blob: 7563611c6615435e8a05ecb0758335c4d3fc0d54 [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 Wronskif66e9282006-10-03 23:23:27 +02005 * Michal Wronski (michal.wronski@gmail.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 *
George C. Wilson20ca73b2006-05-24 16:09:55 -050011 * Audit: George Wilson (ltcgcw@us.ibm.com)
12 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * This file is released under the GPL.
14 */
15
Randy.Dunlapc59ede72006-01-11 12:17:46 -080016#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#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>
George C. Wilson20ca73b2006-05-24 16:09:55 -050029#include <linux/audit.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070030#include <linux/signal.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080031#include <linux/mutex.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070032#include <linux/nsproxy.h>
33#include <linux/pid.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <net/sock.h>
36#include "util.h"
37
38#define MQUEUE_MAGIC 0x19800202
39#define DIRENT_SIZE 20
40#define FILENT_SIZE 80
41
42#define SEND 0
43#define RECV 1
44
45#define STATE_NONE 0
46#define STATE_PENDING 1
47#define STATE_READY 2
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049/* 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 */
54
Joe Kortyb231cca42008-10-18 20:28:32 -070055/*
56 * Define the ranges various user-specified maximum values can
57 * be set to.
58 */
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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64struct 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 */
69};
70
71struct mqueue_inode_info {
72 spinlock_t lock;
73 struct inode vfs_inode;
74 wait_queue_head_t wait_q;
75
76 struct msg_msg **messages;
77 struct mq_attr attr;
78
79 struct sigevent notify;
Cedric Le Goatera03fcb72006-10-02 02:17:26 -070080 struct pid* notify_owner;
Adrian Bunk338cec32005-09-10 00:26:54 -070081 struct user_struct *user; /* user who created, for accounting */
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 struct sock *notify_sock;
83 struct sk_buff *notify_cookie;
84
85 /* for tasks waiting for free space and messages, respectively */
86 struct ext_wait_queue e_wait_q[2];
87
88 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
89};
90
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080091static const struct inode_operations mqueue_dir_inode_operations;
Arjan van de Ven9a321442007-02-12 00:55:35 -080092static const struct file_operations mqueue_file_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093static struct super_operations mqueue_super_ops;
94static void remove_notification(struct mqueue_inode_info *info);
95
96static spinlock_t mq_lock;
Christoph Lametere18b8902006-12-06 20:33:20 -080097static struct kmem_cache *mqueue_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098static struct vfsmount *mqueue_mnt;
99
100static unsigned int queues_count;
101static unsigned int queues_max = DFLT_QUEUESMAX;
102static unsigned int msg_max = DFLT_MSGMAX;
103static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
104
105static struct ctl_table_header * mq_sysctl_table;
106
107static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
108{
109 return container_of(inode, struct mqueue_inode_info, vfs_inode);
110}
111
112static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
113 struct mq_attr *attr)
114{
David Howells86a264a2008-11-14 10:39:18 +1100115 struct user_struct *u = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 struct inode *inode;
117
118 inode = new_inode(sb);
119 if (inode) {
120 inode->i_mode = mode;
David Howells414c0702008-11-14 10:39:06 +1100121 inode->i_uid = current_fsuid();
122 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 inode->i_blocks = 0;
124 inode->i_mtime = inode->i_ctime = inode->i_atime =
125 CURRENT_TIME;
126
127 if (S_ISREG(mode)) {
128 struct mqueue_inode_info *info;
129 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 unsigned long mq_bytes, mq_msg_tblsz;
131
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;
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700141 info->notify_owner = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 info->qsize = 0;
143 info->user = NULL; /* set when all is ok */
144 memset(&info->attr, 0, sizeof(info->attr));
Joe Kortyb231cca42008-10-18 20:28:32 -0700145 info->attr.mq_maxmsg = msg_max;
146 info->attr.mq_msgsize = msgsize_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 if (attr) {
148 info->attr.mq_maxmsg = attr->mq_maxmsg;
149 info->attr.mq_msgsize = attr->mq_msgsize;
150 }
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));
154
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;
161 }
162 u->mq_bytes += mq_bytes;
163 spin_unlock(&mq_lock);
164
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;
171 }
172 /* all is ok */
173 info->user = get_uid(u);
174 } else if (S_ISDIR(mode)) {
Dave Hansend8c76e62006-09-30 23:29:04 -0700175 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 /* 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;
180 }
181 }
182 return inode;
183out_inode:
184 make_bad_inode(inode);
185 iput(inode);
186 return NULL;
187}
188
