]> git.ipfire.org Git - thirdparty/coreutils.git/commitdiff
tail -F can fail to track a file after it's been rotated
authorJim Meyering <meyering@redhat.com>
Sun, 15 Nov 2009 08:25:29 +0000 (09:25 +0100)
committerJim Meyering <meyering@redhat.com>
Mon, 16 Nov 2009 08:30:43 +0000 (09:30 +0100)
Tailing forever and by-name (--follow=name, -F), tail would
sometimes fail to follow a file that had been removed via rename.
If you can't apply this patch and have tail 7.6 or newer, you can
work around the bug via the undocumented --disable-inotify option.
* src/tail.c (tail_forever_inotify): When tailing by name (-F),
do not un-watch a file upon receipt of the IN_MOVE_SELF event.
Reported by Arjan Opmeer in http://bugs.debian.org/548439.
* NEWS (Bug fixes): Mention it.
Also see http://marc.info/?l=coreutils-bug&m=125829031916515
* tests/Makefile.am (TESTS): Add tail-2/inotify-rotate.
* tests/tail-2/inotify-rotate: New test.

NEWS
THANKS
src/tail.c
tests/Makefile.am
tests/tail-2/inotify-rotate [new file with mode: 0755]

diff --git a/NEWS b/NEWS
index c9fe6caae436a202b68a6d99b39872360d21c41d..e3763cd19d4cf681b168b3b7e068a68acc1a98ee 100644 (file)
--- a/NEWS
+++ b/NEWS
@@ -48,6 +48,11 @@ GNU coreutils NEWS                                    -*- outline -*-
   were first renamed or unlinked or never modified.
   [The race was introduced in coreutils-7.5]
 
+  tail -F (inotify-enabled) now consistently tails a file that has been
+  replaced via renaming.  That operation provokes either of two sequences
+  of inotify events.  The less common sequence is now handled as well.
+  [The bug came with the implementation change in coreutils-7.5]
+
   timeout now doesn't exit unless the command it is monitoring does,
   for any specified signal. [bug introduced in coreutils-7.0].
 
diff --git a/THANKS b/THANKS
index 7955da76c3a3705cc017a6bb40246fb7d44c08f6..f9bc476393fe455fdaa071d8906a58d2b4d8a8de 100644 (file)
--- a/THANKS
+++ b/THANKS
@@ -52,6 +52,7 @@ Andy Longton                        alongton@metamark.com
 Anthony Thyssen                     anthony@griffith.edu.au
 Antonio Rendas                      ajrendas@yahoo.com
 Ariel Faigon                        ariel@cthulhu.engr.sgi.com
+Arjan Opmeer                        arjan.opmeer@gmail.com
 Arne H. Juul                        arnej@solan.unit.no
 Arne Henrik Juul                    arnej@imf.unit.no
 Arnold Robbins                      arnold@skeeve.com
@@ -401,7 +402,7 @@ Max Chang                           maxchang@ucla.edu
 Meelis Roos                         mroos@tartu.cyber.ee
 Michael                             michael@aplatform.com
 Michael ???                         michael@roka.net
-Michael Bacarella                   mbac@netgraft.com>
+Michael Bacarella                   mbac@netgraft.com
 Michael Deutschmann                 michael@talamasca.ocis.net
 Michael Elizabeth Chastain          mec.gnu@mindspring.com
 Michael Gaughen                     mgaughen@polyserve.com
index 09afeec9948b5a54564879bce052027869ca07bc..9a2f5ae828d9bbd6160dbd8d144bd67fbb820426 100644 (file)
@@ -1404,7 +1404,13 @@ tail_forever_inotify (int wd, struct File_spec *f, size_t n_files,
 
       if (ev->mask & (IN_ATTRIB | IN_DELETE_SELF | IN_MOVE_SELF))
         {
-          if (ev->mask & (IN_DELETE_SELF | IN_MOVE_SELF))
+          /* For IN_DELETE_SELF, we always want to remove the watch.
+             However, for IN_MOVE_SELF (the file we're watching has
+             been clobbered via a rename), when tailing by NAME, we
+             must continue to watch the file.  It's only when following
+             by file descriptor that we must remove the watch.  */
+          if ((ev->mask & IN_DELETE_SELF)
+              || ((ev->mask & IN_MOVE_SELF) && follow_mode == Follow_descriptor))
             {
               inotify_rm_watch (wd, f[i].wd);
               hash_delete (wd_table, &(f[i]));
index f74d0a10bb18f8beec4776ed374dbe91d784a189..fe8f3475443756126c67b59ee3ffbc9f4844d0f8 100644 (file)
@@ -80,6 +80,7 @@ TESTS =                                               \
   rm/ext3-perf                                 \
   rm/cycle                                     \
   cp/link-heap                                 \
+  tail-2/inotify-rotate                                \
   chmod/no-x                                   \
   chgrp/basic                                  \
   rm/dangling-symlink                          \
diff --git a/tests/tail-2/inotify-rotate b/tests/tail-2/inotify-rotate
new file mode 100755 (executable)
index 0000000..f45486a
--- /dev/null
@@ -0,0 +1,55 @@
+#!/bin/sh
+# ensure that tail -F handles rotation
+
+# Copyright (C) 2009 Free Software Foundation, Inc.
+
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+if test "$VERBOSE" = yes; then
+  set -x
+  tail --version
+fi
+
+. $srcdir/test-lib.sh
+expensive_
+
+# For details, see
+# http://lists.gnu.org/archive/html/bug-coreutils/2009-11/msg00213.html
+
+# Perform at least this many iterations, because on multi-core systems
+# the offending sequence of events can be surprisingly uncommon.
+for i in $(seq 50); do
+    echo $i
+    rm -rf k x out
+    :>k && :>x && timeout 10 tail -F k > out 2>&1 &
+    pid=$!
+    sleep .1
+    echo b > k;
+    # wait for b to appear in out
+    while :; do grep b out > /dev/null && break; done
+    mv x k
+    # wait for tail to detect the rename
+    while :; do grep tail: out > /dev/null && break; done
+    echo ok >> k
+    found=0
+    # wait up to 10 seconds for "ok" to appear in out
+    for j in $(seq 100); do
+        grep ok out > /dev/null && { found=1; break; }
+        sleep 0.1
+    done
+    kill $pid
+    test $found = 0 && { fail=1; cat out; break; }
+done
+
+Exit $fail