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 #include <linux/ipc_namespace.h>
  35
  36 #include <net/sock.h>
  37 #include "util.h"
  38
  39 #define MQUEUE_MAGIC    0x19800202
  40 #define DIRENT_SIZE     20
  41 #define FILENT_SIZE     80
  42
  43 #define SEND            0
  44 #define RECV            1
  45
  46 #define STATE_NONE      0
  47 #define STATE_PENDING   1
  48 #define STATE_READY     2
  49
  50 struct ext_wait_queue {         /* queue of sleeping tasks */
  51         struct task_struct *task;
  52         struct list_head list;
  53         struct msg_msg *msg;    /* ptr of loaded message */
  54         int state;              /* one of STATE_* values */
  55 };
  56
  57 struct mqueue_inode_info {
  58         spinlock_t lock;
  59         struct inode vfs_inode;
  60         wait_queue_head_t wait_q;
  61
  62         struct msg_msg **messages;
  63         struct mq_attr attr;
  64
  65         struct sigevent notify;
  66         struct pid* notify_owner;
  67         struct user_struct *user;       /* user who created, for accounting */
  68         struct sock *notify_sock;
  69         struct sk_buff *notify_cookie;
  70
  71         /* for tasks waiting for free space and messages, respectively */
  72         struct ext_wait_queue e_wait_q[2];
  73
  74         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  75 };
  76
  77 static const struct inode_operations mqueue_dir_inode_operations;
  78 static const struct file_operations mqueue_file_operations;
  79 static const struct super_operations mqueue_super_ops;
  80 static void remove_notification(struct mqueue_inode_info *info);
  81
  82 static struct kmem_cache *mqueue_inode_cachep;
  83
  84 static struct ctl_table_header * mq_sysctl_table;
  85
  86 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  87 {
  88         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  89 }
  90
  91 /*
  92  * This routine should be called with the mq_lock held.
  93  */
  94 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
  95 {
  96         return get_ipc_ns(inode->i_sb->s_fs_info);
  97 }
  98
  99 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 100 {
 101         struct ipc_namespace *ns;
 102
 103         spin_lock(&mq_lock);
 104         ns = __get_ns_from_inode(inode);
 105         spin_unlock(&mq_lock);
 106         return ns;
 107 }
 108
 109 static struct inode *mqueue_get_inode(struct super_block *sb,
 110                 struct ipc_namespace *ipc_ns, int mode,
 111                 struct mq_attr *attr)
 112 {
 113         struct user_struct *u = current_user();
 114         struct inode *inode;
 115
 116         inode = new_inode(sb);
 117         if (inode) {
 118                 inode->i_mode = mode;
 119                 inode->i_uid = current_fsuid();
 120                 inode->i_gid = current_fsgid();
 121                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 122                                 CURRENT_TIME;
 123
 124                 if (S_ISREG(mode)) {
 125                         struct mqueue_inode_info *info;
 126                         struct task_struct *p = current;
 127                         unsigned long mq_bytes, mq_msg_tblsz;
 128
 129                         inode->i_fop = &mqueue_file_operations;
 130                         inode->i_size = FILENT_SIZE;
 131                         /* mqueue specific info */
 132                         info = MQUEUE_I(inode);
 133                         spin_lock_init(&info->lock);
 134                         init_waitqueue_head(&info->wait_q);
 135                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 136                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 137                         info->messages = NULL;
 138                         info->notify_owner = NULL;
 139                         info->qsize = 0;
 140                         info->user = NULL;      /* set when all is ok */
 141                         memset(&info->attr, 0, sizeof(info->attr));
 142                         info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
 143                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
 144                         if (attr) {
 145                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 146                                 info->attr.mq_msgsize = attr->mq_msgsize;
 147                         }
 148                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 149                         mq_bytes = (mq_msg_tblsz +
 150                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 151
 152                         spin_lock(&mq_lock);
 153                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 154                             u->mq_bytes + mq_bytes >
 155                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 156                                 spin_unlock(&mq_lock);
 157                                 goto out_inode;
 158                         }
 159                         u->mq_bytes += mq_bytes;
 160                         spin_unlock(&mq_lock);
 161
 162                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 163                         if (!info->messages) {
 164                                 spin_lock(&mq_lock);
 165                                 u->mq_bytes -= mq_bytes;
 166                                 spin_unlock(&mq_lock);
 167                                 goto out_inode;
 168                         }
 169                         /* all is ok */
 170                         info->user = get_uid(u);
