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@gmail.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  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/pid.h>
  34
  35 #include <net/sock.h>
  36 #include "util.h"
  37
  38 #define MQUEUE_MAGIC    0x19800202
  39 #define DIRENT_SIZE     20
  40 #define FILENT_SIZE     80
  41
  42 #define SEND            0
  43 #define RECV            1
  44
  45 #define STATE_NONE      0
  46 #define STATE_PENDING   1
  47 #define STATE_READY     2
  48
  49 /* default values */
  50 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  51 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  52 #define HARD_MSGMAX     (131072/sizeof(void*))
  53 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  54
  55
  56 struct ext_wait_queue {         /* queue of sleeping tasks */
  57         struct task_struct *task;
  58         struct list_head list;
  59         struct msg_msg *msg;    /* ptr of loaded message */
  60         int state;              /* one of STATE_* values */
  61 };
  62
  63 struct mqueue_inode_info {
  64         spinlock_t lock;
  65         struct inode vfs_inode;
  66         wait_queue_head_t wait_q;
  67
  68         struct msg_msg **messages;
  69         struct mq_attr attr;
  70
  71         struct sigevent notify;
  72         struct pid* notify_owner;
  73         struct user_struct *user;       /* user who created, for accounting */
  74         struct sock *notify_sock;
  75         struct sk_buff *notify_cookie;
  76
  77         /* for tasks waiting for free space and messages, respectively */
  78         struct ext_wait_queue e_wait_q[2];
  79
  80         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  81 };
  82
  83 static const struct inode_operations mqueue_dir_inode_operations;
  84 static const struct file_operations mqueue_file_operations;
  85 static struct super_operations mqueue_super_ops;
  86 static void remove_notification(struct mqueue_inode_info *info);
  87
  88 static spinlock_t mq_lock;
  89 static struct kmem_cache *mqueue_inode_cachep;
  90 static struct vfsmount *mqueue_mnt;
  91
  92 static unsigned int queues_count;
  93 static unsigned int queues_max  = DFLT_QUEUESMAX;
  94 static unsigned int msg_max     = DFLT_MSGMAX;
  95 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  96
  97 static struct ctl_table_header * mq_sysctl_table;
  98
  99 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 100 {
 101         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 102 }
 103
 104 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 105                                                         struct mq_attr *attr)
 106 {
 107         struct inode *inode;
 108
 109         inode = new_inode(sb);
 110         if (inode) {
 111                 inode->i_mode = mode;
 112                 inode->i_uid = current->fsuid;
 113                 inode->i_gid = current->fsgid;
 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 = NULL;
 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                         inc_nlink(inode);
 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 int mqueue_get_sb(struct file_system_type *fs_type,
 204                          int flags, const char *dev_name,
 205                          void *data, struct vfsmount *mnt)
 206 {
 207         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
 208 }
 209
 210 static void init_once(void *foo)
 211 {
 212         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 213
 214         inode_init_once(&p->vfs_inode);
 215 }
 216
 217 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 218 {
 219         struct mqueue_inode_info *ei;
 220
 221         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 222         if (!ei)
 223                 return NULL;
 224         return &ei->vfs_inode;
 225 }
 226
 227 static void mqueue_destroy_inode(struct inode *inode)
 228 {
 229         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 230 }
 231
 232 static void mqueue_delete_inode(struct inode *inode)
 233 {
 234         struct mqueue_inode_info *info;
 235         struct user_struct *user;
 236         unsigned long mq_bytes;
 237         int i;
 238
 239         if (S_ISDIR(inode->i_mode)) {
 240                 clear_inode(inode);
 241                 return;
 242         }
 243         info = MQUEUE_I(inode);
 244         spin_lock(&info->lock);
 245         for (i = 0; i < info->attr.mq_curmsgs; i++)
 246                 free_msg(info->messages[i]);
 247         kfree(info->messages);
 248         spin_unlock(&info->lock);
 249
 250         clear_inode(inode);
 251
 252         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 253                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 254         user = info->user;
 255         if (user) {
 256                 spin_lock(&mq_lock);
 257                 user->mq_bytes -= mq_bytes;
 258                 queues_count--;
 259                 spin_unlock(&mq_lock);
 260                 free_uid(user);
 261         }
 262 }
 263
 264 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 265                                 int mode, struct nameidata *nd)
 266 {
 267         struct inode *inode;
 268         struct mq_attr *attr = dentry->d_fsdata;
 269         int error;
 270
 271         spin_lock(&mq_lock);
 272         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 273                 error = -ENOSPC;
 274                 goto out_lock;
 275         }
 276         queues_count++;
 277         spin_unlock(&mq_lock);
 278
 279         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 280         if (!inode) {
 281                 error = -ENOMEM;
 282                 spin_lock(&mq_lock);
 283                 queues_count--;
 284                 goto out_lock;
 285         }
 286
 287         dir->i_size += DIRENT_SIZE;
 288         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 289
 290         d_instantiate(dentry, inode);
 291         dget(dentry);
 292         return 0;
 293 out_lock:
 294         spin_unlock(&mq_lock);
 295         return error;
 296 }
 297
 298 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 299 {
 300         struct inode *inode = dentry->d_inode;
 301
 302         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 303         dir->i_size -= DIRENT_SIZE;
 304         drop_nlink(inode);
 305         dput(dentry);
 306         return 0;
 307 }
 308
 309 /*
 310 *       This is routine for system read from queue file.
