blob: 4e776f9c80e7c1fbe0da01a1dcdc9f63c04ebd89 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
Michal Wronski65163fd2005-11-06 21:31:54 +01005 * Michal Wronski (Michal.Wronski@motorola.com)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul (manfred@colorfullife.com)
10 *
11 * This file is released under the GPL.
12 */
13
Randy.Dunlapc59ede72006-01-11 12:17:46 -080014#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/init.h>
16#include <linux/pagemap.h>
17#include <linux/file.h>
18#include <linux/mount.h>
19#include <linux/namei.h>
20#include <linux/sysctl.h>
21#include <linux/poll.h>
22#include <linux/mqueue.h>
23#include <linux/msg.h>
24#include <linux/skbuff.h>
25#include <linux/netlink.h>
26#include <linux/syscalls.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070027#include <linux/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <net/sock.h>
29#include "util.h"
30
31#define MQUEUE_MAGIC 0x19800202
32#define DIRENT_SIZE 20
33#define FILENT_SIZE 80
34
35#define SEND 0
36#define RECV 1
37
38#define STATE_NONE 0
39#define STATE_PENDING 1
40#define STATE_READY 2
41
42/* used by sysctl */
43#define FS_MQUEUE 1
44#define CTL_QUEUESMAX 2
45#define CTL_MSGMAX 3
46#define CTL_MSGSIZEMAX 4
47
48/* default values */
49#define DFLT_QUEUESMAX 256 /* max number of message queues */
50#define DFLT_MSGMAX 10 /* max number of messages in each queue */
51#define HARD_MSGMAX (131072/sizeof(void*))
52#define DFLT_MSGSIZEMAX 8192 /* max message size */
53
54#define NOTIFY_COOKIE_LEN 32
55
56struct ext_wait_queue { /* queue of sleeping tasks */
57 struct task_struct *task;
58 struct list_head list;
59 struct msg_msg *msg; /* ptr of loaded message */
60 int state; /* one of STATE_* values */
61};
62
63struct mqueue_inode_info {
64 spinlock_t lock;
65 struct inode vfs_inode;
66 wait_queue_head_t wait_q;
67
68 struct msg_msg **messages;
69 struct mq_attr attr;
70
71 struct sigevent notify;
72 pid_t notify_owner;
Adrian Bunk338cec32005-09-10 00:26:54 -070073 struct user_struct *user; /* user who created, for accounting */
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 struct sock *notify_sock;
75 struct sk_buff *notify_cookie;
76
77 /* for tasks waiting for free space and messages, respectively */
78 struct ext_wait_queue e_wait_q[2];
79
80 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
81};
82
83static struct inode_operations mqueue_dir_inode_operations;
84static struct file_operations mqueue_file_operations;
85static struct super_operations mqueue_super_ops;
86static void remove_notification(struct mqueue_inode_info *info);
87
88static spinlock_t mq_lock;
89static kmem_cache_t *mqueue_inode_cachep;
90static struct vfsmount *mqueue_mnt;
91
92static unsigned int queues_count;
93static unsigned int queues_max = DFLT_QUEUESMAX;
94static unsigned int msg_max = DFLT_MSGMAX;
95static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
96
97static struct ctl_table_header * mq_sysctl_table;
98
99static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100{
101 return container_of(inode, struct mqueue_inode_info, vfs_inode);
102}
103
104static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
105 struct mq_attr *attr)
106{
107 struct inode *inode;
108
109 inode = new_inode(sb);
110 if (inode) {
111 inode->i_mode = mode;
112 inode->i_uid = current->fsuid;
113 inode->i_gid = current->fsgid;
114 inode->i_blksize = PAGE_CACHE_SIZE;
115 inode->i_blocks = 0;
116 inode->i_mtime = inode->i_ctime = inode->i_atime =
117 CURRENT_TIME;
118
119 if (S_ISREG(mode)) {
120 struct mqueue_inode_info *info;
121 struct task_struct *p = current;
122 struct user_struct *u = p->user;
123 unsigned long mq_bytes, mq_msg_tblsz;
124
125 inode->i_fop = &mqueue_file_operations;
126 inode->i_size = FILENT_SIZE;
127 /* mqueue specific info */
128 info = MQUEUE_I(inode);
129 spin_lock_init(&info->lock);
130 init_waitqueue_head(&info->wait_q);
131 INIT_LIST_HEAD(&info->e_wait_q[0].list);
132 INIT_LIST_HEAD(&info->e_wait_q[1].list);
133 info->messages = NULL;
134 info->notify_owner = 0;
135 info->qsize = 0;
136 info->user = NULL; /* set when all is ok */
137 memset(&info->attr, 0, sizeof(info->attr));
138 info->attr.mq_maxmsg = DFLT_MSGMAX;
139 info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
140 if (attr) {
