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