 311 *       To avoid mess with doing here some sort of mq_receive we allow
 312 *       to read only queue size & notification info (the only values
 313 *       that are interesting from user point of view and aren't accessible
 314 *       through std routines)
 315 */
 316 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 317                                 size_t count, loff_t *off)
 318 {
 319         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 320         char buffer[FILENT_SIZE];
 321         ssize_t ret;
 322
 323         spin_lock(&info->lock);
 324         snprintf(buffer, sizeof(buffer),
 325                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 326                         info->qsize,
 327                         info->notify_owner ? info->notify.sigev_notify : 0,
 328                         (info->notify_owner &&
 329                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 330                                 info->notify.sigev_signo : 0,
 331                         pid_vnr(info->notify_owner));
 332         spin_unlock(&info->lock);
 333         buffer[sizeof(buffer)-1] = '\0';
 334
 335         ret = simple_read_from_buffer(u_data, count, off, buffer,
 336                                 strlen(buffer));
 337         if (ret <= 0)
 338                 return ret;
 339
 340         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 341         return ret;
 342 }
 343
 344 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 345 {
 346         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 347
 348         spin_lock(&info->lock);
 349         if (task_tgid(current) == info->notify_owner)
 350                 remove_notification(info);
 351
 352         spin_unlock(&info->lock);
 353         return 0;
 354 }
 355
 356 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 357 {
 358         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 359         int retval = 0;
 360
 361         poll_wait(filp, &info->wait_q, poll_tab);
 362
 363         spin_lock(&info->lock);
 364         if (info->attr.mq_curmsgs)
 365                 retval = POLLIN | POLLRDNORM;
 366
 367         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 368                 retval |= POLLOUT | POLLWRNORM;
 369         spin_unlock(&info->lock);
 370
 371         return retval;
 372 }
 373
 374 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 375 static void wq_add(struct mqueue_inode_info *info, int sr,
 376                         struct ext_wait_queue *ewp)
 377 {
 378         struct ext_wait_queue *walk;
 379
 380         ewp->task = current;
 381
 382         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 383                 if (walk->task->static_prio <= current->static_prio) {
 384                         list_add_tail(&ewp->list, &walk->list);
 385                         return;
 386                 }
 387         }
 388         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 389 }
 390
 391 /*
 392  * Puts current task to sleep. Caller must hold queue lock. After return
 393  * lock isn't held.
