ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (wrona@mat.uni.torun.pl)
   6  *
   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
  14 #include <linux/init.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/file.h>
  17 #include <linux/mount.h>
  18 #include <linux/namei.h>
  19 #include <linux/sysctl.h>
  20 #include <linux/poll.h>
  21 #include <linux/mqueue.h>
  22 #include <linux/msg.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/netlink.h>
  25 #include <net/sock.h>
  26 #include "util.h"
  27
  28 #define MQUEUE_MAGIC    0x19800202
  29 #define DIRENT_SIZE     20
  30 #define FILENT_SIZE     80
  31
  32 #define SEND            0
  33 #define RECV            1
  34
  35 #define STATE_NONE      0
  36 #define STATE_PENDING   1
  37 #define STATE_READY     2
  38
  39 /* used by sysctl */
  40 #define FS_MQUEUE       1
  41 #define CTL_QUEUESMAX   2
  42 #define CTL_MSGMAX      3
  43 #define CTL_MSGSIZEMAX  4
  44
  45 /* default values */
  46 #define DFLT_QUEUESMAX  64      /* max number of message queues */
  47 #define DFLT_MSGMAX     40      /* max number of messages in each queue */
  48 #define HARD_MSGMAX     (131072/sizeof(void*))
  49 #define DFLT_MSGSIZEMAX 16384   /* max message size */
  50
  51 #define NOTIFY_COOKIE_LEN       32
  52
  53 struct ext_wait_queue {         /* queue of sleeping tasks */
  54         struct task_struct *task;
  55         struct list_head list;
  56         struct msg_msg *msg;    /* ptr of loaded message */
  57         int state;              /* one of STATE_* values */
  58 };
  59
  60 struct mqueue_inode_info {
  61         spinlock_t lock;
  62         struct inode vfs_inode;
  63         wait_queue_head_t wait_q;
  64
  65         struct msg_msg **messages;
  66         struct mq_attr attr;
  67
  68         struct sigevent notify;
  69         pid_t notify_owner;
  70         struct sock *notify_sock;
  71         struct sk_buff *notify_cookie;
  72
  73         /* for tasks waiting for free space and messages, respectively */
  74         struct ext_wait_queue e_wait_q[2];
  75
  76         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  77 };
  78
  79 static struct inode_operations mqueue_dir_inode_operations;
  80 static struct file_operations mqueue_file_operations;
  81 static struct super_operations mqueue_super_ops;
  82 static void remove_notification(struct mqueue_inode_info *info);
  83
  84 static spinlock_t mq_lock;
  85 static kmem_cache_t *mqueue_inode_cachep;
  86 static struct vfsmount *mqueue_mnt;
  87
  88 static unsigned int queues_count;
  89 static unsigned int queues_max  = DFLT_QUEUESMAX;
  90 static unsigned int msg_max     = DFLT_MSGMAX;
  91 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  92
  93 static struct ctl_table_header * mq_sysctl_table;
  94
  95 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  96 {
  97         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  98 }
  99
 100 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 101                                                         struct mq_attr *attr)
 102 {
 103         struct inode *inode;
 104
 105         inode = new_inode(sb);
 106         if (inode) {
 107                 inode->i_mode = mode;
 108                 inode->i_uid = current->fsuid;
 109                 inode->i_gid = current->fsgid;
 110                 inode->i_blksize = PAGE_CACHE_SIZE;
 111                 inode->i_blocks = 0;
 112                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 113                                 CURRENT_TIME;
 114
 115                 if (S_ISREG(mode)) {
 116                         struct mqueue_inode_info *info;
 117
 118                         inode->i_fop = &mqueue_file_operations;
 119                         inode->i_size = FILENT_SIZE;
 120                         /* mqueue specific info */
 121                         info = MQUEUE_I(inode);
 122                         spin_lock_init(&info->lock);
 123                         init_waitqueue_head(&info->wait_q);
 124                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 125                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 126                         info->notify_owner = 0;
 127                         info->qsize = 0;
 128                         memset(&info->attr, 0, sizeof(info->attr));
 129                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 130                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 131                         if (attr) {
 132                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 133                                 info->attr.mq_msgsize = attr->mq_msgsize;
