ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    ([email protected])
   5  *                          Michal Wronski          ([email protected])
   6  *
   7  * Spinlocks:               Mohamed Abbas           ([email protected])
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          ([email protected])
  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 <linux/syscalls.h>
  26 #include <net/sock.h>
  27 #include "util.h"
  28
  29 #define MQUEUE_MAGIC    0x19800202
  30 #define DIRENT_SIZE     20
  31 #define FILENT_SIZE     80
  32
  33 #define SEND            0
  34 #define RECV            1
  35
  36 #define STATE_NONE      0
  37 #define STATE_PENDING   1
  38 #define STATE_READY     2
  39
  40 /* used by sysctl */
  41 #define FS_MQUEUE       1
  42 #define CTL_QUEUESMAX   2
  43 #define CTL_MSGMAX      3
  44 #define CTL_MSGSIZEMAX  4
  45
  46 /* default values */
  47 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  48 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  49 #define HARD_MSGMAX     (131072/sizeof(void*))
  50 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  51
  52 #define NOTIFY_COOKIE_LEN       32
  53
  54 struct ext_wait_queue {         /* queue of sleeping tasks */
  55         struct task_struct *task;
  56         struct list_head list;
  57         struct msg_msg *msg;    /* ptr of loaded message */
  58         int state;              /* one of STATE_* values */
  59 };
  60
  61 struct mqueue_inode_info {
  62         spinlock_t lock;
  63         struct inode vfs_inode;
  64         wait_queue_head_t wait_q;
  65
  66         struct msg_msg **messages;
  67         struct mq_attr attr;
  68
  69         struct sigevent notify;
  70         pid_t notify_owner;
  71         struct user_struct *user;       /* user who created, for accouting */
  72         struct sock *notify_sock;
  73         struct sk_buff *notify_cookie;
  74
  75         /* for tasks waiting for free space and messages, respectively */
  76         struct ext_wait_queue e_wait_q[2];
  77
  78         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  79 };
  80
  81 static struct inode_operations mqueue_dir_inode_operations;
  82 static struct file_operations mqueue_file_operations;
  83 static struct super_operations mqueue_super_ops;
  84 static void remove_notification(struct mqueue_inode_info *info);
  85
  86 static spinlock_t mq_lock;
  87 static kmem_cache_t *mqueue_inode_cachep;
  88 static struct vfsmount *mqueue_mnt;
  89
  90 static unsigned int queues_count;
  91 static unsigned int queues_max  = DFLT_QUEUESMAX;
  92 static unsigned int msg_max     = DFLT_MSGMAX;
  93 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  94
  95 static struct ctl_table_header * mq_sysctl_table;
  96
  97 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  98 {
  99         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 100 }
 101
 102 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 103                                                         struct mq_attr *attr)
 104 {
 105         struct inode *inode;
 106
 107         inode = new_inode(sb);
 108         if (inode) {
 109                 inode->i_mode = mode;
 110                 inode->i_uid = current->fsuid;
 111                 inode->i_gid = current->fsgid;
 112                 inode->i_blksize = PAGE_CACHE_SIZE;
 113                 inode->i_blocks = 0;
 114                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 115                                 CURRENT_TIME;
 116
 117                 if (S_ISREG(mode)) {
 118                         struct mqueue_inode_info *info;
 119                         struct task_struct *p = current;
 120                         struct user_struct *u = p->user;
 121                         unsigned long mq_bytes, mq_msg_tblsz;
 122
 123                         inode->i_fop = &mqueue_file_operations;
 124                         inode->i_size = FILENT_SIZE;
 125                         /* mqueue specific info */
 126                         info = MQUEUE_I(inode);
 127                         spin_lock_init(&info->lock);
 128                         init_waitqueue_head(&info->wait_q);
 129                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 130                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 131                         info->messages = NULL;
 132                         info->notify_owner = 0;
 133                         info->qsize = 0;
 134                         info->user = NULL;      /* set when all is ok */
 135                         memset(&info->attr, 0, sizeof(info->attr));
 136                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 137                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 138                         if (attr) {
 139                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 140                                 info->attr.mq_msgsize = attr->mq_msgsize;
 141                         }
 142                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 143                         mq_bytes = (mq_msg_tblsz +
 144                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 145
 146                         spin_lock(&mq_lock);
 147                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 148                             u->mq_bytes + mq_bytes >
 149                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 150                                 spin_unlock(&mq_lock);
