3 * Copyright (C) 1992 Krishna Balasubramanian
5 * Sep 1997 - Call suser() last after "normal" permission checks so we
6 * get BSD style process accounting right.
7 * Occurs in several places in the IPC code.
8 * Chris Evans, <chris@ferret.lmh.ox.ac.uk>
9 * Nov 1999 - ipc helper functions, unified SMP locking
10 * Manfred Spraul <manfreds@colorfullife.com>
11 * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary().
12 * Mingming Cao <cmm@us.ibm.com>
15 #include <linux/config.h>
17 #include <linux/shm.h>
18 #include <linux/init.h>
19 #include <linux/msg.h>
20 #include <linux/smp_lock.h>
21 #include <linux/vmalloc.h>
22 #include <linux/slab.h>
23 #include <linux/highuid.h>
24 #include <linux/security.h>
25 #include <linux/rcupdate.h>
26 #include <linux/workqueue.h>
27 #include <linux/vs_base.h>
29 #include <asm/unistd.h>
31 #include <asm/unistd.h>
36 * ipc_init - initialise IPC subsystem
38 * The various system5 IPC resources (semaphores, messages and shared
39 * memory are initialised
42 static int __init ipc_init(void)
52 * ipc_init_ids - initialise IPC identifiers
53 * @ids: Identifier set
54 * @size: Number of identifiers
56 * Given a size for the ipc identifier range (limited below IPCMNI)
57 * set up the sequence range to use then allocate and initialise the
61 void __init ipc_init_ids(struct ipc_ids* ids, int size)
64 sema_init(&ids->sem,1);
73 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
74 if(seq_limit > USHRT_MAX)
75 ids->seq_max = USHRT_MAX;
77 ids->seq_max = seq_limit;
80 ids->entries = ipc_rcu_alloc(sizeof(struct ipc_id)*size);
82 if(ids->entries == NULL) {
83 printk(KERN_ERR "ipc_init_ids() failed, ipc service disabled.\n");
86 for(i=0;i<ids->size;i++)
87 ids->entries[i].p = NULL;
91 * ipc_findkey - find a key in an ipc identifier set
92 * @ids: Identifier set
93 * @key: The key to find
95 * Requires ipc_ids.sem locked.
96 * Returns the identifier if found or -1 if not.
99 int ipc_findkey(struct ipc_ids* ids, key_t key)
102 struct kern_ipc_perm* p;
103 int max_id = ids->max_id;
106 * rcu_dereference() is not needed here
107 * since ipc_ids.sem is held
109 for (id = 0; id <= max_id; id++) {
110 p = ids->entries[id].p;
113 if (!vx_check(p->xid, VX_IDENT))
122 * Requires ipc_ids.sem locked
124 static int grow_ary(struct ipc_ids* ids, int newsize)
132 if(newsize <= ids->size)
135 new = ipc_rcu_alloc(sizeof(struct ipc_id)*newsize);
138 memcpy(new, ids->entries, sizeof(struct ipc_id)*ids->size);
139 for(i=ids->size;i<newsize;i++) {
145 * before setting the ids->entries to the new array, there must be a
146 * smp_wmb() to make sure the memcpyed contents of the new array are
147 * visible before the new array becomes visible.
149 smp_wmb(); /* prevent seeing new array uninitialized. */
151 smp_wmb(); /* prevent indexing into old array based on new size. */
159 * ipc_addid - add an IPC identifier
160 * @ids: IPC identifier set
161 * @new: new IPC permission set
162 * @size: new size limit for the id array
164 * Add an entry 'new' to the IPC arrays. The permissions object is
165 * initialised and the first free entry is set up and the id assigned
166 * is returned. The list is returned in a locked state on success.
167 * On failure the list is not locked and -1 is returned.
169 * Called with ipc_ids.sem held.