141 info->attr.mq_maxmsg = attr->mq_maxmsg;
142 info->attr.mq_msgsize = attr->mq_msgsize;
143 }
144 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
145 mq_bytes = (mq_msg_tblsz +
146 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
147
148 spin_lock(&mq_lock);
149 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
150 u->mq_bytes + mq_bytes >
151 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
152 spin_unlock(&mq_lock);
153 goto out_inode;
154 }
155 u->mq_bytes += mq_bytes;
156 spin_unlock(&mq_lock);
157
158 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
159 if (!info->messages) {
160 spin_lock(&mq_lock);
161 u->mq_bytes -= mq_bytes;
162 spin_unlock(&mq_lock);
163 goto out_inode;
164 }
165 /* all is ok */
166 info->user = get_uid(u);
167 } else if (S_ISDIR(mode)) {
168 inode->i_nlink++;
169 /* Some things misbehave if size == 0 on a directory */
170 inode->i_size = 2 * DIRENT_SIZE;
171 inode->i_op = &mqueue_dir_inode_operations;
172 inode->i_fop = &simple_dir_operations;
173 }
174 }
175 return inode;
176out_inode:
177 make_bad_inode(inode);
178 iput(inode);
179 return NULL;
180}
181
182static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
183{
184 struct inode *inode;
185
186 sb->s_blocksize = PAGE_CACHE_SIZE;
187 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
188 sb->s_magic = MQUEUE_MAGIC;
189 sb->s_op = &mqueue_super_ops;
190
191 inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
192 if (!inode)
193 return -ENOMEM;
194
195 sb->s_root = d_alloc_root(inode);
196 if (!sb->s_root) {
197 iput(inode);
198 return -ENOMEM;
199 }
200
201 return 0;
202}
203
204static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
205 int flags, const char *dev_name,
206 void *data)
207{
208 return get_sb_single(fs_type, flags, data, mqueue_fill_super);
209}
210
211static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
212{
213 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
214
215 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
216 SLAB_CTOR_CONSTRUCTOR)
217 inode_init_once(&p->vfs_inode);
218}
219
220static struct inode *mqueue_alloc_inode(struct super_block *sb)
221{
222 struct mqueue_inode_info *ei;
223
224 ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
225 if (!ei)
226 return NULL;
227 return &ei->vfs_inode;
228}
229
230static void mqueue_destroy_inode(struct inode *inode)
231{
232 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
233}
234
235static void mqueue_delete_inode(struct inode *inode)
236{
237 struct mqueue_inode_info *info;
238 struct user_struct *user;
239 unsigned long mq_bytes;
240 int i;
241
242 if (S_ISDIR(inode->i_mode)) {
243 clear_inode(inode);
244 return;
245 }
246 info = MQUEUE_I(inode);
247 spin_lock(&info->lock);
248 for (i = 0; i < info->attr.mq_curmsgs; i++)
249 free_msg(info->messages[i]);
250 kfree(info->messages);
251 spin_unlock(&info->lock);
252
253 clear_inode(inode);
254
255 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
256 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
257 user = info->user;
258 if (user) {
259 spin_lock(&mq_lock);
260 user->mq_bytes -= mq_bytes;
261 queues_count--;
262 spin_unlock(&mq_lock);
263 free_uid(user);
264 }
265}
266
267static int mqueue_create(struct inode *dir, struct dentry *dentry,
268 int mode, struct nameidata *nd)
269{
270 struct inode *inode;
271 struct mq_attr *attr = dentry->d_fsdata;
272 int error;
273
274 spin_lock(&mq_lock);
275 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
276 error = -ENOSPC;
277 goto out_lock;
278 }
279 queues_count++;
280 spin_unlock(&mq_lock);
281
282 inode = mqueue_get_inode(dir->i_sb, mode, attr);
283 if (!inode) {
284 error = -ENOMEM;
285 spin_lock(&mq_lock);
286 queues_count--;
287 goto out_lock;
288 }
289
290 dir->i_size += DIRENT_SIZE;
291 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
292
293 d_instantiate(dentry, inode);
294 dget(dentry);
295 return 0;
296out_lock:
297 spin_unlock(&mq_lock);
298 return error;
299}
300
301static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
302{
303 struct inode *inode = dentry->d_inode;
304
305 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
306 dir->i_size -= DIRENT_SIZE;
307 inode->i_nlink--;
308 dput(dentry);
309 return 0;
310}
311
312/*
313* This is routine for system read from queue file.
