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)
368 if (unlikely((ssize_t) count < 0))
374 if (unlikely(pos < 0)) {
375 if (!unsigned_offsets(file))
377 if (count >= -pos) /* both values are in 0..LLONG_MAX */
379 } else if (unlikely((loff_t) (pos + count) < 0)) {
380 if (!unsigned_offsets(file))
385 return security_file_permission(file,
386 read_write == READ ? MAY_READ : MAY_WRITE);
389 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
391 struct iovec iov = { .iov_base = buf, .iov_len = len };
393 struct iov_iter iter;
396 init_sync_kiocb(&kiocb, filp);
397 kiocb.ki_pos = (ppos ? *ppos : 0);
398 iov_iter_init(&iter, READ, &iov, 1, len);
400 ret = call_read_iter(filp, &kiocb, &iter);
401 BUG_ON(ret == -EIOCBQUEUED);
403 *ppos = kiocb.ki_pos;
407 static int warn_unsupported(struct file *file, const char *op)
410 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
411 op, file, current->pid, current->comm);
415 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
419 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
422 struct iov_iter iter;
425 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
427 if (!(file->f_mode & FMODE_CAN_READ))
430 * Also fail if ->read_iter and ->read are both wired up as that
431 * implies very convoluted semantics.
433 if (unlikely(!file->f_op->read_iter || file->f_op->read))
434 return warn_unsupported(file, "read");
436 init_sync_kiocb(&kiocb, file);
437 kiocb.ki_pos = pos ? *pos : 0;
438 iov_iter_kvec(&iter, READ, &iov, 1, iov.iov_len);
439 ret = file->f_op->read_iter(&kiocb, &iter);
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 .iov_base = (void *)buf,
515 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
518 struct iov_iter iter;
521 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
523 if (!(file->f_mode & FMODE_CAN_WRITE))
526 * Also fail if ->write_iter and ->write are both wired up as that
527 * implies very convoluted semantics.
529 if (unlikely(!file->f_op->write_iter || file->f_op->write))
530 return warn_unsupported(file, "write");
532 init_sync_kiocb(&kiocb, file);
533 kiocb.ki_pos = pos ? *pos : 0;
534 iov_iter_kvec(&iter, WRITE, &iov, 1, iov.iov_len);
535 ret = file->f_op->write_iter(&kiocb, &iter);
539 fsnotify_modify(file);
540 add_wchar(current, ret);
546 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
547 * but autofs is one of the few internal kernel users that actually
548 * wants this _and_ can be built as a module. So we need to export
549 * this symbol for autofs, even though it really isn't appropriate
550 * for any other kernel modules.
552 EXPORT_SYMBOL_GPL(__kernel_write);
554 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
559 ret = rw_verify_area(WRITE, file, pos, count);
563 file_start_write(file);
564 ret = __kernel_write(file, buf, count, pos);
565 file_end_write(file);
568 EXPORT_SYMBOL(kernel_write);
570 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
574 if (!(file->f_mode & FMODE_WRITE))
576 if (!(file->f_mode & FMODE_CAN_WRITE))
578 if (unlikely(!access_ok(buf, count)))
581 ret = rw_verify_area(WRITE, file, pos, count);
584 if (count > MAX_RW_COUNT)
585 count = MAX_RW_COUNT;
586 file_start_write(file);
587 if (file->f_op->write)
588 ret = file->f_op->write(file, buf, count, pos);
589 else if (file->f_op->write_iter)
590 ret = new_sync_write(file, buf, count, pos);
594 fsnotify_modify(file);
595 add_wchar(current, ret);
598 file_end_write(file);
602 /* file_ppos returns &file->f_pos or NULL if file is stream */
603 static inline loff_t *file_ppos(struct file *file)
605 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
608 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
610 struct fd f = fdget_pos(fd);
611 ssize_t ret = -EBADF;
614 loff_t pos, *ppos = file_ppos(f.file);
619 ret = vfs_read(f.file, buf, count, ppos);
620 if (ret >= 0 && ppos)
627 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
629 return ksys_read(fd, buf, count);
632 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
634 struct fd f = fdget_pos(fd);
635 ssize_t ret = -EBADF;
638 loff_t pos, *ppos = file_ppos(f.file);
643 ret = vfs_write(f.file, buf, count, ppos);
644 if (ret >= 0 && ppos)
652 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
655 return ksys_write(fd, buf, count);
658 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
662 ssize_t ret = -EBADF;
670 if (f.file->f_mode & FMODE_PREAD)
671 ret = vfs_read(f.file, buf, count, &pos);
678 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
679 size_t, count, loff_t, pos)
