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(struct kmem_cache *cachep, 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         size_t slen;
 322         loff_t o;
 323
 324         if (!count)
 325                 return 0;
 326
 327         spin_lock(&info->lock);
 328         snprintf(buffer, sizeof(buffer),
 329                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 330                         info->qsize,
 331                         info->notify_owner ? info->notify.sigev_notify : 0,
 332                         (info->notify_owner &&
 333                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 334                                 info->notify.sigev_signo : 0,
 335                         pid_nr_ns(info->notify_owner,
 336                                 current->nsproxy->pid_ns));
 337         spin_unlock(&info->lock);
 338         buffer[sizeof(buffer)-1] = '\0';
 339         slen = strlen(buffer)+1;
 340
 341         o = *off;
 342         if (o > slen)
 343                 return 0;
 344
 345         if (o + count > slen)
 346                 count = slen - o;
 347
 348         if (copy_to_user(u_data, buffer + o, count))
 349                 return -EFAULT;
 350
 351         *off = o + count;
 352         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 353         return count;
 354 }
 355
 356 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 357 {
 358         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 359
 360         spin_lock(&info->lock);
 361         if (task_tgid(current) == info->notify_owner)
 362                 remove_notification(info);
 363
 364         spin_unlock(&info->lock);
 365         return 0;
 366 }
 367
 368 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 369 {
 370         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 371         int retval = 0;
 372
 373         poll_wait(filp, &info->wait_q, poll_tab);
 374
 375         spin_lock(&info->lock);
 376         if (info->attr.mq_curmsgs)
 377                 retval = POLLIN | POLLRDNORM;
 378
 379         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 380                 retval |= POLLOUT | POLLWRNORM;
 381         spin_unlock(&info->lock);
 382
 383         return retval;
 384 }
 385
 386 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 387 static void wq_add(struct mqueue_inode_info *info, int sr,
 388                         struct ext_wait_queue *ewp)
 389 {
 390         struct ext_wait_queue *walk;
 391
 392         ewp->task = current;
 393
 394         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 395                 if (walk->task->static_prio <= current->static_prio) {
 396                         list_add_tail(&ewp->list, &walk->list);
 397                         return;
 398                 }
 399         }
 400         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 401 }
 402
 403 /*
 404  * Puts current task to sleep. Caller must hold queue lock. After return
 405  * lock isn't held.
 406  * sr: SEND or RECV
 407  */
 408 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 409                         long timeout, struct ext_wait_queue *ewp)
 410 {
 411         int retval;
 412         signed long time;
 413
 414         wq_add(info, sr, ewp);
 415
 416         for (;;) {
 417                 set_current_state(TASK_INTERRUPTIBLE);
 418
 419                 spin_unlock(&info->lock);
 420                 time = schedule_timeout(timeout);
 421
 422                 while (ewp->state == STATE_PENDING)
 423                         cpu_relax();
 424
 425                 if (ewp->state == STATE_READY) {
 426                         retval = 0;
 427                         goto out;
 428                 }
 429                 spin_lock(&info->lock);
 430                 if (ewp->state == STATE_READY) {
 431                         retval = 0;
 432                         goto out_unlock;
 433                 }
 434                 if (signal_pending(current)) {
 435                         retval = -ERESTARTSYS;
 436                         break;
 437                 }
 438                 if (time == 0) {
 439                         retval = -ETIMEDOUT;
 440                         break;
 441                 }
 442         }
 443         list_del(&ewp->list);
 444 out_unlock:
 445         spin_unlock(&info->lock);
 446 out:
 447         return retval;
 448 }
 449
 450 /*
 451  * Returns waiting task that should be serviced first or NULL if none exists
 452  */
 453 static struct ext_wait_queue *wq_get_first_waiter(
 454                 struct mqueue_inode_info *info, int sr)
 455 {
 456         struct list_head *ptr;
 457
 458         ptr = info->e_wait_q[sr].list.prev;
 459         if (ptr == &info->e_wait_q[sr].list)
 460                 return NULL;
 461         return list_entry(ptr, struct ext_wait_queue, list);
 462 }
 463
 464 /* Auxiliary functions to manipulate messages' list */
 465 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 466 {
