projects/linux·blob v2.6.13 · refs

862 lines · 18.9 KiB

1/*
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 <[email protected]>
12 *
13 * mostly rewritten, threaded and wake-one semantics added
14 * MSGMAX limit removed, sysctl's added
15 * (c) 1999 Manfred Spraul <[email protected]>
16 */
17
18#include <linux/config.h>
19#include <linux/slab.h>
20#include <linux/msg.h>
21#include <linux/spinlock.h>
22#include <linux/init.h>
23#include <linux/proc_fs.h>
24#include <linux/list.h>
25#include <linux/security.h>
26#include <linux/sched.h>
27#include <linux/syscalls.h>
28#include <linux/audit.h>
29#include <asm/current.h>
30#include <asm/uaccess.h>
31#include "util.h"
32
33/* sysctl: */
34int msg_ctlmax = MSGMAX;
35int msg_ctlmnb = MSGMNB;
36int msg_ctlmni = MSGMNI;
37
38/* one msg_receiver structure for each sleeping receiver */
39struct msg_receiver {
40 struct list_head r_list;
41 struct task_struct* r_tsk;
42
43 int r_mode;
44 long r_msgtype;
45 long r_maxsize;
46
47 struct msg_msg* volatile r_msg;
48};
49
50/* one msg_sender for each sleeping sender */
51struct msg_sender {
52 struct list_head list;
53 struct task_struct* tsk;
54};
55
56#define SEARCH_ANY 1
57#define SEARCH_EQUAL 2
58#define SEARCH_NOTEQUAL 3
59#define SEARCH_LESSEQUAL 4
60
61static atomic_t msg_bytes = ATOMIC_INIT(0);
62static atomic_t msg_hdrs = ATOMIC_INIT(0);
63
64static struct ipc_ids msg_ids;
65
66#define msg_lock(id) ((struct msg_queue*)ipc_lock(&msg_ids,id))
67#define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
68#define msg_rmid(id) ((struct msg_queue*)ipc_rmid(&msg_ids,id))
69#define msg_checkid(msq, msgid) \
70 ipc_checkid(&msg_ids,&msq->q_perm,msgid)
71#define msg_buildid(id, seq) \
72 ipc_buildid(&msg_ids, id, seq)
73
74static void freeque (struct msg_queue *msq, int id);
75static int newque (key_t key, int msgflg);
76#ifdef CONFIG_PROC_FS
77static int sysvipc_msg_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
78#endif
79
80void __init msg_init (void)
81{
82 ipc_init_ids(&msg_ids,msg_ctlmni);
83
84#ifdef CONFIG_PROC_FS
85 create_proc_read_entry("sysvipc/msg", 0, NULL, sysvipc_msg_read_proc, NULL);
86#endif
87}
88
89static int newque (key_t key, int msgflg)
90{
91 int id;
92 int retval;
93 struct msg_queue *msq;
94
95 msq = ipc_rcu_alloc(sizeof(*msq));
96 if (!msq)
97 return -ENOMEM;
98
99 msq->q_perm.mode = (msgflg & S_IRWXUGO);
100 msq->q_perm.key = key;
101
102 msq->q_perm.security = NULL;
103 retval = security_msg_queue_alloc(msq);
104 if (retval) {
105 ipc_rcu_putref(msq);
106 return retval;
107 }
108
109 id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
110 if(id == -1) {
111 security_msg_queue_free(msq);
112 ipc_rcu_putref(msq);
113 return -ENOSPC;
114 }
115
116 msq->q_stime = msq->q_rtime = 0;
117 msq->q_ctime = get_seconds();