 151                                 goto out_inode;
 152                         }
 153                         u->mq_bytes += mq_bytes;
 154                         spin_unlock(&mq_lock);
 155
 156                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 157                         if (!info->messages) {
 158                                 spin_lock(&mq_lock);
 159                                 u->mq_bytes -= mq_bytes;
 160                                 spin_unlock(&mq_lock);
 161                                 goto out_inode;
 162                         }
 163                         /* all is ok */
 164                         info->user = get_uid(u);
 165                 } else if (S_ISDIR(mode)) {
 166                         inode->i_nlink++;
 167                         /* Some things misbehave if size == 0 on a directory */
 168                         inode->i_size = 2 * DIRENT_SIZE;
 169                         inode->i_op = &mqueue_dir_inode_operations;
 170                         inode->i_fop = &simple_dir_operations;
 171                 }
 172         }
 173         return inode;
 174 out_inode:
 175         make_bad_inode(inode);
 176         iput(inode);
 177         return NULL;
 178 }
 179
 180 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 181 {
 182         struct inode *inode;
 183
 184         sb->s_blocksize = PAGE_CACHE_SIZE;
 185         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 186         sb->s_magic = MQUEUE_MAGIC;
 187         sb->s_op = &mqueue_super_ops;
 188
 189         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 190         if (!inode)
 191                 return -ENOMEM;
 192
 193         sb->s_root = d_alloc_root(inode);
 194         if (!sb->s_root) {
 195                 iput(inode);
 196                 return -ENOMEM;
 197         }
 198
 199         return 0;
 200 }
 201
 202 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 203                                          int flags, const char *dev_name,
 204                                          void *data)
 205 {
 206         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 207 }
 208
 209 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 210 {
 211         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 212
 213         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 214                 SLAB_CTOR_CONSTRUCTOR)
 215                 inode_init_once(&p->vfs_inode);
 216 }
 217
 218 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 219 {
 220         struct mqueue_inode_info *ei;
 221
 222         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 223         if (!ei)
 224                 return NULL;
 225         return &ei->vfs_inode;
 226 }
 227
 228 static void mqueue_destroy_inode(struct inode *inode)
 229 {
 230         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 231 }
 232
 233 static void mqueue_delete_inode(struct inode *inode)
 234 {
 235         struct mqueue_inode_info *info;
 236         struct user_struct *user;
 237         unsigned long mq_bytes;
 238         int i;
 239
 240         if (S_ISDIR(inode->i_mode)) {
 241                 clear_inode(inode);
 242                 return;
 243         }
 244         info = MQUEUE_I(inode);
 245         spin_lock(&info->lock);
 246         for (i = 0; i < info->attr.mq_curmsgs; i++)
 247                 free_msg(info->messages[i]);
 248         kfree(info->messages);
 249         spin_unlock(&info->lock);
 250
 251         clear_inode(inode);
 252
 253         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 254                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 255         user = info->user;
 256         if (user) {
 257                 spin_lock(&mq_lock);
 258                 user->mq_bytes -= mq_bytes;
 259                 queues_count--;
 260                 spin_unlock(&mq_lock);
 261                 free_uid(user);
 262         }
 263 }
 264
 265 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 266                                 int mode, struct nameidata *nd)
 267 {
 268         struct inode *inode;
 269         struct mq_attr *attr = dentry->d_fsdata;
 270         int error;
 271
 272         spin_lock(&mq_lock);
 273         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 274                 error = -ENOSPC;
 275                 goto out_lock;
 276         }
 277         queues_count++;
 278         spin_unlock(&mq_lock);
 279
 280         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 281         if (!inode) {
 282                 error = -ENOMEM;
 283                 spin_lock(&mq_lock);
 284                 queues_count--;
 285                 goto out_lock;
 286         }
 287
 288         dir->i_size += DIRENT_SIZE;
 289         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 290
 291         d_instantiate(dentry, inode);
 292         dget(dentry);
 293         return 0;
 294 out_lock:
 295         spin_unlock(&mq_lock);
 296         return error;
 297 }
 298
 299 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 300 {
 301         struct inode *inode = dentry->d_inode;
 302
 303         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 304         dir->i_size -= DIRENT_SIZE;
 305         inode->i_nlink--;
 306         dput(dentry);
 307         return 0;
 308 }
 309
 310 /*
 311 *       This is routine for system read from queue file.