 394  * sr: SEND or RECV
 395  */
 396 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 397                         long timeout, struct ext_wait_queue *ewp)
 398 {
 399         int retval;
 400         signed long time;
 401
 402         wq_add(info, sr, ewp);
 403
 404         for (;;) {
 405                 set_current_state(TASK_INTERRUPTIBLE);
 406
 407                 spin_unlock(&info->lock);
 408                 time = schedule_timeout(timeout);
 409
 410                 while (ewp->state == STATE_PENDING)
 411                         cpu_relax();
 412
 413                 if (ewp->state == STATE_READY) {
 414                         retval = 0;
 415                         goto out;
 416                 }
 417                 spin_lock(&info->lock);
 418                 if (ewp->state == STATE_READY) {
 419                         retval = 0;
 420                         goto out_unlock;
 421                 }
 422                 if (signal_pending(current)) {
 423                         retval = -ERESTARTSYS;
 424                         break;
 425                 }
 426                 if (time == 0) {
 427                         retval = -ETIMEDOUT;
 428                         break;
 429                 }
 430         }
 431         list_del(&ewp->list);
 432 out_unlock:
 433         spin_unlock(&info->lock);
 434 out:
 435         return retval;
 436 }
 437
 438 /*
 439  * Returns waiting task that should be serviced first or NULL if none exists
 440  */
 441 static struct ext_wait_queue *wq_get_first_waiter(
 442                 struct mqueue_inode_info *info, int sr)
 443 {
 444         struct list_head *ptr;
 445
 446         ptr = info->e_wait_q[sr].list.prev;
 447         if (ptr == &info->e_wait_q[sr].list)
 448                 return NULL;
 449         return list_entry(ptr, struct ext_wait_queue, list);
 450 }
 451
 452 /* Auxiliary functions to manipulate messages' list */
 453 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 454 {
 455         int k;
 456
 457         k = info->attr.mq_curmsgs - 1;
 458         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 459                 info->messages[k + 1] = info->messages[k];
 460                 k--;
 461         }
 462         info->attr.mq_curmsgs++;
 463         info->qsize += ptr->m_ts;
 464         info->messages[k + 1] = ptr;
 465 }
 466
 467 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 468 {
 469         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 470         return info->messages[info->attr.mq_curmsgs];
 471 }
 472
 473 static inline void set_cookie(struct sk_buff *skb, char code)
 474 {
 475         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 476 }
 477
 478 /*
 479  * The next function is only to split too long sys_mq_timedsend
 480  */
 481 static void __do_notify(struct mqueue_inode_info *info)
 482 {
 483         /* notification
 484          * invoked when there is registered process and there isn't process
 485          * waiting synchronously for message AND state of queue changed from
 486          * empty to not empty. Here we are sure that no one is waiting
 487          * synchronously. */
 488         if (info->notify_owner &&
 489             info->attr.mq_curmsgs == 1) {
 490                 struct siginfo sig_i;
 491                 switch (info->notify.sigev_notify) {
 492                 case SIGEV_NONE:
 493                         break;
 494                 case SIGEV_SIGNAL:
 495                         /* sends signal */
 496
 497                         sig_i.si_signo = info->notify.sigev_signo;
 498                         sig_i.si_errno = 0;
 499                         sig_i.si_code = SI_MESGQ;
 500                         sig_i.si_value = info->notify.sigev_value;
 501                         sig_i.si_pid = task_tgid_nr_ns(current,
 502                                                 ns_of_pid(info->notify_owner));
 503                         sig_i.si_uid = current->uid;
 504
 505                         kill_pid_info(info->notify.sigev_signo,
 506                                       &sig_i, info->notify_owner);
 507                         break;
 508                 case SIGEV_THREAD:
 509                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 510                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 511                         break;
 512                 }
 513                 /* after notification unregisters process */
 514                 put_pid(info->notify_owner);
 515                 info->notify_owner = NULL;
 516         }
 517         wake_up(&info->wait_q);
 518 }
 519
 520 static long prepare_timeout(const struct timespec __user *u_arg)
 521 {
 522         struct timespec ts, nowts;
 523         long timeout;
 524
 525         if (u_arg) {
 526                 if (unlikely(copy_from_user(&ts, u_arg,
 527                                         sizeof(struct timespec))))
 528                         return -EFAULT;
 529
