ipc/namespace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/ipc/namespace.c
   4  * Copyright (C) 2006 Pavel Emelyanov <xemul@openvz.org> OpenVZ, SWsoft Inc.
   5  */
   6
   7 #include <linux/ipc.h>
   8 #include <linux/msg.h>
   9 #include <linux/ipc_namespace.h>
  10 #include <linux/rcupdate.h>
  11 #include <linux/nsproxy.h>
  12 #include <linux/slab.h>
  13 #include <linux/cred.h>
  14 #include <linux/fs.h>
  15 #include <linux/mount.h>
  16 #include <linux/user_namespace.h>
  17 #include <linux/proc_ns.h>
  18 #include <linux/sched/task.h>
  19
  20 #include "util.h"
  21
  22 /*
  23  * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
  24  */
  25 static void free_ipc(struct work_struct *unused);
  26 static DECLARE_WORK(free_ipc_work, free_ipc);
  27
  28 static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns)
  29 {
  30         return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES);
  31 }
  32
  33 static void dec_ipc_namespaces(struct ucounts *ucounts)
  34 {
  35         dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES);
  36 }
  37
  38 static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns,
  39                                            struct ipc_namespace *old_ns)
  40 {
  41         struct ipc_namespace *ns;
  42         struct ucounts *ucounts;
  43         int err;
  44
  45         err = -ENOSPC;
  46  again:
  47         ucounts = inc_ipc_namespaces(user_ns);
  48         if (!ucounts) {
  49                 /*
  50                  * IPC namespaces are freed asynchronously, by free_ipc_work.
  51                  * If frees were pending, flush_work will wait, and
  52                  * return true. Fail the allocation if no frees are pending.
  53                  */
  54                 if (flush_work(&free_ipc_work))
  55                         goto again;
  56                 goto fail;
  57         }
  58
  59         err = -ENOMEM;
  60         ns = kzalloc(sizeof(struct ipc_namespace), GFP_KERNEL_ACCOUNT);
  61         if (ns == NULL)
  62                 goto fail_dec;
  63
  64         err = ns_alloc_inum(&ns->ns);
  65         if (err)
  66                 goto fail_free;
  67         ns->ns.ops = &ipcns_operations;
  68
  69         refcount_set(&ns->ns.count, 1);
  70         ns->user_ns = get_user_ns(user_ns);
  71         ns->ucounts = ucounts;
  72
  73         err = mq_init_ns(ns);
  74         if (err)
  75                 goto fail_put;
  76
  77         err = -ENOMEM;
  78         if (!setup_mq_sysctls(ns))
  79                 goto fail_put;
  80
  81         if (!setup_ipc_sysctls(ns))
  82                 goto fail_mq;
  83
  84         err = msg_init_ns(ns);
  85         if (err)
  86                 goto fail_ipc;
  87
  88         sem_init_ns(ns);
  89         shm_init_ns(ns);
  90
  91         return ns;
  92
  93 fail_ipc:
  94         retire_ipc_sysctls(ns);
  95 fail_mq:
  96         retire_mq_sysctls(ns);
  97
  98 fail_put:
  99         put_user_ns(ns->user_ns);
 100         ns_free_inum(&ns->ns);
 101 fail_free:
 102         kfree(ns);
 103 fail_dec:
 104         dec_ipc_namespaces(ucounts);
 105 fail:
 106         return ERR_PTR(err);
 107 }
 108
 109 struct ipc_namespace *copy_ipcs(unsigned long flags,
 110         struct user_namespace *user_ns, struct ipc_namespace *ns)
 111 {
 112         if (!(flags & CLONE_NEWIPC))
 113                 return get_ipc_ns(ns);
 114         return create_ipc_ns(user_ns, ns);
 115 }
 116
 117 /*
 118  * free_ipcs - free all ipcs of one type
 119  * @ns:   the namespace to remove the ipcs from
 120  * @ids:  the table of ipcs to free
 121  * @free: the function called to free each individual ipc
 122  *
 123  * Called for each kind of ipc when an ipc_namespace exits.