 312 *       To avoid mess with doing here some sort of mq_receive we allow
 313 *       to read only queue size & notification info (the only values
 314 *       that are interesting from user point of view and aren't accessible
 315 *       through std routines)
 316 */
 317 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 318                                 size_t count, loff_t * off)
 319 {
 320         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 321         char buffer[FILENT_SIZE];
 322         size_t slen;
 323         loff_t o;
 324
 325         if (!count)
 326                 return 0;
 327
 328         spin_lock(&info->lock);
 329         snprintf(buffer, sizeof(buffer),
 330                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 331                         info->qsize,
 332                         info->notify_owner ? info->notify.sigev_notify : 0,
 333                         (info->notify_owner &&
 334                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 335                                 info->notify.sigev_signo : 0,
 336                         info->notify_owner);
 337         spin_unlock(&info->lock);
 338         buffer[sizeof(buffer)-1] = '\0';
 339         slen = strlen(buffer)+1;
 340
 341         o = *off;
 342         if (o > slen)
 343                 return 0;
 344
 345         if (o + count > slen)
 346                 count = slen - o;
 347
 348         if (copy_to_user(u_data, buffer + o, count))
 349                 return -EFAULT;
 350
 351         *off = o + count;
 352         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 353         return count;
 354 }
 355
 356 static int mqueue_flush_file(struct file *filp)
 357 {
 358         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 359
 360         spin_lock(&info->lock);
 361         if (current->tgid == info->notify_owner)
 362                 remove_notification(info);
 363
 364         spin_unlock(&info->lock);
 365         return 0;
 366 }
 367
 368 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 369 {
 370         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 371         int retval = 0;
 372
 373         poll_wait(filp, &info->wait_q, poll_tab);
 374
 375         spin_lock(&info->lock);
 376         if (info->attr.mq_curmsgs)
 377                 retval = POLLIN | POLLRDNORM;
 378
 379         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 380                 retval |= POLLOUT | POLLWRNORM;
 381         spin_unlock(&info->lock);
 382
 383         return retval;
 384 }
 385
 386 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 387 static void wq_add(struct mqueue_inode_info *info, int sr,
 388                         struct ext_wait_queue *ewp)
 389 {
 390         struct ext_wait_queue *walk;
 391
 392         ewp->task = current;
 393
 394         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 395                 if (walk->task->static_prio <= current->static_prio) {
 396                         list_add_tail(&ewp->list, &walk->list);
 397                         return;
 398                 }
 399         }
 400         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 401 }
 402
 403 /*
 404  * Puts current task to sleep. Caller must hold queue lock. After return
 405  * lock isn't held.
 406  * sr: SEND or RECV
 407  */
 408 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 409                         long timeout, struct ext_wait_queue *ewp)
 410 {
 411         int retval;
 412         signed long time;
 413
 414         wq_add(info, sr, ewp);
 415
 416         for (;;) {
 417                 set_current_state(TASK_INTERRUPTIBLE);
 418
 419                 spin_unlock(&info->lock);
 420                 time = schedule_timeout(timeout);
 421
 422                 while (ewp->state == STATE_PENDING)
 423                         cpu_relax();
 424
 425                 if (ewp->state == STATE_READY) {
 426                         retval = 0;
 427                         goto out;
 428                 }
 429                 spin_lock(&info->lock);
 430                 if (ewp->state == STATE_READY) {
 431                         retval = 0;
 432                         goto out_unlock;
 433                 }
 434                 if (signal_pending(current)) {
 435                         retval = -ERESTARTSYS;
 436                         break;
 437                 }
 438                 if (time == 0) {
 439                         retval = -ETIMEDOUT;
 440                         break;
 441                 }
 442         }
 443         list_del(&ewp->list);
 444 out_unlock:
 445         spin_unlock(&info->lock);
 446 out:
 447         return retval;
 448 }
 449
 450 /*
 451  * Returns waiting task that should be serviced first or NULL if none exists
 452  */
 453 static struct ext_wait_queue *wq_get_first_waiter(
 454                 struct mqueue_inode_info *info, int sr)
 455 {
 456         struct list_head *ptr;
 457
 458         ptr = info->e_wait_q[sr].list.prev;
 459         if (ptr == &info->e_wait_q[sr].list)
 460                 return NULL;
 461         return list_entry(ptr, struct ext_wait_queue, list);
 462 }
 463
 464 /* Auxiliary functions to manipulate messages' list */
 465 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 466 {
 467         int k;
 468
 469         k = info->attr.mq_curmsgs - 1;
 470         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 471                 info->messages[k + 1] = info->messages[k];
 472                 k--;
 473         }
 474         info->attr.mq_curmsgs++;
 475         info->qsize += ptr->m_ts;
 476         info->messages[k + 1] = ptr;
 477 }
 478
 479 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 480 {
