ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (michal.wronski@gmail.com)
   6  *
   7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          (manfred@colorfullife.com)
  10  *
  11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/pid.h>
  34
  35 #include <net/sock.h>
  36 #include "util.h"
  37
  38 #define MQUEUE_MAGIC    0x19800202
  39 #define DIRENT_SIZE     20
  40 #define FILENT_SIZE     80
  41
  42 #define SEND            0
  43 #define RECV            1
  44
  45 #define STATE_NONE      0
  46 #define STATE_PENDING   1
  47 #define STATE_READY     2
  48
  49 /* default values */
  50 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  51 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  52 #define HARD_MSGMAX     (131072/sizeof(void*))
  53 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  54
  55 /*
  56  * Define the ranges various user-specified maximum values can
  57  * be set to.
  58  */
  59 #define MIN_MSGMAX      1               /* min value for msg_max */
  60 #define MAX_MSGMAX      HARD_MSGMAX     /* max value for msg_max */
  61 #define MIN_MSGSIZEMAX  128             /* min value for msgsize_max */
  62 #define MAX_MSGSIZEMAX  (8192*128)      /* max value for msgsize_max */
  63
  64 struct ext_wait_queue {         /* queue of sleeping tasks */
  65         struct task_struct *task;
  66         struct list_head list;
  67         struct msg_msg *msg;    /* ptr of loaded message */
  68         int state;              /* one of STATE_* values */
  69 };
  70
  71 struct mqueue_inode_info {
  72         spinlock_t lock;
  73         struct inode vfs_inode;
  74         wait_queue_head_t wait_q;
  75
  76         struct msg_msg **messages;
  77         struct mq_attr attr;
  78
  79         struct sigevent notify;
  80         struct pid* notify_owner;
  81         struct user_struct *user;       /* user who created, for accounting */
  82         struct sock *notify_sock;
  83         struct sk_buff *notify_cookie;
  84
  85         /* for tasks waiting for free space and messages, respectively */
  86         struct ext_wait_queue e_wait_q[2];
  87
  88         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  89 };
  90
  91 static const struct inode_operations mqueue_dir_inode_operations;
  92 static const struct file_operations mqueue_file_operations;
  93 static struct super_operations mqueue_super_ops;
  94 static void remove_notification(struct mqueue_inode_info *info);
  95
  96 static spinlock_t mq_lock;
  97 static struct kmem_cache *mqueue_inode_cachep;
  98 static struct vfsmount *mqueue_mnt;
  99
 100 static unsigned int queues_count;
 101 static unsigned int queues_max  = DFLT_QUEUESMAX;
 102 static unsigned int msg_max     = DFLT_MSGMAX;
 103 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
 104
 105 static struct ctl_table_header * mq_sysctl_table;
 106
 107 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 108 {
 109         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 110 }
 111
 112 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 113                                                         struct mq_attr *attr)
 114 {
 115         struct user_struct *u = current_user();
 116         struct inode *inode;
 117
 118         inode = new_inode(sb);
 119         if (inode) {
 120                 inode->i_mode = mode;
 121                 inode->i_uid = current_fsuid();
 122                 inode->i_gid = current_fsgid();
 123                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 124                                 CURRENT_TIME;
 125
 126                 if (S_ISREG(mode)) {
 127                         struct mqueue_inode_info *info;
 128                         struct task_struct *p = current;
 129                         unsigned long mq_bytes, mq_msg_tblsz;
 130
 131                         inode->i_fop = &mqueue_file_operations;
 132                         inode->i_size = FILENT_SIZE;
 133                         /* mqueue specific info */
 134                         info = MQUEUE_I(inode);
 135                         spin_lock_init(&info->lock);
 136                         init_waitqueue_head(&info->wait_q);
 137                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 138                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 139                         info->messages = NULL;
 140                         info->notify_owner = NULL;
 141                         info->qsize = 0;
 142                         info->user = NULL;      /* set when all is ok */
 143                         memset(&info->attr, 0, sizeof(info->attr));
 144                         info->attr.mq_maxmsg = msg_max;
 145                         info->attr.mq_msgsize = msgsize_max;
 146                         if (attr) {
 147                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 148                                 info->attr.mq_msgsize = attr->mq_msgsize;
 149                         }
 150                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 151                         mq_bytes = (mq_msg_tblsz +
 152                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 153
 154                         spin_lock(&mq_lock);
 155                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 156                             u->mq_bytes + mq_bytes >
