2 * linux/fs/read_write.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/smp_lock.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/module.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include "read_write.h"
21 #include <asm/uaccess.h>
22 #include <asm/unistd.h>
24 const struct file_operations generic_ro_fops = {
25 .llseek = generic_file_llseek,
27 .aio_read = generic_file_aio_read,
28 .mmap = generic_file_readonly_mmap,
29 .splice_read = generic_file_splice_read,
32 EXPORT_SYMBOL(generic_ro_fops);
35 * generic_file_llseek_unlocked - lockless generic llseek implementation
36 * @file: file structure to seek on
37 * @offset: file offset to seek to
38 * @origin: type of seek
40 * Updates the file offset to the value specified by @offset and @origin.
41 * Locking must be provided by the caller.
44 generic_file_llseek_unlocked(struct file *file, loff_t offset, int origin)
46 struct inode *inode = file->f_mapping->host;
50 offset += inode->i_size;
54 * Here we special-case the lseek(fd, 0, SEEK_CUR)
55 * position-querying operation. Avoid rewriting the "same"
56 * f_pos value back to the file because a concurrent read(),
57 * write() or lseek() might have altered it
61 offset += file->f_pos;
65 if (offset < 0 || offset > inode->i_sb->s_maxbytes)
68 /* Special lock needed here? */
69 if (offset != file->f_pos) {
76 EXPORT_SYMBOL(generic_file_llseek_unlocked);
79 * generic_file_llseek - generic llseek implementation for regular files
80 * @file: file structure to seek on
81 * @offset: file offset to seek to
82 * @origin: type of seek
84 * This is a generic implemenation of ->llseek useable for all normal local
85 * filesystems. It just updates the file offset to the value specified by
86 * @offset and @origin under i_mutex.
88 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
92 mutex_lock(&file->f_dentry->d_inode->i_mutex);
93 rval = generic_file_llseek_unlocked(file, offset, origin);
94 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
98 EXPORT_SYMBOL(generic_file_llseek);
101 * noop_llseek - No Operation Performed llseek implementation
102 * @file: file structure to seek on
103 * @offset: file offset to seek to
104 * @origin: type of seek
106 * This is an implementation of ->llseek useable for the rare special case when
107 * userspace expects the seek to succeed but the (device) file is actually not
108 * able to perform the seek. In this case you use noop_llseek() instead of
109 * falling back to the default implementation of ->llseek.
111 loff_t noop_llseek(struct file *file, loff_t offset, int origin)
115 EXPORT_SYMBOL(noop_llseek);
117 loff_t no_llseek(struct file *file, loff_t offset, int origin)
121 EXPORT_SYMBOL(no_llseek);
123 loff_t default_llseek(struct file *file, loff_t offset, int origin)
127 mutex_lock(&file->f_dentry->d_inode->i_mutex);
130 offset += i_size_read(file->f_path.dentry->d_inode);
134 retval = file->f_pos;
137 offset += file->f_pos;
141 if (offset != file->f_pos) {
142 file->f_pos = offset;
148 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
151 EXPORT_SYMBOL(default_llseek);
153 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
155 loff_t (*fn)(struct file *, loff_t, int);
158 if (file->f_mode & FMODE_LSEEK) {
159 if (file->f_op && file->f_op->llseek)
160 fn = file->f_op->llseek;
162 return fn(file, offset, origin);
164 EXPORT_SYMBOL(vfs_llseek);
166 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
173 file = fget_light(fd, &fput_needed);
178 if (origin <= SEEK_MAX) {
179 loff_t res = vfs_llseek(file, offset, origin);
181 if (res != (loff_t)retval)
182 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
184 fput_light(file, fput_needed);
189 #ifdef __ARCH_WANT_SYS_LLSEEK
190 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
191 unsigned long, offset_low, loff_t __user *, result,
192 unsigned int, origin)
200 file = fget_light(fd, &fput_needed);
205 if (origin > SEEK_MAX)
208 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
211 retval = (int)offset;
214 if (!copy_to_user(result, &offset, sizeof(offset)))
218 fput_light(file, fput_needed);
225 * rw_verify_area doesn't like huge counts. We limit
226 * them to something that fits in "int" so that others
227 * won't have to do range checks all the time.