314* To avoid mess with doing here some sort of mq_receive we allow
315* to read only queue size & notification info (the only values
316* that are interesting from user point of view and aren't accessible
317* through std routines)
318*/
319static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
320 size_t count, loff_t * off)
321{
322 struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
323 char buffer[FILENT_SIZE];
324 size_t slen;
325 loff_t o;
326
327 if (!count)
328 return 0;
329
330 spin_lock(&info->lock);
331 snprintf(buffer, sizeof(buffer),
332 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
333 info->qsize,
334 info->notify_owner ? info->notify.sigev_notify : 0,
335 (info->notify_owner &&
336 info->notify.sigev_notify == SIGEV_SIGNAL) ?
337 info->notify.sigev_signo : 0,
338 info->notify_owner);
339 spin_unlock(&info->lock);
340 buffer[sizeof(buffer)-1] = '\0';
341 slen = strlen(buffer)+1;
342
343 o = *off;
344 if (o > slen)
345 return 0;
346
347 if (o + count > slen)
348 count = slen - o;
349
350 if (copy_to_user(u_data, buffer + o, count))
351 return -EFAULT;
352
353 *off = o + count;
354 filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
355 return count;
356}
357
358static int mqueue_flush_file(struct file *filp)
359{
360 struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
361
362 spin_lock(&info->lock);
363 if (current->tgid == info->notify_owner)
364 remove_notification(info);
365
366 spin_unlock(&info->lock);
367 return 0;
368}
369
370static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
371{
372 struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
373 int retval = 0;
374
375 poll_wait(filp, &info->wait_q, poll_tab);
376
377 spin_lock(&info->lock);
378 if (info->attr.mq_curmsgs)
379 retval = POLLIN | POLLRDNORM;
380
381 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
382 retval |= POLLOUT | POLLWRNORM;
383 spin_unlock(&info->lock);
384
385 return retval;
386}
387
388/* Adds current to info->e_wait_q[sr] before element with smaller prio */
389static void wq_add(struct mqueue_inode_info *info, int sr,
390 struct ext_wait_queue *ewp)
391{
392 struct ext_wait_queue *walk;
393
394 ewp->task = current;
395
396 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
397 if (walk->task->static_prio <= current->static_prio) {
398 list_add_tail(&ewp->list, &walk->list);
399 return;
400 }
401 }
402 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
403}
404
405/*
406 * Puts current task to sleep. Caller must hold queue lock. After return
407 * lock isn't held.
408 * sr: SEND or RECV
409 */
410static int wq_sleep(struct mqueue_inode_info *info, int sr,
411 long timeout, struct ext_wait_queue *ewp)
412{
413 int retval;
414 signed long time;
415
416 wq_add(info, sr, ewp);
417
418 for (;;) {
419 set_current_state(TASK_INTERRUPTIBLE);
420
421 spin_unlock(&info->lock);
422 time = schedule_timeout(timeout);
423
424 while (ewp->state == STATE_PENDING)
425 cpu_relax();
426
427 if (ewp->state == STATE_READY) {
428 retval = 0;
429 goto out;
430 }
431 spin_lock(&info->lock);
432 if (ewp->state == STATE_READY) {
433 retval = 0;
434 goto out_unlock;
435 }
436 if (signal_pending(current)) {
437 retval = -ERESTARTSYS;
438 break;
439 }
440 if (time == 0) {
441 retval = -ETIMEDOUT;
442 break;
443 }
444 }
445 list_del(&ewp->list);
446out_unlock:
447 spin_unlock(&info->lock);
448out:
449 return retval;
450}
451
452/*
453 * Returns waiting task that should be serviced first or NULL if none exists
454 */
455static struct ext_wait_queue *wq_get_first_waiter(
456 struct mqueue_inode_info *info, int sr)
457{
458 struct list_head *ptr;
459
460 ptr = info->e_wait_q[sr].list.prev;
461 if (ptr == &info->e_wait_q[sr].list)
462 return NULL;
463 return list_entry(ptr, struct ext_wait_queue, list);
464}
465
466/* Auxiliary functions to manipulate messages' list */
467static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
468{
469 int k;
470
471 k = info->attr.mq_curmsgs - 1;
472 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
473 info->messages[k + 1] = info->messages[k];
474 k--;
475 }
476 info->attr.mq_curmsgs++;
477 info->qsize += ptr->m_ts;
478 info->messages[k + 1] = ptr;
479}
480
481static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
482{