 157                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 158                                 spin_unlock(&mq_lock);
 159                                 goto out_inode;
 160                         }
 161                         u->mq_bytes += mq_bytes;
 162                         spin_unlock(&mq_lock);
 163
 164                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 165                         if (!info->messages) {
 166                                 spin_lock(&mq_lock);
 167                                 u->mq_bytes -= mq_bytes;
 168                                 spin_unlock(&mq_lock);
 169                                 goto out_inode;
 170                         }
 171                         /* all is ok */
 172                         info->user = get_uid(u);
 173                 } else if (S_ISDIR(mode)) {
 174                         inc_nlink(inode);
 175                         /* Some things misbehave if size == 0 on a directory */
 176                         inode->i_size = 2 * DIRENT_SIZE;
 177                         inode->i_op = &mqueue_dir_inode_operations;
 178                         inode->i_fop = &simple_dir_operations;
 179                 }
 180         }
 181         return inode;
 182 out_inode:
 183         make_bad_inode(inode);
 184         iput(inode);
 185         return NULL;
 186 }
 187
 188 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 189 {
 190         struct inode *inode;
 191
 192         sb->s_blocksize = PAGE_CACHE_SIZE;
 193         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 194         sb->s_magic = MQUEUE_MAGIC;
 195         sb->s_op = &mqueue_super_ops;
 196
 197         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 198         if (!inode)
 199                 return -ENOMEM;
 200
 201         sb->s_root = d_alloc_root(inode);
 202         if (!sb->s_root) {
 203                 iput(inode);
 204                 return -ENOMEM;
 205         }
 206
 207         return 0;
 208 }
 209
 210 static int mqueue_get_sb(struct file_system_type *fs_type,
 211                          int flags, const char *dev_name,
 212                          void *data, struct vfsmount *mnt)
 213 {
 214         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
 215 }
 216
 217 static void init_once(void *foo)
 218 {
 219         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 220
 221         inode_init_once(&p->vfs_inode);
 222 }
 223
 224 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 225 {
 226         struct mqueue_inode_info *ei;
 227
 228         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 229         if (!ei)
 230                 return NULL;
 231         return &ei->vfs_inode;
 232 }
 233
 234 static void mqueue_destroy_inode(struct inode *inode)
 235 {
 236         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 237 }
 238
 239 static void mqueue_delete_inode(struct inode *inode)
 240 {
 241         struct mqueue_inode_info *info;
 242         struct user_struct *user;
 243         unsigned long mq_bytes;
 244         int i;
 245
 246         if (S_ISDIR(inode->i_mode)) {
 247                 clear_inode(inode);
 248                 return;
 249         }
 250         info = MQUEUE_I(inode);
 251         spin_lock(&info->lock);
 252         for (i = 0; i < info->attr.mq_curmsgs; i++)
 253                 free_msg(info->messages[i]);
 254         kfree(info->messages);
 255         spin_unlock(&info->lock);
 256
 257         clear_inode(inode);
 258
 259         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 260                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 261         user = info->user;
 262         if (user) {
 263                 spin_lock(&mq_lock);
 264                 user->mq_bytes -= mq_bytes;
 265                 queues_count--;
 266                 spin_unlock(&mq_lock);
 267                 free_uid(user);
 268         }
 269 }
 270
 271 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 272                                 int mode, struct nameidata *nd)
 273 {
 274         struct inode *inode;
 275         struct mq_attr *attr = dentry->d_fsdata;
 276         int error;
 277
 278         spin_lock(&mq_lock);
 279         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 280                 error = -ENOSPC;
 281                 goto out_lock;
 282         }
 283         queues_count++;
 284         spin_unlock(&mq_lock);
 285
 286         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 287         if (!inode) {
 288                 error = -ENOMEM;
 289                 spin_lock(&mq_lock);
 290                 queues_count--;
 291                 goto out_lock;
 292         }
 293
 294         dir->i_size += DIRENT_SIZE;
 295         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 296
 297         d_instantiate(dentry, inode);
 298         dget(dentry);
 299         return 0;
 300 out_lock:
 301         spin_unlock(&mq_lock);
 302         return error;
 303 }
 304
 305 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 306 {
 307         struct inode *inode = dentry->d_inode;
 308
 309         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 310         dir->i_size -= DIRENT_SIZE;
 311         drop_nlink(inode);
 312         dput(dentry);
 313         return 0;
 314 }
 315
 316 /*
 317 *       This is routine for system read from queue file.
