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/vs_context.h>
  27 #include <linux/vs_limit.h>
  28 #include <net/sock.h>
  29 #include "util.h"
  30
  31 #define MQUEUE_MAGIC    0x19800202
  32 #define DIRENT_SIZE     20
  33 #define FILENT_SIZE     80
  34
  35 #define SEND            0
  36 #define RECV            1
  37
  38 #define STATE_NONE      0
  39 #define STATE_PENDING   1
  40 #define STATE_READY     2
  41
  42 /* used by sysctl */
  43 #define FS_MQUEUE       1
  44 #define CTL_QUEUESMAX   2
  45 #define CTL_MSGMAX      3
  46 #define CTL_MSGSIZEMAX  4
  47
  48 /* default values */
  49 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  50 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  51 #define HARD_MSGMAX     (131072/sizeof(void*))
  52 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  53
  54 #define NOTIFY_COOKIE_LEN       32
  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         pid_t notify_owner;
  73         struct user_struct *user;       /* user who created, for accouting */
  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 struct inode_operations mqueue_dir_inode_operations;
  84 static 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 kmem_cache_t *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_blksize = PAGE_CACHE_SIZE;
 115                 inode->i_blocks = 0;
 116                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 117                                 CURRENT_TIME;
 118
 119                 if (S_ISREG(mode)) {
 120                         struct mqueue_inode_info *info;
 121                         struct task_struct *p = current;
 122                         struct user_struct *u = p->user;
 123                         unsigned long mq_bytes, mq_msg_tblsz;
 124
 125                         inode->i_fop = &mqueue_file_operations;
 126                         inode->i_size = FILENT_SIZE;
 127                         /* mqueue specific info */
 128                         info = MQUEUE_I(inode);
 129                         spin_lock_init(&info->lock);
 130                         init_waitqueue_head(&info->wait_q);
 131                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 132                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 133                         info->messages = NULL;
 134                         info->notify_owner = 0;
 135                         info->qsize = 0;
 136                         info->user = NULL;      /* set when all is ok */
 137                         memset(&info->attr, 0, sizeof(info->attr));
 138                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 139                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 140                         if (attr) {
 141                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 142                                 info->attr.mq_msgsize = attr->mq_msgsize;
 143                         }
 144                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 145                         mq_bytes = (mq_msg_tblsz +
 146                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 147
 148                         spin_lock(&mq_lock);
 149                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 150                             u->mq_bytes + mq_bytes >
 151                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur ||
 152                             !vx_ipcmsg_avail(p->vx_info, mq_bytes)) {
 153                                 spin_unlock(&mq_lock);
 154                                 goto out_inode;
 155                         }
 156                         u->mq_bytes += mq_bytes;
 157                         vx_ipcmsg_add(p->vx_info, u, mq_bytes);
 158                         spin_unlock(&mq_lock);
 159
 160                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 161                         if (!info->messages) {
 162                                 spin_lock(&mq_lock);
 163                                 u->mq_bytes -= mq_bytes;
 164                                 vx_ipcmsg_sub(p->vx_info, u, mq_bytes);
 165                                 spin_unlock(&mq_lock);
 166                                 goto out_inode;
 167                         }
 168                         /* all is ok */
 169                         info->user = get_uid(u);
 170                 } else if (S_ISDIR(mode)) {
 171                         inode->i_nlink++;
 172                         /* Some things misbehave if size == 0 on a directory */
 173                         inode->i_size = 2 * DIRENT_SIZE;