172 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
176 size = grow_ary(ids,size);
179 * rcu_dereference()() is not needed here since
180 * ipc_ids.sem is held
182 for (id = 0; id < size; id++) {
183 if(ids->entries[id].p == NULL)
189 if (id > ids->max_id)
192 new->cuid = new->uid = current->euid;
193 new->gid = new->cgid = current->egid;
195 new->seq = ids->seq++;
196 if(ids->seq > ids->seq_max)
199 new->lock = SPIN_LOCK_UNLOCKED;
202 spin_lock(&new->lock);
203 ids->entries[id].p = new;
208 * ipc_rmid - remove an IPC identifier
209 * @ids: identifier set
210 * @id: Identifier to remove
212 * The identifier must be valid, and in use. The kernel will panic if
213 * fed an invalid identifier. The entry is removed and internal
214 * variables recomputed. The object associated with the identifier
216 * ipc_ids.sem and the spinlock for this ID is hold before this function
217 * is called, and remain locked on the exit.
220 struct kern_ipc_perm* ipc_rmid(struct ipc_ids* ids, int id)
222 struct kern_ipc_perm* p;
223 int lid = id % SEQ_MULTIPLIER;
228 * do not need a rcu_dereference()() here to force ordering
229 * on Alpha, since the ipc_ids.sem is held.
231 p = ids->entries[lid].p;
232 ids->entries[lid].p = NULL;
237 if (lid == ids->max_id) {
242 } while (ids->entries[lid].p == NULL);
250 * ipc_alloc - allocate ipc space
251 * @size: size desired
253 * Allocate memory from the appropriate pools and return a pointer to it.
254 * NULL is returned if the allocation fails
257 void* ipc_alloc(int size)
263 out = kmalloc(size, GFP_KERNEL);
268 * ipc_free - free ipc space
269 * @ptr: pointer returned by ipc_alloc
270 * @size: size of block
272 * Free a block created with ipc_alloc. The caller must know the size
273 * used in the allocation call.
276 void ipc_free(void* ptr, int size)
286 * There are three headers that are prepended to the actual allocation:
287 * - during use: ipc_rcu_hdr.
288 * - during the rcu grace period: ipc_rcu_grace.
289 * - [only if vmalloc]: ipc_rcu_sched.
290 * Their lifetime doesn't overlap, thus the headers share the same memory.
291 * Unlike a normal union, they are right-aligned, thus some container_of
292 * forward/backward casting is necessary:
305 /* "void *" makes sure alignment of following data is sane. */
311 struct work_struct work;
312 /* "void *" makes sure alignment of following data is sane. */
316 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
317 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
318 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
319 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
321 static inline int rcu_use_vmalloc(int size)
323 /* Too big for a single page? */
324 if (HDRLEN_KMALLOC + size > PAGE_SIZE)
330 * ipc_rcu_alloc - allocate ipc and rcu space
331 * @size: size desired
333 * Allocate memory for the rcu header structure + the object.
334 * Returns the pointer to the object.
335 * NULL is returned if the allocation fails.
338 void* ipc_rcu_alloc(int size)
342 * We prepend the allocation with the rcu struct, and
343 * workqueue if necessary (for vmalloc).
345 if (rcu_use_vmalloc(size)) {
346 out = vmalloc(HDRLEN_VMALLOC + size);
348 out += HDRLEN_VMALLOC;
349 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
350 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
353 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
355 out += HDRLEN_KMALLOC;
356 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
357 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
364 void ipc_rcu_getref(void *ptr)
366 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
370 * ipc_schedule_free - free ipc + rcu space
372 * Since RCU callback function is called in bh,
373 * we need to defer the vfree to schedule_work
375 static void ipc_schedule_free(struct rcu_head *head)
377 struct ipc_rcu_grace *grace =
378 container_of(head, struct ipc_rcu_grace, rcu);
379 struct ipc_rcu_sched *sched =
380 container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
382 INIT_WORK(&sched->work, vfree, sched);
383 schedule_work(&sched->work);
387 * ipc_immediate_free - free ipc + rcu space
389 * Free from the RCU callback context
392 static void ipc_immediate_free(struct rcu_head *head)
394 struct ipc_rcu_grace *free =
395 container_of(head, struct ipc_rcu_grace, rcu);
399 void ipc_rcu_putref(void *ptr)
401 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
404 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
405 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
408 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
414 * ipcperms - check IPC permissions
415 * @ipcp: IPC permission set
416 * @flag: desired permission set.