 318 *       To avoid mess with doing here some sort of mq_receive we allow
 319 *       to read only queue size & notification info (the only values
 320 *       that are interesting from user point of view and aren't accessible
 321 *       through std routines)
 322 */
 323 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 324                                 size_t count, loff_t *off)
 325 {
 326         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 327         char buffer[FILENT_SIZE];
 328         ssize_t ret;
 329
 330         spin_lock(&info->lock);
 331         snprintf(buffer, sizeof(buffer),
 332                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 333                         info->qsize,
 334                         info->notify_owner ? info->notify.sigev_notify : 0,
 335                         (info->notify_owner &&
 336                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 337                                 info->notify.sigev_signo : 0,
 338                         pid_vnr(info->notify_owner));
 339         spin_unlock(&info->lock);
 340         buffer[sizeof(buffer)-1] = '\0';
 341
 342         ret = simple_read_from_buffer(u_data, count, off, buffer,
 343                                 strlen(buffer));
 344         if (ret <= 0)
 345                 return ret;
 346
 347         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 348         return ret;
 349 }
 350
 351 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 352 {
 353         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 354
 355         spin_lock(&info->lock);
 356         if (task_tgid(current) == info->notify_owner)
 357                 remove_notification(info);
 358
 359         spin_unlock(&info->lock);
 360         return 0;
 361 }
 362
 363 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 364 {
 365         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 366         int retval = 0;
 367
 368         poll_wait(filp, &info->wait_q, poll_tab);
 369
 370         spin_lock(&info->lock);
 371         if (info->attr.mq_curmsgs)
 372                 retval = POLLIN | POLLRDNORM;
 373
 374         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 375                 retval |= POLLOUT | POLLWRNORM;
 376         spin_unlock(&info->lock);
 377
 378         return retval;
 379 }
 380
 381 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 382 static void wq_add(struct mqueue_inode_info *info, int sr,
 383                         struct ext_wait_queue *ewp)
 384 {
 385         struct ext_wait_queue *walk;
 386
 387         ewp->task = current;
 388
 389         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 390                 if (walk->task->static_prio <= current->static_prio) {
 391                         list_add_tail(&ewp->list, &walk->list);
 392                         return;
 393                 }
 394         }
 395         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 396 }
 397
 398 /*
 399  * Puts current task to sleep. Caller must hold queue lock. After return
 400  * lock isn't held.
 401  * sr: SEND or RECV
 402  */
 403 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 404                         long timeout, struct ext_wait_queue *ewp)
 405 {
 406         int retval;
 407         signed long time;
 408
 409         wq_add(info, sr, ewp);
 410
 411         for (;;) {
 412                 set_current_state(TASK_INTERRUPTIBLE);
 413
 414                 spin_unlock(&info->lock);
 415                 time = schedule_timeout(timeout);
 416
 417                 while (ewp->state == STATE_PENDING)
 418                         cpu_relax();
 419
 420                 if (ewp->state == STATE_READY) {
 421                         retval = 0;
 422                         goto out;
 423                 }
 424                 spin_lock(&info->lock);
 425                 if (ewp->state == STATE_READY) {
 426                         retval = 0;
 427                         goto out_unlock;
 428                 }
 429                 if (signal_pending(current)) {
 430                         retval = -ERESTARTSYS;
 431                         break;
 432                 }
 433                 if (time == 0) {
 434                         retval = -ETIMEDOUT;
 435                         break;
 436                 }
 437         }
 438         list_del(&ewp->list);
 439 out_unlock:
 440         spin_unlock(&info->lock);
 441 out:
 442         return retval;
 443 }
 444
 445 /*
 446  * Returns waiting task that should be serviced first or NULL if none exists
 447  */
 448 static struct ext_wait_queue *wq_get_first_waiter(
 449                 struct mqueue_inode_info *info, int sr)
 450 {
 451         struct list_head *ptr;
 452
 453         ptr = info->e_wait_q[sr].list.prev;
 454         if (ptr == &info->e_wait_q[sr].list)
 455                 return NULL;
 456         return list_entry(ptr, struct ext_wait_queue, list);
 457 }
 458
 459 /* Auxiliary functions to manipulate messages' list */
 460 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 461 {
 462         int k;
 463
 464         k = info->attr.mq_curmsgs - 1;
 465         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 466                 info->messages[k + 1] = info->messages[k];
 467                 k--;
 468         }
 469         info->attr.mq_curmsgs++;
 470         info->qsize += ptr->m_ts;
 471         info->messages[k + 1] = ptr;
 472 }
 473
 474 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 475 {
 476         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 477         return info->messages[info->attr.mq_curmsgs];
 478 }
 479
 480 static inline void set_cookie(struct sk_buff *skb, char code)
 481 {
 482         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 483 }
 484
 485 /*
 486  * The next function is only to split too long sys_mq_timedsend
 487  */
 488 static void __do_notify(struct mqueue_inode_info *info)
 489 {
 490         /* notification
 491          * invoked when there is registered process and there isn't process
 492          * waiting synchronously for message AND state of queue changed from