229 #define MAX_RW_COUNT (INT_MAX & PAGE_CACHE_MASK)
231 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
235 int retval = -EINVAL;
237 inode = file->f_path.dentry->d_inode;
238 if (unlikely((ssize_t) count < 0))
241 if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
244 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
245 retval = locks_mandatory_area(
246 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
247 inode, file, pos, count);
251 retval = security_file_permission(file,
252 read_write == READ ? MAY_READ : MAY_WRITE);
255 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
258 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
260 set_current_state(TASK_UNINTERRUPTIBLE);
261 if (!kiocbIsKicked(iocb))
264 kiocbClearKicked(iocb);
265 __set_current_state(TASK_RUNNING);
268 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
270 struct iovec iov = { .iov_base = buf, .iov_len = len };
274 init_sync_kiocb(&kiocb, filp);
275 kiocb.ki_pos = *ppos;
277 kiocb.ki_nbytes = len;
280 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
281 if (ret != -EIOCBRETRY)
283 wait_on_retry_sync_kiocb(&kiocb);
286 if (-EIOCBQUEUED == ret)
287 ret = wait_on_sync_kiocb(&kiocb);
288 *ppos = kiocb.ki_pos;
292 EXPORT_SYMBOL(do_sync_read);
294 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
298 if (!(file->f_mode & FMODE_READ))
300 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
302 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
305 ret = rw_verify_area(READ, file, pos, count);
308 if (file->f_op->read)
309 ret = file->f_op->read(file, buf, count, pos);
311 ret = do_sync_read(file, buf, count, pos);
313 fsnotify_access(file);
314 add_rchar(current, ret);
322 EXPORT_SYMBOL(vfs_read);
324 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
326 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
330 init_sync_kiocb(&kiocb, filp);
331 kiocb.ki_pos = *ppos;
333 kiocb.ki_nbytes = len;
336 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
337 if (ret != -EIOCBRETRY)
339 wait_on_retry_sync_kiocb(&kiocb);
342 if (-EIOCBQUEUED == ret)
343 ret = wait_on_sync_kiocb(&kiocb);
344 *ppos = kiocb.ki_pos;
348 EXPORT_SYMBOL(do_sync_write);
350 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
354 if (!(file->f_mode & FMODE_WRITE))
356 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
358 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
361 ret = rw_verify_area(WRITE, file, pos, count);
364 if (file->f_op->write)
365 ret = file->f_op->write(file, buf, count, pos);
367 ret = do_sync_write(file, buf, count, pos);
369 fsnotify_modify(file);
370 add_wchar(current, ret);
378 EXPORT_SYMBOL(vfs_write);
380 static inline loff_t file_pos_read(struct file *file)
385 static inline void file_pos_write(struct file *file, loff_t pos)
390 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
393 ssize_t ret = -EBADF;
396 file = fget_light(fd, &fput_needed);
398 loff_t pos = file_pos_read(file);
399 ret = vfs_read(file, buf, count, &pos);
400 file_pos_write(file, pos);
401 fput_light(file, fput_needed);
407 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
411 ssize_t ret = -EBADF;
414 file = fget_light(fd, &fput_needed);
416 loff_t pos = file_pos_read(file);
417 ret = vfs_write(file, buf, count, &pos);
418 file_pos_write(file, pos);
419 fput_light(file, fput_needed);
425 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
426 size_t count, loff_t pos)
429 ssize_t ret = -EBADF;
435 file = fget_light(fd, &fput_needed);
438 if (file->f_mode & FMODE_PREAD)
439 ret = vfs_read(file, buf, count, &pos);
440 fput_light(file, fput_needed);
445 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
446 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
448 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
449 (size_t) count, pos);
451 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
454 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
455 size_t count, loff_t pos)
458 ssize_t ret = -EBADF;
464 file = fget_light(fd, &fput_needed);
467 if (file->f_mode & FMODE_PWRITE)
468 ret = vfs_write(file, buf, count, &pos);
469 fput_light(file, fput_needed);
474 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
475 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
477 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
478 (size_t) count, pos);
480 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
484 * Reduce an iovec's length in-place. Return the resulting number of segments
486 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
488 unsigned long seg = 0;
491 while (seg < nr_segs) {
493 if (len + iov->iov_len >= to) {
494 iov->iov_len = to - len;
502 EXPORT_SYMBOL(iov_shorten);
504 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
505 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
510 init_sync_kiocb(&kiocb, filp);
511 kiocb.ki_pos = *ppos;
513 kiocb.ki_nbytes = len;
516 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
517 if (ret != -EIOCBRETRY)
519 wait_on_retry_sync_kiocb(&kiocb);
522 if (ret == -EIOCBQUEUED)
523 ret = wait_on_sync_kiocb(&kiocb);
524 *ppos = kiocb.ki_pos;
528 /* Do it by hand, with file-ops */
529 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
530 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
532 struct iovec *vector = iov;
535 while (nr_segs > 0) {
540 base = vector->iov_base;
541 len = vector->iov_len;
545 nr = fn(filp, base, len, ppos);
560 /* A write operation does a read from user space and vice versa */
561 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
563 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
564 unsigned long nr_segs, unsigned long fast_segs,
565 struct iovec *fast_pointer,
566 struct iovec **ret_pointer)
570 struct iovec *iov = fast_pointer;
573 * SuS says "The readv() function *may* fail if the iovcnt argument
574 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
575 * traditionally returned zero for zero segments, so...
