ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (wrona@mat.uni.torun.pl)
   6  *
   7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          (manfred@colorfullife.com)
  10  *
  11  * This file is released under the GPL.
  12  */
  13
  14 #include <linux/init.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/file.h>
  17 #include <linux/mount.h>
  18 #include <linux/namei.h>
  19 #include <linux/sysctl.h>
  20 #include <linux/poll.h>
  21 #include <linux/mqueue.h>
  22 #include <linux/msg.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/netlink.h>
  25 #include <linux/syscalls.h>
  26 #include <linux/signal.h>
  27 #include <linux/vs_context.h>
  28 #include <linux/vs_limit.h>
  29 #include <net/sock.h>
  30 #include "util.h"
  31
  32 #define MQUEUE_MAGIC    0x19800202
  33 #define DIRENT_SIZE     20
  34 #define FILENT_SIZE     80
  35
  36 #define SEND            0
  37 #define RECV            1
  38
  39 #define STATE_NONE      0
  40 #define STATE_PENDING   1
  41 #define STATE_READY     2
  42
  43 /* used by sysctl */
  44 #define FS_MQUEUE       1
  45 #define CTL_QUEUESMAX   2
  46 #define CTL_MSGMAX      3
  47 #define CTL_MSGSIZEMAX  4
  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 #define NOTIFY_COOKIE_LEN       32
  56
  57 struct ext_wait_queue {         /* queue of sleeping tasks */
  58         struct task_struct *task;
  59         struct list_head list;
  60         struct msg_msg *msg;    /* ptr of loaded message */
  61         int state;              /* one of STATE_* values */
  62 };
  63
  64 struct mqueue_inode_info {
  65         spinlock_t lock;
  66         struct inode vfs_inode;
  67         wait_queue_head_t wait_q;
  68
  69         struct msg_msg **messages;
  70         struct mq_attr attr;
  71
  72         struct sigevent notify;
  73         pid_t notify_owner;
  74         struct user_struct *user;       /* user who created, for accouting */
  75         struct sock *notify_sock;
  76         struct sk_buff *notify_cookie;
  77
  78         /* for tasks waiting for free space and messages, respectively */
  79         struct ext_wait_queue e_wait_q[2];
  80
  81         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  82 };
  83
  84 static struct inode_operations mqueue_dir_inode_operations;
  85 static struct file_operations mqueue_file_operations;
  86 static struct super_operations mqueue_super_ops;
  87 static void remove_notification(struct mqueue_inode_info *info);
  88
  89 static spinlock_t mq_lock;
  90 static kmem_cache_t *mqueue_inode_cachep;
  91 static struct vfsmount *mqueue_mnt;
  92
  93 static unsigned int queues_count;
  94 static unsigned int queues_max  = DFLT_QUEUESMAX;
  95 static unsigned int msg_max     = DFLT_MSGMAX;
  96 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  97
  98 static struct ctl_table_header * mq_sysctl_table;
  99
 100 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 101 {
 102         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 103 }
 104
 105 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 106                                                         struct mq_attr *attr)
 107 {
 108         struct inode *inode;
 109
 110         inode = new_inode(sb);
 111         if (inode) {
 112                 inode->i_mode = mode;
 113                 inode->i_uid = current->fsuid;
 114                 inode->i_gid = current->fsgid;
 115                 inode->i_blksize = PAGE_CACHE_SIZE;
 116                 inode->i_blocks = 0;
 117                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 118                                 CURRENT_TIME;
 119
 120                 if (S_ISREG(mode)) {
 121                         struct mqueue_inode_info *info;
 122                         struct task_struct *p = current;
 123                         struct user_struct *u = p->user;
 124                         unsigned long mq_bytes, mq_msg_tblsz;
 125
 126                         inode->i_fop = &mqueue_file_operations;
 127                         inode->i_size = FILENT_SIZE;
 128                         /* mqueue specific info */
 129                         info = MQUEUE_I(inode);
 130                         spin_lock_init(&info->lock);
 131                         init_waitqueue_head(&info->wait_q);
 132                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 133                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 134                         info->messages = NULL;
 135                         info->notify_owner = 0;
 136                         info->qsize = 0;
 137                         info->user = NULL;      /* set when all is ok */
 138                         memset(&info->attr, 0, sizeof(info->attr));
 139                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 140                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 141                         if (attr) {
 142                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 143                                 info->attr.mq_msgsize = attr->mq_msgsize;
 144                         }
 145                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 146                         mq_bytes = (mq_msg_tblsz +
 147                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 148
 149                         spin_lock(&mq_lock);
 150                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 151                             u->mq_bytes + mq_bytes >
 152                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur ||
 153                             !vx_ipcmsg_avail(p->vx_info, mq_bytes)) {
 154                                 spin_unlock(&mq_lock);