189static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
190{
191 struct inode *inode;
192
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;
197
198 inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
199 if (!inode)
200 return -ENOMEM;
201
202 sb->s_root = d_alloc_root(inode);
203 if (!sb->s_root) {
204 iput(inode);
205 return -ENOMEM;
206 }
207
208 return 0;
209}
210
David Howells454e2392006-06-23 02:02:57 -0700211static int mqueue_get_sb(struct file_system_type *fs_type,
212 int flags, const char *dev_name,
213 void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
David Howells454e2392006-06-23 02:02:57 -0700215 return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700218static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
221
Christoph Lametera35afb82007-05-16 22:10:57 -0700222 inode_init_once(&p->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225static struct inode *mqueue_alloc_inode(struct super_block *sb)
226{
227 struct mqueue_inode_info *ei;
228
Christoph Lametere94b1762006-12-06 20:33:17 -0800229 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 if (!ei)
231 return NULL;
232 return &ei->vfs_inode;
233}
234
235static void mqueue_destroy_inode(struct inode *inode)
236{
237 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
238}
239
240static void mqueue_delete_inode(struct inode *inode)
241{
242 struct mqueue_inode_info *info;
243 struct user_struct *user;
244 unsigned long mq_bytes;
245 int i;
246
247 if (S_ISDIR(inode->i_mode)) {
248 clear_inode(inode);
249 return;
250 }
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);
257
258 clear_inode(inode);
259
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);
269 }
270}
271
272static int mqueue_create(struct inode *dir, struct dentry *dentry,
273 int mode, struct nameidata *nd)
274{
275 struct inode *inode;
276 struct mq_attr *attr = dentry->d_fsdata;
277 int error;
278
279 spin_lock(&mq_lock);
280 if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
281 error = -ENOSPC;
282 goto out_lock;
283 }
284 queues_count++;
285 spin_unlock(&mq_lock);
286
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;
293 }
294
295 dir->i_size += DIRENT_SIZE;
296 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
297
298 d_instantiate(dentry, inode);
299 dget(dentry);
300 return 0;
301out_lock:
302 spin_unlock(&mq_lock);
303 return error;
304}
305
306static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
307{
308 struct inode *inode = dentry->d_inode;
309
310 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
311 dir->i_size -= DIRENT_SIZE;
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700312 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 dput(dentry);
314 return 0;
315}
316
317/*
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)
323*/
324static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700325 size_t count, loff_t *off)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326{
Josef Sipek6d630792006-12-08 02:37:11 -0800327 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 char buffer[FILENT_SIZE];
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700329 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
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,
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800339 pid_vnr(info->notify_owner));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 spin_unlock(&info->lock);
341 buffer[sizeof(buffer)-1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700343 ret = simple_read_from_buffer(u_data, count, off, buffer,
344 strlen(buffer));
345 if (ret <= 0)
346 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Josef Sipek6d630792006-12-08 02:37:11 -0800348 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700349 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700352static int mqueue_flush_file(struct file *filp, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Josef Sipek6d630792006-12-08 02:37:11 -0800354 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356 spin_lock(&info->lock);
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700357 if (task_tgid(current) == info->notify_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 remove_notification(info);
359
360 spin_unlock(&info->lock);
361 return 0;
362}
363
364static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
365{
Josef Sipek6d630792006-12-08 02:37:11 -0800366 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 int retval = 0;
368
369 poll_wait(filp, &info->wait_q, poll_tab);
370
371 spin_lock(&info->lock);
372 if (info->attr.mq_curmsgs)
373 retval = POLLIN | POLLRDNORM;
374
375 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
376 retval |= POLLOUT | POLLWRNORM;
377 spin_unlock(&info->lock);
378
379 return retval;
380}
381
382/* Adds current to info->e_wait_q[sr] before element with smaller prio */
383static void wq_add(struct mqueue_inode_info *info, int sr,
384 struct ext_wait_queue *ewp)
385{
386 struct ext_wait_queue *walk;
387
388 ewp->task = current;
389
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;
394 }
395 }
396 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
397}
398
399/*
400 * Puts current task to sleep. Caller must hold queue lock. After return
401 * lock isn't held.