 530                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 531                         || ts.tv_nsec >= NSEC_PER_SEC))
 532                         return -EINVAL;
 533                 nowts = CURRENT_TIME;
 534                 /* first subtract as jiffies can't be too big */
 535                 ts.tv_sec -= nowts.tv_sec;
 536                 if (ts.tv_nsec < nowts.tv_nsec) {
 537                         ts.tv_nsec += NSEC_PER_SEC;
 538                         ts.tv_sec--;
 539                 }
 540                 ts.tv_nsec -= nowts.tv_nsec;
 541                 if (ts.tv_sec < 0)
 542                         return 0;
 543
 544                 timeout = timespec_to_jiffies(&ts) + 1;
 545         } else
 546                 return MAX_SCHEDULE_TIMEOUT;
 547
 548         return timeout;
 549 }
 550
 551 static void remove_notification(struct mqueue_inode_info *info)
 552 {
 553         if (info->notify_owner != NULL &&
 554             info->notify.sigev_notify == SIGEV_THREAD) {
 555                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 556                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 557         }
 558         put_pid(info->notify_owner);
 559         info->notify_owner = NULL;
 560 }
 561
 562 static int mq_attr_ok(struct mq_attr *attr)
 563 {
 564         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 565                 return 0;
 566         if (capable(CAP_SYS_RESOURCE)) {
 567                 if (attr->mq_maxmsg > HARD_MSGMAX)
 568                         return 0;
 569         } else {
 570                 if (attr->mq_maxmsg > msg_max ||
 571                                 attr->mq_msgsize > msgsize_max)
 572                         return 0;
 573         }
 574         /* check for overflow */
 575         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 576                 return 0;
 577         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 578             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 579             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 580                 return 0;
 581         return 1;
 582 }
 583
 584 /*
 585  * Invoked when creating a new queue via sys_mq_open
 586  */
 587 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 588                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 589 {
 590         struct mq_attr attr;
 591         struct file *result;
 592         int ret;
 593
 594         if (u_attr) {
 595                 ret = -EFAULT;
 596                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 597                         goto out;
 598                 ret = -EINVAL;
 599                 if (!mq_attr_ok(&attr))
 600                         goto out;
 601                 /* store for use during create */
 602                 dentry->d_fsdata = &attr;
 603         }
 604
 605         mode &= ~current->fs->umask;
 606         ret = mnt_want_write(mqueue_mnt);
 607         if (ret)
 608                 goto out;
 609         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 610         dentry->d_fsdata = NULL;
 611         if (ret)
 612                 goto out_drop_write;
 613
 614         result = dentry_open(dentry, mqueue_mnt, oflag);
 615         /*
 616          * dentry_open() took a persistent mnt_want_write(),
 617          * so we can now drop this one.
 618          */
 619         mnt_drop_write(mqueue_mnt);
 620         return result;
 621
 622 out_drop_write:
 623         mnt_drop_write(mqueue_mnt);
 624 out:
 625         dput(dentry);
 626         mntput(mqueue_mnt);
 627         return ERR_PTR(ret);
 628 }
 629
 630 /* Opens existing queue */
 631 static struct file *do_open(struct dentry *dentry, int oflag)
 632 {
 633 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 634                                         MAY_READ | MAY_WRITE };
 635
 636         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 637                 dput(dentry);
 638                 mntput(mqueue_mnt);
 639                 return ERR_PTR(-EINVAL);
 640         }
 641
 642         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 643                 dput(dentry);
 644                 mntput(mqueue_mnt);
 645                 return ERR_PTR(-EACCES);
 646         }
 647
 648         return dentry_open(dentry, mqueue_mnt, oflag);
 649 }
 650
 651 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
 652                 struct mq_attr __user *, u_attr)
 653 {
 654         struct dentry *dentry;
 655         struct file *filp;
 656         char *name;
 657         int fd, error;
 658
 659         error = audit_mq_open(oflag, mode, u_attr);
 660         if (error != 0)
 661                 return error;
 662
 663         if (IS_ERR(name = getname(u_name)))
 664                 return PTR_ERR(name);
 665
 666         fd = get_unused_fd_flags(O_CLOEXEC);
 667         if (fd < 0)
 668                 goto out_putname;
 669
 670         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 671         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 672         if (IS_ERR(dentry)) {