 481         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 482         return info->messages[info->attr.mq_curmsgs];
 483 }
 484
 485 static inline void set_cookie(struct sk_buff *skb, char code)
 486 {
 487         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 488 }
 489
 490 /*
 491  * The next function is only to split too long sys_mq_timedsend
 492  */
 493 static void __do_notify(struct mqueue_inode_info *info)
 494 {
 495         /* notification
 496          * invoked when there is registered process and there isn't process
 497          * waiting synchronously for message AND state of queue changed from
 498          * empty to not empty. Here we are sure that no one is waiting
 499          * synchronously. */
 500         if (info->notify_owner &&
 501             info->attr.mq_curmsgs == 1) {
 502                 struct siginfo sig_i;
 503                 switch (info->notify.sigev_notify) {
 504                 case SIGEV_NONE:
 505                         break;
 506                 case SIGEV_SIGNAL:
 507                         /* sends signal */
 508
 509                         sig_i.si_signo = info->notify.sigev_signo;
 510                         sig_i.si_errno = 0;
 511                         sig_i.si_code = SI_MESGQ;
 512                         sig_i.si_value = info->notify.sigev_value;
 513                         sig_i.si_pid = current->tgid;
 514                         sig_i.si_uid = current->uid;
 515
 516                         kill_proc_info(info->notify.sigev_signo,
 517                                        &sig_i, info->notify_owner);
 518                         break;
 519                 case SIGEV_THREAD:
 520                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 521                         netlink_sendskb(info->notify_sock,
 522                                         info->notify_cookie, 0);
 523                         break;
 524                 }
 525                 /* after notification unregisters process */
 526                 info->notify_owner = 0;
 527         }
 528         wake_up(&info->wait_q);
 529 }
 530
 531 static long prepare_timeout(const struct timespec __user *u_arg)
 532 {
 533         struct timespec ts, nowts;
 534         long timeout;
 535
 536         if (u_arg) {
 537                 if (unlikely(copy_from_user(&ts, u_arg,
 538                                         sizeof(struct timespec))))
 539                         return -EFAULT;
 540
 541                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 542                         || ts.tv_nsec >= NSEC_PER_SEC))
 543                         return -EINVAL;
 544                 nowts = CURRENT_TIME;
 545                 /* first subtract as jiffies can't be too big */
 546                 ts.tv_sec -= nowts.tv_sec;
 547                 if (ts.tv_nsec < nowts.tv_nsec) {
 548                         ts.tv_nsec += NSEC_PER_SEC;
 549                         ts.tv_sec--;
 550                 }
 551                 ts.tv_nsec -= nowts.tv_nsec;
 552                 if (ts.tv_sec < 0)
 553                         return 0;
 554
 555                 timeout = timespec_to_jiffies(&ts) + 1;
 556         } else
 557                 return MAX_SCHEDULE_TIMEOUT;
 558
 559         return timeout;
 560 }
 561
 562 static void remove_notification(struct mqueue_inode_info *info)
 563 {
 564         if (info->notify_owner != 0 &&
 565             info->notify.sigev_notify == SIGEV_THREAD) {
 566                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 567                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 568         }
 569         info->notify_owner = 0;
 570 }
 571
 572 static int mq_attr_ok(struct mq_attr *attr)
 573 {
 574         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 575                 return 0;
 576         if (capable(CAP_SYS_RESOURCE)) {
 577                 if (attr->mq_maxmsg > HARD_MSGMAX)
 578                         return 0;
 579         } else {
 580                 if (attr->mq_maxmsg > msg_max ||
 581                                 attr->mq_msgsize > msgsize_max)
 582                         return 0;
 583         }
 584         /* check for overflow */
 585         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 586                 return 0;
 587         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 588             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 589             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 590                 return 0;
 591         return 1;
 592 }
 593
 594 /*
 595  * Invoked when creating a new queue via sys_mq_open
 596  */
 597 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 598                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 599 {
 600         struct file *filp;
 601         struct mq_attr attr;
 602         int ret;
 603
 604         if (u_attr != NULL) {
 605                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 606                         return ERR_PTR(-EFAULT);
 607                 if (!mq_attr_ok(&attr))
 608                         return ERR_PTR(-EINVAL);
 609                 /* store for use during create */
 610                 dentry->d_fsdata = &attr;
 611         }
 612
 613         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 614         dentry->d_fsdata = NULL;
 615         if (ret)
 616                 return ERR_PTR(ret);
 617
 618         filp = dentry_open(dentry, mqueue_mnt, oflag);
 619         if (!IS_ERR(filp))
 620                 dget(dentry);
 621
 622         return filp;
 623 }
 624
 625 /* Opens existing queue */
 626 static struct file *do_open(struct dentry *dentry, int oflag)
 627 {
 628 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 629                                         MAY_READ | MAY_WRITE };