402 * sr: SEND or RECV
403 */
404static int wq_sleep(struct mqueue_inode_info *info, int sr,
405 long timeout, struct ext_wait_queue *ewp)
406{
407 int retval;
408 signed long time;
409
410 wq_add(info, sr, ewp);
411
412 for (;;) {
413 set_current_state(TASK_INTERRUPTIBLE);
414
415 spin_unlock(&info->lock);
416 time = schedule_timeout(timeout);
417
418 while (ewp->state == STATE_PENDING)
419 cpu_relax();
420
421 if (ewp->state == STATE_READY) {
422 retval = 0;
423 goto out;
424 }
425 spin_lock(&info->lock);
426 if (ewp->state == STATE_READY) {
427 retval = 0;
428 goto out_unlock;
429 }
430 if (signal_pending(current)) {
431 retval = -ERESTARTSYS;
432 break;
433 }
434 if (time == 0) {
435 retval = -ETIMEDOUT;
436 break;
437 }
438 }
439 list_del(&ewp->list);
440out_unlock:
441 spin_unlock(&info->lock);
442out:
443 return retval;
444}
445
446/*
447 * Returns waiting task that should be serviced first or NULL if none exists
448 */
449static struct ext_wait_queue *wq_get_first_waiter(
450 struct mqueue_inode_info *info, int sr)
451{
452 struct list_head *ptr;
453
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);
458}
459
460/* Auxiliary functions to manipulate messages' list */
461static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
462{
463 int k;
464
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--;
469 }
470 info->attr.mq_curmsgs++;
471 info->qsize += ptr->m_ts;
472 info->messages[k + 1] = ptr;
473}
474
475static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
476{
477 info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
478 return info->messages[info->attr.mq_curmsgs];
479}
480
481static inline void set_cookie(struct sk_buff *skb, char code)
482{
483 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
484}
485
486/*
487 * The next function is only to split too long sys_mq_timedsend
488 */
489static void __do_notify(struct mqueue_inode_info *info)
490{
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 */
504
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;
Pavel Emelyanov56496c12008-02-08 04:19:21 -0800509 sig_i.si_pid = task_tgid_vnr(current);
David Howells414c0702008-11-14 10:39:06 +1100510 sig_i.si_uid = current_uid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700512 kill_pid_info(info->notify.sigev_signo,
513 &sig_i, info->notify_owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 break;
515 case SIGEV_THREAD:
516 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
Denis V. Lunev7ee015e2007-10-10 21:14:03 -0700517 netlink_sendskb(info->notify_sock, info->notify_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 break;
519 }
520 /* after notification unregisters process */
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700521 put_pid(info->notify_owner);
522 info->notify_owner = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 }
524 wake_up(&info->wait_q);
525}
526
527static long prepare_timeout(const struct timespec __user *u_arg)
528{
529 struct timespec ts, nowts;
530 long timeout;
531
532 if (u_arg) {
533 if (unlikely(copy_from_user(&ts, u_arg,
534 sizeof(struct timespec))))
535 return -EFAULT;
536
537 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
538 || ts.tv_nsec >= NSEC_PER_SEC))
539 return -EINVAL;
540 nowts = CURRENT_TIME;
541 /* first subtract as jiffies can't be too big */
542 ts.tv_sec -= nowts.tv_sec;
543 if (ts.tv_nsec < nowts.tv_nsec) {
544 ts.tv_nsec += NSEC_PER_SEC;
545 ts.tv_sec--;
546 }
547 ts.tv_nsec -= nowts.tv_nsec;
548 if (ts.tv_sec < 0)
549 return 0;
550
551 timeout = timespec_to_jiffies(&ts) + 1;
552 } else
553 return MAX_SCHEDULE_TIMEOUT;
554
555 return timeout;
556}
557
558static void remove_notification(struct mqueue_inode_info *info)
559{
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700560 if (info->notify_owner != NULL &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 info->notify.sigev_notify == SIGEV_THREAD) {
562 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
Denis V. Lunev7ee015e2007-10-10 21:14:03 -0700563 netlink_sendskb(info->notify_sock, info->notify_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700565 put_pid(info->notify_owner);
566 info->notify_owner = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
568
569static int mq_attr_ok(struct mq_attr *attr)
570{
571 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
572 return 0;
573 if (capable(CAP_SYS_RESOURCE)) {
574 if (attr->mq_maxmsg > HARD_MSGMAX)
575 return 0;
576 } else {
577 if (attr->mq_maxmsg > msg_max ||
578 attr->mq_msgsize > msgsize_max)
579 return 0;
580 }
581 /* check for overflow */
582 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
583 return 0;
584 if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
585 (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
586 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
587 return 0;
588 return 1;
589}
590
591/*
592 * Invoked when creating a new queue via sys_mq_open
593 */
594static struct file *do_create(struct dentry *dir, struct dentry *dentry,
595 int oflag, mode_t mode, struct mq_attr __user *u_attr)
596{
David Howells745ca242008-11-14 10:39:22 +1100597 const struct cred *cred = current_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 struct mq_attr attr;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800599 struct file *result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 int ret;
601
Alexander Viro7c7dce92006-01-14 15:29:55 -0500602 if (u_attr) {
603 ret = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 if (copy_from_user(&attr, u_attr, sizeof(attr)))
Alexander Viro7c7dce92006-01-14 15:29:55 -0500605 goto out;
606 ret = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 if (!mq_attr_ok(&attr))
Alexander Viro7c7dce92006-01-14 15:29:55 -0500608 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 /* store for use during create */
610 dentry->d_fsdata = &attr;
611 }
612
Krzysztof Benedyczak59175832005-09-27 22:21:36 +0200613 mode &= ~current->fs->umask;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800614 ret = mnt_want_write(mqueue_mnt);
615 if (ret)
616 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 ret = vfs_create(dir->d_inode, dentry, mode, NULL);
618 dentry->d_fsdata = NULL;
619 if (ret)
Dave Hansen4a3fd212008-02-15 14:37:48 -0800620 goto out_drop_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
David Howells745ca242008-11-14 10:39:22 +1100622 result = dentry_open(dentry, mqueue_mnt, oflag, cred);
Dave Hansen4a3fd212008-02-15 14:37:48 -0800623 /*
624 * dentry_open() took a persistent mnt_want_write(),
625 * so we can now drop this one.