 673                 error = PTR_ERR(dentry);
 674                 goto out_err;
 675         }
 676         mntget(mqueue_mnt);
 677
 678         if (oflag & O_CREAT) {
 679                 if (dentry->d_inode) {  /* entry already exists */
 680                         audit_inode(name, dentry);
 681                         error = -EEXIST;
 682                         if (oflag & O_EXCL)
 683                                 goto out;
 684                         filp = do_open(dentry, oflag);
 685                 } else {
 686                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 687                                                 oflag, mode, u_attr);
 688                 }
 689         } else {
 690                 error = -ENOENT;
 691                 if (!dentry->d_inode)
 692                         goto out;
 693                 audit_inode(name, dentry);
 694                 filp = do_open(dentry, oflag);
 695         }
 696
 697         if (IS_ERR(filp)) {
 698                 error = PTR_ERR(filp);
 699                 goto out_putfd;
 700         }
 701
 702         fd_install(fd, filp);
 703         goto out_upsem;
 704
 705 out:
 706         dput(dentry);
 707         mntput(mqueue_mnt);
 708 out_putfd:
 709         put_unused_fd(fd);
 710 out_err:
 711         fd = error;
 712 out_upsem:
 713         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 714 out_putname:
 715         putname(name);
 716         return fd;
 717 }
 718
 719 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 720 {
 721         int err;
 722         char *name;
 723         struct dentry *dentry;
 724         struct inode *inode = NULL;
 725
 726         name = getname(u_name);
 727         if (IS_ERR(name))
 728                 return PTR_ERR(name);
 729
 730         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
 731                         I_MUTEX_PARENT);
 732         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 733         if (IS_ERR(dentry)) {
 734                 err = PTR_ERR(dentry);
 735                 goto out_unlock;
 736         }
 737
 738         if (!dentry->d_inode) {
 739                 err = -ENOENT;
 740                 goto out_err;
 741         }
 742
 743         inode = dentry->d_inode;
 744         if (inode)
 745                 atomic_inc(&inode->i_count);
 746         err = mnt_want_write(mqueue_mnt);
 747         if (err)
 748                 goto out_err;
 749         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 750         mnt_drop_write(mqueue_mnt);
 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 for 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 SYSCALL_DEFINE5(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         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
 826         if (ret != 0)
 827                 return ret;
 828
 829         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 830                 return -EINVAL;
 831
 832         timeout = prepare_timeout(u_abs_timeout);
 833
 834         ret = -EBADF;
 835         filp = fget(mqdes);
 836         if (unlikely(!filp))
 837                 goto out;
 838
 839         inode = filp->f_path.dentry->d_inode;
 840         if (unlikely(filp->f_op != &mqueue_file_operations))
 841                 goto out_fput;
 842         info = MQUEUE_I(inode);
 843         audit_inode(NULL, filp->f_path.dentry);
 844
 845         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 846                 goto out_fput;
 847
 848         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 849                 ret = -EMSGSIZE;
 850                 goto out_fput;
 851         }
 852
 853         /* First try to allocate memory, before doing anything with
 854          * existing queues. */
 855         msg_ptr = load_msg(u_msg_ptr, msg_len);
 856         if (IS_ERR(msg_ptr)) {
 857                 ret = PTR_ERR(msg_ptr);
 858                 goto out_fput;
 859         }
 860         msg_ptr->m_ts = msg_len;
 861         msg_ptr->m_type = msg_prio;
 862
 863         spin_lock(&info->lock);
 864
 865         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 866                 if (filp->f_flags & O_NONBLOCK) {
 867                         spin_unlock(&info->lock);
 868                         ret = -EAGAIN;
 869                 } else if (unlikely(timeout < 0)) {
 870                         spin_unlock(&info->lock);
 871                         ret = timeout;
 872                 } else {
 873                         wait.task = current;
 874                         wait.msg = (void *) msg_ptr;
 875                         wait.state = STATE_NONE;
 876                         ret = wq_sleep(info, SEND, timeout, &wait);
 877                 }
 878                 if (ret < 0)
 879                         free_msg(msg_ptr);
 880         } else {
 881                 receiver = wq_get_first_waiter(info, RECV);
 882                 if (receiver) {
