Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'af_unix-remove-old-gc-leftovers'
Kuniyuki Iwashima says:
====================
af_unix: Remove old GC leftovers.
This is a follow-up series for commit 4090fa373f0e ("af_unix: Replace
garbage collection algorithm.") which introduced the new GC for AF_UNIX.
Now we no longer need two ugly tricks for the old GC, let's remove them.
====================
Link: https://lore.kernel.org/r/20240401173125.92184-1-kuniyu@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
+3
-51
-1
include/net/af_unix.h
-1
include/net/af_unix.h
+2
-49
net/unix/af_unix.c
+2
-49
net/unix/af_unix.c
···
1794
1794
if (too_many_unix_fds(current))
1795
1795
return -ETOOMANYREFS;
1796
1796
1797
-
/* Need to duplicate file references for the sake of garbage
1798
-
* collection. Otherwise a socket in the fps might become a
1799
-
* candidate for GC while the skb is not yet queued.
1800
-
*/
1801
-
UNIXCB(skb).fp = scm_fp_dup(scm->fp);
1802
-
if (!UNIXCB(skb).fp)
1803
-
return -ENOMEM;
1797
+
UNIXCB(skb).fp = scm->fp;
1798
+
scm->fp = NULL;
1804
1799
1805
1800
if (unix_prepare_fpl(UNIXCB(skb).fp))
1806
1801
return -ENOMEM;
···
1814
1819
static void unix_peek_fds(struct scm_cookie *scm, struct sk_buff *skb)
1815
1820
{
1816
1821
scm->fp = scm_fp_dup(UNIXCB(skb).fp);
1817
-
1818
-
/*
1819
-
* Garbage collection of unix sockets starts by selecting a set of
1820
-
* candidate sockets which have reference only from being in flight
1821
-
* (total_refs == inflight_refs). This condition is checked once during
1822
-
* the candidate collection phase, and candidates are marked as such, so
1823
-
* that non-candidates can later be ignored. While inflight_refs is
1824
-
* protected by unix_gc_lock, total_refs (file count) is not, hence this
1825
-
* is an instantaneous decision.
1826
-
*
1827
-
* Once a candidate, however, the socket must not be reinstalled into a
1828
-
* file descriptor while the garbage collection is in progress.
1829
-
*
1830
-
* If the above conditions are met, then the directed graph of
1831
-
* candidates (*) does not change while unix_gc_lock is held.
1832
-
*
1833
-
* Any operations that changes the file count through file descriptors
1834
-
* (dup, close, sendmsg) does not change the graph since candidates are
1835
-
* not installed in fds.
1836
-
*
1837
-
* Dequeing a candidate via recvmsg would install it into an fd, but
1838
-
* that takes unix_gc_lock to decrement the inflight count, so it's
1839
-
* serialized with garbage collection.
1840
-
*
1841
-
* MSG_PEEK is special in that it does not change the inflight count,
1842
-
* yet does install the socket into an fd. The following lock/unlock
1843
-
* pair is to ensure serialization with garbage collection. It must be
1844
-
* done between incrementing the file count and installing the file into
1845
-
* an fd.
1846
-
*
1847
-
* If garbage collection starts after the barrier provided by the
1848
-
* lock/unlock, then it will see the elevated refcount and not mark this
1849
-
* as a candidate. If a garbage collection is already in progress
1850
-
* before the file count was incremented, then the lock/unlock pair will
1851
-
* ensure that garbage collection is finished before progressing to
1852
-
* installing the fd.
1853
-
*
1854
-
* (*) A -> B where B is on the queue of A or B is on the queue of C
1855
-
* which is on the queue of listening socket A.
1856
-
*/
1857
-
spin_lock(&unix_gc_lock);
1858
-
spin_unlock(&unix_gc_lock);
1859
1822
}
1860
1823
1861
1824
static void unix_destruct_scm(struct sk_buff *skb)