483 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
484 return info->messages[info->attr.mq_curmsgs];
485}
486
487static inline void set_cookie(struct sk_buff *skb, char code)
488{
489 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
490}
491
492/*
493 * The next function is only to split too long sys_mq_timedsend
494 */
495static void __do_notify(struct mqueue_inode_info *info)
496{
497 /* notification
498 * invoked when there is registered process and there isn't process
499 * waiting synchronously for message AND state of queue changed from
500 * empty to not empty. Here we are sure that no one is waiting
501 * synchronously. */
502 if (info->notify_owner &&
503 info->attr.mq_curmsgs == 1) {
504 struct siginfo sig_i;
505 switch (info->notify.sigev_notify) {
506 case SIGEV_NONE:
507 break;
508 case SIGEV_SIGNAL:
509 /* sends signal */
510
511 sig_i.si_signo = info->notify.sigev_signo;
512 sig_i.si_errno = 0;
513 sig_i.si_code = SI_MESGQ;
514 sig_i.si_value = info->notify.sigev_value;
515 sig_i.si_pid = current->tgid;
516 sig_i.si_uid = current->uid;
517
518 kill_proc_info(info->notify.sigev_signo,
519 &sig_i, info->notify_owner);
520 break;
521 case SIGEV_THREAD:
522 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
523 netlink_sendskb(info->notify_sock,
524 info->notify_cookie, 0);
525 break;
526 }
527 /* after notification unregisters process */
528 info->notify_owner = 0;
529 }
530 wake_up(&info->wait_q);
531}
532
533static long prepare_timeout(const struct timespec __user *u_arg)
534{
535 struct timespec ts, nowts;
536 long timeout;
537
538 if (u_arg) {
539 if (unlikely(copy_from_user(&ts, u_arg,
540 sizeof(struct timespec))))
541 return -EFAULT;
542
543 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
544 || ts.tv_nsec >= NSEC_PER_SEC))
545 return -EINVAL;
546 nowts = CURRENT_TIME;
547 /* first subtract as jiffies can't be too big */
548 ts.tv_sec -= nowts.tv_sec;
549 if (ts.tv_nsec < nowts.tv_nsec) {
550 ts.tv_nsec += NSEC_PER_SEC;
551 ts.tv_sec--;
552 }
553 ts.tv_nsec -= nowts.tv_nsec;
554 if (ts.tv_sec < 0)
555 return 0;
556
557 timeout = timespec_to_jiffies(&ts) + 1;
558 } else
559 return MAX_SCHEDULE_TIMEOUT;
560
561 return timeout;
562}
563
564static void remove_notification(struct mqueue_inode_info *info)
565{
566 if (info->notify_owner != 0 &&
567 info->notify.sigev_notify == SIGEV_THREAD) {
568 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
569 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
570 }
571 info->notify_owner = 0;
572}
573
574static int mq_attr_ok(struct mq_attr *attr)
575{
576 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
577 return 0;
578 if (capable(CAP_SYS_RESOURCE)) {
579 if (attr->mq_maxmsg > HARD_MSGMAX)
580 return 0;
581 } else {
582 if (attr->mq_maxmsg > msg_max ||
583 attr->mq_msgsize > msgsize_max)
584 return 0;
585 }
586 /* check for overflow */
587 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
588 return 0;
589 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
590 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
591 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
592 return 0;
593 return 1;
594}
595
596/*
597 * Invoked when creating a new queue via sys_mq_open
598 */
599static struct file *do_create(struct dentry *dir, struct dentry *dentry,
600 int oflag, mode_t mode, struct mq_attr __user *u_attr)
601{
602 struct file *filp;
603 struct mq_attr attr;
604 int ret;
605
606 if (u_attr != NULL) {
607 if (copy_from_user(&attr, u_attr, sizeof(attr)))
608 return ERR_PTR(-EFAULT);
609 if (!mq_attr_ok(&attr))
610 return ERR_PTR(-EINVAL);
611 /* store for use during create */
612 dentry->d_fsdata = &attr;
613 }
614
Krzysztof Benedyczak59175832005-09-27 22:21:36 +0200615 mode &= ~current->fs->umask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
617 dentry->d_fsdata = NULL;
618 if (ret)
619 return ERR_PTR(ret);
620
621 filp = dentry_open(dentry, mqueue_mnt, oflag);
622 if (!IS_ERR(filp))
623 dget(dentry);
624
625 return filp;
626}
627
628/* Opens existing queue */
629static struct file *do_open(struct dentry *dentry, int oflag)
630{
631static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
632 MAY_READ | MAY_WRITE };
633 struct file *filp;
634
635 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