 155                                 goto out_inode;
 156                         }
 157                         u->mq_bytes += mq_bytes;
 158                         vx_ipcmsg_add(p->vx_info, u, mq_bytes);
 159                         spin_unlock(&mq_lock);
 160
 161                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 162                         if (!info->messages) {
 163                                 spin_lock(&mq_lock);
 164                                 u->mq_bytes -= mq_bytes;
 165                                 vx_ipcmsg_sub(p->vx_info, u, mq_bytes);
 166                                 spin_unlock(&mq_lock);
 167                                 goto out_inode;
 168                         }
 169                         /* all is ok */
 170                         info->user = get_uid(u);
 171                 } else if (S_ISDIR(mode)) {
 172                         inode->i_nlink++;
 173                         /* Some things misbehave if size == 0 on a directory */
 174                         inode->i_size = 2 * DIRENT_SIZE;
 175                         inode->i_op = &mqueue_dir_inode_operations;
 176                         inode->i_fop = &simple_dir_operations;
 177                 }
 178         }
 179         return inode;
 180 out_inode:
 181         make_bad_inode(inode);
 182         iput(inode);
 183         return NULL;
 184 }
 185
 186 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 187 {
 188         struct inode *inode;
 189
 190         sb->s_blocksize = PAGE_CACHE_SIZE;
 191         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 192         sb->s_magic = MQUEUE_MAGIC;
 193         sb->s_op = &mqueue_super_ops;
 194
 195         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 196         if (!inode)
 197                 return -ENOMEM;
 198
 199         sb->s_root = d_alloc_root(inode);
 200         if (!sb->s_root) {
 201                 iput(inode);
 202                 return -ENOMEM;
 203         }
 204
 205         return 0;
 206 }
 207
 208 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 209                                          int flags, const char *dev_name,
 210                                          void *data)
 211 {
 212         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 213 }
 214
 215 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 216 {
 217         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 218
 219         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 220                 SLAB_CTOR_CONSTRUCTOR)
 221                 inode_init_once(&p->vfs_inode);
 222 }
 223
 224 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 225 {
 226         struct mqueue_inode_info *ei;
 227
 228         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 229         if (!ei)
 230                 return NULL;
 231         return &ei->vfs_inode;
 232 }
 233
 234 static void mqueue_destroy_inode(struct inode *inode)
 235 {
 236         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 237 }
 238
 239 static void mqueue_delete_inode(struct inode *inode)
 240 {
 241         struct mqueue_inode_info *info;
 242         struct user_struct *user;
 243         unsigned long mq_bytes;
 244         int i;
 245
 246         if (S_ISDIR(inode->i_mode)) {
 247                 clear_inode(inode);
 248                 return;
 249         }
 250         info = MQUEUE_I(inode);
 251         spin_lock(&info->lock);
 252         for (i = 0; i < info->attr.mq_curmsgs; i++)
 253                 free_msg(info->messages[i]);
 254         kfree(info->messages);
 255         spin_unlock(&info->lock);
 256
 257         clear_inode(inode);
 258
 259         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 260                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 261         user = info->user;
 262         if (user) {
 263                 struct vx_info *vxi = locate_vx_info(user->xid);
 264
 265                 spin_lock(&mq_lock);
 266                 user->mq_bytes -= mq_bytes;
 267                 vx_ipcmsg_sub(vxi, user, mq_bytes);
 268                 queues_count--;
 269                 spin_unlock(&mq_lock);
 270                 put_vx_info(vxi);
 271                 free_uid(user);
 272         }
 273 }
 274
 275 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 276                                 int mode, struct nameidata *nd)
 277 {
 278         struct inode *inode;
 279         struct mq_attr *attr = dentry->d_fsdata;
 280         int error;
 281
 282         spin_lock(&mq_lock);
 283         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 284                 error = -ENOSPC;
 285                 goto out_lock;
 286         }
 287         queues_count++;
 288         spin_unlock(&mq_lock);
 289
 290         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 291         if (!inode) {
 292                 error = -ENOMEM;
 293                 spin_lock(&mq_lock);
 294                 queues_count--;
 295                 goto out_lock;
 296         }
 297
 298         dir->i_size += DIRENT_SIZE;
 299         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 300
 301         d_instantiate(dentry, inode);
 302         dget(dentry);
 303         return 0;
 304 out_lock:
 305         spin_unlock(&mq_lock);
 306         return error;
 307 }
 308
 309 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 310 {
 311         struct inode *inode = dentry->d_inode;
 312
 313         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 314         dir->i_size -= DIRENT_SIZE;
 315         inode->i_nlink--;
 316         dput(dentry);
 317         return 0;
 318 }
 319
 320 /*
 321 *       This is routine for system read from queue file.
