tail: fix handling of inotify event queue overflow

* src/tail.c (tail_forever_inotify): Fallback to polling upon overlow.
* tests/tail/inotify-overflow.sh: Add a new test.
* tests/local.mk: Reference the new test.
* NEWS: Mention the bug fix.
Discovered by Anthropic's OSS Scanner, as bug ANT-2026-UUNT73BK.

NEWS

@@ -52,6 +52,12 @@ GNU coreutils NEWS                                    -*- outline -*-
   it would have failed if the control flag encoding was different.
   [This bug was present in "the beginning".]
 
+  'tail --follow' no longer ignores an overflow of the inotify event queue,
+  and now reverts to polling in that case.  Previously tail could fail
+  to notice new data, or that a file followed by name was replaced,
+  if events arrived faster than tail could process them.
+  [bug introduced in coreutils-7.5]
+
   'unexpand' no longer overflows a heap buffer on non-GLIBC platforms,
   given whitespace characters whose display width is negative.  E.g.,
   this edge case can occur with \u0085 (next line) characters on musl.

src/tail.c

@@ -1766,6 +1766,15 @@ tail_forever_inotify (int wd, struct File_spec *f, int n_files,
       ev = void_ev;
       evbuf_off += sizeof (*ev) + ev->len;
 
+      /* If the kernel discarded events for any of the watched files
+         or directories, so we can no longer depend on our view of them.
+         Revert to polling now.  */
+      if (ev->mask & IN_Q_OVERFLOW)
+        {
+          error (0, 0, _("inotify event queue overflowed"));
+          return;
+        }
+
       /* If a directory is deleted, IN_DELETE_SELF is emitted
          with ev->name of length 0.
          We need to catch it, otherwise it would wait forever,

tests/local.mk

@@ -207,6 +207,7 @@ all_tests =					\
   tests/tail/inotify-rotate-resources.sh	\
   tests/tail/inotify-dir-recreate.sh		\
   tests/tail/inotify-only-regular.sh		\
+  tests/tail/inotify-overflow.sh		\
   tests/tail/quote-headers.sh			\
   tests/chmod/no-x.sh				\
   tests/chgrp/basic.sh				\

tests/tail/inotify-overflow.sh

@@ -0,0 +1,75 @@
+#!/bin/sh
+# Ensure tail -F copes with the kernel discarding inotify events
+
+# Copyright (C) 2026 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 <https://www.gnu.org/licenses/>.
+
+. "${srcdir=.}/tests/init.sh"; path_prepend_ ./src
+print_ver_ tail
+require_inotify_supported_
+
+max_events=$(cat /proc/sys/fs/inotify/max_queued_events) &&
+test "$max_events" -le 100000 ||
+skip_ 'inotify queue limit unknown or too large'
+
+# Terminate any background tail process
+cleanup_()
+{
+  kill -CONT "$pid" 2>/dev/null
+  kill "$pid" 2>/dev/null
+  wait "$pid" 2>/dev/null
+  pid=
+}
+
+# $check_re - string to be found
+check_tail_output_ ()
+{
+  local delay="$1"
+  grep "$check_re" out > /dev/null ||
+    { sleep $delay ; return 1; }
+}
+
+grep_timeout_ ()
+{
+  check_re="$1"
+  retry_delay_ check_tail_output_ .1 7
+}
+
+# Speed up the fallback-to-polling phase.
+fastpoll='-s.1 --max-unchanged-stats=1'
+
+mkdir dir && echo x > dir/file || framework_failure_
+
+tail --debug --pid=$$ -F $fastpoll dir/file >out 2>&1 & pid=$!
+grep_timeout_ 'using notification mode' ||
+  { cat out; skip_ "inotify can't be used"; }
+
+# Fill the event queue while tail is not reading it,
+# so that the events for the following rotation are discarded.
+kill -STOP $pid || framework_failure_
+(cd dir && seq $max_events | xargs touch) || framework_failure_
+mv dir/file dir/file.1 &&
+echo y > dir/file &&
+kill -CONT $pid || framework_failure_
+
+# From coreutils-7.5 to 9.12 inclusive, the new file never noticed.
+grep_timeout_ '^y$' || { cat out; fail=1; }
+grep 'inotify event queue overflowed' out || {
+  cat out
+  test "$fail" = 1 || framework_failure_
+}
+
+
+Exit $fail