636 return ERR_PTR(-EINVAL);
637
638 if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
639 return ERR_PTR(-EACCES);
640
641 filp = dentry_open(dentry, mqueue_mnt, oflag);
642
643 if (!IS_ERR(filp))
644 dget(dentry);
645
646 return filp;
647}
648
649asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
650 struct mq_attr __user *u_attr)
651{
652 struct dentry *dentry;
653 struct file *filp;
654 char *name;
655 int fd, error;
656
657 if (IS_ERR(name = getname(u_name)))
658 return PTR_ERR(name);
659
660 fd = get_unused_fd();
661 if (fd < 0)
662 goto out_putname;
663
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800664 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
666 if (IS_ERR(dentry)) {
667 error = PTR_ERR(dentry);
668 goto out_err;
669 }
670 mntget(mqueue_mnt);
671
672 if (oflag & O_CREAT) {
673 if (dentry->d_inode) { /* entry already exists */
674 filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
675 do_open(dentry, oflag);
676 } else {
677 filp = do_create(mqueue_mnt->mnt_root, dentry,
678 oflag, mode, u_attr);
679 }
680 } else
681 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
682 ERR_PTR(-ENOENT);
683
684 dput(dentry);
685
686 if (IS_ERR(filp)) {
687 error = PTR_ERR(filp);
688 goto out_putfd;
689 }
690
691 set_close_on_exec(fd, 1);
692 fd_install(fd, filp);
693 goto out_upsem;
694
695out_putfd:
696 mntput(mqueue_mnt);
697 put_unused_fd(fd);
698out_err:
699 fd = error;
700out_upsem:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800701 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702out_putname:
703 putname(name);
704 return fd;
705}
706
707asmlinkage long sys_mq_unlink(const char __user *u_name)
708{
709 int err;
710 char *name;
711 struct dentry *dentry;
712 struct inode *inode = NULL;
713
714 name = getname(u_name);
715 if (IS_ERR(name))
716 return PTR_ERR(name);
717
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800718 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
720 if (IS_ERR(dentry)) {
721 err = PTR_ERR(dentry);
722 goto out_unlock;
723 }
724
725 if (!dentry->d_inode) {
726 err = -ENOENT;
727 goto out_err;
728 }
729
730 inode = dentry->d_inode;
731 if (inode)
732 atomic_inc(&inode->i_count);
733
734 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
735out_err:
736 dput(dentry);
737
738out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800739 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 putname(name);
741 if (inode)
742 iput(inode);
743
744 return err;
745}
746
747/* Pipelined send and receive functions.
748 *
749 * If a receiver finds no waiting message, then it registers itself in the
750 * list of waiting receivers. A sender checks that list before adding the new
751 * message into the message array. If there is a waiting receiver, then it
752 * bypasses the message array and directly hands the message over to the
753 * receiver.
754 * The receiver accepts the message and returns without grabbing the queue
755 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
756 * are necessary. The same algorithm is used for sysv semaphores, see
757 * ipc/sem.c fore more details.
758 *
759 * The same algorithm is used for senders.
760 */
761
762/* pipelined_send() - send a message directly to the task waiting in
763 * sys_mq_timedreceive() (without inserting message into a queue).
764 */
765static inline void pipelined_send(struct mqueue_inode_info *info,
766 struct msg_msg *message,
767 struct ext_wait_queue *receiver)
768{
769 receiver->msg = message;
770 list_del(&receiver->list);
771 receiver->state = STATE_PENDING;
772 wake_up_process(receiver->task);
akpm@osdl.orgd59dd462005-05-01 08:58:47 -0700773 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 receiver->state = STATE_READY;
775}
776
777/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
778 * gets its message and put to the queue (we have one free place for sure). */
779static inline void pipelined_receive(struct mqueue_inode_info *info)
780{
781 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
782
783 if (!sender) {
784 /* for poll */
785 wake_up_interruptible(&info->wait_q);
786 return;
787 }
788 msg_insert(sender->msg, info);
789 list_del(&sender->list);
790 sender->state = STATE_PENDING;