 171                 } else if (S_ISDIR(mode)) {
 172                         inc_nlink(inode);
 173                         /* Some things misbehave if size == 0 on a directory */
 174                         inode->i_size = 2 * DIRENT_SIZE;
 175                         inode->i_op = &mqueue_dir_inode_operations;
 176                         inode->i_fop = &simple_dir_operations;
 177                 }
 178         }
 179         return inode;
 180 out_inode:
 181         make_bad_inode(inode);
 182         iput(inode);
 183         return NULL;
 184 }
 185
 186 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 187 {
 188         struct inode *inode;
 189         struct ipc_namespace *ns = data;
 190         int error = 0;
 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, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
 198                                 NULL);
 199         if (!inode) {
 200                 error = -ENOMEM;
 201                 goto out;
 202         }
 203
 204         sb->s_root = d_alloc_root(inode);
 205         if (!sb->s_root) {
 206                 iput(inode);
 207                 error = -ENOMEM;
 208         }
 209
 210 out:
 211         return error;
 212 }
 213
 214 static int mqueue_get_sb(struct file_system_type *fs_type,
 215                          int flags, const char *dev_name,
 216                          void *data, struct vfsmount *mnt)
 217 {
 218         if (!(flags & MS_KERNMOUNT))
 219                 data = current->nsproxy->ipc_ns;
 220         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
 221 }
 222
 223 static void init_once(void *foo)
 224 {
 225         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 226
 227         inode_init_once(&p->vfs_inode);
 228 }
 229
 230 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 231 {
 232         struct mqueue_inode_info *ei;
 233
 234         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 235         if (!ei)
 236                 return NULL;
 237         return &ei->vfs_inode;
 238 }
 239
 240 static void mqueue_destroy_inode(struct inode *inode)
 241 {
 242         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 243 }
 244
 245 static void mqueue_delete_inode(struct inode *inode)
 246 {
 247         struct mqueue_inode_info *info;
 248         struct user_struct *user;
 249         unsigned long mq_bytes;
 250         int i;
 251         struct ipc_namespace *ipc_ns;
 252
 253         if (S_ISDIR(inode->i_mode)) {
 254                 clear_inode(inode);
 255                 return;
 256         }
 257         ipc_ns = get_ns_from_inode(inode);
 258         info = MQUEUE_I(inode);
 259         spin_lock(&info->lock);
 260         for (i = 0; i < info->attr.mq_curmsgs; i++)
 261                 free_msg(info->messages[i]);
 262         kfree(info->messages);
 263         spin_unlock(&info->lock);
 264
 265         clear_inode(inode);
 266
 267         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 268                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 269         user = info->user;
 270         if (user) {
 271                 spin_lock(&mq_lock);
 272                 user->mq_bytes -= mq_bytes;
 273                 /*
 274                  * get_ns_from_inode() ensures that the
 275                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 276                  * to which we now hold a reference, or it is NULL.
 277                  * We can't put it here under mq_lock, though.
 278                  */
 279                 if (ipc_ns)
 280                         ipc_ns->mq_queues_count--;
 281                 spin_unlock(&mq_lock);
 282                 free_uid(user);
 283         }
 284         if (ipc_ns)
 285                 put_ipc_ns(ipc_ns);
 286 }
 287
 288 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 289                                 int mode, struct nameidata *nd)
 290 {
 291         struct inode *inode;
 292         struct mq_attr *attr = dentry->d_fsdata;
 293         int error;
 294         struct ipc_namespace *ipc_ns;
 295
 296         spin_lock(&mq_lock);
 297         ipc_ns = __get_ns_from_inode(dir);
 298         if (!ipc_ns) {
 299                 error = -EACCES;
 300                 goto out_unlock;
 301         }
 302         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 303                         !capable(CAP_SYS_RESOURCE)) {
 304                 error = -ENOSPC;
 305                 goto out_unlock;
 306         }
 307         ipc_ns->mq_queues_count++;
 308         spin_unlock(&mq_lock);
 309
 310         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 311         if (!inode) {
 312                 error = -ENOMEM;
 313                 spin_lock(&mq_lock);
 314                 ipc_ns->mq_queues_count--;
 315                 goto out_unlock;
 316         }
 317
 318         put_ipc_ns(ipc_ns);
 319         dir->i_size += DIRENT_SIZE;
 320         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 321
 322         d_instantiate(dentry, inode);
 323         dget(dentry);
 324         return 0;
 325 out_unlock:
 326         spin_unlock(&mq_lock);
 327         if (ipc_ns)
 328                 put_ipc_ns(ipc_ns);
 329         return error;
 330 }
 331
 332 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 333 {
 334         struct inode *inode = dentry->d_inode;
 335
 336         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 337         dir->i_size -= DIRENT_SIZE;
 338         drop_nlink(inode);
 339         dput(dentry);
 340         return 0;
 341 }
 342
 343 /*
 344 *       This is routine for system read from queue file.