 174                         inode->i_op = &mqueue_dir_inode_operations;
 175                         inode->i_fop = &simple_dir_operations;
 176                 }
 177         }
 178         return inode;
 179 out_inode:
 180         make_bad_inode(inode);
 181         iput(inode);
 182         return NULL;
 183 }
 184
 185 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 186 {
 187         struct inode *inode;
 188
 189         sb->s_blocksize = PAGE_CACHE_SIZE;
 190         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 191         sb->s_magic = MQUEUE_MAGIC;
 192         sb->s_op = &mqueue_super_ops;
 193
 194         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 195         if (!inode)
 196                 return -ENOMEM;
 197
 198         sb->s_root = d_alloc_root(inode);
 199         if (!sb->s_root) {
 200                 iput(inode);
 201                 return -ENOMEM;
 202         }
 203
 204         return 0;
 205 }
 206
 207 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 208                                          int flags, const char *dev_name,
 209                                          void *data)
 210 {
 211         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 212 }
 213
 214 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 215 {
 216         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 217
 218         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 219                 SLAB_CTOR_CONSTRUCTOR)
 220                 inode_init_once(&p->vfs_inode);
 221 }
 222
 223 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 224 {
 225         struct mqueue_inode_info *ei;
 226
 227         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 228         if (!ei)
 229                 return NULL;
 230         return &ei->vfs_inode;
 231 }
 232
 233 static void mqueue_destroy_inode(struct inode *inode)
 234 {
 235         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 236 }
 237
 238 static void mqueue_delete_inode(struct inode *inode)
 239 {
 240         struct mqueue_inode_info *info;
 241         struct user_struct *user;
 242         unsigned long mq_bytes;
 243         int i;
 244
 245         if (S_ISDIR(inode->i_mode)) {
 246                 clear_inode(inode);
 247                 return;
 248         }
 249         info = MQUEUE_I(inode);
 250         spin_lock(&info->lock);
 251         for (i = 0; i < info->attr.mq_curmsgs; i++)
 252                 free_msg(info->messages[i]);
 253         kfree(info->messages);
 254         spin_unlock(&info->lock);
 255
 256         clear_inode(inode);
 257
 258         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 259                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 260         user = info->user;
 261         if (user) {
 262                 struct vx_info *vxi = locate_vx_info(user->xid);
 263
 264                 spin_lock(&mq_lock);
 265                 user->mq_bytes -= mq_bytes;
 266                 vx_ipcmsg_sub(vxi, user, mq_bytes);
 267                 queues_count--;
 268                 spin_unlock(&mq_lock);
 269                 put_vx_info(vxi);
 270                 free_uid(user);
 271         }
 272 }
 273
 274 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 275                                 int mode, struct nameidata *nd)
 276 {
 277         struct inode *inode;
 278         struct mq_attr *attr = dentry->d_fsdata;
 279         int error;
 280
 281         spin_lock(&mq_lock);
 282         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 283                 error = -ENOSPC;
 284                 goto out_lock;
 285         }
 286         queues_count++;
 287         spin_unlock(&mq_lock);
 288
 289         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 290         if (!inode) {
 291                 error = -ENOMEM;
 292                 spin_lock(&mq_lock);
 293                 queues_count--;
 294                 goto out_lock;
 295         }
 296
 297         dir->i_size += DIRENT_SIZE;
 298         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 299
 300         d_instantiate(dentry, inode);
 301         dget(dentry);
 302         return 0;
 303 out_lock:
 304         spin_unlock(&mq_lock);
 305         return error;
 306 }
 307
 308 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 309 {
 310         struct inode *inode = dentry->d_inode;
 311
 312         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 313         dir->i_size -= DIRENT_SIZE;
 314         inode->i_nlink--;
 315         dput(dentry);
 316         return 0;
 317 }
 318
 319 /*
 320 *       This is routine for system read from queue file.
