blob: fbf757064a328a11ea100b2975a6ef180430502c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/ipc/msg.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 *
5 * Removed all the remaining kerneld mess
6 * Catch the -EFAULT stuff properly
7 * Use GFP_KERNEL for messages as in 1.2
8 * Fixed up the unchecked user space derefs
9 * Copyright (C) 1998 Alan Cox & Andi Kleen
10 *
11 * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
12 *
13 * mostly rewritten, threaded and wake-one semantics added
14 * MSGMAX limit removed, sysctl's added
Christian Kujau624dffc2006-01-15 02:43:54 +010015 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016 */
17
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/config.h>
20#include <linux/slab.h>
21#include <linux/msg.h>
22#include <linux/spinlock.h>
23#include <linux/init.h>
24#include <linux/proc_fs.h>
25#include <linux/list.h>
26#include <linux/security.h>
27#include <linux/sched.h>
28#include <linux/syscalls.h>
29#include <linux/audit.h>
Mike Waychison19b49462005-09-06 15:17:10 -070030#include <linux/seq_file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/current.h>
32#include <asm/uaccess.h>
33#include "util.h"
34
35/* sysctl: */
36int msg_ctlmax = MSGMAX;
37int msg_ctlmnb = MSGMNB;
38int msg_ctlmni = MSGMNI;
39
40/* one msg_receiver structure for each sleeping receiver */
41struct msg_receiver {
42 struct list_head r_list;
43 struct task_struct* r_tsk;
44
45 int r_mode;
46 long r_msgtype;
47 long r_maxsize;
48
49 struct msg_msg* volatile r_msg;
50};
51
52/* one msg_sender for each sleeping sender */
53struct msg_sender {
54 struct list_head list;
55 struct task_struct* tsk;
56};
57
58#define SEARCH_ANY 1
59#define SEARCH_EQUAL 2
60#define SEARCH_NOTEQUAL 3
61#define SEARCH_LESSEQUAL 4
62
63static atomic_t msg_bytes = ATOMIC_INIT(0);
64static atomic_t msg_hdrs = ATOMIC_INIT(0);
65
66static struct ipc_ids msg_ids;
67
68#define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
69#define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
70#define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
71#define msg_checkid(msq, msgid) \
72 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
73#define msg_buildid(id, seq) \
74 ipc_buildid(&msg_ids, id, seq)
75
76static void freeque (struct msg_queue *msq, int id);
77static int newque (key_t key, int msgflg);
78#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -070079static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#endif
81
82void __init msg_init (void)
83{
84 ipc_init_ids(&msg_ids,msg_ctlmni);
Mike Waychison19b49462005-09-06 15:17:10 -070085 ipc_init_proc_interface("sysvipc/msg",
86 " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
87 &msg_ids,
88 sysvipc_msg_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089}
90
91static int newque (key_t key, int msgflg)
92{
93 int id;
94 int retval;
95 struct msg_queue *msq;
96
97 msq = ipc_rcu_alloc(sizeof(*msq));
98 if (!msq)
99 return -ENOMEM;
100
101 msq->q_perm.mode = (msgflg & S_IRWXUGO);
102 msq->q_perm.key = key;
103
104 msq->q_perm.security = NULL;
105 retval = security_msg_queue_alloc(msq);
106 if (retval) {
107 ipc_rcu_putref(msq);
108 return retval;
109 }
110
111 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
112 if(id == -1) {
113 security_msg_queue_free(msq);
114 ipc_rcu_putref(msq);
115 return -ENOSPC;
116 }
117
Mike Waychison19b49462005-09-06 15:17:10 -0700118 msq->q_id = msg_buildid(id,msq->q_perm.seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 msq->q_stime = msq->q_rtime = 0;
