f)->f_mode & FMODE_PWRITE)
1146 ret = vfs_writev(fd_file(f), vec, vlen, &pos, flags);
1150 add_wchar(current, ret);
1155 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1156 unsigned long, vlen)
1158 return do_readv(fd, vec, vlen, 0);
1161 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1162 unsigned long, vlen)
1164 return do_writev(fd, vec, vlen, 0);
1167 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1168 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1170 loff_t pos = pos_from_hilo(pos_h, pos_l);
1172 return do_preadv(fd, vec, vlen, pos, 0);
1175 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1176 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1179 loff_t pos = pos_from_hilo(pos_h, pos_l);
1182 return do_readv(fd, vec, vlen, flags);
1184 return do_preadv(fd, vec, vlen, pos, flags);
1187 SYSCALL_DEFINE5(pwritev, 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_pwritev(fd, vec, vlen, pos, 0);
1195 SYSCALL_DEFINE6(pwritev2, 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_writev(fd, vec, vlen, flags);
1204 return do_pwritev(fd, vec, vlen, pos, flags);
1208 * Various compat syscalls. Note that they all pretend to take a native
1209 * iovec - import_iovec will properly treat those as compat_iovecs based on
1210 * in_compat_syscall().
1212 #ifdef CONFIG_COMPAT
1213 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1214 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1215 const struct iovec __user *, vec,
1216 unsigned long, vlen, loff_t, pos)
1218 return do_preadv(fd, vec, vlen, pos, 0);
1222 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1223 const struct iovec __user *, vec,
1224 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1226 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1228 return do_preadv(fd, vec, vlen, pos, 0);
1231 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1232 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1233 const struct iovec __user *, vec,
1234 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1237 return do_readv(fd, vec, vlen, flags);
1238 return do_preadv(fd, vec, vlen, pos, flags);
1242 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1243 const struct iovec __user *, vec,
1244 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1247 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1250 return do_readv(fd, vec, vlen, flags);
1251 return do_preadv(fd, vec, vlen, pos, flags);
1254 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1255 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1256 const struct iovec __user *, vec,
1257 unsigned long, vlen, loff_t, pos)
1259 return do_pwritev(fd, vec, vlen, pos, 0);
1263 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1264 const struct iovec __user *,vec,
1265 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1267 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1269 return do_pwritev(fd, vec, vlen, pos, 0);
1272 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1273 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1274 const struct iovec __user *, vec,
1275 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1278 return do_writev(fd, vec, vlen, flags);
1279 return do_pwritev(fd, vec, vlen, pos, flags);
1283 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1284 const struct iovec __user *,vec,
1285 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1287 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1290 return do_writev(fd, vec, vlen, flags);
1291 return do_pwritev(fd, vec, vlen, pos, flags);
1293 #endif /* CONFIG_COMPAT */
1295 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1296 size_t count, loff_t max)
1298 struct inode *in_inode, *out_inode;
1299 struct pipe_inode_info *opipe;
1306 * Get input file, and verify that it is ok..
1308 CLASS(fd, in)(in_fd);
1311 if (!(fd_file(in)->f_mode & FMODE_READ))
1314 pos = fd_file(in)->f_pos;
1317 if (!(fd_file(in)->f_mode & FMODE_PREAD))
1320 retval = rw_verify_area(READ, fd_file(in), &pos, count);
1323 if (count > MAX_RW_COUNT)
1324 count = MAX_RW_COUNT;
1327 * Get output file, and verify that it is ok..
1329 CLASS(fd, out)(out_fd);
1332 if (!(fd_file(out)->f_mode & FMODE_WRITE))
1334 in_inode = file_inode(fd_file(in));
1335 out_inode = file_inode(fd_file(out));
1336 out_pos = fd_file(out)->f_pos;
1339 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1341 if (unlikely(pos + count > max)) {
1350 * We need to debate whether we can enable this or not. The
1351 * man page documents EAGAIN return for the output at least,
1352 * and the application is arguably buggy if it doesn't expect
1353 * EAGAIN on a non-blocking file descriptor.