626 */
627 mnt_drop_write(mqueue_mnt);
628 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Dave Hansen4a3fd212008-02-15 14:37:48 -0800630out_drop_write:
631 mnt_drop_write(mqueue_mnt);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500632out:
633 dput(dentry);
634 mntput(mqueue_mnt);
635 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636}
637
638/* Opens existing queue */
639static struct file *do_open(struct dentry *dentry, int oflag)
640{
David Howells745ca242008-11-14 10:39:22 +1100641 const struct cred *cred = current_cred();
642
643 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
644 MAY_READ | MAY_WRITE };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Alexander Viro7c7dce92006-01-14 15:29:55 -0500646 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
647 dput(dentry);
648 mntput(mqueue_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 return ERR_PTR(-EINVAL);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Al Virof419a2e2008-07-22 00:07:17 -0400652 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
Alexander Viro7c7dce92006-01-14 15:29:55 -0500653 dput(dentry);
654 mntput(mqueue_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 return ERR_PTR(-EACCES);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
David Howells745ca242008-11-14 10:39:22 +1100658 return dentry_open(dentry, mqueue_mnt, oflag, cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
660
661asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
662 struct mq_attr __user *u_attr)
663{
664 struct dentry *dentry;
665 struct file *filp;
666 char *name;
667 int fd, error;
668
George C. Wilson20ca73b2006-05-24 16:09:55 -0500669 error = audit_mq_open(oflag, mode, u_attr);
670 if (error != 0)
671 return error;
672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (IS_ERR(name = getname(u_name)))
674 return PTR_ERR(name);
675
Ulrich Drepper269f2132008-05-03 15:28:45 -0400676 fd = get_unused_fd_flags(O_CLOEXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 if (fd < 0)
678 goto out_putname;
679
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800680 mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
682 if (IS_ERR(dentry)) {
683 error = PTR_ERR(dentry);
684 goto out_err;
685 }
686 mntget(mqueue_mnt);
687
688 if (oflag & O_CREAT) {
689 if (dentry->d_inode) { /* entry already exists */
Al Viro5a190ae2007-06-07 12:19:32 -0400690 audit_inode(name, dentry);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500691 error = -EEXIST;
692 if (oflag & O_EXCL)
693 goto out;
694 filp = do_open(dentry, oflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 } else {
696 filp = do_create(mqueue_mnt->mnt_root, dentry,
697 oflag, mode, u_attr);
698 }
Alexander Viro7c7dce92006-01-14 15:29:55 -0500699 } else {
700 error = -ENOENT;
701 if (!dentry->d_inode)
702 goto out;
Al Viro5a190ae2007-06-07 12:19:32 -0400703 audit_inode(name, dentry);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500704 filp = do_open(dentry, oflag);
705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 if (IS_ERR(filp)) {
708 error = PTR_ERR(filp);
709 goto out_putfd;
710 }
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 fd_install(fd, filp);
713 goto out_upsem;
714
Alexander Viro7c7dce92006-01-14 15:29:55 -0500715out:
716 dput(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 mntput(mqueue_mnt);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500718out_putfd:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 put_unused_fd(fd);
720out_err:
721 fd = error;
722out_upsem:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800723 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724out_putname:
725 putname(name);
726 return fd;
727}
728
729asmlinkage long sys_mq_unlink(const char __user *u_name)
730{
731 int err;
732 char *name;
733 struct dentry *dentry;
734 struct inode *inode = NULL;
735
736 name = getname(u_name);
737 if (IS_ERR(name))
738 return PTR_ERR(name);
739
Peter Zijlstra7a434812007-03-06 01:42:09 -0800740 mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
741 I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
743 if (IS_ERR(dentry)) {
744 err = PTR_ERR(dentry);
745 goto out_unlock;
746 }
747
748 if (!dentry->d_inode) {
749 err = -ENOENT;
750 goto out_err;
751 }
752
753 inode = dentry->d_inode;
754 if (inode)
755 atomic_inc(&inode->i_count);
Dave Hansen06227532008-02-15 14:37:34 -0800756 err = mnt_want_write(mqueue_mnt);
757 if (err)
758 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
Dave Hansen06227532008-02-15 14:37:34 -0800760 mnt_drop_write(mqueue_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761out_err:
762 dput(dentry);
763
764out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800765 mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 putname(name);
767 if (inode)
768 iput(inode);
769
770 return err;
771}
772
773/* Pipelined send and receive functions.