118 msq->q_cbytes = msq->q_qnum = 0;
119 msq->q_qbytes = msg_ctlmnb;
120 msq->q_lspid = msq->q_lrpid = 0;
121 INIT_LIST_HEAD(&msq->q_messages);
122 INIT_LIST_HEAD(&msq->q_receivers);
123 INIT_LIST_HEAD(&msq->q_senders);
124 msg_unlock(msq);
125
126 return msg_buildid(id,msq->q_perm.seq);
127}
128
129static inline void ss_add(struct msg_queue* msq, struct msg_sender* mss)
130{
131 mss->tsk=current;
132 current->state=TASK_INTERRUPTIBLE;
133 list_add_tail(&mss->list,&msq->q_senders);
134}
135
136static inline void ss_del(struct msg_sender* mss)
137{
138 if(mss->list.next != NULL)
139 list_del(&mss->list);
140}
141
142static void ss_wakeup(struct list_head* h, int kill)
143{
144 struct list_head *tmp;
145
146 tmp = h->next;
147 while (tmp != h) {
148 struct msg_sender* mss;
149
150 mss = list_entry(tmp,struct msg_sender,list);
151 tmp = tmp->next;
152 if(kill)
153 mss->list.next=NULL;
154 wake_up_process(mss->tsk);
155 }
156}
157
158static void expunge_all(struct msg_queue* msq, int res)
159{
160 struct list_head *tmp;
161
162 tmp = msq->q_receivers.next;
163 while (tmp != &msq->q_receivers) {
164 struct msg_receiver* msr;
165
166 msr = list_entry(tmp,struct msg_receiver,r_list);
167 tmp = tmp->next;
168 msr->r_msg = NULL;
169 wake_up_process(msr->r_tsk);
170 smp_mb();
171 msr->r_msg = ERR_PTR(res);
172 }
173}
174/*
175 * freeque() wakes up waiters on the sender and receiver waiting queue,
176 * removes the message queue from message queue ID
177 * array, and cleans up all the messages associated with this queue.
178 *
179 * msg_ids.sem and the spinlock for this message queue is hold
180 * before freeque() is called. msg_ids.sem remains locked on exit.
181 */
182static void freeque (struct msg_queue *msq, int id)
183{
184 struct list_head *tmp;
185
186 expunge_all(msq,-EIDRM);
187 ss_wakeup(&msq->q_senders,1);
188 msq = msg_rmid(id);
189 msg_unlock(msq);
190
191 tmp = msq->q_messages.next;
192 while(tmp != &msq->q_messages) {
193 struct msg_msg* msg = list_entry(tmp,struct msg_msg,m_list);
194 tmp = tmp->next;
195 atomic_dec(&msg_hdrs);
196 free_msg(msg);
197 }
198 atomic_sub(msq->q_cbytes, &msg_bytes);
199 security_msg_queue_free(msq);
200 ipc_rcu_putref(msq);
201}
202
203asmlinkage long sys_msgget (key_t key, int msgflg)
204{
205 int id, ret = -EPERM;
206 struct msg_queue *msq;
207
208 down(&msg_ids.sem);
209 if (key == IPC_PRIVATE)
210 ret = newque(key, msgflg);
211 else if ((id = ipc_findkey(&msg_ids, key)) == -1) { /* key not used */
212 if (!(msgflg & IPC_CREAT))
213 ret = -ENOENT;
214 else
215 ret = newque(key, msgflg);
216 } else if (msgflg & IPC_CREAT && msgflg & IPC_EXCL) {
217 ret = -EEXIST;
218 } else {
219 msq = msg_lock(id);
220 if(msq==NULL)
221 BUG();
222 if (ipcperms(&msq->q_perm, msgflg))
223 ret = -EACCES;
