thinkpad-acpi: fix ALSA callback return status
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / ipc / mqueue.c
blob0cefc8e023ff06387897e96ef505feadfa9d79aa
1 /*
2 * POSIX message queues filesystem for Linux.
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
5 * Michal Wronski (michal.wronski@gmail.com)
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul (manfred@colorfullife.com)
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
13 * This file is released under the GPL.
16 #include <linux/capability.h>
17 #include <linux/init.h>
18 #include <linux/pagemap.h>
19 #include <linux/file.h>
20 #include <linux/mount.h>
21 #include <linux/namei.h>
22 #include <linux/sysctl.h>
23 #include <linux/poll.h>
24 #include <linux/mqueue.h>
25 #include <linux/msg.h>
26 #include <linux/skbuff.h>
27 #include <linux/netlink.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/signal.h>
31 #include <linux/mutex.h>
32 #include <linux/nsproxy.h>
33 #include <linux/pid.h>
35 #include <net/sock.h>
36 #include "util.h"
38 #define MQUEUE_MAGIC 0x19800202
39 #define DIRENT_SIZE 20
40 #define FILENT_SIZE 80
42 #define SEND 0
43 #define RECV 1
45 #define STATE_NONE 0
46 #define STATE_PENDING 1
47 #define STATE_READY 2
49 /* default values */
50 #define DFLT_QUEUESMAX 256 /* max number of message queues */
51 #define DFLT_MSGMAX 10 /* max number of messages in each queue */
52 #define HARD_MSGMAX (131072/sizeof(void*))
53 #define DFLT_MSGSIZEMAX 8192 /* max message size */
56 * Define the ranges various user-specified maximum values can
57 * be set to.
59 #define MIN_MSGMAX 1 /* min value for msg_max */
60 #define MAX_MSGMAX HARD_MSGMAX /* max value for msg_max */
61 #define MIN_MSGSIZEMAX 128 /* min value for msgsize_max */
62 #define MAX_MSGSIZEMAX (8192*128) /* max value for msgsize_max */
64 struct ext_wait_queue { /* queue of sleeping tasks */
65 struct task_struct *task;
66 struct list_head list;
67 struct msg_msg *msg; /* ptr of loaded message */
68 int state; /* one of STATE_* values */
71 struct mqueue_inode_info {
72 spinlock_t lock;
73 struct inode vfs_inode;
74 wait_queue_head_t wait_q;
76 struct msg_msg **messages;
77 struct mq_attr attr;
79 struct sigevent notify;
80 struct pid* notify_owner;
81 struct user_struct *user; /* user who created, for accounting */
82 struct sock *notify_sock;
83 struct sk_buff *notify_cookie;
85 /* for tasks waiting for free space and messages, respectively */
86 struct ext_wait_queue e_wait_q[2];
88 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
91 static const struct inode_operations mqueue_dir_inode_operations;
92 static const struct file_operations mqueue_file_operations;
93 static struct super_operations mqueue_super_ops;
94 static void remove_notification(struct mqueue_inode_info *info);
96 static spinlock_t mq_lock;
97 static struct kmem_cache *mqueue_inode_cachep;
98 static struct vfsmount *mqueue_mnt;
100 static unsigned int queues_count;
101 static unsigned int queues_max = DFLT_QUEUESMAX;
102 static unsigned int msg_max = DFLT_MSGMAX;
103 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
105 static struct ctl_table_header * mq_sysctl_table;
107 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
109 return container_of(inode, struct mqueue_inode_info, vfs_inode);
112 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
113 struct mq_attr *attr)
115 struct inode *inode;
117 inode = new_inode(sb);
118 if (inode) {
119 inode->i_mode = mode;
120 inode->i_uid = current->fsuid;
121 inode->i_gid = current->fsgid;
122 inode->i_blocks = 0;
123 inode->i_mtime = inode->i_ctime = inode->i_atime =
124 CURRENT_TIME;
126 if (S_ISREG(mode)) {
127 struct mqueue_inode_info *info;
128 struct task_struct *p = current;
129 struct user_struct *u = p->user;
130 unsigned long mq_bytes, mq_msg_tblsz;
132 inode->i_fop = &mqueue_file_operations;
133 inode->i_size = FILENT_SIZE;
134 /* mqueue specific info */
135 info = MQUEUE_I(inode);
136 spin_lock_init(&info->lock);
137 init_waitqueue_head(&info->wait_q);
138 INIT_LIST_HEAD(&info->e_wait_q[0].list);
139 INIT_LIST_HEAD(&info->e_wait_q[1].list);
140 info->messages = NULL;
141 info->notify_owner = NULL;