120 msq->q_ctime = get_seconds();
121 msq->q_cbytes = msq->q_qnum = 0;
122 msq->q_qbytes = msg_ctlmnb;
123 msq->q_lspid = msq->q_lrpid = 0;
124 INIT_LIST_HEAD(&msq->q_messages);
125 INIT_LIST_HEAD(&msq->q_receivers);
126 INIT_LIST_HEAD(&msq->q_senders);
127 msg_unlock(msq);
128
Mike Waychison19b49462005-09-06 15:17:10 -0700129 return msq->q_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130}
131
132static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss)
133{
134 mss->tsk=current;
135 current->state=TASK_INTERRUPTIBLE;
136 list_add_tail(&mss->list,&msq->q_senders);
137}
138
139static inline void ss_del(struct msg_sender* mss)
140{
141 if(mss->list.next != NULL)
142 list_del(&mss->list);
143}
144
145static void ss_wakeup(struct list_head* h, int kill)
146{
147 struct list_head *tmp;
148
149 tmp = h->next;
150 while (tmp != h) {
151 struct msg_sender* mss;
152
153 mss = list_entry(tmp,struct msg_sender,list);
154 tmp = tmp->next;
155 if(kill)
156 mss->list.next=NULL;
157 wake_up_process(mss->tsk);
158 }
159}
160
161static void expunge_all(struct msg_queue* msq, int res)
162{
163 struct list_head *tmp;
164
165 tmp = msq->q_receivers.next;
166 while (tmp != &msq->q_receivers) {
167 struct msg_receiver* msr;
168
169 msr = list_entry(tmp,struct msg_receiver,r_list);
170 tmp = tmp->next;
171 msr->r_msg = NULL;
172 wake_up_process(msr->r_tsk);
173 smp_mb();
174 msr->r_msg = ERR_PTR(res);
175 }
176}
177/*
178 * freeque() wakes up waiters on the sender and receiver waiting queue,
179 * removes the message queue from message queue ID
180 * array, and cleans up all the messages associated with this queue.
181 *
182 * msg_ids.sem and the spinlock for this message queue is hold
183 * before freeque() is called. msg_ids.sem remains locked on exit.
184 */
185static void freeque (struct msg_queue *msq, int id)
186{
187 struct list_head *tmp;
188
189 expunge_all(msq,-EIDRM);
190 ss_wakeup(&msq->q_senders,1);
191 msq = msg_rmid(id);
192 msg_unlock(msq);
193
194 tmp = msq->q_messages.next;
195 while(tmp != &msq->q_messages) {
196 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list);
197 tmp = tmp->next;
198 atomic_dec(&msg_hdrs);
199 free_msg(msg);
200 }
201 atomic_sub(msq->q_cbytes, &msg_bytes);
202 security_msg_queue_free(msq);
203 ipc_rcu_putref(msq);
204}
205
206asmlinkage long sys_msgget (key_t key, int msgflg)
207{
208 int id, ret = -EPERM;
209 struct msg_queue *msq;
210
211 down(&msg_ids.sem);
212 if (key == IPC_PRIVATE)
213 ret = newque(key, msgflg);
214 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
215 if (!(msgflg & IPC_CREAT))
216 ret = -ENOENT;
217 else
218 ret = newque(key, msgflg);
219 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
220 ret = -EEXIST;
221 } else {
222 msq = msg_lock(id);
223 if(msq==NULL)
224 BUG();
225 if (ipcperms(&msq->q_perm, msgflg))
226 ret = -EACCES;
227 else {
228 int qid = msg_buildid(id, msq->q_perm.seq);
229 ret = security_msg_queue_associate(msq, msgflg);
230 if (!ret)
231 ret = qid;
232 }
233 msg_unlock(msq);
234 }
235 up(&msg_ids.sem);
236 return ret;
237}
238
239static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
240{
241 switch(version) {
242 case IPC_64:
243 return copy_to_user (buf, in, sizeof(*in));
244 case IPC_OLD:
245 {
246 struct msqid_ds out;
247
248 memset(&out,0,sizeof(out));
249
250 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
251
252 out.msg_stime = in->msg_stime;