 134                         }
 135                         info->messages = kmalloc(info->attr.mq_maxmsg * sizeof(struct msg_msg *), GFP_KERNEL);
 136                         if (!info->messages) {
 137                                 make_bad_inode(inode);
 138                                 iput(inode);
 139                                 inode = NULL;
 140                         }
 141                 } else if (S_ISDIR(mode)) {
 142                         inode->i_nlink++;
 143                         /* Some things misbehave if size == 0 on a directory */
 144                         inode->i_size = 2 * DIRENT_SIZE;
 145                         inode->i_op = &mqueue_dir_inode_operations;
 146                         inode->i_fop = &simple_dir_operations;
 147                 }
 148         }
 149         return inode;
 150 }
 151
 152 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 153 {
 154         struct inode *inode;
 155
 156         sb->s_blocksize = PAGE_CACHE_SIZE;
 157         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 158         sb->s_magic = MQUEUE_MAGIC;
 159         sb->s_op = &mqueue_super_ops;
 160
 161         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 162         if (!inode)
 163                 return -ENOMEM;
 164
 165         sb->s_root = d_alloc_root(inode);
 166         if (!sb->s_root) {
 167                 iput(inode);
 168                 return -ENOMEM;
 169         }
 170
 171         return 0;
 172 }
 173
 174 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 175                                          int flags, const char *dev_name,
 176                                          void *data)
 177 {
 178         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 179 }
 180
 181 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 182 {
 183         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 184
 185         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 186                 SLAB_CTOR_CONSTRUCTOR)
 187                 inode_init_once(&p->vfs_inode);
 188 }
 189
 190 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 191 {
 192         struct mqueue_inode_info *ei;
 193
 194         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 195         if (!ei)
 196                 return NULL;
 197         return &ei->vfs_inode;
 198 }
 199
 200 static void mqueue_destroy_inode(struct inode *inode)
 201 {
 202         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 203 }
 204
 205 static void mqueue_delete_inode(struct inode *inode)
 206 {
 207         struct mqueue_inode_info *info;
 208         int i;
 209
 210         if (S_ISDIR(inode->i_mode)) {
 211                 clear_inode(inode);
 212                 return;
 213         }
 214         info = MQUEUE_I(inode);
 215         spin_lock(&info->lock);
 216         for (i = 0; i < info->attr.mq_curmsgs; i++)
 217                 free_msg(info->messages[i]);
 218         kfree(info->messages);
 219         spin_unlock(&info->lock);
 220
 221         clear_inode(inode);
 222
 223         if (info->messages) {
 224                 spin_lock(&mq_lock);
 225                 queues_count--;
 226                 spin_unlock(&mq_lock);
 227         }
 228 }
 229
 230 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 231                                 int mode, struct nameidata *nd)
 232 {
 233         struct inode *inode;
 234         struct mq_attr *attr = dentry->d_fsdata;
 235         int error;
 236
 237         spin_lock(&mq_lock);
 238         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 239                 error = -ENOSPC;
 240                 goto out_lock;
 241         }
 242         queues_count++;
 243         spin_unlock(&mq_lock);
 244
 245         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 246         if (!inode) {
 247                 error = -ENOMEM;
 248                 spin_lock(&mq_lock);
 249                 queues_count--;
 250                 goto out_lock;
 251         }
 252
 253         dir->i_size += DIRENT_SIZE;
 254         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 255
 256         d_instantiate(dentry, inode);
 257         dget(dentry);
 258         return 0;
 259 out_lock:
 260         spin_unlock(&mq_lock);
 261         return error;
 262 }
 263
 264 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 265 {
 266         struct inode *inode = dentry->d_inode;
 267
 268         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 269         dir->i_size -= DIRENT_SIZE;
 270         inode->i_nlink--;
 271         dput(dentry);
 272         return 0;
 273 }
 274
 275 /*
 276 *       This is routine for system read from queue file.
