File name is not persisted if opened with O_SYNC and O_TRUNC flags
Subsystems (2)
Detailed description
====================
Hello, there seems to be an issue with O_SYNC flag when used together with O_TRUNC on various file systems.
Opening a file with O_SYNC (or using fsync(fd)) should persist directory entry.
However, if O_SYNC is used together with O_TRUNC the file will be missing if system crashes.
According to POSIX this is OK, but most file systems provide stronger guarantees (would be actually nice to have a more recent documentation on this behavior). This happens on Btrfs, ext4, XFS, F2FS and likely other file systems.
System info
===========
Linux version 6.19-rc7, also tested on 6.17
How to reproduce
================
```
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
int main() {
int status;
status = creat("file", S_IRWXU | S_IRWXG | S_IROTH | S_IXOTH);
printf("CREAT: %d\n", status);
close(status);
status = open("file", O_RDWR | O_TRUNC | O_SYNC);
printf("OPEN: %d\n", status);
}
// after the crash `file` is missing
```Steps:
1. Create and mount new file system in default configuration.
2. Change directory to root of the file system and run the compiled test.
3. Cause hard system crash (e.g. QEMU system_reset command).
4. Remount file system after crash.
5. Observe that file is missing.
Quoted 7 lines · Vyacheslav Kovalevsky
Detailed description
====================Hello, there seems to be an issue with
O_SYNCflag when used together withO_TRUNCon various file systems.
Opening a file withO_SYNC(or using fsync(fd)) should persist directory entry.
No, it should not. I'm not sure who hallucinated, but O_SYNC has always always applied to persistency semantics after writes and nothing else.
Quoted 3 lines · Christoph Hellwig
Quoted 1 line · Vyacheslav Kovalevsky
Detailed description ==================== Hello, there seems to be an issue with
O_SYNCflag when used together withO_TRUNCon various file systems. Opening a file withO_SYNC(or using fsync(fd)) should persist directory entry.No, it should not. I'm not sure who hallucinated, but
O_SYNChas always always applied to persistency semantics after writes and nothing else.
You are right, opening file with O_SYNC does not persist anything and ftruncate or O_TRUNC do not count as write I/O it seems. Also found an error related to these assumptions in our testing tool. Thanks.