mqueue.c revision d6629859b36d953a4b1369b749f178736911bf10
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/vmalloc.h> 28#include <linux/netlink.h> 29#include <linux/syscalls.h> 30#include <linux/audit.h> 31#include <linux/signal.h> 32#include <linux/mutex.h> 33#include <linux/nsproxy.h> 34#include <linux/pid.h> 35#include <linux/ipc_namespace.h> 36#include <linux/user_namespace.h> 37#include <linux/slab.h> 38 39#include <net/sock.h> 40#include "util.h" 41 42#define MQUEUE_MAGIC 0x19800202 43#define DIRENT_SIZE 20 44#define FILENT_SIZE 80 45 46#define SEND 0 47#define RECV 1 48 49#define STATE_NONE 0 50#define STATE_PENDING 1 51#define STATE_READY 2 52 53struct posix_msg_tree_node { 54 struct rb_node rb_node; 55 struct list_head msg_list; 56 int priority; 57}; 58 59struct ext_wait_queue { /* queue of sleeping tasks */ 60 struct task_struct *task; 61 struct list_head list; 62 struct msg_msg *msg; /* ptr of loaded message */ 63 int state; /* one of STATE_* values */ 64}; 65 66struct mqueue_inode_info { 67 spinlock_t lock; 68 struct inode vfs_inode; 69 wait_queue_head_t wait_q; 70 71 struct rb_root msg_tree; 72 struct mq_attr attr; 73 74 struct sigevent notify; 75 struct pid* notify_owner; 76 struct user_namespace *notify_user_ns; 77 struct user_struct *user; /* user who created, for accounting */ 78 struct sock *notify_sock; 79 struct sk_buff *notify_cookie; 80 81 /* for tasks waiting for free space and messages, respectively */ 82 struct ext_wait_queue e_wait_q[2]; 83 84 unsigned long qsize; /* size of queue in memory (sum of all msgs) */ 85}; 86 87static const struct inode_operations mqueue_dir_inode_operations; 88static const struct file_operations mqueue_file_operations; 89static const struct super_operations mqueue_super_ops; 90static void remove_notification(struct mqueue_inode_info *info); 91 92static struct kmem_cache *mqueue_inode_cachep; 93 94static struct ctl_table_header * mq_sysctl_table; 95 96static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode) 97{ 98 return container_of(inode, struct mqueue_inode_info, vfs_inode); 99} 100 101/* 102 * This routine should be called with the mq_lock held. 103 */ 104static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode) 105{ 106 return get_ipc_ns(inode->i_sb->s_fs_info); 107} 108 109static struct ipc_namespace *get_ns_from_inode(struct inode *inode) 110{ 111 struct ipc_namespace *ns; 112 113 spin_lock(&mq_lock); 114 ns = __get_ns_from_inode(inode); 115 spin_unlock(&mq_lock); 116 return ns; 117} 118 119/* Auxiliary functions to manipulate messages' list */ 120static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info) 121{ 122 struct rb_node **p, *parent = NULL; 123 struct posix_msg_tree_node *leaf; 124 125 p = &info->msg_tree.rb_node; 126 while (*p) { 127 parent = *p; 128 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); 129 130 if (likely(leaf->priority == msg->m_type)) 131 goto insert_msg; 132 else if (msg->m_type < leaf->priority) 133 p = &(*p)->rb_left; 134 else 135 p = &(*p)->rb_right; 136 } 137 leaf = kzalloc(sizeof(*leaf), GFP_ATOMIC); 138 if (!leaf) 139 return -ENOMEM; 140 rb_init_node(&leaf->rb_node); 141 INIT_LIST_HEAD(&leaf->msg_list); 142 leaf->priority = msg->m_type; 143 rb_link_node(&leaf->rb_node, parent, p); 144 rb_insert_color(&leaf->rb_node, &info->msg_tree); 145 info->qsize += sizeof(struct posix_msg_tree_node); 146insert_msg: 147 info->attr.mq_curmsgs++; 148 info->qsize += msg->m_ts; 149 list_add_tail(&msg->m_list, &leaf->msg_list); 150 return 0; 151} 152 153static inline struct msg_msg *msg_get(struct mqueue_inode_info *info) 154{ 155 struct rb_node **p, *parent = NULL; 156 struct posix_msg_tree_node *leaf; 157 struct msg_msg *msg; 158 159try_again: 160 p = &info->msg_tree.rb_node; 161 while (*p) { 162 parent = *p; 163 /* 164 * During insert, low priorities go to the left and high to the 165 * right. On receive, we want the highest priorities first, so 166 * walk all the way to the right. 