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 <linux/syscalls.h>
  26 #include <linux/signal.h>
  27 #include <net/sock.h>
  28 #include "util.h"
  29
  30 #define MQUEUE_MAGIC    0x19800202
  31 #define DIRENT_SIZE     20
  32 #define FILENT_SIZE     80
  33
  34 #define SEND            0
  35 #define RECV            1
  36
  37 #define STATE_NONE      0
  38 #define STATE_PENDING   1
  39 #define STATE_READY     2
  40
  41 /* used by sysctl */
  42 #define FS_MQUEUE       1
  43 #define CTL_QUEUESMAX   2
  44 #define CTL_MSGMAX      3
  45 #define CTL_MSGSIZEMAX  4
  46
  47 /* default values */
  48 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  49 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  50 #define HARD_MSGMAX     (131072/sizeof(void*))
  51 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  52
  53 #define NOTIFY_COOKIE_LEN       32
  54
  55 struct ext_wait_queue {         /* queue of sleeping tasks */
  56         struct task_struct *task;
  57         struct list_head list;
  58         struct msg_msg *msg;    /* ptr of loaded message */
  59         int state;              /* one of STATE_* values */
  60 };
  61
  62 struct mqueue_inode_info {
  63         spinlock_t lock;
  64         struct inode vfs_inode;
  65         wait_queue_head_t wait_q;
  66
  67         struct msg_msg **messages;
  68         struct mq_attr attr;
  69
  70         struct sigevent notify;
  71         pid_t notify_owner;
  72         struct user_struct *user;       /* user who created, for accounting */
  73         struct sock *notify_sock;
  74         struct sk_buff *notify_cookie;
  75
  76         /* for tasks waiting for free space and messages, respectively */
  77         struct ext_wait_queue e_wait_q[2];
  78
  79         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  80 };
  81
  82 static struct inode_operations mqueue_dir_inode_operations;
  83 static struct file_operations mqueue_file_operations;
  84 static struct super_operations mqueue_super_ops;
  85 static void remove_notification(struct mqueue_inode_info *info);
  86
  87 static spinlock_t mq_lock;
  88 static kmem_cache_t *mqueue_inode_cachep;
  89 static struct vfsmount *mqueue_mnt;
  90
  91 static unsigned int queues_count;
  92 static unsigned int queues_max  = DFLT_QUEUESMAX;
  93 static unsigned int msg_max     = DFLT_MSGMAX;
  94 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  95
  96 static struct ctl_table_header * mq_sysctl_table;
  97
  98 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  99 {
 100         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 101 }
 102
 103 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 104                                                         struct mq_attr *attr)
 105 {
 106         struct inode *inode;
 107
 108         inode = new_inode(sb);
 109         if (inode) {
 110                 inode->i_mode = mode;
 111                 inode->i_uid = current->fsuid;
 112                 inode->i_gid = current->fsgid;
 113                 inode->i_blksize = PAGE_CACHE_SIZE;
 114                 inode->i_blocks = 0;
 115                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 116                                 CURRENT_TIME;
 117
 118                 if (S_ISREG(mode)) {
 119                         struct mqueue_inode_info *info;
 120                         struct task_struct *p = current;
 121                         struct user_struct *u = p->user;
 122                         unsigned long mq_bytes, mq_msg_tblsz;
 123
 124                         inode->i_fop = &mqueue_file_operations;
 125                         inode->i_size = FILENT_SIZE;
 126                         /* mqueue specific info */
 127                         info = MQUEUE_I(inode);
 128                         spin_lock_init(&info->lock);
 129                         init_waitqueue_head(&info->wait_q);
 130                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 131                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 132                         info->messages = NULL;
 133                         info->notify_owner = 0;
 134                         info->qsize = 0;
 135                         info->user = NULL;      /* set when all is ok */
 136                         memset(&info->attr, 0, sizeof(info->attr));
 137                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 138                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 139                         if (attr) {
 140                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 141                                 info->attr.mq_msgsize = attr->mq_msgsize;
 142                         }
 143                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 144                         mq_bytes = (mq_msg_tblsz +
 145                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 146
 147                         spin_lock(&mq_lock);
 148                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 149                             u->mq_bytes + mq_bytes >
 150                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 151                                 spin_unlock(&mq_lock);
 152                                 goto out_inode;
 153                         }