 322 *       To avoid mess with doing here some sort of mq_receive we allow
 323 *       to read only queue size & notification info (the only values
 324 *       that are interesting from user point of view and aren't accessible
 325 *       through std routines)
 326 */
 327 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 328                                 size_t count, loff_t * off)
 329 {
 330         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 331         char buffer[FILENT_SIZE];
 332         size_t slen;
 333         loff_t o;
 334
 335         if (!count)
 336                 return 0;
 337
 338         spin_lock(&info->lock);
 339         snprintf(buffer, sizeof(buffer),
 340                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 341                         info->qsize,
 342                         info->notify_owner ? info->notify.sigev_notify : 0,
 343                         (info->notify_owner &&
 344                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 345                                 info->notify.sigev_signo : 0,
 346                         info->notify_owner);
 347         spin_unlock(&info->lock);
 348         buffer[sizeof(buffer)-1] = '\0';
 349         slen = strlen(buffer)+1;
 350
 351         o = *off;
 352         if (o > slen)
 353                 return 0;
 354
 355         if (o + count > slen)
 356                 count = slen - o;
 357
 358         if (copy_to_user(u_data, buffer + o, count))
 359                 return -EFAULT;
 360
 361         *off = o + count;
 362         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 363         return count;
 364 }
 365
 366 static int mqueue_flush_file(struct file *filp)
 367 {
 368         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 369
 370         spin_lock(&info->lock);
 371         if (current->tgid == info->notify_owner)
 372                 remove_notification(info);
 373
 374         spin_unlock(&info->lock);
 375         return 0;
 376 }
 377
 378 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 379 {
 380         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 381         int retval = 0;
 382
 383         poll_wait(filp, &info->wait_q, poll_tab);
 384
 385         spin_lock(&info->lock);
 386         if (info->attr.mq_curmsgs)
 387                 retval = POLLIN | POLLRDNORM;
 388
 389         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 390                 retval |= POLLOUT | POLLWRNORM;
 391         spin_unlock(&info->lock);
 392
 393         return retval;
 394 }
 395
 396 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 397 static void wq_add(struct mqueue_inode_info *info, int sr,
 398                         struct ext_wait_queue *ewp)
 399 {
 400         struct ext_wait_queue *walk;
 401
 402         ewp->task = current;
 403
 404         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 405                 if (walk->task->static_prio <= current->static_prio) {
 406                         list_add_tail(&ewp->list, &walk->list);
 407                         return;
 408                 }
 409         }
 410         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 411 }
 412
 413 /*
 414  * Puts current task to sleep. Caller must hold queue lock. After return
 415  * lock isn't held.
 416  * sr: SEND or RECV
 417  */
 418 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 419                         long timeout, struct ext_wait_queue *ewp)
 420 {
 421         int retval;
 422         signed long time;
 423
 424         wq_add(info, sr, ewp);
 425
 426         for (;;) {
 427                 set_current_state(TASK_INTERRUPTIBLE);
 428
 429                 spin_unlock(&info->lock);
 430                 time = schedule_timeout(timeout);
 431
 432                 while (ewp->state == STATE_PENDING)
 433                         cpu_relax();
 434
 435                 if (ewp->state == STATE_READY) {
 436                         retval = 0;
 437                         goto out;
 438                 }
 439                 spin_lock(&info->lock);
 440                 if (ewp->state == STATE_READY) {
 441                         retval = 0;
 442                         goto out_unlock;
 443                 }
 444                 if (signal_pending(current)) {
 445                         retval = -ERESTARTSYS;
 446                         break;
 447                 }
 448                 if (time == 0) {
 449                         retval = -ETIMEDOUT;
 450                         break;
 451                 }
 452         }
 453         list_del(&ewp->list);
 454 out_unlock:
 455         spin_unlock(&info->lock);
 456 out:
 457         return retval;
 458 }
 459
 460 /*
 461  * Returns waiting task that should be serviced first or NULL if none exists
 462  */
 463 static struct ext_wait_queue *wq_get_first_waiter(
 464                 struct mqueue_inode_info *info, int sr)
 465 {
 466         struct list_head *ptr;
 467
 468         ptr = info->e_wait_q[sr].list.prev;
 469         if (ptr == &info->e_wait_q[sr].list)
 470                 return NULL;
 471         return list_entry(ptr, struct ext_wait_queue, list);
 472 }
 473
 474 /* Auxiliary functions to manipulate messages' list */
 475 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 476 {
 477         int k;
 478
 479         k = info->attr.mq_curmsgs - 1;
 480         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 481                 info->messages[k + 1] = info->messages[k];
 482                 k--;
 483         }
 484         info->attr.mq_curmsgs++;
 485         info->qsize += ptr->m_ts;
 486         info->messages[k + 1] = ptr;
 487 }
 488
 489 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 490 {