167 */ 168 p = &(*p)->rb_right; 169 } 170 if (!parent) { 171 if (info->attr.mq_curmsgs) { 172 pr_warn_once("Inconsistency in POSIX message queue, " 173 "no tree element, but supposedly messages " 174 "should exist!\n"); 175 info->attr.mq_curmsgs = 0; 176 } 177 return NULL; 178 } 179 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); 180 if (list_empty(&leaf->msg_list)) { 181 pr_warn_once("Inconsistency in POSIX message queue, " 182 "empty leaf node but we haven't implemented " 183 "lazy leaf delete!\n"); 184 rb_erase(&leaf->rb_node, &info->msg_tree); 185 info->qsize -= sizeof(struct posix_msg_tree_node); 186 kfree(leaf); 187 goto try_again; 188 } else { 189 msg = list_first_entry(&leaf->msg_list, 190 struct msg_msg, m_list); 191 list_del(&msg->m_list); 192 if (list_empty(&leaf->msg_list)) { 193 rb_erase(&leaf->rb_node, &info->msg_tree); 194 info->qsize -= sizeof(struct posix_msg_tree_node); 195 kfree(leaf); 196 } 197 } 198 info->attr.mq_curmsgs--; 199 info->qsize -= msg->m_ts; 200 return msg; 201} 202 203static struct inode *mqueue_get_inode(struct super_block *sb, 204 struct ipc_namespace *ipc_ns, umode_t mode, 205 struct mq_attr *attr) 206{ 207 struct user_struct *u = current_user(); 208 struct inode *inode; 209 int ret = -ENOMEM; 210 211 inode = new_inode(sb); 212 if (!inode) 213 goto err; 214 215 inode->i_ino = get_next_ino(); 216 inode->i_mode = mode; 217 inode->i_uid = current_fsuid(); 218 inode->i_gid = current_fsgid(); 219 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME; 220 221 if (S_ISREG(mode)) { 222 struct mqueue_inode_info *info; 223 unsigned long mq_bytes, mq_treesize; 224 225 inode->i_fop = &mqueue_file_operations; 226 inode->i_size = FILENT_SIZE; 227 /* mqueue specific info */ 228 info = MQUEUE_I(inode); 229 spin_lock_init(&info->lock); 230 init_waitqueue_head(&info->wait_q); 231 INIT_LIST_HEAD(&info->e_wait_q[0].list); 232 INIT_LIST_HEAD(&info->e_wait_q[1].list); 233 info->notify_owner = NULL; 234 info->notify_user_ns = NULL; 235 info->qsize = 0; 236 info->user = NULL; /* set when all is ok */ 237 info->msg_tree = RB_ROOT; 238 memset(&info->attr, 0, sizeof(info->attr)); 239 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max, 240 ipc_ns->mq_msg_default); 241 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max, 242 ipc_ns->mq_msgsize_default); 243 if (attr) { 244 info->attr.mq_maxmsg = attr->mq_maxmsg; 245 info->attr.mq_msgsize = attr->mq_msgsize; 246 } 247 /* 248 * We used to allocate a static array of pointers and account 249 * the size of that array as well as one msg_msg struct per 250 * possible message into the queue size. That's no longer 251 * accurate as the queue is now an rbtree and will grow and 252 * shrink depending on usage patterns. We can, however, still 253 * account one msg_msg struct per message, but the nodes are 254 * allocated depending on priority usage, and most programs 255 * only use one, or a handful, of priorities. However, since 256 * this is pinned memory, we need to assume worst case, so 257 * that means the min(mq_maxmsg, max_priorities) * struct 258 * posix_msg_tree_node. 259 */ 260 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + 261 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * 262 sizeof(struct posix_msg_tree_node); 263 264 mq_bytes = mq_treesize + (info->attr.mq_maxmsg * 265 info->attr.mq_msgsize); 