 154                         u->mq_bytes += mq_bytes;
 155                         spin_unlock(&mq_lock);
 156
 157                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 158                         if (!info->messages) {
 159                                 spin_lock(&mq_lock);
 160                                 u->mq_bytes -= mq_bytes;
 161                                 spin_unlock(&mq_lock);
 162                                 goto out_inode;
 163                         }
 164                         /* all is ok */
 165                         info->user = get_uid(u);
 166                 } else if (S_ISDIR(mode)) {
 167                         inode->i_nlink++;
 168                         /* Some things misbehave if size == 0 on a directory */
 169                         inode->i_size = 2 * DIRENT_SIZE;
 170                         inode->i_op = &mqueue_dir_inode_operations;
 171                         inode->i_fop = &simple_dir_operations;
 172                 }
 173         }
 174         return inode;
 175 out_inode:
 176         make_bad_inode(inode);
 177         iput(inode);
 178         return NULL;
 179 }
 180
 181 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 182 {
 183         struct inode *inode;
 184
 185         sb->s_blocksize = PAGE_CACHE_SIZE;
 186         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 187         sb->s_magic = MQUEUE_MAGIC;
 188         sb->s_op = &mqueue_super_ops;
 189
 190         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 191         if (!inode)
 192                 return -ENOMEM;
 193
 194         sb->s_root = d_alloc_root(inode);
 195         if (!sb->s_root) {
 196                 iput(inode);
 197                 return -ENOMEM;
 198         }
 199
 200         return 0;
 201 }
 202
 203 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 204                                          int flags, const char *dev_name,
 205                                          void *data)
 206 {
 207         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 208 }
 209
 210 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 211 {
 212         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 213
 214         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 215                 SLAB_CTOR_CONSTRUCTOR)
 216                 inode_init_once(&p->vfs_inode);
 217 }
 218
 219 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 220 {
 221         struct mqueue_inode_info *ei;
 222
 223         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 224         if (!ei)
 225                 return NULL;
 226         return &ei->vfs_inode;
 227 }
 228
 229 static void mqueue_destroy_inode(struct inode *inode)
 230 {
 231         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 232 }
 233
 234 static void mqueue_delete_inode(struct inode *inode)
 235 {
 236         struct mqueue_inode_info *info;
 237         struct user_struct *user;
 238         unsigned long mq_bytes;
 239         int i;
 240
 241         if (S_ISDIR(inode->i_mode)) {
 242                 clear_inode(inode);
 243                 return;
 244         }
 245         info = MQUEUE_I(inode);
 246         spin_lock(&info->lock);
 247         for (i = 0; i < info->attr.mq_curmsgs; i++)
 248                 free_msg(info->messages[i]);
 249         kfree(info->messages);
 250         spin_unlock(&info->lock);
 251
 252         clear_inode(inode);
 253
 254         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 255                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 256         user = info->user;
 257         if (user) {
 258                 spin_lock(&mq_lock);
 259                 user->mq_bytes -= mq_bytes;
 260                 queues_count--;
 261                 spin_unlock(&mq_lock);
 262                 free_uid(user);
 263         }
 264 }
 265
 266 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 267                                 int mode, struct nameidata *nd)
 268 {
 269         struct inode *inode;
 270         struct mq_attr *attr = dentry->d_fsdata;
 271         int error;
 272
 273         spin_lock(&mq_lock);
 274         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 275                 error = -ENOSPC;
 276                 goto out_lock;
 277         }
 278         queues_count++;
 279         spin_unlock(&mq_lock);
 280
 281         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 282         if (!inode) {
 283                 error = -ENOMEM;
 284                 spin_lock(&mq_lock);
 285                 queues_count--;
 286                 goto out_lock;
 287         }
 288
 289         dir->i_size += DIRENT_SIZE;
 290         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 291
 292         d_instantiate(dentry, inode);
 293         dget(dentry);
 294         return 0;
 295 out_lock:
 296         spin_unlock(&mq_lock);
 297         return error;
 298 }
 299
 300 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 301 {
 302         struct inode *inode = dentry->d_inode;
 303
 304         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 305         dir->i_size -= DIRENT_SIZE;
 306         inode->i_nlink--;
 307         dput(dentry);
 308         return 0;
 309 }
 310
 311 /*
 312 *       This is routine for system read from queue file.