 491         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 492         return info->messages[info->attr.mq_curmsgs];
 493 }
 494
 495 static inline void set_cookie(struct sk_buff *skb, char code)
 496 {
 497         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 498 }
 499
 500 /*
 501  * The next function is only to split too long sys_mq_timedsend
 502  */
 503 static void __do_notify(struct mqueue_inode_info *info)
 504 {
 505         /* notification
 506          * invoked when there is registered process and there isn't process
 507          * waiting synchronously for message AND state of queue changed from
 508          * empty to not empty. Here we are sure that no one is waiting
 509          * synchronously. */
 510         if (info->notify_owner &&
 511             info->attr.mq_curmsgs == 1) {
 512                 struct siginfo sig_i;
 513                 switch (info->notify.sigev_notify) {
 514                 case SIGEV_NONE:
 515                         break;
 516                 case SIGEV_SIGNAL:
 517                         /* sends signal */
 518
 519                         sig_i.si_signo = info->notify.sigev_signo;
 520                         sig_i.si_errno = 0;
 521                         sig_i.si_code = SI_MESGQ;
 522                         sig_i.si_value = info->notify.sigev_value;
 523                         sig_i.si_pid = current->tgid;
 524                         sig_i.si_uid = current->uid;
 525
 526                         kill_proc_info(info->notify.sigev_signo,
 527                                        &sig_i, info->notify_owner);
 528                         break;
 529                 case SIGEV_THREAD:
 530                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 531                         netlink_sendskb(info->notify_sock,
 532                                         info->notify_cookie, 0);
 533                         break;
 534                 }
 535                 /* after notification unregisters process */
 536                 info->notify_owner = 0;
 537         }
 538         wake_up(&info->wait_q);
 539 }
 540
 541 static long prepare_timeout(const struct timespec __user *u_arg)
 542 {
 543         struct timespec ts, nowts;
 544         long timeout;
 545
 546         if (u_arg) {
 547                 if (unlikely(copy_from_user(&ts, u_arg,
 548                                         sizeof(struct timespec))))
 549                         return -EFAULT;
 550
 551                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 552                         || ts.tv_nsec >= NSEC_PER_SEC))
 553                         return -EINVAL;
 554                 nowts = CURRENT_TIME;
 555                 /* first subtract as jiffies can't be too big */
 556                 ts.tv_sec -= nowts.tv_sec;
 557                 if (ts.tv_nsec < nowts.tv_nsec) {
 558                         ts.tv_nsec += NSEC_PER_SEC;
 559                         ts.tv_sec--;
 560                 }
 561                 ts.tv_nsec -= nowts.tv_nsec;
 562                 if (ts.tv_sec < 0)
 563                         return 0;
 564
 565                 timeout = timespec_to_jiffies(&ts) + 1;
 566         } else
 567                 return MAX_SCHEDULE_TIMEOUT;
 568
 569         return timeout;
 570 }
 571
 572 static void remove_notification(struct mqueue_inode_info *info)
 573 {
 574         if (info->notify_owner != 0 &&
 575             info->notify.sigev_notify == SIGEV_THREAD) {
 576                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 577                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 578         }
 579         info->notify_owner = 0;
 580 }
 581
 582 static int mq_attr_ok(struct mq_attr *attr)
 583 {
 584         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 585                 return 0;
 586         if (capable(CAP_SYS_RESOURCE)) {
 587                 if (attr->mq_maxmsg > HARD_MSGMAX)
 588                         return 0;
 589         } else {
 590                 if (attr->mq_maxmsg > msg_max ||
 591                                 attr->mq_msgsize > msgsize_max)
 592                         return 0;
 593         }
 594         /* check for overflow */
 595         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 596                 return 0;
 597         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 598             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 599             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 600                 return 0;
 601         return 1;
 602 }
 603
 604 /*
 605  * Invoked when creating a new queue via sys_mq_open
 606  */
 607 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 608                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 609 {
 610         struct file *filp;
 611         struct mq_attr attr;
 612         int ret;
 613
 614         if (u_attr != NULL) {
 615                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 616                         return ERR_PTR(-EFAULT);
 617                 if (!mq_attr_ok(&attr))
 618                         return ERR_PTR(-EINVAL);
 619                 /* store for use during create */
 620                 dentry->d_fsdata = &attr;
 621         }
 622
 623         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 624         dentry->d_fsdata = NULL;
 625         if (ret)
 626                 return ERR_PTR(ret);
 627
 628         filp = dentry_open(dentry, mqueue_mnt, oflag);
 629         if (!IS_ERR(filp))
 630                 dget(dentry);
 631
 632         return filp;
 633 }
 634
 635 /* Opens existing queue */
 636 static struct file *do_open(struct dentry *dentry, int oflag)
 637 {
 638 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 639                                         MAY_READ | MAY_WRITE };
