Implement RWF_SYNC/RWF_DSYNC flag support in SCALE's pwritev2 syscall handler.

Recommendation for implementation (IBM should implement):

1. Add IOCB_SYNC/IOCB_DSYNC Handling in gpfs_f_write_iter()

int gpfs_f_write_iter(struct kiocb *iocb, struct iov_iter *iter) {

ssize_t ret;

// Current code: just writes, ignores flags 

ret = generic_file_write_iter(iocb, iter);

// REQUIRED for RWF_SYNC support: 

if (ret > 0 && (iocb->ki_flags & (IOCB_SYNC | IOCB_DSYNC)))  {

int sync_type = (iocb->ki_flags & IOCB_DSYNC) ? GPFS_DATASYNC : GPFS_FULLSYNC;

// Flush buffered writes and wait for disk

gpfs_fsync_range(iocb->ki_filp, iocb->ki_pos - ret, iocb->ki_pos, sync_type);

}

return ret;

}

2. Ensure NFS Server Integration

// In GPFS NFS export server code:

// When handling NFSv3 WRITE RPC with stable = FILE_SYNC:

if (write_stable == FILE_SYNC) {

// Must set flags that gpfs_f_write_iter will see

iocb->ki_flags |= IOCB_SYNC | IOCB_DSYNC;

// Now when NFS server calls write_iter, GPFS will sync

}

3. Test Coverage (From Ticket TS022243547 writepause Test)

// Test that should pass:

int fd = open("file", O_RDWR); // NO O_SYNC!

// Write via pwritev2 with RWF_SYNC

struct iovec iov = {buffer, size};

pwritev2(fd, &iov, 1, 0, RWF_SYNC);

// Simulate sudden power loss

// (In real test: echo b > /proc/sysrq-trigger)

// After reboot:

// Data MUST be present on disk

// Test currently fails on GPFS, passes on XFS

Why:

• Standard Linux interface (Linux 4.6+, 2016)
• Required for NFS FILE_SYNC protocol compliance
• Prevents data loss in durability-critical workloads
• Matches industry-standard filesystem behavior (XFS, ext4, Btrfs, etc.)

Impact if not fixed:

• Data loss risk during failures
• NFS clients cannot rely on FILE_SYNC durability
• Breaks distributed consensus, transaction journals, and database WAL guarantees
• Creates data inconsistency across replicated systems

Test case: The writepause program from the ticket (TS022243547) demonstrates the bug and validates the fix.