 493          * empty to not empty. Here we are sure that no one is waiting
 494          * synchronously. */
 495         if (info->notify_owner &&
 496             info->attr.mq_curmsgs == 1) {
 497                 struct siginfo sig_i;
 498                 switch (info->notify.sigev_notify) {
 499                 case SIGEV_NONE:
 500                         break;
 501                 case SIGEV_SIGNAL:
 502                         /* sends signal */
 503
 504                         sig_i.si_signo = info->notify.sigev_signo;
 505                         sig_i.si_errno = 0;
 506                         sig_i.si_code = SI_MESGQ;
 507                         sig_i.si_value = info->notify.sigev_value;
 508                         sig_i.si_pid = task_tgid_vnr(current);
 509                         sig_i.si_uid = current_uid();
 510
 511                         kill_pid_info(info->notify.sigev_signo,
 512                                       &sig_i, info->notify_owner);
 513                         break;
 514                 case SIGEV_THREAD:
 515                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 516                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 517                         break;
 518                 }
 519                 /* after notification unregisters process */
 520                 put_pid(info->notify_owner);
 521                 info->notify_owner = NULL;
 522         }
 523         wake_up(&info->wait_q);
 524 }
 525
 526 static long prepare_timeout(struct timespec *p)
 527 {
 528         struct timespec nowts;
 529         long timeout;
 530
 531         if (p) {
 532                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
 533                         || p->tv_nsec >= NSEC_PER_SEC))
 534                         return -EINVAL;
 535                 nowts = CURRENT_TIME;
 536                 /* first subtract as jiffies can't be too big */
 537                 p->tv_sec -= nowts.tv_sec;
 538                 if (p->tv_nsec < nowts.tv_nsec) {
 539                         p->tv_nsec += NSEC_PER_SEC;
 540                         p->tv_sec--;
 541                 }
 542                 p->tv_nsec -= nowts.tv_nsec;
 543                 if (p->tv_sec < 0)
 544                         return 0;
 545
 546                 timeout = timespec_to_jiffies(p) + 1;
 547         } else
 548                 return MAX_SCHEDULE_TIMEOUT;
 549
 550         return timeout;
 551 }
 552
 553 static void remove_notification(struct mqueue_inode_info *info)
 554 {
 555         if (info->notify_owner != NULL &&
 556             info->notify.sigev_notify == SIGEV_THREAD) {
 557                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 558                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 559         }
 560         put_pid(info->notify_owner);
 561         info->notify_owner = NULL;
 562 }
 563
 564 static int mq_attr_ok(struct mq_attr *attr)
 565 {
 566         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 567                 return 0;
 568         if (capable(CAP_SYS_RESOURCE)) {
 569                 if (attr->mq_maxmsg > HARD_MSGMAX)
 570                         return 0;
 571         } else {
 572                 if (attr->mq_maxmsg > msg_max ||
 573                                 attr->mq_msgsize > msgsize_max)
 574                         return 0;
 575         }
 576         /* check for overflow */
 577         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 578                 return 0;
 579         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 580             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 581             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 582                 return 0;
 583         return 1;
 584 }
 585
 586 /*
 587  * Invoked when creating a new queue via sys_mq_open
 588  */
 589 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 590                         int oflag, mode_t mode, struct mq_attr *attr)
 591 {
 592         const struct cred *cred = current_cred();
 593         struct file *result;
 594         int ret;
 595
 596         if (attr) {
 597                 ret = -EINVAL;
 598                 if (!mq_attr_ok(attr))
 599                         goto out;
 600                 /* store for use during create */
 601                 dentry->d_fsdata = attr;
 602         }
 603
 604         mode &= ~current->fs->umask;
 605         ret = mnt_want_write(mqueue_mnt);
 606         if (ret)
 607                 goto out;
 608         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 609         dentry->d_fsdata = NULL;
 610         if (ret)
 611                 goto out_drop_write;
 612
 613         result = dentry_open(dentry, mqueue_mnt, oflag, cred);
 614         /*
 615          * dentry_open() took a persistent mnt_want_write(),
 616          * so we can now drop this one.
 617          */
 618         mnt_drop_write(mqueue_mnt);
 619         return result;
 620
 621 out_drop_write:
 622         mnt_drop_write(mqueue_mnt);
 623 out:
 624         dput(dentry);
 625         mntput(mqueue_mnt);
 626         return ERR_PTR(ret);
 627 }
 628
 629 /* Opens existing queue */
 630 static struct file *do_open(struct dentry *dentry, int oflag)
 631 {
 632         const struct cred *cred = current_cred();
 633
 634         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 635                                                   MAY_READ | MAY_WRITE };
 636
 637         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 638                 dput(dentry);
 639                 mntput(mqueue_mnt);
 640                 return ERR_PTR(-EINVAL);
 641         }
 642
 643         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 644                 dput(dentry);
 645                 mntput(mqueue_mnt);
 646                 return ERR_PTR(-EACCES);
 647         }
 648
 649         return dentry_open(dentry, mqueue_mnt, oflag, cred);
 650 }
 651
 652 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 653                                 struct mq_attr __user *u_attr)
 654 {
 655         struct dentry *dentry;
 656         struct file *filp;