142 info->qsize = 0;
143 info->user = NULL; /* set when all is ok */
144 memset(&info->attr, 0, sizeof(info->attr));
145 info->attr.mq_maxmsg = msg_max;
146 info->attr.mq_msgsize = msgsize_max;
147 if (attr) {
148 info->attr.mq_maxmsg = attr->mq_maxmsg;
149 info->attr.mq_msgsize = attr->mq_msgsize;
151 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
152 mq_bytes = (mq_msg_tblsz +
153 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
155 spin_lock(&mq_lock);
156 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
157 u->mq_bytes + mq_bytes >
158 p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
159 spin_unlock(&mq_lock);
160 goto out_inode;
162 u->mq_bytes += mq_bytes;
163 spin_unlock(&mq_lock);
165 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
166 if (!info->messages) {
167 spin_lock(&mq_lock);
168 u->mq_bytes -= mq_bytes;
169 spin_unlock(&mq_lock);
170 goto out_inode;
172 /* all is ok */
173 info->user = get_uid(u);
174 } else if (S_ISDIR(mode)) {
175 inc_nlink(inode);
176 /* Some things misbehave if size == 0 on a directory */
177 inode->i_size = 2 * DIRENT_SIZE;
178 inode->i_op = &mqueue_dir_inode_operations;
179 inode->i_fop = &simple_dir_operations;
182 return inode;
183 out_inode:
184 make_bad_inode(inode);
185 iput(inode);
186 return NULL;
189 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
191 struct inode *inode;
193 sb->s_blocksize = PAGE_CACHE_SIZE;
194 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
195 sb->s_magic = MQUEUE_MAGIC;
196 sb->s_op = &mqueue_super_ops;
198 inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
199 if (!inode)
200 return -ENOMEM;
202 sb->s_root = d_alloc_root(inode);
203 if (!sb->s_root) {
204 iput(inode);
205 return -ENOMEM;
208 return 0;
211 static int mqueue_get_sb(struct file_system_type *fs_type,
212 int flags, const char *dev_name,
213 void *data, struct vfsmount *mnt)
215 return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
218 static void init_once(void *foo)
220 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
222 inode_init_once(&p->vfs_inode);
225 static struct inode *mqueue_alloc_inode(struct super_block *sb)
227 struct mqueue_inode_info *ei;
229 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
230 if (!ei)
231 return NULL;
232 return &ei->vfs_inode;
235 static void mqueue_destroy_inode(struct inode *inode)
237 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
240 static void mqueue_delete_inode(struct inode *inode)
242 struct mqueue_inode_info *info;
243 struct user_struct *user;
244 unsigned long mq_bytes;
245 int i;
247 if (S_ISDIR(inode->i_mode)) {
248 clear_inode(inode);
249 return;
251 info = MQUEUE_I(inode);
252 spin_lock(&info->lock);
253 for (i = 0; i < info->attr.mq_curmsgs; i++)
254 free_msg(info->messages[i]);
255 kfree(info->messages);
256 spin_unlock(&info->lock);
258 clear_inode(inode);
260 mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
261 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
262 user = info->user;
263 if (user) {
264 spin_lock(&mq_lock);
265 user->mq_bytes -= mq_bytes;
266 queues_count--;
267 spin_unlock(&mq_lock);
268 free_uid(user);
272 static int mqueue_create(struct inode *dir, struct dentry *dentry,
273 int mode, struct nameidata *nd)
275 struct inode *inode;
276 struct mq_attr *attr = dentry->d_fsdata;
277 int error;
279 spin_lock(&mq_lock);
280 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
281 error = -ENOSPC;
282 goto out_lock;
284 queues_count++;
285 spin_unlock(&mq_lock);
287 inode = mqueue_get_inode(dir->i_sb, mode, attr);
288 if (!inode) {
289 error = -ENOMEM;
290 spin_lock(&mq_lock);
291 queues_count--;
292 goto out_lock;
295 dir->i_size += DIRENT_SIZE;
296 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
298 d_instantiate(dentry, inode);
299 dget(dentry);
300 return 0;
301 out_lock:
302 spin_unlock(&mq_lock);
303 return error;
306 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
308 struct inode *inode = dentry->d_inode;
310 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
311 dir->i_size -= DIRENT_SIZE;
312 drop_nlink(inode);
313 dput(dentry);
314 return 0;
318 * This is routine for system read from queue file.