224 else {
225 int qid = msg_buildid(id, msq->q_perm.seq);
226 ret = security_msg_queue_associate(msq, msgflg);
227 if (!ret)
228 ret = qid;
229 }
230 msg_unlock(msq);
231 }
232 up(&msg_ids.sem);
233 return ret;
234}
235
236static inline unsigned long copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
237{
238 switch(version) {
239 case IPC_64:
240 return copy_to_user (buf, in, sizeof(*in));
241 case IPC_OLD:
242 {
243 struct msqid_ds out;
244
245 memset(&out,0,sizeof(out));
246
247 ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
248
249 out.msg_stime = in->msg_stime;
250 out.msg_rtime = in->msg_rtime;
251 out.msg_ctime = in->msg_ctime;
252
253 if(in->msg_cbytes > USHRT_MAX)
254 out.msg_cbytes = USHRT_MAX;
255 else
256 out.msg_cbytes = in->msg_cbytes;
257 out.msg_lcbytes = in->msg_cbytes;
258
259 if(in->msg_qnum > USHRT_MAX)
260 out.msg_qnum = USHRT_MAX;
261 else
262 out.msg_qnum = in->msg_qnum;
263
264 if(in->msg_qbytes > USHRT_MAX)
265 out.msg_qbytes = USHRT_MAX;
266 else
267 out.msg_qbytes = in->msg_qbytes;
268 out.msg_lqbytes = in->msg_qbytes;
269
270 out.msg_lspid = in->msg_lspid;
271 out.msg_lrpid = in->msg_lrpid;
272
273 return copy_to_user (buf, &out, sizeof(out));
274 }
275 default:
276 return -EINVAL;
277 }
278}
279
280struct msq_setbuf {
281 unsigned long qbytes;
282 uid_t uid;
283 gid_t gid;
284 mode_t mode;
285};
286
287static inline unsigned long copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version)
288{
289 switch(version) {
290 case IPC_64:
291 {
292 struct msqid64_ds tbuf;
293
294 if (copy_from_user (&tbuf, buf, sizeof (tbuf)))
295 return -EFAULT;
296
297 out->qbytes = tbuf.msg_qbytes;
298 out->uid = tbuf.msg_perm.uid;
299 out->gid = tbuf.msg_perm.gid;
300 out->mode = tbuf.msg_perm.mode;
301
302 return 0;
303 }
304 case IPC_OLD:
305 {
306 struct msqid_ds tbuf_old;
307
308 if (copy_from_user (&tbuf_old, buf, sizeof (tbuf_old)))
309 return -EFAULT;
310
311 out->uid = tbuf_old.msg_perm.uid;
312 out->gid = tbuf_old.msg_perm.gid;
313 out->mode = tbuf_old.msg_perm.mode;
314
315 if(tbuf_old.msg_qbytes == 0)
316 out->qbytes = tbuf_old.msg_lqbytes;
317 else
318 out->qbytes = tbuf_old.msg_qbytes;
319
320 return 0;
321 }
322 default:
323 return -EINVAL;
324 }
325}
326
327asmlinkage long sys_msgctl (int msqid, int cmd, struct msqid_ds __user *buf)
328{
329 int err, version;
330 struct msg_queue *msq;
331 struct msq_setbuf setbuf;
332 struct kern_ipc_perm *ipcp;
333
334 if (msqid < 0 || cmd < 0)
335 return -EINVAL;
336
337 version = ipc_parse_version(&cmd);
338
339 switch (cmd) {
340 case IPC_INFO:
341 case MSG_INFO:
342 {
343 struct msginfo msginfo;
344 int max_id;
345 if (!buf)
346 return -EFAULT;
347 /* We must not return kernel stack data.
348 * due to padding, it's not enough
349 * to set all member fields.