 321 *       To avoid mess with doing here some sort of mq_receive we allow
 322 *       to read only queue size & notification info (the only values
 323 *       that are interesting from user point of view and aren't accessible
 324 *       through std routines)
 325 */
 326 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 327                                 size_t count, loff_t * off)
 328 {
 329         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 330         char buffer[FILENT_SIZE];
 331         size_t slen;
 332         loff_t o;
 333
 334         if (!count)
 335                 return 0;
 336
 337         spin_lock(&info->lock);
 338         snprintf(buffer, sizeof(buffer),
 339                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 340                         info->qsize,
 341                         info->notify_owner ? info->notify.sigev_notify : 0,
 342                         (info->notify_owner &&
 343                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 344                                 info->notify.sigev_signo : 0,
 345                         info->notify_owner);
 346         spin_unlock(&info->lock);
 347         buffer[sizeof(buffer)-1] = '\0';
 348         slen = strlen(buffer)+1;
 349
 350         o = *off;
 351         if (o > slen)
 352                 return 0;
 353
 354         if (o + count > slen)
 355                 count = slen - o;
 356
 357         if (copy_to_user(u_data, buffer + o, count))
 358                 return -EFAULT;
 359
 360         *off = o + count;
 361         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 362         return count;
 363 }
 364
 365 static int mqueue_flush_file(struct file *filp)
 366 {
 367         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 368
 369         spin_lock(&info->lock);
 370         if (current->tgid == info->notify_owner)
 371                 remove_notification(info);
 372
 373         spin_unlock(&info->lock);
 374         return 0;
 375 }
 376
 377 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 378 {
 379         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 380         int retval = 0;
 381
 382         poll_wait(filp, &info->wait_q, poll_tab);
 383
 384         spin_lock(&info->lock);
 385         if (info->attr.mq_curmsgs)
 386                 retval = POLLIN | POLLRDNORM;
 387
 388         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 389                 retval |= POLLOUT | POLLWRNORM;
 390         spin_unlock(&info->lock);
 391
 392         return retval;
 393 }
 394
 395 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 396 static void wq_add(struct mqueue_inode_info *info, int sr,
 397                         struct ext_wait_queue *ewp)
 398 {
 399         struct ext_wait_queue *walk;
 400
 401         ewp->task = current;
 402
 403         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 404                 if (walk->task->static_prio <= current->static_prio) {
 405                         list_add_tail(&ewp->list, &walk->list);
 406                         return;
 407                 }
 408         }
 409         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 410 }
 411
 412 /*
 413  * Puts current task to sleep. Caller must hold queue lock. After return
 414  * lock isn't held.
 415  * sr: SEND or RECV
 416  */
 417 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 418                         long timeout, struct ext_wait_queue *ewp)
 419 {
 420         int retval;
 421         signed long time;
 422
 423         wq_add(info, sr, ewp);
 424
 425         for (;;) {
 426                 set_current_state(TASK_INTERRUPTIBLE);
 427
 428                 spin_unlock(&info->lock);
 429                 time = schedule_timeout(timeout);
 430
 431                 while (ewp->state == STATE_PENDING)
 432                         cpu_relax();
 433
 434                 if (ewp->state == STATE_READY) {
 435                         retval = 0;
 436                         goto out;
 437                 }
 438                 spin_lock(&info->lock);
 439                 if (ewp->state == STATE_READY) {
 440                         retval = 0;
 441                         goto out_unlock;
 442                 }
 443                 if (signal_pending(current)) {
 444                         retval = -ERESTARTSYS;
 445                         break;
 446                 }
 447                 if (time == 0) {
 448                         retval = -ETIMEDOUT;
 449                         break;
 450                 }
 451         }
 452         list_del(&ewp->list);
 453 out_unlock:
 454         spin_unlock(&info->lock);
 455 out:
 456         return retval;
 457 }
 458
 459 /*
 460  * Returns waiting task that should be serviced first or NULL if none exists
 461  */
 462 static struct ext_wait_queue *wq_get_first_waiter(
 463                 struct mqueue_inode_info *info, int sr)
 464 {
 465         struct list_head *ptr;
 466
 467         ptr = info->e_wait_q[sr].list.prev;
 468         if (ptr == &info->e_wait_q[sr].list)