791 wake_up_process(sender->task);
akpm@osdl.orgd59dd462005-05-01 08:58:47 -0700792 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 sender->state = STATE_READY;
794}
795
796asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
797 size_t msg_len, unsigned int msg_prio,
798 const struct timespec __user *u_abs_timeout)
799{
800 struct file *filp;
801 struct inode *inode;
802 struct ext_wait_queue wait;
803 struct ext_wait_queue *receiver;
804 struct msg_msg *msg_ptr;
805 struct mqueue_inode_info *info;
806 long timeout;
807 int ret;
808
809 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
810 return -EINVAL;
811
812 timeout = prepare_timeout(u_abs_timeout);
813
814 ret = -EBADF;
815 filp = fget(mqdes);
816 if (unlikely(!filp))
817 goto out;
818
819 inode = filp->f_dentry->d_inode;
820 if (unlikely(filp->f_op != &mqueue_file_operations))
821 goto out_fput;
822 info = MQUEUE_I(inode);
823
824 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
825 goto out_fput;
826
827 if (unlikely(msg_len > info->attr.mq_msgsize)) {
828 ret = -EMSGSIZE;
829 goto out_fput;
830 }
831
832 /* First try to allocate memory, before doing anything with
833 * existing queues. */
834 msg_ptr = load_msg(u_msg_ptr, msg_len);
835 if (IS_ERR(msg_ptr)) {
836 ret = PTR_ERR(msg_ptr);
837 goto out_fput;
838 }
839 msg_ptr->m_ts = msg_len;
840 msg_ptr->m_type = msg_prio;
841
842 spin_lock(&info->lock);
843
844 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
845 if (filp->f_flags & O_NONBLOCK) {
846 spin_unlock(&info->lock);
847 ret = -EAGAIN;
848 } else if (unlikely(timeout < 0)) {
849 spin_unlock(&info->lock);
850 ret = timeout;
851 } else {
852 wait.task = current;
853 wait.msg = (void *) msg_ptr;
854 wait.state = STATE_NONE;
855 ret = wq_sleep(info, SEND, timeout, &wait);
856 }
857 if (ret < 0)
858 free_msg(msg_ptr);
859 } else {
860 receiver = wq_get_first_waiter(info, RECV);
861 if (receiver) {
862 pipelined_send(info, msg_ptr, receiver);
863 } else {
864 /* adds message to the queue */
865 msg_insert(msg_ptr, info);
866 __do_notify(info);
867 }
868 inode->i_atime = inode->i_mtime = inode->i_ctime =
869 CURRENT_TIME;
870 spin_unlock(&info->lock);
871 ret = 0;
872 }
873out_fput:
874 fput(filp);
875out:
876 return ret;
877}
878
879asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
880 size_t msg_len, unsigned int __user *u_msg_prio,
881 const struct timespec __user *u_abs_timeout)
882{
883 long timeout;
884 ssize_t ret;
885 struct msg_msg *msg_ptr;
886 struct file *filp;
887 struct inode *inode;
888 struct mqueue_inode_info *info;
889 struct ext_wait_queue wait;
890
891 timeout = prepare_timeout(u_abs_timeout);
892
893 ret = -EBADF;
894 filp = fget(mqdes);
895 if (unlikely(!filp))
896 goto out;
897
898 inode = filp->f_dentry->d_inode;
899 if (unlikely(filp->f_op != &mqueue_file_operations))
900 goto out_fput;
901 info = MQUEUE_I(inode);
902
903 if (unlikely(!(filp->f_mode & FMODE_READ)))
904 goto out_fput;
905
906 /* checks if buffer is big enough */
907 if (unlikely(msg_len < info->attr.mq_msgsize)) {
908 ret = -EMSGSIZE;
909 goto out_fput;
910 }
911
912 spin_lock(&info->lock);
913 if (info->attr.mq_curmsgs == 0) {
914 if (filp->f_flags & O_NONBLOCK) {
915 spin_unlock(&info->lock);
916 ret = -EAGAIN;
917 msg_ptr = NULL;
918 } else if (unlikely(timeout < 0)) {
919 spin_unlock(&info->lock);
920 ret = timeout;
921 msg_ptr = NULL;
922 } else {
923 wait.task = current;
924 wait.state = STATE_NONE;
925 ret = wq_sleep(info, RECV, timeout, &wait);
926 msg_ptr = wait.msg;
927 }
928 } else {
929 msg_ptr = msg_get(info);
930
931 inode->i_atime = inode->i_mtime = inode->i_ctime =
932 CURRENT_TIME;
933
934 /* There is now free space in queue. */
935 pipelined_receive(info);
936 spin_unlock(&info->lock);
937 ret = 0;
938 }
939 if (ret == 0) {
940 ret = msg_ptr->m_ts;
941
942 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
943 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
944 ret = -EFAULT;
945 }
946 free_msg(msg_ptr);
947 }
948out_fput:
949 fput(filp);
950out:
951 return ret;
952}
953
954/*
955 * Notes: the case when user wants us to deregister (with NULL as pointer)
956 * and he isn't currently owner of notification, will be silently discarded.
957 * It isn't explicitly defined in the POSIX.
