How does the open system call work

优质

小牛编辑

129浏览

2023-12-01

Introduction

This is the fifth part of the chapter that describes system calls mechanism in the Linux kernel. Previous parts of this chapter described this mechanism in general. Now I will try to describe implementation of different system calls in the Linux kernel. Previous parts from this chapter and parts from other chapters of the books describe mostly deep parts of the Linux kernel that are faintly visible or fully invisible from the userspace. But the Linux kernel code is not only about itself. The vast of the Linux kernel code provides ability to our code. Due to the linux kernel our programs can read/write from/to files and don't know anything about sectors, tracks and other parts of a disk structures, we can send data over network and don't build encapsulated network packets by hand and etc.

I don't know how about you, but it is interesting to me not only how an operating system works, but how do my software interacts with it. As you may know, our programs interacts with the kernel through the special mechanism which is called system call. So, I've decided to write series of parts which will describe implementation and behavior of system calls which we are using every day like read, write, open, close, dup and etc.

I have decided to start from the description of the open system call. if you have written at least one C program, you should know that before we are able to read/write or execute other manipulations with a file we need to open it with the open function:

#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/stat.h>
#include <sys/types.h>

int main(int argc, char *argv) {
        int fd = open("test", O_RDONLY);

        if fd < 0 {
                perror("Opening of the file is failed\n");
        }
        else {
                printf("file sucessfully opened\n");
        }

        close(fd); 
        return 0;
}

In this case, the open is the function from standard library, but not system call. The standard library will call related system call for us. The open call will return a file descriptor which is just a unique number within our process which is associated with the opened file. Now as we opened a file and got file descriptor as result of open call, we may start to interact with this file. We can write into, read from it and etc. List of opened file by a process is available via proc filesystem:

$ sudo ls /proc/1/fd/

0  10  12  14  16  2   21  23  25  27  29  30  32  34  36  38  4   41  43  45  47  49  50  53  55  58  6   61  63  67  8
1  11  13  15  19  20  22  24  26  28  3   31  33  35  37  39  40  42  44  46  48  5   51  54  57  59  60  62  65  7   9

I am not going to describe more details about the open routine from the userspace view in this post, but mostly from the kernel side. if you are not very familiar with, you can get more info in the man page.

So let's start.

Definition of the open system call

If you have read the fourth part of the linux-insides book, you should know that system calls are defined with the help of SYSCALL_DEFINE macro. So, the open system call is not exception.

Definition of the open system call is located in the fs/open.c source code file and looks pretty small for the first view:

SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
{
	if (force_o_largefile())
		flags |= O_LARGEFILE;

	return do_sys_open(AT_FDCWD, filename, flags, mode);
}

As you may guess, the do_sys_open function from the same source code file does the main job. But before this function will be called, let's consider the if clause from which the implementation of the open system call starts:

if (force_o_largefile())
	flags |= O_LARGEFILE;

Here we apply the O_LARGEFILE flag to the flags which were passed to open system call in a case when the force_o_largefile() will return true. What is O_LARGEFILE? We may read this in the man page for the open(2) system call:

O_LARGEFILE
(LFS) Allow files whose sizes cannot be represented in an off_t (but can be represented in an off64_t) to be opened.

As we may read in the GNU C Library Reference Manual:

off_t
This is a signed integer type used to represent file sizes. In the GNU C Library, this type is no narrower than int. If the source is compiled with _FILE_OFFSET_BITS == 64 this type is transparently replaced by off64_t.

and

off64_t
This type is used similar to off_t. The difference is that even on 32 bit machines, where the off_t type would have 32 bits, off64_t has 64 bits and so is able to address files up to 2^63 bytes in length. When compiling with _FILE_OFFSET_BITS == 64 this type is available under the name off_t.

So it is not hard to guess that the off_t, off64_t and O_LARGEFILE are about a file size. In the case of the Linux kernel, the O_LARGEFILE is used to disallow opening large files on 32bit systems if the caller didn't specify O_LARGEFILE flag during opening of a file. On 64bit systems we force on this flag in open system call. And the force_o_largefile macro from the include/linux/fcntl.h linux kernel header file confirms this:

#ifndef force_o_largefile
#define force_o_largefile() (BITS_PER_LONG != 32)
#endif

This macro may be architecture-specific as for example for IA-64 architecture, but in our case the x86_64 does not provide definition of the force_o_largefile and it will be used from include/linux/fcntl.h.