266 267 spin_lock(&mq_lock); 268 if (u->mq_bytes + mq_bytes < u->mq_bytes || 269 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) { 270 spin_unlock(&mq_lock); 271 /* mqueue_evict_inode() releases info->messages */ 272 ret = -EMFILE; 273 goto out_inode; 274 } 275 u->mq_bytes += mq_bytes; 276 spin_unlock(&mq_lock); 277 278 /* all is ok */ 279 info->user = get_uid(u); 280 } else if (S_ISDIR(mode)) { 281 inc_nlink(inode); 282 /* Some things misbehave if size == 0 on a directory */ 283 inode->i_size = 2 * DIRENT_SIZE; 284 inode->i_op = &mqueue_dir_inode_operations; 285 inode->i_fop = &simple_dir_operations; 286 } 287 288 return inode; 289out_inode: 290 iput(inode); 291err: 292 return ERR_PTR(ret); 293} 294 295static int mqueue_fill_super(struct super_block *sb, void *data, int silent) 296{ 297 struct inode *inode; 298 struct ipc_namespace *ns = data; 299 300 sb->s_blocksize = PAGE_CACHE_SIZE; 301 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 302 sb->s_magic = MQUEUE_MAGIC; 303 sb->s_op = &mqueue_super_ops; 304 305 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL); 306 if (IS_ERR(inode)) 307 return PTR_ERR(inode); 308 309 sb->s_root = d_make_root(inode); 310 if (!sb->s_root) 311 return -ENOMEM; 312 return 0; 313} 314 315static struct dentry *mqueue_mount(struct file_system_type *fs_type, 316 int flags, const char *dev_name, 317 void *data) 318{ 319 if (!(flags & MS_KERNMOUNT)) 320 data = current->nsproxy->ipc_ns; 321 return mount_ns(fs_type, flags, data, mqueue_fill_super); 322} 323 324static void init_once(void *foo) 325{ 326 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo; 327 328 inode_init_once(&p->vfs_inode); 329} 330 331static struct inode *mqueue_alloc_inode(struct super_block *sb) 332{ 333 struct mqueue_inode_info *ei; 334 335 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL); 336 if (!ei) 337 return NULL; 338 return &ei->vfs_inode; 339} 340 341static void mqueue_i_callback(struct rcu_head *head) 342{ 343 struct inode *inode = container_of(head, struct inode, i_rcu); 344 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode)); 345} 346 347static void mqueue_destroy_inode(struct inode *inode) 348{ 349 call_rcu(&inode->i_rcu, mqueue_i_callback); 350} 351 352static void mqueue_evict_inode(struct inode *inode) 353{ 354 struct mqueue_inode_info *info; 355 struct user_struct *user; 356 unsigned long mq_bytes, mq_treesize; 357 struct ipc_namespace *ipc_ns; 358 struct msg_msg *msg; 359 360 clear_inode(inode); 361 362 if (S_ISDIR(inode->i_mode)) 363 return; 364 365 ipc_ns = get_ns_from_inode(inode); 366 info = MQUEUE_I(inode); 367 spin_lock(&info->lock); 368 while ((msg = msg_get(info)) != NULL) 369 free_msg(msg); 370 spin_unlock(&info->lock); 371 372 /* Total amount of bytes accounted for the mqueue */ 373 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + 374 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * 375 sizeof(struct posix_msg_tree_node); 376 377 mq_bytes = mq_treesize + (info->attr.mq_maxmsg * 378 info->attr.mq_msgsize); 379 380 user = info->user; 381 if (user) { 382 spin_lock(&mq_lock); 383 user->mq_bytes -= mq_bytes; 384 /* 385 * get_ns_from_inode() ensures that the 386 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns 387 * to which we now hold a reference, or it is NULL. 388 * We can't put it here under mq_lock, though. 