958 */
959asmlinkage long sys_mq_notify(mqd_t mqdes,
960 const struct sigevent __user *u_notification)
961{
962 int ret;
963 struct file *filp;
964 struct sock *sock;
965 struct inode *inode;
966 struct sigevent notification;
967 struct mqueue_inode_info *info;
968 struct sk_buff *nc;
969
970 nc = NULL;
971 sock = NULL;
972 if (u_notification != NULL) {
973 if (copy_from_user(&notification, u_notification,
974 sizeof(struct sigevent)))
975 return -EFAULT;
976
977 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
978 notification.sigev_notify != SIGEV_SIGNAL &&
979 notification.sigev_notify != SIGEV_THREAD))
980 return -EINVAL;
981 if (notification.sigev_notify == SIGEV_SIGNAL &&
Jesper Juhl7ed20e12005-05-01 08:59:14 -0700982 !valid_signal(notification.sigev_signo)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 return -EINVAL;
984 }
985 if (notification.sigev_notify == SIGEV_THREAD) {
986 /* create the notify skb */
987 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
988 ret = -ENOMEM;
989 if (!nc)
990 goto out;
991 ret = -EFAULT;
992 if (copy_from_user(nc->data,
993 notification.sigev_value.sival_ptr,
994 NOTIFY_COOKIE_LEN)) {
995 goto out;
996 }
997
998 /* TODO: add a header? */
999 skb_put(nc, NOTIFY_COOKIE_LEN);
1000 /* and attach it to the socket */
1001retry:
1002 filp = fget(notification.sigev_signo);
1003 ret = -EBADF;
1004 if (!filp)
1005 goto out;
1006 sock = netlink_getsockbyfilp(filp);
1007 fput(filp);
1008 if (IS_ERR(sock)) {
1009 ret = PTR_ERR(sock);
1010 sock = NULL;
1011 goto out;
1012 }
1013
1014 ret = netlink_attachskb(sock, nc, 0, MAX_SCHEDULE_TIMEOUT);
1015 if (ret == 1)
1016 goto retry;
1017 if (ret) {
1018 sock = NULL;
1019 nc = NULL;
1020 goto out;
1021 }
1022 }
1023 }
1024
1025 ret = -EBADF;
1026 filp = fget(mqdes);
1027 if (!filp)
1028 goto out;
1029
1030 inode = filp->f_dentry->d_inode;
1031 if (unlikely(filp->f_op != &mqueue_file_operations))
1032 goto out_fput;
1033 info = MQUEUE_I(inode);
1034
1035 ret = 0;
1036 spin_lock(&info->lock);
1037 if (u_notification == NULL) {
1038 if (info->notify_owner == current->tgid) {
1039 remove_notification(info);
1040 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1041 }
1042 } else if (info->notify_owner != 0) {
1043 ret = -EBUSY;
1044 } else {
1045 switch (notification.sigev_notify) {
1046 case SIGEV_NONE:
1047 info->notify.sigev_notify = SIGEV_NONE;
1048 break;
1049 case SIGEV_THREAD:
1050 info->notify_sock = sock;
1051 info->notify_cookie = nc;
1052 sock = NULL;
1053 nc = NULL;
1054 info->notify.sigev_notify = SIGEV_THREAD;
1055 break;
1056 case SIGEV_SIGNAL:
1057 info->notify.sigev_signo = notification.sigev_signo;
1058 info->notify.sigev_value = notification.sigev_value;
1059 info->notify.sigev_notify = SIGEV_SIGNAL;
1060 break;
1061 }
1062 info->notify_owner = current->tgid;
1063 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1064 }
1065 spin_unlock(&info->lock);
1066out_fput:
1067 fput(filp);
1068out:
1069 if (sock) {
1070 netlink_detachskb(sock, nc);
1071 } else if (nc) {
1072 dev_kfree_skb(nc);
1073 }
1074 return ret;
1075}
1076
1077asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1078 const struct mq_attr __user *u_mqstat,
1079 struct mq_attr __user *u_omqstat)
1080{
1081 int ret;
1082 struct mq_attr mqstat, omqstat;
1083 struct file *filp;
1084 struct inode *inode;
1085 struct mqueue_inode_info *info;
1086
1087 if (u_mqstat != NULL) {
1088 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1089 return -EFAULT;
1090 if (mqstat.mq_flags & (~O_NONBLOCK))
1091 return -EINVAL;
1092 }
1093
1094 ret = -EBADF;
1095 filp = fget(mqdes);
1096 if (!filp)
1097 goto out;
1098
1099 inode = filp->f_dentry->d_inode;
1100 if (unlikely(filp->f_op != &mqueue_file_operations))
1101 goto out_fput;
1102 info = MQUEUE_I(inode);
1103