So, as we may see the force_o_largefile is just a macro which expands to the true value in our case of x86_64 architecture. As we are considering 64-bit architecture, the force_o_largefile will be expanded to true and the O_LARGEFILE flag will be added to the set of flags which were passed to the open system call.

Now as we considered meaning of the O_LARGEFILE flag and force_o_largefile macro, we can proceed to the consideration of the implementation of the do_sys_open function. As I wrote above, this function is defined in the same source code file and looks:

long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
{
	struct open_flags op;
	int fd = build_open_flags(flags, mode, &op);
	struct filename *tmp;

	if (fd)
		return fd;

	tmp = getname(filename);
	if (IS_ERR(tmp))
		return PTR_ERR(tmp);

	fd = get_unused_fd_flags(flags);
	if (fd >= 0) {
		struct file *f = do_filp_open(dfd, tmp, &op);
		if (IS_ERR(f)) {
			put_unused_fd(fd);
			fd = PTR_ERR(f);
		} else {
			fsnotify_open(f);
			fd_install(fd, f);
		}
	}
	putname(tmp);
	return fd;
}

Let's try to understand how the do_sys_open works step by step.

open(2) flags

As you know the open system call takes set of flags as second argument that control opening a file and mode as third argument that specifies permission the permissions of a file if it is created. The do_sys_open function starts from the call of the build_open_flags function which does some checks that set of the given flags is valid and handles different conditions of flags and mode.

Let's look at the implementation of the build_open_flags. This function is defined in the same kernel file and takes three arguments:

flags - flags that control opening of a file;
mode - permissions for newly created file;

The last argument - op is represented with the open_flags structure:

struct open_flags {
        int open_flag;
        umode_t mode;
        int acc_mode;
        int intent;
        int lookup_flags;
};

which is defined in the fs/internal.h header file and as we may see it holds information about flags and access mode for internal kernel purposes. As you already may guess the main goal of the build_open_flags function is to fill an instance of this structure.

Implementation of the build_open_flags function starts from the definition of local variables and one of them is:

int acc_mode = ACC_MODE(flags);

This local variable represents access mode and its initial value will be equal to the value of expanded ACC_MODE macro. This macro is defined in the include/linux/fs.h and looks pretty interesting:

#define ACC_MODE(x) ("\004\002\006\006"[(x)&O_ACCMODE])
#define O_ACCMODE   00000003

The "\004\002\006\006" is an array of four chars:

"\004\002\006\006" == {'\004', '\002', '\006', '\006'}

So, the ACC_MODE macro just expands to the accession to this array by [(x) & O_ACCMODE] index. As we just saw, the O_ACCMODE is 00000003. By applying x & O_ACCMODE we will take the two least significant bits which are represents read, write or read/write access modes:

#define O_RDONLY        00000000
#define O_WRONLY        00000001
#define O_RDWR          00000002

After getting value from the array by the calculated index, the ACC_MODE will be expanded to access mode mask of a file which will hold MAY_WRITE, MAY_READ and other information.

We may see following condition after we have calculated initial access mode:

if (flags & (O_CREAT | __O_TMPFILE))
	op->mode = (mode & S_IALLUGO) | S_IFREG;
else
	op->mode = 0;

Here we reset permissions in open_flags instance if a opened file wasn't temporary and wasn't open for creation. This is because:

if neither O_CREAT nor O_TMPFILE is specified, then mode is ignored.

In other case if O_CREAT or O_TMPFILE were passed we canonicalize it to a regular file because a directory should be created with the opendir system call.

At the next step we check that a file is not tried to be opened via fanotify and without the O_CLOEXEC flag:

flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;

We do this to not leak a file descriptor. By default, the new file descriptor is set to remain open across an execve system call, but the open system call supports O_CLOEXEC flag that can be used to change this default behaviour. So we do this to prevent leaking of a file descriptor when one thread opens a file to set O_CLOEXEC flag and in the same time the second process does a fork + execve and as you may remember that child will have copies of the parent's set of open file descriptors.