 469                 return NULL;
 470         return list_entry(ptr, struct ext_wait_queue, list);
 471 }
 472
 473 /* Auxiliary functions to manipulate messages' list */
 474 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 475 {
 476         int k;
 477
 478         k = info->attr.mq_curmsgs - 1;
 479         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 480                 info->messages[k + 1] = info->messages[k];
 481                 k--;
 482         }
 483         info->attr.mq_curmsgs++;
 484         info->qsize += ptr->m_ts;
 485         info->messages[k + 1] = ptr;
 486 }
 487
 488 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 489 {
 490         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 491         return info->messages[info->attr.mq_curmsgs];
 492 }
 493
 494 static inline void set_cookie(struct sk_buff *skb, char code)
 495 {
 496         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 497 }
 498
 499 /*
 500  * The next function is only to split too long sys_mq_timedsend
 501  */
 502 static void __do_notify(struct mqueue_inode_info *info)
 503 {
 504         /* notification
 505          * invoked when there is registered process and there isn't process
 506          * waiting synchronously for message AND state of queue changed from
 507          * empty to not empty. Here we are sure that no one is waiting
 508          * synchronously. */
 509         if (info->notify_owner &&
 510             info->attr.mq_curmsgs == 1) {
 511                 struct siginfo sig_i;
 512                 switch (info->notify.sigev_notify) {
 513                 case SIGEV_NONE:
 514                         break;
 515                 case SIGEV_SIGNAL:
 516                         /* sends signal */
 517
 518                         sig_i.si_signo = info->notify.sigev_signo;
 519                         sig_i.si_errno = 0;
 520                         sig_i.si_code = SI_MESGQ;
 521                         sig_i.si_value = info->notify.sigev_value;
 522                         sig_i.si_pid = current->tgid;
 523                         sig_i.si_uid = current->uid;
 524
 525                         kill_proc_info(info->notify.sigev_signo,
 526                                        &sig_i, info->notify_owner);
 527                         break;
 528                 case SIGEV_THREAD:
 529                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 530                         netlink_sendskb(info->notify_sock,
 531                                         info->notify_cookie, 0);
 532                         break;
 533                 }
 534                 /* after notification unregisters process */
 535                 info->notify_owner = 0;
 536         }
 537         wake_up(&info->wait_q);
 538 }
 539
 540 static long prepare_timeout(const struct timespec __user *u_arg)
 541 {
 542         struct timespec ts, nowts;
 543         long timeout;
 544
 545         if (u_arg) {
 546                 if (unlikely(copy_from_user(&ts, u_arg,
 547                                         sizeof(struct timespec))))
 548                         return -EFAULT;
 549
 550                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 551                         || ts.tv_nsec >= NSEC_PER_SEC))
 552                         return -EINVAL;
 553                 nowts = CURRENT_TIME;
 554                 /* first subtract as jiffies can't be too big */
 555                 ts.tv_sec -= nowts.tv_sec;
 556                 if (ts.tv_nsec < nowts.tv_nsec) {
 557                         ts.tv_nsec += NSEC_PER_SEC;
 558                         ts.tv_sec--;
 559                 }
 560                 ts.tv_nsec -= nowts.tv_nsec;
 561                 if (ts.tv_sec < 0)
 562                         return 0;
 563
 564                 timeout = timespec_to_jiffies(&ts) + 1;
 565         } else
 566                 return MAX_SCHEDULE_TIMEOUT;
 567
 568         return timeout;
 569 }
 570
 571 static void remove_notification(struct mqueue_inode_info *info)
 572 {
 573         if (info->notify_owner != 0 &&
 574             info->notify.sigev_notify == SIGEV_THREAD) {
 575                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 576                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 577         }
 578         info->notify_owner = 0;
 579 }
 580
 581 static int mq_attr_ok(struct mq_attr *attr)
 582 {
 583         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 584                 return 0;
 585         if (capable(CAP_SYS_RESOURCE)) {
 586                 if (attr->mq_maxmsg > HARD_MSGMAX)
 587                         return 0;
 588         } else {
 589                 if (attr->mq_maxmsg > msg_max ||
 590                                 attr->mq_msgsize > msgsize_max)
 591                         return 0;
 592         }
 593         /* check for overflow */
 594         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 595                 return 0;