 277 *       To avoid mess with doing here some sort of mq_receive we allow
 278 *       to read only queue size & notification info (the only values
 279 *       that are interesting from user point of view and aren't accessible
 280 *       through std routines)
 281 */
 282 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 283                                 size_t count, loff_t * off)
 284 {
 285         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 286         char buffer[FILENT_SIZE];
 287         size_t slen;
 288         loff_t o;
 289
 290         if (!count)
 291                 return 0;
 292
 293         spin_lock(&info->lock);
 294         snprintf(buffer, sizeof(buffer),
 295                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 296                         info->qsize,
 297                         info->notify_owner ? info->notify.sigev_notify : 0,
 298                         (info->notify_owner &&
 299                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 300                                 info->notify.sigev_signo : 0,
 301                         info->notify_owner);
 302         spin_unlock(&info->lock);
 303         buffer[sizeof(buffer)-1] = '\0';
 304         slen = strlen(buffer)+1;
 305
 306         o = *off;
 307         if (o > slen)
 308                 return 0;
 309
 310         if (o + count > slen)
 311                 count = slen - o;
 312
 313         if (copy_to_user(u_data, buffer + o, count))
 314                 return -EFAULT;
 315
 316         *off = o + count;
 317         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 318         return count;
 319 }
 320
 321 static int mqueue_flush_file(struct file *filp)
 322 {
 323         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 324
 325         spin_lock(&info->lock);
 326         if (current->tgid == info->notify_owner)
 327                 remove_notification(info);
 328
 329         spin_unlock(&info->lock);
 330         return 0;
 331 }
 332
 333 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 334 {
 335         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 336         int retval = 0;
 337
 338         poll_wait(filp, &info->wait_q, poll_tab);
 339
 340         spin_lock(&info->lock);
 341         if (info->attr.mq_curmsgs)
 342                 retval = POLLIN | POLLRDNORM;
 343
 344         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 345                 retval |= POLLOUT | POLLWRNORM;
 346         spin_unlock(&info->lock);
 347
 348         return retval;
 349 }
 350
 351 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 352 static void wq_add(struct mqueue_inode_info *info, int sr,
 353                         struct ext_wait_queue *ewp)
 354 {
 355         struct ext_wait_queue *walk;
 356
 357         ewp->task = current;
 358
 359         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 360                 if (walk->task->static_prio <= current->static_prio) {
 361                         list_add_tail(&ewp->list, &walk->list);
 362                         return;
 363                 }
 364         }
 365         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 366 }
 367
 368 /*
 369  * Puts current task to sleep. Caller must hold queue lock. After return
 370  * lock isn't held.
 371  * sr: SEND or RECV
 372  */
 373 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 374                         long timeout, struct ext_wait_queue *ewp)
 375 {
 376         int retval;
 377         signed long time;
 378
 379         wq_add(info, sr, ewp);
 380
 381         for (;;) {
 382                 set_current_state(TASK_INTERRUPTIBLE);
 383
 384                 spin_unlock(&info->lock);
 385                 time = schedule_timeout(timeout);
 386
 387                 while (ewp->state == STATE_PENDING)
 388                         cpu_relax();
 389
 390                 if (ewp->state == STATE_READY) {
 391                         retval = 0;
 392                         goto out;
 393                 }
 394                 spin_lock(&info->lock);
 395                 if (ewp->state == STATE_READY) {
 396                         retval = 0;
 397                         goto out_unlock;
 398                 }
 399                 if (signal_pending(current)) {
 400                         retval = -ERESTARTSYS;
 401                         break;
 402                 }
 403                 if (time == 0) {
 404                         retval = -ETIMEDOUT;
 405                         break;
 406                 }
 407         }
 408         list_del(&ewp->list);
 409 out_unlock:
 410         spin_unlock(&info->lock);
 411 out:
 412         return retval;
 413 }
 414
 415 /*
 416  * Returns waiting task that should be serviced first or NULL if none exists
 417  */
 418 static struct ext_wait_queue *wq_get_first_waiter(
 419                 struct mqueue_inode_info *info, int sr)
 420 {
 421         struct list_head *ptr;
 422
 423         ptr = info->e_wait_q[sr].list.prev;
 424         if (ptr == &info->e_wait_q[sr].list)
 425                 return NULL;
 426         return list_entry(ptr, struct ext_wait_queue, list);
 427 }
 428
 429 /* Auxiliary functions to manipulate messages' list */
 430 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 431 {
 432         int k;
 433
 434         k = info->attr.mq_curmsgs - 1;
 435         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 436                 info->messages[k + 1] = info->messages[k];
 437                 k--;
 438         }
 439         info->attr.mq_curmsgs++;
 440         info->qsize += ptr->m_ts;
 441         info->messages[k + 1] = ptr;
 442 }
 443
