Why does forking my process cause the file to be read infinitely

I am surprised that there is a problem, but it does seem to be a problem on Linux (I tested on Ubuntu 16.04 LTS running in a VMWare Fusion VM on my Mac) — but it was not a problem on my Mac running macOS 10.13.4 (High Sierra), and I wouldn't expect it to be a problem on other variants of Unix either.

As I noted in a comment:

There's an open file description and an open file descriptor behind each stream. When the process forks, the child has its own set of open file descriptors (and file streams), but each file descriptor in the child shares the open file description with the parent. IF (and that's a big 'if') the child process closing the file descriptors first did the equivalent of lseek(fd, 0, SEEK_SET), then that would also position the file descriptor for the parent process, and that could lead to an infinite loop. However, I've never heard of a library that does that seek; there's no reason to do it.

See POSIX open() and fork() for more information about open file descriptors and open file descriptions.

The open file descriptors are private to a process; the open file descriptions are shared by all copies of the file descriptor created by an initial 'open file' operation. One of the key properties of the open file description is the current seek position. That means that a child process can change the current seek position for a parent — because it is in the shared open file description.

neof97.c

I used the following code — a mildly adapted version of the original that compiles cleanly with rigorous compilation options:

#include "posixver.h"
#include <stdio.h>
#include <stdlib.h>
#include <sys/wait.h>
#include <unistd.h>

enum { MAX = 100 };

int main(void)
{
    if (freopen("input.txt", "r", stdin) == 0)
        return 1;
    char s[MAX];
    for (int i = 0; i < 30 && fgets(s, MAX, stdin) != NULL; i++)
    {
        // Commenting out this region fixes the issue
        int status;
        pid_t pid = fork();
        if (pid == 0)
        {
            exit(0);
        }
        else
        {
            waitpid(pid, &status, 0);
        }
        // End region
        printf("%s", s);
    }
    return 0;
}

One of the modifications limits the number of cycles (children) to just 30. I used a data file with 4 lines of 20 random letters plus a newline (84 bytes total):

ywYaGKiRtAwzaBbuzvNb
eRsjPoBaIdxZZtJWfSty
uGnxGhSluywhlAEBIXNP
plRXLszVvPgZhAdTLlYe

I ran the command under strace on Ubuntu:

$ strace -ff -o st-out -- neof97
ywYaGKiRtAwzaBbuzvNb
eRsjPoBaIdxZZtJWfSty
uGnxGhSluywhlAEBIXNP
plRXLszVvPgZhAdTLlYe
…
uGnxGhSluywhlAEBIXNP
plRXLszVvPgZhAdTLlYe
ywYaGKiRtAwzaBbuzvNb
eRsjPoBaIdxZZtJWfSty
$

There were 31 files with names of the form st-out.808## where the hashes were 2-digit numbers. The main process file was quite large; the others were small, with one of the sizes 66, 110, 111, or 137:

$ cat st-out.80833
lseek(0, -63, SEEK_CUR)                 = 21
exit_group(0)                           = ?
+++ exited with 0 +++
$ cat st-out.80834
lseek(0, -42, SEEK_CUR)                 = -1 EINVAL (Invalid argument)
exit_group(0)                           = ?
+++ exited with 0 +++
$ cat st-out.80835
lseek(0, -21, SEEK_CUR)                 = 0
exit_group(0)                           = ?
+++ exited with 0 +++
$ cat st-out.80836
exit_group(0)                           = ?
+++ exited with 0 +++
$

It just so happened that the first 4 children each exhibited one of the four behaviours — and each further set of 4 children exhibited the same pattern.

This shows that three out of four of the children were indeed doing an lseek() on standard input before exiting. Obviously, I have now seen a library do it. I have no idea why it is thought to be a good idea, though, but empirically, that is what is happening.

neof67.c

This version of the code, using a separate file stream (and file descriptor) and fopen() instead of freopen() also runs into the problem.