 883                         pipelined_send(info, msg_ptr, receiver);
 884                 } else {
 885                         /* adds message to the queue */
 886                         msg_insert(msg_ptr, info);
 887                         __do_notify(info);
 888                 }
 889                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 890                                 CURRENT_TIME;
 891                 spin_unlock(&info->lock);
 892                 ret = 0;
 893         }
 894 out_fput:
 895         fput(filp);
 896 out:
 897         return ret;
 898 }
 899
 900 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 901                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 902                 const struct timespec __user *, u_abs_timeout)
 903 {
 904         long timeout;
 905         ssize_t ret;
 906         struct msg_msg *msg_ptr;
 907         struct file *filp;
 908         struct inode *inode;
 909         struct mqueue_inode_info *info;
 910         struct ext_wait_queue wait;
 911
 912         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
 913         if (ret != 0)
 914                 return ret;
 915
 916         timeout = prepare_timeout(u_abs_timeout);
 917
 918         ret = -EBADF;
 919         filp = fget(mqdes);
 920         if (unlikely(!filp))
 921                 goto out;
 922
 923         inode = filp->f_path.dentry->d_inode;
 924         if (unlikely(filp->f_op != &mqueue_file_operations))
 925                 goto out_fput;
 926         info = MQUEUE_I(inode);
 927         audit_inode(NULL, filp->f_path.dentry);
 928
 929         if (unlikely(!(filp->f_mode & FMODE_READ)))
 930                 goto out_fput;
 931
 932         /* checks if buffer is big enough */
 933         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 934                 ret = -EMSGSIZE;
 935                 goto out_fput;
 936         }
 937
 938         spin_lock(&info->lock);
 939         if (info->attr.mq_curmsgs == 0) {
 940                 if (filp->f_flags & O_NONBLOCK) {
 941                         spin_unlock(&info->lock);
 942                         ret = -EAGAIN;
 943                         msg_ptr = NULL;
 944                 } else if (unlikely(timeout < 0)) {
 945                         spin_unlock(&info->lock);
 946                         ret = timeout;
 947                         msg_ptr = NULL;
 948                 } else {
 949                         wait.task = current;
 950                         wait.state = STATE_NONE;
 951                         ret = wq_sleep(info, RECV, timeout, &wait);
 952                         msg_ptr = wait.msg;
 953                 }
 954         } else {
 955                 msg_ptr = msg_get(info);
 956
 957                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 958                                 CURRENT_TIME;
 959
 960                 /* There is now free space in queue. */
 961                 pipelined_receive(info);
 962                 spin_unlock(&info->lock);
 963                 ret = 0;
 964         }
 965         if (ret == 0) {
 966                 ret = msg_ptr->m_ts;
 967
 968                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 969                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 970                         ret = -EFAULT;
 971                 }
 972                 free_msg(msg_ptr);
 973         }
 974 out_fput:
 975         fput(filp);
 976 out:
 977         return ret;
 978 }
 979
 980 /*
 981  * Notes: the case when user wants us to deregister (with NULL as pointer)
 982  * and he isn't currently owner of notification, will be silently discarded.
 983  * It isn't explicitly defined in the POSIX.
 984  */
 985 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
 986                 const struct sigevent __user *, u_notification)
 987 {
 988         int ret;
 989         struct file *filp;
 990         struct sock *sock;
 991         struct inode *inode;
 992         struct sigevent notification;
 993         struct mqueue_inode_info *info;
 994         struct sk_buff *nc;
 995
 996         ret = audit_mq_notify(mqdes, u_notification);
 997         if (ret != 0)
 998                 return ret;
 999
1000         nc = NULL;
1001         sock = NULL;
1002         if (u_notification != NULL) {
1003                 if (copy_from_user(&notification, u_notification,
1004                                         sizeof(struct sigevent)))
1005                         return -EFAULT;
1006
1007                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1008                              notification.sigev_notify != SIGEV_SIGNAL &&
1009                              notification.sigev_notify != SIGEV_THREAD))
1010                         return -EINVAL;
1011                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1012                         !valid_signal(notification.sigev_signo)) {
1013                         return -EINVAL;
1014                 }
1015                 if (notification.sigev_notify == SIGEV_THREAD) {