 444 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 445 {
 446         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 447         return info->messages[info->attr.mq_curmsgs];
 448 }
 449
 450 static inline void set_cookie(struct sk_buff *skb, char code)
 451 {
 452         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 453 }
 454
 455 /*
 456  * The next function is only to split too long sys_mq_timedsend
 457  */
 458 static void __do_notify(struct mqueue_inode_info *info)
 459 {
 460         /* notification
 461          * invoked when there is registered process and there isn't process
 462          * waiting synchronously for message AND state of queue changed from
 463          * empty to not empty. Here we are sure that no one is waiting
 464          * synchronously. */
 465         if (info->notify_owner &&
 466             info->attr.mq_curmsgs == 1) {
 467                 struct siginfo sig_i;
 468                 switch (info->notify.sigev_notify) {
 469                 case SIGEV_NONE:
 470                         break;
 471                 case SIGEV_SIGNAL:
 472                         /* sends signal */
 473
 474                         sig_i.si_signo = info->notify.sigev_signo;
 475                         sig_i.si_errno = 0;
 476                         sig_i.si_code = SI_MESGQ;
 477                         sig_i.si_value = info->notify.sigev_value;
 478                         sig_i.si_pid = current->tgid;
 479                         sig_i.si_uid = current->uid;
 480
 481                         kill_proc_info(info->notify.sigev_signo,
 482                                        &sig_i, info->notify_owner);
 483                         break;
 484                 case SIGEV_THREAD:
 485                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 486                         netlink_sendskb(info->notify_sock,
 487                                         info->notify_cookie, 0);
 488                         break;
 489                 }
 490                 /* after notification unregisters process */
 491                 info->notify_owner = 0;
 492         }
 493         wake_up(&info->wait_q);
 494 }
 495
 496 static long prepare_timeout(const struct timespec __user *u_arg)
 497 {
 498         struct timespec ts, nowts;
 499         long timeout;
 500
 501         if (u_arg) {
 502                 if (unlikely(copy_from_user(&ts, u_arg,
 503                                         sizeof(struct timespec))))
 504                         return -EFAULT;
 505
 506                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 507                         || ts.tv_nsec >= NSEC_PER_SEC))
 508                         return -EINVAL;
 509                 nowts = CURRENT_TIME;
 510                 /* first subtract as jiffies can't be too big */
 511                 ts.tv_sec -= nowts.tv_sec;
 512                 if (ts.tv_nsec < nowts.tv_nsec) {
 513                         ts.tv_nsec += NSEC_PER_SEC;
 514                         ts.tv_sec--;
 515                 }
 516                 ts.tv_nsec -= nowts.tv_nsec;
 517                 if (ts.tv_sec < 0)
 518                         return 0;
 519
 520                 timeout = timespec_to_jiffies(&ts) + 1;
 521         } else
 522                 return MAX_SCHEDULE_TIMEOUT;
 523
 524         return timeout;
 525 }
 526
 527 static void remove_notification(struct mqueue_inode_info *info)
 528 {
 529         if (info->notify_owner != 0 &&
 530             info->notify.sigev_notify == SIGEV_THREAD) {
 531                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 532                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 533         }
 534         info->notify_owner = 0;
 535 }
 536
 537 /*
 538  * Invoked when creating a new queue via sys_mq_open
 539  */
 540 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 541                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 542 {
 543         struct file *filp;
 544         struct mq_attr attr;
 545         int ret;
 546
 547         if (u_attr != NULL) {
 548                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 549                         return ERR_PTR(-EFAULT);
 550
 551                 if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0)
 552                         return ERR_PTR(-EINVAL);
 553                 if (capable(CAP_SYS_RESOURCE)) {
 554                         if (attr.mq_maxmsg > HARD_MSGMAX)
 555                                 return ERR_PTR(-EINVAL);
 556                 } else {
 557                         if (attr.mq_maxmsg > msg_max ||
 558                                         attr.mq_msgsize > msgsize_max)
 559                                 return ERR_PTR(-EINVAL);
 560                 }
 561                 /* store for use during create */
 562                 dentry->d_fsdata = &attr;
 563         }
 564
 565         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 566         dentry->d_fsdata = NULL;
 567         if (ret)
 568                 return ERR_PTR(ret);
 569
 570         filp = dentry_open(dentry, mqueue_mnt, oflag);
 571         if (!IS_ERR(filp))
 572                 dget(dentry);
 573
 574         return filp;
 575 }
 576
 577 /* Opens existing queue */
 578 static struct file *do_open(struct dentry *dentry, int oflag)
 579 {
 580 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 581                                         MAY_READ | MAY_WRITE };
 582         struct file *filp;
 583
 584         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 585                 return ERR_PTR(-EINVAL);
 586
 587         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