 313 *       To avoid mess with doing here some sort of mq_receive we allow
 314 *       to read only queue size & notification info (the only values
 315 *       that are interesting from user point of view and aren't accessible
 316 *       through std routines)
 317 */
 318 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 319                                 size_t count, loff_t * off)
 320 {
 321         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 322         char buffer[FILENT_SIZE];
 323         size_t slen;
 324         loff_t o;
 325
 326         if (!count)
 327                 return 0;
 328
 329         spin_lock(&info->lock);
 330         snprintf(buffer, sizeof(buffer),
 331                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 332                         info->qsize,
 333                         info->notify_owner ? info->notify.sigev_notify : 0,
 334                         (info->notify_owner &&
 335                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 336                                 info->notify.sigev_signo : 0,
 337                         info->notify_owner);
 338         spin_unlock(&info->lock);
 339         buffer[sizeof(buffer)-1] = '\0';
 340         slen = strlen(buffer)+1;
 341
 342         o = *off;
 343         if (o > slen)
 344                 return 0;
 345
 346         if (o + count > slen)
 347                 count = slen - o;
 348
 349         if (copy_to_user(u_data, buffer + o, count))
 350                 return -EFAULT;
 351
 352         *off = o + count;
 353         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 354         return count;
 355 }
 356
 357 static int mqueue_flush_file(struct file *filp)
 358 {
 359         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 360
 361         spin_lock(&info->lock);
 362         if (current->tgid == info->notify_owner)
 363                 remove_notification(info);
 364
 365         spin_unlock(&info->lock);
 366         return 0;
 367 }
 368
 369 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 370 {
 371         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 372         int retval = 0;
 373
 374         poll_wait(filp, &info->wait_q, poll_tab);
 375
 376         spin_lock(&info->lock);
 377         if (info->attr.mq_curmsgs)
 378                 retval = POLLIN | POLLRDNORM;
 379
 380         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 381                 retval |= POLLOUT | POLLWRNORM;
 382         spin_unlock(&info->lock);
 383
 384         return retval;
 385 }
 386
 387 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 388 static void wq_add(struct mqueue_inode_info *info, int sr,
 389                         struct ext_wait_queue *ewp)
 390 {
 391         struct ext_wait_queue *walk;
 392
 393         ewp->task = current;
 394
 395         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 396                 if (walk->task->static_prio <= current->static_prio) {
 397                         list_add_tail(&ewp->list, &walk->list);
 398                         return;
 399                 }
 400         }
 401         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 402 }
 403
 404 /*
 405  * Puts current task to sleep. Caller must hold queue lock. After return
 406  * lock isn't held.
 407  * sr: SEND or RECV
 408  */
 409 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 410                         long timeout, struct ext_wait_queue *ewp)
 411 {
 412         int retval;
 413         signed long time;
 414
 415         wq_add(info, sr, ewp);
 416
 417         for (;;) {
 418                 set_current_state(TASK_INTERRUPTIBLE);
 419
 420                 spin_unlock(&info->lock);
 421                 time = schedule_timeout(timeout);
 422
 423                 while (ewp->state == STATE_PENDING)
 424                         cpu_relax();
 425
 426                 if (ewp->state == STATE_READY) {
 427                         retval = 0;
 428                         goto out;
 429                 }
 430                 spin_lock(&info->lock);
 431                 if (ewp->state == STATE_READY) {
 432                         retval = 0;
 433                         goto out_unlock;
 434                 }
 435                 if (signal_pending(current)) {
 436                         retval = -ERESTARTSYS;
 437                         break;
 438                 }
 439                 if (time == 0) {
 440                         retval = -ETIMEDOUT;
 441                         break;
 442                 }
 443         }
 444         list_del(&ewp->list);
 445 out_unlock:
 446         spin_unlock(&info->lock);
 447 out:
 448         return retval;
 449 }
 450
 451 /*
 452  * Returns waiting task that should be serviced first or NULL if none exists
 453  */
 454 static struct ext_wait_queue *wq_get_first_waiter(
 455                 struct mqueue_inode_info *info, int sr)
 456 {
 457         struct list_head *ptr;
 458
 459         ptr = info->e_wait_q[sr].list.prev;
 460         if (ptr == &info->e_wait_q[sr].list)
 461                 return NULL;
 462         return list_entry(ptr, struct ext_wait_queue, list);
 463 }
 464
 465 /* Auxiliary functions to manipulate messages' list */
 466 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 467 {
 468         int k;
 469
 470         k = info->attr.mq_curmsgs - 1;
 471         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 472                 info->messages[k + 1] = info->messages[k];
 473                 k--;
 474         }
 475         info->attr.mq_curmsgs++;
 476         info->qsize += ptr->m_ts;
 477         info->messages[k + 1] = ptr;
 478 }
 479
 480 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 481 {
