cgroups: use flex_array in attach_proc
authorBen Blum <bblum@andrew.cmu.edu>
Thu, 26 May 2011 23:25:21 +0000 (16:25 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 27 May 2011 00:12:34 +0000 (17:12 -0700)
Convert cgroup_attach_proc to use flex_array.

The cgroup_attach_proc implementation requires a pre-allocated array to
store task pointers to atomically move a thread-group, but asking for a
monolithic array with kmalloc() may be unreliable for very large groups.
Using flex_array provides the same functionality with less risk of
failure.

This is a post-patch for cgroup-procs-write.patch.

Signed-off-by: Ben Blum <bblum@andrew.cmu.edu>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Matt Helsley <matthltc@us.ibm.com>
Reviewed-by: Paul Menage <menage@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/cgroup.c

index 5e6a9745f0e75500157830bc80781882367b2cfc..00a884342d3db7176b98b48baef632557bbf2764 100644 (file)
@@ -57,6 +57,7 @@
 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
 #include <linux/eventfd.h>
 #include <linux/poll.h>
+#include <linux/flex_array.h> /* used in cgroup_attach_proc */
 
 #include <asm/atomic.h>
 
@@ -1995,7 +1996,7 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
        struct cgroupfs_root *root = cgrp->root;
        /* threadgroup list cursor and array */
        struct task_struct *tsk;
-       struct task_struct **group;
+       struct flex_array *group;
        /*
         * we need to make sure we have css_sets for all the tasks we're
         * going to move -before- we actually start moving them, so that in
@@ -2012,9 +2013,15 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
         * and if threads exit, this will just be an over-estimate.
         */
        group_size = get_nr_threads(leader);
-       group = kmalloc(group_size * sizeof(*group), GFP_KERNEL);
+       /* flex_array supports very large thread-groups better than kmalloc. */
+       group = flex_array_alloc(sizeof(struct task_struct *), group_size,
+                                GFP_KERNEL);
        if (!group)
                return -ENOMEM;
+       /* pre-allocate to guarantee space while iterating in rcu read-side. */
+       retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
+       if (retval)
+               goto out_free_group_list;
 
        /* prevent changes to the threadgroup list while we take a snapshot. */
        rcu_read_lock();
@@ -2037,7 +2044,12 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
                /* as per above, nr_threads may decrease, but not increase. */
                BUG_ON(i >= group_size);
                get_task_struct(tsk);
-               group[i] = tsk;
+               /*
+                * saying GFP_ATOMIC has no effect here because we did prealloc
+                * earlier, but it's good form to communicate our expectations.
+                */
+               retval = flex_array_put_ptr(group, i, tsk, GFP_ATOMIC);
+               BUG_ON(retval != 0);
                i++;
        } while_each_thread(leader, tsk);
        /* remember the number of threads in the array for later. */
@@ -2059,7 +2071,8 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
                if (ss->can_attach_task) {
                        /* run on each task in the threadgroup. */
                        for (i = 0; i < group_size; i++) {
-                               retval = ss->can_attach_task(cgrp, group[i]);
+                               tsk = flex_array_get_ptr(group, i);
+                               retval = ss->can_attach_task(cgrp, tsk);
                                if (retval) {
                                        failed_ss = ss;
                                        cancel_failed_ss = true;
@@ -2075,7 +2088,7 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
         */
        INIT_LIST_HEAD(&newcg_list);
        for (i = 0; i < group_size; i++) {
-               tsk = group[i];
+               tsk = flex_array_get_ptr(group, i);
                /* nothing to do if this task is already in the cgroup */
                oldcgrp = task_cgroup_from_root(tsk, root);
                if (cgrp == oldcgrp)
@@ -2114,7 +2127,7 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
                        ss->pre_attach(cgrp);
        }
        for (i = 0; i < group_size; i++) {
-               tsk = group[i];
+               tsk = flex_array_get_ptr(group, i);
                /* leave current thread as it is if it's already there */
                oldcgrp = task_cgroup_from_root(tsk, root);
                if (cgrp == oldcgrp)
@@ -2167,10 +2180,12 @@ out_cancel_attach:
                }
        }
        /* clean up the array of referenced threads in the group. */
-       for (i = 0; i < group_size; i++)
-               put_task_struct(group[i]);
+       for (i = 0; i < group_size; i++) {
+               tsk = flex_array_get_ptr(group, i);
+               put_task_struct(tsk);
+       }
 out_free_group_list:
-       kfree(group);
+       flex_array_free(group);
        return retval;
 }