 640         struct file *filp;
 641
 642         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 643                 return ERR_PTR(-EINVAL);
 644
 645         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
 646                 return ERR_PTR(-EACCES);
 647
 648         filp = dentry_open(dentry, mqueue_mnt, oflag);
 649
 650         if (!IS_ERR(filp))
 651                 dget(dentry);
 652
 653         return filp;
 654 }
 655
 656 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 657                                 struct mq_attr __user *u_attr)
 658 {
 659         struct dentry *dentry;
 660         struct file *filp;
 661         char *name;
 662         int fd, error;
 663
 664         if (IS_ERR(name = getname(u_name)))
 665                 return PTR_ERR(name);
 666
 667         fd = get_unused_fd();
 668         if (fd < 0)
 669                 goto out_putname;
 670
 671         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 672         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 673         if (IS_ERR(dentry)) {
 674                 error = PTR_ERR(dentry);
 675                 goto out_err;
 676         }
 677         mntget(mqueue_mnt);
 678
 679         if (oflag & O_CREAT) {
 680                 if (dentry->d_inode) {  /* entry already exists */
 681                         filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
 682                                         do_open(dentry, oflag);
 683                 } else {
 684                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 685                                                 oflag, mode, u_attr);
 686                 }
 687         } else
 688                 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
 689                                         ERR_PTR(-ENOENT);
 690
 691         dput(dentry);
 692
 693         if (IS_ERR(filp)) {
 694                 error = PTR_ERR(filp);
 695                 goto out_putfd;
 696         }
 697
 698         set_close_on_exec(fd, 1);
 699         fd_install(fd, filp);
 700         goto out_upsem;
 701
 702 out_putfd:
 703         mntput(mqueue_mnt);
 704         put_unused_fd(fd);
 705 out_err:
 706         fd = error;
 707 out_upsem:
 708         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 709 out_putname:
 710         putname(name);
 711         return fd;
 712 }
 713
 714 asmlinkage long sys_mq_unlink(const char __user *u_name)
 715 {
 716         int err;
 717         char *name;
 718         struct dentry *dentry;
 719         struct inode *inode = NULL;
 720
 721         name = getname(u_name);
 722         if (IS_ERR(name))
 723                 return PTR_ERR(name);
 724
 725         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 726         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 727         if (IS_ERR(dentry)) {
 728                 err = PTR_ERR(dentry);
 729                 goto out_unlock;
 730         }
 731
 732         if (!dentry->d_inode) {
 733                 err = -ENOENT;
 734                 goto out_err;
 735         }
 736
 737         inode = dentry->d_inode;
 738         if (inode)
 739                 atomic_inc(&inode->i_count);
 740
 741         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 742 out_err:
 743         dput(dentry);
 744
 745 out_unlock:
 746         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 747         putname(name);
 748         if (inode)
 749                 iput(inode);
 750
 751         return err;
 752 }
 753
 754 /* Pipelined send and receive functions.
 755  *
 756  * If a receiver finds no waiting message, then it registers itself in the
 757  * list of waiting receivers. A sender checks that list before adding the new
 758  * message into the message array. If there is a waiting receiver, then it
 759  * bypasses the message array and directly hands the message over to the
 760  * receiver.
 761  * The receiver accepts the message and returns without grabbing the queue
 762  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 763  * are necessary. The same algorithm is used for sysv semaphores, see
 764  * ipc/sem.c fore more details.
 765  *
 766  * The same algorithm is used for senders.
 767  */
 768
 769 /* pipelined_send() - send a message directly to the task waiting in
 770  * sys_mq_timedreceive() (without inserting message into a queue).
 771  */
 772 static inline void pipelined_send(struct mqueue_inode_info *info,
 773                                   struct msg_msg *message,
 774                                   struct ext_wait_queue *receiver)
 775 {
 776         receiver->msg = message;
 777         list_del(&receiver->list);
 778         receiver->state = STATE_PENDING;
 779         wake_up_process(receiver->task);
 780         smp_wmb();
 781         receiver->state = STATE_READY;
 782 }
 783
 784 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 785  * gets its message and put to the queue (we have one free place for sure). */
 786 static inline void pipelined_receive(struct mqueue_inode_info *info)
 787 {
 788         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 789
 790         if (!sender) {
 791                 /* for poll */
 792                 wake_up_interruptible(&info->wait_q);
 793                 return;
 794         }
 795         msg_insert(sender->msg, info);
 796         list_del(&sender->list);
 797         sender->state = STATE_PENDING;
 798         wake_up_process(sender->task);
 799         smp_wmb();
 800         sender->state = STATE_READY;
 801 }
 802
 803 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 804         size_t msg_len, unsigned int msg_prio,
 805         const struct timespec __user *u_abs_timeout)
 806 {
 807         struct file *filp;
