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 <manfred@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/capability.h>
24 #include <linux/highuid.h>
25 #include <linux/security.h>
26 #include <linux/rcupdate.h>
27 #include <linux/workqueue.h>
28 #include <linux/seq_file.h>
29 #include <linux/proc_fs.h>
30 #include <asm/unistd.h>
34 struct ipc_proc_iface {
38 int (*show)(struct seq_file *, void *);
42 * ipc_init - initialise IPC subsystem
44 * The various system5 IPC resources (semaphores, messages and shared
45 * memory are initialised
48 static int __init ipc_init(void)
58 * ipc_init_ids - initialise IPC identifiers
59 * @ids: Identifier set
60 * @size: Number of identifiers
62 * Given a size for the ipc identifier range (limited below IPCMNI)
63 * set up the sequence range to use then allocate and initialise the
67 void __init ipc_init_ids(struct ipc_ids* ids, int size)
70 sema_init(&ids->sem,1);
78 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
79 if(seq_limit > USHRT_MAX)
80 ids->seq_max = USHRT_MAX;
82 ids->seq_max = seq_limit;
85 ids->entries = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*size +
86 sizeof(struct ipc_id_ary));
88 if(ids->entries == NULL) {
89 printk(KERN_ERR "ipc_init_ids() failed, ipc service disabled.\n");
91 ids->entries = &ids->nullentry;
93 ids->entries->size = size;
95 ids->entries->p[i] = NULL;
99 static struct file_operations sysvipc_proc_fops;
101 * ipc_init_proc_interface - Create a proc interface for sysipc types
102 * using a seq_file interface.
103 * @path: Path in procfs
104 * @header: Banner to be printed at the beginning of the file.
105 * @ids: ipc id table to iterate.
106 * @show: show routine.
108 void __init ipc_init_proc_interface(const char *path, const char *header,
110 int (*show)(struct seq_file *, void *))
112 struct proc_dir_entry *pde;
113 struct ipc_proc_iface *iface;
115 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
119 iface->header = header;
123 pde = create_proc_entry(path,
124 S_IRUGO, /* world readable */
125 NULL /* parent dir */);
128 pde->proc_fops = &sysvipc_proc_fops;
136 * ipc_findkey - find a key in an ipc identifier set
137 * @ids: Identifier set
138 * @key: The key to find
140 * Requires ipc_ids.sem locked.
141 * Returns the identifier if found or -1 if not.
144 int ipc_findkey(struct ipc_ids* ids, key_t key)
147 struct kern_ipc_perm* p;
148 int max_id = ids->max_id;
151 * rcu_dereference() is not needed here
152 * since ipc_ids.sem is held
154 for (id = 0; id <= max_id; id++) {
155 p = ids->entries->p[id];
158 if (!vx_check(p->xid, VX_IDENT))
167 * Requires ipc_ids.sem locked
169 static int grow_ary(struct ipc_ids* ids, int newsize)
171 struct ipc_id_ary* new;
172 struct ipc_id_ary* old;
174 int size = ids->entries->size;
181 new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*newsize +
182 sizeof(struct ipc_id_ary));
186 memcpy(new->p, ids->entries->p, sizeof(struct kern_ipc_perm *)*size);
187 for(i=size;i<newsize;i++) {
193 * Use rcu_assign_pointer() to make sure the memcpyed contents
194 * of the new array are visible before the new array becomes visible.
196 rcu_assign_pointer(ids->entries, new);
203 * ipc_addid - add an IPC identifier
204 * @ids: IPC identifier set
205 * @new: new IPC permission set
206 * @size: new size limit for the id array
208 * Add an entry 'new' to the IPC arrays. The permissions object is
209 * initialised and the first free entry is set up and the id assigned
210 * is returned. The list is returned in a locked state on success.
211 * On failure the list is not locked and -1 is returned.
213 * Called with ipc_ids.sem held.
216 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
220 size = grow_ary(ids,size);
223 * rcu_dereference()() is not needed here since
224 * ipc_ids.sem is held
226 for (id = 0; id < size; id++) {
227 if(ids->entries->p[id] == NULL)
233 if (id > ids->max_id)
236 new->cuid = new->uid = current->euid;
237 new->gid = new->cgid = current->egid;
239 new->seq = ids->seq++;
240 if(ids->seq > ids->seq_max)
243 spin_lock_init(&new->lock);
246 spin_lock(&new->lock);
247 ids->entries->p[id] = new;
252 * ipc_rmid - remove an IPC identifier
253 * @ids: identifier set
254 * @id: Identifier to remove
256 * The identifier must be valid, and in use. The kernel will panic if
257 * fed an invalid identifier. The entry is removed and internal
258 * variables recomputed. The object associated with the identifier
260 * ipc_ids.sem and the spinlock for this ID is hold before this function
261 * is called, and remain locked on the exit.