 482         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 483         return info->messages[info->attr.mq_curmsgs];
 484 }
 485
 486 static inline void set_cookie(struct sk_buff *skb, char code)
 487 {
 488         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 489 }
 490
 491 /*
 492  * The next function is only to split too long sys_mq_timedsend
 493  */
 494 static void __do_notify(struct mqueue_inode_info *info)
 495 {
 496         /* notification
 497          * invoked when there is registered process and there isn't process
 498          * waiting synchronously for message AND state of queue changed from
 499          * empty to not empty. Here we are sure that no one is waiting
 500          * synchronously. */
 501         if (info->notify_owner &&
 502             info->attr.mq_curmsgs == 1) {
 503                 struct siginfo sig_i;
 504                 switch (info->notify.sigev_notify) {
 505                 case SIGEV_NONE:
 506                         break;
 507                 case SIGEV_SIGNAL:
 508                         /* sends signal */
 509
 510                         sig_i.si_signo = info->notify.sigev_signo;
 511                         sig_i.si_errno = 0;
 512                         sig_i.si_code = SI_MESGQ;
 513                         sig_i.si_value = info->notify.sigev_value;
 514                         sig_i.si_pid = current->tgid;
 515                         sig_i.si_uid = current->uid;
 516
 517                         kill_proc_info(info->notify.sigev_signo,
 518                                        &sig_i, info->notify_owner);
 519                         break;
 520                 case SIGEV_THREAD:
 521                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 522                         netlink_sendskb(info->notify_sock,
 523                                         info->notify_cookie, 0);
 524                         break;
 525                 }
 526                 /* after notification unregisters process */
 527                 info->notify_owner = 0;
 528         }
 529         wake_up(&info->wait_q);
 530 }
 531
 532 static long prepare_timeout(const struct timespec __user *u_arg)
 533 {
 534         struct timespec ts, nowts;
 535         long timeout;
 536
 537         if (u_arg) {
 538                 if (unlikely(copy_from_user(&ts, u_arg,
 539                                         sizeof(struct timespec))))
 540                         return -EFAULT;
 541
 542                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 543                         || ts.tv_nsec >= NSEC_PER_SEC))
 544                         return -EINVAL;
 545                 nowts = CURRENT_TIME;
 546                 /* first subtract as jiffies can't be too big */
 547                 ts.tv_sec -= nowts.tv_sec;
 548                 if (ts.tv_nsec < nowts.tv_nsec) {
 549                         ts.tv_nsec += NSEC_PER_SEC;
 550                         ts.tv_sec--;
 551                 }
 552                 ts.tv_nsec -= nowts.tv_nsec;
 553                 if (ts.tv_sec < 0)
 554                         return 0;
 555
 556                 timeout = timespec_to_jiffies(&ts) + 1;
 557         } else
 558                 return MAX_SCHEDULE_TIMEOUT;
 559
 560         return timeout;
 561 }
 562
 563 static void remove_notification(struct mqueue_inode_info *info)
 564 {
 565         if (info->notify_owner != 0 &&
 566             info->notify.sigev_notify == SIGEV_THREAD) {
 567                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 568                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 569         }
 570         info->notify_owner = 0;
 571 }
 572
 573 static int mq_attr_ok(struct mq_attr *attr)
 574 {
 575         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 576                 return 0;
 577         if (capable(CAP_SYS_RESOURCE)) {
 578                 if (attr->mq_maxmsg > HARD_MSGMAX)
 579                         return 0;
 580         } else {
 581                 if (attr->mq_maxmsg > msg_max ||
 582                                 attr->mq_msgsize > msgsize_max)
 583                         return 0;
 584         }
 585         /* check for overflow */
 586         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 587                 return 0;
 588         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 589             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 590             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 591                 return 0;
 592         return 1;
 593 }
 594
 595 /*
 596  * Invoked when creating a new queue via sys_mq_open
 597  */
 598 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 599                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 600 {
 601         struct file *filp;
 602         struct mq_attr attr;
 603         int ret;
 604
 605         if (u_attr != NULL) {
 606                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 607                         return ERR_PTR(-EFAULT);
 608                 if (!mq_attr_ok(&attr))
 609                         return ERR_PTR(-EINVAL);
 610                 /* store for use during create */
 611                 dentry->d_fsdata = &attr;
 612         }
 613
 614         mode &= ~current->fs->umask;
 615         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 616         dentry->d_fsdata = NULL;
 617         if (ret)
 618                 return ERR_PTR(ret);
 619
 620         filp = dentry_open(dentry, mqueue_mnt, oflag);
 621         if (!IS_ERR(filp))
 622                 dget(dentry);
 623
 624         return filp;
 625 }
 626
 627 /* Opens existing queue */
 628 static struct file *do_open(struct dentry *dentry, int oflag)
 629 {
 630 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 631                                         MAY_READ | MAY_WRITE };