 808         struct inode *inode;
 809         struct ext_wait_queue wait;
 810         struct ext_wait_queue *receiver;
 811         struct msg_msg *msg_ptr;
 812         struct mqueue_inode_info *info;
 813         long timeout;
 814         int ret;
 815
 816         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 817                 return -EINVAL;
 818
 819         timeout = prepare_timeout(u_abs_timeout);
 820
 821         ret = -EBADF;
 822         filp = fget(mqdes);
 823         if (unlikely(!filp))
 824                 goto out;
 825
 826         inode = filp->f_dentry->d_inode;
 827         if (unlikely(filp->f_op != &mqueue_file_operations))
 828                 goto out_fput;
 829         info = MQUEUE_I(inode);
 830
 831         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 832                 goto out_fput;
 833
 834         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 835                 ret = -EMSGSIZE;
 836                 goto out_fput;
 837         }
 838
 839         /* First try to allocate memory, before doing anything with
 840          * existing queues. */
 841         msg_ptr = load_msg(u_msg_ptr, msg_len);
 842         if (IS_ERR(msg_ptr)) {
 843                 ret = PTR_ERR(msg_ptr);
 844                 goto out_fput;
 845         }
 846         msg_ptr->m_ts = msg_len;
 847         msg_ptr->m_type = msg_prio;
 848
 849         spin_lock(&info->lock);
 850
 851         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 852                 if (filp->f_flags & O_NONBLOCK) {
 853                         spin_unlock(&info->lock);
 854                         ret = -EAGAIN;
 855                 } else if (unlikely(timeout < 0)) {
 856                         spin_unlock(&info->lock);
 857                         ret = timeout;
 858                 } else {
 859                         wait.task = current;
 860                         wait.msg = (void *) msg_ptr;
 861                         wait.state = STATE_NONE;
 862                         ret = wq_sleep(info, SEND, timeout, &wait);
 863                 }
 864                 if (ret < 0)
 865                         free_msg(msg_ptr);
 866         } else {
 867                 receiver = wq_get_first_waiter(info, RECV);
 868                 if (receiver) {
 869                         pipelined_send(info, msg_ptr, receiver);
 870                 } else {
 871                         /* adds message to the queue */
 872                         msg_insert(msg_ptr, info);
 873                         __do_notify(info);
 874                 }
 875                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 876                                 CURRENT_TIME;
 877                 spin_unlock(&info->lock);
 878                 ret = 0;
 879         }
 880 out_fput:
 881         fput(filp);
 882 out:
 883         return ret;
 884 }
 885
 886 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 887         size_t msg_len, unsigned int __user *u_msg_prio,
 888         const struct timespec __user *u_abs_timeout)
 889 {
 890         long timeout;
 891         ssize_t ret;
 892         struct msg_msg *msg_ptr;
 893         struct file *filp;
 894         struct inode *inode;
 895         struct mqueue_inode_info *info;
 896         struct ext_wait_queue wait;
 897
 898         timeout = prepare_timeout(u_abs_timeout);
 899
 900         ret = -EBADF;
 901         filp = fget(mqdes);
 902         if (unlikely(!filp))
 903                 goto out;
 904
 905         inode = filp->f_dentry->d_inode;
 906         if (unlikely(filp->f_op != &mqueue_file_operations))
 907                 goto out_fput;
 908         info = MQUEUE_I(inode);
 909
 910         if (unlikely(!(filp->f_mode & FMODE_READ)))
 911                 goto out_fput;
 912
 913         /* checks if buffer is big enough */
 914         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 915                 ret = -EMSGSIZE;
 916                 goto out_fput;
 917         }
 918
 919         spin_lock(&info->lock);
 920         if (info->attr.mq_curmsgs == 0) {
 921                 if (filp->f_flags & O_NONBLOCK) {
 922                         spin_unlock(&info->lock);
 923                         ret = -EAGAIN;
 924                         msg_ptr = NULL;
 925                 } else if (unlikely(timeout < 0)) {
 926                         spin_unlock(&info->lock);
 927                         ret = timeout;
 928                         msg_ptr = NULL;
 929                 } else {
 930                         wait.task = current;
 931                         wait.state = STATE_NONE;
 932                         ret = wq_sleep(info, RECV, timeout, &wait);
 933                         msg_ptr = wait.msg;
 934                 }
 935         } else {
 936                 msg_ptr = msg_get(info);
 937
 938                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 939                                 CURRENT_TIME;
 940
 941                 /* There is now free space in queue. */
 942                 pipelined_receive(info);
 943                 spin_unlock(&info->lock);
 944                 ret = 0;
 945         }
 946         if (ret == 0) {
 947                 ret = msg_ptr->m_ts;
 948
 949                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 950                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 951                         ret = -EFAULT;
 952                 }
 953                 free_msg(msg_ptr);
 954         }
 955 out_fput:
 956         fput(filp);
 957 out:
 958         return ret;
 959 }
 960
 961 /*
 962  * Notes: the case when user wants us to deregister (with NULL as pointer)
 963  * and he isn't currently owner of notification, will be silently discarded.
 964  * It isn't explicitly defined in the POSIX.