681 return ksys_pread64(fd, buf, count, pos);
684 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
685 size_t count, loff_t pos)
688 ssize_t ret = -EBADF;
696 if (f.file->f_mode & FMODE_PWRITE)
697 ret = vfs_write(f.file, buf, count, &pos);
704 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
705 size_t, count, loff_t, pos)
707 return ksys_pwrite64(fd, buf, count, pos);
710 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
711 loff_t *ppos, int type, rwf_t flags)
716 init_sync_kiocb(&kiocb, filp);
717 ret = kiocb_set_rw_flags(&kiocb, flags);
720 kiocb.ki_pos = (ppos ? *ppos : 0);
723 ret = call_read_iter(filp, &kiocb, iter);
725 ret = call_write_iter(filp, &kiocb, iter);
726 BUG_ON(ret == -EIOCBQUEUED);
728 *ppos = kiocb.ki_pos;
732 /* Do it by hand, with file-ops */
733 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
734 loff_t *ppos, int type, rwf_t flags)
738 if (flags & ~RWF_HIPRI)
741 while (iov_iter_count(iter)) {
742 struct iovec iovec = iov_iter_iovec(iter);
746 nr = filp->f_op->read(filp, iovec.iov_base,
747 iovec.iov_len, ppos);
749 nr = filp->f_op->write(filp, iovec.iov_base,
750 iovec.iov_len, ppos);
759 if (nr != iovec.iov_len)
761 iov_iter_advance(iter, nr);
767 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
768 loff_t *pos, rwf_t flags)
773 if (!(file->f_mode & FMODE_READ))
775 if (!(file->f_mode & FMODE_CAN_READ))
778 tot_len = iov_iter_count(iter);
781 ret = rw_verify_area(READ, file, pos, tot_len);
785 if (file->f_op->read_iter)
786 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
788 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
791 fsnotify_access(file);
795 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
796 struct iov_iter *iter)
801 if (!file->f_op->read_iter)
803 if (!(file->f_mode & FMODE_READ))
805 if (!(file->f_mode & FMODE_CAN_READ))
808 tot_len = iov_iter_count(iter);
811 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
815 ret = call_read_iter(file, iocb, iter);
818 fsnotify_access(file);
821 EXPORT_SYMBOL(vfs_iocb_iter_read);
823 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
826 if (!file->f_op->read_iter)
828 return do_iter_read(file, iter, ppos, flags);
830 EXPORT_SYMBOL(vfs_iter_read);
832 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
833 loff_t *pos, rwf_t flags)
838 if (!(file->f_mode & FMODE_WRITE))
840 if (!(file->f_mode & FMODE_CAN_WRITE))
843 tot_len = iov_iter_count(iter);
846 ret = rw_verify_area(WRITE, file, pos, tot_len);
850 if (file->f_op->write_iter)
851 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
853 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
855 fsnotify_modify(file);
859 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
860 struct iov_iter *iter)
865 if (!file->f_op->write_iter)
867 if (!(file->f_mode & FMODE_WRITE))
869 if (!(file->f_mode & FMODE_CAN_WRITE))
872 tot_len = iov_iter_count(iter);
875 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
879 ret = call_write_iter(file, iocb, iter);
881 fsnotify_modify(file);
885 EXPORT_SYMBOL(vfs_iocb_iter_write);
887 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
890 if (!file->f_op->write_iter)
892 return do_iter_write(file, iter, ppos, flags);
894 EXPORT_SYMBOL(vfs_iter_write);
896 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
897 unsigned long vlen, loff_t *pos, rwf_t flags)
899 struct iovec iovstack[UIO_FASTIOV];
900 struct iovec *iov = iovstack;
901 struct iov_iter iter;
904 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
906 ret = do_iter_read(file, &iter, pos, flags);
913 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
914 unsigned long vlen, loff_t *pos, rwf_t flags)
916 struct iovec iovstack[UIO_FASTIOV];
917 struct iovec *iov = iovstack;
918 struct iov_iter iter;
921 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
923 file_start_write(file);
924 ret = do_iter_write(file, &iter, pos, flags);
925 file_end_write(file);
931 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
932 unsigned long vlen, rwf_t flags)
934 struct fd f = fdget_pos(fd);
935 ssize_t ret = -EBADF;
938 loff_t pos, *ppos = file_ppos(f.file);
943 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
944 if (ret >= 0 && ppos)
950 add_rchar(current, ret);
955 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
956 unsigned long vlen, rwf_t flags)
958 struct fd f = fdget_pos(fd);
959 ssize_t ret = -EBADF;
962 loff_t pos, *ppos = file_ppos(f.file);
967 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
968 if (ret >= 0 && ppos)
974 add_wchar(current, ret);