389 */ 390 if (ipc_ns) 391 ipc_ns->mq_queues_count--; 392 spin_unlock(&mq_lock); 393 free_uid(user); 394 } 395 if (ipc_ns) 396 put_ipc_ns(ipc_ns); 397} 398 399static int mqueue_create(struct inode *dir, struct dentry *dentry, 400 umode_t mode, struct nameidata *nd) 401{ 402 struct inode *inode; 403 struct mq_attr *attr = dentry->d_fsdata; 404 int error; 405 struct ipc_namespace *ipc_ns; 406 407 spin_lock(&mq_lock); 408 ipc_ns = __get_ns_from_inode(dir); 409 if (!ipc_ns) { 410 error = -EACCES; 411 goto out_unlock; 412 } 413 if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX || 414 (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max && 415 !capable(CAP_SYS_RESOURCE))) { 416 error = -ENOSPC; 417 goto out_unlock; 418 } 419 ipc_ns->mq_queues_count++; 420 spin_unlock(&mq_lock); 421 422 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr); 423 if (IS_ERR(inode)) { 424 error = PTR_ERR(inode); 425 spin_lock(&mq_lock); 426 ipc_ns->mq_queues_count--; 427 goto out_unlock; 428 } 429 430 put_ipc_ns(ipc_ns); 431 dir->i_size += DIRENT_SIZE; 432 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME; 433 434 d_instantiate(dentry, inode); 435 dget(dentry); 436 return 0; 437out_unlock: 438 spin_unlock(&mq_lock); 439 if (ipc_ns) 440 put_ipc_ns(ipc_ns); 441 return error; 442} 443 444static int mqueue_unlink(struct inode *dir, struct dentry *dentry) 445{ 446 struct inode *inode = dentry->d_inode; 447 448 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME; 449 dir->i_size -= DIRENT_SIZE; 450 drop_nlink(inode); 451 dput(dentry); 452 return 0; 453} 454 455/* 456* This is routine for system read from queue file. 457* To avoid mess with doing here some sort of mq_receive we allow 458* to read only queue size & notification info (the only values 459* that are interesting from user point of view and aren't accessible 460* through std routines) 461*/ 462static ssize_t mqueue_read_file(struct file *filp, char __user *u_data, 463 size_t count, loff_t *off) 464{ 465 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode); 466 char buffer[FILENT_SIZE]; 467 ssize_t ret; 468 469 spin_lock(&info->lock); 470 snprintf(buffer, sizeof(buffer), 471 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n", 472 info->qsize, 473 info->notify_owner ? info->notify.sigev_notify : 0, 474 (info->notify_owner && 475 info->notify.sigev_notify == SIGEV_SIGNAL) ? 476 info->notify.sigev_signo : 0, 477 pid_vnr(info->notify_owner)); 478 spin_unlock(&info->lock); 479 buffer[sizeof(buffer)-1] = '\0'; 480 481 ret = simple_read_from_buffer(u_data, count, off, buffer, 482 strlen(buffer)); 483 if (ret <= 0) 484 return ret; 485 486 filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME; 487 return ret; 488} 489 490static int mqueue_flush_file(struct file *filp, fl_owner_t id) 491{ 492 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode); 493 494 spin_lock(&info->lock); 495 if (task_tgid(current) == info->notify_owner) 496 remove_notification(info); 497 498 spin_unlock(&info->lock); 499 return 0; 500} 501 502static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab) 503{ 504 struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode); 505 int retval = 0; 506 507 poll_wait(filp, &info->wait_q, poll_tab); 508 509 spin_lock(&info->lock); 510 if (info->attr.mq_curmsgs) 511 retval = POLLIN | POLLRDNORM; 512 513 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg) 514 retval |= POLLOUT | POLLWRNORM; 515 spin_unlock(&info->lock); 516 517 return retval; 518} 519 520/* Adds current to info->e_wait_q[sr] before element with smaller prio */ 521static void wq_add(struct mqueue_inode_info *info, int sr, 522 struct ext_wait_queue *ewp) 523{ 524 struct ext_wait_queue *walk; 525 526 ewp->task = current; 527 528 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) { 529 if (walk->task->static_prio <= current->static_prio) { 530 list_add_tail(&ewp->list, &walk->list); 531 return; 532 } 533 } 534 list_add_tail(&ewp->list, &info->e_wait_q[sr].list); 