 657         char *name;
 658         struct mq_attr attr;
 659         int fd, error;
 660
 661         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 662                 return -EFAULT;
 663
 664         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 665
 666         if (IS_ERR(name = getname(u_name)))
 667                 return PTR_ERR(name);
 668
 669         fd = get_unused_fd_flags(O_CLOEXEC);
 670         if (fd < 0)
 671                 goto out_putname;
 672
 673         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 674         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 675         if (IS_ERR(dentry)) {
 676                 error = PTR_ERR(dentry);
 677                 goto out_err;
 678         }
 679         mntget(mqueue_mnt);
 680
 681         if (oflag & O_CREAT) {
 682                 if (dentry->d_inode) {  /* entry already exists */
 683                         audit_inode(name, dentry);
 684                         error = -EEXIST;
 685                         if (oflag & O_EXCL)
 686                                 goto out;
 687                         filp = do_open(dentry, oflag);
 688                 } else {
 689                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 690                                                 oflag, mode,
 691                                                 u_attr ? &attr : NULL);
 692                 }
 693         } else {
 694                 error = -ENOENT;
 695                 if (!dentry->d_inode)
 696                         goto out;
 697                 audit_inode(name, dentry);
 698                 filp = do_open(dentry, oflag);
 699         }
 700
 701         if (IS_ERR(filp)) {
 702                 error = PTR_ERR(filp);
 703                 goto out_putfd;
 704         }
 705
 706         fd_install(fd, filp);
 707         goto out_upsem;
 708
 709 out:
 710         dput(dentry);
 711         mntput(mqueue_mnt);
 712 out_putfd:
 713         put_unused_fd(fd);
 714 out_err:
 715         fd = error;
 716 out_upsem:
 717         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 718 out_putname:
 719         putname(name);
 720         return fd;
 721 }
 722
 723 asmlinkage long sys_mq_unlink(const char __user *u_name)
 724 {
 725         int err;
 726         char *name;
 727         struct dentry *dentry;
 728         struct inode *inode = NULL;
 729
 730         name = getname(u_name);
 731         if (IS_ERR(name))
 732                 return PTR_ERR(name);
 733
 734         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
 735                         I_MUTEX_PARENT);
 736         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 737         if (IS_ERR(dentry)) {
 738                 err = PTR_ERR(dentry);
 739                 goto out_unlock;
 740         }
 741
 742         if (!dentry->d_inode) {
 743                 err = -ENOENT;
 744                 goto out_err;
 745         }
 746
 747         inode = dentry->d_inode;
 748         if (inode)
 749                 atomic_inc(&inode->i_count);
 750         err = mnt_want_write(mqueue_mnt);
 751         if (err)
 752                 goto out_err;
 753         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 754         mnt_drop_write(mqueue_mnt);
 755 out_err:
 756         dput(dentry);
 757
 758 out_unlock:
 759         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 760         putname(name);
 761         if (inode)
 762                 iput(inode);
 763
 764         return err;
 765 }
 766
 767 /* Pipelined send and receive functions.
 768  *
 769  * If a receiver finds no waiting message, then it registers itself in the
 770  * list of waiting receivers. A sender checks that list before adding the new
 771  * message into the message array. If there is a waiting receiver, then it
 772  * bypasses the message array and directly hands the message over to the
 773  * receiver.
 774  * The receiver accepts the message and returns without grabbing the queue
 775  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 776  * are necessary. The same algorithm is used for sysv semaphores, see
 777  * ipc/sem.c for more details.
 778  *
 779  * The same algorithm is used for senders.
 780  */
 781
 782 /* pipelined_send() - send a message directly to the task waiting in
 783  * sys_mq_timedreceive() (without inserting message into a queue).
 784  */
 785 static inline void pipelined_send(struct mqueue_inode_info *info,
 786                                   struct msg_msg *message,
 787                                   struct ext_wait_queue *receiver)
 788 {
 789         receiver->msg = message;
 790         list_del(&receiver->list);
 791         receiver->state = STATE_PENDING;
 792         wake_up_process(receiver->task);
 793         smp_wmb();
 794         receiver->state = STATE_READY;
 795 }
 796
 797 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 798  * gets its message and put to the queue (we have one free place for sure). */
 799 static inline void pipelined_receive(struct mqueue_inode_info *info)
 800 {
 801         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 802
 803         if (!sender) {
 804                 /* for poll */
 805                 wake_up_interruptible(&info->wait_q);
 806                 return;
 807         }
 808         msg_insert(sender->msg, info);
 809         list_del(&sender->list);
 810         sender->state = STATE_PENDING;
 811         wake_up_process(sender->task);
 812         smp_wmb();
 813         sender->state = STATE_READY;
 814 }
 815
 816 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 817         size_t msg_len, unsigned int msg_prio,
 818         const struct timespec __user *u_abs_timeout)
 819 {
 820         struct file *filp;
 821         struct inode *inode;
 822         struct ext_wait_queue wait;
 823         struct ext_wait_queue *receiver;
 824         struct msg_msg *msg_ptr;
 825         struct mqueue_inode_info *info;
 826         struct timespec ts, *p = NULL;
 827         long timeout;
 828         int ret;
 829
 830         if (u_abs_timeout) {
 831                 if (copy_from_user(&ts, u_abs_timeout,