253 out.msg_rtime = in->msg_rtime;
254 out.msg_ctime = in->msg_ctime;
255
256 if(in->msg_cbytes > USHRT_MAX)
257 out.msg_cbytes = USHRT_MAX;
258 else
259 out.msg_cbytes = in->msg_cbytes;
260 out.msg_lcbytes = in->msg_cbytes;
261
262 if(in->msg_qnum > USHRT_MAX)
263 out.msg_qnum = USHRT_MAX;
264 else
265 out.msg_qnum = in->msg_qnum;
266
267 if(in->msg_qbytes > USHRT_MAX)
268 out.msg_qbytes = USHRT_MAX;
269 else
270 out.msg_qbytes = in->msg_qbytes;
271 out.msg_lqbytes = in->msg_qbytes;
272
273 out.msg_lspid = in->msg_lspid;
274 out.msg_lrpid = in->msg_lrpid;
275
276 return copy_to_user (buf, &out, sizeof(out));
277 }
278 default:
279 return -EINVAL;
280 }
281}
282
283struct msq_setbuf {
284 unsigned long qbytes;
285 uid_t uid;
286 gid_t gid;
287 mode_t mode;
288};
289
290static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
291{
292 switch(version) {
293 case IPC_64:
294 {
295 struct msqid64_ds tbuf;
296
297 if (copy_from_user (&tbuf, buf, sizeof (tbuf)))
298 return -EFAULT;
299
300 out->qbytes = tbuf.msg_qbytes;
301 out->uid = tbuf.msg_perm.uid;
302 out->gid = tbuf.msg_perm.gid;
303 out->mode = tbuf.msg_perm.mode;
304
305 return 0;
306 }
307 case IPC_OLD:
308 {
309 struct msqid_ds tbuf_old;
310
311 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old)))
312 return -EFAULT;
313
314 out->uid = tbuf_old.msg_perm.uid;
315 out->gid = tbuf_old.msg_perm.gid;
316 out->mode = tbuf_old.msg_perm.mode;
317
318 if(tbuf_old.msg_qbytes == 0)
319 out->qbytes = tbuf_old.msg_lqbytes;
320 else
321 out->qbytes = tbuf_old.msg_qbytes;
322
323 return 0;
324 }
325 default:
326 return -EINVAL;
327 }
328}
329
330asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf)
331{
332 int err, version;
333 struct msg_queue *msq;
334 struct msq_setbuf setbuf;
335 struct kern_ipc_perm *ipcp;
336
337 if (msqid < 0 || cmd < 0)
338 return -EINVAL;
339
340 version = ipc_parse_version(&cmd);
341
342 switch (cmd) {
343 case IPC_INFO:
344 case MSG_INFO:
345 {
346 struct msginfo msginfo;
347 int max_id;
348 if (!buf)
349 return -EFAULT;
350 /* We must not return kernel stack data.
351 * due to padding, it's not enough
352 * to set all member fields.
353 */
354
355 err = security_msg_queue_msgctl(NULL, cmd);
356 if (err)
357 return err;
358
359 memset(&msginfo,0,sizeof(msginfo));
360 msginfo.msgmni = msg_ctlmni;
361 msginfo.msgmax = msg_ctlmax;
362 msginfo.msgmnb = msg_ctlmnb;
363 msginfo.msgssz = MSGSSZ;
364 msginfo.msgseg = MSGSEG;
365 down(&msg_ids.sem);
366 if (cmd == MSG_INFO) {
367 msginfo.msgpool = msg_ids.in_use;
368 msginfo.msgmap = atomic_read(&msg_hdrs);
369 msginfo.msgtql = atomic_read(&msg_bytes);
370 } else {
371 msginfo.msgmap = MSGMAP;
372 msginfo.msgpool = MSGPOOL;
373 msginfo.msgtql = MSGTQL;
374 }
375 max_id = msg_ids.max_id;
376 up(&msg_ids.sem);
377 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo)))
378 return -EFAULT;
379 return (max_id < 0) ? 0: max_id;
380 }
381 case MSG_STAT:
382 case IPC_STAT:
383 {
384 struct msqid64_ds tbuf;
385 int success_return;
386 if (!buf)
387 return -EFAULT;
388 if(cmd == MSG_STAT && msqid >= msg_ids.entries->size)
389 return -EINVAL;
390
391 memset(&tbuf,0,sizeof(tbuf));
392
393 msq = msg_lock(msqid);
394 if (msq == NULL)
395 return -EINVAL;
396
397 if(cmd == MSG_STAT) {
398 success_return = msg_buildid(msqid, msq->q_perm.seq);
399 } else {
400 err = -EIDRM;
401 if (msg_checkid(msq,msqid))
402 goto out_unlock;
403 success_return = 0;
404 }
405 err = -EACCES;
406 if (ipcperms (&msq->q_perm, S_IRUGO))