 467         int k;
 468
 469         k = info->attr.mq_curmsgs - 1;
 470         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 471                 info->messages[k + 1] = info->messages[k];
 472                 k--;
 473         }
 474         info->attr.mq_curmsgs++;
 475         info->qsize += ptr->m_ts;
 476         info->messages[k + 1] = ptr;
 477 }
 478
 479 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 480 {
 481         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 482         return info->messages[info->attr.mq_curmsgs];
 483 }
 484
 485 static inline void set_cookie(struct sk_buff *skb, char code)
 486 {
 487         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 488 }
 489
 490 /*
 491  * The next function is only to split too long sys_mq_timedsend
 492  */
 493 static void __do_notify(struct mqueue_inode_info *info)
 494 {
 495         /* notification
 496          * invoked when there is registered process and there isn't process
 497          * waiting synchronously for message AND state of queue changed from
 498          * empty to not empty. Here we are sure that no one is waiting
 499          * synchronously. */
 500         if (info->notify_owner &&
 501             info->attr.mq_curmsgs == 1) {
 502                 struct siginfo sig_i;
 503                 switch (info->notify.sigev_notify) {
 504                 case SIGEV_NONE:
 505                         break;
 506                 case SIGEV_SIGNAL:
 507                         /* sends signal */
 508
 509                         sig_i.si_signo = info->notify.sigev_signo;
 510                         sig_i.si_errno = 0;
 511                         sig_i.si_code = SI_MESGQ;
 512                         sig_i.si_value = info->notify.sigev_value;
 513                         sig_i.si_pid = task_pid_vnr(current);
 514                         sig_i.si_uid = current->uid;
 515
 516                         kill_pid_info(info->notify.sigev_signo,
 517                                       &sig_i, info->notify_owner);
 518                         break;
 519                 case SIGEV_THREAD:
 520                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 521                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 522                         break;
 523                 }
 524                 /* after notification unregisters process */
 525                 put_pid(info->notify_owner);
 526                 info->notify_owner = NULL;
 527         }
 528         wake_up(&info->wait_q);
 529 }
 530
 531 static long prepare_timeout(const struct timespec __user *u_arg)
 532 {
 533         struct timespec ts, nowts;
 534         long timeout;
 535
 536         if (u_arg) {
 537                 if (unlikely(copy_from_user(&ts, u_arg,
 538                                         sizeof(struct timespec))))
 539                         return -EFAULT;
 540
 541                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 542                         || ts.tv_nsec >= NSEC_PER_SEC))
 543                         return -EINVAL;
 544                 nowts = CURRENT_TIME;
 545                 /* first subtract as jiffies can't be too big */
 546                 ts.tv_sec -= nowts.tv_sec;
 547                 if (ts.tv_nsec < nowts.tv_nsec) {
 548                         ts.tv_nsec += NSEC_PER_SEC;
 549                         ts.tv_sec--;
 550                 }
 551                 ts.tv_nsec -= nowts.tv_nsec;
 552                 if (ts.tv_sec < 0)
 553                         return 0;
 554
 555                 timeout = timespec_to_jiffies(&ts) + 1;
 556         } else
 557                 return MAX_SCHEDULE_TIMEOUT;
 558
 559         return timeout;
 560 }
 561
 562 static void remove_notification(struct mqueue_inode_info *info)
 563 {
 564         if (info->notify_owner != NULL &&
 565             info->notify.sigev_notify == SIGEV_THREAD) {
 566                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 567                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 568         }
 569         put_pid(info->notify_owner);
 570         info->notify_owner = NULL;
 571 }
 572
 573 static int mq_attr_ok(struct mq_attr *attr)
 574 {
 575         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 576                 return 0;
 577         if (capable(CAP_SYS_RESOURCE)) {
 578                 if (attr->mq_maxmsg > HARD_MSGMAX)
 579                         return 0;
 580         } else {
 581                 if (attr->mq_maxmsg > msg_max ||
 582                                 attr->mq_msgsize > msgsize_max)
 583                         return 0;
 584         }
 585         /* check for overflow */
 586         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 587                 return 0;
 588         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 589             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 590             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 591                 return 0;