 965  */
 966 asmlinkage long sys_mq_notify(mqd_t mqdes,
 967                                 const struct sigevent __user *u_notification)
 968 {
 969         int ret;
 970         struct file *filp;
 971         struct sock *sock;
 972         struct inode *inode;
 973         struct sigevent notification;
 974         struct mqueue_inode_info *info;
 975         struct sk_buff *nc;
 976
 977         nc = NULL;
 978         sock = NULL;
 979         if (u_notification != NULL) {
 980                 if (copy_from_user(&notification, u_notification,
 981                                         sizeof(struct sigevent)))
 982                         return -EFAULT;
 983
 984                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 985                              notification.sigev_notify != SIGEV_SIGNAL &&
 986                              notification.sigev_notify != SIGEV_THREAD))
 987                         return -EINVAL;
 988                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 989                         !valid_signal(notification.sigev_signo)) {
 990                         return -EINVAL;
 991                 }
 992                 if (notification.sigev_notify == SIGEV_THREAD) {
 993                         /* create the notify skb */
 994                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
 995                         ret = -ENOMEM;
 996                         if (!nc)
 997                                 goto out;
 998                         ret = -EFAULT;
 999                         if (copy_from_user(nc->data,
1000                                         notification.sigev_value.sival_ptr,
1001                                         NOTIFY_COOKIE_LEN)) {
1002                                 goto out;
1003                         }
1004
1005                         /* TODO: add a header? */
1006                         skb_put(nc, NOTIFY_COOKIE_LEN);
1007                         /* and attach it to the socket */
1008 retry:
1009                         filp = fget(notification.sigev_signo);
1010                         ret = -EBADF;
1011                         if (!filp)
1012                                 goto out;
1013                         sock = netlink_getsockbyfilp(filp);
1014                         fput(filp);
1015                         if (IS_ERR(sock)) {
1016                                 ret = PTR_ERR(sock);
1017                                 sock = NULL;
1018                                 goto out;
1019                         }
1020
1021                         ret = netlink_attachskb(sock, nc, 0, MAX_SCHEDULE_TIMEOUT);
1022                         if (ret == 1)
1023                                 goto retry;
1024                         if (ret) {
1025                                 sock = NULL;
1026                                 nc = NULL;
1027                                 goto out;
1028                         }
1029                 }
1030         }
1031
1032         ret = -EBADF;
1033         filp = fget(mqdes);
1034         if (!filp)
1035                 goto out;
1036
1037         inode = filp->f_dentry->d_inode;
1038         if (unlikely(filp->f_op != &mqueue_file_operations))
1039                 goto out_fput;
1040         info = MQUEUE_I(inode);
1041
1042         ret = 0;
1043         spin_lock(&info->lock);
1044         if (u_notification == NULL) {
1045                 if (info->notify_owner == current->tgid) {
1046                         remove_notification(info);
1047                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1048                 }
1049         } else if (info->notify_owner != 0) {
1050                 ret = -EBUSY;
1051         } else {
1052                 switch (notification.sigev_notify) {
1053                 case SIGEV_NONE:
1054                         info->notify.sigev_notify = SIGEV_NONE;
1055                         break;
1056                 case SIGEV_THREAD:
1057                         info->notify_sock = sock;
1058                         info->notify_cookie = nc;
1059                         sock = NULL;
1060                         nc = NULL;
1061                         info->notify.sigev_notify = SIGEV_THREAD;
1062                         break;
1063                 case SIGEV_SIGNAL:
1064                         info->notify.sigev_signo = notification.sigev_signo;
1065                         info->notify.sigev_value = notification.sigev_value;
1066                         info->notify.sigev_notify = SIGEV_SIGNAL;
1067                         break;
1068                 }
1069                 info->notify_owner = current->tgid;
1070                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1071         }
1072         spin_unlock(&info->lock);
1073 out_fput:
1074         fput(filp);
1075 out:
1076         if (sock) {
1077                 netlink_detachskb(sock, nc);
1078         } else if (nc) {
1079                 dev_kfree_skb(nc);
1080         }
1081         return ret;
1082 }
1083
1084 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1085                         const struct mq_attr __user *u_mqstat,
1086                         struct mq_attr __user *u_omqstat)
1087 {
1088         int ret;
1089         struct mq_attr mqstat, omqstat;
1090         struct file *filp;
1091         struct inode *inode;
1092         struct mqueue_inode_info *info;
1093
1094         if (u_mqstat != NULL) {
1095                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1096                         return -EFAULT;
1097                 if (mqstat.mq_flags & (~O_NONBLOCK))
1098                         return -EINVAL;
