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  64      /* max number of message queues */
  47 #define DFLT_MSGMAX     40      /* max number of messages in each queue */
  48 #define HARD_MSGMAX     (131072/sizeof(void*))
  49 #define DFLT_MSGSIZEMAX 16384   /* 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 sock *notify_sock;
  71         struct sk_buff *notify_cookie;
  72
  73         /* for tasks waiting for free space and messages, respectively */
  74         struct ext_wait_queue e_wait_q[2];
  75
  76         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  77 };
  78
  79 static struct inode_operations mqueue_dir_inode_operations;
  80 static struct file_operations mqueue_file_operations;
  81 static struct super_operations mqueue_super_ops;
  82 static void remove_notification(struct mqueue_inode_info *info);
  83
  84 static spinlock_t mq_lock;
  85 static kmem_cache_t *mqueue_inode_cachep;
  86 static struct vfsmount *mqueue_mnt;
  87
  88 static unsigned int queues_count;
  89 static unsigned int queues_max  = DFLT_QUEUESMAX;
  90 static unsigned int msg_max     = DFLT_MSGMAX;
  91 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  92
  93 static struct ctl_table_header * mq_sysctl_table;
  94
  95 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  96 {
  97         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  98 }
  99
 100 static struct inode *mqueue_get_inode(struct super_block *sb, int mode)
 101 {
 102         struct inode *inode;
 103
 104         inode = new_inode(sb);
 105         if (inode) {
 106                 inode->i_mode = mode;
 107                 inode->i_uid = current->fsuid;
 108                 inode->i_gid = current->fsgid;
 109                 inode->i_blksize = PAGE_CACHE_SIZE;
 110                 inode->i_blocks = 0;
 111                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 112                                 CURRENT_TIME;
 113
 114                 if (S_ISREG(mode)) {
 115                         struct mqueue_inode_info *info;
 116
 117                         inode->i_fop = &mqueue_file_operations;
 118                         inode->i_size = FILENT_SIZE;
 119                         /* mqueue specific info */
 120                         info = MQUEUE_I(inode);
 121                         spin_lock_init(&info->lock);
 122                         init_waitqueue_head(&info->wait_q);
 123                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 124                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 125                         info->notify_owner = 0;
 126                         info->qsize = 0;
 127                         memset(&info->attr, 0, sizeof(info->attr));
 128                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 129                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 130                         info->messages = kmalloc(DFLT_MSGMAX * sizeof(struct msg_msg *), GFP_KERNEL);
 131                         if (!info->messages) {
 132                                 make_bad_inode(inode);
 133                                 iput(inode);
 134                                 inode = NULL;
 135                         }
 136                 } else if (S_ISDIR(mode)) {
 137                         inode->i_nlink++;
 138                         /* Some things misbehave if size == 0 on a directory */
 139                         inode->i_size = 2 * DIRENT_SIZE;
 140                         inode->i_op = &mqueue_dir_inode_operations;
 141                         inode->i_fop = &simple_dir_operations;
 142                 }
 143         }
 144         return inode;
 145 }
 146
 147 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 148 {
 149         struct inode *inode;
 150
 151         sb->s_blocksize = PAGE_CACHE_SIZE;
 152         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 153         sb->s_magic = MQUEUE_MAGIC;
 154         sb->s_op = &mqueue_super_ops;
 155
 156         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO);
 157         if (!inode)
 158                 return -ENOMEM;
 159
 160         sb->s_root = d_alloc_root(inode);
 161         if (!sb->s_root) {
 162                 iput(inode);
 163                 return -ENOMEM;
 164         }
 165
 166         return 0;
 167 }
 168
 169 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 170                                          int flags, const char *dev_name,
 171                                          void *data)
 172 {
 173         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 174 }
 175
 176 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 177 {
 178         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 179
 180         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 181                 SLAB_CTOR_CONSTRUCTOR)
 182                 inode_init_once(&p->vfs_inode);
 183 }
 184
 185 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 186 {
 187         struct mqueue_inode_info *ei;
 188
 189         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 190         if (!ei)
 191                 return NULL;
 192         return &ei->vfs_inode;
 193 }
 194
 195 static void mqueue_destroy_inode(struct inode *inode)
 196 {
 197         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 198 }
 199
 200 static void mqueue_delete_inode(struct inode *inode)