 592         return 1;
 593 }
 594
 595 /*
 596  * Invoked when creating a new queue via sys_mq_open
 597  */
 598 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 599                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 600 {
 601         struct mq_attr attr;
 602         int ret;
 603
 604         if (u_attr) {
 605                 ret = -EFAULT;
 606                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 607                         goto out;
 608                 ret = -EINVAL;
 609                 if (!mq_attr_ok(&attr))
 610                         goto out;
 611                 /* store for use during create */
 612                 dentry->d_fsdata = &attr;
 613         }
 614
 615         mode &= ~current->fs->umask;
 616         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 617         dentry->d_fsdata = NULL;
 618         if (ret)
 619                 goto out;
 620
 621         return dentry_open(dentry, mqueue_mnt, oflag);
 622
 623 out:
 624         dput(dentry);
 625         mntput(mqueue_mnt);
 626         return ERR_PTR(ret);
 627 }
 628
 629 /* Opens existing queue */
 630 static struct file *do_open(struct dentry *dentry, int oflag)
 631 {
 632 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 633                                         MAY_READ | MAY_WRITE };
 634
 635         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 636                 dput(dentry);
 637                 mntput(mqueue_mnt);
 638                 return ERR_PTR(-EINVAL);
 639         }
 640
 641         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 642                 dput(dentry);
 643                 mntput(mqueue_mnt);
 644                 return ERR_PTR(-EACCES);
 645         }
 646
 647         return dentry_open(dentry, mqueue_mnt, oflag);
 648 }
 649
 650 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 651                                 struct mq_attr __user *u_attr)
 652 {
 653         struct dentry *dentry;
 654         struct file *filp;
 655         char *name;
 656         int fd, error;
 657
 658         error = audit_mq_open(oflag, mode, u_attr);
 659         if (error != 0)
 660                 return error;
 661
 662         if (IS_ERR(name = getname(u_name)))
 663                 return PTR_ERR(name);
 664
 665         fd = get_unused_fd();
 666         if (fd < 0)
 667                 goto out_putname;
 668
 669         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 670         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 671         if (IS_ERR(dentry)) {
 672                 error = PTR_ERR(dentry);
 673                 goto out_err;
 674         }
 675         mntget(mqueue_mnt);
 676
 677         if (oflag & O_CREAT) {
 678                 if (dentry->d_inode) {  /* entry already exists */
 679                         audit_inode(name, dentry);
 680                         error = -EEXIST;
 681                         if (oflag & O_EXCL)
 682                                 goto out;
 683                         filp = do_open(dentry, oflag);
 684                 } else {
 685                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 686                                                 oflag, mode, u_attr);
 687                 }
 688         } else {
 689                 error = -ENOENT;
 690                 if (!dentry->d_inode)
 691                         goto out;
 692                 audit_inode(name, dentry);
 693                 filp = do_open(dentry, oflag);
 694         }
 695
 696         if (IS_ERR(filp)) {
 697                 error = PTR_ERR(filp);
 698                 goto out_putfd;
 699         }
 700
 701         set_close_on_exec(fd, 1);
 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 asmlinkage long sys_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
 747         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 748 out_err:
 749         dput(dentry);
 750
 751 out_unlock:
 752         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 753         putname(name);
 754         if (inode)
 755                 iput(inode);
 756
 757         return err;
 758 }
 759
 760 /* Pipelined send and receive functions.
 761  *
 762  * If a receiver finds no waiting message, then it registers itself in the
 763  * list of waiting receivers. A sender checks that list before adding the new
 764  * message into the message array. If there is a waiting receiver, then it
 765  * bypasses the message array and directly hands the message over to the
 766  * receiver.
 767  * The receiver accepts the message and returns without grabbing the queue
 768  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 769  * are necessary. The same algorithm is used for sysv semaphores, see
 770  * ipc/sem.c for more details.
 771  *
 772  * The same algorithm is used for senders.
 773  */
 774
 775 /* pipelined_send() - send a message directly to the task waiting in
 776  * sys_mq_timedreceive() (without inserting message into a queue).