At the next step we check that if our flags contains O_SYNC flag, we apply O_DSYNC flag too:

if (flags & __O_SYNC)
	flags |= O_DSYNC;

The O_SYNC flag guarantees that the any write call will not return before all data has been transferred to the disk. The O_DSYNC is like O_SYNC except that there is no requirement to wait for any metadata (like atime, mtime and etc.) changes will be written. We apply O_DSYNC in a case of __O_SYNC because it is implemented as __O_SYNC|O_DSYNC in the Linux kernel.

After this we must be sure that if a user wants to create temporary file, the flags should contain O_TMPFILE_MASK or in other words it should contain or O_CREAT or O_TMPFILE or both and also it should be writeable:

if (flags & __O_TMPFILE) {
	if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
		return -EINVAL;
	if (!(acc_mode & MAY_WRITE))
		return -EINVAL;
} else if (flags & O_PATH) {
       	flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
        acc_mode = 0;
}

as it is written in in the manual page:

O_TMPFILE must be specified with one of O_RDWR or O_WRONLY

If we didn't pass O_TMPFILE for creation of a temporary file, we check the O_PATH flag at the next condition. The O_PATH flag allows us to obtain a file descriptor that may be used for two following purposes:

to indicate a location in the filesystem tree;
to perform operations that act purely at the file descriptor level.

So, in this case the file itself is not opened, but operations like dup, fcntl and other can be used. So, if all file content related operations like read, write and other are permitted, only O_DIRECTORY | O_NOFOLLOW | O_PATH flags can be used. We have finished with flags for this moment in the build_open_flags for this moment and we may fill our open_flags->open_flag with them:

op->open_flag = flags;

Now we have filled open_flag field which represents flags that will control opening of a file and mode that will represent umask of a new file if we open file for creation. There are still to fill last flags in the our open_flags structure. The next is op->acc_mode which represents access mode to a opened file. We already filled the acc_mode local variable with the initial value at the beginning of the build_open_flags and now we check last two flags related to access mode:

if (flags & O_TRUNC)
        acc_mode |= MAY_WRITE;
if (flags & O_APPEND)
	acc_mode |= MAY_APPEND;
op->acc_mode = acc_mode;

These flags are - O_TRUNC that will truncate an opened file to length 0 if it existed before we open it and the O_APPEND flag allows to open a file in append mode. So the opened file will be appended during write but not overwritten.

The next field of the open_flags structure is - intent. It allows us to know about our intention or in other words what do we really want to do with file, open it, create, rename it or something else. So we set it to zero if our flags contains the O_PATH flag as we can't do anything related to a file content with this flag:

op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;

or just to LOOKUP_OPEN intention. Additionally we set LOOKUP_CREATE intention if we want to create new file and to be sure that a file didn't exist before with O_EXCL flag:

if (flags & O_CREAT) {
	op->intent |= LOOKUP_CREATE;
	if (flags & O_EXCL)
		op->intent |= LOOKUP_EXCL;
}

The last flag of the open_flags structure is the lookup_flags:

if (flags & O_DIRECTORY)
	lookup_flags |= LOOKUP_DIRECTORY;
if (!(flags & O_NOFOLLOW))
	lookup_flags |= LOOKUP_FOLLOW;
op->lookup_flags = lookup_flags;

return 0;

We fill it with LOOKUP_DIRECTORY if we want to open a directory and LOOKUP_FOLLOW if we don't want to follow (open) symlink. That's all. It is the end of the build_open_flags function. The open_flags structure is filled with modes and flags for a file opening and we can return back to the do_sys_open.

Actual opening of a file

At the next step after build_open_flags function is finished and we have formed flags and modes for our file we should get the filename structure with the help of the getname function by name of a file which was passed to the open system call:

tmp = getname(filename);
if (IS_ERR(tmp))
	return PTR_ERR(tmp);

The getname function is defined in the fs/namei.c source code file and looks:

struct filename *
getname(const char __user * filename)
{
        return getname_flags(filename, 0, NULL);
}

So, it just calls the getname_flags function and returns its result. The main goal of the getname_flags function is to copy a file path given from userland to kernel space. The filename structure is defined in the include/linux/fs.h linux kernel header file and contains following fields:

name - pointer to a file path in kernel space;
uptr - original pointer from userland;
aname - filename from audit context;
refcnt - reference counter;
iname - a filename in a case when it will be less than PATH_MAX.