319 * To avoid mess with doing here some sort of mq_receive we allow
320 * to read only queue size & notification info (the only values
321 * that are interesting from user point of view and aren't accessible
322 * through std routines)
324 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
325 size_t count, loff_t *off)
327 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
328 char buffer[FILENT_SIZE];
329 ssize_t ret;
331 spin_lock(&info->lock);
332 snprintf(buffer, sizeof(buffer),
333 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
334 info->qsize,
335 info->notify_owner ? info->notify.sigev_notify : 0,
336 (info->notify_owner &&
337 info->notify.sigev_notify == SIGEV_SIGNAL) ?
338 info->notify.sigev_signo : 0,
339 pid_vnr(info->notify_owner));
340 spin_unlock(&info->lock);
341 buffer[sizeof(buffer)-1] = '\0';
343 ret = simple_read_from_buffer(u_data, count, off, buffer,
344 strlen(buffer));
345 if (ret <= 0)
346 return ret;
348 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
349 return ret;
352 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
354 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
356 spin_lock(&info->lock);
357 if (task_tgid(current) == info->notify_owner)
358 remove_notification(info);
360 spin_unlock(&info->lock);
361 return 0;
364 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
366 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
367 int retval = 0;
369 poll_wait(filp, &info->wait_q, poll_tab);
371 spin_lock(&info->lock);
372 if (info->attr.mq_curmsgs)
373 retval = POLLIN | POLLRDNORM;
375 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
376 retval |= POLLOUT | POLLWRNORM;
377 spin_unlock(&info->lock);
379 return retval;
382 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
383 static void wq_add(struct mqueue_inode_info *info, int sr,
384 struct ext_wait_queue *ewp)
386 struct ext_wait_queue *walk;
388 ewp->task = current;
390 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
391 if (walk->task->static_prio <= current->static_prio) {
392 list_add_tail(&ewp->list, &walk->list);
393 return;
396 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
400 * Puts current task to sleep. Caller must hold queue lock. After return
401 * lock isn't held.
402 * sr: SEND or RECV
404 static int wq_sleep(struct mqueue_inode_info *info, int sr,
405 long timeout, struct ext_wait_queue *ewp)
407 int retval;
408 signed long time;
410 wq_add(info, sr, ewp);
412 for (;;) {
413 set_current_state(TASK_INTERRUPTIBLE);
415 spin_unlock(&info->lock);
416 time = schedule_timeout(timeout);
418 while (ewp->state == STATE_PENDING)
419 cpu_relax();
421 if (ewp->state == STATE_READY) {
422 retval = 0;
423 goto out;
425 spin_lock(&info->lock);
426 if (ewp->state == STATE_READY) {
427 retval = 0;
428 goto out_unlock;
430 if (signal_pending(current)) {
431 retval = -ERESTARTSYS;
432 break;
434 if (time == 0) {
435 retval = -ETIMEDOUT;
436 break;
439 list_del(&ewp->list);
440 out_unlock:
441 spin_unlock(&info->lock);
442 out:
443 return retval;
447 * Returns waiting task that should be serviced first or NULL if none exists
449 static struct ext_wait_queue *wq_get_first_waiter(
450 struct mqueue_inode_info *info, int sr)
452 struct list_head *ptr;
454 ptr = info->e_wait_q[sr].list.prev;
455 if (ptr == &info->e_wait_q[sr].list)
456 return NULL;
457 return list_entry(ptr, struct ext_wait_queue, list);
460 /* Auxiliary functions to manipulate messages' list */
461 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
463 int k;
465 k = info->attr.mq_curmsgs - 1;
466 while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
467 info->messages[k + 1] = info->messages[k];
468 k--;
470 info->attr.mq_curmsgs++;
471 info->qsize += ptr->m_ts;
472 info->messages[k + 1] = ptr;
475 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
477 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