1016                         long timeo;
1017
1018                         /* create the notify skb */
1019                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1020                         ret = -ENOMEM;
1021                         if (!nc)
1022                                 goto out;
1023                         ret = -EFAULT;
1024                         if (copy_from_user(nc->data,
1025                                         notification.sigev_value.sival_ptr,
1026                                         NOTIFY_COOKIE_LEN)) {
1027                                 goto out;
1028                         }
1029
1030                         /* TODO: add a header? */
1031                         skb_put(nc, NOTIFY_COOKIE_LEN);
1032                         /* and attach it to the socket */
1033 retry:
1034                         filp = fget(notification.sigev_signo);
1035                         ret = -EBADF;
1036                         if (!filp)
1037                                 goto out;
1038                         sock = netlink_getsockbyfilp(filp);
1039                         fput(filp);
1040                         if (IS_ERR(sock)) {
1041                                 ret = PTR_ERR(sock);
1042                                 sock = NULL;
1043                                 goto out;
1044                         }
1045
1046                         timeo = MAX_SCHEDULE_TIMEOUT;
1047                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1048                         if (ret == 1)
1049                                 goto retry;
1050                         if (ret) {
1051                                 sock = NULL;
1052                                 nc = NULL;
1053                                 goto out;
1054                         }
1055                 }
1056         }
1057
1058         ret = -EBADF;
1059         filp = fget(mqdes);
1060         if (!filp)
1061                 goto out;
1062
1063         inode = filp->f_path.dentry->d_inode;
1064         if (unlikely(filp->f_op != &mqueue_file_operations))
1065                 goto out_fput;
1066         info = MQUEUE_I(inode);
1067
1068         ret = 0;
1069         spin_lock(&info->lock);
1070         if (u_notification == NULL) {
1071                 if (info->notify_owner == task_tgid(current)) {
1072                         remove_notification(info);
1073                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1074                 }
1075         } else if (info->notify_owner != NULL) {
1076                 ret = -EBUSY;
1077         } else {
1078                 switch (notification.sigev_notify) {
1079                 case SIGEV_NONE:
1080                         info->notify.sigev_notify = SIGEV_NONE;
1081                         break;
1082                 case SIGEV_THREAD:
1083                         info->notify_sock = sock;
1084                         info->notify_cookie = nc;
1085                         sock = NULL;
1086                         nc = NULL;
1087                         info->notify.sigev_notify = SIGEV_THREAD;
1088                         break;
1089                 case SIGEV_SIGNAL:
1090                         info->notify.sigev_signo = notification.sigev_signo;
1091                         info->notify.sigev_value = notification.sigev_value;
1092                         info->notify.sigev_notify = SIGEV_SIGNAL;
1093                         break;
1094                 }
1095
1096                 info->notify_owner = get_pid(task_tgid(current));
1097                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1098         }
1099         spin_unlock(&info->lock);
1100 out_fput:
1101         fput(filp);
1102 out:
1103         if (sock) {
1104                 netlink_detachskb(sock, nc);
1105         } else if (nc) {
1106                 dev_kfree_skb(nc);
1107         }
1108         return ret;
1109 }
1110
1111 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1112                 const struct mq_attr __user *, u_mqstat,
1113                 struct mq_attr __user *, u_omqstat)
1114 {
1115         int ret;
1116         struct mq_attr mqstat, omqstat;
1117         struct file *filp;
1118         struct inode *inode;
1119         struct mqueue_inode_info *info;
1120
1121         if (u_mqstat != NULL) {
1122                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1123                         return -EFAULT;
1124                 if (mqstat.mq_flags & (~O_NONBLOCK))
1125                         return -EINVAL;
1126         }
1127
1128         ret = -EBADF;
1129         filp = fget(mqdes);
1130         if (!filp)
1131                 goto out;
1132
1133         inode = filp->f_path.dentry->d_inode;
1134         if (unlikely(filp->f_op != &mqueue_file_operations))
1135                 goto out_fput;
1136         info = MQUEUE_I(inode);
1137
1138         spin_lock(&info->lock);
1139
1140         omqstat = info->attr;
1141         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1142         if (u_mqstat) {