1104 spin_lock(&info->lock);
1105
1106 omqstat = info->attr;
1107 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1108 if (u_mqstat) {
1109 if (mqstat.mq_flags & O_NONBLOCK)
1110 filp->f_flags |= O_NONBLOCK;
1111 else
1112 filp->f_flags &= ~O_NONBLOCK;
1113
1114 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1115 }
1116
1117 spin_unlock(&info->lock);
1118
1119 ret = 0;
1120 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1121 sizeof(struct mq_attr)))
1122 ret = -EFAULT;
1123
1124out_fput:
1125 fput(filp);
1126out:
1127 return ret;
1128}
1129
1130static struct inode_operations mqueue_dir_inode_operations = {
1131 .lookup = simple_lookup,
1132 .create = mqueue_create,
1133 .unlink = mqueue_unlink,
1134};
1135
1136static struct file_operations mqueue_file_operations = {
1137 .flush = mqueue_flush_file,
1138 .poll = mqueue_poll_file,
1139 .read = mqueue_read_file,
1140};
1141
1142static struct super_operations mqueue_super_ops = {
1143 .alloc_inode = mqueue_alloc_inode,
1144 .destroy_inode = mqueue_destroy_inode,
1145 .statfs = simple_statfs,
1146 .delete_inode = mqueue_delete_inode,
1147 .drop_inode = generic_delete_inode,
1148};
1149
1150static struct file_system_type mqueue_fs_type = {
1151 .name = "mqueue",
1152 .get_sb = mqueue_get_sb,
1153 .kill_sb = kill_litter_super,
1154};
1155
1156static int msg_max_limit_min = DFLT_MSGMAX;
1157static int msg_max_limit_max = HARD_MSGMAX;
1158
1159static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1160static int msg_maxsize_limit_max = INT_MAX;
1161
1162static ctl_table mq_sysctls[] = {
1163 {
1164 .ctl_name = CTL_QUEUESMAX,
1165 .procname = "queues_max",
1166 .data = &queues_max,
1167 .maxlen = sizeof(int),
1168 .mode = 0644,
1169 .proc_handler = &proc_dointvec,
1170 },
1171 {
1172 .ctl_name = CTL_MSGMAX,
1173 .procname = "msg_max",
1174 .data = &msg_max,
1175 .maxlen = sizeof(int),
1176 .mode = 0644,
1177 .proc_handler = &proc_dointvec_minmax,
1178 .extra1 = &msg_max_limit_min,
1179 .extra2 = &msg_max_limit_max,
1180 },
1181 {
1182 .ctl_name = CTL_MSGSIZEMAX,
1183 .procname = "msgsize_max",
1184 .data = &msgsize_max,
1185 .maxlen = sizeof(int),
1186 .mode = 0644,
1187 .proc_handler = &proc_dointvec_minmax,
1188 .extra1 = &msg_maxsize_limit_min,
1189 .extra2 = &msg_maxsize_limit_max,
1190 },
1191 { .ctl_name = 0 }
1192};
1193
1194static ctl_table mq_sysctl_dir[] = {
1195 {
1196 .ctl_name = FS_MQUEUE,
1197 .procname = "mqueue",
1198 .mode = 0555,
1199 .child = mq_sysctls,
1200 },
1201 { .ctl_name = 0 }
1202};
1203
1204static ctl_table mq_sysctl_root[] = {
1205 {
1206 .ctl_name = CTL_FS,
1207 .procname = "fs",
1208 .mode = 0555,
1209 .child = mq_sysctl_dir,
1210 },
1211 { .ctl_name = 0 }
1212};
1213
1214static int __init init_mqueue_fs(void)
1215{
1216 int error;
1217
1218 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1219 sizeof(struct mqueue_inode_info), 0,
1220 SLAB_HWCACHE_ALIGN, init_once, NULL);
1221 if (mqueue_inode_cachep == NULL)
1222 return -ENOMEM;
1223
1224 /* ignore failues - they are not fatal */
1225 mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1226
1227 error = register_filesystem(&mqueue_fs_type);
1228 if (error)
1229 goto out_sysctl;
1230
1231 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1232 error = PTR_ERR(mqueue_mnt);
1233 goto out_filesystem;
1234 }
1235
1236 /* internal initialization - not common for vfs */
1237 queues_count = 0;
1238 spin_lock_init(&mq_lock);
1239
1240 return 0;
1241
1242out_filesystem:
1243 unregister_filesystem(&mqueue_fs_type);
1244out_sysctl:
1245 if (mq_sysctl_table)
1246 unregister_sysctl_table(mq_sysctl_table);
1247 if (kmem_cache_destroy(mqueue_inode_cachep)) {
1248 printk(KERN_INFO
1249 "mqueue_inode_cache: not all structures were freed\n");
1250 }
1251 return error;
1252}
1253
1254__initcall(init_mqueue_fs);