#include "posixver.h"
#include <stdio.h>
#include <stdlib.h>
#include <sys/wait.h>
#include <unistd.h>

enum { MAX = 100 };

int main(void)
{
    FILE *fp = fopen("input.txt", "r");
    if (fp == 0)
        return 1;
    char s[MAX];
    for (int i = 0; i < 30 && fgets(s, MAX, fp) != NULL; i++)
    {
        // Commenting out this region fixes the issue
        int status;
        pid_t pid = fork();
        if (pid == 0)
        {
            exit(0);
        }
        else
        {
            waitpid(pid, &status, 0);
        }
        // End region
        printf("%s", s);
    }
    return 0;
}

This also exhibits the same behaviour, except that the file descriptor on which the seek occurs is 3 instead of 0. So, two of my hypotheses are disproven — it's related to freopen() and stdin; both are shown incorrect by the second test code.

Preliminary diagnosis

IMO, this is a bug. You should not be able to run into this problem. It is most likely a bug in the Linux (GNU C) library rather than the kernel. It is caused by the lseek() in the child processes. It is not clear (because I've not gone to look at the source code) what the library is doing or why.


GLIBC Bug 23151

GLIBC Bug 23151 - A forked process with unclosed file does lseek before exit and can cause infinite loop in parent I/O.

The bug was created 2018-05-08 US/Pacific, and was closed as INVALID by 2018-05-09. The reason given was:

Please read http://pubs.opengroup.org/onlinepubs/9699919799/functions/V2_chap02.html#tag_15_05_01, especially this paragraph:

Note that after a fork(), two handles exist where one existed before. […]

POSIX

The complete section of POSIX referred to (apart from verbiage noting that this is not covered by the C standard) is this:

2.5.1 Interaction of File Descriptors and Standard I/O Streams

An open file description may be accessed through a file descriptor, which is created using functions such as open() or pipe(), or through a stream, which is created using functions such as fopen() or popen(). Either a file descriptor or a stream is called a "handle" on the open file description to which it refers; an open file description may have several handles.

Handles can be created or destroyed by explicit user action, without affecting the underlying open file description. Some of the ways to create them include fcntl(), dup(), fdopen(), fileno(), and fork(). They can be destroyed by at least fclose(), close(), and the exec functions.

A file descriptor that is never used in an operation that could affect the file offset (for example, read(), write(), or lseek()) is not considered a handle for this discussion, but could give rise to one (for example, as a consequence of fdopen(), dup(), or fork()). This exception does not include the file descriptor underlying a stream, whether created with fopen() or fdopen(), so long as it is not used directly by the application to affect the file offset. The read() and write() functions implicitly affect the file offset; lseek() explicitly affects it.

The result of function calls involving any one handle (the "active handle") is defined elsewhere in this volume of POSIX.1-2017, but if two or more handles are used, and any one of them is a stream, the application shall ensure that their actions are coordinated as described below. If this is not done, the result is undefined.

A handle which is a stream is considered to be closed when either an fclose(), or freopen() with non-full(1) filename, is executed on it (for freopen() with a null filename, it is implementation-defined whether a new handle is created or the existing one reused), or when the process owning that stream terminates with exit(), abort(), or due to a signal. A file descriptor is closed by close(), _exit(), or the exec() functions when FD_CLOEXEC is set on that file descriptor.

(1)[sic] Using 'non-full' is probably a typo for 'non-null'.

For a handle to become the active handle, the application shall ensure that the actions below are performed between the last use of the handle (the current active handle) and the first use of the second handle (the future active handle). The second handle then becomes the active handle. All activity by the application affecting the file offset on the first handle shall be suspended until it again becomes the active file handle. (If a stream function has as an underlying function one that affects the file offset, the stream function shall be considered to affect the file offset.)

The handles need not be in the same process for these rules to apply.

Note that after a fork(), two handles exist where one existed before. The application shall ensure that, if both handles can ever be accessed, they are both in a state where the other could become the active handle first. The application shall prepare for a fork() exactly as if it were a change of active handle. (If the only action performed by one of the processes is one of the exec() functions or _exit() (not exit()), the handle is never accessed in that process.)

For the first handle, the first applicable condition below applies. After the actions required below are taken, if the handle is still open, the application can close it.

  • If it is a file descriptor, no action is required.

  • If the only further action to be performed on any handle to this open file descriptor is to close it, no action need be taken.

  • If it is a stream which is unbuffered, no action need be taken.

  • If it is a stream which is line buffered, and the last byte written to the stream was a <newline> (that is, as if a putc('\n') was the most recent operation on that stream), no action need be taken.