535} 536 537/* 538 * Puts current task to sleep. Caller must hold queue lock. After return 539 * lock isn't held. 540 * sr: SEND or RECV 541 */ 542static int wq_sleep(struct mqueue_inode_info *info, int sr, 543 ktime_t *timeout, struct ext_wait_queue *ewp) 544{ 545 int retval; 546 signed long time; 547 548 wq_add(info, sr, ewp); 549 550 for (;;) { 551 set_current_state(TASK_INTERRUPTIBLE); 552 553 spin_unlock(&info->lock); 554 time = schedule_hrtimeout_range_clock(timeout, 0, 555 HRTIMER_MODE_ABS, CLOCK_REALTIME); 556 557 while (ewp->state == STATE_PENDING) 558 cpu_relax(); 559 560 if (ewp->state == STATE_READY) { 561 retval = 0; 562 goto out; 563 } 564 spin_lock(&info->lock); 565 if (ewp->state == STATE_READY) { 566 retval = 0; 567 goto out_unlock; 568 } 569 if (signal_pending(current)) { 570 retval = -ERESTARTSYS; 571 break; 572 } 573 if (time == 0) { 574 retval = -ETIMEDOUT; 575 break; 576 } 577 } 578 list_del(&ewp->list); 579out_unlock: 580 spin_unlock(&info->lock); 581out: 582 return retval; 583} 584 585/* 586 * Returns waiting task that should be serviced first or NULL if none exists 587 */ 588static struct ext_wait_queue *wq_get_first_waiter( 589 struct mqueue_inode_info *info, int sr) 590{ 591 struct list_head *ptr; 592 593 ptr = info->e_wait_q[sr].list.prev; 594 if (ptr == &info->e_wait_q[sr].list) 595 return NULL; 596 return list_entry(ptr, struct ext_wait_queue, list); 597} 598 599 600static inline void set_cookie(struct sk_buff *skb, char code) 601{ 602 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code; 603} 604 605/* 606 * The next function is only to split too long sys_mq_timedsend 607 */ 608static void __do_notify(struct mqueue_inode_info *info) 609{ 610 /* notification 611 * invoked when there is registered process and there isn't process 612 * waiting synchronously for message AND state of queue changed from 613 * empty to not empty. Here we are sure that no one is waiting 614 * synchronously. */ 615 if (info->notify_owner && 616 info->attr.mq_curmsgs == 1) { 617 struct siginfo sig_i; 618 switch (info->notify.sigev_notify) { 619 case SIGEV_NONE: 620 break; 621 case SIGEV_SIGNAL: 622 /* sends signal */ 623 624 sig_i.si_signo = info->notify.sigev_signo; 625 sig_i.si_errno = 0; 626 sig_i.si_code = SI_MESGQ; 627 sig_i.si_value = info->notify.sigev_value; 628 /* map current pid/uid into info->owner's namespaces */ 629 rcu_read_lock(); 630 sig_i.si_pid = task_tgid_nr_ns(current, 631 ns_of_pid(info->notify_owner)); 632 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid()); 633 rcu_read_unlock(); 634 635 kill_pid_info(info->notify.sigev_signo, 636 &sig_i, info->notify_owner); 637 break; 638 case SIGEV_THREAD: 639 set_cookie(info->notify_cookie, NOTIFY_WOKENUP); 640 netlink_sendskb(info->notify_sock, info->notify_cookie); 641 break; 642 } 643 /* after notification unregisters process */ 644 put_pid(info->notify_owner); 645 put_user_ns(info->notify_user_ns); 646 info->notify_owner = NULL; 647 info->notify_user_ns = NULL; 648 } 649 wake_up(&info->wait_q); 650} 651 652static int prepare_timeout(const struct timespec __user *u_abs_timeout, 653 ktime_t *expires, struct timespec *ts) 654{ 655 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec))) 656 return -EFAULT; 657 if (!timespec_valid(ts)) 658 return -EINVAL; 659 660 *expires = timespec_to_ktime(*ts); 661 return 0; 662} 663 664static void remove_notification(struct mqueue_inode_info *info) 665{ 666 if (info->notify_owner != NULL && 667 info->notify.sigev_notify == SIGEV_THREAD) { 668 set_cookie(info->notify_cookie, NOTIFY_REMOVED); 669 netlink_sendskb(info->notify_sock, info->notify_cookie); 670 } 671 