 201 {
 202         struct mqueue_inode_info *info;
 203         int i;
 204
 205         if (S_ISDIR(inode->i_mode)) {
 206                 clear_inode(inode);
 207                 return;
 208         }
 209         info = MQUEUE_I(inode);
 210         spin_lock(&info->lock);
 211         for (i = 0; i < info->attr.mq_curmsgs; i++)
 212                 free_msg(info->messages[i]);
 213         kfree(info->messages);
 214         spin_unlock(&info->lock);
 215
 216         clear_inode(inode);
 217
 218         if (info->messages) {
 219                 spin_lock(&mq_lock);
 220                 queues_count--;
 221                 spin_unlock(&mq_lock);
 222         }
 223 }
 224
 225 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 226                                 int mode, struct nameidata *nd)
 227 {
 228         struct inode *inode;
 229         int error;
 230
 231         spin_lock(&mq_lock);
 232         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 233                 error = -ENOSPC;
 234                 goto out_lock;
 235         }
 236         queues_count++;
 237         spin_unlock(&mq_lock);
 238
 239         inode = mqueue_get_inode(dir->i_sb, mode);
 240         if (!inode) {
 241                 error = -ENOMEM;
 242                 spin_lock(&mq_lock);
 243                 queues_count--;
 244                 goto out_lock;
 245         }
 246
 247         dir->i_size += DIRENT_SIZE;
 248         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 249
 250         d_instantiate(dentry, inode);
 251         dget(dentry);
 252         return 0;
 253 out_lock:
 254         spin_unlock(&mq_lock);
 255         return error;
 256 }
 257
 258 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 259 {
 260         struct inode *inode = dentry->d_inode;
 261
 262         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 263         dir->i_size -= DIRENT_SIZE;
 264         inode->i_nlink--;
 265         dput(dentry);
 266         return 0;
 267 }
 268
 269 /*
 270 *       This is routine for system read from queue file.
 271 *       To avoid mess with doing here some sort of mq_receive we allow
 272 *       to read only queue size & notification info (the only values
 273 *       that are interesting from user point of view and aren't accessible
 274 *       through std routines)
 275 */
 276 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 277                                 size_t count, loff_t * off)
 278 {
 279         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 280         char buffer[FILENT_SIZE];
 281         size_t slen;
 282         loff_t o;
 283
 284         if (!count)
 285                 return 0;
 286
 287         spin_lock(&info->lock);
 288         snprintf(buffer, sizeof(buffer),
 289                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 290                         info->qsize,
 291                         info->notify_owner ? info->notify.sigev_notify : 0,
 292                         (info->notify_owner &&
 293                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 294                                 info->notify.sigev_signo : 0,
 295                         info->notify_owner);
 296         spin_unlock(&info->lock);
 297         buffer[sizeof(buffer)-1] = '\0';
 298         slen = strlen(buffer)+1;
 299
 300         o = *off;
 301         if (o > slen)
 302                 return 0;
 303
 304         if (o + count > slen)
 305                 count = slen - o;
 306
 307         if (copy_to_user(u_data, buffer + o, count))
 308                 return -EFAULT;
 309
 310         *off = o + count;
 311         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 312         return count;
 313 }
 314
 315 static int mqueue_flush_file(struct file *filp)
 316 {
 317         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 318
 319         spin_lock(&info->lock);
 320         if (current->tgid == info->notify_owner)
 321                 remove_notification(info);
 322
 323         spin_unlock(&info->lock);
 324         return 0;
 325 }
 326
 327 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 328 {
 329         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 330         int retval = 0;
 331
 332         poll_wait(filp, &info->wait_q, poll_tab);
 333
 334         spin_lock(&info->lock);
 335         if (info->attr.mq_curmsgs)
 336                 retval = POLLIN | POLLRDNORM;
 337
 338         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 339                 retval |= POLLOUT | POLLWRNORM;
 340         spin_unlock(&info->lock);
 341
 342         return retval;
 343 }
 344
 345 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 346 static void wq_add(struct mqueue_inode_info *info, int sr,
 347                         struct ext_wait_queue *ewp)
 348 {
 349         struct ext_wait_queue *walk;
 350
 351         ewp->task = current;
 352
 353         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 354                 if (walk->task->static_prio <= current->static_prio) {
 355                         list_add_tail(&ewp->list, &walk->list);
 356                         return;
 357                 }
 358         }
 359         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 360 }
 361
 362 /*
 363  * Puts current task to sleep. Caller must hold queue lock. After return
 364  * lock isn't held.