 832                                         sizeof(struct timespec)))
 833                         return -EFAULT;
 834                 p = &ts;
 835         }
 836
 837         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 838                 return -EINVAL;
 839
 840         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
 841         timeout = prepare_timeout(p);
 842
 843         ret = -EBADF;
 844         filp = fget(mqdes);
 845         if (unlikely(!filp))
 846                 goto out;
 847
 848         inode = filp->f_path.dentry->d_inode;
 849         if (unlikely(filp->f_op != &mqueue_file_operations))
 850                 goto out_fput;
 851         info = MQUEUE_I(inode);
 852         audit_inode(NULL, filp->f_path.dentry);
 853
 854         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 855                 goto out_fput;
 856
 857         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 858                 ret = -EMSGSIZE;
 859                 goto out_fput;
 860         }
 861
 862         /* First try to allocate memory, before doing anything with
 863          * existing queues. */
 864         msg_ptr = load_msg(u_msg_ptr, msg_len);
 865         if (IS_ERR(msg_ptr)) {
 866                 ret = PTR_ERR(msg_ptr);
 867                 goto out_fput;
 868         }
 869         msg_ptr->m_ts = msg_len;
 870         msg_ptr->m_type = msg_prio;
 871
 872         spin_lock(&info->lock);
 873
 874         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 875                 if (filp->f_flags & O_NONBLOCK) {
 876                         spin_unlock(&info->lock);
 877                         ret = -EAGAIN;
 878                 } else if (unlikely(timeout < 0)) {
 879                         spin_unlock(&info->lock);
 880                         ret = timeout;
 881                 } else {
 882                         wait.task = current;
 883                         wait.msg = (void *) msg_ptr;
 884                         wait.state = STATE_NONE;
 885                         ret = wq_sleep(info, SEND, timeout, &wait);
 886                 }
 887                 if (ret < 0)
 888                         free_msg(msg_ptr);
 889         } else {
 890                 receiver = wq_get_first_waiter(info, RECV);
 891                 if (receiver) {
 892                         pipelined_send(info, msg_ptr, receiver);
 893                 } else {
 894                         /* adds message to the queue */
 895                         msg_insert(msg_ptr, info);
 896                         __do_notify(info);
 897                 }
 898                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 899                                 CURRENT_TIME;
 900                 spin_unlock(&info->lock);
 901                 ret = 0;
 902         }
 903 out_fput:
 904         fput(filp);
 905 out:
 906         return ret;
 907 }
 908
 909 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 910         size_t msg_len, unsigned int __user *u_msg_prio,
 911         const struct timespec __user *u_abs_timeout)
 912 {
 913         long timeout;
 914         ssize_t ret;
 915         struct msg_msg *msg_ptr;
 916         struct file *filp;
 917         struct inode *inode;
 918         struct mqueue_inode_info *info;
 919         struct ext_wait_queue wait;
 920         struct timespec ts, *p = NULL;
 921
 922         if (u_abs_timeout) {
 923                 if (copy_from_user(&ts, u_abs_timeout,
 924                                         sizeof(struct timespec)))
 925                         return -EFAULT;
 926                 p = &ts;
 927         }
 928
 929         audit_mq_sendrecv(mqdes, msg_len, 0, p);
 930         timeout = prepare_timeout(p);
 931
 932         ret = -EBADF;
 933         filp = fget(mqdes);
 934         if (unlikely(!filp))
 935                 goto out;
 936
 937         inode = filp->f_path.dentry->d_inode;
 938         if (unlikely(filp->f_op != &mqueue_file_operations))
 939                 goto out_fput;
 940         info = MQUEUE_I(inode);
 941         audit_inode(NULL, filp->f_path.dentry);
 942
 943         if (unlikely(!(filp->f_mode & FMODE_READ)))
 944                 goto out_fput;
 945
 946         /* checks if buffer is big enough */
 947         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 948                 ret = -EMSGSIZE;
 949                 goto out_fput;
 950         }
 951
 952         spin_lock(&info->lock);
 953         if (info->attr.mq_curmsgs == 0) {
 954                 if (filp->f_flags & O_NONBLOCK) {
 955                         spin_unlock(&info->lock);
 956                         ret = -EAGAIN;
 957                         msg_ptr = NULL;
 958                 } else if (unlikely(timeout < 0)) {
 959                         spin_unlock(&info->lock);
 960                         ret = timeout;
 961                         msg_ptr = NULL;
 962                 } else {
 963                         wait.task = current;
 964                         wait.state = STATE_NONE;
 965                         ret = wq_sleep(info, RECV, timeout, &wait);
 966                         msg_ptr = wait.msg;
 967                 }
 968         } else {
 969                 msg_ptr = msg_get(info);
 970
 971                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 972                                 CURRENT_TIME;
 973
 974                 /* There is now free space in queue. */
 975                 pipelined_receive(info);
 976                 spin_unlock(&info->lock);
 977                 ret = 0;
 978         }
 979         if (ret == 0) {
 980                 ret = msg_ptr->m_ts;
 981
 982                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 983                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 984                         ret = -EFAULT;
 985                 }
 986                 free_msg(msg_ptr);
 987         }
 988 out_fput:
 989         fput(filp);
 990 out:
 991         return ret;
 992 }
 993
 994 /*
 995  * Notes: the case when user wants us to deregister (with NULL as pointer)
 996  * and he isn't currently owner of notification, will be silently discarded.