350 */
351
352 err = security_msg_queue_msgctl(NULL, cmd);
353 if (err)
354 return err;
355
356 memset(&msginfo,0,sizeof(msginfo));
357 msginfo.msgmni = msg_ctlmni;
358 msginfo.msgmax = msg_ctlmax;
359 msginfo.msgmnb = msg_ctlmnb;
360 msginfo.msgssz = MSGSSZ;
361 msginfo.msgseg = MSGSEG;
362 down(&msg_ids.sem);
363 if (cmd == MSG_INFO) {
364 msginfo.msgpool = msg_ids.in_use;
365 msginfo.msgmap = atomic_read(&msg_hdrs);
366 msginfo.msgtql = atomic_read(&msg_bytes);
367 } else {
368 msginfo.msgmap = MSGMAP;
369 msginfo.msgpool = MSGPOOL;
370 msginfo.msgtql = MSGTQL;
371 }
372 max_id = msg_ids.max_id;
373 up(&msg_ids.sem);
374 if (copy_to_user (buf, &msginfo, sizeof(struct msginfo)))
375 return -EFAULT;
376 return (max_id < 0) ? 0: max_id;
377 }
378 case MSG_STAT:
379 case IPC_STAT:
380 {
381 struct msqid64_ds tbuf;
382 int success_return;
383 if (!buf)
384 return -EFAULT;
385 if(cmd == MSG_STAT && msqid >= msg_ids.entries->size)
386 return -EINVAL;
387
388 memset(&tbuf,0,sizeof(tbuf));
389
390 msq = msg_lock(msqid);
391 if (msq == NULL)
392 return -EINVAL;
393
394 if(cmd == MSG_STAT) {
395 success_return = msg_buildid(msqid, msq->q_perm.seq);
396 } else {
397 err = -EIDRM;
398 if (msg_checkid(msq,msqid))
399 goto out_unlock;
400 success_return = 0;
401 }
402 err = -EACCES;
403 if (ipcperms (&msq->q_perm, S_IRUGO))
404 goto out_unlock;
405
406 err = security_msg_queue_msgctl(msq, cmd);
407 if (err)
408 goto out_unlock;
409
410 kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
411 tbuf.msg_stime = msq->q_stime;
412 tbuf.msg_rtime = msq->q_rtime;
413 tbuf.msg_ctime = msq->q_ctime;
414 tbuf.msg_cbytes = msq->q_cbytes;
415 tbuf.msg_qnum = msq->q_qnum;
416 tbuf.msg_qbytes = msq->q_qbytes;
417 tbuf.msg_lspid = msq->q_lspid;
418 tbuf.msg_lrpid = msq->q_lrpid;
419 msg_unlock(msq);
420 if (copy_msqid_to_user(buf, &tbuf, version))
421 return -EFAULT;
422 return success_return;
423 }
424 case IPC_SET:
425 if (!buf)
426 return -EFAULT;
427 if (copy_msqid_from_user (&setbuf, buf, version))
428 return -EFAULT;
429 if ((err = audit_ipc_perms(setbuf.qbytes, setbuf.uid, setbuf.gid, setbuf.mode)))
430 return err;
431 break;
432 case IPC_RMID:
433 break;
434 default:
435 return -EINVAL;
436 }
437
438 down(&msg_ids.sem);
439 msq = msg_lock(msqid);
440 err=-EINVAL;
441 if (msq == NULL)
442 goto out_up;
443
444 err = -EIDRM;
445 if (msg_checkid(msq,msqid))
446 goto out_unlock_up;
447 ipcp = &msq->q_perm;
448 err = -EPERM;
449 if (current->euid != ipcp->cuid &&
450 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN))
451 /* We _could_ check for CAP_CHOWN above, but we don't */
452 goto out_unlock_up;
453
454 err = security_msg_queue_msgctl(msq, cmd);
455 if (err)
456 goto out_unlock_up;
457
458 switch (cmd) {
459 case IPC_SET:
460 {
461 err = -EPERM;
462 if (setbuf.qbytes > msg_ctlmnb && !capable(CAP_SYS_RESOURCE))
463 goto out_unlock_up;
464
465 msq->q_qbytes = setbuf.qbytes;
466
467 ipcp->uid = setbuf.uid;
468 ipcp->gid = setbuf.gid;
469 ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
470 (S_IRWXUGO & setbuf.mode);
471 msq->q_ctime = get_seconds();
472 /* sleeping receivers might be excluded by
473 * stricter permissions.
474 */
475 expunge_all(msq,-EAGAIN);
476 /* sleeping senders might be able to send
477 * due to a larger queue size.