 596         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 597             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 598             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 599                 return 0;
 600         return 1;
 601 }
 602
 603 /*
 604  * Invoked when creating a new queue via sys_mq_open
 605  */
 606 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 607                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 608 {
 609         struct file *filp;
 610         struct mq_attr attr;
 611         int ret;
 612
 613         if (u_attr != NULL) {
 614                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 615                         return ERR_PTR(-EFAULT);
 616                 if (!mq_attr_ok(&attr))
 617                         return ERR_PTR(-EINVAL);
 618                 /* store for use during create */
 619                 dentry->d_fsdata = &attr;
 620         }
 621
 622         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 623         dentry->d_fsdata = NULL;
 624         if (ret)
 625                 return ERR_PTR(ret);
 626
 627         filp = dentry_open(dentry, mqueue_mnt, oflag);
 628         if (!IS_ERR(filp))
 629                 dget(dentry);
 630
 631         return filp;
 632 }
 633
 634 /* Opens existing queue */
 635 static struct file *do_open(struct dentry *dentry, int oflag)
 636 {
 637 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 638                                         MAY_READ | MAY_WRITE };
 639         struct file *filp;
 640
 641         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 642                 return ERR_PTR(-EINVAL);
 643
 644         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
 645                 return ERR_PTR(-EACCES);
 646
 647         filp = dentry_open(dentry, mqueue_mnt, oflag);
 648
 649         if (!IS_ERR(filp))
 650                 dget(dentry);
 651
 652         return filp;
 653 }
 654
 655 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 656                                 struct mq_attr __user *u_attr)
 657 {
 658         struct dentry *dentry;
 659         struct file *filp;
 660         char *name;
 661         int fd, error;
 662
 663         if (IS_ERR(name = getname(u_name)))
 664                 return PTR_ERR(name);
 665
 666         fd = get_unused_fd();
 667         if (fd < 0)
 668                 goto out_putname;
 669
 670         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 671         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 672         if (IS_ERR(dentry)) {
 673                 error = PTR_ERR(dentry);
 674                 goto out_err;
 675         }
 676         mntget(mqueue_mnt);
 677
 678         if (oflag & O_CREAT) {
 679                 if (dentry->d_inode) {  /* entry already exists */
 680                         filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
 681                                         do_open(dentry, oflag);
 682                 } else {
 683                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 684                                                 oflag, mode, u_attr);
 685                 }
 686         } else
 687                 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
 688                                         ERR_PTR(-ENOENT);
 689
 690         dput(dentry);
 691
 692         if (IS_ERR(filp)) {
 693                 error = PTR_ERR(filp);
 694                 goto out_putfd;
 695         }
 696
 697         set_close_on_exec(fd, 1);
 698         fd_install(fd, filp);
 699         goto out_upsem;
 700
 701 out_putfd:
 702         mntput(mqueue_mnt);
 703         put_unused_fd(fd);
 704 out_err:
 705         fd = error;
 706 out_upsem:
 707         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 708 out_putname:
 709         putname(name);
 710         return fd;
 711 }
 712
 713 asmlinkage long sys_mq_unlink(const char __user *u_name)
 714 {
 715         int err;
 716         char *name;
 717         struct dentry *dentry;
 718         struct inode *inode = NULL;
 719
 720         name = getname(u_name);
 721         if (IS_ERR(name))
 722                 return PTR_ERR(name);
 723
 724         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 725         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 726         if (IS_ERR(dentry)) {
 727                 err = PTR_ERR(dentry);
 728                 goto out_unlock;
 729         }
 730
 731         if (!dentry->d_inode) {
 732                 err = -ENOENT;
 733                 goto out_err;
 734         }
 735
 736         inode = dentry->d_inode;
 737         if (inode)
 738                 atomic_inc(&inode->i_count);
 739
 740         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 741 out_err:
 742         dput(dentry);
 743
 744 out_unlock:
 745         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 746         putname(name);
 747         if (inode)
 748                 iput(inode);
 749
 750         return err;
 751 }
 752
 753 /* Pipelined send and receive functions.