 345 *       To avoid mess with doing here some sort of mq_receive we allow
 346 *       to read only queue size & notification info (the only values
 347 *       that are interesting from user point of view and aren't accessible
 348 *       through std routines)
 349 */
 350 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 351                                 size_t count, loff_t *off)
 352 {
 353         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 354         char buffer[FILENT_SIZE];
 355         ssize_t ret;
 356
 357         spin_lock(&info->lock);
 358         snprintf(buffer, sizeof(buffer),
 359                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 360                         info->qsize,
 361                         info->notify_owner ? info->notify.sigev_notify : 0,
 362                         (info->notify_owner &&
 363                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 364                                 info->notify.sigev_signo : 0,
 365                         pid_vnr(info->notify_owner));
 366         spin_unlock(&info->lock);
 367         buffer[sizeof(buffer)-1] = '\0';
 368
 369         ret = simple_read_from_buffer(u_data, count, off, buffer,
 370                                 strlen(buffer));
 371         if (ret <= 0)
 372                 return ret;
 373
 374         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 375         return ret;
 376 }
 377
 378 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 379 {
 380         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 381
 382         spin_lock(&info->lock);
 383         if (task_tgid(current) == info->notify_owner)
 384                 remove_notification(info);
 385
 386         spin_unlock(&info->lock);
 387         return 0;
 388 }
 389
 390 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 391 {
 392         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 393         int retval = 0;
 394
 395         poll_wait(filp, &info->wait_q, poll_tab);
 396
 397         spin_lock(&info->lock);
 398         if (info->attr.mq_curmsgs)
 399                 retval = POLLIN | POLLRDNORM;
 400
 401         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 402                 retval |= POLLOUT | POLLWRNORM;
 403         spin_unlock(&info->lock);
 404
 405         return retval;
 406 }
 407
 408 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 409 static void wq_add(struct mqueue_inode_info *info, int sr,
 410                         struct ext_wait_queue *ewp)
 411 {
 412         struct ext_wait_queue *walk;
 413
 414         ewp->task = current;
 415
 416         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 417                 if (walk->task->static_prio <= current->static_prio) {
 418                         list_add_tail(&ewp->list, &walk->list);
 419                         return;
 420                 }
 421         }
 422         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 423 }
 424
 425 /*
 426  * Puts current task to sleep. Caller must hold queue lock. After return
 427  * lock isn't held.
 428  * sr: SEND or RECV
 429  */
 430 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 431                         long timeout, struct ext_wait_queue *ewp)
 432 {
 433         int retval;
 434         signed long time;
 435
 436         wq_add(info, sr, ewp);
 437
 438         for (;;) {
 439                 set_current_state(TASK_INTERRUPTIBLE);
 440
 441                 spin_unlock(&info->lock);
 442                 time = schedule_timeout(timeout);
 443
 444                 while (ewp->state == STATE_PENDING)
 445                         cpu_relax();
 446
 447                 if (ewp->state == STATE_READY) {
 448                         retval = 0;
 449                         goto out;
 450                 }
 451                 spin_lock(&info->lock);
 452                 if (ewp->state == STATE_READY) {
 453                         retval = 0;
 454                         goto out_unlock;
 455                 }
 456                 if (signal_pending(current)) {
 457                         retval = -ERESTARTSYS;
 458                         break;
 459                 }
 460                 if (time == 0) {
 461                         retval = -ETIMEDOUT;
 462                         break;
 463                 }
 464         }
 465         list_del(&ewp->list);
 466 out_unlock:
 467         spin_unlock(&info->lock);
 468 out:
 469         return retval;
 470 }
 471
 472 /*
 473  * Returns waiting task that should be serviced first or NULL if none exists
 474  */
 475 static struct ext_wait_queue *wq_get_first_waiter(
 476                 struct mqueue_inode_info *info, int sr)
 477 {
 478         struct list_head *ptr;
 479
 480         ptr = info->e_wait_q[sr].list.prev;
 481         if (ptr == &info->e_wait_q[sr].list)
 482                 return NULL;
 483         return list_entry(ptr, struct ext_wait_queue, list);
 484 }
 485
 486 /* Auxiliary functions to manipulate messages' list */
 487 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 488 {
 489         int k;
 490
 491         k = info->attr.mq_curmsgs - 1;
 492         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 493                 info->messages[k + 1] = info->messages[k];
 494                 k--;
 495         }
 496         info->attr.mq_curmsgs++;
 497         info->qsize += ptr->m_ts;
 498         info->messages[k + 1] = ptr;
 499 }
 500
 501 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 502 {
 503         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 504         return info->messages[info->attr.mq_curmsgs];