 630         struct file *filp;
 631
 632         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 633                 return ERR_PTR(-EINVAL);
 634
 635         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
 636                 return ERR_PTR(-EACCES);
 637
 638         filp = dentry_open(dentry, mqueue_mnt, oflag);
 639
 640         if (!IS_ERR(filp))
 641                 dget(dentry);
 642
 643         return filp;
 644 }
 645
 646 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 647                                 struct mq_attr __user *u_attr)
 648 {
 649         struct dentry *dentry;
 650         struct file *filp;
 651         char *name;
 652         int fd, error;
 653
 654         if (IS_ERR(name = getname(u_name)))
 655                 return PTR_ERR(name);
 656
 657         fd = get_unused_fd();
 658         if (fd < 0)
 659                 goto out_putname;
 660
 661         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 662         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 663         if (IS_ERR(dentry)) {
 664                 error = PTR_ERR(dentry);
 665                 goto out_err;
 666         }
 667         mntget(mqueue_mnt);
 668
 669         if (oflag & O_CREAT) {
 670                 if (dentry->d_inode) {  /* entry already exists */
 671                         filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
 672                                         do_open(dentry, oflag);
 673                 } else {
 674                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 675                                                 oflag, mode, u_attr);
 676                 }
 677         } else
 678                 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
 679                                         ERR_PTR(-ENOENT);
 680
 681         dput(dentry);
 682
 683         if (IS_ERR(filp)) {
 684                 error = PTR_ERR(filp);
 685                 goto out_putfd;
 686         }
 687
 688         set_close_on_exec(fd, 1);
 689         fd_install(fd, filp);
 690         goto out_upsem;
 691
 692 out_putfd:
 693         mntput(mqueue_mnt);
 694         put_unused_fd(fd);
 695 out_err:
 696         fd = error;
 697 out_upsem:
 698         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 699 out_putname:
 700         putname(name);
 701         return fd;
 702 }
 703
 704 asmlinkage long sys_mq_unlink(const char __user *u_name)
 705 {
 706         int err;
 707         char *name;
 708         struct dentry *dentry;
 709         struct inode *inode = NULL;
 710
 711         name = getname(u_name);
 712         if (IS_ERR(name))
 713                 return PTR_ERR(name);
 714
 715         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 716         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 717         if (IS_ERR(dentry)) {
 718                 err = PTR_ERR(dentry);
 719                 goto out_unlock;
 720         }
 721
 722         if (!dentry->d_inode) {
 723                 err = -ENOENT;
 724                 goto out_err;
 725         }
 726
 727         inode = dentry->d_inode;
 728         if (inode)
 729                 atomic_inc(&inode->i_count);
 730
 731         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 732 out_err:
 733         dput(dentry);
 734
 735 out_unlock:
 736         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 737         putname(name);
 738         if (inode)
 739                 iput(inode);
 740
 741         return err;
 742 }
 743
 744 /* Pipelined send and receive functions.
 745  *
 746  * If a receiver finds no waiting message, then it registers itself in the
 747  * list of waiting receivers. A sender checks that list before adding the new
 748  * message into the message array. If there is a waiting receiver, then it
 749  * bypasses the message array and directly hands the message over to the
 750  * receiver.
 751  * The receiver accepts the message and returns without grabbing the queue
 752  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 753  * are necessary. The same algorithm is used for sysv semaphores, see
 754  * ipc/sem.c fore more details.
 755  *
 756  * The same algorithm is used for senders.
 757  */
 758
 759 /* pipelined_send() - send a message directly to the task waiting in
 760  * sys_mq_timedreceive() (without inserting message into a queue).
 761  */
 762 static inline void pipelined_send(struct mqueue_inode_info *info,
 763                                   struct msg_msg *message,
 764                                   struct ext_wait_queue *receiver)
 765 {
 766         receiver->msg = message;
 767         list_del(&receiver->list);
 768         receiver->state = STATE_PENDING;
 769         wake_up_process(receiver->task);
 770         wmb();
 771         receiver->state = STATE_READY;
 772 }
 773
 774 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 775  * gets its message and put to the queue (we have one free place for sure). */
 776 static inline void pipelined_receive(struct mqueue_inode_info *info)
 777 {
 778         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 779
 780         if (!sender) {
 781                 /* for poll */
 782                 wake_up_interruptible(&info->wait_q);
 783                 return;
 784         }
 785         msg_insert(sender->msg, info);
 786         list_del(&sender->list);
 787         sender->state = STATE_PENDING;
 788         wake_up_process(sender->task);
 789         wmb();
 790         sender->state = STATE_READY;
 791 }
 792
 793 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 794         size_t msg_len, unsigned int msg_prio,
 795         const struct timespec __user *u_abs_timeout)
 796 {
 797         struct file *filp;