1355 if (fd_file(in)->f_flags & O_NONBLOCK)
1356 fl = SPLICE_F_NONBLOCK;
1358 opipe = get_pipe_info(fd_file(out), true);
1360 retval = rw_verify_area(WRITE, fd_file(out), &out_pos, count);
1363 retval = do_splice_direct(fd_file(in), &pos, fd_file(out), &out_pos,
1366 if (fd_file(out)->f_flags & O_NONBLOCK)
1367 fl |= SPLICE_F_NONBLOCK;
1369 retval = splice_file_to_pipe(fd_file(in), opipe, &pos, count, fl);
1373 add_rchar(current, retval);
1374 add_wchar(current, retval);
1375 fsnotify_access(fd_file(in));
1376 fsnotify_modify(fd_file(out));
1377 fd_file(out)->f_pos = out_pos;
1381 fd_file(in)->f_pos = pos;
1387 retval = -EOVERFLOW;
1391 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1398 if (unlikely(get_user(off, offset)))
1401 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1402 if (unlikely(put_user(pos, offset)))
1407 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1410 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1416 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1418 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1419 if (unlikely(put_user(pos, offset)))
1424 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1427 #ifdef CONFIG_COMPAT
1428 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1429 compat_off_t __user *, offset, compat_size_t, count)
1436 if (unlikely(get_user(off, offset)))
1439 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1440 if (unlikely(put_user(pos, offset)))
1445 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1448 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1449 compat_loff_t __user *, offset, compat_size_t, count)
1455 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1457 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1458 if (unlikely(put_user(pos, offset)))
1463 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1468 * Performs necessary checks before doing a file copy
1470 * Can adjust amount of bytes to copy via @req_count argument.
1471 * Returns appropriate error code that caller should return or
1472 * zero in case the copy should be allowed.
1474 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1475 struct file *file_out, loff_t pos_out,
1476 size_t *req_count, unsigned int flags)
1478 struct inode *inode_in = file_inode(file_in);
1479 struct inode *inode_out = file_inode(file_out);
1480 uint64_t count = *req_count;
1484 ret = generic_file_rw_checks(file_in, file_out);
1489 * We allow some filesystems to handle cross sb copy, but passing
1490 * a file of the wrong filesystem type to filesystem driver can result
1491 * in an attempt to dereference the wrong type of ->private_data, so
1492 * avoid doing that until we really have a good reason.
1494 * nfs and cifs define several different file_system_type structures
1495 * and several different sets of file_operations, but they all end up
1496 * using the same ->copy_file_range() function pointer.
1498 if (flags & COPY_FILE_SPLICE) {
1499 /* cross sb splice is allowed */
1500 } else if (file_out->f_op->copy_file_range) {
1501 if (file_in->f_op->copy_file_range !=
1502 file_out->f_op->copy_file_range)
1504 } else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {
1508 /* Don't touch certain kinds of inodes */
1509 if (IS_IMMUTABLE(inode_out))
1512 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1515 /* Ensure offsets don't wrap. */
1516 if (pos_in + count < pos_in || pos_out + count < pos_out)
1519 /* Shorten the copy to EOF */
1520 size_in = i_size_read(inode_in);
1521 if (pos_in >= size_in)
1524 count = min(count, size_in - (uint64_t)pos_in);
1526 ret = generic_write_check_limits(file_out, pos_out, &count);
1530 /* Don't allow overlapped copying within the same file. */
1531 if (inode_in == inode_out &&
1532 pos_out + count > pos_in &&
1533 pos_out < pos_in + count)
1541 * copy_file_range() differs from regular file read and write in that it
1542 * specifically allows return partial success. When it does so is up to
1543 * the copy_file_range method.
1545 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1546 struct file *file_out, loff_t pos_out,
1547 size_t len, unsigned int flags)
1550 bool splice = flags & COPY_FILE_SPLICE;
1551 bool samesb = file_inode(file_in)->i_sb == file_inode(file_out)->i_sb;
1553 if (flags & ~COPY_FILE_SPLICE)
1556 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1561 ret = rw_verify_area(READ, file_in, &pos_in, len);
1565 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1572 file_start_write(file_out);
1575 * Cloning is supported by more file systems, so we implement copy on
1576 * same sb using clone, but for filesystems where both clone and copy
1577 * are supported (e.g. nfs,cifs), we only call the copy method.
1579 if (!splice && file_out->f_op->copy_file_range) {
1580 ret = file_out->f_op->copy_file_range(file_in, pos_in,
1583 } else if (!splice && file_in->f_op->remap_file_range && samesb) {
1584 ret = file_in->f_op->remap_file_range(file_in, pos_in,
1586 min_t(loff_t, MAX_RW_COUNT, len),
1587 REMAP_FILE_CAN_SHORTEN);
1588 /* fallback to splice */
1591 } else if (samesb) {
1592 /* Fallback to splice for same sb copy for backward compat */
1596 file_end_write(file_out);
1602 * We can get here for same sb copy of filesystems that do not implement
1603 * ->copy_file_range() in case filesystem does not support clone or in
1604 * case filesystem supports clone but rejected the clone request (e.g.