 588                 return ERR_PTR(-EACCES);
 589
 590         filp = dentry_open(dentry, mqueue_mnt, oflag);
 591
 592         if (!IS_ERR(filp))
 593                 dget(dentry);
 594
 595         return filp;
 596 }
 597
 598 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 599                                 struct mq_attr __user *u_attr)
 600 {
 601         struct dentry *dentry;
 602         struct file *filp;
 603         char *name;
 604         int fd, error;
 605
 606         if (IS_ERR(name = getname(u_name)))
 607                 return PTR_ERR(name);
 608
 609         fd = get_unused_fd();
 610         if (fd < 0)
 611                 goto out_putname;
 612
 613         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 614         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 615         if (IS_ERR(dentry)) {
 616                 error = PTR_ERR(dentry);
 617                 goto out_err;
 618         }
 619         mntget(mqueue_mnt);
 620
 621         if (oflag & O_CREAT) {
 622                 if (dentry->d_inode) {  /* entry already exists */
 623                         filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
 624                                         do_open(dentry, oflag);
 625                 } else {
 626                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 627                                                 oflag, mode, u_attr);
 628                 }
 629         } else
 630                 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
 631                                         ERR_PTR(-ENOENT);
 632
 633         dput(dentry);
 634
 635         if (IS_ERR(filp)) {
 636                 error = PTR_ERR(filp);
 637                 goto out_putfd;
 638         }
 639
 640         set_close_on_exec(fd, 1);
 641         fd_install(fd, filp);
 642         goto out_upsem;
 643
 644 out_putfd:
 645         mntput(mqueue_mnt);
 646         put_unused_fd(fd);
 647 out_err:
 648         fd = error;
 649 out_upsem:
 650         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 651 out_putname:
 652         putname(name);
 653         return fd;
 654 }
 655
 656 asmlinkage long sys_mq_unlink(const char __user *u_name)
 657 {
 658         int err;
 659         char *name;
 660         struct dentry *dentry;
 661         struct inode *inode = NULL;
 662
 663         name = getname(u_name);
 664         if (IS_ERR(name))
 665                 return PTR_ERR(name);
 666
 667         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 668         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 669         if (IS_ERR(dentry)) {
 670                 err = PTR_ERR(dentry);
 671                 goto out_unlock;
 672         }
 673
 674         if (!dentry->d_inode) {
 675                 err = -ENOENT;
 676                 goto out_err;
 677         }
 678
 679         inode = dentry->d_inode;
 680         if (inode)
 681                 atomic_inc(&inode->i_count);
 682
 683         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 684 out_err:
 685         dput(dentry);
 686
 687 out_unlock:
 688         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 689         putname(name);
 690         if (inode)
 691                 iput(inode);
 692
 693         return err;
 694 }
 695
 696 /* Pipelined send and receive functions.
 697  *
 698  * If a receiver finds no waiting message, then it registers itself in the
 699  * list of waiting receivers. A sender checks that list before adding the new
 700  * message into the message array. If there is a waiting receiver, then it
 701  * bypasses the message array and directly hands the message over to the
 702  * receiver.
 703  * The receiver accepts the message and returns without grabbing the queue
 704  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 705  * are necessary. The same algorithm is used for sysv semaphores, see
 706  * ipc/sem.c fore more details.
 707  *
 708  * The same algorithm is used for senders.
 709  */
 710
 711 /* pipelined_send() - send a message directly to the task waiting in
 712  * sys_mq_timedreceive() (without inserting message into a queue).
 713  */
 714 static inline void pipelined_send(struct mqueue_inode_info *info,
 715                                   struct msg_msg *message,
 716                                   struct ext_wait_queue *receiver)
 717 {
 718         receiver->msg = message;
 719         list_del(&receiver->list);
 720         receiver->state = STATE_PENDING;
 721         wake_up_process(receiver->task);
 722         wmb();
 723         receiver->state = STATE_READY;
 724 }
 725
 726 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 727  * gets its message and put to the queue (we have one free place for sure). */
 728 static inline void pipelined_receive(struct mqueue_inode_info *info)
 729 {
 730         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 731
 732         if (!sender) {
 733                 /* for poll */
 734                 wake_up_interruptible(&info->wait_q);
 735                 return;
 736         }
 737         msg_insert(sender->msg, info);
 738         list_del(&sender->list);
 739         sender->state = STATE_PENDING;
 740         wake_up_process(sender->task);
 741         wmb();
 742         sender->state = STATE_READY;
 743 }
 744
 745 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 746         size_t msg_len, unsigned int msg_prio,
 747         const struct timespec __user *u_abs_timeout)
 748 {
 749         struct file *filp;
 750         struct inode *inode;
 751         struct ext_wait_queue wait;
 752         struct ext_wait_queue *receiver;