478 return info->messages[info->attr.mq_curmsgs];
481 static inline void set_cookie(struct sk_buff *skb, char code)
483 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
487 * The next function is only to split too long sys_mq_timedsend
489 static void __do_notify(struct mqueue_inode_info *info)
491 /* notification
492 * invoked when there is registered process and there isn't process
493 * waiting synchronously for message AND state of queue changed from
494 * empty to not empty. Here we are sure that no one is waiting
495 * synchronously. */
496 if (info->notify_owner &&
497 info->attr.mq_curmsgs == 1) {
498 struct siginfo sig_i;
499 switch (info->notify.sigev_notify) {
500 case SIGEV_NONE:
501 break;
502 case SIGEV_SIGNAL:
503 /* sends signal */
505 sig_i.si_signo = info->notify.sigev_signo;
506 sig_i.si_errno = 0;
507 sig_i.si_code = SI_MESGQ;
508 sig_i.si_value = info->notify.sigev_value;
509 sig_i.si_pid = task_tgid_nr_ns(current,
510 ns_of_pid(info->notify_owner));
511 sig_i.si_uid = current->uid;
513 kill_pid_info(info->notify.sigev_signo,
514 &sig_i, info->notify_owner);
515 break;
516 case SIGEV_THREAD:
517 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
518 netlink_sendskb(info->notify_sock, info->notify_cookie);
519 break;
521 /* after notification unregisters process */
522 put_pid(info->notify_owner);
523 info->notify_owner = NULL;
525 wake_up(&info->wait_q);
528 static long prepare_timeout(const struct timespec __user *u_arg)
530 struct timespec ts, nowts;
531 long timeout;
533 if (u_arg) {
534 if (unlikely(copy_from_user(&ts, u_arg,
535 sizeof(struct timespec))))
536 return -EFAULT;
538 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
539 || ts.tv_nsec >= NSEC_PER_SEC))
540 return -EINVAL;
541 nowts = CURRENT_TIME;
542 /* first subtract as jiffies can't be too big */
543 ts.tv_sec -= nowts.tv_sec;
544 if (ts.tv_nsec < nowts.tv_nsec) {
545 ts.tv_nsec += NSEC_PER_SEC;
546 ts.tv_sec--;
548 ts.tv_nsec -= nowts.tv_nsec;
549 if (ts.tv_sec < 0)
550 return 0;
552 timeout = timespec_to_jiffies(&ts) + 1;
553 } else
554 return MAX_SCHEDULE_TIMEOUT;
556 return timeout;
559 static void remove_notification(struct mqueue_inode_info *info)
561 if (info->notify_owner != NULL &&
562 info->notify.sigev_notify == SIGEV_THREAD) {
563 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
564 netlink_sendskb(info->notify_sock, info->notify_cookie);
566 put_pid(info->notify_owner);
567 info->notify_owner = NULL;
570 static int mq_attr_ok(struct mq_attr *attr)
572 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
573 return 0;
574 if (capable(CAP_SYS_RESOURCE)) {
575 if (attr->mq_maxmsg > HARD_MSGMAX)
576 return 0;
577 } else {
578 if (attr->mq_maxmsg > msg_max ||
579 attr->mq_msgsize > msgsize_max)
580 return 0;
582 /* check for overflow */
583 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
584 return 0;
585 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
586 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
587 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
588 return 0;
589 return 1;
593 * Invoked when creating a new queue via sys_mq_open
595 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
596 int oflag, mode_t mode, struct mq_attr __user *u_attr)
598 struct mq_attr attr;
599 struct file *result;
600 int ret;
602 if (u_attr) {
603 ret = -EFAULT;
604 if (copy_from_user(&attr, u_attr, sizeof(attr)))
605 goto out;
606 ret = -EINVAL;
607 if (!mq_attr_ok(&attr))
608 goto out;
609 /* store for use during create */
610 dentry->d_fsdata = &attr;
613 mode &= ~current->fs->umask;
614 ret = mnt_want_write(mqueue_mnt);
615 if (ret)
616 goto out;
617 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
618 dentry->d_fsdata = NULL;
619 if (ret)
620 goto out_drop_write;
622 result = dentry_open(dentry, mqueue_mnt, oflag);
624 * dentry_open() took a persistent mnt_want_write(),
625 * so we can now drop this one.
