ceph: avoid carrying Fw cap during write into page cache
authorSage Weil <sage@newdream.net>
Tue, 26 Jul 2011 18:27:34 +0000 (11:27 -0700)
committerSage Weil <sage@newdream.net>
Tue, 26 Jul 2011 18:27:34 +0000 (11:27 -0700)
The generic_file_aio_write call may block on balance_dirty_pages while we
flush data to the OSDs.  If we hold a reference to the FILE_WR cap during
that interval revocation by the MDS (e.g., to do a stat(2)) may be very
slow.

Reviewed-by: Yehuda Sadeh <yehuda@hq.newdream.net>
Signed-off-by: Sage Weil <sage@newdream.net>
fs/ceph/file.c

index 44e4fe9fba026e9a6c0a49d68b3505a9b1122a32..6c90cf090601996653a033ed13d046ef92c9a4ac 100644 (file)
@@ -713,7 +713,7 @@ retry_snap:
                want = CEPH_CAP_FILE_BUFFER;
        ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
        if (ret < 0)
-               goto out;
+               goto out_put;
 
        dout("aio_write %p %llx.%llx %llu~%u  got cap refs on %s\n",
             inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
@@ -726,8 +726,18 @@ retry_snap:
                ret = ceph_sync_write(file, iov->iov_base, iov->iov_len,
                        &iocb->ki_pos);
        } else {
-               ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
+               /*
+                * buffered write; drop Fw early to avoid slow
+                * revocation if we get stuck on balance_dirty_pages
+                */
+               int dirty;
+
+               spin_lock(&inode->i_lock);
+               dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
+               spin_unlock(&inode->i_lock);
+               ceph_put_cap_refs(ci, got);
 
+               ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
                if ((ret >= 0 || ret == -EIOCBQUEUED) &&
                    ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host)
                     || ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
@@ -735,7 +745,12 @@ retry_snap:
                        if (err < 0)
                                ret = err;
                }
+
+               if (dirty)
+                       __mark_inode_dirty(inode, dirty);
+               goto out;
        }
+
        if (ret >= 0) {
                int dirty;
                spin_lock(&inode->i_lock);
@@ -745,12 +760,13 @@ retry_snap:
                        __mark_inode_dirty(inode, dirty);
        }
 
-out:
+out_put:
        dout("aio_write %p %llx.%llx %llu~%u  dropping cap refs on %s\n",
             inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
             ceph_cap_string(got));
        ceph_put_cap_refs(ci, got);
 
+out:
        if (ret == -EOLDSNAPC) {
                dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
                     inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len);