407 goto out_unlock;
408
409 err = security_msg_queue_msgctl(msq, cmd);
410 if (err)
411 goto out_unlock;
412
413 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
414 tbuf.msg_stime = msq->q_stime;
415 tbuf.msg_rtime = msq->q_rtime;
416 tbuf.msg_ctime = msq->q_ctime;
417 tbuf.msg_cbytes = msq->q_cbytes;
418 tbuf.msg_qnum = msq->q_qnum;
419 tbuf.msg_qbytes = msq->q_qbytes;
420 tbuf.msg_lspid = msq->q_lspid;
421 tbuf.msg_lrpid = msq->q_lrpid;
422 msg_unlock(msq);
423 if (copy_msqid_to_user(buf, &tbuf, version))
424 return -EFAULT;
425 return success_return;
426 }
427 case IPC_SET:
428 if (!buf)
429 return -EFAULT;
430 if (copy_msqid_from_user (&setbuf, buf, version))
431 return -EFAULT;
432 if ((err = audit_ipc_perms(setbuf.qbytes, setbuf.uid, setbuf.gid, setbuf.mode)))
433 return err;
434 break;
435 case IPC_RMID:
436 break;
437 default:
438 return -EINVAL;
439 }
440
441 down(&msg_ids.sem);
442 msq = msg_lock(msqid);
443 err=-EINVAL;
444 if (msq == NULL)
445 goto out_up;
446
447 err = -EIDRM;
448 if (msg_checkid(msq,msqid))
449 goto out_unlock_up;
450 ipcp = &msq->q_perm;
451 err = -EPERM;
452 if (current->euid != ipcp->cuid &&
453 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
454 /* We _could_ check for CAP_CHOWN above, but we don't */
455 goto out_unlock_up;
456
457 err = security_msg_queue_msgctl(msq, cmd);
458 if (err)
459 goto out_unlock_up;
460
461 switch (cmd) {
462 case IPC_SET:
463 {
464 err = -EPERM;
465 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
466 goto out_unlock_up;
467
468 msq->q_qbytes = setbuf.qbytes;
469
470 ipcp->uid = setbuf.uid;
471 ipcp->gid = setbuf.gid;
472 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
473 (S_IRWXUGO & setbuf.mode);
474 msq->q_ctime = get_seconds();
475 /* sleeping receivers might be excluded by
476 * stricter permissions.
477 */
478 expunge_all(msq,-EAGAIN);
479 /* sleeping senders might be able to send
480 * due to a larger queue size.
481 */
482 ss_wakeup(&msq->q_senders,0);
483 msg_unlock(msq);
484 break;
485 }
486 case IPC_RMID:
487 freeque (msq, msqid);
488 break;
489 }
490 err = 0;
491out_up:
492 up(&msg_ids.sem);
493 return err;
494out_unlock_up:
495 msg_unlock(msq);
496 goto out_up;
497out_unlock:
498 msg_unlock(msq);
499 return err;
500}
501
502static int testmsg(struct msg_msg* msg,long type,int mode)
503{
504 switch(mode)
505 {
506 case SEARCH_ANY:
507 return 1;
508 case SEARCH_LESSEQUAL:
509 if(msg->m_type <=type)
510 return 1;
511 break;
512 case SEARCH_EQUAL:
513 if(msg->m_type == type)
514 return 1;
515 break;
516 case SEARCH_NOTEQUAL:
517 if(msg->m_type != type)
518 return 1;
519 break;
520 }
521 return 0;
522}
523
524static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg)
525{
526 struct list_head* tmp;
527
528 tmp = msq->q_receivers.next;
529 while (tmp != &msq->q_receivers) {
530 struct msg_receiver* msr;
531 msr = list_entry(tmp,struct msg_receiver,r_list);
532 tmp = tmp->next;
533 if(testmsg(msg,msr->r_msgtype,msr->r_mode) &&
534 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
535 list_del(&msr->r_list);
536 if(msr->r_maxsize < msg->m_ts) {
537 msr->r_msg = NULL;
538 wake_up_process(msr->r_tsk);
539 smp_mb();
540 msr->r_msg = ERR_PTR(-E2BIG);
541 } else {
542 msr->r_msg = NULL;
543 msq->q_lrpid = msr->r_tsk->pid;
544 msq->q_rtime = get_seconds();
545 wake_up_process(msr->r_tsk);
546 smp_mb();
547 msr->r_msg = msg;
548 return 1;
549 }
550 }
551 }
552 return 0;
553}
554
555asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
556{
557 struct msg_queue *msq;
558 struct msg_msg *msg;
559 long mtype;
560 int err;
561
562 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
563 return -EINVAL;
564 if (get_user(mtype, &msgp->mtype))
565 return -EFAULT;
566 if (mtype < 1)
567 return -EINVAL;
568
569 msg = load_msg(msgp->mtext, msgsz);
570 if(IS_ERR(msg))
571 return PTR_ERR(msg);
572
573 msg->m_type = mtype;
574 msg->m_ts = msgsz;
575
576 msq = msg_lock(msqid);
577 err=-EINVAL;
578 if(msq==NULL)
579 goto out_free;
580
581 err= -EIDRM;
582 if (msg_checkid(msq,msqid))
583 goto out_unlock_free;
584
585 for (;;) {
586 struct msg_sender s;
587
588 err=-EACCES;
589 if (ipcperms(&msq->q_perm, S_IWUGO))
590 goto out_unlock_free;
591
592 err = security_msg_queue_msgsnd(msq, msg, msgflg);
593 if (err)
594 goto out_unlock_free;
595
596 if(msgsz + msq->q_cbytes <= msq->q_qbytes &&
597 1 + msq->q_qnum <= msq->q_qbytes) {
598 break;
599 }
600
601 /* queue full, wait: */
602 if(msgflg&IPC_NOWAIT) {
603 err=-EAGAIN;
604 goto out_unlock_free;
605 }
606 ss_add(msq, &s);
607 ipc_rcu_getref(msq);
608 msg_unlock(msq);
609 schedule();
610
611 ipc_lock_by_ptr(&msq->q_perm);
612 ipc_rcu_putref(msq);
613 if (msq->q_perm.deleted) {
614 err = -EIDRM;
615 goto out_unlock_free;
616 }
617 ss_del(&s);
618
619 if (signal_pending(current)) {
620 err=-ERESTARTNOHAND;
621 goto out_unlock_free;
622 }
623 }
624
625 msq->q_lspid = current->tgid;
626 msq->q_stime = get_seconds();
627
628 if(!pipelined_send(msq,msg)) {
629 /* noone is waiting for this message, enqueue it */
630 list_add_tail(&msg->m_list,&msq->q_messages);
631 msq->q_cbytes += msgsz;
632 msq->q_qnum++;
633 atomic_add(msgsz,&msg_bytes);
634 atomic_inc(&msg_hdrs);
635 }
636
637 err = 0;
638 msg = NULL;
639
640out_unlock_free:
641 msg_unlock(msq);
642out_free:
643 if(msg!=NULL)
644 free_msg(msg);
645 return err;
646}
647
648static inline int convert_mode(long* msgtyp, int msgflg)
649{
650 /*
651 * find message of correct type.
652 * msgtyp = 0 => get first.
653 * msgtyp > 0 => get first message of matching type.
654 * msgtyp < 0 => get message with least type must be < abs(msgtype).
655 */
656 if(*msgtyp==0)
657 return SEARCH_ANY;
658 if(*msgtyp<0) {
659 *msgtyp=-(*msgtyp);
660 return SEARCH_LESSEQUAL;
661 }
662 if(msgflg & MSG_EXCEPT)
663 return SEARCH_NOTEQUAL;
664 return SEARCH_EQUAL;
665}
666
667asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz,
668 long msgtyp, int msgflg)
669{
670 struct msg_queue *msq;
671 struct msg_msg *msg;
672 int mode;
673
674 if (msqid < 0 || (long) msgsz < 0)
675 return -EINVAL;
676 mode = convert_mode(&msgtyp,msgflg);
677
678 msq = msg_lock(msqid);
679 if(msq==NULL)
680 return -EINVAL;
681
682 msg = ERR_PTR(-EIDRM);
683 if (msg_checkid(msq,msqid))
684 goto out_unlock;
685
686 for (;;) {
687 struct msg_receiver msr_d;
688 struct list_head* tmp;
689
690 msg = ERR_PTR(-EACCES);
691 if (ipcperms (&msq->q_perm, S_IRUGO))
692 goto out_unlock;
693
694 msg = ERR_PTR(-EAGAIN);
695 tmp = msq->q_messages.next;
696 while (tmp != &msq->q_messages) {
697 struct msg_msg *walk_msg;
698 walk_msg = list_entry(tmp,struct msg_msg,m_list);
699 if(testmsg(walk_msg,msgtyp,mode) &&
700 !security_msg_queue_msgrcv(msq, walk_msg, current, msgtyp, mode)) {
701 msg = walk_msg;
702 if(mode == SEARCH_LESSEQUAL && walk_msg->m_type != 1) {
703 msg=walk_msg;
704 msgtyp=walk_msg->m_type-1;
705 } else {
706 msg=walk_msg;
707 break;
708 }
709 }
710 tmp = tmp->next;
711 }