 754  *
 755  * If a receiver finds no waiting message, then it registers itself in the
 756  * list of waiting receivers. A sender checks that list before adding the new
 757  * message into the message array. If there is a waiting receiver, then it
 758  * bypasses the message array and directly hands the message over to the
 759  * receiver.
 760  * The receiver accepts the message and returns without grabbing the queue
 761  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 762  * are necessary. The same algorithm is used for sysv semaphores, see
 763  * ipc/sem.c fore more details.
 764  *
 765  * The same algorithm is used for senders.
 766  */
 767
 768 /* pipelined_send() - send a message directly to the task waiting in
 769  * sys_mq_timedreceive() (without inserting message into a queue).
 770  */
 771 static inline void pipelined_send(struct mqueue_inode_info *info,
 772                                   struct msg_msg *message,
 773                                   struct ext_wait_queue *receiver)
 774 {
 775         receiver->msg = message;
 776         list_del(&receiver->list);
 777         receiver->state = STATE_PENDING;
 778         wake_up_process(receiver->task);
 779         wmb();
 780         receiver->state = STATE_READY;
 781 }
 782
 783 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 784  * gets its message and put to the queue (we have one free place for sure). */
 785 static inline void pipelined_receive(struct mqueue_inode_info *info)
 786 {
 787         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 788
 789         if (!sender) {
 790                 /* for poll */
 791                 wake_up_interruptible(&info->wait_q);
 792                 return;
 793         }
 794         msg_insert(sender->msg, info);
 795         list_del(&sender->list);
 796         sender->state = STATE_PENDING;
 797         wake_up_process(sender->task);
 798         wmb();
 799         sender->state = STATE_READY;
 800 }
 801
 802 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 803         size_t msg_len, unsigned int msg_prio,
 804         const struct timespec __user *u_abs_timeout)
 805 {
 806         struct file *filp;
 807         struct inode *inode;
 808         struct ext_wait_queue wait;
 809         struct ext_wait_queue *receiver;
 810         struct msg_msg *msg_ptr;
 811         struct mqueue_inode_info *info;
 812         long timeout;
 813         int ret;
 814
 815         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 816                 return -EINVAL;
 817
 818         timeout = prepare_timeout(u_abs_timeout);
 819
 820         ret = -EBADF;
 821         filp = fget(mqdes);
 822         if (unlikely(!filp))
 823                 goto out;
 824
 825         inode = filp->f_dentry->d_inode;
 826         if (unlikely(filp->f_op != &mqueue_file_operations))
 827                 goto out_fput;
 828         info = MQUEUE_I(inode);
 829
 830         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 831                 goto out_fput;
 832
 833         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 834                 ret = -EMSGSIZE;
 835                 goto out_fput;
 836         }
 837
 838         /* First try to allocate memory, before doing anything with
 839          * existing queues. */
 840         msg_ptr = load_msg(u_msg_ptr, msg_len);
 841         if (IS_ERR(msg_ptr)) {
 842                 ret = PTR_ERR(msg_ptr);
 843                 goto out_fput;
 844         }
 845         msg_ptr->m_ts = msg_len;
 846         msg_ptr->m_type = msg_prio;
 847
 848         spin_lock(&info->lock);
 849
 850         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 851                 if (filp->f_flags & O_NONBLOCK) {
 852                         spin_unlock(&info->lock);
 853                         ret = -EAGAIN;
 854                 } else if (unlikely(timeout < 0)) {
 855                         spin_unlock(&info->lock);
 856                         ret = timeout;
 857                 } else {
 858                         wait.task = current;
 859                         wait.msg = (void *) msg_ptr;
 860                         wait.state = STATE_NONE;
 861                         ret = wq_sleep(info, SEND, timeout, &wait);
 862                 }
 863                 if (ret < 0)
 864                         free_msg(msg_ptr);
 865         } else {
 866                 receiver = wq_get_first_waiter(info, RECV);
 867                 if (receiver) {
 868                         pipelined_send(info, msg_ptr, receiver);
 869                 } else {