478 */
479 ss_wakeup(&msq->q_senders,0);
480 msg_unlock(msq);
481 break;
482 }
483 case IPC_RMID:
484 freeque (msq, msqid);
485 break;
486 }
487 err = 0;
488out_up:
489 up(&msg_ids.sem);
490 return err;
491out_unlock_up:
492 msg_unlock(msq);
493 goto out_up;
494out_unlock:
495 msg_unlock(msq);
496 return err;
497}
498
499static int testmsg(struct msg_msg* msg,long type,int mode)
500{
501 switch(mode)
502 {
503 case SEARCH_ANY:
504 return 1;
505 case SEARCH_LESSEQUAL:
506 if(msg->m_type <=type)
507 return 1;
508 break;
509 case SEARCH_EQUAL:
510 if(msg->m_type == type)
511 return 1;
512 break;
513 case SEARCH_NOTEQUAL:
514 if(msg->m_type != type)
515 return 1;
516 break;
517 }
518 return 0;
519}
520
521static inline int pipelined_send(struct msg_queue* msq, struct msg_msg* msg)
522{
523 struct list_head* tmp;
524
525 tmp = msq->q_receivers.next;
526 while (tmp != &msq->q_receivers) {
527 struct msg_receiver* msr;
528 msr = list_entry(tmp,struct msg_receiver,r_list);
529 tmp = tmp->next;
530 if(testmsg(msg,msr->r_msgtype,msr->r_mode) &&
531 !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
532 list_del(&msr->r_list);
533 if(msr->r_maxsize < msg->m_ts) {
534 msr->r_msg = NULL;
535 wake_up_process(msr->r_tsk);
536 smp_mb();
537 msr->r_msg = ERR_PTR(-E2BIG);
538 } else {
539 msr->r_msg = NULL;
540 msq->q_lrpid = msr->r_tsk->pid;
541 msq->q_rtime = get_seconds();
542 wake_up_process(msr->r_tsk);
543 smp_mb();
544 msr->r_msg = msg;
545 return 1;
546 }
547 }
548 }
549 return 0;
550}
551
552asmlinkage long sys_msgsnd (int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
553{
554 struct msg_queue *msq;
555 struct msg_msg *msg;
556 long mtype;
557 int err;
558
559 if (msgsz > msg_ctlmax || (long) msgsz < 0 || msqid < 0)
560 return -EINVAL;
561 if (get_user(mtype, &msgp->mtype))
562 return -EFAULT;
563 if (mtype < 1)
564 return -EINVAL;
565
566 msg = load_msg(msgp->mtext, msgsz);
567 if(IS_ERR(msg))
568 return PTR_ERR(msg);
569
570 msg->m_type = mtype;
571 msg->m_ts = msgsz;
572
573 msq = msg_lock(msqid);
574 err=-EINVAL;
575 if(msq==NULL)
576 goto out_free;
577
578 err= -EIDRM;
579 if (msg_checkid(msq,msqid))
580 goto out_unlock_free;
581
582 for (;;) {
583 struct msg_sender s;
584
585 err=-EACCES;
586 if (ipcperms(&msq->q_perm, S_IWUGO))
587 goto out_unlock_free;
588
589 err = security_msg_queue_msgsnd(msq, msg, msgflg);
590 if (err)
591 goto out_unlock_free;
592
593 if(msgsz + msq->q_cbytes <= msq->q_qbytes &&
594 1 + msq->q_qnum <= msq->q_qbytes) {
595 break;
596 }
597
598 /* queue full, wait: */
599 if(msgflg&IPC_NOWAIT) {
600 err=-EAGAIN;
601 goto out_unlock_free;
602 }
603 ss_add(msq, &s);
604 ipc_rcu_getref(msq);
605 msg_unlock(msq);
606 schedule();
607
608 ipc_lock_by_ptr(&msq->q_perm);
609 ipc_rcu_putref(msq);
610 if (msq->q_perm.deleted) {
611 err = -EIDRM;
612 goto out_unlock_free;
613 }
614 ss_del(&s);
615
616 if (signal_pending(current)) {
617 err=-ERESTARTNOHAND;
618 goto out_unlock_free;
619 }
620 }
621
622 msq->q_lspid = current->tgid;
623 msq->q_stime = get_seconds();
624
625 if(!pipelined_send(msq,msg)) {
626 /* noone is waiting for this message, enqueue it */
627 list_add_tail(&msg->m_list,&msq->q_messages);
628 msq->q_cbytes += msgsz;
629 msq->q_qnum++;
630 atomic_add(msgsz,&msg_bytes);
631 atomic_inc(&msg_hdrs);
632 }
633
634 err = 0;
635 msg = NULL;
636
637out_unlock_free:
638 msg_unlock(msq);
639out_free:
640 if(msg!=NULL)
641 free_msg(msg);
642 return err;
643}
644
645static inline int convert_mode(long* msgtyp, int msgflg)
646{
647 /*
648 * find message of correct type.