 777  */
 778 static inline void pipelined_send(struct mqueue_inode_info *info,
 779                                   struct msg_msg *message,
 780                                   struct ext_wait_queue *receiver)
 781 {
 782         receiver->msg = message;
 783         list_del(&receiver->list);
 784         receiver->state = STATE_PENDING;
 785         wake_up_process(receiver->task);
 786         smp_wmb();
 787         receiver->state = STATE_READY;
 788 }
 789
 790 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 791  * gets its message and put to the queue (we have one free place for sure). */
 792 static inline void pipelined_receive(struct mqueue_inode_info *info)
 793 {
 794         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 795
 796         if (!sender) {
 797                 /* for poll */
 798                 wake_up_interruptible(&info->wait_q);
 799                 return;
 800         }
 801         msg_insert(sender->msg, info);
 802         list_del(&sender->list);
 803         sender->state = STATE_PENDING;
 804         wake_up_process(sender->task);
 805         smp_wmb();
 806         sender->state = STATE_READY;
 807 }
 808
 809 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 810         size_t msg_len, unsigned int msg_prio,
 811         const struct timespec __user *u_abs_timeout)
 812 {
 813         struct file *filp;
 814         struct inode *inode;
 815         struct ext_wait_queue wait;
 816         struct ext_wait_queue *receiver;
 817         struct msg_msg *msg_ptr;
 818         struct mqueue_inode_info *info;
 819         long timeout;
 820         int ret;
 821
 822         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
 823         if (ret != 0)
 824                 return ret;
 825
 826         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 827                 return -EINVAL;
 828
 829         timeout = prepare_timeout(u_abs_timeout);
 830
 831         ret = -EBADF;
 832         filp = fget(mqdes);
 833         if (unlikely(!filp))
 834                 goto out;
 835
 836         inode = filp->f_path.dentry->d_inode;
 837         if (unlikely(filp->f_op != &mqueue_file_operations))
 838                 goto out_fput;
 839         info = MQUEUE_I(inode);
 840         audit_inode(NULL, filp->f_path.dentry);
 841
 842         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 843                 goto out_fput;
 844
 845         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 846                 ret = -EMSGSIZE;
 847                 goto out_fput;
 848         }
 849
 850         /* First try to allocate memory, before doing anything with
 851          * existing queues. */
 852         msg_ptr = load_msg(u_msg_ptr, msg_len);
 853         if (IS_ERR(msg_ptr)) {
 854                 ret = PTR_ERR(msg_ptr);
 855                 goto out_fput;
 856         }
 857         msg_ptr->m_ts = msg_len;
 858         msg_ptr->m_type = msg_prio;
 859
 860         spin_lock(&info->lock);
 861
 862         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 863                 if (filp->f_flags & O_NONBLOCK) {
 864                         spin_unlock(&info->lock);
 865                         ret = -EAGAIN;
 866                 } else if (unlikely(timeout < 0)) {
 867                         spin_unlock(&info->lock);
 868                         ret = timeout;
 869                 } else {
 870                         wait.task = current;
 871                         wait.msg = (void *) msg_ptr;
 872                         wait.state = STATE_NONE;
 873                         ret = wq_sleep(info, SEND, timeout, &wait);
 874                 }
 875                 if (ret < 0)
 876                         free_msg(msg_ptr);
 877         } else {
 878                 receiver = wq_get_first_waiter(info, RECV);
 879                 if (receiver) {
 880                         pipelined_send(info, msg_ptr, receiver);
 881                 } else {
 882                         /* adds message to the queue */
 883                         msg_insert(msg_ptr, info);
 884                         __do_notify(info);
 885                 }
 886                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 887                                 CURRENT_TIME;
 888                 spin_unlock(&info->lock);
 889                 ret = 0;
 890         }
 891 out_fput:
 892         fput(filp);
 893 out:
 894         return ret;
 895 }
 896
 897 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 898         size_t msg_len, unsigned int __user *u_msg_prio,
 899         const struct timespec __user *u_abs_timeout)
 900 {
 901         long timeout;
 902         ssize_t ret;
 903         struct msg_msg *msg_ptr;