 753         struct msg_msg *msg_ptr;
 754         struct mqueue_inode_info *info;
 755         long timeout;
 756         int ret;
 757
 758         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 759                 return -EINVAL;
 760
 761         timeout = prepare_timeout(u_abs_timeout);
 762
 763         ret = -EBADF;
 764         filp = fget(mqdes);
 765         if (unlikely(!filp))
 766                 goto out;
 767
 768         inode = filp->f_dentry->d_inode;
 769         if (unlikely(filp->f_op != &mqueue_file_operations))
 770                 goto out_fput;
 771         info = MQUEUE_I(inode);
 772
 773         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 774                 goto out_fput;
 775
 776         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 777                 ret = -EMSGSIZE;
 778                 goto out_fput;
 779         }
 780
 781         /* First try to allocate memory, before doing anything with
 782          * existing queues. */
 783         msg_ptr = load_msg(u_msg_ptr, msg_len);
 784         if (unlikely(IS_ERR(msg_ptr))) {
 785                 ret = PTR_ERR(msg_ptr);
 786                 goto out_fput;
 787         }
 788         msg_ptr->m_ts = msg_len;
 789         msg_ptr->m_type = msg_prio;
 790
 791         spin_lock(&info->lock);
 792
 793         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 794                 if (filp->f_flags & O_NONBLOCK) {
 795                         spin_unlock(&info->lock);
 796                         ret = -EAGAIN;
 797                 } else if (unlikely(timeout < 0)) {
 798                         spin_unlock(&info->lock);
 799                         ret = timeout;
 800                 } else {
 801                         wait.task = current;
 802                         wait.msg = (void *) msg_ptr;
 803                         wait.state = STATE_NONE;
 804                         ret = wq_sleep(info, SEND, timeout, &wait);
 805                 }
 806                 if (ret < 0)
 807                         free_msg(msg_ptr);
 808         } else {
 809                 receiver = wq_get_first_waiter(info, RECV);
 810                 if (receiver) {
 811                         pipelined_send(info, msg_ptr, receiver);
 812                 } else {
 813                         /* adds message to the queue */
 814                         msg_insert(msg_ptr, info);
 815                         __do_notify(info);
 816                 }
 817                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 818                                 CURRENT_TIME;
 819                 spin_unlock(&info->lock);
 820                 ret = 0;
 821         }
 822 out_fput:
 823         fput(filp);
 824 out:
 825         return ret;
 826 }
 827
 828 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 829         size_t msg_len, unsigned int __user *u_msg_prio,
 830         const struct timespec __user *u_abs_timeout)
 831 {
 832         long timeout;
 833         ssize_t ret;
 834         struct msg_msg *msg_ptr;
 835         struct file *filp;
 836         struct inode *inode;
 837         struct mqueue_inode_info *info;
 838         struct ext_wait_queue wait;
 839
 840         timeout = prepare_timeout(u_abs_timeout);
 841
 842         ret = -EBADF;
 843         filp = fget(mqdes);
 844         if (unlikely(!filp))
 845                 goto out;
 846
 847         inode = filp->f_dentry->d_inode;
 848         if (unlikely(filp->f_op != &mqueue_file_operations))
 849                 goto out_fput;
 850         info = MQUEUE_I(inode);
 851
 852         if (unlikely(!(filp->f_mode & FMODE_READ)))
 853                 goto out_fput;
 854
 855         /* checks if buffer is big enough */
 856         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 857                 ret = -EMSGSIZE;
 858                 goto out_fput;
 859         }
 860
 861         spin_lock(&info->lock);
 862         if (info->attr.mq_curmsgs == 0) {
 863                 if (filp->f_flags & O_NONBLOCK) {
 864                         spin_unlock(&info->lock);
 865                         ret = -EAGAIN;
 866                         msg_ptr = NULL;
 867                 } else if (unlikely(timeout < 0)) {
 868                         spin_unlock(&info->lock);
 869                         ret = timeout;
 870                         msg_ptr = NULL;
 871                 } else {
 872                         wait.task = current;
 873                         wait.state = STATE_NONE;
 874                         ret = wq_sleep(info, RECV, timeout, &wait);
 875                         msg_ptr = wait.msg;
 876                 }
 877         } else {
 878                 msg_ptr = msg_get(info);
 879
 880                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 881                                 CURRENT_TIME;
 882
 883                 /* There is now free space in queue. */
 884                 pipelined_receive(info);
 885                 spin_unlock(&info->lock);
 886                 ret = 0;
 887         }
 888         if (ret == 0) {
 889                 ret = msg_ptr->m_ts;
 890
 891                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 892                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 893                         ret = -EFAULT;
 894                 }
 895                 free_msg(msg_ptr);
 896         }
 897 out_fput:
 898         fput(filp);
 899 out:
 900         return ret;
 901 }
 902
 903 /*
 904  * Notes: the case when user wants us to deregister (with NULL as pointer)
 905  * and he isn't currently owner of notification, will be silently discarded.
 906  * It isn't explicitly defined in the POSIX.