649 * msgtyp = 0 => get first.
650 * msgtyp > 0 => get first message of matching type.
651 * msgtyp < 0 => get message with least type must be < abs(msgtype).
652 */
653 if(*msgtyp==0)
654 return SEARCH_ANY;
655 if(*msgtyp<0) {
656 *msgtyp=-(*msgtyp);
657 return SEARCH_LESSEQUAL;
658 }
659 if(msgflg & MSG_EXCEPT)
660 return SEARCH_NOTEQUAL;
661 return SEARCH_EQUAL;
662}
663
664asmlinkage long sys_msgrcv (int msqid, struct msgbuf __user *msgp, size_t msgsz,
665 long msgtyp, int msgflg)
666{
667 struct msg_queue *msq;
668 struct msg_msg *msg;
669 int mode;
670
671 if (msqid < 0 || (long) msgsz < 0)
672 return -EINVAL;
673 mode = convert_mode(&msgtyp,msgflg);
674
675 msq = msg_lock(msqid);
676 if(msq==NULL)
677 return -EINVAL;
678
679 msg = ERR_PTR(-EIDRM);
680 if (msg_checkid(msq,msqid))
681 goto out_unlock;
682
683 for (;;) {
684 struct msg_receiver msr_d;
685 struct list_head* tmp;
686
687 msg = ERR_PTR(-EACCES);
688 if (ipcperms (&msq->q_perm, S_IRUGO))
689 goto out_unlock;
690
691 msg = ERR_PTR(-EAGAIN);
692 tmp = msq->q_messages.next;
693 while (tmp != &msq->q_messages) {
694 struct msg_msg *walk_msg;
695 walk_msg = list_entry(tmp,struct msg_msg,m_list);
696 if(testmsg(walk_msg,msgtyp,mode) &&
697 !security_msg_queue_msgrcv(msq, walk_msg, current, msgtyp, mode)) {
698 msg = walk_msg;
699 if(mode == SEARCH_LESSEQUAL && walk_msg->m_type != 1) {
700 msg=walk_msg;
701 msgtyp=walk_msg->m_type-1;
702 } else {
703 msg=walk_msg;
704 break;
705 }
706 }
707 tmp = tmp->next;
708 }
709 if(!IS_ERR(msg)) {
710 /* Found a suitable message. Unlink it from the queue. */
711 if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
712 msg = ERR_PTR(-E2BIG);
713 goto out_unlock;
714 }
715 list_del(&msg->m_list);
716 msq->q_qnum--;
717 msq->q_rtime = get_seconds();
718 msq->q_lrpid = current->tgid;
719 msq->q_cbytes -= msg->m_ts;
720 atomic_sub(msg->m_ts,&msg_bytes);
721 atomic_dec(&msg_hdrs);
722 ss_wakeup(&msq->q_senders,0);
723 msg_unlock(msq);
724 break;
725 }
726 /* No message waiting. Wait for a message */
727 if (msgflg & IPC_NOWAIT) {
728 msg = ERR_PTR(-ENOMSG);
729 goto out_unlock;
730 }
731 list_add_tail(&msr_d.r_list,&msq->q_receivers);
732 msr_d.r_tsk = current;
733 msr_d.r_msgtype = msgtyp;
734 msr_d.r_mode = mode;
735 if(msgflg & MSG_NOERROR)
736 msr_d.r_maxsize = INT_MAX;
737 else
738 msr_d.r_maxsize = msgsz;
739 msr_d.r_msg = ERR_PTR(-EAGAIN);
740 current->state = TASK_INTERRUPTIBLE;
741 msg_unlock(msq);
742
743 schedule();
744
745 /* Lockless receive, part 1:
746 * Disable preemption. We don't hold a reference to the queue
747 * and getting a reference would defeat the idea of a lockless
748 * operation, thus the code relies on rcu to guarantee the
749 * existance of msq:
750 * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
751 * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
752 * rcu_read_lock() prevents preemption between reading r_msg
753 * and the spin_lock() inside ipc_lock_by_ptr().