 904         struct file *filp;
 905         struct inode *inode;
 906         struct mqueue_inode_info *info;
 907         struct ext_wait_queue wait;
 908
 909         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
 910         if (ret != 0)
 911                 return ret;
 912
 913         timeout = prepare_timeout(u_abs_timeout);
 914
 915         ret = -EBADF;
 916         filp = fget(mqdes);
 917         if (unlikely(!filp))
 918                 goto out;
 919
 920         inode = filp->f_path.dentry->d_inode;
 921         if (unlikely(filp->f_op != &mqueue_file_operations))
 922                 goto out_fput;
 923         info = MQUEUE_I(inode);
 924         audit_inode(NULL, filp->f_path.dentry);
 925
 926         if (unlikely(!(filp->f_mode & FMODE_READ)))
 927                 goto out_fput;
 928
 929         /* checks if buffer is big enough */
 930         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 931                 ret = -EMSGSIZE;
 932                 goto out_fput;
 933         }
 934
 935         spin_lock(&info->lock);
 936         if (info->attr.mq_curmsgs == 0) {
 937                 if (filp->f_flags & O_NONBLOCK) {
 938                         spin_unlock(&info->lock);
 939                         ret = -EAGAIN;
 940                         msg_ptr = NULL;
 941                 } else if (unlikely(timeout < 0)) {
 942                         spin_unlock(&info->lock);
 943                         ret = timeout;
 944                         msg_ptr = NULL;
 945                 } else {
 946                         wait.task = current;
 947                         wait.state = STATE_NONE;
 948                         ret = wq_sleep(info, RECV, timeout, &wait);
 949                         msg_ptr = wait.msg;
 950                 }
 951         } else {
 952                 msg_ptr = msg_get(info);
 953
 954                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 955                                 CURRENT_TIME;
 956
 957                 /* There is now free space in queue. */
 958                 pipelined_receive(info);
 959                 spin_unlock(&info->lock);
 960                 ret = 0;
 961         }
 962         if (ret == 0) {
 963                 ret = msg_ptr->m_ts;
 964
 965                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 966                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 967                         ret = -EFAULT;
 968                 }
 969                 free_msg(msg_ptr);
 970         }
 971 out_fput:
 972         fput(filp);
 973 out:
 974         return ret;
 975 }
 976
 977 /*
 978  * Notes: the case when user wants us to deregister (with NULL as pointer)
 979  * and he isn't currently owner of notification, will be silently discarded.
 980  * It isn't explicitly defined in the POSIX.
 981  */
 982 asmlinkage long sys_mq_notify(mqd_t mqdes,
 983                                 const struct sigevent __user *u_notification)
 984 {
 985         int ret;
 986         struct file *filp;
 987         struct sock *sock;
 988         struct inode *inode;
 989         struct sigevent notification;
 990         struct mqueue_inode_info *info;
 991         struct sk_buff *nc;
 992
 993         ret = audit_mq_notify(mqdes, u_notification);
 994         if (ret != 0)
 995                 return ret;
 996
 997         nc = NULL;
 998         sock = NULL;
 999         if (u_notification != NULL) {
1000                 if (copy_from_user(&notification, u_notification,
1001                                         sizeof(struct sigevent)))
1002                         return -EFAULT;
1003
1004                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1005                              notification.sigev_notify != SIGEV_SIGNAL &&
1006                              notification.sigev_notify != SIGEV_THREAD))
1007                         return -EINVAL;
1008                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1009                         !valid_signal(notification.sigev_signo)) {
1010                         return -EINVAL;
1011                 }
1012                 if (notification.sigev_notify == SIGEV_THREAD) {
1013                         long timeo;
1014
1015                         /* create the notify skb */
1016                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1017                         ret = -ENOMEM;
1018                         if (!nc)
1019                                 goto out;
1020                         ret = -EFAULT;
1021                         if (copy_from_user(nc->data,
1022                                         notification.sigev_value.sival_ptr,
1023                                         NOTIFY_COOKIE_LEN)) {