627 mnt_drop_write(mqueue_mnt);
628 return result;
630 out_drop_write:
631 mnt_drop_write(mqueue_mnt);
632 out:
633 dput(dentry);
634 mntput(mqueue_mnt);
635 return ERR_PTR(ret);
638 /* Opens existing queue */
639 static struct file *do_open(struct dentry *dentry, int oflag)
641 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
642 MAY_READ | MAY_WRITE };
644 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
645 dput(dentry);
646 mntput(mqueue_mnt);
647 return ERR_PTR(-EINVAL);
650 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
651 dput(dentry);
652 mntput(mqueue_mnt);
653 return ERR_PTR(-EACCES);
656 return dentry_open(dentry, mqueue_mnt, oflag);
659 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
660 struct mq_attr __user *, u_attr)
662 struct dentry *dentry;
663 struct file *filp;
664 char *name;
665 int fd, error;
667 error = audit_mq_open(oflag, mode, u_attr);
668 if (error != 0)
669 return error;
671 if (IS_ERR(name = getname(u_name)))
672 return PTR_ERR(name);
674 fd = get_unused_fd_flags(O_CLOEXEC);
675 if (fd < 0)
676 goto out_putname;
678 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
679 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
680 if (IS_ERR(dentry)) {
681 error = PTR_ERR(dentry);
682 goto out_err;
684 mntget(mqueue_mnt);
686 if (oflag & O_CREAT) {
687 if (dentry->d_inode) { /* entry already exists */
688 audit_inode(name, dentry);
689 error = -EEXIST;
690 if (oflag & O_EXCL)
691 goto out;
692 filp = do_open(dentry, oflag);
693 } else {
694 filp = do_create(mqueue_mnt->mnt_root, dentry,
695 oflag, mode, u_attr);
697 } else {
698 error = -ENOENT;
699 if (!dentry->d_inode)
700 goto out;
701 audit_inode(name, dentry);
702 filp = do_open(dentry, oflag);
705 if (IS_ERR(filp)) {
706 error = PTR_ERR(filp);
707 goto out_putfd;
710 fd_install(fd, filp);
711 goto out_upsem;
713 out:
714 dput(dentry);
715 mntput(mqueue_mnt);
716 out_putfd:
717 put_unused_fd(fd);
718 out_err:
719 fd = error;
720 out_upsem:
721 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
722 out_putname:
723 putname(name);
724 return fd;
727 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
729 int err;
730 char *name;
731 struct dentry *dentry;
732 struct inode *inode = NULL;
734 name = getname(u_name);
735 if (IS_ERR(name))
736 return PTR_ERR(name);
738 mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
739 I_MUTEX_PARENT);
740 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
741 if (IS_ERR(dentry)) {
742 err = PTR_ERR(dentry);
743 goto out_unlock;
746 if (!dentry->d_inode) {
747 err = -ENOENT;
748 goto out_err;
751 inode = dentry->d_inode;
752 if (inode)
753 atomic_inc(&inode->i_count);
754 err = mnt_want_write(mqueue_mnt);
755 if (err)
756 goto out_err;
757 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
758 mnt_drop_write(mqueue_mnt);
759 out_err:
760 dput(dentry);
762 out_unlock:
763 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
764 putname(name);
765 if (inode)
766 iput(inode);
768 return err;
771 /* Pipelined send and receive functions.
773 * If a receiver finds no waiting message, then it registers itself in the
774 * list of waiting receivers. A sender checks that list before adding the new
775 * message into the message array. If there is a waiting receiver, then it
776 * bypasses the message array and directly hands the message over to the
777 * receiver.
778 * The receiver accepts the message and returns without grabbing the queue
779 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
780 * are necessary. The same algorithm is used for sysv semaphores, see
781 * ipc/sem.c for more details.
783 * The same algorithm is used for senders.
786 /* pipelined_send() - send a message directly to the task waiting in
787 * sys_mq_timedreceive() (without inserting message into a queue).