264 struct kern_ipc_perm* ipc_rmid(struct ipc_ids* ids, int id)
266 struct kern_ipc_perm* p;
267 int lid = id % SEQ_MULTIPLIER;
268 if(lid >= ids->entries->size)
272 * do not need a rcu_dereference()() here to force ordering
273 * on Alpha, since the ipc_ids.sem is held.
275 p = ids->entries->p[lid];
276 ids->entries->p[lid] = NULL;
281 if (lid == ids->max_id) {
286 } while (ids->entries->p[lid] == NULL);
294 * ipc_alloc - allocate ipc space
295 * @size: size desired
297 * Allocate memory from the appropriate pools and return a pointer to it.
298 * NULL is returned if the allocation fails
301 void* ipc_alloc(int size)
307 out = kmalloc(size, GFP_KERNEL);
312 * ipc_free - free ipc space
313 * @ptr: pointer returned by ipc_alloc
314 * @size: size of block
316 * Free a block created with ipc_alloc. The caller must know the size
317 * used in the allocation call.
320 void ipc_free(void* ptr, int size)
330 * There are three headers that are prepended to the actual allocation:
331 * - during use: ipc_rcu_hdr.
332 * - during the rcu grace period: ipc_rcu_grace.
333 * - [only if vmalloc]: ipc_rcu_sched.
334 * Their lifetime doesn't overlap, thus the headers share the same memory.
335 * Unlike a normal union, they are right-aligned, thus some container_of
336 * forward/backward casting is necessary:
349 /* "void *" makes sure alignment of following data is sane. */
355 struct work_struct work;
356 /* "void *" makes sure alignment of following data is sane. */
360 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
361 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
362 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
363 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
365 static inline int rcu_use_vmalloc(int size)
367 /* Too big for a single page? */
368 if (HDRLEN_KMALLOC + size > PAGE_SIZE)
374 * ipc_rcu_alloc - allocate ipc and rcu space
375 * @size: size desired
377 * Allocate memory for the rcu header structure + the object.
378 * Returns the pointer to the object.
379 * NULL is returned if the allocation fails.
382 void* ipc_rcu_alloc(int size)
386 * We prepend the allocation with the rcu struct, and
387 * workqueue if necessary (for vmalloc).
389 if (rcu_use_vmalloc(size)) {
390 out = vmalloc(HDRLEN_VMALLOC + size);
392 out += HDRLEN_VMALLOC;
393 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
394 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
397 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
399 out += HDRLEN_KMALLOC;
400 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
401 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
408 void ipc_rcu_getref(void *ptr)
410 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
414 * ipc_schedule_free - free ipc + rcu space
415 * @head: RCU callback structure for queued work
417 * Since RCU callback function is called in bh,
418 * we need to defer the vfree to schedule_work
420 static void ipc_schedule_free(struct rcu_head *head)
422 struct ipc_rcu_grace *grace =
423 container_of(head, struct ipc_rcu_grace, rcu);
424 struct ipc_rcu_sched *sched =
425 container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
427 INIT_WORK(&sched->work, vfree, sched);
428 schedule_work(&sched->work);
432 * ipc_immediate_free - free ipc + rcu space
433 * @head: RCU callback structure that contains pointer to be freed
435 * Free from the RCU callback context
437 static void ipc_immediate_free(struct rcu_head *head)
439 struct ipc_rcu_grace *free =
440 container_of(head, struct ipc_rcu_grace, rcu);
444 void ipc_rcu_putref(void *ptr)
446 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
449 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
450 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
453 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
459 * ipcperms - check IPC permissions
460 * @ipcp: IPC permission set
461 * @flag: desired permission set.
463 * Check user, group, other permissions for access
464 * to ipc resources. return 0 if allowed
467 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
468 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
469 int requested_mode, granted_mode;
471 if (!vx_check(ipcp->xid, VX_ADMIN|VX_IDENT)) /* maybe just VX_IDENT? */
473 requested_mode = (flag >> 6) | (flag >> 3) | flag;
474 granted_mode = ipcp->mode;
475 if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
477 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
479 /* is there some bit set in requested_mode but not in granted_mode? */
480 if ((requested_mode & ~granted_mode & 0007) &&
481 !capable(CAP_IPC_OWNER))
484 return security_ipc_permission(ipcp, flag);
488 * Functions to convert between the kern_ipc_perm structure and the
489 * old/new ipc_perm structures
493 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
494 * @in: kernel permissions
495 * @out: new style IPC permissions
497 * Turn the kernel object 'in' into a set of permissions descriptions
498 * for returning to userspace (out).