 505 }
 506
 507 static inline void set_cookie(struct sk_buff *skb, char code)
 508 {
 509         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 510 }
 511
 512 /*
 513  * The next function is only to split too long sys_mq_timedsend
 514  */
 515 static void __do_notify(struct mqueue_inode_info *info)
 516 {
 517         /* notification
 518          * invoked when there is registered process and there isn't process
 519          * waiting synchronously for message AND state of queue changed from
 520          * empty to not empty. Here we are sure that no one is waiting
 521          * synchronously. */
 522         if (info->notify_owner &&
 523             info->attr.mq_curmsgs == 1) {
 524                 struct siginfo sig_i;
 525                 switch (info->notify.sigev_notify) {
 526                 case SIGEV_NONE:
 527                         break;
 528                 case SIGEV_SIGNAL:
 529                         /* sends signal */
 530
 531                         sig_i.si_signo = info->notify.sigev_signo;
 532                         sig_i.si_errno = 0;
 533                         sig_i.si_code = SI_MESGQ;
 534                         sig_i.si_value = info->notify.sigev_value;
 535                         sig_i.si_pid = task_tgid_nr_ns(current,
 536                                                 ns_of_pid(info->notify_owner));
 537                         sig_i.si_uid = current_uid();
 538
 539                         kill_pid_info(info->notify.sigev_signo,
 540                                       &sig_i, info->notify_owner);
 541                         break;
 542                 case SIGEV_THREAD:
 543                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 544                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 545                         break;
 546                 }
 547                 /* after notification unregisters process */
 548                 put_pid(info->notify_owner);
 549                 info->notify_owner = NULL;
 550         }
 551         wake_up(&info->wait_q);
 552 }
 553
 554 static long prepare_timeout(struct timespec *p)
 555 {
 556         struct timespec nowts;
 557         long timeout;
 558
 559         if (p) {
 560                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
 561                         || p->tv_nsec >= NSEC_PER_SEC))
 562                         return -EINVAL;
 563                 nowts = CURRENT_TIME;
 564                 /* first subtract as jiffies can't be too big */
 565                 p->tv_sec -= nowts.tv_sec;
 566                 if (p->tv_nsec < nowts.tv_nsec) {
 567                         p->tv_nsec += NSEC_PER_SEC;
 568                         p->tv_sec--;
 569                 }
 570                 p->tv_nsec -= nowts.tv_nsec;
 571                 if (p->tv_sec < 0)
 572                         return 0;
 573
 574                 timeout = timespec_to_jiffies(p) + 1;
 575         } else
 576                 return MAX_SCHEDULE_TIMEOUT;
 577
 578         return timeout;
 579 }
 580
 581 static void remove_notification(struct mqueue_inode_info *info)
 582 {
 583         if (info->notify_owner != NULL &&
 584             info->notify.sigev_notify == SIGEV_THREAD) {
 585                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 586                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 587         }
 588         put_pid(info->notify_owner);
 589         info->notify_owner = NULL;
 590 }
 591
 592 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 593 {
 594         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 595                 return 0;
 596         if (capable(CAP_SYS_RESOURCE)) {
 597                 if (attr->mq_maxmsg > HARD_MSGMAX)
 598                         return 0;
 599         } else {
 600                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 601                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 602                         return 0;
 603         }
 604         /* check for overflow */
 605         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 606                 return 0;
 607         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 608             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 609             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 610                 return 0;
 611         return 1;
 612 }
 613
 614 /*
 615  * Invoked when creating a new queue via sys_mq_open
 616  */
 617 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 618                         struct dentry *dentry, int oflag, mode_t mode,
 619                         struct mq_attr *attr)
 620 {
 621         const struct cred *cred = current_cred();
 622         struct file *result;
 623         int ret;
 624
 625         if (attr) {
 626                 ret = -EINVAL;
 627                 if (!mq_attr_ok(ipc_ns, attr))
 628                         goto out;
 629                 /* store for use during create */
 630                 dentry->d_fsdata = attr;
 631         }
 632
 633         mode &= ~current_umask();
 634         ret = mnt_want_write(ipc_ns->mq_mnt);
 635         if (ret)
 636                 goto out;
 637         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 638         dentry->d_fsdata = NULL;
 639         if (ret)
 640                 goto out_drop_write;
 641
 642         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 643         /*
 644          * dentry_open() took a persistent mnt_want_write(),
 645          * so we can now drop this one.