 365  * sr: SEND or RECV
 366  */
 367 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 368                         long timeout, struct ext_wait_queue *ewp)
 369 {
 370         int retval;
 371         signed long time;
 372
 373         wq_add(info, sr, ewp);
 374
 375         for (;;) {
 376                 set_current_state(TASK_INTERRUPTIBLE);
 377
 378                 spin_unlock(&info->lock);
 379                 time = schedule_timeout(timeout);
 380
 381                 while (ewp->state == STATE_PENDING)
 382                         cpu_relax();
 383
 384                 if (ewp->state == STATE_READY) {
 385                         retval = 0;
 386                         goto out;
 387                 }
 388                 spin_lock(&info->lock);
 389                 if (ewp->state == STATE_READY) {
 390                         retval = 0;
 391                         goto out_unlock;
 392                 }
 393                 if (signal_pending(current)) {
 394                         retval = -ERESTARTSYS;
 395                         break;
 396                 }
 397                 if (time == 0) {
 398                         retval = -ETIMEDOUT;
 399                         break;
 400                 }
 401         }
 402         list_del(&ewp->list);
 403 out_unlock:
 404         spin_unlock(&info->lock);
 405 out:
 406         return retval;
 407 }
 408
 409 /*
 410  * Returns waiting task that should be serviced first or NULL if none exists
 411  */
 412 static struct ext_wait_queue *wq_get_first_waiter(
 413                 struct mqueue_inode_info *info, int sr)
 414 {
 415         struct list_head *ptr;
 416
 417         ptr = info->e_wait_q[sr].list.prev;
 418         if (ptr == &info->e_wait_q[sr].list)
 419                 return NULL;
 420         return list_entry(ptr, struct ext_wait_queue, list);
 421 }
 422
 423 /* Auxiliary functions to manipulate messages' list */
 424 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 425 {
 426         int k;
 427
 428         k = info->attr.mq_curmsgs - 1;
 429         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 430                 info->messages[k + 1] = info->messages[k];
 431                 k--;
 432         }
 433         info->attr.mq_curmsgs++;
 434         info->qsize += ptr->m_ts;
 435         info->messages[k + 1] = ptr;
 436 }
 437
 438 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 439 {
 440         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 441         return info->messages[info->attr.mq_curmsgs];
 442 }
 443
 444 static inline void set_cookie(struct sk_buff *skb, char code)
 445 {
 446         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 447 }
 448
 449 /*
 450  * The next function is only to split too long sys_mq_timedsend
 451  */
 452 static void __do_notify(struct mqueue_inode_info *info)
 453 {
 454         /* notification
 455          * invoked when there is registered process and there isn't process
 456          * waiting synchronously for message AND state of queue changed from
 457          * empty to not empty. Here we are sure that no one is waiting
 458          * synchronously. */
 459         if (info->notify_owner &&
 460             info->attr.mq_curmsgs == 1) {
 461                 struct siginfo sig_i;
 462                 switch (info->notify.sigev_notify) {
 463                 case SIGEV_NONE:
 464                         break;
 465                 case SIGEV_SIGNAL:
 466                         /* sends signal */
 467
 468                         sig_i.si_signo = info->notify.sigev_signo;
 469                         sig_i.si_errno = 0;
 470                         sig_i.si_code = SI_MESGQ;
 471                         sig_i.si_value = info->notify.sigev_value;
 472                         sig_i.si_pid = current->tgid;
 473                         sig_i.si_uid = current->uid;
 474
 475                         kill_proc_info(info->notify.sigev_signo,
 476                                        &sig_i, info->notify_owner);
 477                         break;
 478                 case SIGEV_THREAD:
 479                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 480                         netlink_sendskb(info->notify_sock,
 481                                         info->notify_cookie, 0);
 482                         break;
 483                 }
 484                 /* after notification unregisters process */
 485                 info->notify_owner = 0;
 486         }
 487         wake_up(&info->wait_q);
 488 }
 489
 490 static long prepare_timeout(const struct timespec __user *u_arg)
 491 {
 492         struct timespec ts, nowts;
 493         long timeout;
 494
 495         if (u_arg) {
 496                 if (unlikely(copy_from_user(&ts, u_arg,
 497                                         sizeof(struct timespec))))
 498                         return -EFAULT;
 499
 500                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 501                         || ts.tv_nsec >= NSEC_PER_SEC))
 502                         return -EINVAL;
 503                 nowts = CURRENT_TIME;
 504                 /* first subtract as jiffies can't be too big */
 505                 ts.tv_sec -= nowts.tv_sec;
 506                 if (ts.tv_nsec < nowts.tv_nsec) {
 507                         ts.tv_nsec += NSEC_PER_SEC;
 508                         ts.tv_sec--;
 509                 }
 510                 ts.tv_nsec -= nowts.tv_nsec;
 511                 if (ts.tv_sec < 0)
 512                         return 0;
 513
 514                 timeout = timespec_to_jiffies(&ts) + 1;
 515         } else
 516                 return MAX_SCHEDULE_TIMEOUT;
 517
 518         return timeout;
 519 }
 520
 521 static void remove_notification(struct mqueue_inode_info *info)
 522 {
 523         if (info->notify_owner != 0 &&
 524             info->notify.sigev_notify == SIGEV_THREAD) {
