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         ns->user_ns = get_user_ns(user_ns);
  38
  39         err = mq_init_ns(ns);
  40         if (err) {
  41                 put_user_ns(ns->user_ns);
  42                 ns_free_inum(&ns->ns);
  43                 kfree(ns);
  44                 return ERR_PTR(err);
  45         }
  46
  47         sem_init_ns(ns);
  48         msg_init_ns(ns);
  49         shm_init_ns(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
  99         put_user_ns(ns->user_ns);
 100         ns_free_inum(&ns->ns);
 101         kfree(ns);
 102 }
 103
 104 /*
 105  * put_ipc_ns - drop a reference to an ipc namespace.
 106  * @ns: the namespace to put
 107  *
 108  * If this is the last task in the namespace exiting, and
 109  * it is dropping the refcount to 0, then it can race with
 110  * a task in another ipc namespace but in a mounts namespace
 111  * which has this ipcns's mqueuefs mounted, doing some action
 112  * with one of the mqueuefs files.  That can raise the refcount.
 113  * So dropping the refcount, and raising the refcount when
 114  * accessing it through the VFS, are protected with mq_lock.
 115  *
 116  * (Clearly, a task raising the refcount on its own ipc_ns
 117  * needn't take mq_lock since it can't race with the last task
 118  * in the ipcns exiting).
 119  */
 120 void put_ipc_ns(struct ipc_namespace *ns)
 121 {
 122         if (atomic_dec_and_lock(&ns->count, &mq_lock)) {
 123                 mq_clear_sbinfo(ns);
 124                 spin_unlock(&mq_lock);
 125                 mq_put_mnt(ns);
 126                 free_ipc_ns(ns);
 127         }
 128 }
 129
 130 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
 131 {
 132         return container_of(ns, struct ipc_namespace, ns);
 133 }
 134
 135 static struct ns_common *ipcns_get(struct task_struct *task)
 136 {
 137         struct ipc_namespace *ns = NULL;
 138         struct nsproxy *nsproxy;
 139
 140         task_lock(task);
 141         nsproxy = task->nsproxy;
 142         if (nsproxy)
 143                 ns = get_ipc_ns(nsproxy->ipc_ns);
 144         task_unlock(task);
 145
 146         return ns ? &ns->ns : NULL;
 147 }
 148
 149 static void ipcns_put(struct ns_common *ns)
 150 {
 151         return put_ipc_ns(to_ipc_ns(ns));
 152 }
 153
 154 static int ipcns_install(struct nsproxy *nsproxy, struct ns_common *new)
 155 {
 156         struct ipc_namespace *ns = to_ipc_ns(new);
 157         if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
 158             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 159                 return -EPERM;
 160
 161         /* Ditch state from the old ipc namespace */
 162         exit_sem(current);
 163         put_ipc_ns(nsproxy->ipc_ns);
 164         nsproxy->ipc_ns = get_ipc_ns(ns);
 165         return 0;
 166 }
 167
 168 const struct proc_ns_operations ipcns_operations = {
 169         .name           = "ipc",
 170         .type           = CLONE_NEWIPC,
 171         .get            = ipcns_get,
 172         .put            = ipcns_put,
 173         .install        = ipcns_install,
 174 };