ipc/namespace.c

   1 /*
   2  * linux/ipc/namespace.c
   3  * Copyright (C) 2006 Pavel Emelyanov <xemul@openvz.org> OpenVZ, SWsoft Inc.
   4  */
   5
   6 #include <linux/ipc.h>
   7 #include <linux/msg.h>
   8 #include <linux/ipc_namespace.h>
   9 #include <linux/rcupdate.h>
  10 #include <linux/nsproxy.h>
  11 #include <linux/slab.h>
  12 #include <linux/fs.h>
  13 #include <linux/mount.h>
  14 #include <linux/user_namespace.h>
  15 #include <linux/proc_ns.h>
  16
  17 #include "util.h"
  18
  19 static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
  20                                            struct ipc_namespace *old_ns)
  21 {
  22         struct ipc_namespace *ns;
  23         int err;
  24
  25         ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
  26         if (ns == NULL)
  27                 return ERR_PTR(-ENOMEM);
  28
  29         err = ns_alloc_inum(&ns->ns);
  30         if (err) {
  31                 kfree(ns);
  32                 return ERR_PTR(err);
  33         }
  34         ns->ns.ops = &ipcns_operations;
  35
  36         atomic_set(&ns->count, 1);
  37         err = mq_init_ns(ns);
  38         if (err) {
  39                 ns_free_inum(&ns->ns);
  40                 kfree(ns);
  41                 return ERR_PTR(err);
  42         }
  43         atomic_inc(&nr_ipc_ns);
  44
  45         sem_init_ns(ns);
  46         msg_init_ns(ns);
  47         shm_init_ns(ns);
  48
  49         ns->user_ns = get_user_ns(user_ns);
  50
  51         return ns;
  52 }
  53
  54 struct ipc_namespace *copy_ipcs(unsigned long flags,
  55         struct user_namespace *user_ns, struct ipc_namespace *ns)
  56 {
  57         if (!(flags & CLONE_NEWIPC))
  58                 return get_ipc_ns(ns);
  59         return create_ipc_ns(user_ns, ns);
  60 }
  61
  62 /*
  63  * free_ipcs - free all ipcs of one type
  64  * @ns:   the namespace to remove the ipcs from
  65  * @ids:  the table of ipcs to free
  66  * @free: the function called to free each individual ipc
  67  *
  68  * Called for each kind of ipc when an ipc_namespace exits.
  69  */
  70 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
  71                void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
  72 {
  73         struct kern_ipc_perm *perm;
  74         int next_id;
  75         int total, in_use;
  76
  77         down_write(&ids->rwsem);
  78
  79         in_use = ids->in_use;
  80
  81         for (total = 0, next_id = 0; total < in_use; next_id++) {
  82                 perm = idr_find(&ids->ipcs_idr, next_id);
  83                 if (perm == NULL)
  84                         continue;
  85                 rcu_read_lock();
  86                 ipc_lock_object(perm);
  87                 free(ns, perm);
  88                 total++;
  89         }
  90         up_write(&ids->rwsem);
  91 }
  92
  93 static void free_ipc_ns(struct ipc_namespace *ns)
  94 {
  95         sem_exit_ns(ns);
  96         msg_exit_ns(ns);
  97         shm_exit_ns(ns);
  98         atomic_dec(&nr_ipc_ns);
  99
 100         put_user_ns(ns->user_ns);
 101         ns_free_inum(&ns->ns);
 102         kfree(ns);
 103 }
 104
 105 /*
 106  * put_ipc_ns - drop a reference to an ipc namespace.
 107  * @ns: the namespace to put
 108  *
 109  * If this is the last task in the namespace exiting, and
 110  * it is dropping the refcount to 0, then it can race with
 111  * a task in another ipc namespace but in a mounts namespace
 112  * which has this ipcns's mqueuefs mounted, doing some action
 113  * with one of the mqueuefs files.  That can raise the refcount.
 114  * So dropping the refcount, and raising the refcount when
 115  * accessing it through the VFS, are protected with mq_lock.
 116  *
 117  * (Clearly, a task raising the refcount on its own ipc_ns
 118  * needn't take mq_lock since it can't race with the last task
 119  * in the ipcns exiting).
 120  */
 121 void put_ipc_ns(struct ipc_namespace *ns)
 122 {
 123         if (atomic_dec_and_lock(&ns->count, &mq_lock)) {
 124                 mq_clear_sbinfo(ns);
 125                 spin_unlock(&mq_lock);
 126                 mq_put_mnt(ns);
 127                 free_ipc_ns(ns);
 128         }
 129 }
 130
 131 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
 132 {
 133         return container_of(ns, struct ipc_namespace, ns);
 134 }
 135
 136 static struct ns_common *ipcns_get(struct task_struct *task)
 137 {
 138         struct ipc_namespace *ns = NULL;
 139         struct nsproxy *nsproxy;
 140
 141         task_lock(task);
 142         nsproxy = task->nsproxy;
 143         if (nsproxy)
 144                 ns = get_ipc_ns(nsproxy->ipc_ns);
 145         task_unlock(task);
 146
 147         return ns ? &ns->ns : NULL;
 148 }
 149
 150 static void ipcns_put(struct ns_common *ns)
 151 {
 152         return put_ipc_ns(to_ipc_ns(ns));
 153 }
 154
 155 static int ipcns_install(struct nsproxy *nsproxy, struct ns_common *new)
 156 {
 157         struct ipc_namespace *ns = to_ipc_ns(new);
 158         if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
 159             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 160                 return -EPERM;
 161
 162         /* Ditch state from the old ipc namespace */
 163         exit_sem(current);
 164         put_ipc_ns(nsproxy->ipc_ns);
 165         nsproxy->ipc_ns = get_ipc_ns(ns);
 166         return 0;
 167 }
 168
 169 const struct proc_ns_operations ipcns_operations = {
 170         .name           = "ipc",
 171         .type           = CLONE_NEWIPC,
 172         .get            = ipcns_get,
 173         .put            = ipcns_put,
 174         .install        = ipcns_install,
 175 };