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(struct file *file, loff_t offset, int origin)
38 struct inode *inode = file->f_mapping->host;
42 offset += inode->i_size;
45 offset += file->f_pos;
48 if (offset>=0 && offset<=inode->i_sb->s_maxbytes) {
49 /* Special lock needed here? */
50 if (offset != file->f_pos) {
58 EXPORT_SYMBOL(generic_file_llseek_unlocked);
60 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
63 mutex_lock(&file->f_dentry->d_inode->i_mutex);
64 n = generic_file_llseek_unlocked(file, offset, origin);
65 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
68 EXPORT_SYMBOL(generic_file_llseek);
70 loff_t no_llseek(struct file *file, loff_t offset, int origin)
74 EXPORT_SYMBOL(no_llseek);
76 loff_t default_llseek(struct file *file, loff_t offset, int origin)
83 offset += i_size_read(file->f_path.dentry->d_inode);
86 offset += file->f_pos;
90 if (offset != file->f_pos) {
99 EXPORT_SYMBOL(default_llseek);
101 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
103 loff_t (*fn)(struct file *, loff_t, int);
106 if (file->f_mode & FMODE_LSEEK) {
108 if (file->f_op && file->f_op->llseek)
109 fn = file->f_op->llseek;
111 return fn(file, offset, origin);
113 EXPORT_SYMBOL(vfs_llseek);
115 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, origin)
122 file = fget_light(fd, &fput_needed);
127 if (origin <= SEEK_MAX) {
128 loff_t res = vfs_llseek(file, offset, origin);
130 if (res != (loff_t)retval)
131 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
133 fput_light(file, fput_needed);
138 #ifdef __ARCH_WANT_SYS_LLSEEK
139 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
140 unsigned long, offset_low, loff_t __user *, result,
141 unsigned int, origin)
149 file = fget_light(fd, &fput_needed);
154 if (origin > SEEK_MAX)
157 offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
160 retval = (int)offset;
163 if (!copy_to_user(result, &offset, sizeof(offset)))
167 fput_light(file, fput_needed);
174 * rw_verify_area doesn't like huge counts. We limit
175 * them to something that fits in "int" so that others
176 * won't have to do range checks all the time.
178 #define MAX_RW_COUNT (INT_MAX & PAGE_CACHE_MASK)
180 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
184 int retval = -EINVAL;
186 inode = file->f_path.dentry->d_inode;
187 if (unlikely((ssize_t) count < 0))
190 if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
193 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
194 retval = locks_mandatory_area(
195 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
196 inode, file, pos, count);
200 retval = security_file_permission(file,
201 read_write == READ ? MAY_READ : MAY_WRITE);
204 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
207 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
209 set_current_state(TASK_UNINTERRUPTIBLE);
210 if (!kiocbIsKicked(iocb))
213 kiocbClearKicked(iocb);
214 __set_current_state(TASK_RUNNING);
217 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
219 struct iovec iov = { .iov_base = buf, .iov_len = len };
223 init_sync_kiocb(&kiocb, filp);
224 kiocb.ki_pos = *ppos;
228 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
229 if (ret != -EIOCBRETRY)
231 wait_on_retry_sync_kiocb(&kiocb);
234 if (-EIOCBQUEUED == ret)
235 ret = wait_on_sync_kiocb(&kiocb);
236 *ppos = kiocb.ki_pos;
240 EXPORT_SYMBOL(do_sync_read);
242 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
246 if (!(file->f_mode & FMODE_READ))
248 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
250 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
253 ret = rw_verify_area(READ, file, pos, count);
256 if (file->f_op->read)
257 ret = file->f_op->read(file, buf, count, pos);
259 ret = do_sync_read(file, buf, count, pos);
261 fsnotify_access(file->f_path.dentry);
262 add_rchar(current, ret);
270 EXPORT_SYMBOL(vfs_read);
272 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
274 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
278 init_sync_kiocb(&kiocb, filp);
279 kiocb.ki_pos = *ppos;