As I already wrote above, the main goal of the getname_flags function is to copy name of a file which was passed to the open system call from user space to kernel space with the strncpy_from_user function. The next step after a filename will be copied to kernel space is getting of new non-busy file descriptor:

fd = get_unused_fd_flags(flags);

The get_unused_fd_flags function takes table of open files of the current process, minimum (0) and maximum (RLIMIT_NOFILE) possible number of a file descriptor in the system and flags that we have passed to the open system call and allocates file descriptor and mark it busy in the file descriptor table of the current process. The get_unused_fd_flags function sets or clears the O_CLOEXEC flag depends on its state in the passed flags.

The last and main step in the do_sys_open is the do_filp_open function:

struct file *f = do_filp_open(dfd, tmp, &op);

if (IS_ERR(f)) {
	put_unused_fd(fd);
	fd = PTR_ERR(f);
} else {
	fsnotify_open(f);
	fd_install(fd, f);
}

The main goal of this function is to resolve given path name into file structure which represents an opened file of a process. If something going wrong and execution of the do_filp_open function will be failed, we should free new file descriptor with the put_unused_fd or in other way the file structure returned by the do_filp_open will be stored in the file descriptor table of the current process.

Now let's take a short look at the implementation of the do_filp_open function. This function is defined in the fs/namei.c linux kernel source code file and starts from initialization of the nameidata structure. This structure will provide a link to a file inode. Actually this is one of the main point of the do_filp_open function to acquire an inode by the filename given to open system call. After the nameidata structure will be initialized, the path_openat function will be called:

filp = path_openat(&nd, op, flags | LOOKUP_RCU);

if (unlikely(filp == ERR_PTR(-ECHILD)))
	filp = path_openat(&nd, op, flags);
if (unlikely(filp == ERR_PTR(-ESTALE)))
	filp = path_openat(&nd, op, flags | LOOKUP_REVAL);

Note that it is called three times. Actually, the Linux kernel will open the file in RCU mode. This is the most efficient way to open a file. If this try will be failed, the kernel enters the normal mode. The third call is relatively rare, only in the nfs file system is likely to be used. The path_openat function executes path lookup or in other words it tries to find a dentry (what the Linux kernel uses to keep track of the hierarchy of files in directories) corresponding to a path.

The path_openat function starts from the call of the get_empty_flip() function that allocates a new file structure with some additional checks like do we exceed amount of opened files in the system or not and etc. After we have got allocated new file structure we call the do_tmpfile or do_o_path functions in a case if we have passed O_TMPFILE | O_CREATE or O_PATH flags during call of the open system call. These both cases are quite specific, so let's consider quite usual case when we want to open already existed file and want to read/write from/to it.

In this case the path_init function will be called. This function performs some preporatory work before actual path lookup. This includes search of start position of path traversal and its metadata like inode of the path, dentry inode and etc. This can be root directory - / or current directory as in our case, because we use AT_CWD as starting point (see call of the do_sys_open at the beginning of the post).

The next step after the path_init is the loop which executes the link_path_walk and do_last. The first function executes name resolution or in other words this function starts process of walking along a given path. It handles everything step by step except the last component of a file path. This handling includes checking of a permissions and getting a file component. As a file component is gotten, it is passed to walk_component that updates current directory entry from the dcache or asks underlying filesystem. This repeats before all path's components will not be handled in such way. After the link_path_walk will be executed, the do_last function will populate a file structure based on the result of the link_path_walk. As we reached last component of the given file path the vfs_open function from the do_last will be called.

This function is defined in the fs/open.c linux kernel source code file and the main goal of this function is to call an open operation of underlying filesystem.

That's all for now. We didn't consider full implementation of the open system call. We skip some parts like handling case when we want to open a file from other filesystem with different mount point, resolving symlinks and etc., but it should be not so hard to follow this stuff. This stuff does not included in generic implementation of open system call and depends on underlying filesystem. If you are interested in, you may lookup the file_operations.open callback function for a certain filesystem.

Conclusion

This is the end of the fifth part of the implementation of different system calls in the Linux kernel. If you have questions or suggestions, ping me on twitter 0xAX, drop me an email, or just create an issue. In the next part, we will continue to dive into system calls in the Linux kernel and see the implementation of the read system call.

Please note that English is not my first language and I am really sorry for any inconvenience. If you find any mistakes please send me PR to linux-insides.