789 static inline void pipelined_send(struct mqueue_inode_info *info,
790 struct msg_msg *message,
791 struct ext_wait_queue *receiver)
793 receiver->msg = message;
794 list_del(&receiver->list);
795 receiver->state = STATE_PENDING;
796 wake_up_process(receiver->task);
797 smp_wmb();
798 receiver->state = STATE_READY;
801 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
802 * gets its message and put to the queue (we have one free place for sure). */
803 static inline void pipelined_receive(struct mqueue_inode_info *info)
805 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
807 if (!sender) {
808 /* for poll */
809 wake_up_interruptible(&info->wait_q);
810 return;
812 msg_insert(sender->msg, info);
813 list_del(&sender->list);
814 sender->state = STATE_PENDING;
815 wake_up_process(sender->task);
816 smp_wmb();
817 sender->state = STATE_READY;
820 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
821 size_t, msg_len, unsigned int, msg_prio,
822 const struct timespec __user *, u_abs_timeout)
824 struct file *filp;
825 struct inode *inode;
826 struct ext_wait_queue wait;
827 struct ext_wait_queue *receiver;
828 struct msg_msg *msg_ptr;
829 struct mqueue_inode_info *info;
830 long timeout;
831 int ret;
833 ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
834 if (ret != 0)
835 return ret;
837 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
838 return -EINVAL;
840 timeout = prepare_timeout(u_abs_timeout);
842 ret = -EBADF;
843 filp = fget(mqdes);
844 if (unlikely(!filp))
845 goto out;
847 inode = filp->f_path.dentry->d_inode;
848 if (unlikely(filp->f_op != &mqueue_file_operations))
849 goto out_fput;
850 info = MQUEUE_I(inode);
851 audit_inode(NULL, filp->f_path.dentry);
853 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
854 goto out_fput;
856 if (unlikely(msg_len > info->attr.mq_msgsize)) {
857 ret = -EMSGSIZE;
858 goto out_fput;
861 /* First try to allocate memory, before doing anything with
862 * existing queues. */
863 msg_ptr = load_msg(u_msg_ptr, msg_len);
864 if (IS_ERR(msg_ptr)) {
865 ret = PTR_ERR(msg_ptr);
866 goto out_fput;
868 msg_ptr->m_ts = msg_len;
869 msg_ptr->m_type = msg_prio;
871 spin_lock(&info->lock);
873 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
874 if (filp->f_flags & O_NONBLOCK) {
875 spin_unlock(&info->lock);
876 ret = -EAGAIN;
877 } else if (unlikely(timeout < 0)) {
878 spin_unlock(&info->lock);
879 ret = timeout;
880 } else {
881 wait.task = current;
882 wait.msg = (void *) msg_ptr;
883 wait.state = STATE_NONE;
884 ret = wq_sleep(info, SEND, timeout, &wait);
886 if (ret < 0)
887 free_msg(msg_ptr);
888 } else {
889 receiver = wq_get_first_waiter(info, RECV);
890 if (receiver) {
891 pipelined_send(info, msg_ptr, receiver);
892 } else {
893 /* adds message to the queue */
894 msg_insert(msg_ptr, info);
895 __do_notify(info);
897 inode->i_atime = inode->i_mtime = inode->i_ctime =
898 CURRENT_TIME;
899 spin_unlock(&info->lock);
900 ret = 0;
902 out_fput:
903 fput(filp);
904 out:
905 return ret;
908 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
909 size_t, msg_len, unsigned int __user *, u_msg_prio,
910 const struct timespec __user *, u_abs_timeout)
912 long timeout;
913 ssize_t ret;
914 struct msg_msg *msg_ptr;
915 struct file *filp;
916 struct inode *inode;
917 struct mqueue_inode_info *info;
918 struct ext_wait_queue wait;
920 ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
921 if (ret != 0)
922 return ret;
924 timeout = prepare_timeout(u_abs_timeout);
926 ret = -EBADF;
927 filp = fget(mqdes);
928 if (unlikely(!filp))
929 goto out;
931 inode = filp->f_path.dentry->d_inode;
932 if (unlikely(filp->f_op != &mqueue_file_operations))
933 goto out_fput;
934 info = MQUEUE_I(inode);
935 audit_inode(NULL, filp->f_path.dentry);
937 if (unlikely(!(filp->f_mode & FMODE_READ)))
938 goto out_fput;
940 /* checks if buffer is big enough */
941 if (unlikely(msg_len < info->attr.mq_msgsize)) {
942 ret = -EMSGSIZE;
943 goto out_fput;
946 spin_lock(&info->lock);
947 if (info->attr.mq_curmsgs == 0) {
948 if (filp->f_flags & O_NONBLOCK) {
949 spin_unlock(&info->lock);
950 ret = -EAGAIN;
951 msg_ptr = NULL;
952 } else if (unlikely(timeout < 0)) {
953 spin_unlock(&info->lock);
954 ret = timeout;
955 msg_ptr = NULL;
956 } else {
957 wait.task = current;
958 wait.state = STATE_NONE;
959 ret = wq_sleep(info, RECV, timeout, &wait);
960 msg_ptr = wait.msg;
962 } else {
963 msg_ptr = msg_get(info);
965 inode->i_atime = inode->i_mtime = inode->i_ctime =
966 CURRENT_TIME;
968 /* There is now free space in queue. */
969 pipelined_receive(info);
970 spin_unlock(&info->lock);
971 ret = 0;
973 if (ret == 0) {
974 ret = msg_ptr->m_ts;
976 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
977 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
978 ret = -EFAULT;
980 free_msg(msg_ptr);
982 out_fput:
983 fput(filp);
984 out:
985 return ret;
989 * Notes: the case when user wants us to deregister (with NULL as pointer)
990 * and he isn't currently owner of notification, will be silently discarded.
