| 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_fs.h> |
| 16 | |
| 17 | #include "util.h" |
| 18 | |
| 19 | static struct ipc_namespace *create_ipc_ns(struct task_struct *tsk, |
| 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 | atomic_set(&ns->count, 1); |
| 30 | err = mq_init_ns(ns); |
| 31 | if (err) { |
| 32 | kfree(ns); |
| 33 | return ERR_PTR(err); |
| 34 | } |
| 35 | atomic_inc(&nr_ipc_ns); |
| 36 | |
| 37 | sem_init_ns(ns); |
| 38 | msg_init_ns(ns); |
| 39 | shm_init_ns(ns); |
| 40 | |
| 41 | /* |
| 42 | * msgmni has already been computed for the new ipc ns. |
| 43 | * Thus, do the ipcns creation notification before registering that |
| 44 | * new ipcns in the chain. |
| 45 | */ |
| 46 | ipcns_notify(IPCNS_CREATED); |
| 47 | register_ipcns_notifier(ns); |
| 48 | |
| 49 | ns->user_ns = get_user_ns(task_cred_xxx(tsk, user_ns)); |
| 50 | |
| 51 | return ns; |
| 52 | } |
| 53 | |
| 54 | struct ipc_namespace *copy_ipcs(unsigned long flags, |
| 55 | struct task_struct *tsk) |
| 56 | { |
| 57 | struct ipc_namespace *ns = tsk->nsproxy->ipc_ns; |
| 58 | |
| 59 | if (!(flags & CLONE_NEWIPC)) |
| 60 | return get_ipc_ns(ns); |
| 61 | return create_ipc_ns(tsk, ns); |
| 62 | } |
| 63 | |
| 64 | /* |
| 65 | * free_ipcs - free all ipcs of one type |
| 66 | * @ns: the namespace to remove the ipcs from |
| 67 | * @ids: the table of ipcs to free |
| 68 | * @free: the function called to free each individual ipc |
| 69 | * |
| 70 | * Called for each kind of ipc when an ipc_namespace exits. |
| 71 | */ |
| 72 | void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids, |
| 73 | void (*free)(struct ipc_namespace *, struct kern_ipc_perm *)) |
| 74 | { |
| 75 | struct kern_ipc_perm *perm; |
| 76 | int next_id; |
| 77 | int total, in_use; |
| 78 | |
| 79 | down_write(&ids->rw_mutex); |
| 80 | |
| 81 | in_use = ids->in_use; |
| 82 | |
| 83 | for (total = 0, next_id = 0; total < in_use; next_id++) { |
| 84 | perm = idr_find(&ids->ipcs_idr, next_id); |
| 85 | if (perm == NULL) |
| 86 | continue; |
| 87 | ipc_lock_by_ptr(perm); |
| 88 | free(ns, perm); |
| 89 | total++; |
| 90 | } |
| 91 | up_write(&ids->rw_mutex); |
| 92 | } |
| 93 | |
| 94 | static void free_ipc_ns(struct ipc_namespace *ns) |
| 95 | { |
| 96 | /* |
| 97 | * Unregistering the hotplug notifier at the beginning guarantees |
| 98 | * that the ipc namespace won't be freed while we are inside the |
| 99 | * callback routine. Since the blocking_notifier_chain_XXX routines |
| 100 | * hold a rw lock on the notifier list, unregister_ipcns_notifier() |
| 101 | * won't take the rw lock before blocking_notifier_call_chain() has |
| 102 | * released the rd lock. |
| 103 | */ |
| 104 | unregister_ipcns_notifier(ns); |
| 105 | sem_exit_ns(ns); |
| 106 | msg_exit_ns(ns); |
| 107 | shm_exit_ns(ns); |
| 108 | atomic_dec(&nr_ipc_ns); |
| 109 | |
| 110 | /* |
| 111 | * Do the ipcns removal notification after decrementing nr_ipc_ns in |
| 112 | * order to have a correct value when recomputing msgmni. |
| 113 | */ |
| 114 | ipcns_notify(IPCNS_REMOVED); |
| 115 | put_user_ns(ns->user_ns); |
| 116 | kfree(ns); |
| 117 | } |
| 118 | |
| 119 | /* |
| 120 | * put_ipc_ns - drop a reference to an ipc namespace. |
| 121 | * @ns: the namespace to put |
| 122 | * |
| 123 | * If this is the last task in the namespace exiting, and |
| 124 | * it is dropping the refcount to 0, then it can race with |
| 125 | * a task in another ipc namespace but in a mounts namespace |
| 126 | * which has this ipcns's mqueuefs mounted, doing some action |
| 127 | * with one of the mqueuefs files. That can raise the refcount. |
| 128 | * So dropping the refcount, and raising the refcount when |
| 129 | * accessing it through the VFS, are protected with mq_lock. |
| 130 | * |
| 131 | * (Clearly, a task raising the refcount on its own ipc_ns |
| 132 | * needn't take mq_lock since it can't race with the last task |
| 133 | * in the ipcns exiting). |
| 134 | */ |
| 135 | void put_ipc_ns(struct ipc_namespace *ns) |
| 136 | { |
| 137 | if (atomic_dec_and_lock(&ns->count, &mq_lock)) { |
| 138 | mq_clear_sbinfo(ns); |
| 139 | spin_unlock(&mq_lock); |
| 140 | mq_put_mnt(ns); |
| 141 | free_ipc_ns(ns); |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | static void *ipcns_get(struct task_struct *task) |
| 146 | { |
| 147 | struct ipc_namespace *ns = NULL; |
| 148 | struct nsproxy *nsproxy; |
| 149 | |
| 150 | rcu_read_lock(); |
| 151 | nsproxy = task_nsproxy(task); |
| 152 | if (nsproxy) |
| 153 | ns = get_ipc_ns(nsproxy->ipc_ns); |
| 154 | rcu_read_unlock(); |
| 155 | |
| 156 | return ns; |
| 157 | } |
| 158 | |
| 159 | static void ipcns_put(void *ns) |
| 160 | { |
| 161 | return put_ipc_ns(ns); |
| 162 | } |
| 163 | |
| 164 | static int ipcns_install(struct nsproxy *nsproxy, void *ns) |
| 165 | { |
| 166 | /* Ditch state from the old ipc namespace */ |
| 167 | exit_sem(current); |
| 168 | put_ipc_ns(nsproxy->ipc_ns); |
| 169 | nsproxy->ipc_ns = get_ipc_ns(ns); |
| 170 | return 0; |
| 171 | } |
| 172 | |
| 173 | const struct proc_ns_operations ipcns_operations = { |
| 174 | .name = "ipc", |
| 175 | .type = CLONE_NEWIPC, |
| 176 | .get = ipcns_get, |
| 177 | .put = ipcns_put, |
| 178 | .install = ipcns_install, |
| 179 | }; |