774 *
775 * If a receiver finds no waiting message, then it registers itself in the
776 * list of waiting receivers. A sender checks that list before adding the new
777 * message into the message array. If there is a waiting receiver, then it
778 * bypasses the message array and directly hands the message over to the
779 * receiver.
780 * The receiver accepts the message and returns without grabbing the queue
781 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
782 * are necessary. The same algorithm is used for sysv semaphores, see
Serge E. Hallyn7b7a3172006-03-28 01:56:23 -0800783 * ipc/sem.c for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 *
785 * The same algorithm is used for senders.
786 */
787
788/* pipelined_send() - send a message directly to the task waiting in
789 * sys_mq_timedreceive() (without inserting message into a queue).
790 */
791static inline void pipelined_send(struct mqueue_inode_info *info,
792 struct msg_msg *message,
793 struct ext_wait_queue *receiver)
794{
795 receiver->msg = message;
796 list_del(&receiver->list);
797 receiver->state = STATE_PENDING;
798 wake_up_process(receiver->task);
akpm@osdl.orgd59dd462005-05-01 08:58:47 -0700799 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 receiver->state = STATE_READY;
801}
802
803/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
804 * gets its message and put to the queue (we have one free place for sure). */
805static inline void pipelined_receive(struct mqueue_inode_info *info)
806{
807 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
808
809 if (!sender) {
810 /* for poll */
811 wake_up_interruptible(&info->wait_q);
812 return;
813 }
814 msg_insert(sender->msg, info);
815 list_del(&sender->list);
816 sender->state = STATE_PENDING;
817 wake_up_process(sender->task);
akpm@osdl.orgd59dd462005-05-01 08:58:47 -0700818 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 sender->state = STATE_READY;
820}
821
822asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
823 size_t msg_len, unsigned int msg_prio,
824 const struct timespec __user *u_abs_timeout)
825{
826 struct file *filp;
827 struct inode *inode;
828 struct ext_wait_queue wait;
829 struct ext_wait_queue *receiver;
830 struct msg_msg *msg_ptr;
831 struct mqueue_inode_info *info;
832 long timeout;
833 int ret;
834
George C. Wilson20ca73b2006-05-24 16:09:55 -0500835 ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
836 if (ret != 0)
837 return ret;
838
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
840 return -EINVAL;
841
842 timeout = prepare_timeout(u_abs_timeout);
843
844 ret = -EBADF;
845 filp = fget(mqdes);
846 if (unlikely(!filp))
847 goto out;
848
Josef Sipek6d630792006-12-08 02:37:11 -0800849 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 if (unlikely(filp->f_op != &mqueue_file_operations))
851 goto out_fput;
852 info = MQUEUE_I(inode);
Al Viro5a190ae2007-06-07 12:19:32 -0400853 audit_inode(NULL, filp->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 if (unlikely(!(filp->f_mode & FMODE_WRITE)))
856 goto out_fput;
857
858 if (unlikely(msg_len > info->attr.mq_msgsize)) {
859 ret = -EMSGSIZE;
860 goto out_fput;
861 }
862
863 /* First try to allocate memory, before doing anything with
864 * existing queues. */
865 msg_ptr = load_msg(u_msg_ptr, msg_len);
866 if (IS_ERR(msg_ptr)) {
867 ret = PTR_ERR(msg_ptr);