991 * It isn't explicitly defined in the POSIX.
993 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
994 const struct sigevent __user *, u_notification)
996 int ret;
997 struct file *filp;
998 struct sock *sock;
999 struct inode *inode;
1000 struct sigevent notification;
1001 struct mqueue_inode_info *info;
1002 struct sk_buff *nc;
1004 ret = audit_mq_notify(mqdes, u_notification);
1005 if (ret != 0)
1006 return ret;
1008 nc = NULL;
1009 sock = NULL;
1010 if (u_notification != NULL) {
1011 if (copy_from_user(&notification, u_notification,
1012 sizeof(struct sigevent)))
1013 return -EFAULT;
1015 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1016 notification.sigev_notify != SIGEV_SIGNAL &&
1017 notification.sigev_notify != SIGEV_THREAD))
1018 return -EINVAL;
1019 if (notification.sigev_notify == SIGEV_SIGNAL &&
1020 !valid_signal(notification.sigev_signo)) {
1021 return -EINVAL;
1023 if (notification.sigev_notify == SIGEV_THREAD) {
1024 long timeo;
1026 /* create the notify skb */
1027 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1028 ret = -ENOMEM;
1029 if (!nc)
1030 goto out;
1031 ret = -EFAULT;
1032 if (copy_from_user(nc->data,
1033 notification.sigev_value.sival_ptr,
1034 NOTIFY_COOKIE_LEN)) {
1035 goto out;
1038 /* TODO: add a header? */
1039 skb_put(nc, NOTIFY_COOKIE_LEN);
1040 /* and attach it to the socket */
1041 retry:
1042 filp = fget(notification.sigev_signo);
1043 ret = -EBADF;
1044 if (!filp)
1045 goto out;
1046 sock = netlink_getsockbyfilp(filp);
1047 fput(filp);
1048 if (IS_ERR(sock)) {
1049 ret = PTR_ERR(sock);
1050 sock = NULL;
1051 goto out;
1054 timeo = MAX_SCHEDULE_TIMEOUT;
1055 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1056 if (ret == 1)
1057 goto retry;
1058 if (ret) {
1059 sock = NULL;
1060 nc = NULL;
1061 goto out;
1066 ret = -EBADF;
1067 filp = fget(mqdes);
1068 if (!filp)
1069 goto out;
1071 inode = filp->f_path.dentry->d_inode;
1072 if (unlikely(filp->f_op != &mqueue_file_operations))
1073 goto out_fput;
1074 info = MQUEUE_I(inode);
1076 ret = 0;
1077 spin_lock(&info->lock);
1078 if (u_notification == NULL) {
1079 if (info->notify_owner == task_tgid(current)) {
1080 remove_notification(info);
1081 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1083 } else if (info->notify_owner != NULL) {
1084 ret = -EBUSY;
1085 } else {
1086 switch (notification.sigev_notify) {
1087 case SIGEV_NONE:
1088 info->notify.sigev_notify = SIGEV_NONE;
1089 break;
1090 case SIGEV_THREAD:
1091 info->notify_sock = sock;
1092 info->notify_cookie = nc;
1093 sock = NULL;
1094 nc = NULL;
1095 info->notify.sigev_notify = SIGEV_THREAD;
1096 break;
1097 case SIGEV_SIGNAL:
1098 info->notify.sigev_signo = notification.sigev_signo;
1099 info->notify.sigev_value = notification.sigev_value;
1100 info->notify.sigev_notify = SIGEV_SIGNAL;
1101 break;
1104 info->notify_owner = get_pid(task_tgid(current));
1105 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1107 spin_unlock(&info->lock);
1108 out_fput:
1109 fput(filp);
1110 out:
1111 if (sock) {
1112 netlink_detachskb(sock, nc);
1113 } else if (nc) {
1114 dev_kfree_skb(nc);
1116 return ret;
1119 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1120 const struct mq_attr __user *, u_mqstat,
1121 struct mq_attr __user *, u_omqstat)
1123 int ret;
1124 struct mq_attr mqstat, omqstat;
1125 struct file *filp;
1126 struct inode *inode;
1127 struct mqueue_inode_info *info;
1129 if (u_mqstat != NULL) {
1130 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1131 return -EFAULT;
1132 if (mqstat.mq_flags & (~O_NONBLOCK))
1133 return -EINVAL;
1136 ret = -EBADF;
1137 filp = fget(mqdes);
1138 if (!filp)