 646          */
 647         mnt_drop_write(ipc_ns->mq_mnt);
 648         return result;
 649
 650 out_drop_write:
 651         mnt_drop_write(ipc_ns->mq_mnt);
 652 out:
 653         dput(dentry);
 654         mntput(ipc_ns->mq_mnt);
 655         return ERR_PTR(ret);
 656 }
 657
 658 /* Opens existing queue */
 659 static struct file *do_open(struct ipc_namespace *ipc_ns,
 660                                 struct dentry *dentry, int oflag)
 661 {
 662         const struct cred *cred = current_cred();
 663
 664         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 665                                                   MAY_READ | MAY_WRITE };
 666
 667         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 668                 dput(dentry);
 669                 mntput(ipc_ns->mq_mnt);
 670                 return ERR_PTR(-EINVAL);
 671         }
 672
 673         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 674                 dput(dentry);
 675                 mntput(ipc_ns->mq_mnt);
 676                 return ERR_PTR(-EACCES);
 677         }
 678
 679         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 680 }
 681
 682 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
 683                 struct mq_attr __user *, u_attr)
 684 {
 685         struct dentry *dentry;
 686         struct file *filp;
 687         char *name;
 688         struct mq_attr attr;
 689         int fd, error;
 690         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 691
 692         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 693                 return -EFAULT;
 694
 695         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 696
 697         if (IS_ERR(name = getname(u_name)))
 698                 return PTR_ERR(name);
 699
 700         fd = get_unused_fd_flags(O_CLOEXEC);
 701         if (fd < 0)
 702                 goto out_putname;
 703
 704         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 705         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 706         if (IS_ERR(dentry)) {
 707                 error = PTR_ERR(dentry);
 708                 goto out_err;
 709         }
 710         mntget(ipc_ns->mq_mnt);
 711
 712         if (oflag & O_CREAT) {
 713                 if (dentry->d_inode) {  /* entry already exists */
 714                         audit_inode(name, dentry);
 715                         error = -EEXIST;
 716                         if (oflag & O_EXCL)
 717                                 goto out;
 718                         filp = do_open(ipc_ns, dentry, oflag);
 719                 } else {
 720                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 721                                                 dentry, oflag, mode,
 722                                                 u_attr ? &attr : NULL);
 723                 }
 724         } else {
 725                 error = -ENOENT;
 726                 if (!dentry->d_inode)
 727                         goto out;
 728                 audit_inode(name, dentry);
 729                 filp = do_open(ipc_ns, dentry, oflag);
 730         }
 731
 732         if (IS_ERR(filp)) {
 733                 error = PTR_ERR(filp);
 734                 goto out_putfd;
 735         }
 736
 737         fd_install(fd, filp);
 738         goto out_upsem;
 739
 740 out:
 741         dput(dentry);
 742         mntput(ipc_ns->mq_mnt);
 743 out_putfd:
 744         put_unused_fd(fd);
 745 out_err:
 746         fd = error;
 747 out_upsem:
 748         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 749 out_putname:
 750         putname(name);
 751         return fd;
 752 }
 753
 754 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 755 {
 756         int err;
 757         char *name;
 758         struct dentry *dentry;
 759         struct inode *inode = NULL;
 760         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 761
 762         name = getname(u_name);
 763         if (IS_ERR(name))
 764                 return PTR_ERR(name);
 765
 766         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 767                         I_MUTEX_PARENT);
 768         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 769         if (IS_ERR(dentry)) {
 770                 err = PTR_ERR(dentry);
 771                 goto out_unlock;
 772         }
 773
 774         if (!dentry->d_inode) {
 775                 err = -ENOENT;
 776                 goto out_err;
 777         }
 778
 779         inode = dentry->d_inode;
 780         if (inode)
 781                 atomic_inc(&inode->i_count);
 782         err = mnt_want_write(ipc_ns->mq_mnt);
 783         if (err)
 784                 goto out_err;
 785         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 786         mnt_drop_write(ipc_ns->mq_mnt);
 787 out_err:
 788         dput(dentry);
 789
 790 out_unlock:
 791         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 792         putname(name);
 793         if (inode)
 794                 iput(inode);
 795
 796         return err;
 797 }
 798
 799 /* Pipelined send and receive functions.
 800  *
 801  * If a receiver finds no waiting message, then it registers itself in the
 802  * list of waiting receivers. A sender checks that list before adding the new
 803  * message into the message array. If there is a waiting receiver, then it
 804  * bypasses the message array and directly hands the message over to the
 805  * receiver.
 806  * The receiver accepts the message and returns without grabbing the queue
 807  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 808  * are necessary. The same algorithm is used for sysv semaphores, see
 809  * ipc/sem.c for more details.
 810  *
 811  * The same algorithm is used for senders.
 812  */
 813
 814 /* pipelined_send() - send a message directly to the task waiting in
 815  * sys_mq_timedreceive() (without inserting message into a queue).