 870                         /* adds message to the queue */
 871                         msg_insert(msg_ptr, info);
 872                         __do_notify(info);
 873                 }
 874                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 875                                 CURRENT_TIME;
 876                 spin_unlock(&info->lock);
 877                 ret = 0;
 878         }
 879 out_fput:
 880         fput(filp);
 881 out:
 882         return ret;
 883 }
 884
 885 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 886         size_t msg_len, unsigned int __user *u_msg_prio,
 887         const struct timespec __user *u_abs_timeout)
 888 {
 889         long timeout;
 890         ssize_t ret;
 891         struct msg_msg *msg_ptr;
 892         struct file *filp;
 893         struct inode *inode;
 894         struct mqueue_inode_info *info;
 895         struct ext_wait_queue wait;
 896
 897         timeout = prepare_timeout(u_abs_timeout);
 898
 899         ret = -EBADF;
 900         filp = fget(mqdes);
 901         if (unlikely(!filp))
 902                 goto out;
 903
 904         inode = filp->f_dentry->d_inode;
 905         if (unlikely(filp->f_op != &mqueue_file_operations))
 906                 goto out_fput;
 907         info = MQUEUE_I(inode);
 908
 909         if (unlikely(!(filp->f_mode & FMODE_READ)))
 910                 goto out_fput;
 911
 912         /* checks if buffer is big enough */
 913         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 914                 ret = -EMSGSIZE;
 915                 goto out_fput;
 916         }
 917
 918         spin_lock(&info->lock);
 919         if (info->attr.mq_curmsgs == 0) {
 920                 if (filp->f_flags & O_NONBLOCK) {
 921                         spin_unlock(&info->lock);
 922                         ret = -EAGAIN;
 923                         msg_ptr = NULL;
 924                 } else if (unlikely(timeout < 0)) {
 925                         spin_unlock(&info->lock);
 926                         ret = timeout;
 927                         msg_ptr = NULL;
 928                 } else {
 929                         wait.task = current;
 930                         wait.state = STATE_NONE;
 931                         ret = wq_sleep(info, RECV, timeout, &wait);
 932                         msg_ptr = wait.msg;
 933                 }
 934         } else {
 935                 msg_ptr = msg_get(info);
 936
 937                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 938                                 CURRENT_TIME;
 939
 940                 /* There is now free space in queue. */
 941                 pipelined_receive(info);
 942                 spin_unlock(&info->lock);
 943                 ret = 0;
 944         }
 945         if (ret == 0) {
 946                 ret = msg_ptr->m_ts;
 947
 948                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 949                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 950                         ret = -EFAULT;
 951                 }
 952                 free_msg(msg_ptr);
 953         }
 954 out_fput:
 955         fput(filp);
 956 out:
 957         return ret;
 958 }
 959
 960 /*
 961  * Notes: the case when user wants us to deregister (with NULL as pointer)
 962  * and he isn't currently owner of notification, will be silently discarded.
 963  * It isn't explicitly defined in the POSIX.
 964  */
 965 asmlinkage long sys_mq_notify(mqd_t mqdes,
 966                                 const struct sigevent __user *u_notification)
 967 {
 968         int ret;
 969         struct file *filp;
 970         struct sock *sock;
 971         struct inode *inode;
 972         struct sigevent notification;
 973         struct mqueue_inode_info *info;
 974         struct sk_buff *nc;
 975
 976         nc = NULL;
 977         sock = NULL;
 978         if (u_notification != NULL) {
 979                 if (copy_from_user(&notification, u_notification,
 980                                         sizeof(struct sigevent)))
 981                         return -EFAULT;
 982
 983                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 984                              notification.sigev_notify != SIGEV_SIGNAL &&
 985                              notification.sigev_notify != SIGEV_THREAD))
 986                         return -EINVAL;
 987                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 988                         (notification.sigev_signo < 0 ||
 989                          notification.sigev_signo > _NSIG)) {
 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);