583 * First get the "struct iovec" from user memory and
584 * verify all the pointers
586 if (nr_segs > UIO_MAXIOV) {
590 if (nr_segs > fast_segs) {
591 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
597 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
603 * According to the Single Unix Specification we should return EINVAL
604 * if an element length is < 0 when cast to ssize_t or if the
605 * total length would overflow the ssize_t return value of the
609 for (seg = 0; seg < nr_segs; seg++) {
610 void __user *buf = iov[seg].iov_base;
611 ssize_t len = (ssize_t)iov[seg].iov_len;
613 /* see if we we're about to use an invalid len or if
614 * it's about to overflow ssize_t */
615 if (len < 0 || (ret + len < ret)) {
619 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
631 static ssize_t do_readv_writev(int type, struct file *file,
632 const struct iovec __user * uvector,
633 unsigned long nr_segs, loff_t *pos)
636 struct iovec iovstack[UIO_FASTIOV];
637 struct iovec *iov = iovstack;
647 ret = rw_copy_check_uvector(type, uvector, nr_segs,
648 ARRAY_SIZE(iovstack), iovstack, &iov);
653 ret = rw_verify_area(type, file, pos, tot_len);
659 fn = file->f_op->read;
660 fnv = file->f_op->aio_read;
662 fn = (io_fn_t)file->f_op->write;
663 fnv = file->f_op->aio_write;
667 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
670 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
675 if ((ret + (type == READ)) > 0) {
677 fsnotify_access(file);
679 fsnotify_modify(file);
684 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
685 unsigned long vlen, loff_t *pos)
687 if (!(file->f_mode & FMODE_READ))
689 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
692 return do_readv_writev(READ, file, vec, vlen, pos);
695 EXPORT_SYMBOL(vfs_readv);
697 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
698 unsigned long vlen, loff_t *pos)
700 if (!(file->f_mode & FMODE_WRITE))
702 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
705 return do_readv_writev(WRITE, file, vec, vlen, pos);
708 EXPORT_SYMBOL(vfs_writev);
710 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
714 ssize_t ret = -EBADF;
717 file = fget_light(fd, &fput_needed);
719 loff_t pos = file_pos_read(file);
720 ret = vfs_readv(file, vec, vlen, &pos);
721 file_pos_write(file, pos);
722 fput_light(file, fput_needed);
726 add_rchar(current, ret);
731 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
735 ssize_t ret = -EBADF;
738 file = fget_light(fd, &fput_needed);
740 loff_t pos = file_pos_read(file);
741 ret = vfs_writev(file, vec, vlen, &pos);
742 file_pos_write(file, pos);
743 fput_light(file, fput_needed);
747 add_wchar(current, ret);
752 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
754 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
755 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
758 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
759 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
761 loff_t pos = pos_from_hilo(pos_h, pos_l);
763 ssize_t ret = -EBADF;
769 file = fget_light(fd, &fput_needed);
772 if (file->f_mode & FMODE_PREAD)
773 ret = vfs_readv(file, vec, vlen, &pos);
774 fput_light(file, fput_needed);
778 add_rchar(current, ret);
783 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
784 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
786 loff_t pos = pos_from_hilo(pos_h, pos_l);
788 ssize_t ret = -EBADF;
794 file = fget_light(fd, &fput_needed);
797 if (file->f_mode & FMODE_PWRITE)
798 ret = vfs_writev(file, vec, vlen, &pos);
799 fput_light(file, fput_needed);
803 add_wchar(current, ret);
808 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
809 size_t count, loff_t max)
811 struct file * in_file, * out_file;
812 struct inode * in_inode, * out_inode;
815 int fput_needed_in, fput_needed_out, fl;
818 * Get input file, and verify that it is ok..
821 in_file = fget_light(in_fd, &fput_needed_in);
824 if (!(in_file->f_mode & FMODE_READ))
828 ppos = &in_file->f_pos;
830 if (!(in_file->f_mode & FMODE_PREAD))
832 retval = rw_verify_area(READ, in_file, ppos, count);
838 * Get output file, and verify that it is ok..
841 out_file = fget_light(out_fd, &fput_needed_out);
844 if (!(out_file->f_mode & FMODE_WRITE))
847 in_inode = in_file->f_path.dentry->d_inode;
848 out_inode = out_file->f_path.dentry->d_inode;
849 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
855 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
858 if (unlikely(pos + count > max)) {
868 * We need to debate whether we can enable this or not. The
869 * man page documents EAGAIN return for the output at least,
870 * and the application is arguably buggy if it doesn't expect
871 * EAGAIN on a non-blocking file descriptor.
873 if (in_file->f_flags & O_NONBLOCK)
874 fl = SPLICE_F_NONBLOCK;
876 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
879 add_rchar(current, retval);
880 add_wchar(current, retval);
889 fput_light(out_file, fput_needed_out);
891 fput_light(in_file, fput_needed_in);
896 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
903 if (unlikely(get_user(off, offset)))
906 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
907 if (unlikely(put_user(pos, offset)))
912 return do_sendfile(out_fd, in_fd, NULL, count, 0);
915 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
921 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
923 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
924 if (unlikely(put_user(pos, offset)))
929 return do_sendfile(out_fd, in_fd, NULL, count, 0);