 798         struct inode *inode;
 799         struct ext_wait_queue wait;
 800         struct ext_wait_queue *receiver;
 801         struct msg_msg *msg_ptr;
 802         struct mqueue_inode_info *info;
 803         long timeout;
 804         int ret;
 805
 806         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 807                 return -EINVAL;
 808
 809         timeout = prepare_timeout(u_abs_timeout);
 810
 811         ret = -EBADF;
 812         filp = fget(mqdes);
 813         if (unlikely(!filp))
 814                 goto out;
 815
 816         inode = filp->f_dentry->d_inode;
 817         if (unlikely(filp->f_op != &mqueue_file_operations))
 818                 goto out_fput;
 819         info = MQUEUE_I(inode);
 820
 821         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 822                 goto out_fput;
 823
 824         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 825                 ret = -EMSGSIZE;
 826                 goto out_fput;
 827         }
 828
 829         /* First try to allocate memory, before doing anything with
 830          * existing queues. */
 831         msg_ptr = load_msg(u_msg_ptr, msg_len);
 832         if (IS_ERR(msg_ptr)) {
 833                 ret = PTR_ERR(msg_ptr);
 834                 goto out_fput;
 835         }
 836         msg_ptr->m_ts = msg_len;
 837         msg_ptr->m_type = msg_prio;
 838
 839         spin_lock(&info->lock);
 840
 841         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 842                 if (filp->f_flags & O_NONBLOCK) {
 843                         spin_unlock(&info->lock);
 844                         ret = -EAGAIN;
 845                 } else if (unlikely(timeout < 0)) {
 846                         spin_unlock(&info->lock);
 847                         ret = timeout;
 848                 } else {
 849                         wait.task = current;
 850                         wait.msg = (void *) msg_ptr;
 851                         wait.state = STATE_NONE;
 852                         ret = wq_sleep(info, SEND, timeout, &wait);
 853                 }
 854                 if (ret < 0)
 855                         free_msg(msg_ptr);
 856         } else {
 857                 receiver = wq_get_first_waiter(info, RECV);
 858                 if (receiver) {
 859                         pipelined_send(info, msg_ptr, receiver);
 860                 } else {
 861                         /* adds message to the queue */
 862                         msg_insert(msg_ptr, info);
 863                         __do_notify(info);
 864                 }
 865                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 866                                 CURRENT_TIME;
 867                 spin_unlock(&info->lock);
 868                 ret = 0;
 869         }
 870 out_fput:
 871         fput(filp);
 872 out:
 873         return ret;
 874 }
 875
 876 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 877         size_t msg_len, unsigned int __user *u_msg_prio,
 878         const struct timespec __user *u_abs_timeout)
 879 {
 880         long timeout;
 881         ssize_t ret;
 882         struct msg_msg *msg_ptr;
 883         struct file *filp;
 884         struct inode *inode;
 885         struct mqueue_inode_info *info;
 886         struct ext_wait_queue wait;
 887
 888         timeout = prepare_timeout(u_abs_timeout);
 889
 890         ret = -EBADF;
 891         filp = fget(mqdes);
 892         if (unlikely(!filp))
 893                 goto out;
 894
 895         inode = filp->f_dentry->d_inode;
 896         if (unlikely(filp->f_op != &mqueue_file_operations))
 897                 goto out_fput;
 898         info = MQUEUE_I(inode);
 899
 900         if (unlikely(!(filp->f_mode & FMODE_READ)))
 901                 goto out_fput;
 902
 903         /* checks if buffer is big enough */
 904         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 905                 ret = -EMSGSIZE;
 906                 goto out_fput;
 907         }
 908
 909         spin_lock(&info->lock);
 910         if (info->attr.mq_curmsgs == 0) {
 911                 if (filp->f_flags & O_NONBLOCK) {
 912                         spin_unlock(&info->lock);
 913                         ret = -EAGAIN;
 914                         msg_ptr = NULL;
 915                 } else if (unlikely(timeout < 0)) {
 916                         spin_unlock(&info->lock);
 917                         ret = timeout;
 918                         msg_ptr = NULL;
 919                 } else {
 920                         wait.task = current;
 921                         wait.state = STATE_NONE;
 922                         ret = wq_sleep(info, RECV, timeout, &wait);
 923                         msg_ptr = wait.msg;
 924                 }
 925         } else {
 926                 msg_ptr = msg_get(info);
 927
 928                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 929                                 CURRENT_TIME;
 930
 931                 /* There is now free space in queue. */
 932                 pipelined_receive(info);
 933                 spin_unlock(&info->lock);
 934                 ret = 0;
 935         }
 936         if (ret == 0) {
 937                 ret = msg_ptr->m_ts;
 938
 939                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 940                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 941                         ret = -EFAULT;
 942                 }
 943                 free_msg(msg_ptr);
 944         }
 945 out_fput:
 946         fput(filp);
 947 out:
 948         return ret;
 949 }
 950
 951 /*
 952  * Notes: the case when user wants us to deregister (with NULL as pointer)
 953  * and he isn't currently owner of notification, will be silently discarded.
 954  * It isn't explicitly defined in the POSIX.