1605 * because it was not block aligned).
1607 * In both cases, fall back to kernel copy so we are able to maintain a
1608 * consistent story about which filesystems support copy_file_range()
1609 * and which filesystems do not, that will allow userspace tools to
1610 * make consistent desicions w.r.t using copy_file_range().
1612 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE
1613 * for server-side-copy between any two sb.
1615 * In any case, we call do_splice_direct() and not splice_file_range(),
1616 * without file_start_write() held, to avoid possible deadlocks related
1617 * to splicing from input file, while file_start_write() is held on
1618 * the output file on a different sb.
1620 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1621 min_t(size_t, len, MAX_RW_COUNT), 0);
1624 fsnotify_access(file_in);
1625 add_rchar(current, ret);
1626 fsnotify_modify(file_out);
1627 add_wchar(current, ret);
1635 EXPORT_SYMBOL(vfs_copy_file_range);
1637 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1638 int, fd_out, loff_t __user *, off_out,
1639 size_t, len, unsigned int, flags)
1643 ssize_t ret = -EBADF;
1645 CLASS(fd, f_in)(fd_in);
1649 CLASS(fd, f_out)(fd_out);
1650 if (fd_empty(f_out))
1654 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1657 pos_in = fd_file(f_in)->f_pos;
1661 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1664 pos_out = fd_file(f_out)->f_pos;
1670 ret = vfs_copy_file_range(fd_file(f_in), pos_in, fd_file(f_out), pos_out, len,
1677 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1680 fd_file(f_in)->f_pos = pos_in;
1684 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1687 fd_file(f_out)->f_pos = pos_out;
1694 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1695 * LFS limits. If pos is under the limit it becomes a short access. If it
1696 * exceeds the limit we return -EFBIG.
1698 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1700 struct inode *inode = file->f_mapping->host;
1701 loff_t max_size = inode->i_sb->s_maxbytes;
1702 loff_t limit = rlimit(RLIMIT_FSIZE);
1704 if (limit != RLIM_INFINITY) {
1706 send_sig(SIGXFSZ, current, 0);
1709 *count = min(*count, limit - pos);
1712 if (!(file->f_flags & O_LARGEFILE))
1713 max_size = MAX_NON_LFS;
1715 if (unlikely(pos >= max_size))
1718 *count = min(*count, max_size - pos);
1722 EXPORT_SYMBOL_GPL(generic_write_check_limits);
1724 /* Like generic_write_checks(), but takes size of write instead of iter. */
1725 int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1727 struct file *file = iocb->ki_filp;
1728 struct inode *inode = file->f_mapping->host;
1730 if (IS_SWAPFILE(inode))
1736 if (iocb->ki_flags & IOCB_APPEND)
1737 iocb->ki_pos = i_size_read(inode);
1739 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1740 !((iocb->ki_flags & IOCB_DIRECT) ||
1741 (file->f_op->fop_flags & FOP_BUFFER_WASYNC)))
1744 return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1746 EXPORT_SYMBOL(generic_write_checks_count);
1749 * Performs necessary checks before doing a write
1751 * Can adjust writing position or amount of bytes to write.
1752 * Returns appropriate error code that caller should return or
1753 * zero in case that write should be allowed.
1755 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1757 loff_t count = iov_iter_count(from);
1760 ret = generic_write_checks_count(iocb, &count);
1764 iov_iter_truncate(from, count);
1765 return iov_iter_count(from);
1767 EXPORT_SYMBOL(generic_write_checks);
1770 * Performs common checks before doing a file copy/clone
1771 * from @file_in to @file_out.
1773 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1775 struct inode *inode_in = file_inode(file_in);
1776 struct inode *inode_out = file_inode(file_out);
1778 /* Don't copy dirs, pipes, sockets... */
1779 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1781 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1784 if (!(file_in->f_mode & FMODE_READ) ||
1785 !(file_out->f_mode & FMODE_WRITE) ||
1786 (file_out->f_flags & O_APPEND))
1792 int generic_atomic_write_valid(struct kiocb *iocb, struct iov_iter *iter)
1794 size_t len = iov_iter_count(iter);
1796 if (!iter_is_ubuf(iter))
1799 if (!is_power_of_2(len))
1802 if (!IS_ALIGNED(iocb->ki_pos, len))
1805 if (!(iocb->ki_flags & IOCB_DIRECT))
1810 EXPORT_SYMBOL_GPL(generic_atomic_write_valid);