 632         struct file *filp;
 633
 634         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 635                 return ERR_PTR(-EINVAL);
 636
 637         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
 638                 return ERR_PTR(-EACCES);
 639
 640         filp = dentry_open(dentry, mqueue_mnt, oflag);
 641
 642         if (!IS_ERR(filp))
 643                 dget(dentry);
 644
 645         return filp;
 646 }
 647
 648 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 649                                 struct mq_attr __user *u_attr)
 650 {
 651         struct dentry *dentry;
 652         struct file *filp;
 653         char *name;
 654         int fd, error;
 655
 656         if (IS_ERR(name = getname(u_name)))
 657                 return PTR_ERR(name);
 658
 659         fd = get_unused_fd();
 660         if (fd < 0)
 661                 goto out_putname;
 662
 663         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 664         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 665         if (IS_ERR(dentry)) {
 666                 error = PTR_ERR(dentry);
 667                 goto out_err;
 668         }
 669         mntget(mqueue_mnt);
 670
 671         if (oflag & O_CREAT) {
 672                 if (dentry->d_inode) {  /* entry already exists */
 673                         filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
 674                                         do_open(dentry, oflag);
 675                 } else {
 676                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 677                                                 oflag, mode, u_attr);
 678                 }
 679         } else
 680                 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
 681                                         ERR_PTR(-ENOENT);
 682
 683         dput(dentry);
 684
 685         if (IS_ERR(filp)) {
 686                 error = PTR_ERR(filp);
 687                 goto out_putfd;
 688         }
 689
 690         set_close_on_exec(fd, 1);
 691         fd_install(fd, filp);
 692         goto out_upsem;
 693
 694 out_putfd:
 695         mntput(mqueue_mnt);
 696         put_unused_fd(fd);
 697 out_err:
 698         fd = error;
 699 out_upsem:
 700         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 701 out_putname:
 702         putname(name);
 703         return fd;
 704 }
 705
 706 asmlinkage long sys_mq_unlink(const char __user *u_name)
 707 {
 708         int err;
 709         char *name;
 710         struct dentry *dentry;
 711         struct inode *inode = NULL;
 712
 713         name = getname(u_name);
 714         if (IS_ERR(name))
 715                 return PTR_ERR(name);
 716
 717         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 718         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 719         if (IS_ERR(dentry)) {
 720                 err = PTR_ERR(dentry);
 721                 goto out_unlock;
 722         }
 723
 724         if (!dentry->d_inode) {
 725                 err = -ENOENT;
 726                 goto out_err;
 727         }
 728
 729         inode = dentry->d_inode;
 730         if (inode)
 731                 atomic_inc(&inode->i_count);
 732
 733         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 734 out_err:
 735         dput(dentry);
 736
 737 out_unlock:
 738         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 739         putname(name);
 740         if (inode)
 741                 iput(inode);
 742
 743         return err;
 744 }
 745
 746 /* Pipelined send and receive functions.
 747  *
 748  * If a receiver finds no waiting message, then it registers itself in the
 749  * list of waiting receivers. A sender checks that list before adding the new
 750  * message into the message array. If there is a waiting receiver, then it
 751  * bypasses the message array and directly hands the message over to the
 752  * receiver.
 753  * The receiver accepts the message and returns without grabbing the queue
 754  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 755  * are necessary. The same algorithm is used for sysv semaphores, see
 756  * ipc/sem.c fore more details.
 757  *
 758  * The same algorithm is used for senders.
 759  */
 760
 761 /* pipelined_send() - send a message directly to the task waiting in
 762  * sys_mq_timedreceive() (without inserting message into a queue).
 763  */
 764 static inline void pipelined_send(struct mqueue_inode_info *info,
 765                                   struct msg_msg *message,
 766                                   struct ext_wait_queue *receiver)
 767 {
 768         receiver->msg = message;
 769         list_del(&receiver->list);
 770         receiver->state = STATE_PENDING;
 771         wake_up_process(receiver->task);
 772         smp_wmb();
 773         receiver->state = STATE_READY;
 774 }
 775
 776 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 777  * gets its message and put to the queue (we have one free place for sure). */
 778 static inline void pipelined_receive(struct mqueue_inode_info *info)
 779 {
 780         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 781
 782         if (!sender) {
 783                 /* for poll */
 784                 wake_up_interruptible(&info->wait_q);
 785                 return;
 786         }
 787         msg_insert(sender->msg, info);
 788         list_del(&sender->list);
 789         sender->state = STATE_PENDING;
 790         wake_up_process(sender->task);
 791         smp_wmb();
 792         sender->state = STATE_READY;
 793 }
 794
 795 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 796         size_t msg_len, unsigned int msg_prio,
 797         const struct timespec __user *u_abs_timeout)
 798 {
 799         struct file *filp;