 525                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 526                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 527         }
 528         info->notify_owner = 0;
 529 }
 530
 531 /*
 532  * Invoked when creating a new queue via sys_mq_open
 533  */
 534 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 535                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 536 {
 537         struct file *filp;
 538         struct inode *inode;
 539         struct mqueue_inode_info *info;
 540         struct msg_msg **msgs = NULL;
 541         struct mq_attr attr;
 542         int ret;
 543
 544         if (u_attr != NULL) {
 545                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 546                         return ERR_PTR(-EFAULT);
 547
 548                 if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0)
 549                         return ERR_PTR(-EINVAL);
 550                 if (capable(CAP_SYS_RESOURCE)) {
 551                         if (attr.mq_maxmsg > HARD_MSGMAX)
 552                                 return ERR_PTR(-EINVAL);
 553                 } else {
 554                         if (attr.mq_maxmsg > msg_max ||
 555                                         attr.mq_msgsize > msgsize_max)
 556                                 return ERR_PTR(-EINVAL);
 557                 }
 558                 msgs = kmalloc(attr.mq_maxmsg * sizeof(*msgs), GFP_KERNEL);
 559                 if (!msgs)
 560                         return ERR_PTR(-ENOMEM);
 561         } else {
 562                 msgs = NULL;
 563         }
 564
 565         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 566         if (ret) {
 567                 kfree(msgs);
 568                 return ERR_PTR(ret);
 569         }
 570
 571         inode = dentry->d_inode;
 572         info = MQUEUE_I(inode);
 573
 574         if (msgs) {
 575                 info->attr.mq_maxmsg = attr.mq_maxmsg;
 576                 info->attr.mq_msgsize = attr.mq_msgsize;
 577                 kfree(info->messages);
 578                 info->messages = msgs;
 579         }
 580
 581         filp = dentry_open(dentry, mqueue_mnt, oflag);
 582         if (!IS_ERR(filp))
 583                 dget(dentry);
 584
 585         return filp;
 586 }
 587
 588 /* Opens existing queue */
 589 static struct file *do_open(struct dentry *dentry, int oflag)
 590 {
 591 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 592                                         MAY_READ | MAY_WRITE };
 593         struct file *filp;
 594
 595         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 596                 return ERR_PTR(-EINVAL);
 597
 598         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL))
 599                 return ERR_PTR(-EACCES);
 600
 601         filp = dentry_open(dentry, mqueue_mnt, oflag);
 602
 603         if (!IS_ERR(filp))
 604                 dget(dentry);
 605
 606         return filp;
 607 }
 608
 609 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 610                                 struct mq_attr __user *u_attr)
 611 {
 612         struct dentry *dentry;
 613         struct file *filp;
 614         char *name;
 615         int fd, error;
 616
 617         if (IS_ERR(name = getname(u_name)))
 618                 return PTR_ERR(name);
 619
 620         fd = get_unused_fd();
 621         if (fd < 0)
 622                 goto out_putname;
 623
 624         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 625         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 626         if (IS_ERR(dentry)) {
 627                 error = PTR_ERR(dentry);
 628                 goto out_err;
 629         }
 630         mntget(mqueue_mnt);
 631
 632         if (oflag & O_CREAT) {
 633                 if (dentry->d_inode) {  /* entry already exists */
 634                         filp = (oflag & O_EXCL) ? ERR_PTR(-EEXIST) :
 635                                         do_open(dentry, oflag);
 636                 } else {
 637                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 638                                                 oflag, mode, u_attr);
 639                 }
 640         } else
 641                 filp = (dentry->d_inode) ? do_open(dentry, oflag) :
 642                                         ERR_PTR(-ENOENT);
 643
 644         dput(dentry);
 645
 646         if (IS_ERR(filp)) {
 647                 error = PTR_ERR(filp);
 648                 goto out_putfd;
 649         }
 650
 651         set_close_on_exec(fd, 1);
 652         fd_install(fd, filp);
 653         goto out_upsem;
 654
 655 out_putfd:
 656         mntput(mqueue_mnt);
 657         put_unused_fd(fd);
 658 out_err:
 659         fd = error;
 660 out_upsem:
 661         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 662 out_putname:
 663         putname(name);
 664         return fd;
 665 }
 666
 667 asmlinkage long sys_mq_unlink(const char __user *u_name)
 668 {
 669         int err;
 670         char *name;
 671         struct dentry *dentry;
 672         struct inode *inode = NULL;
 673
 674         name = getname(u_name);
 675         if (IS_ERR(name))
 676                 return PTR_ERR(name);
 677
 678         down(&mqueue_mnt->mnt_root->d_inode->i_sem);
 679         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 680         if (IS_ERR(dentry)) {
 681                 err = PTR_ERR(dentry);
 682                 goto out_unlock;
 683         }
 684
 685         if (!dentry->d_inode) {
 686                 err = -ENOENT;
 687                 goto out_err;
 688         }
 689
 690         inode = dentry->d_inode;
 691         if (inode)
 692                 atomic_inc(&inode->i_count);
 693
 694         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 695 out_err:
 696         dput(dentry);
 697
 698 out_unlock:
 699         up(&mqueue_mnt->mnt_root->d_inode->i_sem);
 700         putname(name);
 701         if (inode)
 702                 iput(inode);
 703
 704         return err;
 705 }
 706
 707 /* Pipelined send and receive functions.