 816  */
 817 static inline void pipelined_send(struct mqueue_inode_info *info,
 818                                   struct msg_msg *message,
 819                                   struct ext_wait_queue *receiver)
 820 {
 821         receiver->msg = message;
 822         list_del(&receiver->list);
 823         receiver->state = STATE_PENDING;
 824         wake_up_process(receiver->task);
 825         smp_wmb();
 826         receiver->state = STATE_READY;
 827 }
 828
 829 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 830  * gets its message and put to the queue (we have one free place for sure). */
 831 static inline void pipelined_receive(struct mqueue_inode_info *info)
 832 {
 833         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 834
 835         if (!sender) {
 836                 /* for poll */
 837                 wake_up_interruptible(&info->wait_q);
 838                 return;
 839         }
 840         msg_insert(sender->msg, info);
 841         list_del(&sender->list);
 842         sender->state = STATE_PENDING;
 843         wake_up_process(sender->task);
 844         smp_wmb();
 845         sender->state = STATE_READY;
 846 }
 847
 848 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 849                 size_t, msg_len, unsigned int, msg_prio,
 850                 const struct timespec __user *, u_abs_timeout)
 851 {
 852         struct file *filp;
 853         struct inode *inode;
 854         struct ext_wait_queue wait;
 855         struct ext_wait_queue *receiver;
 856         struct msg_msg *msg_ptr;
 857         struct mqueue_inode_info *info;
 858         struct timespec ts, *p = NULL;
 859         long timeout;
 860         int ret;
 861
 862         if (u_abs_timeout) {
 863                 if (copy_from_user(&ts, u_abs_timeout,
 864                                         sizeof(struct timespec)))
 865                         return -EFAULT;
 866                 p = &ts;
 867         }
 868
 869         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 870                 return -EINVAL;
 871
 872         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
 873         timeout = prepare_timeout(p);
 874
 875         ret = -EBADF;
 876         filp = fget(mqdes);
 877         if (unlikely(!filp))
 878                 goto out;
 879
 880         inode = filp->f_path.dentry->d_inode;
 881         if (unlikely(filp->f_op != &mqueue_file_operations))
 882                 goto out_fput;
 883         info = MQUEUE_I(inode);
 884         audit_inode(NULL, filp->f_path.dentry);
 885
 886         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 887                 goto out_fput;
 888
 889         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 890                 ret = -EMSGSIZE;
 891                 goto out_fput;
 892         }
 893
 894         /* First try to allocate memory, before doing anything with
 895          * existing queues. */
 896         msg_ptr = load_msg(u_msg_ptr, msg_len);
 897         if (IS_ERR(msg_ptr)) {
 898                 ret = PTR_ERR(msg_ptr);
 899                 goto out_fput;
 900         }
 901         msg_ptr->m_ts = msg_len;
 902         msg_ptr->m_type = msg_prio;
 903
 904         spin_lock(&info->lock);
 905
 906         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 907                 if (filp->f_flags & O_NONBLOCK) {
 908                         spin_unlock(&info->lock);
 909                         ret = -EAGAIN;
 910                 } else if (unlikely(timeout < 0)) {
 911                         spin_unlock(&info->lock);
 912                         ret = timeout;
 913                 } else {
 914                         wait.task = current;
 915                         wait.msg = (void *) msg_ptr;
 916                         wait.state = STATE_NONE;
 917                         ret = wq_sleep(info, SEND, timeout, &wait);
 918                 }
 919                 if (ret < 0)
 920                         free_msg(msg_ptr);
 921         } else {
 922                 receiver = wq_get_first_waiter(info, RECV);
 923                 if (receiver) {
 924                         pipelined_send(info, msg_ptr, receiver);
 925                 } else {
 926                         /* adds message to the queue */
 927                         msg_insert(msg_ptr, info);
 928                         __do_notify(info);
 929                 }
 930                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 931                                 CURRENT_TIME;
 932                 spin_unlock(&info->lock);
 933                 ret = 0;
 934         }
 935 out_fput:
 936         fput(filp);
 937 out:
 938         return ret;
 939 }
 940
 941 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 942                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 943                 const struct timespec __user *, u_abs_timeout)
 944 {
 945         long timeout;
 946         ssize_t ret;
 947         struct msg_msg *msg_ptr;
 948         struct file *filp;
 949         struct inode *inode;
 950         struct mqueue_inode_info *info;
 951         struct ext_wait_queue wait;
 952         struct timespec ts, *p = NULL;
 953
 954         if (u_abs_timeout) {
 955                 if (copy_from_user(&ts, u_abs_timeout,
 956                                         sizeof(struct timespec)))
 957                         return -EFAULT;
 958                 p = &ts;
 959         }
 960
 961         audit_mq_sendrecv(mqdes, msg_len, 0, p);
 962         timeout = prepare_timeout(p);
 963
 964         ret = -EBADF;
 965         filp = fget(mqdes);
 966         if (unlikely(!filp))
 967                 goto out;
 968
 969         inode = filp->f_path.dentry->d_inode;
 970         if (unlikely(filp->f_op != &mqueue_file_operations))
 971                 goto out_fput;
 972         info = MQUEUE_I(inode);
 973         audit_inode(NULL, filp->f_path.dentry);
 974
 975         if (unlikely(!(filp->f_mode & FMODE_READ)))
 976                 goto out_fput;
 977
 978         /* checks if buffer is big enough */
 979         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 980                 ret = -EMSGSIZE;
 981                 goto out_fput;
 982         }
 983
 984         spin_lock(&info->lock);
 985         if (info->attr.mq_curmsgs == 0) {
 986                 if (filp->f_flags & O_NONBLOCK) {
 987                         spin_unlock(&info->lock);
 988                         ret = -EAGAIN;
 989                         msg_ptr = NULL;
 990                 } else if (unlikely(timeout < 0)) {
 991                         spin_unlock(&info->lock);
 992                         ret = timeout;
 993                         msg_ptr = NULL;
 994                 } else {
 995                         wait.task = current;
 996                         wait.state = STATE_NONE;
 997                         ret = wq_sleep(info, RECV, timeout, &wait);
 998                         msg_ptr = wait.msg;
 999                 }
1000         } else {
1001                 msg_ptr = msg_get(info);
1002
1003                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1004                                 CURRENT_TIME;
1005
1006                 /* There is now free space in queue. */
1007                 pipelined_receive(info);
1008                 spin_unlock(&info->lock);
1009                 ret = 0;
1010         }
1011         if (ret == 0) {
1012                 ret = msg_ptr->m_ts;
1013
1014                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1015                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1016                         ret = -EFAULT;
1017                 }
1018                 free_msg(msg_ptr);
1019         }
1020 out_fput:
1021         fput(filp);
1022 out:
1023         return ret;
1024 }
1025
1026 /*
1027  * Notes: the case when user wants us to deregister (with NULL as pointer)
1028  * and he isn't currently owner of notification, will be silently discarded.