 124  */
 125 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids,
 126                void (*free)(struct ipc_namespace *, struct kern_ipc_perm *))
 127 {
 128         struct kern_ipc_perm *perm;
 129         int next_id;
 130         int total, in_use;
 131
 132         down_write(&ids->rwsem);
 133
 134         in_use = ids->in_use;
 135
 136         for (total = 0, next_id = 0; total < in_use; next_id++) {
 137                 perm = idr_find(&ids->ipcs_idr, next_id);
 138                 if (perm == NULL)
 139                         continue;
 140                 rcu_read_lock();
 141                 ipc_lock_object(perm);
 142                 free(ns, perm);
 143                 total++;
 144         }
 145         up_write(&ids->rwsem);
 146 }
 147
 148 static void free_ipc_ns(struct ipc_namespace *ns)
 149 {
 150         /*
 151          * Caller needs to wait for an RCU grace period to have passed
 152          * after making the mount point inaccessible to new accesses.
 153          */
 154         mntput(ns->mq_mnt);
 155         sem_exit_ns(ns);
 156         msg_exit_ns(ns);
 157         shm_exit_ns(ns);
 158
 159         retire_mq_sysctls(ns);
 160         retire_ipc_sysctls(ns);
 161
 162         dec_ipc_namespaces(ns->ucounts);
 163         put_user_ns(ns->user_ns);
 164         ns_free_inum(&ns->ns);
 165         kfree(ns);
 166 }
 167
 168 static LLIST_HEAD(free_ipc_list);
 169 static void free_ipc(struct work_struct *unused)
 170 {
 171         struct llist_node *node = llist_del_all(&free_ipc_list);
 172         struct ipc_namespace *n, *t;
 173
 174         llist_for_each_entry_safe(n, t, node, mnt_llist)
 175                 mnt_make_shortterm(n->mq_mnt);
 176
 177         /* Wait for any last users to have gone away. */
 178         synchronize_rcu();
 179
 180         llist_for_each_entry_safe(n, t, node, mnt_llist)
 181                 free_ipc_ns(n);
 182 }
 183
 184 /*
 185  * put_ipc_ns - drop a reference to an ipc namespace.
 186  * @ns: the namespace to put
 187  *
 188  * If this is the last task in the namespace exiting, and
 189  * it is dropping the refcount to 0, then it can race with
 190  * a task in another ipc namespace but in a mounts namespace
 191  * which has this ipcns's mqueuefs mounted, doing some action
 192  * with one of the mqueuefs files.  That can raise the refcount.
 193  * So dropping the refcount, and raising the refcount when
 194  * accessing it through the VFS, are protected with mq_lock.
 195  *
 196  * (Clearly, a task raising the refcount on its own ipc_ns
 197  * needn't take mq_lock since it can't race with the last task
 198  * in the ipcns exiting).
 199  */
 200 void put_ipc_ns(struct ipc_namespace *ns)
 201 {
 202         if (refcount_dec_and_lock(&ns->ns.count, &mq_lock)) {
 203                 mq_clear_sbinfo(ns);
 204                 spin_unlock(&mq_lock);
 205
 206                 if (llist_add(&ns->mnt_llist, &free_ipc_list))
 207                         schedule_work(&free_ipc_work);
 208         }
 209 }
 210
 211 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns)
 212 {
 213         return container_of(ns, struct ipc_namespace, ns);
 214 }
 215
 216 static struct ns_common *ipcns_get(struct task_struct *task)
 217 {
 218         struct ipc_namespace *ns = NULL;
 219         struct nsproxy *nsproxy;
 220
 221         task_lock(task);
 222         nsproxy = task->nsproxy;
 223         if (nsproxy)
 224                 ns = get_ipc_ns(nsproxy->ipc_ns);
 225         task_unlock(task);
 226
 227         return ns ? &ns->ns : NULL;
 228 }
 229
 230 static void ipcns_put(struct ns_common *ns)
 231 {
 232         return put_ipc_ns(to_ipc_ns(ns));
 233 }
 234
 235 static int ipcns_install(struct nsset *nsset, struct ns_common *new)
 236 {
 237         struct nsproxy *nsproxy = nsset->nsproxy;
 238         struct ipc_namespace *ns = to_ipc_ns(new);
 239         if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) ||
 240             !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
 241                 return -EPERM;
 242
 243         put_ipc_ns(nsproxy->ipc_ns);
 244         nsproxy->ipc_ns = get_ipc_ns(ns);
 245         return 0;
 246 }
 247
 248 static struct user_namespace *ipcns_owner(struct ns_common *ns)
 249 {
 250         return to_ipc_ns(ns)->user_ns;
 251 }
 252
 253 const struct proc_ns_operations ipcns_operations = {
 254         .name           = "ipc",
 255         .type           = CLONE_NEWIPC,
 256         .get            = ipcns_get,
 257         .put            = ipcns_put,
 258         .install        = ipcns_install,
 259         .owner          = ipcns_owner,
 260 };