 907  */
 908 asmlinkage long sys_mq_notify(mqd_t mqdes,
 909                                 const struct sigevent __user *u_notification)
 910 {
 911         int ret;
 912         struct file *filp;
 913         struct sock *sock;
 914         struct inode *inode;
 915         struct sigevent notification;
 916         struct mqueue_inode_info *info;
 917         struct sk_buff *nc;
 918
 919         nc = NULL;
 920         sock = NULL;
 921         if (u_notification != NULL) {
 922                 if (copy_from_user(&notification, u_notification,
 923                                         sizeof(struct sigevent)))
 924                         return -EFAULT;
 925
 926                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 927                              notification.sigev_notify != SIGEV_SIGNAL &&
 928                              notification.sigev_notify != SIGEV_THREAD))
 929                         return -EINVAL;
 930                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 931                         (notification.sigev_signo < 0 ||
 932                          notification.sigev_signo > _NSIG)) {
 933                         return -EINVAL;
 934                 }
 935                 if (notification.sigev_notify == SIGEV_THREAD) {
 936                         /* create the notify skb */
 937                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
 938                         ret = -ENOMEM;
 939                         if (!nc)
 940                                 goto out;
 941                         ret = -EFAULT;
 942                         if (copy_from_user(nc->data,
 943                                         notification.sigev_value.sival_ptr,
 944                                         NOTIFY_COOKIE_LEN)) {
 945                                 goto out;
 946                         }
 947
 948                         /* TODO: add a header? */
 949                         skb_put(nc, NOTIFY_COOKIE_LEN);
 950                         /* and attach it to the socket */
 951 retry:
 952                         filp = fget(notification.sigev_signo);
 953                         ret = -EBADF;
 954                         if (!filp)
 955                                 goto out;
 956                         sock = netlink_getsockbyfilp(filp);
 957                         fput(filp);
 958                         if (IS_ERR(sock)) {
 959                                 ret = PTR_ERR(sock);
 960                                 sock = NULL;
 961                                 goto out;
 962                         }
 963
 964                         ret = netlink_attachskb(sock, nc, 0, MAX_SCHEDULE_TIMEOUT);
 965                         if (ret == 1)
 966                                 goto retry;
 967                         if (ret) {
 968                                 sock = NULL;
 969                                 nc = NULL;
 970                                 goto out;
 971                         }
 972                 }
 973         }
 974
 975         ret = -EBADF;
 976         filp = fget(mqdes);
 977         if (!filp)
 978                 goto out;
 979
 980         inode = filp->f_dentry->d_inode;
 981         if (unlikely(filp->f_op != &mqueue_file_operations))
 982                 goto out_fput;
 983         info = MQUEUE_I(inode);
 984
 985         ret = 0;
 986         spin_lock(&info->lock);
 987         if (u_notification == NULL) {
 988                 if (info->notify_owner == current->tgid) {
 989                         remove_notification(info);
 990                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
 991                 }
 992         } else if (info->notify_owner != 0) {
 993                 ret = -EBUSY;
 994         } else {
 995                 switch (notification.sigev_notify) {
 996                 case SIGEV_NONE:
 997                         info->notify.sigev_notify = SIGEV_NONE;
 998                         break;
 999                 case SIGEV_THREAD:
1000                         info->notify_sock = sock;
1001                         info->notify_cookie = nc;
1002                         sock = NULL;
1003                         nc = NULL;
1004                         info->notify.sigev_notify = SIGEV_THREAD;
1005                         break;
1006                 case SIGEV_SIGNAL:
1007                         info->notify.sigev_signo = notification.sigev_signo;
1008                         info->notify.sigev_value = notification.sigev_value;
1009                         info->notify.sigev_notify = SIGEV_SIGNAL;
1010                         break;
1011                 }
1012                 info->notify_owner = current->tgid;
1013                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1014         }
1015         spin_unlock(&info->lock);
1016 out_fput:
1017         fput(filp);
1018 out:
1019         if (sock) {
1020                 netlink_detachskb(sock, nc);
1021         } else if (nc) {
1022                 dev_kfree_skb(nc);
1023         }
1024         return ret;
1025 }
1026
1027 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1028                         const struct mq_attr __user *u_mqstat,
1029                         struct mq_attr __user *u_omqstat)
1030 {
1031         int ret;
1032         struct mq_attr mqstat, omqstat;
1033         struct file *filp;
1034         struct inode *inode;
1035         struct mqueue_inode_info *info;
1036
1037         if (u_mqstat != NULL) {
1038                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1039                         return -EFAULT;
1040                 if (mqstat.mq_flags & (~O_NONBLOCK))
1041                         return -EINVAL;