979 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
981 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
982 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
985 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
986 unsigned long vlen, loff_t pos, rwf_t flags)
989 ssize_t ret = -EBADF;
997 if (f.file->f_mode & FMODE_PREAD)
998 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1003 add_rchar(current, ret);
1008 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1009 unsigned long vlen, loff_t pos, rwf_t flags)
1012 ssize_t ret = -EBADF;
1020 if (f.file->f_mode & FMODE_PWRITE)
1021 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1026 add_wchar(current, ret);
1031 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1032 unsigned long, vlen)
1034 return do_readv(fd, vec, vlen, 0);
1037 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1038 unsigned long, vlen)
1040 return do_writev(fd, vec, vlen, 0);
1043 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1044 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1046 loff_t pos = pos_from_hilo(pos_h, pos_l);
1048 return do_preadv(fd, vec, vlen, pos, 0);
1051 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1052 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1055 loff_t pos = pos_from_hilo(pos_h, pos_l);
1058 return do_readv(fd, vec, vlen, flags);
1060 return do_preadv(fd, vec, vlen, pos, flags);
1063 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1064 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1066 loff_t pos = pos_from_hilo(pos_h, pos_l);
1068 return do_pwritev(fd, vec, vlen, pos, 0);
1071 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1072 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1075 loff_t pos = pos_from_hilo(pos_h, pos_l);
1078 return do_writev(fd, vec, vlen, flags);
1080 return do_pwritev(fd, vec, vlen, pos, flags);
1084 * Various compat syscalls. Note that they all pretend to take a native
1085 * iovec - import_iovec will properly treat those as compat_iovecs based on
1086 * in_compat_syscall().
1088 #ifdef CONFIG_COMPAT
1089 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1090 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1091 const struct iovec __user *, vec,
1092 unsigned long, vlen, loff_t, pos)
1094 return do_preadv(fd, vec, vlen, pos, 0);
1098 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1099 const struct iovec __user *, vec,
1100 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1102 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1104 return do_preadv(fd, vec, vlen, pos, 0);
1107 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1108 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1109 const struct iovec __user *, vec,
1110 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1113 return do_readv(fd, vec, vlen, flags);
1114 return do_preadv(fd, vec, vlen, pos, flags);
1118 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1119 const struct iovec __user *, vec,
1120 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1123 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1126 return do_readv(fd, vec, vlen, flags);
1127 return do_preadv(fd, vec, vlen, pos, flags);
1130 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1131 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1132 const struct iovec __user *, vec,
1133 unsigned long, vlen, loff_t, pos)
1135 return do_pwritev(fd, vec, vlen, pos, 0);
1139 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1140 const struct iovec __user *,vec,
1141 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1143 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1145 return do_pwritev(fd, vec, vlen, pos, 0);
1148 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1149 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1150 const struct iovec __user *, vec,
1151 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1154 return do_writev(fd, vec, vlen, flags);
1155 return do_pwritev(fd, vec, vlen, pos, flags);
1159 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1160 const struct iovec __user *,vec,
1161 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1163 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1166 return do_writev(fd, vec, vlen, flags);
1167 return do_pwritev(fd, vec, vlen, pos, flags);
1169 #endif /* CONFIG_COMPAT */
1171 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1172 size_t count, loff_t max)
1175 struct inode *in_inode, *out_inode;
1176 struct pipe_inode_info *opipe;
1183 * Get input file, and verify that it is ok..