868 goto out_fput;
869 }
870 msg_ptr->m_ts = msg_len;
871 msg_ptr->m_type = msg_prio;
872
873 spin_lock(&info->lock);
874
875 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
876 if (filp->f_flags & O_NONBLOCK) {
877 spin_unlock(&info->lock);
878 ret = -EAGAIN;
879 } else if (unlikely(timeout < 0)) {
880 spin_unlock(&info->lock);
881 ret = timeout;
882 } else {
883 wait.task = current;
884 wait.msg = (void *) msg_ptr;
885 wait.state = STATE_NONE;
886 ret = wq_sleep(info, SEND, timeout, &wait);
887 }
888 if (ret < 0)
889 free_msg(msg_ptr);
890 } else {
891 receiver = wq_get_first_waiter(info, RECV);
892 if (receiver) {
893 pipelined_send(info, msg_ptr, receiver);
894 } else {
895 /* adds message to the queue */
896 msg_insert(msg_ptr, info);
897 __do_notify(info);
898 }
899 inode->i_atime = inode->i_mtime = inode->i_ctime =
900 CURRENT_TIME;
901 spin_unlock(&info->lock);
902 ret = 0;
903 }
904out_fput:
905 fput(filp);
906out:
907 return ret;
908}
909
910asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
911 size_t msg_len, unsigned int __user *u_msg_prio,
912 const struct timespec __user *u_abs_timeout)
913{
914 long timeout;
915 ssize_t ret;
916 struct msg_msg *msg_ptr;
917 struct file *filp;
918 struct inode *inode;
919 struct mqueue_inode_info *info;
920 struct ext_wait_queue wait;
921
George C. Wilson20ca73b2006-05-24 16:09:55 -0500922 ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
923 if (ret != 0)
924 return ret;
925
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 timeout = prepare_timeout(u_abs_timeout);
927
928 ret = -EBADF;
929 filp = fget(mqdes);
930 if (unlikely(!filp))
931 goto out;
932
Josef Sipek6d630792006-12-08 02:37:11 -0800933 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 if (unlikely(filp->f_op != &mqueue_file_operations))
935 goto out_fput;
936 info = MQUEUE_I(inode);
Al Viro5a190ae2007-06-07 12:19:32 -0400937 audit_inode(NULL, filp->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
939 if (unlikely(!(filp->f_mode & FMODE_READ)))
940 goto out_fput;
941
942 /* checks if buffer is big enough */
943 if (unlikely(msg_len < info->attr.mq_msgsize)) {
944 ret = -EMSGSIZE;
945 goto out_fput;
946 }
947
948 spin_lock(&info->lock);
949 if (info->attr.mq_curmsgs == 0) {
950 if (filp->f_flags & O_NONBLOCK) {
951 spin_unlock(&info->lock);
952 ret = -EAGAIN;
953 msg_ptr = NULL;
954 } else if (unlikely(timeout < 0)) {
955 spin_unlock(&info->lock);
956 ret = timeout;
957 msg_ptr = NULL;
958 } else {
959 wait.task = current;
960 wait.state = STATE_NONE;
961 ret = wq_sleep(info, RECV, timeout, &wait);
962 msg_ptr = wait.msg;
963 }
964 } else {
965 msg_ptr = msg_get(info);
966
967 inode->i_atime = inode->i_mtime = inode->i_ctime =
968 CURRENT_TIME;
969
970 /* There is now free space in queue. */
971 pipelined_receive(info);
972 spin_unlock(&info->lock);
973 ret = 0;
974 }
975 if (ret == 0) {
976 ret = msg_ptr->m_ts;
977
978 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
979 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
980 ret = -EFAULT;
981 }
982 free_msg(msg_ptr);
983 }
984out_fput:
985 fput(filp);
986out:
987 return ret;
988}
989
990/*
991 * Notes: the case when user wants us to deregister (with NULL as pointer)
992 * and he isn't currently owner of notification, will be silently discarded.
993 * It isn't explicitly defined in the POSIX.