 708  *
 709  * If a receiver finds no waiting message, then it registers itself in the
 710  * list of waiting receivers. A sender checks that list before adding the new
 711  * message into the message array. If there is a waiting receiver, then it
 712  * bypasses the message array and directly hands the message over to the
 713  * receiver.
 714  * The receiver accepts the message and returns without grabbing the queue
 715  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 716  * are necessary. The same algorithm is used for sysv semaphores, see
 717  * ipc/sem.c fore more details.
 718  *
 719  * The same algorithm is used for senders.
 720  */
 721
 722 /* pipelined_send() - send a message directly to the task waiting in
 723  * sys_mq_timedreceive() (without inserting message into a queue).
 724  */
 725 static inline void pipelined_send(struct mqueue_inode_info *info,
 726                                   struct msg_msg *message,
 727                                   struct ext_wait_queue *receiver)
 728 {
 729         receiver->msg = message;
 730         list_del(&receiver->list);
 731         receiver->state = STATE_PENDING;
 732         wake_up_process(receiver->task);
 733         wmb();
 734         receiver->state = STATE_READY;
 735 }
 736
 737 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 738  * gets its message and put to the queue (we have one free place for sure). */
 739 static inline void pipelined_receive(struct mqueue_inode_info *info)
 740 {
 741         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 742
 743         if (!sender) {
 744                 /* for poll */
 745                 wake_up_interruptible(&info->wait_q);
 746                 return;
 747         }
 748         msg_insert(sender->msg, info);
 749         list_del(&sender->list);
 750         sender->state = STATE_PENDING;
 751         wake_up_process(sender->task);
 752         wmb();
 753         sender->state = STATE_READY;
 754 }
 755
 756 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 757         size_t msg_len, unsigned int msg_prio,
 758         const struct timespec __user *u_abs_timeout)
 759 {
 760         struct file *filp;
 761         struct inode *inode;
 762         struct ext_wait_queue wait;
 763         struct ext_wait_queue *receiver;
 764         struct msg_msg *msg_ptr;
 765         struct mqueue_inode_info *info;
 766         long timeout;
 767         int ret;
 768
 769         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 770                 return -EINVAL;
 771
 772         timeout = prepare_timeout(u_abs_timeout);
 773
 774         ret = -EBADF;
 775         filp = fget(mqdes);
 776         if (unlikely(!filp))
 777                 goto out;
 778
 779         inode = filp->f_dentry->d_inode;
 780         if (unlikely(filp->f_op != &mqueue_file_operations))
 781                 goto out_fput;
 782         info = MQUEUE_I(inode);
 783
 784         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 785                 goto out_fput;
 786
 787         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 788                 ret = -EMSGSIZE;
 789                 goto out_fput;
 790         }
 791
 792         /* First try to allocate memory, before doing anything with
 793          * existing queues. */
 794         msg_ptr = load_msg((void *)u_msg_ptr, msg_len);
 795         if (unlikely(IS_ERR(msg_ptr))) {
 796                 ret = PTR_ERR(msg_ptr);
 797                 goto out_fput;
 798         }
 799         msg_ptr->m_ts = msg_len;
 800         msg_ptr->m_type = msg_prio;
 801
 802         spin_lock(&info->lock);
 803
 804         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 805                 if (filp->f_flags & O_NONBLOCK) {
 806                         spin_unlock(&info->lock);
 807                         ret = -EAGAIN;
 808                 } else if (unlikely(timeout < 0)) {
 809                         spin_unlock(&info->lock);
 810                         ret = timeout;