283 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
284 if (ret != -EIOCBRETRY)
286 wait_on_retry_sync_kiocb(&kiocb);
289 if (-EIOCBQUEUED == ret)
290 ret = wait_on_sync_kiocb(&kiocb);
291 *ppos = kiocb.ki_pos;
295 EXPORT_SYMBOL(do_sync_write);
297 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
301 if (!(file->f_mode & FMODE_WRITE))
303 if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
305 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
308 ret = rw_verify_area(WRITE, file, pos, count);
311 if (file->f_op->write)
312 ret = file->f_op->write(file, buf, count, pos);
314 ret = do_sync_write(file, buf, count, pos);
316 fsnotify_modify(file->f_path.dentry);
317 add_wchar(current, ret);
325 EXPORT_SYMBOL(vfs_write);
327 static inline loff_t file_pos_read(struct file *file)
332 static inline void file_pos_write(struct file *file, loff_t pos)
337 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
340 ssize_t ret = -EBADF;
343 file = fget_light(fd, &fput_needed);
345 loff_t pos = file_pos_read(file);
346 ret = vfs_read(file, buf, count, &pos);
347 file_pos_write(file, pos);
348 fput_light(file, fput_needed);
354 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
358 ssize_t ret = -EBADF;
361 file = fget_light(fd, &fput_needed);
363 loff_t pos = file_pos_read(file);
364 ret = vfs_write(file, buf, count, &pos);
365 file_pos_write(file, pos);
366 fput_light(file, fput_needed);
372 SYSCALL_DEFINE(pread64)(unsigned int fd, char __user *buf,
373 size_t count, loff_t pos)
376 ssize_t ret = -EBADF;
382 file = fget_light(fd, &fput_needed);
385 if (file->f_mode & FMODE_PREAD)
386 ret = vfs_read(file, buf, count, &pos);
387 fput_light(file, fput_needed);
392 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
393 asmlinkage long SyS_pread64(long fd, long buf, long count, loff_t pos)
395 return SYSC_pread64((unsigned int) fd, (char __user *) buf,
396 (size_t) count, pos);
398 SYSCALL_ALIAS(sys_pread64, SyS_pread64);
401 SYSCALL_DEFINE(pwrite64)(unsigned int fd, const char __user *buf,
402 size_t count, loff_t pos)
405 ssize_t ret = -EBADF;
411 file = fget_light(fd, &fput_needed);
414 if (file->f_mode & FMODE_PWRITE)
415 ret = vfs_write(file, buf, count, &pos);
416 fput_light(file, fput_needed);
421 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
422 asmlinkage long SyS_pwrite64(long fd, long buf, long count, loff_t pos)
424 return SYSC_pwrite64((unsigned int) fd, (const char __user *) buf,
425 (size_t) count, pos);
427 SYSCALL_ALIAS(sys_pwrite64, SyS_pwrite64);
431 * Reduce an iovec's length in-place. Return the resulting number of segments
433 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
435 unsigned long seg = 0;
438 while (seg < nr_segs) {
440 if (len + iov->iov_len >= to) {
441 iov->iov_len = to - len;
449 EXPORT_SYMBOL(iov_shorten);
451 ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
452 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
457 init_sync_kiocb(&kiocb, filp);
458 kiocb.ki_pos = *ppos;
460 kiocb.ki_nbytes = len;
463 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
464 if (ret != -EIOCBRETRY)
466 wait_on_retry_sync_kiocb(&kiocb);
469 if (ret == -EIOCBQUEUED)
470 ret = wait_on_sync_kiocb(&kiocb);
471 *ppos = kiocb.ki_pos;
475 /* Do it by hand, with file-ops */
476 ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
477 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
479 struct iovec *vector = iov;
482 while (nr_segs > 0) {
487 base = vector->iov_base;
488 len = vector->iov_len;
492 nr = fn(filp, base, len, ppos);
507 /* A write operation does a read from user space and vice versa */
508 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
510 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
511 unsigned long nr_segs, unsigned long fast_segs,
512 struct iovec *fast_pointer,
513 struct iovec **ret_pointer)
517 struct iovec *iov = fast_pointer;
520 * SuS says "The readv() function *may* fail if the iovcnt argument
521 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
522 * traditionally returned zero for zero segments, so...
