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/cred.h> 13 #include <linux/fs.h> 14 #include <linux/mount.h> 15 #include <linux/user_namespace.h> 16 #include <linux/proc_ns.h> 17 18 #include "util.h" 19 20 static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns) 21 { 22 return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES); 23 } 24 25 static void dec_ipc_namespaces(struct ucounts *ucounts) 26 { 27 dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES); 28 } 29 30 static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns, 31 struct ipc_namespace *old_ns) 32 { 33 struct ipc_namespace *ns; 34 struct ucounts *ucounts; 35 int err; 36 37 err = -ENOSPC; 38 ucounts = inc_ipc_namespaces(user_ns); 39 if (!ucounts) 40 goto fail; 41 42 err = -ENOMEM; 43 ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL); 44 if (ns == NULL) 45 goto fail_dec; 46 47 err = ns_alloc_inum(&ns->ns); 48 if (err) 49 goto fail_free; 50 ns->ns.ops = &ipcns_operations; 51 52 atomic_set(&ns->count, 1); 53 ns->user_ns = get_user_ns(user_ns); 54 ns->ucounts = ucounts; 55 56 err = mq_init_ns(ns); 57 if (err) 58 goto fail_put; 59 60 sem_init_ns(ns); 61 msg_init_ns(ns); 62 shm_init_ns(ns); 63 64 return ns; 65 66 fail_put: 67 put_user_ns(ns->user_ns); 68 ns_free_inum(&ns->ns); 69 fail_free: 70 kfree(ns); 71 fail_dec: 72 dec_ipc_namespaces(ucounts); 73 fail: 74 return ERR_PTR(err); 75 } 76 77 struct ipc_namespace *copy_ipcs(unsigned long flags, 78 struct user_namespace *user_ns, struct ipc_namespace *ns) 79 { 80 if (!(flags & CLONE_NEWIPC)) 81 return get_ipc_ns(ns); 82 return create_ipc_ns(user_ns, ns); 83 } 84 85 /* 86 * free_ipcs - free all ipcs of one type 87 * @ns: the namespace to remove the ipcs from 88 * @ids: the table of ipcs to free 89 * @free: the function called to free each individual ipc 90 * 91 * Called for each kind of ipc when an ipc_namespace exits. 92 */ 93 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids, 94 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *)) 95 { 96 struct kern_ipc_perm *perm; 97 int next_id; 98 int total, in_use; 99 100 down_write(&ids->rwsem); 101 102 in_use = ids->in_use; 103 104 for (total = 0, next_id = 0; total < in_use; next_id++) { 105 perm = idr_find(&ids->ipcs_idr, next_id); 106 if (perm == NULL) 107 continue; 108 rcu_read_lock(); 109 ipc_lock_object(perm); 110 free(ns, perm); 111 total++; 112 } 113 up_write(&ids->rwsem); 114 } 115 116 static void free_ipc_ns(struct ipc_namespace *ns) 117 { 118 sem_exit_ns(ns); 119 msg_exit_ns(ns); 120 shm_exit_ns(ns); 121 122 dec_ipc_namespaces(ns->ucounts); 123 put_user_ns(ns->user_ns); 124 ns_free_inum(&ns->ns); 125 kfree(ns); 126 } 127 128 /* 129 * put_ipc_ns - drop a reference to an ipc namespace. 130 * @ns: the namespace to put 131 * 132 * If this is the last task in the namespace exiting, and 133 * it is dropping the refcount to 0, then it can race with 134 * a task in another ipc namespace but in a mounts namespace 135 * which has this ipcns's mqueuefs mounted, doing some action 136 * with one of the mqueuefs files. That can raise the refcount. 137 * So dropping the refcount, and raising the refcount when 138 * accessing it through the VFS, are protected with mq_lock. 139 * 140 * (Clearly, a task raising the refcount on its own ipc_ns 141 * needn't take mq_lock since it can't race with the last task 142 * in the ipcns exiting). 143 */ 144 void put_ipc_ns(struct ipc_namespace *ns) 145 { 146 if (atomic_dec_and_lock(&ns->count, &mq_lock)) { 147 mq_clear_sbinfo(ns); 148 spin_unlock(&mq_lock); 149 mq_put_mnt(ns); 150 free_ipc_ns(ns); 151 } 152 } 153 154 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns) 155 { 156 return container_of(ns, struct ipc_namespace, ns); 157 } 158 159 static struct ns_common *ipcns_get(struct task_struct *task) 160 { 161 struct ipc_namespace *ns = NULL; 162 struct nsproxy *nsproxy; 163 164 task_lock(task); 165 nsproxy = task->nsproxy; 166 if (nsproxy) 167 ns = get_ipc_ns(nsproxy->ipc_ns); 168 task_unlock(task); 169 170 return ns ? &ns->ns : NULL; 171 } 172 173 static void ipcns_put(struct ns_common *ns) 174 { 175 return put_ipc_ns(to_ipc_ns(ns)); 176 } 177 178 static int ipcns_install(struct nsproxy *nsproxy, struct ns_common *new) 179 { 180 struct ipc_namespace *ns = to_ipc_ns(new); 181 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) || 182 !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 183 return -EPERM; 184 185 /* Ditch state from the old ipc namespace */ 186 exit_sem(current); 187 put_ipc_ns(nsproxy->ipc_ns); 188 nsproxy->ipc_ns = get_ipc_ns(ns); 189 return 0; 190 } 191 192 static struct user_namespace *ipcns_owner(struct ns_common *ns) 193 { 194 return to_ipc_ns(ns)->user_ns; 195 } 196 197 const struct proc_ns_operations ipcns_operations = { 198 .name = "ipc", 199 .type = CLONE_NEWIPC, 200 .get = ipcns_get, 201 .put = ipcns_put, 202 .install = ipcns_install, 203 .owner = ipcns_owner, 204 }; 205