1029  * It isn't explicitly defined in the POSIX.
1030  */
1031 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1032                 const struct sigevent __user *, u_notification)
1033 {
1034         int ret;
1035         struct file *filp;
1036         struct sock *sock;
1037         struct inode *inode;
1038         struct sigevent notification;
1039         struct mqueue_inode_info *info;
1040         struct sk_buff *nc;
1041
1042         if (u_notification) {
1043                 if (copy_from_user(&notification, u_notification,
1044                                         sizeof(struct sigevent)))
1045                         return -EFAULT;
1046         }
1047
1048         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1049
1050         nc = NULL;
1051         sock = NULL;
1052         if (u_notification != NULL) {
1053                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1054                              notification.sigev_notify != SIGEV_SIGNAL &&
1055                              notification.sigev_notify != SIGEV_THREAD))
1056                         return -EINVAL;
1057                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1058                         !valid_signal(notification.sigev_signo)) {
1059                         return -EINVAL;
1060                 }
1061                 if (notification.sigev_notify == SIGEV_THREAD) {
1062                         long timeo;
1063
1064                         /* create the notify skb */
1065                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1066                         ret = -ENOMEM;
1067                         if (!nc)
1068                                 goto out;
1069                         ret = -EFAULT;
1070                         if (copy_from_user(nc->data,
1071                                         notification.sigev_value.sival_ptr,
1072                                         NOTIFY_COOKIE_LEN)) {
1073                                 goto out;
1074                         }
1075
1076                         /* TODO: add a header? */
1077                         skb_put(nc, NOTIFY_COOKIE_LEN);
1078                         /* and attach it to the socket */
1079 retry:
1080                         filp = fget(notification.sigev_signo);
1081                         ret = -EBADF;
1082                         if (!filp)
1083                                 goto out;
1084                         sock = netlink_getsockbyfilp(filp);
1085                         fput(filp);
1086                         if (IS_ERR(sock)) {
1087                                 ret = PTR_ERR(sock);
1088                                 sock = NULL;
1089                                 goto out;
1090                         }
1091
1092                         timeo = MAX_SCHEDULE_TIMEOUT;
1093                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1094                         if (ret == 1)
1095                                 goto retry;
1096                         if (ret) {
1097                                 sock = NULL;
1098                                 nc = NULL;
1099                                 goto out;
1100                         }
1101                 }
1102         }
1103
1104         ret = -EBADF;
1105         filp = fget(mqdes);
1106         if (!filp)
1107                 goto out;
1108
1109         inode = filp->f_path.dentry->d_inode;
1110         if (unlikely(filp->f_op != &mqueue_file_operations))
1111                 goto out_fput;
1112         info = MQUEUE_I(inode);
1113
1114         ret = 0;
1115         spin_lock(&info->lock);
1116         if (u_notification == NULL) {
1117                 if (info->notify_owner == task_tgid(current)) {
1118                         remove_notification(info);
1119                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1120                 }
1121         } else if (info->notify_owner != NULL) {
1122                 ret = -EBUSY;
1123         } else {
1124                 switch (notification.sigev_notify) {
1125                 case SIGEV_NONE:
1126                         info->notify.sigev_notify = SIGEV_NONE;
1127                         break;
1128                 case SIGEV_THREAD:
1129                         info->notify_sock = sock;
1130                         info->notify_cookie = nc;
1131                         sock = NULL;
1132                         nc = NULL;
1133                         info->notify.sigev_notify = SIGEV_THREAD;
1134                         break;
1135                 case SIGEV_SIGNAL:
1136                         info->notify.sigev_signo = notification.sigev_signo;
1137                         info->notify.sigev_value = notification.sigev_value;
1138                         info->notify.sigev_notify = SIGEV_SIGNAL;
1139                         break;
1140                 }
1141
1142                 info->notify_owner = get_pid(task_tgid(current));
1143                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1144         }
1145         spin_unlock(&info->lock);
1146 out_fput:
1147         fput(filp);
1148 out:
1149         if (sock) {
1150                 netlink_detachskb(sock, nc);
1151         } else if (nc) {
1152                 dev_kfree_skb(nc);
1153         }
1154         return ret;
1155 }
1156
1157 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1158                 const struct mq_attr __user *, u_mqstat,
1159                 struct mq_attr __user *, u_omqstat)
1160 {
1161         int ret;
1162         struct mq_attr mqstat, omqstat;
1163         struct file *filp;
1164         struct inode *inode;
1165         struct mqueue_inode_info *info;
1166
1167         if (u_mqstat != NULL) {
1168                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1169                         return -EFAULT;
1170                 if (mqstat.mq_flags & (~O_NONBLOCK))
1171                         return -EINVAL;
1172         }
1173
1174         ret = -EBADF;
1175         filp = fget(mqdes);
1176         if (!filp)
1177                 goto out;
1178
1179         inode = filp->f_path.dentry->d_inode;
1180         if (unlikely(filp->f_op != &mqueue_file_operations))
1181                 goto out_fput;
1182         info = MQUEUE_I(inode);
1183
1184         spin_lock(&info->lock);
1185
1186         omqstat = info->attr;
1187         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1188         if (u_mqstat) {
1189                 audit_mq_getsetattr(mqdes, &mqstat);
1190                 spin_lock(&filp->f_lock);
1191                 if (mqstat.mq_flags & O_NONBLOCK)
1192                         filp->f_flags |= O_NONBLOCK;
1193                 else
1194                         filp->f_flags &= ~O_NONBLOCK;
1195                 spin_unlock(&filp->f_lock);
1196
1197                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1198         }
1199
1200         spin_unlock(&info->lock);
1201
1202         ret = 0;
1203         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1204                                                 sizeof(struct mq_attr)))
1205                 ret = -EFAULT;
1206
1207 out_fput:
1208         fput(filp);
1209 out:
1210         return ret;
1211 }
1212
1213 static const struct inode_operations mqueue_dir_inode_operations = {
1214         .lookup = simple_lookup,
1215         .create = mqueue_create,
1216         .unlink = mqueue_unlink,
1217 };
1218
1219 static const struct file_operations mqueue_file_operations = {
1220         .flush = mqueue_flush_file,
1221         .poll = mqueue_poll_file,
1222         .read = mqueue_read_file,
1223 };
1224
1225 static const struct super_operations mqueue_super_ops = {
1226         .alloc_inode = mqueue_alloc_inode,
1227         .destroy_inode = mqueue_destroy_inode,
1228         .statfs = simple_statfs,
1229         .delete_inode = mqueue_delete_inode,
1230         .drop_inode = generic_delete_inode,
1231 };
1232
1233 static struct file_system_type mqueue_fs_type = {
1234         .name = "mqueue",
1235         .get_sb = mqueue_get_sb,
1236         .kill_sb = kill_litter_super,
1237 };
1238
1239 int mq_init_ns(struct ipc_namespace *ns)
1240 {
1241         ns->mq_queues_count  = 0;
1242         ns->mq_queues_max    = DFLT_QUEUESMAX;
1243         ns->mq_msg_max       = DFLT_MSGMAX;
1244         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1245
1246         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1247         if (IS_ERR(ns->mq_mnt)) {
1248                 int err = PTR_ERR(ns->mq_mnt);
1249                 ns->mq_mnt = NULL;
1250                 return err;
1251         }
1252         return 0;
1253 }
1254
1255 void mq_clear_sbinfo(struct ipc_namespace *ns)
1256 {
1257         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1258 }
1259
1260 void mq_put_mnt(struct ipc_namespace *ns)
1261 {
1262         mntput(ns->mq_mnt);
1263 }
1264
1265 static int __init init_mqueue_fs(void)
1266 {
1267         int error;
1268
1269         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1270                                 sizeof(struct mqueue_inode_info), 0,
1271                                 SLAB_HWCACHE_ALIGN, init_once);
1272         if (mqueue_inode_cachep == NULL)
1273                 return -ENOMEM;
1274
1275         /* ignore failues - they are not fatal */
1276         mq_sysctl_table = mq_register_sysctl_table();
1277
1278         error = register_filesystem(&mqueue_fs_type);
1279         if (error)
1280                 goto out_sysctl;
1281
1282         spin_lock_init(&mq_lock);
1283
1284         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1285         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1286                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1287                 goto out_filesystem;
1288         }
1289
1290         return 0;
1291
1292 out_filesystem:
1293         unregister_filesystem(&mqueue_fs_type);
1294 out_sysctl:
1295         if (mq_sysctl_table)
1296                 unregister_sysctl_table(mq_sysctl_table);
1297         kmem_cache_destroy(mqueue_inode_cachep);
1298         return error;
1299 }
1300
1301 __initcall(init_mqueue_fs);