1099         }
1100
1101         ret = -EBADF;
1102         filp = fget(mqdes);
1103         if (!filp)
1104                 goto out;
1105
1106         inode = filp->f_dentry->d_inode;
1107         if (unlikely(filp->f_op != &mqueue_file_operations))
1108                 goto out_fput;
1109         info = MQUEUE_I(inode);
1110
1111         spin_lock(&info->lock);
1112
1113         omqstat = info->attr;
1114         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1115         if (u_mqstat) {
1116                 if (mqstat.mq_flags & O_NONBLOCK)
1117                         filp->f_flags |= O_NONBLOCK;
1118                 else
1119                         filp->f_flags &= ~O_NONBLOCK;
1120
1121                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1122         }
1123
1124         spin_unlock(&info->lock);
1125
1126         ret = 0;
1127         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1128                                                 sizeof(struct mq_attr)))
1129                 ret = -EFAULT;
1130
1131 out_fput:
1132         fput(filp);
1133 out:
1134         return ret;
1135 }
1136
1137 static struct inode_operations mqueue_dir_inode_operations = {
1138         .lookup = simple_lookup,
1139         .create = mqueue_create,
1140         .unlink = mqueue_unlink,
1141 };
1142
1143 static struct file_operations mqueue_file_operations = {
1144         .flush = mqueue_flush_file,
1145         .poll = mqueue_poll_file,
1146         .read = mqueue_read_file,
1147 };
1148
1149 static struct super_operations mqueue_super_ops = {
1150         .alloc_inode = mqueue_alloc_inode,
1151         .destroy_inode = mqueue_destroy_inode,
1152         .statfs = simple_statfs,
1153         .delete_inode = mqueue_delete_inode,
1154         .drop_inode = generic_delete_inode,
1155 };
1156
1157 static struct file_system_type mqueue_fs_type = {
1158         .name = "mqueue",
1159         .get_sb = mqueue_get_sb,
1160         .kill_sb = kill_litter_super,
1161 };
1162
1163 static int msg_max_limit_min = DFLT_MSGMAX;
1164 static int msg_max_limit_max = HARD_MSGMAX;
1165
1166 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1167 static int msg_maxsize_limit_max = INT_MAX;
1168
1169 static ctl_table mq_sysctls[] = {
1170         {
1171                 .ctl_name       = CTL_QUEUESMAX,
1172                 .procname       = "queues_max",
1173                 .data           = &queues_max,
1174                 .maxlen         = sizeof(int),
1175                 .mode           = 0644,
1176                 .proc_handler   = &proc_dointvec,
1177         },
1178         {
1179                 .ctl_name       = CTL_MSGMAX,
1180                 .procname       = "msg_max",
1181                 .data           = &msg_max,
1182                 .maxlen         = sizeof(int),
1183                 .mode           = 0644,
1184                 .proc_handler   = &proc_dointvec_minmax,
1185                 .extra1         = &msg_max_limit_min,
1186                 .extra2         = &msg_max_limit_max,
1187         },
1188         {
1189                 .ctl_name       = CTL_MSGSIZEMAX,
1190                 .procname       = "msgsize_max",
1191                 .data           = &msgsize_max,
1192                 .maxlen         = sizeof(int),
1193                 .mode           = 0644,
1194                 .proc_handler   = &proc_dointvec_minmax,
1195                 .extra1         = &msg_maxsize_limit_min,
1196                 .extra2         = &msg_maxsize_limit_max,
1197         },
1198         { .ctl_name = 0 }
1199 };
1200
1201 static ctl_table mq_sysctl_dir[] = {
1202         {
1203                 .ctl_name       = FS_MQUEUE,
1204                 .procname       = "mqueue",
1205                 .mode           = 0555,
1206                 .child          = mq_sysctls,
1207         },
1208         { .ctl_name = 0 }
1209 };
1210
1211 static ctl_table mq_sysctl_root[] = {
1212         {
1213                 .ctl_name       = CTL_FS,
1214                 .procname       = "fs",
1215                 .mode           = 0555,
1216                 .child          = mq_sysctl_dir,
1217         },
1218         { .ctl_name = 0 }
1219 };
1220
1221 static int __init init_mqueue_fs(void)
1222 {
1223         int error;
1224
1225         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1226                                 sizeof(struct mqueue_inode_info), 0,
1227                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1228         if (mqueue_inode_cachep == NULL)
1229                 return -ENOMEM;
1230
1231         /* ignore failues - they are not fatal */
1232         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1233
1234         error = register_filesystem(&mqueue_fs_type);
1235         if (error)
1236                 goto out_sysctl;
1237
1238         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1239                 error = PTR_ERR(mqueue_mnt);
1240                 goto out_filesystem;
1241         }
1242
1243         /* internal initialization - not common for vfs */
1244         queues_count = 0;
1245         spin_lock_init(&mq_lock);
1246
1247         return 0;
1248
1249 out_filesystem:
1250         unregister_filesystem(&mqueue_fs_type);
1251 out_sysctl:
1252         if (mq_sysctl_table)
1253                 unregister_sysctl_table(mq_sysctl_table);
1254         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1255                 printk(KERN_INFO
1256                         "mqueue_inode_cache: not all structures were freed\n");
1257         }
1258         return error;
1259 }
1260
1261 __initcall(init_mqueue_fs);