lsns: continue the executing even if opening a /proc/$pid fails

In the original code, lsns printed nothing if it failed in opening the
last dntry in /proc/[0-9]* though lsns should work partially.

The original behavior caused the combination of the following two
test cases failed:

    $ tests/ts/lsns/filter & tests/ts/lsns/ioctl_ns &
    [1] 19178
    [2] 19179
    $          lsns: ownership and hierarchy        ...         \
    lsns: -Q, --filter option            ... FAILED
     FAILED

    [1]-  Done                    tests/ts/lsns/filter
    [2]+  Done                    tests/ts/lsns/ioctl_ns

Signed-off-by: Masatake YAMATO <yamato@redhat.com>

1 files changed, 11 insertions, 1 deletions

diff --git a/sys-utils/lsns.c b/sys-utils/lsns.c
index bc966dadb5..0c66d648af 100644
--- a/sys-utils/lsns.c
+++ b/sys-utils/lsns.c
@@ -602,7 +602,17 @@ static int read_processes(struct lsns *ls)
 		DBG(PROC, ul_debug("reading %d", (int) pid));
 		rc = procfs_process_init_path(pc, pid);
 		if (rc < 0) {
-			DBG(PROC, ul_debug("failed in reading /proc/%d", (int) pid));
+			DBG(PROC, ul_debug("failed in reading /proc/%d (rc: %d)", (int) pid, rc));
+			/* This failure is acceptable. If a process ($pid) owning
+			 * a namespace is gone while running this lsns process,
+			 * procfs_process_init_path(pc, $pid) may fail.
+			 *
+			 * We must reset this `rc' here. If this `d' is the last
+			 * dentry in `dir', this read_processes() invocation
+			 * returns this `rc'. In the caller context, the
+			 * non-zero value returned from read_processes() makes
+			 * lsns prints nothing. We should avoid the behavior. */
+			rc = 0;
 			continue;
 		}