 955  */
 956 asmlinkage long sys_mq_notify(mqd_t mqdes,
 957                                 const struct sigevent __user *u_notification)
 958 {
 959         int ret;
 960         struct file *filp;
 961         struct sock *sock;
 962         struct inode *inode;
 963         struct sigevent notification;
 964         struct mqueue_inode_info *info;
 965         struct sk_buff *nc;
 966
 967         nc = NULL;
 968         sock = NULL;
 969         if (u_notification != NULL) {
 970                 if (copy_from_user(&notification, u_notification,
 971                                         sizeof(struct sigevent)))
 972                         return -EFAULT;
 973
 974                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 975                              notification.sigev_notify != SIGEV_SIGNAL &&
 976                              notification.sigev_notify != SIGEV_THREAD))
 977                         return -EINVAL;
 978                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 979                         (notification.sigev_signo < 0 ||
 980                          notification.sigev_signo > _NSIG)) {
 981                         return -EINVAL;
 982                 }
 983                 if (notification.sigev_notify == SIGEV_THREAD) {
 984                         /* create the notify skb */
 985                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
 986                         ret = -ENOMEM;
 987                         if (!nc)
 988                                 goto out;
 989                         ret = -EFAULT;
 990                         if (copy_from_user(nc->data,
 991                                         notification.sigev_value.sival_ptr,
 992                                         NOTIFY_COOKIE_LEN)) {
 993                                 goto out;
 994                         }
 995
 996                         /* TODO: add a header? */
 997                         skb_put(nc, NOTIFY_COOKIE_LEN);
 998                         /* and attach it to the socket */
 999 retry:
1000                         filp = fget(notification.sigev_signo);
1001                         ret = -EBADF;
1002                         if (!filp)
1003                                 goto out;
1004                         sock = netlink_getsockbyfilp(filp);
1005                         fput(filp);
1006                         if (IS_ERR(sock)) {
1007                                 ret = PTR_ERR(sock);
1008                                 sock = NULL;
1009                                 goto out;
1010                         }
1011
1012                         ret = netlink_attachskb(sock, nc, 0, MAX_SCHEDULE_TIMEOUT);
1013                         if (ret == 1)
1014                                 goto retry;
1015                         if (ret) {
1016                                 sock = NULL;
1017                                 nc = NULL;
1018                                 goto out;
1019                         }
1020                 }
1021         }
1022
1023         ret = -EBADF;
1024         filp = fget(mqdes);
1025         if (!filp)
1026                 goto out;
1027
1028         inode = filp->f_dentry->d_inode;
1029         if (unlikely(filp->f_op != &mqueue_file_operations))
1030                 goto out_fput;
1031         info = MQUEUE_I(inode);
1032
1033         ret = 0;
1034         spin_lock(&info->lock);
1035         if (u_notification == NULL) {
1036                 if (info->notify_owner == current->tgid) {
1037                         remove_notification(info);
1038                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1039                 }
1040         } else if (info->notify_owner != 0) {
1041                 ret = -EBUSY;
1042         } else {
1043                 switch (notification.sigev_notify) {
1044                 case SIGEV_NONE:
1045                         info->notify.sigev_notify = SIGEV_NONE;
1046                         break;
1047                 case SIGEV_THREAD:
1048                         info->notify_sock = sock;
1049                         info->notify_cookie = nc;
1050                         sock = NULL;
1051                         nc = NULL;
1052                         info->notify.sigev_notify = SIGEV_THREAD;
1053                         break;
1054                 case SIGEV_SIGNAL:
1055                         info->notify.sigev_signo = notification.sigev_signo;
1056                         info->notify.sigev_value = notification.sigev_value;
1057                         info->notify.sigev_notify = SIGEV_SIGNAL;
1058                         break;
1059                 }
1060                 info->notify_owner = current->tgid;
1061                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1062         }
1063         spin_unlock(&info->lock);
1064 out_fput:
1065         fput(filp);
1066 out:
1067         if (sock) {
1068                 netlink_detachskb(sock, nc);
1069         } else if (nc) {
1070                 dev_kfree_skb(nc);
1071         }
1072         return ret;
1073 }
1074
1075 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1076                         const struct mq_attr __user *u_mqstat,
1077                         struct mq_attr __user *u_omqstat)
1078 {
1079         int ret;
1080         struct mq_attr mqstat, omqstat;
1081         struct file *filp;
1082         struct inode *inode;
1083         struct mqueue_inode_info *info;
1084
1085         if (u_mqstat != NULL) {
1086                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1087                         return -EFAULT;
1088                 if (mqstat.mq_flags & (~O_NONBLOCK))