1042         }
1043
1044         ret = -EBADF;
1045         filp = fget(mqdes);
1046         if (!filp)
1047                 goto out;
1048
1049         inode = filp->f_dentry->d_inode;
1050         if (unlikely(filp->f_op != &mqueue_file_operations))
1051                 goto out_fput;
1052         info = MQUEUE_I(inode);
1053
1054         spin_lock(&info->lock);
1055
1056         omqstat = info->attr;
1057         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1058         if (u_mqstat) {
1059                 if (mqstat.mq_flags & O_NONBLOCK)
1060                         filp->f_flags |= O_NONBLOCK;
1061                 else
1062                         filp->f_flags &= ~O_NONBLOCK;
1063
1064                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1065         }
1066
1067         spin_unlock(&info->lock);
1068
1069         ret = 0;
1070         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1071                                                 sizeof(struct mq_attr)))
1072                 ret = -EFAULT;
1073
1074 out_fput:
1075         fput(filp);
1076 out:
1077         return ret;
1078 }
1079
1080 static struct inode_operations mqueue_dir_inode_operations = {
1081         .lookup = simple_lookup,
1082         .create = mqueue_create,
1083         .unlink = mqueue_unlink,
1084 };
1085
1086 static struct file_operations mqueue_file_operations = {
1087         .flush = mqueue_flush_file,
1088         .poll = mqueue_poll_file,
1089         .read = mqueue_read_file,
1090 };
1091
1092 static struct super_operations mqueue_super_ops = {
1093         .alloc_inode = mqueue_alloc_inode,
1094         .destroy_inode = mqueue_destroy_inode,
1095         .statfs = simple_statfs,
1096         .delete_inode = mqueue_delete_inode,
1097         .drop_inode = generic_delete_inode,
1098 };
1099
1100 static struct file_system_type mqueue_fs_type = {
1101         .name = "mqueue",
1102         .get_sb = mqueue_get_sb,
1103         .kill_sb = kill_litter_super,
1104 };
1105
1106 static int msg_max_limit_min = DFLT_MSGMAX;
1107 static int msg_max_limit_max = HARD_MSGMAX;
1108
1109 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1110 static int msg_maxsize_limit_max = INT_MAX;
1111
1112 static ctl_table mq_sysctls[] = {
1113         {
1114                 .ctl_name       = CTL_QUEUESMAX,
1115                 .procname       = "queues_max",
1116                 .data           = &queues_max,
1117                 .maxlen         = sizeof(int),
1118                 .mode           = 0644,
1119                 .proc_handler   = &proc_dointvec,
1120         },
1121         {
1122                 .ctl_name       = CTL_MSGMAX,
1123                 .procname       = "msg_max",
1124                 .data           = &msg_max,
1125                 .maxlen         = sizeof(int),
1126                 .mode           = 0644,
1127                 .proc_handler   = &proc_dointvec_minmax,
1128                 .extra1         = &msg_max_limit_min,
1129                 .extra2         = &msg_max_limit_max,
1130         },
1131         {
1132                 .ctl_name       = CTL_MSGSIZEMAX,
1133                 .procname       = "msgsize_max",
1134                 .data           = &msgsize_max,
1135                 .maxlen         = sizeof(int),
1136                 .mode           = 0644,
1137                 .proc_handler   = &proc_dointvec_minmax,
1138                 .extra1         = &msg_maxsize_limit_min,
1139                 .extra2         = &msg_maxsize_limit_max,
1140         },
1141         { .ctl_name = 0 }
1142 };
1143
1144 static ctl_table mq_sysctl_dir[] = {
1145         {
1146                 .ctl_name       = FS_MQUEUE,
1147                 .procname       = "mqueue",
1148                 .mode           = 0555,
1149                 .child          = mq_sysctls,
1150         },
1151         { .ctl_name = 0 }
1152 };
1153
1154 static ctl_table mq_sysctl_root[] = {
1155         {
1156                 .ctl_name       = CTL_FS,
1157                 .procname       = "fs",
1158                 .mode           = 0555,
1159                 .child          = mq_sysctl_dir,
1160         },
1161         { .ctl_name = 0 }
1162 };
1163
1164 static int __init init_mqueue_fs(void)
1165 {
1166         int error;
1167
1168         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1169                                 sizeof(struct mqueue_inode_info), 0,
1170                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1171         if (mqueue_inode_cachep == NULL)
1172                 return -ENOMEM;
1173
1174         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1175         if (!mq_sysctl_table) {
1176                 error = -ENOMEM;
1177                 goto out_cache;
1178         }
1179
1180         error = register_filesystem(&mqueue_fs_type);
1181         if (error)
1182                 goto out_sysctl;
1183
1184         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1185                 error = PTR_ERR(mqueue_mnt);
1186                 goto out_filesystem;
1187         }
1188
1189         /* internal initialization - not common for vfs */
1190         queues_count = 0;
1191         spin_lock_init(&mq_lock);
1192
1193         return 0;
1194
1195 out_filesystem:
1196         unregister_filesystem(&mqueue_fs_type);
1197 out_sysctl:
1198         unregister_sysctl_table(mq_sysctl_table);
1199 out_cache:
1200         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1201                 printk(KERN_INFO
1202                         "mqueue_inode_cache: not all structures were freed\n");
1203         }
1204         return error;
1205 }
1206
1207 __initcall(init_mqueue_fs);