 811                 } else {
 812                         wait.task = current;
 813                         wait.msg = (void *) msg_ptr;
 814                         wait.state = STATE_NONE;
 815                         ret = wq_sleep(info, SEND, timeout, &wait);
 816                 }
 817                 if (ret < 0)
 818                         free_msg(msg_ptr);
 819         } else {
 820                 receiver = wq_get_first_waiter(info, RECV);
 821                 if (receiver) {
 822                         pipelined_send(info, msg_ptr, receiver);
 823                 } else {
 824                         /* adds message to the queue */
 825                         msg_insert(msg_ptr, info);
 826                         __do_notify(info);
 827                 }
 828                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 829                                 CURRENT_TIME;
 830                 spin_unlock(&info->lock);
 831                 ret = 0;
 832         }
 833 out_fput:
 834         fput(filp);
 835 out:
 836         return ret;
 837 }
 838
 839 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 840         size_t msg_len, unsigned int __user *u_msg_prio,
 841         const struct timespec __user *u_abs_timeout)
 842 {
 843         long timeout;
 844         ssize_t ret;
 845         struct msg_msg *msg_ptr;
 846         struct file *filp;
 847         struct inode *inode;
 848         struct mqueue_inode_info *info;
 849         struct ext_wait_queue wait;
 850
 851         timeout = prepare_timeout(u_abs_timeout);
 852
 853         ret = -EBADF;
 854         filp = fget(mqdes);
 855         if (unlikely(!filp))
 856                 goto out;
 857
 858         inode = filp->f_dentry->d_inode;
 859         if (unlikely(filp->f_op != &mqueue_file_operations))
 860                 goto out_fput;
 861         info = MQUEUE_I(inode);
 862
 863         if (unlikely(!(filp->f_mode & FMODE_READ)))
 864                 goto out_fput;
 865
 866         /* checks if buffer is big enough */
 867         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 868                 ret = -EMSGSIZE;
 869                 goto out_fput;
 870         }
 871
 872         spin_lock(&info->lock);
 873         if (info->attr.mq_curmsgs == 0) {
 874                 if (filp->f_flags & O_NONBLOCK) {
 875                         spin_unlock(&info->lock);
 876                         ret = -EAGAIN;
 877                         msg_ptr = NULL;
 878                 } else if (unlikely(timeout < 0)) {
 879                         spin_unlock(&info->lock);
 880                         ret = timeout;
 881                         msg_ptr = NULL;
 882                 } else {
 883                         wait.task = current;
 884                         wait.state = STATE_NONE;
 885                         ret = wq_sleep(info, RECV, timeout, &wait);
 886                         msg_ptr = wait.msg;
 887                 }
 888         } else {
 889                 msg_ptr = msg_get(info);
 890
 891                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 892                                 CURRENT_TIME;
 893
 894                 /* There is now free space in queue. */
 895                 pipelined_receive(info);
 896                 spin_unlock(&info->lock);
 897                 ret = 0;
 898         }
 899         if (ret == 0) {
 900                 ret = msg_ptr->m_ts;
 901
 902                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 903                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 904                         ret = -EFAULT;
 905                 }
 906                 free_msg(msg_ptr);
 907         }
 908 out_fput:
 909         fput(filp);
 910 out:
 911         return ret;
 912 }
 913
 914 /*
 915  * Notes: the case when user wants us to deregister (with NULL as pointer)
 916  * and he isn't currently owner of notification, will be silently discarded.
 917  * It isn't explicitly defined in the POSIX.