  • If it is a stream which is open for writing or appending (but not also open for reading), the application shall either perform an fflush(), or the stream shall be closed.

  • If the stream is open for reading and it is at the end of the file (feof() is true), no action need be taken.

  • If the stream is open with a mode that allows reading and the underlying open file description refers to a device that is capable of seeking, the application shall either perform an fflush(), or the stream shall be closed.

For the second handle:

  • If any previous active handle has been used by a function that explicitly changed the file offset, except as required above for the first handle, the application shall perform an lseek() or fseek() (as appropriate to the type of handle) to an appropriate location.

If the active handle ceases to be accessible before the requirements on the first handle, above, have been met, the state of the open file description becomes undefined. This might occur during functions such as a fork() or _exit().

The exec() functions make inaccessible all streams that are open at the time they are called, independent of which streams or file descriptors may be available to the new process image.

When these rules are followed, regardless of the sequence of handles used, implementations shall ensure that an application, even one consisting of several processes, shall yield correct results: no data shall be lost or duplicated when writing, and all data shall be written in order, except as requested by seeks. It is implementation-defined whether, and under what conditions, all input is seen exactly once.

Each function that operates on a stream is said to have zero or more "underlying functions". This means that the stream function shares certain traits with the underlying functions, but does not require that there be any relation between the implementations of the stream function and its underlying functions.

Exegesis

That is hard reading! If you're not clear on the distinction between open file descriptor and open file description, read the specification of open() and fork() (and dup() or dup2()). The definitions for file descriptor and open file description are also relevant, if terse.

In the context of the code in this question (and also for Unwanted child processes being created while file reading), we have a file stream handle open for reading only which has not yet encountered EOF (so feof() would not return true, even though the read position is at the end of the file).

One of the crucial parts of the specification is: The application shall prepare for a fork() exactly as if it were a change of active handle.

This means that the steps outlined for 'first file handle' are relevant, and stepping through them, the first applicable condition is the last:

  • If the stream is open with a mode that allows reading and the underlying open file description refers to a device that is capable of seeking, the application shall either perform an fflush(), or the stream shall be closed.

If you look at the definition for fflush(), you find:

If stream points to an output stream or an update stream in which the most recent operation was not input, fflush() shall cause any unwritten data for that stream to be written to the file, [CX] ⌦ and the last data modification and last file status change timestamps of the underlying file shall be marked for update.

For a stream open for reading with an underlying file description, if the file is not already at EOF, and the file is one capable of seeking, the file offset of the underlying open file description shall be set to the file position of the stream, and any characters pushed back onto the stream by ungetc() or ungetwc() that have not subsequently been read from the stream shall be discarded (without further changing the file offset). ⌫

It isn't exactly clear what happens if you apply fflush() to an input stream associated with a non-seekable file, but that isn't our immediate concern. However, if you're writing generic library code, then you might need to know whether the underlying file descriptor is seekable before doing a fflush() on the stream. Alternatively, use fflush(NULL) to have the system do whatever is necessary for all I/O streams, noting that this will lose any pushed-back characters (via ungetc() etc).

The lseek() operations shown in the strace output seem to be implementing the fflush() semantics associating the file offset of the open file description with the file position of the stream.

So, for the code in this question, it seems that fflush(stdin) is necessary before the fork() to ensure consistency. Not doing that leads to undefined behaviour ('if this is not done, the result is undefined') — such as looping indefinitely.


The exit() call closes all open file handles. After the fork, the child and parent have identical copies of the execution stack, including the FileHandle pointer. When the child exits, it closes the file and resets the pointer.

  int main(){
        freopen("input.txt", "r", stdin);
        char s[MAX];
        prompt(s);
        int i = 0;
        char* ret = fgets(s, MAX, stdin);
        while (ret != NULL) {
            //Commenting out this region fixes the issue
            int status;
            pid_t pid = fork();   // At this point both processes has a copy of the filehandle
            if (pid == 0) {
                exit(0);          // At this point the child closes the filehandle
            } else {
                waitpid(pid, &status, 0);
            }
            //End region
            printf("%s", s);
            ret = fgets(s, MAX, stdin);
        }
    }