Commit | Line | Data |
---|---|---|
ae5e1b22 PE |
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> | |
7eafd7c7 SH |
12 | #include <linux/fs.h> |
13 | #include <linux/mount.h> | |
b515498f | 14 | #include <linux/user_namespace.h> |
ae5e1b22 PE |
15 | |
16 | #include "util.h" | |
17 | ||
b515498f | 18 | static struct ipc_namespace *create_ipc_ns(struct ipc_namespace *old_ns) |
ae5e1b22 | 19 | { |
ae5e1b22 | 20 | struct ipc_namespace *ns; |
7eafd7c7 | 21 | int err; |
ae5e1b22 | 22 | |
ae5e1b22 PE |
23 | ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL); |
24 | if (ns == NULL) | |
ed2ddbf8 | 25 | return ERR_PTR(-ENOMEM); |
ae5e1b22 | 26 | |
7eafd7c7 SH |
27 | atomic_set(&ns->count, 1); |
28 | err = mq_init_ns(ns); | |
29 | if (err) { | |
30 | kfree(ns); | |
31 | return ERR_PTR(err); | |
32 | } | |
4d89dc6a ND |
33 | atomic_inc(&nr_ipc_ns); |
34 | ||
ed2ddbf8 PP |
35 | sem_init_ns(ns); |
36 | msg_init_ns(ns); | |
37 | shm_init_ns(ns); | |
ae5e1b22 | 38 | |
e2c284d8 ND |
39 | /* |
40 | * msgmni has already been computed for the new ipc ns. | |
41 | * Thus, do the ipcns creation notification before registering that | |
42 | * new ipcns in the chain. | |
43 | */ | |
44 | ipcns_notify(IPCNS_CREATED); | |
b6b337ad ND |
45 | register_ipcns_notifier(ns); |
46 | ||
b515498f SH |
47 | ns->user_ns = old_ns->user_ns; |
48 | get_user_ns(ns->user_ns); | |
49 | ||
ae5e1b22 | 50 | return ns; |
ae5e1b22 PE |
51 | } |
52 | ||
53 | struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns) | |
54 | { | |
ae5e1b22 | 55 | if (!(flags & CLONE_NEWIPC)) |
64424289 | 56 | return get_ipc_ns(ns); |
b515498f | 57 | return create_ipc_ns(ns); |
ae5e1b22 PE |
58 | } |
59 | ||
01b8b07a PP |
60 | /* |
61 | * free_ipcs - free all ipcs of one type | |
62 | * @ns: the namespace to remove the ipcs from | |
63 | * @ids: the table of ipcs to free | |
64 | * @free: the function called to free each individual ipc | |
65 | * | |
66 | * Called for each kind of ipc when an ipc_namespace exits. | |
67 | */ | |
68 | void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids, | |
69 | void (*free)(struct ipc_namespace *, struct kern_ipc_perm *)) | |
70 | { | |
71 | struct kern_ipc_perm *perm; | |
72 | int next_id; | |
73 | int total, in_use; | |
74 | ||
75 | down_write(&ids->rw_mutex); | |
76 | ||
77 | in_use = ids->in_use; | |
78 | ||
79 | for (total = 0, next_id = 0; total < in_use; next_id++) { | |
80 | perm = idr_find(&ids->ipcs_idr, next_id); | |
81 | if (perm == NULL) | |
82 | continue; | |
83 | ipc_lock_by_ptr(perm); | |
84 | free(ns, perm); | |
85 | total++; | |
86 | } | |
87 | up_write(&ids->rw_mutex); | |
88 | } | |
89 | ||
b4188def AD |
90 | static void free_ipc_ns(struct ipc_namespace *ns) |
91 | { | |
92 | /* | |
93 | * Unregistering the hotplug notifier at the beginning guarantees | |
94 | * that the ipc namespace won't be freed while we are inside the | |
95 | * callback routine. Since the blocking_notifier_chain_XXX routines | |
96 | * hold a rw lock on the notifier list, unregister_ipcns_notifier() | |
97 | * won't take the rw lock before blocking_notifier_call_chain() has | |
98 | * released the rd lock. | |
99 | */ | |
100 | unregister_ipcns_notifier(ns); | |
101 | sem_exit_ns(ns); | |
102 | msg_exit_ns(ns); | |
103 | shm_exit_ns(ns); | |
104 | kfree(ns); | |
105 | atomic_dec(&nr_ipc_ns); | |
106 | ||
107 | /* | |
108 | * Do the ipcns removal notification after decrementing nr_ipc_ns in | |
109 | * order to have a correct value when recomputing msgmni. | |
110 | */ | |
111 | ipcns_notify(IPCNS_REMOVED); | |
b515498f | 112 | put_user_ns(ns->user_ns); |
b4188def AD |
113 | } |
114 | ||
7eafd7c7 SH |
115 | /* |
116 | * put_ipc_ns - drop a reference to an ipc namespace. | |
117 | * @ns: the namespace to put | |
118 | * | |
119 | * If this is the last task in the namespace exiting, and | |
120 | * it is dropping the refcount to 0, then it can race with | |
121 | * a task in another ipc namespace but in a mounts namespace | |
122 | * which has this ipcns's mqueuefs mounted, doing some action | |
123 | * with one of the mqueuefs files. That can raise the refcount. | |
124 | * So dropping the refcount, and raising the refcount when | |
125 | * accessing it through the VFS, are protected with mq_lock. | |
126 | * | |
127 | * (Clearly, a task raising the refcount on its own ipc_ns | |
128 | * needn't take mq_lock since it can't race with the last task | |
129 | * in the ipcns exiting). | |
130 | */ | |
131 | void put_ipc_ns(struct ipc_namespace *ns) | |
ae5e1b22 | 132 | { |
7eafd7c7 SH |
133 | if (atomic_dec_and_lock(&ns->count, &mq_lock)) { |
134 | mq_clear_sbinfo(ns); | |
135 | spin_unlock(&mq_lock); | |
136 | mq_put_mnt(ns); | |
137 | free_ipc_ns(ns); | |
138 | } | |
139 | } |