 918  */
 919 asmlinkage long sys_mq_notify(mqd_t mqdes,
 920                                 const struct sigevent __user *u_notification)
 921 {
 922         int ret;
 923         struct file *filp;
 924         struct sock *sock;
 925         struct inode *inode;
 926         struct sigevent notification;
 927         struct mqueue_inode_info *info;
 928         struct sk_buff *nc;
 929
 930         nc = NULL;
 931         sock = NULL;
 932         if (u_notification != NULL) {
 933                 if (copy_from_user(&notification, u_notification,
 934                                         sizeof(struct sigevent)))
 935                         return -EFAULT;
 936
 937                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 938                              notification.sigev_notify != SIGEV_SIGNAL &&
 939                              notification.sigev_notify != SIGEV_THREAD))
 940                         return -EINVAL;
 941                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 942                         (notification.sigev_signo < 0 ||
 943                          notification.sigev_signo > _NSIG)) {
 944                         return -EINVAL;
 945                 }
 946                 if (notification.sigev_notify == SIGEV_THREAD) {
 947                         /* create the notify skb */
 948                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
 949                         ret = -ENOMEM;
 950                         if (!nc)
 951                                 goto out;
 952                         ret = -EFAULT;
 953                         if (copy_from_user(nc->data,
 954                                         notification.sigev_value.sival_ptr,
 955                                         NOTIFY_COOKIE_LEN)) {
 956                                 goto out;
 957                         }
 958
 959                         /* TODO: add a header? */
 960                         skb_put(nc, NOTIFY_COOKIE_LEN);
 961                         /* and attach it to the socket */
 962 retry:
 963                         filp = fget(notification.sigev_signo);
 964                         ret = -EBADF;
 965                         if (!filp)
 966                                 goto out;
 967                         sock = netlink_getsockbyfilp(filp);
 968                         fput(filp);
 969                         if (IS_ERR(sock)) {
 970                                 ret = PTR_ERR(sock);
 971                                 sock = NULL;
 972                                 goto out;
 973                         }
 974
 975                         ret = netlink_attachskb(sock, nc, 0, MAX_SCHEDULE_TIMEOUT);
 976                         if (ret == 1)
 977                                 goto retry;
 978                         if (ret) {
 979                                 sock = NULL;
 980                                 nc = NULL;
 981                                 goto out;
 982                         }
 983                 }
 984         }
 985
 986         ret = -EBADF;
 987         filp = fget(mqdes);
 988         if (!filp)
 989                 goto out;
 990
 991         inode = filp->f_dentry->d_inode;
 992         if (unlikely(filp->f_op != &mqueue_file_operations))
 993                 goto out_fput;
 994         info = MQUEUE_I(inode);
 995
 996         ret = 0;
 997         spin_lock(&info->lock);
 998         if (u_notification == NULL) {
 999                 if (info->notify_owner == current->tgid) {
1000                         remove_notification(info);
1001                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1002                 }
1003         } else if (info->notify_owner != 0) {
1004                 ret = -EBUSY;
1005         } else {
1006                 switch (notification.sigev_notify) {
1007                 case SIGEV_NONE:
1008                         info->notify.sigev_notify = SIGEV_NONE;
1009                         break;
1010                 case SIGEV_THREAD:
1011                         info->notify_sock = sock;
1012                         info->notify_cookie = nc;
1013                         sock = NULL;
1014                         nc = NULL;
1015                         info->notify.sigev_notify = SIGEV_THREAD;
1016                         break;
1017                 case SIGEV_SIGNAL:
1018                         info->notify.sigev_signo = notification.sigev_signo;
1019                         info->notify.sigev_value = notification.sigev_value;
1020                         info->notify.sigev_notify = SIGEV_SIGNAL;
1021                         break;
1022                 }
1023                 info->notify_owner = current->tgid;
1024                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1025         }
1026         spin_unlock(&info->lock);
1027 out_fput:
1028         fput(filp);
1029 out:
1030         if (sock) {
1031                 netlink_detachskb(sock, nc);
1032         } else if (nc) {
1033                 dev_kfree_skb(nc);
1034         }
1035         return ret;
1036 }
1037
1038 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1039                         const struct mq_attr __user *u_mqstat,
1040                         struct mq_attr __user *u_omqstat)
1041 {
1042         int ret;
1043         struct mq_attr mqstat, omqstat;
1044         struct file *filp;
1045         struct inode *inode;
1046         struct mqueue_inode_info *info;
1047
1048         if (u_mqstat != NULL) {
1049                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1050                         return -EFAULT;
1051                 if (mqstat.mq_flags & (~O_NONBLOCK))
1052                         return -EINVAL;