 800         struct inode *inode;
 801         struct ext_wait_queue wait;
 802         struct ext_wait_queue *receiver;
 803         struct msg_msg *msg_ptr;
 804         struct mqueue_inode_info *info;
 805         long timeout;
 806         int ret;
 807
 808         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 809                 return -EINVAL;
 810
 811         timeout = prepare_timeout(u_abs_timeout);
 812
 813         ret = -EBADF;
 814         filp = fget(mqdes);
 815         if (unlikely(!filp))
 816                 goto out;
 817
 818         inode = filp->f_dentry->d_inode;
 819         if (unlikely(filp->f_op != &mqueue_file_operations))
 820                 goto out_fput;
 821         info = MQUEUE_I(inode);
 822
 823         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 824                 goto out_fput;
 825
 826         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 827                 ret = -EMSGSIZE;
 828                 goto out_fput;
 829         }
 830
 831         /* First try to allocate memory, before doing anything with
 832          * existing queues. */
 833         msg_ptr = load_msg(u_msg_ptr, msg_len);
 834         if (IS_ERR(msg_ptr)) {
 835                 ret = PTR_ERR(msg_ptr);
 836                 goto out_fput;
 837         }
 838         msg_ptr->m_ts = msg_len;
 839         msg_ptr->m_type = msg_prio;
 840
 841         spin_lock(&info->lock);
 842
 843         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 844                 if (filp->f_flags & O_NONBLOCK) {
 845                         spin_unlock(&info->lock);
 846                         ret = -EAGAIN;
 847                 } else if (unlikely(timeout < 0)) {
 848                         spin_unlock(&info->lock);
 849                         ret = timeout;
 850                 } else {
 851                         wait.task = current;
 852                         wait.msg = (void *) msg_ptr;
 853                         wait.state = STATE_NONE;
 854                         ret = wq_sleep(info, SEND, timeout, &wait);
 855                 }
 856                 if (ret < 0)
 857                         free_msg(msg_ptr);
 858         } else {
 859                 receiver = wq_get_first_waiter(info, RECV);
 860                 if (receiver) {
 861                         pipelined_send(info, msg_ptr, receiver);
 862                 } else {
 863                         /* adds message to the queue */
 864                         msg_insert(msg_ptr, info);
 865                         __do_notify(info);
 866                 }
 867                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 868                                 CURRENT_TIME;
 869                 spin_unlock(&info->lock);
 870                 ret = 0;
 871         }
 872 out_fput:
 873         fput(filp);
 874 out:
 875         return ret;
 876 }
 877
 878 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 879         size_t msg_len, unsigned int __user *u_msg_prio,
 880         const struct timespec __user *u_abs_timeout)
 881 {
 882         long timeout;
 883         ssize_t ret;
 884         struct msg_msg *msg_ptr;
 885         struct file *filp;
 886         struct inode *inode;
 887         struct mqueue_inode_info *info;
 888         struct ext_wait_queue wait;
 889
 890         timeout = prepare_timeout(u_abs_timeout);
 891
 892         ret = -EBADF;
 893         filp = fget(mqdes);
 894         if (unlikely(!filp))
 895                 goto out;
 896
 897         inode = filp->f_dentry->d_inode;
 898         if (unlikely(filp->f_op != &mqueue_file_operations))
 899                 goto out_fput;
 900         info = MQUEUE_I(inode);
 901
 902         if (unlikely(!(filp->f_mode & FMODE_READ)))
 903                 goto out_fput;
 904
 905         /* checks if buffer is big enough */
 906         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 907                 ret = -EMSGSIZE;
 908                 goto out_fput;
 909         }
 910
 911         spin_lock(&info->lock);
 912         if (info->attr.mq_curmsgs == 0) {
 913                 if (filp->f_flags & O_NONBLOCK) {
 914                         spin_unlock(&info->lock);
 915                         ret = -EAGAIN;
 916                         msg_ptr = NULL;
 917                 } else if (unlikely(timeout < 0)) {
 918                         spin_unlock(&info->lock);
 919                         ret = timeout;
 920                         msg_ptr = NULL;
 921                 } else {
 922                         wait.task = current;
 923                         wait.state = STATE_NONE;
 924                         ret = wq_sleep(info, RECV, timeout, &wait);
 925                         msg_ptr = wait.msg;
 926                 }
 927         } else {
 928                 msg_ptr = msg_get(info);
 929
 930                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 931                                 CURRENT_TIME;
 932
 933                 /* There is now free space in queue. */
 934                 pipelined_receive(info);
 935                 spin_unlock(&info->lock);
 936                 ret = 0;
 937         }
 938         if (ret == 0) {
 939                 ret = msg_ptr->m_ts;
 940
 941                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 942                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 943                         ret = -EFAULT;
 944                 }
 945                 free_msg(msg_ptr);
 946         }
 947 out_fput:
 948         fput(filp);
 949 out:
 950         return ret;
 951 }
 952
 953 /*
 954  * Notes: the case when user wants us to deregister (with NULL as pointer)
 955  * and he isn't currently owner of notification, will be silently discarded.
 956  * It isn't explicitly defined in the POSIX.