418 * Check user, group, other permissions for access
419 * to ipc resources. return 0 if allowed
422 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
423 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
424 int requested_mode, granted_mode;
426 if (!vx_check(ipcp->xid, VX_ADMIN|VX_IDENT)) /* maybe just VX_IDENT? */
428 requested_mode = (flag >> 6) | (flag >> 3) | flag;
429 granted_mode = ipcp->mode;
430 if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
432 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
434 /* is there some bit set in requested_mode but not in granted_mode? */
435 if ((requested_mode & ~granted_mode & 0007) &&
436 !capable(CAP_IPC_OWNER))
439 return security_ipc_permission(ipcp, flag);
443 * Functions to convert between the kern_ipc_perm structure and the
444 * old/new ipc_perm structures
448 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
449 * @in: kernel permissions
450 * @out: new style IPC permissions
452 * Turn the kernel object 'in' into a set of permissions descriptions
453 * for returning to userspace (out).
457 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
462 out->cuid = in->cuid;
463 out->cgid = in->cgid;
464 out->mode = in->mode;
469 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
470 * @in: new style IPC permissions
471 * @out: old style IPC permissions
473 * Turn the new style permissions object in into a compatibility
474 * object and store it into the 'out' pointer.
477 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
480 SET_UID(out->uid, in->uid);
481 SET_GID(out->gid, in->gid);
482 SET_UID(out->cuid, in->cuid);
483 SET_GID(out->cgid, in->cgid);
484 out->mode = in->mode;
489 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
490 * is called with shm_ids.sem locked. Since grow_ary() is also called with
491 * shm_ids.sem down(for Shared Memory), there is no need to add read
492 * barriers here to gurantee the writes in grow_ary() are seen in order
495 * However ipc_get() itself does not necessary require ipc_ids.sem down. So
496 * if in the future ipc_get() is used by other places without ipc_ids.sem
497 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
499 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id)
501 struct kern_ipc_perm* out;
502 int lid = id % SEQ_MULTIPLIER;
505 out = ids->entries[lid].p;
509 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
511 struct kern_ipc_perm* out;
512 int lid = id % SEQ_MULTIPLIER;
513 struct ipc_id* entries;
516 if(lid >= ids->size) {
522 * Note: The following two read barriers are corresponding
523 * to the two write barriers in grow_ary(). They guarantee
524 * the writes are seen in the same order on the read side.
525 * smp_rmb() has effect on all CPUs. rcu_dereference()
526 * is used if there are data dependency between two reads, and
527 * has effect only on Alpha.
529 smp_rmb(); /* prevent indexing old array with new size */
530 entries = rcu_dereference(ids->entries);
531 out = entries[lid].p;
536 spin_lock(&out->lock);
538 /* ipc_rmid() may have already freed the ID while ipc_lock
539 * was spinning: here verify that the structure is still valid
542 spin_unlock(&out->lock);
549 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
552 spin_lock(&perm->lock);
555 void ipc_unlock(struct kern_ipc_perm* perm)
557 spin_unlock(&perm->lock);
561 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
563 return SEQ_MULTIPLIER*seq + id;
566 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
568 if(uid/SEQ_MULTIPLIER != ipcp->seq)
573 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
577 * ipc_parse_version - IPC call version
578 * @cmd: pointer to command
580 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
581 * The cmd value is turned from an encoding command and version into
582 * just the command code.
585 int ipc_parse_version (int *cmd)
595 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */