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          (Michal.Wronski@motorola.com)
   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/capability.h>
  15 #include <linux/init.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/file.h>
  18 #include <linux/mount.h>
  19 #include <linux/namei.h>
  20 #include <linux/sysctl.h>
  21 #include <linux/poll.h>
  22 #include <linux/mqueue.h>
  23 #include <linux/msg.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/netlink.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/signal.h>
  28 #include <linux/vs_context.h>
  29 #include <linux/vs_limit.h>
  30 #include <net/sock.h>
  31 #include "util.h"
  32
  33 #define MQUEUE_MAGIC    0x19800202
  34 #define DIRENT_SIZE     20
  35 #define FILENT_SIZE     80
  36
  37 #define SEND            0
  38 #define RECV            1
  39
  40 #define STATE_NONE      0
  41 #define STATE_PENDING   1
  42 #define STATE_READY     2
  43
  44 /* used by sysctl */
  45 #define FS_MQUEUE       1
  46 #define CTL_QUEUESMAX   2
  47 #define CTL_MSGMAX      3
  48 #define CTL_MSGSIZEMAX  4
  49
  50 /* default values */
  51 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  52 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  53 #define HARD_MSGMAX     (131072/sizeof(void*))
  54 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  55
  56 #define NOTIFY_COOKIE_LEN       32
  57
  58 struct ext_wait_queue {         /* queue of sleeping tasks */
  59         struct task_struct *task;
  60         struct list_head list;
  61         struct msg_msg *msg;    /* ptr of loaded message */
  62         int state;              /* one of STATE_* values */
  63 };
  64
  65 struct mqueue_inode_info {
  66         spinlock_t lock;
  67         struct inode vfs_inode;
  68         wait_queue_head_t wait_q;
  69
  70         struct msg_msg **messages;
  71         struct mq_attr attr;
  72
  73         struct sigevent notify;
  74         pid_t notify_owner;
  75         struct user_struct *user;       /* user who created, for accounting */
  76         struct sock *notify_sock;
  77         struct sk_buff *notify_cookie;
  78
  79         /* for tasks waiting for free space and messages, respectively */
  80         struct ext_wait_queue e_wait_q[2];
  81
  82         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  83 };
  84
  85 static struct inode_operations mqueue_dir_inode_operations;
  86 static struct file_operations mqueue_file_operations;
  87 static struct super_operations mqueue_super_ops;
  88 static void remove_notification(struct mqueue_inode_info *info);
  89
  90 static spinlock_t mq_lock;
  91 static kmem_cache_t *mqueue_inode_cachep;
  92 static struct vfsmount *mqueue_mnt;
  93
  94 static unsigned int queues_count;
  95 static unsigned int queues_max  = DFLT_QUEUESMAX;
  96 static unsigned int msg_max     = DFLT_MSGMAX;
  97 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  98
  99 static struct ctl_table_header * mq_sysctl_table;
 100
 101 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 102 {
 103         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 104 }
 105
 106 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 107                                                         struct mq_attr *attr)
 108 {
 109         struct inode *inode;
 110
 111         inode = new_inode(sb);
 112         if (inode) {
 113                 inode->i_mode = mode;
 114                 inode->i_uid = current->fsuid;
 115                 inode->i_gid = current->fsgid;
 116                 inode->i_blksize = PAGE_CACHE_SIZE;
 117                 inode->i_blocks = 0;
 118                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 119                                 CURRENT_TIME;
 120
 121                 if (S_ISREG(mode)) {
 122                         struct mqueue_inode_info *info;
 123                         struct task_struct *p = current;
 124                         struct user_struct *u = p->user;
 125                         unsigned long mq_bytes, mq_msg_tblsz;
 126
 127                         inode->i_fop = &mqueue_file_operations;
 128                         inode->i_size = FILENT_SIZE;
 129                         /* mqueue specific info */
 130                         info = MQUEUE_I(inode);
 131                         spin_lock_init(&info->lock);
 132                         init_waitqueue_head(&info->wait_q);
 133                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 134                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 135                         info->messages = NULL;
 136                         info->notify_owner = 0;
 137                         info->qsize = 0;
 138                         info->user = NULL;      /* set when all is ok */
 139                         memset(&info->attr, 0, sizeof(info->attr));
 140                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 141                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 142                         if (attr) {
 143                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 144                                 info->attr.mq_msgsize = attr->mq_msgsize;
 145                         }
 146                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 147                         mq_bytes = (mq_msg_tblsz +
 148                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 149
 150                         spin_lock(&mq_lock);
 151                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 152                             u->mq_bytes + mq_bytes >
 153                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur ||
 154                             !vx_ipcmsg_avail(p->vx_info, mq_bytes)) {
 155                                 spin_unlock(&mq_lock);
 156                                 goto out_inode;
 157                         }
 158                         u->mq_bytes += mq_bytes;
 159                         vx_ipcmsg_add(p->vx_info, u, mq_bytes);
 160                         spin_unlock(&mq_lock);
 161
 162                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 163                         if (!info->messages) {
 164                                 spin_lock(&mq_lock);
 165                                 u->mq_bytes -= mq_bytes;
 166                                 vx_ipcmsg_sub(p->vx_info, u, mq_bytes);
 167                                 spin_unlock(&mq_lock);
 168                                 goto out_inode;
 169                         }
 170                         /* all is ok */
 171                         info->user = get_uid(u);
 172                 } else if (S_ISDIR(mode)) {
 173                         inode->i_nlink++;
 174                         /* Some things misbehave if size == 0 on a directory */
 175                         inode->i_size = 2 * DIRENT_SIZE;
 176                         inode->i_op = &mqueue_dir_inode_operations;
 177                         inode->i_fop = &simple_dir_operations;
 178                 }
 179         }
 180         return inode;
 181 out_inode:
 182         make_bad_inode(inode);
 183         iput(inode);
 184         return NULL;
 185 }
 186
 187 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 188 {
 189         struct inode *inode;
 190
 191         sb->s_blocksize = PAGE_CACHE_SIZE;
 192         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 193         sb->s_magic = MQUEUE_MAGIC;
 194         sb->s_op = &mqueue_super_ops;
 195
 196         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 197         if (!inode)
 198                 return -ENOMEM;
 199
 200         sb->s_root = d_alloc_root(inode);
 201         if (!sb->s_root) {
 202                 iput(inode);
 203                 return -ENOMEM;
 204         }
 205
 206         return 0;
 207 }
 208
 209 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 210                                          int flags, const char *dev_name,
 211                                          void *data)
 212 {
 213         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 214 }
 215
 216 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 217 {
 218         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 219
 220         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 221                 SLAB_CTOR_CONSTRUCTOR)
 222                 inode_init_once(&p->vfs_inode);
 223 }
 224
 225 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 226 {
 227         struct mqueue_inode_info *ei;
 228
 229         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 230         if (!ei)
 231                 return NULL;
 232         return &ei->vfs_inode;
 233 }
 234
 235 static void mqueue_destroy_inode(struct inode *inode)
 236 {
 237         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 238 }
 239
 240 static void mqueue_delete_inode(struct inode *inode)
 241 {
 242         struct mqueue_inode_info *info;
 243         struct user_struct *user;
 244         unsigned long mq_bytes;
 245         int i;
 246
 247         if (S_ISDIR(inode->i_mode)) {
 248                 clear_inode(inode);
 249                 return;
 250         }
 251         info = MQUEUE_I(inode);
 252         spin_lock(&info->lock);
 253         for (i = 0; i < info->attr.mq_curmsgs; i++)
 254                 free_msg(info->messages[i]);
 255         kfree(info->messages);
 256         spin_unlock(&info->lock);
 257
 258         clear_inode(inode);
 259
 260         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 261                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 262         user = info->user;
 263         if (user) {
 264                 struct vx_info *vxi = lookup_vx_info(user->xid);
 265
 266                 spin_lock(&mq_lock);
 267                 user->mq_bytes -= mq_bytes;
 268                 vx_ipcmsg_sub(vxi, user, mq_bytes);
 269                 queues_count--;
 270                 spin_unlock(&mq_lock);
 271                 put_vx_info(vxi);
 272                 free_uid(user);
 273         }
 274 }
 275
 276 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 277                                 int mode, struct nameidata *nd)
 278 {
 279         struct inode *inode;
 280         struct mq_attr *attr = dentry->d_fsdata;
 281         int error;
 282
 283         spin_lock(&mq_lock);
 284         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 285                 error = -ENOSPC;
 286                 goto out_lock;
 287         }
 288         queues_count++;
 289         spin_unlock(&mq_lock);
 290
 291         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 292         if (!inode) {
 293                 error = -ENOMEM;
 294                 spin_lock(&mq_lock);
 295                 queues_count--;
 296                 goto out_lock;
 297         }
 298
 299         dir->i_size += DIRENT_SIZE;
 300         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 301
 302         d_instantiate(dentry, inode);
 303         dget(dentry);
 304         return 0;
 305 out_lock:
 306         spin_unlock(&mq_lock);
 307         return error;
 308 }
 309
 310 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 311 {
 312         struct inode *inode = dentry->d_inode;
 313
 314         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 315         dir->i_size -= DIRENT_SIZE;
 316         inode->i_nlink--;
 317         dput(dentry);
 318         return 0;
 319 }
 320
 321 /*
 322 *       This is routine for system read from queue file.
 323 *       To avoid mess with doing here some sort of mq_receive we allow
 324 *       to read only queue size & notification info (the only values
 325 *       that are interesting from user point of view and aren't accessible
 326 *       through std routines)
 327 */
 328 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 329                                 size_t count, loff_t * off)
 330 {
 331         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 332         char buffer[FILENT_SIZE];
 333         size_t slen;
 334         loff_t o;
 335
 336         if (!count)
 337                 return 0;
 338
 339         spin_lock(&info->lock);
 340         snprintf(buffer, sizeof(buffer),
 341                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 342                         info->qsize,
 343                         info->notify_owner ? info->notify.sigev_notify : 0,
 344                         (info->notify_owner &&
 345                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 346                                 info->notify.sigev_signo : 0,
 347                         info->notify_owner);
 348         spin_unlock(&info->lock);
 349         buffer[sizeof(buffer)-1] = '\0';
 350         slen = strlen(buffer)+1;
 351
 352         o = *off;
 353         if (o > slen)
 354                 return 0;
 355
 356         if (o + count > slen)
 357                 count = slen - o;
 358
 359         if (copy_to_user(u_data, buffer + o, count))
 360                 return -EFAULT;
 361
 362         *off = o + count;
 363         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 364         return count;
 365 }
 366
 367 static int mqueue_flush_file(struct file *filp)
 368 {
 369         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 370
 371         spin_lock(&info->lock);
 372         if (current->tgid == info->notify_owner)
 373                 remove_notification(info);
 374
 375         spin_unlock(&info->lock);
 376         return 0;
 377 }
 378
 379 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 380 {
 381         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 382         int retval = 0;
 383
 384         poll_wait(filp, &info->wait_q, poll_tab);
 385
 386         spin_lock(&info->lock);
 387         if (info->attr.mq_curmsgs)
 388                 retval = POLLIN | POLLRDNORM;
 389
 390         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 391                 retval |= POLLOUT | POLLWRNORM;
 392         spin_unlock(&info->lock);
 393
 394         return retval;
 395 }
 396
 397 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 398 static void wq_add(struct mqueue_inode_info *info, int sr,
 399                         struct ext_wait_queue *ewp)
 400 {
 401         struct ext_wait_queue *walk;
 402
 403         ewp->task = current;
 404
 405         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 406                 if (walk->task->static_prio <= current->static_prio) {
 407                         list_add_tail(&ewp->list, &walk->list);
 408                         return;
 409                 }
 410         }
 411         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 412 }
 413
 414 /*
 415  * Puts current task to sleep. Caller must hold queue lock. After return
 416  * lock isn't held.
 417  * sr: SEND or RECV
 418  */
 419 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 420                         long timeout, struct ext_wait_queue *ewp)
 421 {
 422         int retval;
 423         signed long time;
 424
 425         wq_add(info, sr, ewp);
 426
 427         for (;;) {
 428                 set_current_state(TASK_INTERRUPTIBLE);
 429
 430                 spin_unlock(&info->lock);
 431                 time = schedule_timeout(timeout);
 432
 433                 while (ewp->state == STATE_PENDING)
 434                         cpu_relax();
 435
 436                 if (ewp->state == STATE_READY) {
 437                         retval = 0;
 438                         goto out;
 439                 }
 440                 spin_lock(&info->lock);
 441                 if (ewp->state == STATE_READY) {
 442                         retval = 0;
 443                         goto out_unlock;
 444                 }
 445                 if (signal_pending(current)) {
 446                         retval = -ERESTARTSYS;
 447                         break;
 448                 }
 449                 if (time == 0) {
 450                         retval = -ETIMEDOUT;
 451                         break;
 452                 }
 453         }
 454         list_del(&ewp->list);
 455 out_unlock:
 456         spin_unlock(&info->lock);
 457 out:
 458         return retval;
 459 }
 460
 461 /*
 462  * Returns waiting task that should be serviced first or NULL if none exists
 463  */
 464 static struct ext_wait_queue *wq_get_first_waiter(
 465                 struct mqueue_inode_info *info, int sr)
 466 {
 467         struct list_head *ptr;
 468
 469         ptr = info->e_wait_q[sr].list.prev;
 470         if (ptr == &info->e_wait_q[sr].list)
 471                 return NULL;
 472         return list_entry(ptr, struct ext_wait_queue, list);
 473 }
 474
 475 /* Auxiliary functions to manipulate messages' list */
 476 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 477 {
 478         int k;
 479
 480         k = info->attr.mq_curmsgs - 1;
 481         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 482                 info->messages[k + 1] = info->messages[k];
 483                 k--;
 484         }
 485         info->attr.mq_curmsgs++;
 486         info->qsize += ptr->m_ts;
 487         info->messages[k + 1] = ptr;
 488 }
 489
 490 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 491 {
 492         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 493         return info->messages[info->attr.mq_curmsgs];
 494 }
 495
 496 static inline void set_cookie(struct sk_buff *skb, char code)
 497 {
 498         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 499 }
 500
 501 /*
 502  * The next function is only to split too long sys_mq_timedsend
 503  */
 504 static void __do_notify(struct mqueue_inode_info *info)
 505 {
 506         /* notification
 507          * invoked when there is registered process and there isn't process
 508          * waiting synchronously for message AND state of queue changed from
 509          * empty to not empty. Here we are sure that no one is waiting
 510          * synchronously. */
 511         if (info->notify_owner &&
 512             info->attr.mq_curmsgs == 1) {
 513                 struct siginfo sig_i;
 514                 switch (info->notify.sigev_notify) {
 515                 case SIGEV_NONE:
 516                         break;
 517                 case SIGEV_SIGNAL:
 518                         /* sends signal */
 519
 520                         sig_i.si_signo = info->notify.sigev_signo;
 521                         sig_i.si_errno = 0;
 522                         sig_i.si_code = SI_MESGQ;
 523                         sig_i.si_value = info->notify.sigev_value;
 524                         sig_i.si_pid = current->tgid;
 525                         sig_i.si_uid = current->uid;
 526
 527                         kill_proc_info(info->notify.sigev_signo,
 528                                        &sig_i, info->notify_owner);
 529                         break;
 530                 case SIGEV_THREAD:
 531                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 532                         netlink_sendskb(info->notify_sock,
 533                                         info->notify_cookie, 0);
 534                         break;
 535                 }
 536                 /* after notification unregisters process */
 537                 info->notify_owner = 0;
 538         }
 539         wake_up(&info->wait_q);
 540 }
 541
 542 static long prepare_timeout(const struct timespec __user *u_arg)
 543 {
 544         struct timespec ts, nowts;
 545         long timeout;
 546
 547         if (u_arg) {
 548                 if (unlikely(copy_from_user(&ts, u_arg,
 549                                         sizeof(struct timespec))))
 550                         return -EFAULT;
 551
 552                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 553                         || ts.tv_nsec >= NSEC_PER_SEC))
 554                         return -EINVAL;
 555                 nowts = CURRENT_TIME;
 556                 /* first subtract as jiffies can't be too big */
 557                 ts.tv_sec -= nowts.tv_sec;
 558                 if (ts.tv_nsec < nowts.tv_nsec) {
 559                         ts.tv_nsec += NSEC_PER_SEC;
 560                         ts.tv_sec--;
 561                 }
 562                 ts.tv_nsec -= nowts.tv_nsec;
 563                 if (ts.tv_sec < 0)
 564                         return 0;
 565
 566                 timeout = timespec_to_jiffies(&ts) + 1;
 567         } else
 568                 return MAX_SCHEDULE_TIMEOUT;
 569
 570         return timeout;
 571 }
 572
 573 static void remove_notification(struct mqueue_inode_info *info)
 574 {
 575         if (info->notify_owner != 0 &&
 576             info->notify.sigev_notify == SIGEV_THREAD) {
 577                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 578                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 579         }
 580         info->notify_owner = 0;
 581 }
 582
 583 static int mq_attr_ok(struct mq_attr *attr)
 584 {
 585         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 586                 return 0;
 587         if (capable(CAP_SYS_RESOURCE)) {
 588                 if (attr->mq_maxmsg > HARD_MSGMAX)
 589                         return 0;
 590         } else {
 591                 if (attr->mq_maxmsg > msg_max ||
 592                                 attr->mq_msgsize > msgsize_max)
 593                         return 0;
 594         }
 595         /* check for overflow */
 596         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 597                 return 0;
 598         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 599             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 600             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 601                 return 0;
 602         return 1;
 603 }
 604
 605 /*
 606  * Invoked when creating a new queue via sys_mq_open
 607  */
 608 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 609                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 610 {
 611         struct mq_attr attr;
 612         int ret;
 613
 614         if (u_attr) {
 615                 ret = -EFAULT;
 616                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 617                         goto out;
 618                 ret = -EINVAL;
 619                 if (!mq_attr_ok(&attr))
 620                         goto out;
 621                 /* store for use during create */
 622                 dentry->d_fsdata = &attr;
 623         }
 624
 625         mode &= ~current->fs->umask;
 626         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 627         dentry->d_fsdata = NULL;
 628         if (ret)
 629                 goto out;
 630
 631         return dentry_open(dentry, mqueue_mnt, oflag);
 632
 633 out:
 634         dput(dentry);
 635         mntput(mqueue_mnt);
 636         return ERR_PTR(ret);
 637 }
 638
 639 /* Opens existing queue */
 640 static struct file *do_open(struct dentry *dentry, int oflag)
 641 {
 642 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 643                                         MAY_READ | MAY_WRITE };
 644
 645         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 646                 dput(dentry);
 647                 mntput(mqueue_mnt);
 648                 return ERR_PTR(-EINVAL);
 649         }
 650
 651         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 652                 dput(dentry);
 653                 mntput(mqueue_mnt);
 654                 return ERR_PTR(-EACCES);
 655         }
 656
 657         return dentry_open(dentry, mqueue_mnt, oflag);
 658 }
 659
 660 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 661                                 struct mq_attr __user *u_attr)
 662 {
 663         struct dentry *dentry;
 664         struct file *filp;
 665         char *name;
 666         int fd, error;
 667
 668         if (IS_ERR(name = getname(u_name)))
 669                 return PTR_ERR(name);
 670
 671         fd = get_unused_fd();
 672         if (fd < 0)
 673                 goto out_putname;
 674
 675         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 676         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 677         if (IS_ERR(dentry)) {
 678                 error = PTR_ERR(dentry);
 679                 goto out_err;
 680         }
 681         mntget(mqueue_mnt);
 682
 683         if (oflag & O_CREAT) {
 684                 if (dentry->d_inode) {  /* entry already exists */
 685                         error = -EEXIST;
 686                         if (oflag & O_EXCL)
 687                                 goto out;
 688                         filp = do_open(dentry, oflag);
 689                 } else {
 690                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 691                                                 oflag, mode, u_attr);
 692                 }
 693         } else {
 694                 error = -ENOENT;
 695                 if (!dentry->d_inode)
 696                         goto out;
 697                 filp = do_open(dentry, oflag);
 698         }
 699
 700         if (IS_ERR(filp)) {
 701                 error = PTR_ERR(filp);
 702                 goto out_putfd;
 703         }
 704
 705         set_close_on_exec(fd, 1);
 706         fd_install(fd, filp);
 707         goto out_upsem;
 708
 709 out:
 710         dput(dentry);
 711         mntput(mqueue_mnt);
 712 out_putfd:
 713         put_unused_fd(fd);
 714 out_err:
 715         fd = error;
 716 out_upsem:
 717         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 718 out_putname:
 719         putname(name);
 720         return fd;
 721 }
 722
 723 asmlinkage long sys_mq_unlink(const char __user *u_name)
 724 {
 725         int err;
 726         char *name;
 727         struct dentry *dentry;
 728         struct inode *inode = NULL;
 729
 730         name = getname(u_name);
 731         if (IS_ERR(name))
 732                 return PTR_ERR(name);
 733
 734         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 735         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 736         if (IS_ERR(dentry)) {
 737                 err = PTR_ERR(dentry);
 738                 goto out_unlock;
 739         }
 740
 741         if (!dentry->d_inode) {
 742                 err = -ENOENT;
 743                 goto out_err;
 744         }
 745
 746         inode = dentry->d_inode;
 747         if (inode)
 748                 atomic_inc(&inode->i_count);
 749
 750         err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
 751 out_err:
 752         dput(dentry);
 753
 754 out_unlock:
 755         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 756         putname(name);
 757         if (inode)
 758                 iput(inode);
 759
 760         return err;
 761 }
 762
 763 /* Pipelined send and receive functions.
 764  *
 765  * If a receiver finds no waiting message, then it registers itself in the
 766  * list of waiting receivers. A sender checks that list before adding the new
 767  * message into the message array. If there is a waiting receiver, then it
 768  * bypasses the message array and directly hands the message over to the
 769  * receiver.
 770  * The receiver accepts the message and returns without grabbing the queue
 771  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 772  * are necessary. The same algorithm is used for sysv semaphores, see
 773  * ipc/sem.c fore more details.
 774  *
 775  * The same algorithm is used for senders.
 776  */
 777
 778 /* pipelined_send() - send a message directly to the task waiting in
 779  * sys_mq_timedreceive() (without inserting message into a queue).
 780  */
 781 static inline void pipelined_send(struct mqueue_inode_info *info,
 782                                   struct msg_msg *message,
 783                                   struct ext_wait_queue *receiver)
 784 {
 785         receiver->msg = message;
 786         list_del(&receiver->list);
 787         receiver->state = STATE_PENDING;
 788         wake_up_process(receiver->task);
 789         smp_wmb();
 790         receiver->state = STATE_READY;
 791 }
 792
 793 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 794  * gets its message and put to the queue (we have one free place for sure). */
 795 static inline void pipelined_receive(struct mqueue_inode_info *info)
 796 {
 797         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 798
 799         if (!sender) {
 800                 /* for poll */
 801                 wake_up_interruptible(&info->wait_q);
 802                 return;
 803         }
 804         msg_insert(sender->msg, info);
 805         list_del(&sender->list);
 806         sender->state = STATE_PENDING;
 807         wake_up_process(sender->task);
 808         smp_wmb();
 809         sender->state = STATE_READY;
 810 }
 811
 812 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 813         size_t msg_len, unsigned int msg_prio,
 814         const struct timespec __user *u_abs_timeout)
 815 {
 816         struct file *filp;
 817         struct inode *inode;
 818         struct ext_wait_queue wait;
 819         struct ext_wait_queue *receiver;
 820         struct msg_msg *msg_ptr;
 821         struct mqueue_inode_info *info;
 822         long timeout;
 823         int ret;
 824
 825         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 826                 return -EINVAL;
 827
 828         timeout = prepare_timeout(u_abs_timeout);
 829
 830         ret = -EBADF;
 831         filp = fget(mqdes);
 832         if (unlikely(!filp))
 833                 goto out;
 834
 835         inode = filp->f_dentry->d_inode;
 836         if (unlikely(filp->f_op != &mqueue_file_operations))
 837                 goto out_fput;
 838         info = MQUEUE_I(inode);
 839
 840         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 841                 goto out_fput;
 842
 843         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 844                 ret = -EMSGSIZE;
 845                 goto out_fput;
 846         }
 847
 848         /* First try to allocate memory, before doing anything with
 849          * existing queues. */
 850         msg_ptr = load_msg(u_msg_ptr, msg_len);
 851         if (IS_ERR(msg_ptr)) {
 852                 ret = PTR_ERR(msg_ptr);
 853                 goto out_fput;
 854         }
 855         msg_ptr->m_ts = msg_len;
 856         msg_ptr->m_type = msg_prio;
 857
 858         spin_lock(&info->lock);
 859
 860         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 861                 if (filp->f_flags & O_NONBLOCK) {
 862                         spin_unlock(&info->lock);
 863                         ret = -EAGAIN;
 864                 } else if (unlikely(timeout < 0)) {
 865                         spin_unlock(&info->lock);
 866                         ret = timeout;
 867                 } else {
 868                         wait.task = current;
 869                         wait.msg = (void *) msg_ptr;
 870                         wait.state = STATE_NONE;
 871                         ret = wq_sleep(info, SEND, timeout, &wait);
 872                 }
 873                 if (ret < 0)
 874                         free_msg(msg_ptr);
 875         } else {
 876                 receiver = wq_get_first_waiter(info, RECV);
 877                 if (receiver) {
 878                         pipelined_send(info, msg_ptr, receiver);
 879                 } else {
 880                         /* adds message to the queue */
 881                         msg_insert(msg_ptr, info);
 882                         __do_notify(info);
 883                 }
 884                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 885                                 CURRENT_TIME;
 886                 spin_unlock(&info->lock);
 887                 ret = 0;
 888         }
 889 out_fput:
 890         fput(filp);
 891 out:
 892         return ret;
 893 }
 894
 895 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 896         size_t msg_len, unsigned int __user *u_msg_prio,
 897         const struct timespec __user *u_abs_timeout)
 898 {
 899         long timeout;
 900         ssize_t ret;
 901         struct msg_msg *msg_ptr;
 902         struct file *filp;
 903         struct inode *inode;
 904         struct mqueue_inode_info *info;
 905         struct ext_wait_queue wait;
 906
 907         timeout = prepare_timeout(u_abs_timeout);
 908
 909         ret = -EBADF;
 910         filp = fget(mqdes);
 911         if (unlikely(!filp))
 912                 goto out;
 913
 914         inode = filp->f_dentry->d_inode;
 915         if (unlikely(filp->f_op != &mqueue_file_operations))
 916                 goto out_fput;
 917         info = MQUEUE_I(inode);
 918
 919         if (unlikely(!(filp->f_mode & FMODE_READ)))
 920                 goto out_fput;
 921
 922         /* checks if buffer is big enough */
 923         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 924                 ret = -EMSGSIZE;
 925                 goto out_fput;
 926         }
 927
 928         spin_lock(&info->lock);
 929         if (info->attr.mq_curmsgs == 0) {
 930                 if (filp->f_flags & O_NONBLOCK) {
 931                         spin_unlock(&info->lock);
 932                         ret = -EAGAIN;
 933                         msg_ptr = NULL;
 934                 } else if (unlikely(timeout < 0)) {
 935                         spin_unlock(&info->lock);
 936                         ret = timeout;
 937                         msg_ptr = NULL;
 938                 } else {
 939                         wait.task = current;
 940                         wait.state = STATE_NONE;
 941                         ret = wq_sleep(info, RECV, timeout, &wait);
 942                         msg_ptr = wait.msg;
 943                 }
 944         } else {
 945                 msg_ptr = msg_get(info);
 946
 947                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 948                                 CURRENT_TIME;
 949
 950                 /* There is now free space in queue. */
 951                 pipelined_receive(info);
 952                 spin_unlock(&info->lock);
 953                 ret = 0;
 954         }
 955         if (ret == 0) {
 956                 ret = msg_ptr->m_ts;
 957
 958                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 959                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 960                         ret = -EFAULT;
 961                 }
 962                 free_msg(msg_ptr);
 963         }
 964 out_fput:
 965         fput(filp);
 966 out:
 967         return ret;
 968 }
 969
 970 /*
 971  * Notes: the case when user wants us to deregister (with NULL as pointer)
 972  * and he isn't currently owner of notification, will be silently discarded.
 973  * It isn't explicitly defined in the POSIX.
 974  */
 975 asmlinkage long sys_mq_notify(mqd_t mqdes,
 976                                 const struct sigevent __user *u_notification)
 977 {
 978         int ret;
 979         struct file *filp;
 980         struct sock *sock;
 981         struct inode *inode;
 982         struct sigevent notification;
 983         struct mqueue_inode_info *info;
 984         struct sk_buff *nc;
 985
 986         nc = NULL;
 987         sock = NULL;
 988         if (u_notification != NULL) {
 989                 if (copy_from_user(&notification, u_notification,
 990                                         sizeof(struct sigevent)))
 991                         return -EFAULT;
 992
 993                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 994                              notification.sigev_notify != SIGEV_SIGNAL &&
 995                              notification.sigev_notify != SIGEV_THREAD))
 996                         return -EINVAL;
 997                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 998                         !valid_signal(notification.sigev_signo)) {
 999                         return -EINVAL;
1000                 }
1001                 if (notification.sigev_notify == SIGEV_THREAD) {
1002                         /* create the notify skb */
1003                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1004                         ret = -ENOMEM;
1005                         if (!nc)
1006                                 goto out;
1007                         ret = -EFAULT;
1008                         if (copy_from_user(nc->data,
1009                                         notification.sigev_value.sival_ptr,
1010                                         NOTIFY_COOKIE_LEN)) {
1011                                 goto out;
1012                         }
1013
1014                         /* TODO: add a header? */
1015                         skb_put(nc, NOTIFY_COOKIE_LEN);
1016                         /* and attach it to the socket */
1017 retry:
1018                         filp = fget(notification.sigev_signo);
1019                         ret = -EBADF;
1020                         if (!filp)
1021                                 goto out;
1022                         sock = netlink_getsockbyfilp(filp);
1023                         fput(filp);
1024                         if (IS_ERR(sock)) {
1025                                 ret = PTR_ERR(sock);
1026                                 sock = NULL;
1027                                 goto out;
1028                         }
1029
1030                         ret = netlink_attachskb(sock, nc, 0,
1031                                         MAX_SCHEDULE_TIMEOUT, NULL);
1032                         if (ret == 1)
1033                                 goto retry;
1034                         if (ret) {
1035                                 sock = NULL;
1036                                 nc = NULL;
1037                                 goto out;
1038                         }
1039                 }
1040         }
1041
1042         ret = -EBADF;
1043         filp = fget(mqdes);
1044         if (!filp)
1045                 goto out;
1046
1047         inode = filp->f_dentry->d_inode;
1048         if (unlikely(filp->f_op != &mqueue_file_operations))
1049                 goto out_fput;
1050         info = MQUEUE_I(inode);
1051
1052         ret = 0;
1053         spin_lock(&info->lock);
1054         if (u_notification == NULL) {
1055                 if (info->notify_owner == current->tgid) {
1056                         remove_notification(info);
1057                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1058                 }
1059         } else if (info->notify_owner != 0) {
1060                 ret = -EBUSY;
1061         } else {
1062                 switch (notification.sigev_notify) {
1063                 case SIGEV_NONE:
1064                         info->notify.sigev_notify = SIGEV_NONE;
1065                         break;
1066                 case SIGEV_THREAD:
1067                         info->notify_sock = sock;
1068                         info->notify_cookie = nc;
1069                         sock = NULL;
1070                         nc = NULL;
1071                         info->notify.sigev_notify = SIGEV_THREAD;
1072                         break;
1073                 case SIGEV_SIGNAL:
1074                         info->notify.sigev_signo = notification.sigev_signo;
1075                         info->notify.sigev_value = notification.sigev_value;
1076                         info->notify.sigev_notify = SIGEV_SIGNAL;
1077                         break;
1078                 }
1079                 info->notify_owner = current->tgid;
1080                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1081         }
1082         spin_unlock(&info->lock);
1083 out_fput:
1084         fput(filp);
1085 out:
1086         if (sock) {
1087                 netlink_detachskb(sock, nc);
1088         } else if (nc) {
1089                 dev_kfree_skb(nc);
1090         }
1091         return ret;
1092 }
1093
1094 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1095                         const struct mq_attr __user *u_mqstat,
1096                         struct mq_attr __user *u_omqstat)
1097 {
1098         int ret;
1099         struct mq_attr mqstat, omqstat;
1100         struct file *filp;
1101         struct inode *inode;
1102         struct mqueue_inode_info *info;
1103
1104         if (u_mqstat != NULL) {
1105                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1106                         return -EFAULT;
1107                 if (mqstat.mq_flags & (~O_NONBLOCK))
1108                         return -EINVAL;
1109         }
1110
1111         ret = -EBADF;
1112         filp = fget(mqdes);
1113         if (!filp)
1114                 goto out;
1115
1116         inode = filp->f_dentry->d_inode;
1117         if (unlikely(filp->f_op != &mqueue_file_operations))
1118                 goto out_fput;
1119         info = MQUEUE_I(inode);
1120
1121         spin_lock(&info->lock);
1122
1123         omqstat = info->attr;
1124         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1125         if (u_mqstat) {
1126                 if (mqstat.mq_flags & O_NONBLOCK)
1127                         filp->f_flags |= O_NONBLOCK;
1128                 else
1129                         filp->f_flags &= ~O_NONBLOCK;
1130
1131                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1132         }
1133
1134         spin_unlock(&info->lock);
1135
1136         ret = 0;
1137         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1138                                                 sizeof(struct mq_attr)))
1139                 ret = -EFAULT;
1140
1141 out_fput:
1142         fput(filp);
1143 out:
1144         return ret;
1145 }
1146
1147 static struct inode_operations mqueue_dir_inode_operations = {
1148         .lookup = simple_lookup,
1149         .create = mqueue_create,
1150         .unlink = mqueue_unlink,
1151 };
1152
1153 static struct file_operations mqueue_file_operations = {
1154         .flush = mqueue_flush_file,
1155         .poll = mqueue_poll_file,
1156         .read = mqueue_read_file,
1157 };
1158
1159 static struct super_operations mqueue_super_ops = {
1160         .alloc_inode = mqueue_alloc_inode,
1161         .destroy_inode = mqueue_destroy_inode,
1162         .statfs = simple_statfs,
1163         .delete_inode = mqueue_delete_inode,
1164         .drop_inode = generic_delete_inode,
1165 };
1166
1167 static struct file_system_type mqueue_fs_type = {
1168         .name = "mqueue",
1169         .get_sb = mqueue_get_sb,
1170         .kill_sb = kill_litter_super,
1171 };
1172
1173 static int msg_max_limit_min = DFLT_MSGMAX;
1174 static int msg_max_limit_max = HARD_MSGMAX;
1175
1176 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1177 static int msg_maxsize_limit_max = INT_MAX;
1178
1179 static ctl_table mq_sysctls[] = {
1180         {
1181                 .ctl_name       = CTL_QUEUESMAX,
1182                 .procname       = "queues_max",
1183                 .data           = &queues_max,
1184                 .maxlen         = sizeof(int),
1185                 .mode           = 0644,
1186                 .proc_handler   = &proc_dointvec,
1187         },
1188         {
1189                 .ctl_name       = CTL_MSGMAX,
1190                 .procname       = "msg_max",
1191                 .data           = &msg_max,
1192                 .maxlen         = sizeof(int),
1193                 .mode           = 0644,
1194                 .proc_handler   = &proc_dointvec_minmax,
1195                 .extra1         = &msg_max_limit_min,
1196                 .extra2         = &msg_max_limit_max,
1197         },
1198         {
1199                 .ctl_name       = CTL_MSGSIZEMAX,
1200                 .procname       = "msgsize_max",
1201                 .data           = &msgsize_max,
1202                 .maxlen         = sizeof(int),
1203                 .mode           = 0644,
1204                 .proc_handler   = &proc_dointvec_minmax,
1205                 .extra1         = &msg_maxsize_limit_min,
1206                 .extra2         = &msg_maxsize_limit_max,
1207         },
1208         { .ctl_name = 0 }
1209 };
1210
1211 static ctl_table mq_sysctl_dir[] = {
1212         {
1213                 .ctl_name       = FS_MQUEUE,
1214                 .procname       = "mqueue",
1215                 .mode           = 0555,
1216                 .child          = mq_sysctls,
1217         },
1218         { .ctl_name = 0 }
1219 };
1220
1221 static ctl_table mq_sysctl_root[] = {
1222         {
1223                 .ctl_name       = CTL_FS,
1224                 .procname       = "fs",
1225                 .mode           = 0555,
1226                 .child          = mq_sysctl_dir,
1227         },
1228         { .ctl_name = 0 }
1229 };
1230
1231 static int __init init_mqueue_fs(void)
1232 {
1233         int error;
1234
1235         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1236                                 sizeof(struct mqueue_inode_info), 0,
1237                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1238         if (mqueue_inode_cachep == NULL)
1239                 return -ENOMEM;
1240
1241         /* ignore failues - they are not fatal */
1242         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1243
1244         error = register_filesystem(&mqueue_fs_type);
1245         if (error)
1246                 goto out_sysctl;
1247
1248         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1249                 error = PTR_ERR(mqueue_mnt);
1250                 goto out_filesystem;
1251         }
1252
1253         /* internal initialization - not common for vfs */
1254         queues_count = 0;
1255         spin_lock_init(&mq_lock);
1256
1257         return 0;
1258
1259 out_filesystem:
1260         unregister_filesystem(&mqueue_fs_type);
1261 out_sysctl:
1262         if (mq_sysctl_table)
1263                 unregister_sysctl_table(mq_sysctl_table);
1264         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1265                 printk(KERN_INFO
1266                         "mqueue_inode_cache: not all structures were freed\n");
1267         }
1268         return error;
1269 }
1270
1271 __initcall(init_mqueue_fs);