 957  */
 958 asmlinkage long sys_mq_notify(mqd_t mqdes,
 959                                 const struct sigevent __user *u_notification)
 960 {
 961         int ret;
 962         struct file *filp;
 963         struct sock *sock;
 964         struct inode *inode;
 965         struct sigevent notification;
 966         struct mqueue_inode_info *info;
 967         struct sk_buff *nc;
 968
 969         nc = NULL;
 970         sock = NULL;
 971         if (u_notification != NULL) {
 972                 if (copy_from_user(&notification, u_notification,
 973                                         sizeof(struct sigevent)))
 974                         return -EFAULT;
 975
 976                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 977                              notification.sigev_notify != SIGEV_SIGNAL &&
 978                              notification.sigev_notify != SIGEV_THREAD))
 979                         return -EINVAL;
 980                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 981                         !valid_signal(notification.sigev_signo)) {
 982                         return -EINVAL;
 983                 }
 984                 if (notification.sigev_notify == SIGEV_THREAD) {
 985                         /* create the notify skb */
 986                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
 987                         ret = -ENOMEM;
 988                         if (!nc)
 989                                 goto out;
 990                         ret = -EFAULT;
 991                         if (copy_from_user(nc->data,
 992                                         notification.sigev_value.sival_ptr,
 993                                         NOTIFY_COOKIE_LEN)) {
 994                                 goto out;
 995                         }
 996
 997                         /* TODO: add a header? */
 998                         skb_put(nc, NOTIFY_COOKIE_LEN);
 999                         /* and attach it to the socket */
1000 retry:
1001                         filp = fget(notification.sigev_signo);
1002                         ret = -EBADF;
1003                         if (!filp)
1004                                 goto out;
1005                         sock = netlink_getsockbyfilp(filp);
1006                         fput(filp);
1007                         if (IS_ERR(sock)) {
1008                                 ret = PTR_ERR(sock);
1009                                 sock = NULL;
1010                                 goto out;
1011                         }
1012
1013                         ret = netlink_attachskb(sock, nc, 0, MAX_SCHEDULE_TIMEOUT);
1014                         if (ret == 1)
1015                                 goto retry;
1016                         if (ret) {
1017                                 sock = NULL;
1018                                 nc = NULL;
1019                                 goto out;
1020                         }
1021                 }
1022         }
1023
1024         ret = -EBADF;
1025         filp = fget(mqdes);
1026         if (!filp)
1027                 goto out;
1028
1029         inode = filp->f_dentry->d_inode;
1030         if (unlikely(filp->f_op != &mqueue_file_operations))
1031                 goto out_fput;
1032         info = MQUEUE_I(inode);
1033
1034         ret = 0;
1035         spin_lock(&info->lock);
1036         if (u_notification == NULL) {
1037                 if (info->notify_owner == current->tgid) {
1038                         remove_notification(info);
1039                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1040                 }
1041         } else if (info->notify_owner != 0) {
1042                 ret = -EBUSY;
1043         } else {
1044                 switch (notification.sigev_notify) {
1045                 case SIGEV_NONE:
1046                         info->notify.sigev_notify = SIGEV_NONE;
1047                         break;
1048                 case SIGEV_THREAD:
1049                         info->notify_sock = sock;
1050                         info->notify_cookie = nc;
1051                         sock = NULL;
1052                         nc = NULL;
1053                         info->notify.sigev_notify = SIGEV_THREAD;
1054                         break;
1055                 case SIGEV_SIGNAL:
1056                         info->notify.sigev_signo = notification.sigev_signo;
1057                         info->notify.sigev_value = notification.sigev_value;
1058                         info->notify.sigev_notify = SIGEV_SIGNAL;
1059                         break;
1060                 }
1061                 info->notify_owner = current->tgid;
1062                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1063         }
1064         spin_unlock(&info->lock);
1065 out_fput:
1066         fput(filp);
1067 out:
1068         if (sock) {
1069                 netlink_detachskb(sock, nc);
1070         } else if (nc) {
1071                 dev_kfree_skb(nc);
1072         }
1073         return ret;
1074 }
1075
1076 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1077                         const struct mq_attr __user *u_mqstat,
1078                         struct mq_attr __user *u_omqstat)
1079 {
1080         int ret;
1081         struct mq_attr mqstat, omqstat;
1082         struct file *filp;
1083         struct inode *inode;
1084         struct mqueue_inode_info *info;
1085
1086         if (u_mqstat != NULL) {
1087                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1088                         return -EFAULT;
1089                 if (mqstat.mq_flags & (~O_NONBLOCK))
1090                         return -EINVAL;
