代码拉取完成,页面将自动刷新
同步操作将从 src-anolis-os/systemd 强制同步,此操作会覆盖自 Fork 仓库以来所做的任何修改,且无法恢复!!!
确定后同步将在后台操作,完成时将刷新页面,请耐心等待。
From 4436c767d45c6568cbdb7aa73e5efba6d0f31f1f Mon Sep 17 00:00:00 2001
From: Lennart Poettering <lennart@poettering.net>
Date: Tue, 9 Nov 2021 00:11:38 +0100
Subject: [PATCH] sd-event: don't destroy inotify data structures from inotify
event handler
This fixes a bad memory access when we destroy an inotify source handler
from the handler itself, and thus destroy the associated inotify_data
structures.
Fixes: #20177
(cherry picked from commit 53baf2efa420cab6c4b1904c9a0c46a0c4ec80a1)
Resolves: #2211358
---
src/libsystemd/sd-event/sd-event.c | 45 +++++++++++++++++++++++++++---
1 file changed, 41 insertions(+), 4 deletions(-)
diff --git a/src/libsystemd/sd-event/sd-event.c b/src/libsystemd/sd-event/sd-event.c
index 5e22190366..01a97a4801 100644
--- a/src/libsystemd/sd-event/sd-event.c
+++ b/src/libsystemd/sd-event/sd-event.c
@@ -263,6 +263,11 @@ struct inotify_data {
* the events locally if they can't be coalesced). */
unsigned n_pending;
+ /* If this counter is non-zero, don't GC the inotify data object even if not used to watch any inode
+ * anymore. This is useful to pin the object for a bit longer, after the last event source needing it
+ * is gone. */
+ unsigned n_busy;
+
/* A linked list of all inotify objects with data already read, that still need processing. We keep this list
* to make it efficient to figure out what inotify objects to process data on next. */
LIST_FIELDS(struct inotify_data, buffered);
@@ -1845,6 +1850,29 @@ static void event_free_inode_data(
free(d);
}
+static void event_gc_inotify_data(
+ sd_event *e,
+ struct inotify_data *d) {
+
+ assert(e);
+
+ /* GCs the inotify data object if we don't need it anymore. That's the case if we don't want to watch
+ * any inode with it anymore, which in turn happens if no event source of this priority is interested
+ * in any inode any longer. That said, we maintain an extra busy counter: if non-zero we'll delay GC
+ * (under the expectation that the GC is called again once the counter is decremented). */
+
+ if (!d)
+ return;
+
+ if (!hashmap_isempty(d->inodes))
+ return;
+
+ if (d->n_busy > 0)
+ return;
+
+ event_free_inotify_data(e, d);
+}
+
static void event_gc_inode_data(
sd_event *e,
struct inode_data *d) {
@@ -1862,8 +1890,7 @@ static void event_gc_inode_data(
inotify_data = d->inotify_data;
event_free_inode_data(e, d);
- if (inotify_data && hashmap_isempty(inotify_data->inodes))
- event_free_inotify_data(e, inotify_data);
+ event_gc_inotify_data(e, inotify_data);
}
static int event_make_inode_data(
@@ -3447,13 +3474,23 @@ static int source_dispatch(sd_event_source *s) {
sz = offsetof(struct inotify_event, name) + d->buffer.ev.len;
assert(d->buffer_filled >= sz);
+ /* If the inotify callback destroys the event source then this likely means we don't need to
+ * watch the inode anymore, and thus also won't need the inotify object anymore. But if we'd
+ * free it immediately, then we couldn't drop the event from the inotify event queue without
+ * memory corruption anymore, as below. Hence, let's not free it immediately, but mark it
+ * "busy" with a counter (which will ensure it's not GC'ed away prematurely). Let's then
+ * explicitly GC it after we are done dropping the inotify event from the buffer. */
+ d->n_busy++;
r = s->inotify.callback(s, &d->buffer.ev, s->userdata);
+ d->n_busy--;
- /* When no event is pending anymore on this inotify object, then let's drop the event from the
- * buffer. */
+ /* When no event is pending anymore on this inotify object, then let's drop the event from
+ * the inotify event queue buffer. */
if (d->n_pending == 0)
event_inotify_data_drop(e, d, sz);
+ /* Now we don't want to access 'd' anymore, it's OK to GC now. */
+ event_gc_inotify_data(e, d);
break;
}
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。