502 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
507 out->cuid = in->cuid;
508 out->cgid = in->cgid;
509 out->mode = in->mode;
514 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
515 * @in: new style IPC permissions
516 * @out: old style IPC permissions
518 * Turn the new style permissions object in into a compatibility
519 * object and store it into the 'out' pointer.
522 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
525 SET_UID(out->uid, in->uid);
526 SET_GID(out->gid, in->gid);
527 SET_UID(out->cuid, in->cuid);
528 SET_GID(out->cgid, in->cgid);
529 out->mode = in->mode;
534 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
535 * is called with shm_ids.sem locked. Since grow_ary() is also called with
536 * shm_ids.sem down(for Shared Memory), there is no need to add read
537 * barriers here to gurantee the writes in grow_ary() are seen in order
540 * However ipc_get() itself does not necessary require ipc_ids.sem down. So
541 * if in the future ipc_get() is used by other places without ipc_ids.sem
542 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
544 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id)
546 struct kern_ipc_perm* out;
547 int lid = id % SEQ_MULTIPLIER;
548 if(lid >= ids->entries->size)
550 out = ids->entries->p[lid];
554 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
556 struct kern_ipc_perm* out;
557 int lid = id % SEQ_MULTIPLIER;
558 struct ipc_id_ary* entries;
561 entries = rcu_dereference(ids->entries);
562 if(lid >= entries->size) {
566 out = entries->p[lid];
571 spin_lock(&out->lock);
573 /* ipc_rmid() may have already freed the ID while ipc_lock
574 * was spinning: here verify that the structure is still valid
577 spin_unlock(&out->lock);
584 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
587 spin_lock(&perm->lock);
590 void ipc_unlock(struct kern_ipc_perm* perm)
592 spin_unlock(&perm->lock);
596 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
598 return SEQ_MULTIPLIER*seq + id;
601 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
603 if(uid/SEQ_MULTIPLIER != ipcp->seq)
608 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
612 * ipc_parse_version - IPC call version
613 * @cmd: pointer to command
615 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
616 * The cmd value is turned from an encoding command and version into
617 * just the command code.
620 int ipc_parse_version (int *cmd)
630 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
632 #ifdef CONFIG_PROC_FS
633 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
635 struct ipc_proc_iface *iface = s->private;
636 struct kern_ipc_perm *ipc = it;
639 /* If we had an ipc id locked before, unlock it */
640 if (ipc && ipc != SEQ_START_TOKEN)
644 * p = *pos - 1 (because id 0 starts at position 1)
645 * + 1 (because we increment the position by one)
647 for (p = *pos; p <= iface->ids->max_id; p++) {
648 if ((ipc = ipc_lock(iface->ids, p)) != NULL) {
654 /* Out of range - return NULL to terminate iteration */
659 * File positions: pos 0 -> header, pos n -> ipc id + 1.
660 * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
662 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
664 struct ipc_proc_iface *iface = s->private;
665 struct kern_ipc_perm *ipc;
669 * Take the lock - this will be released by the corresponding
672 down(&iface->ids->sem);
674 /* pos < 0 is invalid */
678 /* pos == 0 means header */
680 return SEQ_START_TOKEN;
682 /* Find the (pos-1)th ipc */
683 for (p = *pos - 1; p <= iface->ids->max_id; p++) {
684 if ((ipc = ipc_lock(iface->ids, p)) != NULL) {
692 static void sysvipc_proc_stop(struct seq_file *s, void *it)
694 struct kern_ipc_perm *ipc = it;
695 struct ipc_proc_iface *iface = s->private;
697 /* If we had a locked segment, release it */
698 if (ipc && ipc != SEQ_START_TOKEN)
701 /* Release the lock we took in start() */
702 up(&iface->ids->sem);
705 static int sysvipc_proc_show(struct seq_file *s, void *it)
707 struct ipc_proc_iface *iface = s->private;
709 if (it == SEQ_START_TOKEN)
710 return seq_puts(s, iface->header);
712 return iface->show(s, it);
715 static struct seq_operations sysvipc_proc_seqops = {
716 .start = sysvipc_proc_start,
717 .stop = sysvipc_proc_stop,
718 .next = sysvipc_proc_next,
719 .show = sysvipc_proc_show,
722 static int sysvipc_proc_open(struct inode *inode, struct file *file) {
724 struct seq_file *seq;
726 ret = seq_open(file, &sysvipc_proc_seqops);
728 seq = file->private_data;
729 seq->private = PDE(inode)->data;
734 static struct file_operations sysvipc_proc_fops = {
735 .open = sysvipc_proc_open,
738 .release = seq_release,
740 #endif /* CONFIG_PROC_FS */