 997  * It isn't explicitly defined in the POSIX.
 998  */
 999 asmlinkage long sys_mq_notify(mqd_t mqdes,
1000                                 const struct sigevent __user *u_notification)
1001 {
1002         int ret;
1003         struct file *filp;
1004         struct sock *sock;
1005         struct inode *inode;
1006         struct sigevent notification;
1007         struct mqueue_inode_info *info;
1008         struct sk_buff *nc;
1009
1010         if (u_notification) {
1011                 if (copy_from_user(&notification, u_notification,
1012                                         sizeof(struct sigevent)))
1013                         return -EFAULT;
1014         }
1015
1016         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1017
1018         nc = NULL;
1019         sock = NULL;
1020         if (u_notification != NULL) {
1021                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1022                              notification.sigev_notify != SIGEV_SIGNAL &&
1023                              notification.sigev_notify != SIGEV_THREAD))
1024                         return -EINVAL;
1025                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1026                         !valid_signal(notification.sigev_signo)) {
1027                         return -EINVAL;
1028                 }
1029                 if (notification.sigev_notify == SIGEV_THREAD) {
1030                         long timeo;
1031
1032                         /* create the notify skb */
1033                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1034                         ret = -ENOMEM;
1035                         if (!nc)
1036                                 goto out;
1037                         ret = -EFAULT;
1038                         if (copy_from_user(nc->data,
1039                                         notification.sigev_value.sival_ptr,
1040                                         NOTIFY_COOKIE_LEN)) {
1041                                 goto out;
1042                         }
1043
1044                         /* TODO: add a header? */
1045                         skb_put(nc, NOTIFY_COOKIE_LEN);
1046                         /* and attach it to the socket */
1047 retry:
1048                         filp = fget(notification.sigev_signo);
1049                         ret = -EBADF;
1050                         if (!filp)
1051                                 goto out;
1052                         sock = netlink_getsockbyfilp(filp);
1053                         fput(filp);
1054                         if (IS_ERR(sock)) {
1055                                 ret = PTR_ERR(sock);
1056                                 sock = NULL;
1057                                 goto out;
1058                         }
1059
1060                         timeo = MAX_SCHEDULE_TIMEOUT;
1061                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1062                         if (ret == 1)
1063                                 goto retry;
1064                         if (ret) {
1065                                 sock = NULL;
1066                                 nc = NULL;
1067                                 goto out;
1068                         }
1069                 }
1070         }
1071
1072         ret = -EBADF;
1073         filp = fget(mqdes);
1074         if (!filp)
1075                 goto out;
1076
1077         inode = filp->f_path.dentry->d_inode;
1078         if (unlikely(filp->f_op != &mqueue_file_operations))
1079                 goto out_fput;
1080         info = MQUEUE_I(inode);
1081
1082         ret = 0;
1083         spin_lock(&info->lock);
1084         if (u_notification == NULL) {
1085                 if (info->notify_owner == task_tgid(current)) {
1086                         remove_notification(info);
1087                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1088                 }
1089         } else if (info->notify_owner != NULL) {
1090                 ret = -EBUSY;
1091         } else {
1092                 switch (notification.sigev_notify) {
1093                 case SIGEV_NONE:
1094                         info->notify.sigev_notify = SIGEV_NONE;
1095                         break;
1096                 case SIGEV_THREAD:
1097                         info->notify_sock = sock;
1098                         info->notify_cookie = nc;
1099                         sock = NULL;
1100                         nc = NULL;
1101                         info->notify.sigev_notify = SIGEV_THREAD;
1102                         break;
1103                 case SIGEV_SIGNAL:
1104                         info->notify.sigev_signo = notification.sigev_signo;
1105                         info->notify.sigev_value = notification.sigev_value;
1106                         info->notify.sigev_notify = SIGEV_SIGNAL;
1107                         break;
1108                 }
1109
1110                 info->notify_owner = get_pid(task_tgid(current));
1111                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1112         }
1113         spin_unlock(&info->lock);
1114 out_fput:
1115         fput(filp);
1116 out:
1117         if (sock) {
1118                 netlink_detachskb(sock, nc);
1119         } else if (nc) {
1120                 dev_kfree_skb(nc);
1121         }
1122         return ret;
1123 }
1124
1125 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1126                         const struct mq_attr __user *u_mqstat,
1127                         struct mq_attr __user *u_omqstat)
1128 {
1129         int ret;
1130         struct mq_attr mqstat, omqstat;
1131         struct file *filp;
1132         struct inode *inode;
