[LU-15441] WBC: Write hang with O_DSYNC file flag

• Status: Open
• Project: Lustre
• Type: Technical task
• Priority: Minor
• Reporter: Qian Yingjin
• Assignee: Qian Yingjin
• Resolution: Unresolved

Description

When benchmark with filebench oltp.f workload, the writer hanged.

After investigated, we found the following dead lock in WBC code:

 static ssize_t memfs_file_write_iter(struct kiocb *iocb, struct iov_iter *iter)
{
    down_read(&wbci->wbci_rw_sem);
 repeat:
    if (wbc_inode_data_caching(wbci)) {
        rc = generic_file_write_iter(iocb, iter);
        if (rc == -ENOSPC)    
             GOTO(repeat, rc);
    } else {
        rc = ll_i2sbi(inode)->ll_fop->write_iter(iocb, iter);
    }
    up_read(&wbci->wbci_rw_sem);
    ....
}

generic_file_write_iter() ->generic_write_sync():
static inline ssize_t generic_write_sync(struct kiocb *iocb, ssize_t count)
{
	if (iocb->ki_flags & IOCB_DSYNC) {
		int ret = vfs_fsync_range(iocb->ki_filp,
				iocb->ki_pos - count, iocb->ki_pos - 1,
				(iocb->ki_flags & IOCB_SYNC) ? 0 : 1);
          ...

static int memfs_fsync(struct file *file, loff_t start,
                       loff_t end, int datasync)
{
    if (S_ISREG(inode->i_mode)) {
               down_write(&wbci->wbci_rw_sem); //**
               rc = wbcfs_commit_cache_pages(inode);
               up_write(&wbci->wbci_rw_sem); //**
  }
  ...
}

Comments

[Gerrit Updater]

"Yingjin Qian <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/46059
Subject: LU-15441 wbc: write hang with O_DSYNC flag
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 8e02e26c2ffca05bb9907d1e76dfd768b8290c04