File I/O: Further Details

Prev: file-io-the-universal-io-model Next: processes

Atomicity and Race Conditions

All system calls are executed atomically, so there are no conditions where a system call can interrupt another one.

Creating a file exclusively

The O_EXCL flag passed to open can be used to guarantee exclusive creation of a file.

Without this flag it’s not possible to guarantee exclusively opening a file, because it would take more than one syscall.

Appending data to a file

Another use case (common in logging) is multiple processes appending data to the same file. Again, we require one syscall.

To do so, we can use O_APPEND to open the file exclusively in append mode.

File Control Operations: fcntl

fcntl is used for control operations on an open file descriptor.

#include <fcntl.h>
int fcntl(int fd, int cmd, ...); // Return on success depends on cmd, or -1 on error.

Open File Status Flags

fcntl can be used to modify the access mode and open file status flags of a file.

To get them, pass F_GETFL to fcntl.

int flags = fcntl(fd, F_GETFL);

And check the flags for synchronized writes:

if (flags & O_SYNC)

Flags can be set using F_SETFL:

int flags;
flags = fcntl(fd, F_GETFL);
flags |= O_APPEND; // add O_APPEND to the flags
fcntl(fd, F_SETFL, flags); // set the flags

Relationship between File Descriptors and Open Files

Multiple file descriptors may refer to the same open file. They could be open in the same process or different processes.

Duplicating File Descriptors

File descriptors can be duplicated using the dup, dup2 system calls, which is what is done when piping at shell:

$ ./myscript > results.log 2>&1

dup looks like this: it returns a value that corresponds to the new file descriptor on success, which is the lowest unused file descriptor.

#include <unistd.h>
int dup(int oldfd); // Returns (new) file descriptor on success, or –1 on error

But if you want to ensure the file descriptor has a specific number, use dup2:

#include <unistd.h>
int dup2(int oldfd, int newfd); // Returns (new) file descriptor on success, or –1 on error

This closes the newfd if it exists, then returns the value again.

You can also use fcntl, which duplicates oldfd, and returning a value at least as large as startfd.

int newfd = fcntl(oldfd, F_DUPED, startfd);

Another call, dup3 also exists:

#define _GNU_SOURCE
#include <unistd.h>
int dup3(int oldfd, int newfd, int flags); // Returns (new) file descriptor on success, or –1 on error

The supported flag is O_CLOEXEC, which allows the kernel to enable the close-on-exec flag for the new file descriptor.

File I/O at a specified offset: pread and pwrite

pread and pwrite allow you to pass an offset, and a file, to read or write starting at an offset.

#include <unistd.h>
ssize_t pread(int fd, void *buf, size_t count, off_t offset); // Returns number of bytes read, 0 on EOF, or –1 on error
ssize_t pwrite(int fd, const void *buf, size_t count, off_t offset); // Returns number of bytes written, or –1 on error

This is useful for multithreaded programs, because the file offset for each open file is global to all threads, so trying to use lseek on one thread would mess up the offset for other threads.

Scatter-Gather I/O: readv and writev

#include <sys/uio.h>
ssize_t readv(int fd, const struct iovec *iov, int iovcnt); // Returns number of bytes read, 0 on EOF, or –1 on error
ssize_t writev(int fd, const struct iovec *iov, int iovcnt); // Returns number of bytes written, or –1 on error

readv and writev allow for scatter-gather I/O, which transfer multiple buffers of data to and from files in a single system call.

Each element of iov is a struct:

struct iovec {
    void *iov_base; /* Start address of buffer */
    size_t iov_len; /* Number of bytes to transfer to/from buffer */
};

Scatter Input

readv reads a contiguous sequence of bytes from the file at fd and then scatters these bytes into the buffers at iov. It fills up each buffer in order, before proceeding to the next buffer.

readv completes atomically, which is an improvement over read. readv returns the number of bytes read or 0 if it encountered EOF.

Gather Output

writev performs gather output, which concatenates data from all the buffers specified by iov and writes them as a sequence of contigious bytes to the file referred to by the file descriptor fd.

writev also completes atomically, and also can partially write, so checking its return value (same as readv is important.

Performing scatter-gather I/O at a specified offset

Linux also allows for scatter-gather I/O at a specified offset with preadv and pwritev.

#define _BSD_SOURCE
#include <sys/uio.h>
ssize_t preadv(int fd, const struct iovec *iov, int iovcnt, off_t offset); // Returns number of bytes read, 0 on EOF, or –1 on error
ssize_t pwritev(int fd, const struct iovec *iov, int iovcnt, off_t offset); // Returns number of bytes written, or –1 on error

This is useful for multithreaded applications that also want to use scatter-gather I/O.

Truncating a file: truncate and ftruncate

truncate and ftruncate set the size of a file to the value specified by length.

#include <unistd.h>
int truncate(const char *pathname, off_t length);
int ftruncate(int fd, off_t length); // Both return 0 on success, or –1 on error

If the file is longer than length, the excess data is lost. If the file is shorter than length, it is extended with null bytes or a hole.

truncate takes a pathname, and ftruncate takes a file descriptor.

Nonblocking I/O

The O_NONBLOCK flag when opening a flag can be used for nonblocking I/O, which means:

• If the file can’t be opened immediately, then open returns an error instead of blocking.
• After calling open, subsequent I/O operations are also non-blocking. If an I/O system call can’t complete immediately, either partial data transfer is performed, or the system call fails with EAGAIN or EWOULDBLOCK.

I/O on Large Files

Since off_t used to hold a file offset is a signed long, on 32-bit architectures, this would limit files to 2GB in size, which disks were much larger than. There was a large file summit (LFS) interface that allowed for the extension of SUSv2 to support larger files.

The lfs interface has the same name as normal functions but with 64 appended, like open64.

As well, lfs includes some other structs, like stat64 and off64_t.

We can also just set the _FILE_OFFSET_BITS macro to 64, which converts all system calls into their LFS equivalents automatically.

The /dev/fd directory

The virtual directory /dev/fd contains filenames /dev/fd/n, where 0..n is a number corresponding to one of the open file descriptors for the process.

Opening one of the files in /dev/fd is the same as duplicating the file descriptor.

Creating Temporary Files

Creating temporary files can be done with mkstemp or tmpfile, from glibc.

#include <stdlib.h>
int mkstemp(char *template); // Returns file descriptor on success, or –1 on error

The template argument is a pathname in which the last 6 characters must be XXXXXX, which will be replaced to create a unique filename.

mkstemp creates the file with read and writer permissions for the file owner, no permissions for other users, and opens it with O_EXCL, so the caller has exclusive access to the file.

tmpfile creates a uniquely named temporary file that is opened for reading and writing. It is similarly opened with O_EXCL.

#include <stdio.h>
FILE *tmpfile(void); // Returns file pointer on success, or NULL on error

Prev: file-io-the-universal-io-model Next: processes