1024                                 goto out;
1025                         }
1026
1027                         /* TODO: add a header? */
1028                         skb_put(nc, NOTIFY_COOKIE_LEN);
1029                         /* and attach it to the socket */
1030 retry:
1031                         filp = fget(notification.sigev_signo);
1032                         ret = -EBADF;
1033                         if (!filp)
1034                                 goto out;
1035                         sock = netlink_getsockbyfilp(filp);
1036                         fput(filp);
1037                         if (IS_ERR(sock)) {
1038                                 ret = PTR_ERR(sock);
1039                                 sock = NULL;
1040                                 goto out;
1041                         }
1042
1043                         timeo = MAX_SCHEDULE_TIMEOUT;
1044                         ret = netlink_attachskb(sock, nc, 0, &timeo, NULL);
1045                         if (ret == 1)
1046                                 goto retry;
1047                         if (ret) {
1048                                 sock = NULL;
1049                                 nc = NULL;
1050                                 goto out;
1051                         }
1052                 }
1053         }
1054
1055         ret = -EBADF;
1056         filp = fget(mqdes);
1057         if (!filp)
1058                 goto out;
1059
1060         inode = filp->f_path.dentry->d_inode;
1061         if (unlikely(filp->f_op != &mqueue_file_operations))
1062                 goto out_fput;
1063         info = MQUEUE_I(inode);
1064
1065         ret = 0;
1066         spin_lock(&info->lock);
1067         if (u_notification == NULL) {
1068                 if (info->notify_owner == task_tgid(current)) {
1069                         remove_notification(info);
1070                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1071                 }
1072         } else if (info->notify_owner != NULL) {
1073                 ret = -EBUSY;
1074         } else {
1075                 switch (notification.sigev_notify) {
1076                 case SIGEV_NONE:
1077                         info->notify.sigev_notify = SIGEV_NONE;
1078                         break;
1079                 case SIGEV_THREAD:
1080                         info->notify_sock = sock;
1081                         info->notify_cookie = nc;
1082                         sock = NULL;
1083                         nc = NULL;
1084                         info->notify.sigev_notify = SIGEV_THREAD;
1085                         break;
1086                 case SIGEV_SIGNAL:
1087                         info->notify.sigev_signo = notification.sigev_signo;
1088                         info->notify.sigev_value = notification.sigev_value;
1089                         info->notify.sigev_notify = SIGEV_SIGNAL;
1090                         break;
1091                 }
1092
1093                 info->notify_owner = get_pid(task_tgid(current));
1094                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1095         }
1096         spin_unlock(&info->lock);
1097 out_fput:
1098         fput(filp);
1099 out:
1100         if (sock) {
1101                 netlink_detachskb(sock, nc);
1102         } else if (nc) {
1103                 dev_kfree_skb(nc);
1104         }
1105         return ret;
1106 }
1107
1108 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1109                         const struct mq_attr __user *u_mqstat,
1110                         struct mq_attr __user *u_omqstat)
1111 {
1112         int ret;
1113         struct mq_attr mqstat, omqstat;
1114         struct file *filp;
1115         struct inode *inode;
1116         struct mqueue_inode_info *info;
1117
1118         if (u_mqstat != NULL) {
1119                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1120                         return -EFAULT;
1121                 if (mqstat.mq_flags & (~O_NONBLOCK))
1122                         return -EINVAL;
1123         }
1124
1125         ret = -EBADF;
1126         filp = fget(mqdes);
1127         if (!filp)
1128                 goto out;
1129
1130         inode = filp->f_path.dentry->d_inode;
1131         if (unlikely(filp->f_op != &mqueue_file_operations))
1132                 goto out_fput;
1133         info = MQUEUE_I(inode);
1134
1135         spin_lock(&info->lock);
1136
1137         omqstat = info->attr;
1138         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1139         if (u_mqstat) {
1140                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1141                 if (ret != 0)
1142                         goto out;
1143                 if (mqstat.mq_flags & O_NONBLOCK)
