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 = proc_alloc_inum(&ns->proc_inum);
  30         if (err) {
  31                 kfree(ns);
  32                 return ERR_PTR(err);
  33         }
  34
  35         atomic_set(&ns->count, 1);
  36         err = mq_init_ns(ns);
  37         if (err) {
  38                 proc_free_inum(ns->proc_inum);
  39                 kfree(ns);
  40                 return ERR_PTR(err);
  41         }
  42         atomic_inc(&nr_ipc_ns);
  43
  44         sem_init_ns(ns);
  45         msg_init_ns(ns);
  46         shm_init_ns(ns);
  47
  48         ns->user_ns = get_user_ns(user_ns);
  49
  50         return ns;
  51 }
  52
  53 struct ipc_namespace *copy_ipcs(unsigned long flags,
  54         struct user_namespace *user_ns, struct ipc_namespace *ns)
  55 {
  56         if (!(flags & CLONE_NEWIPC))
  57                 return get_ipc_ns(ns);
  58         return create_ipc_ns(user_ns, ns);
  59 }
  60
  61 /*
  62  * free_ipcs - free all ipcs of one type
  63  * @ns:   the namespace to remove the ipcs from
  64  * @ids:  the table of ipcs to free
  65  * @free: the function called to free each individual ipc
  66  *
  67  * Called for each kind of ipc when an ipc_namespace exits.
  68  */
  69 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
  70                void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
  71 {
  72         struct kern_ipc_perm *perm;
  73         int next_id;
  74         int total, in_use;
  75
  76         down_write(&ids->rwsem);
  77
  78         in_use = ids->in_use;
  79
  80         for (total = 0, next_id = 0; total < in_use; next_id++) {
  81                 perm = idr_find(&ids->ipcs_idr, next_id);
  82                 if (perm == NULL)
  83                         continue;
  84                 rcu_read_lock();
  85                 ipc_lock_object(perm);
  86                 free(ns, perm);
  87                 total++;
  88         }
  89         up_write(&ids->rwsem);
  90 }
  91
  92 static void free_ipc_ns(struct ipc_namespace *ns)
  93 {
  94         sem_exit_ns(ns);
  95         msg_exit_ns(ns);
  96         shm_exit_ns(ns);
  97         atomic_dec(&nr_ipc_ns);
  98
  99         put_user_ns(ns->user_ns);
 100         proc_free_inum(ns->proc_inum);
 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 void *ipcns_get(struct task_struct *task)
 131 {
 132         struct ipc_namespace *ns = NULL;
 133         struct nsproxy *nsproxy;
 134
 135         task_lock(task);
 136         nsproxy = task->nsproxy;
 137         if (nsproxy)
 138                 ns = get_ipc_ns(nsproxy->ipc_ns);
 139         task_unlock(task);
 140
 141         return ns;
 142 }
 143
 144 static void ipcns_put(void *ns)
 145 {
 146         return put_ipc_ns(ns);
 147 }
 148
 149 static int ipcns_install(struct nsproxy *nsproxy, void *new)
 150 {
 151         struct ipc_namespace *ns = new;
 152         if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
 153             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 154                 return -EPERM;
 155
 156         /* Ditch state from the old ipc namespace */
 157         exit_sem(current);
 158         put_ipc_ns(nsproxy->ipc_ns);
 159         nsproxy->ipc_ns = get_ipc_ns(ns);
 160         return 0;
 161 }
 162
 163 static unsigned int ipcns_inum(void *vp)
 164 {
 165         struct ipc_namespace *ns = vp;
 166
 167         return ns->proc_inum;
 168 }
 169
 170 const struct proc_ns_operations ipcns_operations = {
 171         .name           = "ipc",
 172         .type           = CLONE_NEWIPC,
 173         .get            = ipcns_get,
 174         .put            = ipcns_put,
 175         .install        = ipcns_install,
 176         .inum           = ipcns_inum,
 177 };