754 */
755 rcu_read_lock();
756
757 /* Lockless receive, part 2:
758 * Wait until pipelined_send or expunge_all are outside of
759 * wake_up_process(). There is a race with exit(), see
760 * ipc/mqueue.c for the details.
761 */
762 msg = (struct msg_msg*) msr_d.r_msg;
763 while (msg == NULL) {
764 cpu_relax();
765 msg = (struct msg_msg*) msr_d.r_msg;
766 }
767
768 /* Lockless receive, part 3:
769 * If there is a message or an error then accept it without
770 * locking.
771 */
772 if(msg != ERR_PTR(-EAGAIN)) {
773 rcu_read_unlock();
774 break;
775 }
776
777 /* Lockless receive, part 3:
778 * Acquire the queue spinlock.
779 */
780 ipc_lock_by_ptr(&msq->q_perm);
781 rcu_read_unlock();
782
783 /* Lockless receive, part 4:
784 * Repeat test after acquiring the spinlock.
785 */
786 msg = (struct msg_msg*)msr_d.r_msg;
787 if(msg != ERR_PTR(-EAGAIN))
788 goto out_unlock;
789
790 list_del(&msr_d.r_list);
791 if (signal_pending(current)) {
792 msg = ERR_PTR(-ERESTARTNOHAND);
793out_unlock:
794 msg_unlock(msq);
795 break;
796 }
797 }
798 if (IS_ERR(msg))
799 return PTR_ERR(msg);
800
801 msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
802 if (put_user (msg->m_type, &msgp->mtype) ||
803 store_msg(msgp->mtext, msg, msgsz)) {
804 msgsz = -EFAULT;
805 }
806 free_msg(msg);
807 return msgsz;
808}
809
810#ifdef CONFIG_PROC_FS
811static int sysvipc_msg_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
812{
813 off_t pos = 0;
814 off_t begin = 0;
815 int i, len = 0;
816
817 down(&msg_ids.sem);
818 len += sprintf(buffer, " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n");
819
820 for(i = 0; i <= msg_ids.max_id; i++) {
821 struct msg_queue * msq;
822 msq = msg_lock(i);
823 if(msq != NULL) {
824 len += sprintf(buffer + len, "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
825 msq->q_perm.key,
826 msg_buildid(i,msq->q_perm.seq),
827 msq->q_perm.mode,
828 msq->q_cbytes,
829 msq->q_qnum,
830 msq->q_lspid,
831 msq->q_lrpid,
832 msq->q_perm.uid,
833 msq->q_perm.gid,
834 msq->q_perm.cuid,
835 msq->q_perm.cgid,
836 msq->q_stime,
837 msq->q_rtime,
838 msq->q_ctime);
839 msg_unlock(msq);
840
841 pos += len;
842 if(pos < offset) {
843 len = 0;
844 begin = pos;
845 }
846 if(pos > offset + length)
847 goto done;
848 }
849
850 }
851 *eof = 1;
852done:
853 up(&msg_ids.sem);
854 *start = buffer + (offset - begin);
855 len -= (offset - begin);
856 if(len > length)
857 len = length;
858 if(len < 0)
859 len = 0;
860 return len;
861}
862#endif