530 * First get the "struct iovec" from user memory and
531 * verify all the pointers
533 if (nr_segs > UIO_MAXIOV) {
537 if (nr_segs > fast_segs) {
538 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
544 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
550 * According to the Single Unix Specification we should return EINVAL
551 * if an element length is < 0 when cast to ssize_t or if the
552 * total length would overflow the ssize_t return value of the
556 for (seg = 0; seg < nr_segs; seg++) {
557 void __user *buf = iov[seg].iov_base;
558 ssize_t len = (ssize_t)iov[seg].iov_len;
560 /* see if we we're about to use an invalid len or if
561 * it's about to overflow ssize_t */
562 if (len < 0 || (ret + len < ret)) {
566 if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
578 static ssize_t do_readv_writev(int type, struct file *file,
579 const struct iovec __user * uvector,
580 unsigned long nr_segs, loff_t *pos)
583 struct iovec iovstack[UIO_FASTIOV];
584 struct iovec *iov = iovstack;
594 ret = rw_copy_check_uvector(type, uvector, nr_segs,
595 ARRAY_SIZE(iovstack), iovstack, &iov);
600 ret = rw_verify_area(type, file, pos, tot_len);
606 fn = file->f_op->read;
607 fnv = file->f_op->aio_read;
609 fn = (io_fn_t)file->f_op->write;
610 fnv = file->f_op->aio_write;
614 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
617 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
622 if ((ret + (type == READ)) > 0) {
624 fsnotify_access(file->f_path.dentry);
626 fsnotify_modify(file->f_path.dentry);
631 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
632 unsigned long vlen, loff_t *pos)
634 if (!(file->f_mode & FMODE_READ))
636 if (!file->f_op || (!file->f_op->aio_read && !file->f_op->read))
639 return do_readv_writev(READ, file, vec, vlen, pos);
642 EXPORT_SYMBOL(vfs_readv);
644 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
645 unsigned long vlen, loff_t *pos)
647 if (!(file->f_mode & FMODE_WRITE))
649 if (!file->f_op || (!file->f_op->aio_write && !file->f_op->write))
652 return do_readv_writev(WRITE, file, vec, vlen, pos);
655 EXPORT_SYMBOL(vfs_writev);
657 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
661 ssize_t ret = -EBADF;
664 file = fget_light(fd, &fput_needed);
666 loff_t pos = file_pos_read(file);
667 ret = vfs_readv(file, vec, vlen, &pos);
668 file_pos_write(file, pos);
669 fput_light(file, fput_needed);
673 add_rchar(current, ret);
678 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
682 ssize_t ret = -EBADF;
685 file = fget_light(fd, &fput_needed);
687 loff_t pos = file_pos_read(file);
688 ret = vfs_writev(file, vec, vlen, &pos);
689 file_pos_write(file, pos);
690 fput_light(file, fput_needed);
694 add_wchar(current, ret);
699 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
700 size_t count, loff_t max)
702 struct file * in_file, * out_file;
703 struct inode * in_inode, * out_inode;
706 int fput_needed_in, fput_needed_out, fl;
709 * Get input file, and verify that it is ok..
712 in_file = fget_light(in_fd, &fput_needed_in);
715 if (!(in_file->f_mode & FMODE_READ))
718 in_inode = in_file->f_path.dentry->d_inode;
721 if (!in_file->f_op || !in_file->f_op->splice_read)
725 ppos = &in_file->f_pos;
727 if (!(in_file->f_mode & FMODE_PREAD))
729 retval = rw_verify_area(READ, in_file, ppos, count);
735 * Get output file, and verify that it is ok..
738 out_file = fget_light(out_fd, &fput_needed_out);
741 if (!(out_file->f_mode & FMODE_WRITE))
744 if (!out_file->f_op || !out_file->f_op->sendpage)
746 out_inode = out_file->f_path.dentry->d_inode;
747 retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
753 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
757 if (unlikely(pos < 0))
759 if (unlikely(pos + count > max)) {
769 * We need to debate whether we can enable this or not. The
770 * man page documents EAGAIN return for the output at least,
771 * and the application is arguably buggy if it doesn't expect
772 * EAGAIN on a non-blocking file descriptor.
774 if (in_file->f_flags & O_NONBLOCK)
775 fl = SPLICE_F_NONBLOCK;
777 retval = do_splice_direct(in_file, ppos, out_file, count, fl);
780 add_rchar(current, retval);
781 add_wchar(current, retval);
790 fput_light(out_file, fput_needed_out);
792 fput_light(in_file, fput_needed_in);
797 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
804 if (unlikely(get_user(off, offset)))
807 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
808 if (unlikely(put_user(pos, offset)))
813 return do_sendfile(out_fd, in_fd, NULL, count, 0);
816 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
822 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
824 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
825 if (unlikely(put_user(pos, offset)))
830 return do_sendfile(out_fd, in_fd, NULL, count, 0);