994 */
995asmlinkage long sys_mq_notify(mqd_t mqdes,
996 const struct sigevent __user *u_notification)
997{
998 int ret;
999 struct file *filp;
1000 struct sock *sock;
1001 struct inode *inode;
1002 struct sigevent notification;
1003 struct mqueue_inode_info *info;
1004 struct sk_buff *nc;
1005
George C. Wilson20ca73b2006-05-24 16:09:55 -05001006 ret = audit_mq_notify(mqdes, u_notification);
1007 if (ret != 0)
1008 return ret;
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 nc = NULL;
1011 sock = NULL;
1012 if (u_notification != NULL) {
1013 if (copy_from_user(&notification, u_notification,
1014 sizeof(struct sigevent)))
1015 return -EFAULT;
1016
1017 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1018 notification.sigev_notify != SIGEV_SIGNAL &&
1019 notification.sigev_notify != SIGEV_THREAD))
1020 return -EINVAL;
1021 if (notification.sigev_notify == SIGEV_SIGNAL &&
Jesper Juhl7ed20e12005-05-01 08:59:14 -07001022 !valid_signal(notification.sigev_signo)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 return -EINVAL;
1024 }
1025 if (notification.sigev_notify == SIGEV_THREAD) {
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001026 long timeo;
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 /* create the notify skb */
1029 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1030 ret = -ENOMEM;
1031 if (!nc)
1032 goto out;
1033 ret = -EFAULT;
1034 if (copy_from_user(nc->data,
1035 notification.sigev_value.sival_ptr,
1036 NOTIFY_COOKIE_LEN)) {
1037 goto out;
1038 }
1039
1040 /* TODO: add a header? */
1041 skb_put(nc, NOTIFY_COOKIE_LEN);
1042 /* and attach it to the socket */
1043retry:
1044 filp = fget(notification.sigev_signo);
1045 ret = -EBADF;
1046 if (!filp)
1047 goto out;
1048 sock = netlink_getsockbyfilp(filp);
1049 fput(filp);
1050 if (IS_ERR(sock)) {
1051 ret = PTR_ERR(sock);
1052 sock = NULL;
1053 goto out;
1054 }
1055
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001056 timeo = MAX_SCHEDULE_TIMEOUT;
Denis V. Lunev9457afe2008-06-05 11:23:39 -07001057 ret = netlink_attachskb(sock, nc, &timeo, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 if (ret == 1)
1059 goto retry;
1060 if (ret) {
1061 sock = NULL;
1062 nc = NULL;
1063 goto out;
1064 }
1065 }
1066 }
1067
1068 ret = -EBADF;
1069 filp = fget(mqdes);
1070 if (!filp)
1071 goto out;
1072
Josef Sipek6d630792006-12-08 02:37:11 -08001073 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (unlikely(filp->f_op != &mqueue_file_operations))
1075 goto out_fput;
1076 info = MQUEUE_I(inode);
1077
1078 ret = 0;
1079 spin_lock(&info->lock);
1080 if (u_notification == NULL) {
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001081 if (info->notify_owner == task_tgid(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 remove_notification(info);
1083 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1084 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001085 } else if (info->notify_owner != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 ret = -EBUSY;
1087 } else {
1088 switch (notification.sigev_notify) {
1089 case SIGEV_NONE:
1090 info->notify.sigev_notify = SIGEV_NONE;
1091 break;
1092 case SIGEV_THREAD:
1093 info->notify_sock = sock;
1094 info->notify_cookie = nc;
1095 sock = NULL;
1096 nc = NULL;
1097 info->notify.sigev_notify = SIGEV_THREAD;
1098 break;
1099 case SIGEV_SIGNAL:
1100 info->notify.sigev_signo = notification.sigev_signo;
1101 info->notify.sigev_value = notification.sigev_value;
1102 info->notify.sigev_notify = SIGEV_SIGNAL;
1103 break;
1104 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001105
1106 info->notify_owner = get_pid(task_tgid(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1108 }
1109 spin_unlock(&info->lock);
1110out_fput:
1111 fput(filp);
1112out:
1113 if (sock) {
1114 netlink_detachskb(sock, nc);
1115 } else if (nc) {
1116 dev_kfree_skb(nc);
1117 }
1118 return ret;
1119}
1120
1121asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1122 const struct mq_attr __user *u_mqstat,
1123 struct mq_attr __user *u_omqstat)
1124{
1125 int ret;
1126 struct mq_attr mqstat, omqstat;
1127 struct file *filp;
1128 struct inode *inode;
1129 struct mqueue_inode_info *info;
1130
1131 if (u_mqstat != NULL) {
1132 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1133 return -EFAULT;
1134 if (mqstat.mq_flags & (~O_NONBLOCK))
1135 return -EINVAL;
1136 }
1137
1138 ret = -EBADF;
1139 filp = fget(mqdes);
1140 if (!filp)
1141 goto out;
1142
Josef Sipek6d630792006-12-08 02:37:11 -08001143 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 if (unlikely(filp->f_op != &mqueue_file_operations))