1144                         filp->f_flags |= O_NONBLOCK;
1145                 else
1146                         filp->f_flags &= ~O_NONBLOCK;
1147
1148                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1149         }
1150
1151         spin_unlock(&info->lock);
1152
1153         ret = 0;
1154         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1155                                                 sizeof(struct mq_attr)))
1156                 ret = -EFAULT;
1157
1158 out_fput:
1159         fput(filp);
1160 out:
1161         return ret;
1162 }
1163
1164 static const struct inode_operations mqueue_dir_inode_operations = {
1165         .lookup = simple_lookup,
1166         .create = mqueue_create,
1167         .unlink = mqueue_unlink,
1168 };
1169
1170 static const struct file_operations mqueue_file_operations = {
1171         .flush = mqueue_flush_file,
1172         .poll = mqueue_poll_file,
1173         .read = mqueue_read_file,
1174 };
1175
1176 static struct super_operations mqueue_super_ops = {
1177         .alloc_inode = mqueue_alloc_inode,
1178         .destroy_inode = mqueue_destroy_inode,
1179         .statfs = simple_statfs,
1180         .delete_inode = mqueue_delete_inode,
1181         .drop_inode = generic_delete_inode,
1182 };
1183
1184 static struct file_system_type mqueue_fs_type = {
1185         .name = "mqueue",
1186         .get_sb = mqueue_get_sb,
1187         .kill_sb = kill_litter_super,
1188 };
1189
1190 static int msg_max_limit_min = DFLT_MSGMAX;
1191 static int msg_max_limit_max = HARD_MSGMAX;
1192
1193 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1194 static int msg_maxsize_limit_max = INT_MAX;
1195
1196 static ctl_table mq_sysctls[] = {
1197         {
1198                 .procname       = "queues_max",
1199                 .data           = &queues_max,
1200                 .maxlen         = sizeof(int),
1201                 .mode           = 0644,
1202                 .proc_handler   = &proc_dointvec,
1203         },
1204         {
1205                 .procname       = "msg_max",
1206                 .data           = &msg_max,
1207                 .maxlen         = sizeof(int),
1208                 .mode           = 0644,
1209                 .proc_handler   = &proc_dointvec_minmax,
1210                 .extra1         = &msg_max_limit_min,
1211                 .extra2         = &msg_max_limit_max,
1212         },
1213         {
1214                 .procname       = "msgsize_max",
1215                 .data           = &msgsize_max,
1216                 .maxlen         = sizeof(int),
1217                 .mode           = 0644,
1218                 .proc_handler   = &proc_dointvec_minmax,
1219                 .extra1         = &msg_maxsize_limit_min,
1220                 .extra2         = &msg_maxsize_limit_max,
1221         },
1222         { .ctl_name = 0 }
1223 };
1224
1225 static ctl_table mq_sysctl_dir[] = {
1226         {
1227                 .procname       = "mqueue",
1228                 .mode           = 0555,
1229                 .child          = mq_sysctls,
1230         },
1231         { .ctl_name = 0 }
1232 };
1233
1234 static ctl_table mq_sysctl_root[] = {
1235         {
1236                 .ctl_name       = CTL_FS,
1237                 .procname       = "fs",
1238                 .mode           = 0555,
1239                 .child          = mq_sysctl_dir,
1240         },
1241         { .ctl_name = 0 }
1242 };
1243
1244 static int __init init_mqueue_fs(void)
1245 {
1246         int error;
1247
1248         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1249                                 sizeof(struct mqueue_inode_info), 0,
1250                                 SLAB_HWCACHE_ALIGN, init_once);
1251         if (mqueue_inode_cachep == NULL)
1252                 return -ENOMEM;
1253
1254         /* ignore failues - they are not fatal */
1255         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1256
1257         error = register_filesystem(&mqueue_fs_type);
1258         if (error)
1259                 goto out_sysctl;
1260
1261         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1262                 error = PTR_ERR(mqueue_mnt);
1263                 goto out_filesystem;
1264         }
1265
1266         /* internal initialization - not common for vfs */
1267         queues_count = 0;
1268         spin_lock_init(&mq_lock);
1269
1270         return 0;
1271
1272 out_filesystem:
1273         unregister_filesystem(&mqueue_fs_type);
1274 out_sysctl:
1275         if (mq_sysctl_table)
1276                 unregister_sysctl_table(mq_sysctl_table);
1277         kmem_cache_destroy(mqueue_inode_cachep);
1278         return error;
1279 }
1280
1281 __initcall(init_mqueue_fs);