releases/3.10.87/ipc-modify-message-queue-accounting-to-not-take-kernel-data-structures-into-account.patch

   1 From de54b9ac253787c366bbfb28d901a31954eb3511 Mon Sep 17 00:00:00 2001
   2 From: Marcus Gelderie <redmnic@gmail.com>
   3 Date: Thu, 6 Aug 2015 15:46:10 -0700
   4 Subject: ipc: modify message queue accounting to not take kernel data structures into account
   5
   6 From: Marcus Gelderie <redmnic@gmail.com>
   7
   8 commit de54b9ac253787c366bbfb28d901a31954eb3511 upstream.
   9
  10 A while back, the message queue implementation in the kernel was
  11 improved to use btrees to speed up retrieval of messages, in commit
  12 d6629859b36d ("ipc/mqueue: improve performance of send/recv").
  13
  14 That patch introducing the improved kernel handling of message queues
  15 (using btrees) has, as a by-product, changed the meaning of the QSIZE
  16 field in the pseudo-file created for the queue.  Before, this field
  17 reflected the size of the user-data in the queue.  Since, it also takes
  18 kernel data structures into account.  For example, if 13 bytes of user
  19 data are in the queue, on my machine the file reports a size of 61
  20 bytes.
  21
  22 There was some discussion on this topic before (for example
  23 https://lkml.org/lkml/2014/10/1/115).  Commenting on a th lkml, Michael
  24 Kerrisk gave the following background
  25 (https://lkml.org/lkml/2015/6/16/74):
  26
  27     The pseudofiles in the mqueue filesystem (usually mounted at
  28     /dev/mqueue) expose fields with metadata describing a message
  29     queue. One of these fields, QSIZE, as originally implemented,
  30     showed the total number of bytes of user data in all messages in
  31     the message queue, and this feature was documented from the
  32     beginning in the mq_overview(7) page. In 3.5, some other (useful)
  33     work happened to break the user-space API in a couple of places,
  34     including the value exposed via QSIZE, which now includes a measure
  35     of kernel overhead bytes for the queue, a figure that renders QSIZE
  36     useless for its original purpose, since there's no way to deduce
  37     the number of overhead bytes consumed by the implementation.
  38     (The other user-space breakage was subsequently fixed.)
  39
  40 This patch removes the accounting of kernel data structures in the
  41 queue.  Reporting the size of these data-structures in the QSIZE field
  42 was a breaking change (see Michael's comment above).  Without the QSIZE
  43 field reporting the total size of user-data in the queue, there is no
  44 way to deduce this number.
  45
  46 It should be noted that the resource limit RLIMIT_MSGQUEUE is counted
  47 against the worst-case size of the queue (in both the old and the new
  48 implementation).  Therefore, the kernel overhead accounting in QSIZE is
  49 not necessary to help the user understand the limitations RLIMIT imposes
  50 on the processes.
  51
  52 Signed-off-by: Marcus Gelderie <redmnic@gmail.com>
  53 Acked-by: Doug Ledford <dledford@redhat.com>
  54 Acked-by: Michael Kerrisk <mtk.manpages@gmail.com>
  55 Acked-by: Davidlohr Bueso <dbueso@suse.de>
  56 Cc: David Howells <dhowells@redhat.com>
  57 Cc: Alexander Viro <viro@zeniv.linux.org.uk>
  58 Cc: John Duffy <jb_duffy@btinternet.com>
  59 Cc: Arto Bendiken <arto@bendiken.net>
  60 Cc: Manfred Spraul <manfred@colorfullife.com>
  61 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  62 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  63 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  64
  65 ---
  66  ipc/mqueue.c |    5 -----
  67  1 file changed, 5 deletions(-)
  68
  69 --- a/ipc/mqueue.c
  70 +++ b/ipc/mqueue.c
  71 @@ -143,7 +143,6 @@ static int msg_insert(struct msg_msg *ms
  72                 if (!leaf)
  73                         return -ENOMEM;
  74                 INIT_LIST_HEAD(&leaf->msg_list);
  75 -               info->qsize += sizeof(*leaf);
  76         }
  77         leaf->priority = msg->m_type;
  78         rb_link_node(&leaf->rb_node, parent, p);
  79 @@ -188,7 +187,6 @@ try_again:
  80                              "lazy leaf delete!\n");
  81                 rb_erase(&leaf->rb_node, &info->msg_tree);
  82                 if (info->node_cache) {
  83 -                       info->qsize -= sizeof(*leaf);
  84                         kfree(leaf);
  85                 } else {
  86                         info->node_cache = leaf;
  87 @@ -201,7 +199,6 @@ try_again:
  88                 if (list_empty(&leaf->msg_list)) {
  89                         rb_erase(&leaf->rb_node, &info->msg_tree);
  90                         if (info->node_cache) {
  91 -                               info->qsize -= sizeof(*leaf);
  92                                 kfree(leaf);
  93                         } else {
  94                                 info->node_cache = leaf;
  95 @@ -1026,7 +1023,6 @@ SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqd
  96                 /* Save our speculative allocation into the cache */
  97                 INIT_LIST_HEAD(&new_leaf->msg_list);
  98                 info->node_cache = new_leaf;
  99 -               info->qsize += sizeof(*new_leaf);
 100                 new_leaf = NULL;
 101         } else {
 102                 kfree(new_leaf);
 103 @@ -1133,7 +1129,6 @@ SYSCALL_DEFINE5(mq_timedreceive, mqd_t,
 104                 /* Save our speculative allocation into the cache */
 105                 INIT_LIST_HEAD(&new_leaf->msg_list);
 106                 info->node_cache = new_leaf;
 107 -               info->qsize += sizeof(*new_leaf);
 108         } else {
 109                 kfree(new_leaf);
 110         }