1189 if (!(in.file->f_mode & FMODE_READ))
1193 pos = in.file->f_pos;
1196 if (!(in.file->f_mode & FMODE_PREAD))
1199 retval = rw_verify_area(READ, in.file, &pos, count);
1202 if (count > MAX_RW_COUNT)
1203 count = MAX_RW_COUNT;
1206 * Get output file, and verify that it is ok..
1209 out = fdget(out_fd);
1212 if (!(out.file->f_mode & FMODE_WRITE))
1214 in_inode = file_inode(in.file);
1215 out_inode = file_inode(out.file);
1216 out_pos = out.file->f_pos;
1219 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1221 if (unlikely(pos + count > max)) {
1222 retval = -EOVERFLOW;
1231 * We need to debate whether we can enable this or not. The
1232 * man page documents EAGAIN return for the output at least,
1233 * and the application is arguably buggy if it doesn't expect
1234 * EAGAIN on a non-blocking file descriptor.
1236 if (in.file->f_flags & O_NONBLOCK)
1237 fl = SPLICE_F_NONBLOCK;
1239 opipe = get_pipe_info(out.file, true);
1241 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1244 file_start_write(out.file);
1245 retval = do_splice_direct(in.file, &pos, out.file, &out_pos,
1247 file_end_write(out.file);
1249 retval = splice_file_to_pipe(in.file, opipe, &pos, count, fl);
1253 add_rchar(current, retval);
1254 add_wchar(current, retval);
1255 fsnotify_access(in.file);
1256 fsnotify_modify(out.file);
1257 out.file->f_pos = out_pos;
1261 in.file->f_pos = pos;
1267 retval = -EOVERFLOW;
1277 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1284 if (unlikely(get_user(off, offset)))
1287 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1288 if (unlikely(put_user(pos, offset)))
1293 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1296 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1302 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1304 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1305 if (unlikely(put_user(pos, offset)))
1310 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1313 #ifdef CONFIG_COMPAT
1314 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1315 compat_off_t __user *, offset, compat_size_t, count)
1322 if (unlikely(get_user(off, offset)))
1325 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1326 if (unlikely(put_user(pos, offset)))
1331 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1334 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1335 compat_loff_t __user *, offset, compat_size_t, count)
1341 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1343 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1344 if (unlikely(put_user(pos, offset)))
1349 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1354 * generic_copy_file_range - copy data between two files
1355 * @file_in: file structure to read from
1356 * @pos_in: file offset to read from
1357 * @file_out: file structure to write data to
1358 * @pos_out: file offset to write data to
1359 * @len: amount of data to copy
1360 * @flags: copy flags
1362 * This is a generic filesystem helper to copy data from one file to another.
1363 * It has no constraints on the source or destination file owners - the files
1364 * can belong to different superblocks and different filesystem types. Short
1365 * copies are allowed.
1367 * This should be called from the @file_out filesystem, as per the
1368 * ->copy_file_range() method.
1370 * Returns the number of bytes copied or a negative error indicating the
1374 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1375 struct file *file_out, loff_t pos_out,
1376 size_t len, unsigned int flags)
1378 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1379 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1381 EXPORT_SYMBOL(generic_copy_file_range);
1383 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1384 struct file *file_out, loff_t pos_out,
1385 size_t len, unsigned int flags)
1388 * Although we now allow filesystems to handle cross sb copy, passing
1389 * a file of the wrong filesystem type to filesystem driver can result
1390 * in an attempt to dereference the wrong type of ->private_data, so
1391 * avoid doing that until we really have a good reason. NFS defines
1392 * several different file_system_type structures, but they all end up
1393 * using the same ->copy_file_range() function pointer.
1395 if (file_out->f_op->copy_file_range &&
1396 file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1397 return file_out->f_op->copy_file_range(file_in, pos_in,
1401 return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1406 * Performs necessary checks before doing a file copy
1408 * Can adjust amount of bytes to copy via @req_count argument.
1409 * Returns appropriate error code that caller should return or
1410 * zero in case the copy should be allowed.