1053         }
1054
1055         ret = -EBADF;
1056         filp = fget(mqdes);
1057         if (!filp)
1058                 goto out;
1059
1060         inode = filp->f_dentry->d_inode;
1061         if (unlikely(filp->f_op != &mqueue_file_operations))
1062                 goto out_fput;
1063         info = MQUEUE_I(inode);
1064
1065         spin_lock(&info->lock);
1066
1067         omqstat = info->attr;
1068         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1069         if (u_mqstat) {
1070                 if (mqstat.mq_flags & O_NONBLOCK)
1071                         filp->f_flags |= O_NONBLOCK;
1072                 else
1073                         filp->f_flags &= ~O_NONBLOCK;
1074
1075                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1076         }
1077
1078         spin_unlock(&info->lock);
1079
1080         ret = 0;
1081         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1082                                                 sizeof(struct mq_attr)))
1083                 ret = -EFAULT;
1084
1085 out_fput:
1086         fput(filp);
1087 out:
1088         return ret;
1089 }
1090
1091 static struct inode_operations mqueue_dir_inode_operations = {
1092         .lookup = simple_lookup,
1093         .create = mqueue_create,
1094         .unlink = mqueue_unlink,
1095 };
1096
1097 static struct file_operations mqueue_file_operations = {
1098         .flush = mqueue_flush_file,
1099         .poll = mqueue_poll_file,
1100         .read = mqueue_read_file,
1101 };
1102
1103 static struct super_operations mqueue_super_ops = {
1104         .alloc_inode = mqueue_alloc_inode,
1105         .destroy_inode = mqueue_destroy_inode,
1106         .statfs = simple_statfs,
1107         .delete_inode = mqueue_delete_inode,
1108         .drop_inode = generic_delete_inode,
1109 };
1110
1111 static struct file_system_type mqueue_fs_type = {
1112         .name = "mqueue",
1113         .get_sb = mqueue_get_sb,
1114         .kill_sb = kill_litter_super,
1115 };
1116
1117 static int msg_max_limit_min = DFLT_MSGMAX;
1118 static int msg_max_limit_max = HARD_MSGMAX;
1119
1120 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1121 static int msg_maxsize_limit_max = INT_MAX;
1122
1123 static ctl_table mq_sysctls[] = {
1124         {
1125                 .ctl_name       = CTL_QUEUESMAX,
1126                 .procname       = "queues_max",
1127                 .data           = &queues_max,
1128                 .maxlen         = sizeof(int),
1129                 .mode           = 0644,
1130                 .proc_handler   = &proc_dointvec,
1131         },
1132         {
1133                 .ctl_name       = CTL_MSGMAX,
1134                 .procname       = "msg_max",
1135                 .data           = &msg_max,
1136                 .maxlen         = sizeof(int),
1137                 .mode           = 0644,
1138                 .proc_handler   = &proc_dointvec_minmax,
1139                 .extra1         = &msg_max_limit_min,
1140                 .extra2         = &msg_max_limit_max,
1141         },
1142         {
1143                 .ctl_name       = CTL_MSGSIZEMAX,
1144                 .procname       = "msgsize_max",
1145                 .data           = &msgsize_max,
1146                 .maxlen         = sizeof(int),
1147                 .mode           = 0644,
1148                 .proc_handler   = &proc_dointvec_minmax,
1149                 .extra1         = &msg_maxsize_limit_min,
1150                 .extra2         = &msg_maxsize_limit_max,
1151         },
1152         { .ctl_name = 0 }
1153 };
1154
1155 static ctl_table mq_sysctl_dir[] = {
1156         {
1157                 .ctl_name       = FS_MQUEUE,
1158                 .procname       = "mqueue",
1159                 .mode           = 0555,
1160                 .child          = mq_sysctls,
1161         },
1162         { .ctl_name = 0 }
1163 };
1164
1165 static ctl_table mq_sysctl_root[] = {
1166         {
1167                 .ctl_name       = CTL_FS,
1168                 .procname       = "fs",
1169                 .mode           = 0555,
1170                 .child          = mq_sysctl_dir,
1171         },
1172         { .ctl_name = 0 }
1173 };
1174
1175 static int __init init_mqueue_fs(void)
1176 {
1177         int error;
1178
1179         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1180                                 sizeof(struct mqueue_inode_info), 0,
1181                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1182         if (mqueue_inode_cachep == NULL)
1183                 return -ENOMEM;
1184
1185         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1186         if (!mq_sysctl_table) {
1187                 error = -ENOMEM;
1188                 goto out_cache;
1189         }
1190
1191         error = register_filesystem(&mqueue_fs_type);
1192         if (error)
1193                 goto out_sysctl;
1194
1195         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1196                 error = PTR_ERR(mqueue_mnt);
1197                 goto out_filesystem;
1198         }
1199
1200         /* internal initialization - not common for vfs */
1201         queues_count = 0;
1202         spin_lock_init(&mq_lock);
1203
1204         return 0;
1205
1206 out_filesystem:
1207         unregister_filesystem(&mqueue_fs_type);
1208 out_sysctl:
1209         unregister_sysctl_table(mq_sysctl_table);
1210 out_cache:
1211         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1212                 printk(KERN_INFO
1213                         "mqueue_inode_cache: not all structures were freed\n");
1214         }
1215         return error;
1216 }
1217
1218 __initcall(init_mqueue_fs);