712 if(!IS_ERR(msg)) {
713 /* Found a suitable message. Unlink it from the queue. */
714 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
715 msg = ERR_PTR(-E2BIG);
716 goto out_unlock;
717 }
718 list_del(&msg->m_list);
719 msq->q_qnum--;
720 msq->q_rtime = get_seconds();
721 msq->q_lrpid = current->tgid;
722 msq->q_cbytes -= msg->m_ts;
723 atomic_sub(msg->m_ts,&msg_bytes);
724 atomic_dec(&msg_hdrs);
725 ss_wakeup(&msq->q_senders,0);
726 msg_unlock(msq);
727 break;
728 }
729 /* No message waiting. Wait for a message */
730 if (msgflg & IPC_NOWAIT) {
731 msg = ERR_PTR(-ENOMSG);
732 goto out_unlock;
733 }
734 list_add_tail(&msr_d.r_list,&msq->q_receivers);
735 msr_d.r_tsk = current;
736 msr_d.r_msgtype = msgtyp;
737 msr_d.r_mode = mode;
738 if(msgflg & MSG_NOERROR)
739 msr_d.r_maxsize = INT_MAX;
740 else
741 msr_d.r_maxsize = msgsz;
742 msr_d.r_msg = ERR_PTR(-EAGAIN);
743 current->state = TASK_INTERRUPTIBLE;
744 msg_unlock(msq);
745
746 schedule();
747
748 /* Lockless receive, part 1:
749 * Disable preemption. We don't hold a reference to the queue
750 * and getting a reference would defeat the idea of a lockless
751 * operation, thus the code relies on rcu to guarantee the
752 * existance of msq:
753 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
754 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
755 * rcu_read_lock() prevents preemption between reading r_msg
756 * and the spin_lock() inside ipc_lock_by_ptr().
757 */
758 rcu_read_lock();
759
760 /* Lockless receive, part 2:
761 * Wait until pipelined_send or expunge_all are outside of
762 * wake_up_process(). There is a race with exit(), see
763 * ipc/mqueue.c for the details.
764 */
765 msg = (struct msg_msg*) msr_d.r_msg;
766 while (msg == NULL) {
767 cpu_relax();
768 msg = (struct msg_msg*) msr_d.r_msg;
769 }
770
771 /* Lockless receive, part 3:
772 * If there is a message or an error then accept it without
773 * locking.
774 */
775 if(msg != ERR_PTR(-EAGAIN)) {
776 rcu_read_unlock();
777 break;
778 }
779
780 /* Lockless receive, part 3:
781 * Acquire the queue spinlock.
782 */
783 ipc_lock_by_ptr(&msq->q_perm);
784 rcu_read_unlock();
785
786 /* Lockless receive, part 4:
787 * Repeat test after acquiring the spinlock.
788 */
789 msg = (struct msg_msg*)msr_d.r_msg;
790 if(msg != ERR_PTR(-EAGAIN))
791 goto out_unlock;
792
793 list_del(&msr_d.r_list);
794 if (signal_pending(current)) {
795 msg = ERR_PTR(-ERESTARTNOHAND);
796out_unlock:
797 msg_unlock(msq);
798 break;
799 }
800 }
801 if (IS_ERR(msg))
802 return PTR_ERR(msg);
803
804 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
805 if (put_user (msg->m_type, &msgp->mtype) ||
806 store_msg(msgp->mtext, msg, msgsz)) {
807 msgsz = -EFAULT;
808 }
809 free_msg(msg);
810 return msgsz;
811}
812
813#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -0700814static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Mike Waychison19b49462005-09-06 15:17:10 -0700816 struct msg_queue *msq = it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Mike Waychison19b49462005-09-06 15:17:10 -0700818 return seq_printf(s,
819 "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
820 msq->q_perm.key,
821 msq->q_id,
822 msq->q_perm.mode,
823 msq->q_cbytes,
824 msq->q_qnum,
825 msq->q_lspid,
826 msq->q_lrpid,
827 msq->q_perm.uid,
828 msq->q_perm.gid,
829 msq->q_perm.cuid,
830 msq->q_perm.cgid,
831 msq->q_stime,
832 msq->q_rtime,
833 msq->q_ctime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
835#endif