1091         }
1092
1093         ret = -EBADF;
1094         filp = fget(mqdes);
1095         if (!filp)
1096                 goto out;
1097
1098         inode = filp->f_dentry->d_inode;
1099         if (unlikely(filp->f_op != &mqueue_file_operations))
1100                 goto out_fput;
1101         info = MQUEUE_I(inode);
1102
1103         spin_lock(&info->lock);
1104
1105         omqstat = info->attr;
1106         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1107         if (u_mqstat) {
1108                 if (mqstat.mq_flags & O_NONBLOCK)
1109                         filp->f_flags |= O_NONBLOCK;
1110                 else
1111                         filp->f_flags &= ~O_NONBLOCK;
1112
1113                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1114         }
1115
1116         spin_unlock(&info->lock);
1117
1118         ret = 0;
1119         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1120                                                 sizeof(struct mq_attr)))
1121                 ret = -EFAULT;
1122
1123 out_fput:
1124         fput(filp);
1125 out:
1126         return ret;
1127 }
1128
1129 static struct inode_operations mqueue_dir_inode_operations = {
1130         .lookup = simple_lookup,
1131         .create = mqueue_create,
1132         .unlink = mqueue_unlink,
1133 };
1134
1135 static struct file_operations mqueue_file_operations = {
1136         .flush = mqueue_flush_file,
1137         .poll = mqueue_poll_file,
1138         .read = mqueue_read_file,
1139 };
1140
1141 static struct super_operations mqueue_super_ops = {
1142         .alloc_inode = mqueue_alloc_inode,
1143         .destroy_inode = mqueue_destroy_inode,
1144         .statfs = simple_statfs,
1145         .delete_inode = mqueue_delete_inode,
1146         .drop_inode = generic_delete_inode,
1147 };
1148
1149 static struct file_system_type mqueue_fs_type = {
1150         .name = "mqueue",
1151         .get_sb = mqueue_get_sb,
1152         .kill_sb = kill_litter_super,
1153 };
1154
1155 static int msg_max_limit_min = DFLT_MSGMAX;
1156 static int msg_max_limit_max = HARD_MSGMAX;
1157
1158 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1159 static int msg_maxsize_limit_max = INT_MAX;
1160
1161 static ctl_table mq_sysctls[] = {
1162         {
1163                 .ctl_name       = CTL_QUEUESMAX,
1164                 .procname       = "queues_max",
1165                 .data           = &queues_max,
1166                 .maxlen         = sizeof(int),
1167                 .mode           = 0644,
1168                 .proc_handler   = &proc_dointvec,
1169         },
1170         {
1171                 .ctl_name       = CTL_MSGMAX,
1172                 .procname       = "msg_max",
1173                 .data           = &msg_max,
1174                 .maxlen         = sizeof(int),
1175                 .mode           = 0644,
1176                 .proc_handler   = &proc_dointvec_minmax,
1177                 .extra1         = &msg_max_limit_min,
1178                 .extra2         = &msg_max_limit_max,
1179         },
1180         {
1181                 .ctl_name       = CTL_MSGSIZEMAX,
1182                 .procname       = "msgsize_max",
1183                 .data           = &msgsize_max,
1184                 .maxlen         = sizeof(int),
1185                 .mode           = 0644,
1186                 .proc_handler   = &proc_dointvec_minmax,
1187                 .extra1         = &msg_maxsize_limit_min,
1188                 .extra2         = &msg_maxsize_limit_max,
1189         },
1190         { .ctl_name = 0 }
1191 };
1192
1193 static ctl_table mq_sysctl_dir[] = {
1194         {
1195                 .ctl_name       = FS_MQUEUE,
1196                 .procname       = "mqueue",
1197                 .mode           = 0555,
1198                 .child          = mq_sysctls,
1199         },
1200         { .ctl_name = 0 }
1201 };
1202
1203 static ctl_table mq_sysctl_root[] = {
1204         {
1205                 .ctl_name       = CTL_FS,
1206                 .procname       = "fs",
1207                 .mode           = 0555,
1208                 .child          = mq_sysctl_dir,
1209         },
1210         { .ctl_name = 0 }
1211 };
1212
1213 static int __init init_mqueue_fs(void)
1214 {
1215         int error;
1216
1217         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1218                                 sizeof(struct mqueue_inode_info), 0,
1219                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1220         if (mqueue_inode_cachep == NULL)
1221                 return -ENOMEM;
1222
1223         /* ignore failues - they are not fatal */
1224         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1225
1226         error = register_filesystem(&mqueue_fs_type);
1227         if (error)
1228                 goto out_sysctl;
1229
1230         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1231                 error = PTR_ERR(mqueue_mnt);
1232                 goto out_filesystem;
1233         }
1234
1235         /* internal initialization - not common for vfs */
1236         queues_count = 0;
1237         spin_lock_init(&mq_lock);
1238
1239         return 0;
1240
1241 out_filesystem:
1242         unregister_filesystem(&mqueue_fs_type);
1243 out_sysctl:
1244         if (mq_sysctl_table)
1245                 unregister_sysctl_table(mq_sysctl_table);
1246         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1247                 printk(KERN_INFO
1248                         "mqueue_inode_cache: not all structures were freed\n");
1249         }
1250         return error;
1251 }
1252
1253 __initcall(init_mqueue_fs);