1145 goto out_fput;
1146 info = MQUEUE_I(inode);
1147
1148 spin_lock(&info->lock);
1149
1150 omqstat = info->attr;
1151 omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1152 if (u_mqstat) {
Al Viro73929062008-12-10 06:58:59 -05001153 audit_mq_getsetattr(mqdes, &mqstat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (mqstat.mq_flags & O_NONBLOCK)
1155 filp->f_flags |= O_NONBLOCK;
1156 else
1157 filp->f_flags &= ~O_NONBLOCK;
1158
1159 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1160 }
1161
1162 spin_unlock(&info->lock);
1163
1164 ret = 0;
1165 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1166 sizeof(struct mq_attr)))
1167 ret = -EFAULT;
1168
1169out_fput:
1170 fput(filp);
1171out:
1172 return ret;
1173}
1174
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001175static const struct inode_operations mqueue_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 .lookup = simple_lookup,
1177 .create = mqueue_create,
1178 .unlink = mqueue_unlink,
1179};
1180
Arjan van de Ven9a321442007-02-12 00:55:35 -08001181static const struct file_operations mqueue_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 .flush = mqueue_flush_file,
1183 .poll = mqueue_poll_file,
1184 .read = mqueue_read_file,
1185};
1186
1187static struct super_operations mqueue_super_ops = {
1188 .alloc_inode = mqueue_alloc_inode,
1189 .destroy_inode = mqueue_destroy_inode,
1190 .statfs = simple_statfs,
1191 .delete_inode = mqueue_delete_inode,
1192 .drop_inode = generic_delete_inode,
1193};
1194
1195static struct file_system_type mqueue_fs_type = {
1196 .name = "mqueue",
1197 .get_sb = mqueue_get_sb,
1198 .kill_sb = kill_litter_super,
1199};
1200
Joe Kortyb231cca42008-10-18 20:28:32 -07001201static int msg_max_limit_min = MIN_MSGMAX;
1202static int msg_max_limit_max = MAX_MSGMAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Joe Kortyb231cca42008-10-18 20:28:32 -07001204static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1205static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207static ctl_table mq_sysctls[] = {
1208 {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 .procname = "queues_max",
1210 .data = &queues_max,
1211 .maxlen = sizeof(int),
1212 .mode = 0644,
1213 .proc_handler = &proc_dointvec,
1214 },
1215 {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 .procname = "msg_max",
1217 .data = &msg_max,
1218 .maxlen = sizeof(int),
1219 .mode = 0644,
1220 .proc_handler = &proc_dointvec_minmax,
1221 .extra1 = &msg_max_limit_min,
1222 .extra2 = &msg_max_limit_max,
1223 },
1224 {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 .procname = "msgsize_max",
1226 .data = &msgsize_max,
1227 .maxlen = sizeof(int),
1228 .mode = 0644,
1229 .proc_handler = &proc_dointvec_minmax,
1230 .extra1 = &msg_maxsize_limit_min,
1231 .extra2 = &msg_maxsize_limit_max,
1232 },
1233 { .ctl_name = 0 }
1234};
1235
1236static ctl_table mq_sysctl_dir[] = {
1237 {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 .procname = "mqueue",
1239 .mode = 0555,
1240 .child = mq_sysctls,
1241 },
1242 { .ctl_name = 0 }
1243};
1244
1245static ctl_table mq_sysctl_root[] = {
1246 {
1247 .ctl_name = CTL_FS,
1248 .procname = "fs",
1249 .mode = 0555,
1250 .child = mq_sysctl_dir,
1251 },
1252 { .ctl_name = 0 }
1253};
1254
1255static int __init init_mqueue_fs(void)
1256{
1257 int error;
1258
1259 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1260 sizeof(struct mqueue_inode_info), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001261 SLAB_HWCACHE_ALIGN, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 if (mqueue_inode_cachep == NULL)
1263 return -ENOMEM;
1264
1265 /* ignore failues - they are not fatal */
Eric W. Biederman0b4d4142007-02-14 00:34:09 -08001266 mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
1268 error = register_filesystem(&mqueue_fs_type);
1269 if (error)
1270 goto out_sysctl;
1271
1272 if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1273 error = PTR_ERR(mqueue_mnt);
1274 goto out_filesystem;
1275 }
1276
1277 /* internal initialization - not common for vfs */
1278 queues_count = 0;
1279 spin_lock_init(&mq_lock);
1280
1281 return 0;
1282
1283out_filesystem:
1284 unregister_filesystem(&mqueue_fs_type);
1285out_sysctl:
1286 if (mq_sysctl_table)
1287 unregister_sysctl_table(mq_sysctl_table);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001288 kmem_cache_destroy(mqueue_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 return error;
1290}
1291
1292__initcall(init_mqueue_fs);