LU-14640 osd: ASSERTION(!PageDirty(lnb[i].lnb_page)

fallocate(PUNCH_HOLE) was leaving the partially-zeroed
page in the buffer cache. This was causing ASSERT when
doing large direct read/write operations. This was see
when executing a fsx run with options:-

$ fsx -c 50 -p 1000 -S 7919 -P /tmp -l 5407677 -N 100000 <file>

Lustre: DEBUG MARKER: GENERIC DEBUG start start
LustreError: 15768:0:(osd_io.c:1563:osd_write_commit())
	ASSERTION( !PageDirty(lnb[i].lnb_page) ) failed:
LustreError: 15768:0:(osd_io.c:1563:osd_write_commit()) LBUG
Pid: 15768, comm: ll_ost_io00_000 3.10.0-957.el7_lustre.x86_64
Call Trace:
[<0>] libcfs_call_trace+0x90/0xf0 [libcfs]
[<0>] lbug_with_loc+0x4c/0xa0 [libcfs]
[<0>] osd_write_commit+0x52c/0x870 [osd_ldiskfs]
[<0>] ofd_commitrw_write+0xe79/0x1510 [ofd]
[<0>] ofd_commitrw+0x2ad/0x9a0 [ofd]
[<0>] tgt_brw_write+0xfd0/0x1cb0 [ptlrpc]
[<0>] tgt_request_handle+0x7ea/0x1750 [ptlrpc]
[<0>] ptlrpc_server_handle_request+0x256/0xb10 [ptlrpc]
[<0>] ptlrpc_main+0xb3c/0x14e0 [ptlrpc]
[<0>] kthread+0xd1/0xe0
[<0>] ret_from_fork_nospec_begin+0xe/0x21
[<0>] 0xfffffffffffffffe
Kernel panic - not syncing: LBUG

Test-case: sanity-benchmark/fsx_partial_punch added

Test-Parameters: testlist=sanity-benchmark
Signed-off-by: Arshad Hussain <arshad.hussain@aeoncomputing.com>
Change-Id: I89fcbc6af0cbf4b544b8d149703053909ecb6cad
Reviewed-on: https://review.whamcloud.com/43462
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Mike Pershin <mpershin@whamcloud.com>
Reviewed-by: Alex Zhuravlev <bzzz@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/osd-ldiskfs/osd_io.c b/lustre/osd-ldiskfs/osd_io.c
index d72896e..40ae417 100644 (file)
--- a/lustre/osd-ldiskfs/osd_io.c
+++ b/lustre/osd-ldiskfs/osd_io.c
@@ -2760,6 +2760,22 @@ void osd_trunc_unlock_all(const struct lu_env *env, struct list_head *list)
        }
 }
 
        }
 }
 

+/* For a partial-page punch, flush punch range to disk immediately */
+static void osd_partial_page_flush_punch(struct osd_device *d,
+                                        struct inode *inode, loff_t start,
+                                        loff_t end)
+{
+       if (osd_use_page_cache(d)) {
+               filemap_fdatawrite_range(inode->i_mapping, start, end);
+       } else {
+               /* Notice we use "wait" version to ensure I/O is complete */
+               filemap_write_and_wait_range(inode->i_mapping, start,
+                                            end);
+               invalidate_mapping_pages(inode->i_mapping, start >> PAGE_SHIFT,
+                                        end >> PAGE_SHIFT);
+       }
+}
+

 /*
  * For a partial-page truncate, flush the page to disk immediately to
  * avoid data corruption during direct disk write.  b=17397
 /*
  * For a partial-page truncate, flush the page to disk immediately to
  * avoid data corruption during direct disk write.  b=17397


@@ -2827,8 +2843,7 @@ void osd_execute_punch(const struct lu_env *env, struct osd_object *obj,
        struct file *file = osd_quasi_file(env, inode);
 
        file->f_op->fallocate(file, mode, start, end - start);
        struct file *file = osd_quasi_file(env, inode);
 
        file->f_op->fallocate(file, mode, start, end - start);

-       osd_partial_page_flush(d, inode, start);
-       osd_partial_page_flush(d, inode, end - 1);
+       osd_partial_page_flush_punch(d, inode, start, end - 1);

 }
 
 void osd_process_truncates(const struct lu_env *env, struct list_head *list)
 }
 
 void osd_process_truncates(const struct lu_env *env, struct list_head *list)

diff --git a/lustre/tests/sanity-benchmark.sh b/lustre/tests/sanity-benchmark.sh
index b01dc45..35a58fc 100644 (file)
--- a/lustre/tests/sanity-benchmark.sh
+++ b/lustre/tests/sanity-benchmark.sh
@@ -199,6 +199,31 @@ test_fsx() {
 }
 run_test fsx "fsx"
 
 }
 run_test fsx "fsx"
 

+test_fsx_partial_punch() {
+       local fsx_count=100000
+       local testfile=$DIR/f0.fsxfile
+       local fsx_size=5407677 # upper bound file size
+       local fsx_seed=7919
+
+       check_set_fallocate
+
+       rm -f $testfile
+       $LFS setstripe -c -1 $testfile
+
+       #
+       # $fsx_seed, $fsx_count and $fsx_size combination almost
+       # always reproduces the LASSERT under LU-14640. Therefore these
+       # constants are used as reproducer vs using a random value and
+       # hoping it hits the error condition
+       #
+       CMD="$FSX -c 50 -p 1000 -S $fsx_seed -P $TMP -l $fsx_size \
+            -N $fsx_count $testfile"
+       echo "Using: $CMD"
+       $CMD || error "fsx failed"
+       rm -f $testfile
+}
+run_test fsx_partial_punch "Verify fsx with partial punch via fallocate"
+

 complete $SECONDS
 check_and_cleanup_lustre
 exit_status
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status