1089                         return -EINVAL;
1090         }
1091
1092         ret = -EBADF;
1093         filp = fget(mqdes);
1094         if (!filp)
1095                 goto out;
1096
1097         inode = filp->f_dentry->d_inode;
1098         if (unlikely(filp->f_op != &mqueue_file_operations))
1099                 goto out_fput;
1100         info = MQUEUE_I(inode);
1101
1102         spin_lock(&info->lock);
1103
1104         omqstat = info->attr;
1105         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1106         if (u_mqstat) {
1107                 if (mqstat.mq_flags & O_NONBLOCK)
1108                         filp->f_flags |= O_NONBLOCK;
1109                 else
1110                         filp->f_flags &= ~O_NONBLOCK;
1111
1112                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1113         }
1114
1115         spin_unlock(&info->lock);
1116
1117         ret = 0;
1118         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1119                                                 sizeof(struct mq_attr)))
1120                 ret = -EFAULT;
1121
1122 out_fput:
1123         fput(filp);
1124 out:
1125         return ret;
1126 }
1127
1128 static struct inode_operations mqueue_dir_inode_operations = {
1129         .lookup = simple_lookup,
1130         .create = mqueue_create,
1131         .unlink = mqueue_unlink,
1132 };
1133
1134 static struct file_operations mqueue_file_operations = {
1135         .flush = mqueue_flush_file,
1136         .poll = mqueue_poll_file,
1137         .read = mqueue_read_file,
1138 };
1139
1140 static struct super_operations mqueue_super_ops = {
1141         .alloc_inode = mqueue_alloc_inode,
1142         .destroy_inode = mqueue_destroy_inode,
1143         .statfs = simple_statfs,
1144         .delete_inode = mqueue_delete_inode,
1145         .drop_inode = generic_delete_inode,
1146 };
1147
1148 static struct file_system_type mqueue_fs_type = {
1149         .name = "mqueue",
1150         .get_sb = mqueue_get_sb,
1151         .kill_sb = kill_litter_super,
1152 };
1153
1154 static int msg_max_limit_min = DFLT_MSGMAX;
1155 static int msg_max_limit_max = HARD_MSGMAX;
1156
1157 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1158 static int msg_maxsize_limit_max = INT_MAX;
1159
1160 static ctl_table mq_sysctls[] = {
1161         {
1162                 .ctl_name       = CTL_QUEUESMAX,
1163                 .procname       = "queues_max",
1164                 .data           = &queues_max,
1165                 .maxlen         = sizeof(int),
1166                 .mode           = 0644,
1167                 .proc_handler   = &proc_dointvec,
1168         },
1169         {
1170                 .ctl_name       = CTL_MSGMAX,
1171                 .procname       = "msg_max",
1172                 .data           = &msg_max,
1173                 .maxlen         = sizeof(int),
1174                 .mode           = 0644,
1175                 .proc_handler   = &proc_dointvec_minmax,
1176                 .extra1         = &msg_max_limit_min,
1177                 .extra2         = &msg_max_limit_max,
1178         },
1179         {
1180                 .ctl_name       = CTL_MSGSIZEMAX,
1181                 .procname       = "msgsize_max",
1182                 .data           = &msgsize_max,
1183                 .maxlen         = sizeof(int),
1184                 .mode           = 0644,
1185                 .proc_handler   = &proc_dointvec_minmax,
1186                 .extra1         = &msg_maxsize_limit_min,
1187                 .extra2         = &msg_maxsize_limit_max,
1188         },
1189         { .ctl_name = 0 }
1190 };
1191
1192 static ctl_table mq_sysctl_dir[] = {
1193         {
1194                 .ctl_name       = FS_MQUEUE,
1195                 .procname       = "mqueue",
1196                 .mode           = 0555,
1197                 .child          = mq_sysctls,
1198         },
1199         { .ctl_name = 0 }
1200 };
1201
1202 static ctl_table mq_sysctl_root[] = {
1203         {
1204                 .ctl_name       = CTL_FS,
1205                 .procname       = "fs",
1206                 .mode           = 0555,
1207                 .child          = mq_sysctl_dir,
1208         },
1209         { .ctl_name = 0 }
1210 };
1211
1212 static int __init init_mqueue_fs(void)
1213 {
1214         int error;
1215
1216         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1217                                 sizeof(struct mqueue_inode_info), 0,
1218                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1219         if (mqueue_inode_cachep == NULL)
1220                 return -ENOMEM;
1221
1222         /* ignore failues - they are not fatal */
1223         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1224
1225         error = register_filesystem(&mqueue_fs_type);
1226         if (error)
1227                 goto out_sysctl;
1228
1229         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1230                 error = PTR_ERR(mqueue_mnt);
1231                 goto out_filesystem;
1232         }
1233
1234         /* internal initialization - not common for vfs */
1235         queues_count = 0;
1236         spin_lock_init(&mq_lock);
1237
1238         return 0;
1239
1240 out_filesystem:
1241         unregister_filesystem(&mqueue_fs_type);
1242 out_sysctl:
1243         if (mq_sysctl_table)
1244                 unregister_sysctl_table(mq_sysctl_table);
1245         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1246                 printk(KERN_INFO
1247                         "mqueue_inode_cache: not all structures were freed\n");
1248         }
1249         return error;
1250 }
1251
1252 __initcall(init_mqueue_fs);