1412 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1413 struct file *file_out, loff_t pos_out,
1414 size_t *req_count, unsigned int flags)
1416 struct inode *inode_in = file_inode(file_in);
1417 struct inode *inode_out = file_inode(file_out);
1418 uint64_t count = *req_count;
1422 ret = generic_file_rw_checks(file_in, file_out);
1426 /* Don't touch certain kinds of inodes */
1427 if (IS_IMMUTABLE(inode_out))
1430 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1433 /* Ensure offsets don't wrap. */
1434 if (pos_in + count < pos_in || pos_out + count < pos_out)
1437 /* Shorten the copy to EOF */
1438 size_in = i_size_read(inode_in);
1439 if (pos_in >= size_in)
1442 count = min(count, size_in - (uint64_t)pos_in);
1444 ret = generic_write_check_limits(file_out, pos_out, &count);
1448 /* Don't allow overlapped copying within the same file. */
1449 if (inode_in == inode_out &&
1450 pos_out + count > pos_in &&
1451 pos_out < pos_in + count)
1459 * copy_file_range() differs from regular file read and write in that it
1460 * specifically allows return partial success. When it does so is up to
1461 * the copy_file_range method.
1463 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1464 struct file *file_out, loff_t pos_out,
1465 size_t len, unsigned int flags)
1472 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1477 ret = rw_verify_area(READ, file_in, &pos_in, len);
1481 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1488 file_start_write(file_out);
1491 * Try cloning first, this is supported by more file systems, and
1492 * more efficient if both clone and copy are supported (e.g. NFS).
1494 if (file_in->f_op->remap_file_range &&
1495 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1498 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1500 min_t(loff_t, MAX_RW_COUNT, len),
1501 REMAP_FILE_CAN_SHORTEN);
1508 ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1510 WARN_ON_ONCE(ret == -EOPNOTSUPP);
1513 fsnotify_access(file_in);
1514 add_rchar(current, ret);
1515 fsnotify_modify(file_out);
1516 add_wchar(current, ret);
1522 file_end_write(file_out);
1526 EXPORT_SYMBOL(vfs_copy_file_range);
1528 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1529 int, fd_out, loff_t __user *, off_out,
1530 size_t, len, unsigned int, flags)
1536 ssize_t ret = -EBADF;
1538 f_in = fdget(fd_in);
1542 f_out = fdget(fd_out);
1548 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1551 pos_in = f_in.file->f_pos;
1555 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1558 pos_out = f_out.file->f_pos;
1561 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1568 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1571 f_in.file->f_pos = pos_in;
1575 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1578 f_out.file->f_pos = pos_out;
1591 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1592 * LFS limits. If pos is under the limit it becomes a short access. If it
1593 * exceeds the limit we return -EFBIG.
1595 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1597 struct inode *inode = file->f_mapping->host;
1598 loff_t max_size = inode->i_sb->s_maxbytes;
1599 loff_t limit = rlimit(RLIMIT_FSIZE);
1601 if (limit != RLIM_INFINITY) {
1603 send_sig(SIGXFSZ, current, 0);
1606 *count = min(*count, limit - pos);
1609 if (!(file->f_flags & O_LARGEFILE))
1610 max_size = MAX_NON_LFS;
1612 if (unlikely(pos >= max_size))
1615 *count = min(*count, max_size - pos);
1621 * Performs necessary checks before doing a write
1623 * Can adjust writing position or amount of bytes to write.
1624 * Returns appropriate error code that caller should return or
1625 * zero in case that write should be allowed.
1627 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1629 struct file *file = iocb->ki_filp;
1630 struct inode *inode = file->f_mapping->host;
1634 if (IS_SWAPFILE(inode))
1637 if (!iov_iter_count(from))
1640 /* FIXME: this is for backwards compatibility with 2.4 */
1641 if (iocb->ki_flags & IOCB_APPEND)
1642 iocb->ki_pos = i_size_read(inode);
1644 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
1647 count = iov_iter_count(from);
1648 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
1652 iov_iter_truncate(from, count);
1653 return iov_iter_count(from);
1655 EXPORT_SYMBOL(generic_write_checks);
1658 * Performs common checks before doing a file copy/clone
1659 * from @file_in to @file_out.
1661 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1663 struct inode *inode_in = file_inode(file_in);
1664 struct inode *inode_out = file_inode(file_out);
1666 /* Don't copy dirs, pipes, sockets... */
1667 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1669 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1672 if (!(file_in->f_mode & FMODE_READ) ||
1673 !(file_out->f_mode & FMODE_WRITE) ||
1674 (file_out->f_flags & O_APPEND))