1139 goto out;
1141 inode = filp->f_path.dentry->d_inode;
1142 if (unlikely(filp->f_op != &mqueue_file_operations))
1143 goto out_fput;
1144 info = MQUEUE_I(inode);
1146 spin_lock(&info->lock);
1148 omqstat = info->attr;
1149 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1150 if (u_mqstat) {
1151 ret = audit_mq_getsetattr(mqdes, &mqstat);
1152 if (ret != 0) {
1153 spin_unlock(&info->lock);
1154 goto out_fput;
1156 if (mqstat.mq_flags & O_NONBLOCK)
1157 filp->f_flags |= O_NONBLOCK;
1158 else
1159 filp->f_flags &= ~O_NONBLOCK;
1161 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1164 spin_unlock(&info->lock);
1166 ret = 0;
1167 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1168 sizeof(struct mq_attr)))
1169 ret = -EFAULT;
1171 out_fput:
1172 fput(filp);
1173 out:
1174 return ret;
1177 static const struct inode_operations mqueue_dir_inode_operations = {
1178 .lookup = simple_lookup,
1179 .create = mqueue_create,
1180 .unlink = mqueue_unlink,
1183 static const struct file_operations mqueue_file_operations = {
1184 .flush = mqueue_flush_file,
1185 .poll = mqueue_poll_file,
1186 .read = mqueue_read_file,
1189 static struct super_operations mqueue_super_ops = {
1190 .alloc_inode = mqueue_alloc_inode,
1191 .destroy_inode = mqueue_destroy_inode,
1192 .statfs = simple_statfs,
1193 .delete_inode = mqueue_delete_inode,
1194 .drop_inode = generic_delete_inode,
1197 static struct file_system_type mqueue_fs_type = {
1198 .name = "mqueue",
1199 .get_sb = mqueue_get_sb,
1200 .kill_sb = kill_litter_super,
1203 static int msg_max_limit_min = MIN_MSGMAX;
1204 static int msg_max_limit_max = MAX_MSGMAX;
1206 static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1207 static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
1209 static ctl_table mq_sysctls[] = {
1211 .procname = "queues_max",
1212 .data = &queues_max,
1213 .maxlen = sizeof(int),
1214 .mode = 0644,
1215 .proc_handler = &proc_dointvec,
1218 .procname = "msg_max",
1219 .data = &msg_max,
1220 .maxlen = sizeof(int),
1221 .mode = 0644,
1222 .proc_handler = &proc_dointvec_minmax,
1223 .extra1 = &msg_max_limit_min,
1224 .extra2 = &msg_max_limit_max,
1227 .procname = "msgsize_max",
1228 .data = &msgsize_max,
1229 .maxlen = sizeof(int),
1230 .mode = 0644,
1231 .proc_handler = &proc_dointvec_minmax,
1232 .extra1 = &msg_maxsize_limit_min,
1233 .extra2 = &msg_maxsize_limit_max,
1235 { .ctl_name = 0 }
1238 static ctl_table mq_sysctl_dir[] = {
1240 .procname = "mqueue",
1241 .mode = 0555,
1242 .child = mq_sysctls,
1244 { .ctl_name = 0 }
1247 static ctl_table mq_sysctl_root[] = {
1249 .ctl_name = CTL_FS,
1250 .procname = "fs",
1251 .mode = 0555,
1252 .child = mq_sysctl_dir,
1254 { .ctl_name = 0 }
1257 static int __init init_mqueue_fs(void)
1259 int error;
1261 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1262 sizeof(struct mqueue_inode_info), 0,
1263 SLAB_HWCACHE_ALIGN, init_once);
1264 if (mqueue_inode_cachep == NULL)
1265 return -ENOMEM;
1267 /* ignore failues - they are not fatal */
1268 mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1270 error = register_filesystem(&mqueue_fs_type);
1271 if (error)
1272 goto out_sysctl;
1274 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1275 error = PTR_ERR(mqueue_mnt);
1276 goto out_filesystem;
1279 /* internal initialization - not common for vfs */
1280 queues_count = 0;
1281 spin_lock_init(&mq_lock);
1283 return 0;
1285 out_filesystem:
1286 unregister_filesystem(&mqueue_fs_type);
1287 out_sysctl:
1288 if (mq_sysctl_table)
1289 unregister_sysctl_table(mq_sysctl_table);
1290 kmem_cache_destroy(mqueue_inode_cachep);
1291 return error;
1294 __initcall(init_mqueue_fs);