1143                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1144                 if (ret != 0) {
1145                         spin_unlock(&info->lock);
1146                         goto out_fput;
1147                 }
1148                 if (mqstat.mq_flags & O_NONBLOCK)
1149                         filp->f_flags |= O_NONBLOCK;
1150                 else
1151                         filp->f_flags &= ~O_NONBLOCK;
1152
1153                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1154         }
1155
1156         spin_unlock(&info->lock);
1157
1158         ret = 0;
1159         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1160                                                 sizeof(struct mq_attr)))
1161                 ret = -EFAULT;
1162
1163 out_fput:
1164         fput(filp);
1165 out:
1166         return ret;
1167 }
1168
1169 static const struct inode_operations mqueue_dir_inode_operations = {
1170         .lookup = simple_lookup,
1171         .create = mqueue_create,
1172         .unlink = mqueue_unlink,
1173 };
1174
1175 static const struct file_operations mqueue_file_operations = {
1176         .flush = mqueue_flush_file,
1177         .poll = mqueue_poll_file,
1178         .read = mqueue_read_file,
1179 };
1180
1181 static struct super_operations mqueue_super_ops = {
1182         .alloc_inode = mqueue_alloc_inode,
1183         .destroy_inode = mqueue_destroy_inode,
1184         .statfs = simple_statfs,
1185         .delete_inode = mqueue_delete_inode,
1186         .drop_inode = generic_delete_inode,
1187 };
1188
1189 static struct file_system_type mqueue_fs_type = {
1190         .name = "mqueue",
1191         .get_sb = mqueue_get_sb,
1192         .kill_sb = kill_litter_super,
1193 };
1194
1195 static int msg_max_limit_min = DFLT_MSGMAX;
1196 static int msg_max_limit_max = HARD_MSGMAX;
1197
1198 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1199 static int msg_maxsize_limit_max = INT_MAX;
1200
1201 static ctl_table mq_sysctls[] = {
1202         {
1203                 .procname       = "queues_max",
1204                 .data           = &queues_max,
1205                 .maxlen         = sizeof(int),
1206                 .mode           = 0644,
1207                 .proc_handler   = &proc_dointvec,
1208         },
1209         {
1210                 .procname       = "msg_max",
1211                 .data           = &msg_max,
1212                 .maxlen         = sizeof(int),
1213                 .mode           = 0644,
1214                 .proc_handler   = &proc_dointvec_minmax,
1215                 .extra1         = &msg_max_limit_min,
1216                 .extra2         = &msg_max_limit_max,
1217         },
1218         {
1219                 .procname       = "msgsize_max",
1220                 .data           = &msgsize_max,
1221                 .maxlen         = sizeof(int),
1222                 .mode           = 0644,
1223                 .proc_handler   = &proc_dointvec_minmax,
1224                 .extra1         = &msg_maxsize_limit_min,
1225                 .extra2         = &msg_maxsize_limit_max,
1226         },
1227         { .ctl_name = 0 }
1228 };
1229
1230 static ctl_table mq_sysctl_dir[] = {
1231         {
1232                 .procname       = "mqueue",
1233                 .mode           = 0555,
1234                 .child          = mq_sysctls,
1235         },
1236         { .ctl_name = 0 }
1237 };
1238
1239 static ctl_table mq_sysctl_root[] = {
1240         {
1241                 .ctl_name       = CTL_FS,
1242                 .procname       = "fs",
1243                 .mode           = 0555,
1244                 .child          = mq_sysctl_dir,
1245         },
1246         { .ctl_name = 0 }
1247 };
1248
1249 static int __init init_mqueue_fs(void)
1250 {
1251         int error;
1252
1253         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1254                                 sizeof(struct mqueue_inode_info), 0,
1255                                 SLAB_HWCACHE_ALIGN, init_once);
1256         if (mqueue_inode_cachep == NULL)
1257                 return -ENOMEM;
1258
1259         /* ignore failues - they are not fatal */
1260         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1261
1262         error = register_filesystem(&mqueue_fs_type);
1263         if (error)
1264                 goto out_sysctl;
1265
1266         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1267                 error = PTR_ERR(mqueue_mnt);
1268                 goto out_filesystem;
1269         }
1270
1271         /* internal initialization - not common for vfs */
1272         queues_count = 0;
1273         spin_lock_init(&mq_lock);
1274
1275         return 0;
1276
1277 out_filesystem:
1278         unregister_filesystem(&mqueue_fs_type);
1279 out_sysctl:
1280         if (mq_sysctl_table)
1281                 unregister_sysctl_table(mq_sysctl_table);
1282         kmem_cache_destroy(mqueue_inode_cachep);
1283         return error;
1284 }
1285
1286 __initcall(init_mqueue_fs);