1133         struct mqueue_inode_info *info;
1134
1135         if (u_mqstat != NULL) {
1136                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1137                         return -EFAULT;
1138                 if (mqstat.mq_flags & (~O_NONBLOCK))
1139                         return -EINVAL;
1140         }
1141
1142         ret = -EBADF;
1143         filp = fget(mqdes);
1144         if (!filp)
1145                 goto out;
1146
1147         inode = filp->f_path.dentry->d_inode;
1148         if (unlikely(filp->f_op != &mqueue_file_operations))
1149                 goto out_fput;
1150         info = MQUEUE_I(inode);
1151
1152         spin_lock(&info->lock);
1153
1154         omqstat = info->attr;
1155         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1156         if (u_mqstat) {
1157                 audit_mq_getsetattr(mqdes, &mqstat);
1158                 if (mqstat.mq_flags & O_NONBLOCK)
1159                         filp->f_flags |= O_NONBLOCK;
1160                 else
1161                         filp->f_flags &= ~O_NONBLOCK;
1162
1163                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1164         }
1165
1166         spin_unlock(&info->lock);
1167
1168         ret = 0;
1169         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1170                                                 sizeof(struct mq_attr)))
1171                 ret = -EFAULT;
1172
1173 out_fput:
1174         fput(filp);
1175 out:
1176         return ret;
1177 }
1178
1179 static const struct inode_operations mqueue_dir_inode_operations = {
1180         .lookup = simple_lookup,
1181         .create = mqueue_create,
1182         .unlink = mqueue_unlink,
1183 };
1184
1185 static const struct file_operations mqueue_file_operations = {
1186         .flush = mqueue_flush_file,
1187         .poll = mqueue_poll_file,
1188         .read = mqueue_read_file,
1189 };
1190
1191 static struct super_operations mqueue_super_ops = {
1192         .alloc_inode = mqueue_alloc_inode,
1193         .destroy_inode = mqueue_destroy_inode,
1194         .statfs = simple_statfs,
1195         .delete_inode = mqueue_delete_inode,
1196         .drop_inode = generic_delete_inode,
1197 };
1198
1199 static struct file_system_type mqueue_fs_type = {
1200         .name = "mqueue",
1201         .get_sb = mqueue_get_sb,
1202         .kill_sb = kill_litter_super,
1203 };
1204
1205 static int msg_max_limit_min = MIN_MSGMAX;
1206 static int msg_max_limit_max = MAX_MSGMAX;
1207
1208 static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1209 static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
1210
1211 static ctl_table mq_sysctls[] = {
1212         {
1213                 .procname       = "queues_max",
1214                 .data           = &queues_max,
1215                 .maxlen         = sizeof(int),
1216                 .mode           = 0644,
1217                 .proc_handler   = &proc_dointvec,
1218         },
1219         {
1220                 .procname       = "msg_max",
1221                 .data           = &msg_max,
1222                 .maxlen         = sizeof(int),
1223                 .mode           = 0644,
1224                 .proc_handler   = &proc_dointvec_minmax,
1225                 .extra1         = &msg_max_limit_min,
1226                 .extra2         = &msg_max_limit_max,
1227         },
1228         {
1229                 .procname       = "msgsize_max",
1230                 .data           = &msgsize_max,
1231                 .maxlen         = sizeof(int),
1232                 .mode           = 0644,
1233                 .proc_handler   = &proc_dointvec_minmax,
1234                 .extra1         = &msg_maxsize_limit_min,
1235                 .extra2         = &msg_maxsize_limit_max,
1236         },
1237         { .ctl_name = 0 }
1238 };
1239
1240 static ctl_table mq_sysctl_dir[] = {
1241         {
1242                 .procname       = "mqueue",
1243                 .mode           = 0555,
1244                 .child          = mq_sysctls,
1245         },
1246         { .ctl_name = 0 }
1247 };
1248
1249 static ctl_table mq_sysctl_root[] = {
1250         {
1251                 .ctl_name       = CTL_FS,
1252                 .procname       = "fs",
1253                 .mode           = 0555,
1254                 .child          = mq_sysctl_dir,
1255         },
1256         { .ctl_name = 0 }
1257 };
1258
1259 static int __init init_mqueue_fs(void)
1260 {
1261         int error;
1262
1263         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1264                                 sizeof(struct mqueue_inode_info), 0,
1265                                 SLAB_HWCACHE_ALIGN, init_once);
1266         if (mqueue_inode_cachep == NULL)
1267                 return -ENOMEM;
1268
1269         /* ignore failues - they are not fatal */
1270         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1271
1272         error = register_filesystem(&mqueue_fs_type);
1273         if (error)
1274                 goto out_sysctl;
1275
1276         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1277                 error = PTR_ERR(mqueue_mnt);
1278                 goto out_filesystem;
1279         }
1280
1281         /* internal initialization - not common for vfs */
1282         queues_count = 0;
1283         spin_lock_init(&mq_lock);
1284
1285         return 0;
1286
1287 out_filesystem:
1288         unregister_filesystem(&mqueue_fs_type);
1289 out_sysctl:
1290         if (mq_sysctl_table)
1291                 unregister_sysctl_table(mq_sysctl_table);
1292         kmem_cache_destroy(mqueue_inode_cachep);
1293         return error;
1294 }
1295
1296 __initcall(init_mqueue_fs);