put_pid(info->notify_owner); 672 put_user_ns(info->notify_user_ns); 673 info->notify_owner = NULL; 674 info->notify_user_ns = NULL; 675} 676 677static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr) 678{ 679 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0) 680 return 0; 681 if (capable(CAP_SYS_RESOURCE)) { 682 if (attr->mq_maxmsg > HARD_MSGMAX || 683 attr->mq_msgsize > HARD_MSGSIZEMAX) 684 return 0; 685 } else { 686 if (attr->mq_maxmsg > ipc_ns->mq_msg_max || 687 attr->mq_msgsize > ipc_ns->mq_msgsize_max) 688 return 0; 689 } 690 /* check for overflow */ 691 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg) 692 return 0; 693 if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize 694 + sizeof (struct msg_msg *))) < 695 (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize)) 696 return 0; 697 return 1; 698} 699 700/* 701 * Invoked when creating a new queue via sys_mq_open 702 */ 703static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir, 704 struct dentry *dentry, int oflag, umode_t mode, 705 struct mq_attr *attr) 706{ 707 const struct cred *cred = current_cred(); 708 struct file *result; 709 int ret; 710 711 if (attr) { 712 if (!mq_attr_ok(ipc_ns, attr)) { 713 ret = -EINVAL; 714 goto out; 715 } 716 /* store for use during create */ 717 dentry->d_fsdata = attr; 718 } 719 720 mode &= ~current_umask(); 721 ret = mnt_want_write(ipc_ns->mq_mnt); 722 if (ret) 723 goto out; 724 ret = vfs_create(dir->d_inode, dentry, mode, NULL); 725 dentry->d_fsdata = NULL; 726 if (ret) 727 goto out_drop_write; 728 729 result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred); 730 /* 731 * dentry_open() took a persistent mnt_want_write(), 732 * so we can now drop this one. 733 */ 734 mnt_drop_write(ipc_ns->mq_mnt); 735 return result; 736 737out_drop_write: 738 mnt_drop_write(ipc_ns->mq_mnt); 739out: 740 dput(dentry); 741 mntput(ipc_ns->mq_mnt); 742 return ERR_PTR(ret); 743} 744 745/* Opens existing queue */ 746static struct file *do_open(struct ipc_namespace *ipc_ns, 747 struct dentry *dentry, int oflag) 748{ 749 int ret; 750 const struct cred *cred = current_cred(); 751 752 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE, 753 MAY_READ | MAY_WRITE }; 754 755 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) { 756 ret = -EINVAL; 757 goto err; 758 } 759 760 if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) { 761 ret = -EACCES; 762 goto err; 763 } 764 765 return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred); 766 767err: 768 dput(dentry); 769 mntput(ipc_ns->mq_mnt); 770 return ERR_PTR(ret); 771} 772 773SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode, 774 struct mq_attr __user *, u_attr) 775{ 776 struct dentry *dentry; 777 struct file *filp; 778 char *name; 779 struct mq_attr attr; 780 int fd, error; 781 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns; 782 783 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr))) 784 return -EFAULT; 785 786 audit_mq_open(oflag, mode, u_attr ? &attr : NULL); 787 788 if (IS_ERR(name = getname(u_name))) 789 return PTR_ERR(name); 790 791 fd = get_unused_fd_flags(O_CLOEXEC); 792 if (fd < 0) 793 goto out_putname; 794 795 mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex); 796 dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name)); 797 if (IS_ERR(dentry)) { 798 error = PTR_ERR(dentry); 799 goto out_putfd; 800 } 801 mntget(ipc_ns->mq_mnt); 802 803 if (oflag & O_CREAT) { 804 if (dentry->d_inode) { /* entry already exists */ 805 audit_inode(name, dentry); 806 if (oflag & O_EXCL) { 807 error = -EEXIST; 808 goto out; 809 } 810 filp = do_open(ipc_ns, dentry, oflag); 811 } else { 812 filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root, 813 dentry, oflag, mode, 814 u_attr ? &attr : NULL); 815 } 816 } else { 817 if (!dentry->d_inode) { 818 error = -ENOENT; 819 goto out; 820 } 821 audit_inode(name, dentry); 822 filp = do_open(ipc_ns, dentry, oflag); 823 } 824 825 if (IS_ERR(filp)) { 826 error = PTR_ERR(filp); 827 goto out_putfd; 828 } 829 830 fd_install(fd, filp); 831 goto out_upsem; 832 833out: 834 dput(dentry); 835 mntput(ipc_ns->mq_mnt); 836out_putfd: 837 put_unused_fd(fd); 838 fd = error; 839out_upsem: 840 mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex); 841out_putname: 842 putname(name); 843 return fd; 844} 845 846SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name) 847{ 848 int err; 849 char *name; 850 struct dentry *dentry; 851 struct inode *inode = NULL; 852 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns; 853 854 name = getname(u_name); 855 if (IS_ERR(name)) 856 return PTR_ERR(name); 857 858 mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex, 859 I_MUTEX_PARENT); 860 dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name)); 861 if (IS_ERR(dentry)) { 862 err = PTR_ERR(dentry); 863 goto out_unlock; 864 } 865 866 if (!dentry->d_inode) { 867 err = -ENOENT; 868 goto out_err; 869 } 870 871 inode = dentry->d_inode; 872 if (inode) 873 ihold(inode); 874 err = mnt_want_write(ipc_ns->mq_mnt); 875 if (err) 876 goto out_err; 877 err = vfs_unlink(dentry->d_parent->d_inode, dentry); 878 mnt_drop_write(ipc_ns->mq_mnt); 879out_err: 880 dput(dentry); 881 882out_unlock: 883 mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex); 884 putname(name); 885 if (inode) 886 iput(inode); 887 888 return err; 889} 890 891/* Pipelined send and receive functions. 892 * 893 * If a receiver finds no waiting message, then it registers itself in the 894 * list of waiting receivers. A sender checks that list before adding the new 895 * message into the message array. If there is a waiting receiver, then it 896 * bypasses the message array and directly hands the message over to the 897 * receiver. 898 * The receiver accepts the message and returns without grabbing the queue 899 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers 900 * are necessary. The same algorithm is used for sysv semaphores, see 901 * ipc/sem.c for more details. 902 * 903 * The same algorithm is used for senders. 904 */ 905 906/* pipelined_send() - send a message directly to the task waiting in 907 * sys_mq_timedreceive() (without inserting message into a queue). 908 */ 909static inline void pipelined_send(struct mqueue_inode_info *info, 910 struct msg_msg *message, 911 struct ext_wait_queue *receiver) 912{ 913 receiver->msg = message; 914 list_del(&receiver->list); 915 receiver->state = STATE_PENDING; 916 wake_up_process(receiver->task); 917 smp_wmb(); 918 receiver->state = STATE_READY; 919} 920 921/* pipelined_receive() - if there is task waiting in sys_mq_timedsend() 922 * gets its message and put to the queue (we have one free place for sure). */ 923static inline void pipelined_receive(struct mqueue_inode_info *info) 924{ 925 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND); 926 927 if (!sender) { 928 /* for poll */ 929 wake_up_interruptible(&info->wait_q); 930 return; 931 } 932 if (msg_insert(sender->msg, info)) 933 return; 934 list_del(&sender->list); 935 sender->state = STATE_PENDING; 936 wake_up_process(sender->task); 937 smp_wmb(); 938 sender->state = STATE_READY; 939} 940 941SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr, 942 size_t, msg_len, unsigned int, msg_prio, 943 const struct timespec __user *, u_abs_timeout) 944{ 945 struct file *filp; 946 struct inode *inode; 947 struct ext_wait_queue wait; 948 struct ext_wait_queue *receiver; 949 struct msg_msg *msg_ptr; 950 struct mqueue_inode_info *info; 951 ktime_t expires, *timeout = NULL; 952 struct timespec ts; 953 int ret; 954 955 if (u_abs_timeout) { 956 int res = prepare_timeout(u_abs_timeout, &expires, &ts); 957 if (res) 958 return res; 959 timeout = &expires; 960 } 961 962 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX)) 963 return -EINVAL; 964 965 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL); 966 967 filp = fget(mqdes); 968 if (unlikely(!filp)) { 969 ret = -EBADF; 970 goto out; 971 } 972 973 inode = filp->f_path.dentry->d_inode; 974 if (unlikely(filp->f_op != &mqueue_file_operations)) { 975 ret = -EBADF; 976 goto out_fput; 977 } 978 info = MQUEUE_I(inode); 979 audit_inode(NULL, filp->f_path.dentry); 980 981 if (unlikely(!(filp->f_mode & FMODE_WRITE))) { 982 ret = -EBADF; 983 goto out_fput; 984 } 985 986 if (unlikely(msg_len > info->attr.mq_msgsize)) { 987 ret = -EMSGSIZE; 988 goto out_fput; 989 } 990 991 /* First try to allocate memory, before doing anything with 992 * existing queues. */ 993 msg_ptr = load_msg(u_msg_ptr, msg_len); 994 if (IS_ERR(msg_ptr)) { 995 ret = PTR_ERR(msg_ptr); 996 goto out_fput; 997 } 998 msg_ptr->m_ts = msg_len; 999 msg_ptr->m_type = msg_prio; 1000 1001 spin_lock(&info->lock); 1002 1003 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) { 1004 if (filp->f_flags & O_NONBLOCK) { 1005 spin_unlock(&info->lock); 1006 ret = -EAGAIN; 1007 } else { 1008 wait.task = current; 1009 wait.msg = (void *) msg_ptr; 1010 wait.state = STATE_NONE; 1011 ret = wq_sleep(info, SEND, timeout, &wait); 1012 } 1013 if (ret < 0) 1014 free_msg(msg_ptr); 1015 } else { 1016 receiver = wq_get_first_waiter(info, RECV); 1017 if (receiver) { 1018 pipelined_send(info, msg_ptr, receiver); 1019 } else { 1020 /* adds message to the queue */ 1021 if (msg_insert(msg_ptr, info)) { 1022 free_msg(msg_ptr); 1023 ret = -ENOMEM; 1024 spin_unlock(&info->lock); 1025 goto out_fput; 1026 } 1027 __do_notify(info); 1028 } 1029 inode->i_atime = inode->i_mtime = inode->i_ctime = 1030 CURRENT_TIME; 1031 spin_unlock(&info->lock); 1032 ret = 0; 1033 } 1034out_fput: 1035 fput(filp); 1036out: 1037 return ret; 1038} 1039 1040SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr, 1041 size_t, msg_len, unsigned int __user *, u_msg_prio, 1042 const struct timespec __user *, u_abs_timeout) 1043{ 1044 ssize_t ret; 1045 struct msg_msg *msg_ptr; 1046 struct file *filp; 1047 struct inode *inode; 1048 struct mqueue_inode_info *info; 1049 struct ext_wait_queue wait; 1050 ktime_t expires, *timeout = NULL; 1051 struct timespec ts; 1052 1053 if (u_abs_timeout) { 1054 int res = prepare_timeout(u_abs_timeout, &expires, &ts); 1055 if (res) 1056 return res; 1057 timeout = &expires; 1058 } 1059 1060 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL); 1061 1062 filp = fget(mqdes); 1063 if (unlikely(!filp)) { 1064 ret = -EBADF; 1065 goto out; 1066 } 1067 1068 inode = filp->f_path.dentry->d_inode; 1069 if (unlikely(filp->f_op != &mqueue_file_operations)) { 1070 ret = -EBADF; 1071 goto out_fput; 1072 } 1073 info = MQUEUE_I(inode); 1074 audit_inode(NULL, filp->f_path.dentry); 1075 1076 if (unlikely(!(filp->f_mode & FMODE_READ))) { 1077 ret = -EBADF; 1078 goto out_fput; 1079 } 1080 1081 /* checks if buffer is big enough */ 1082 if (unlikely(msg_len < info->attr.mq_msgsize)) { 1083 ret = -EMSGSIZE; 1084 goto out_fput; 1085 } 1086 1087 spin_lock(&info->lock); 1088 if (info->attr.mq_curmsgs == 0) { 1089 if (filp->f_flags & O_NONBLOCK) { 1090 spin_unlock(&info->lock); 1091 ret = -EAGAIN; 1092 } else { 1093 wait.task = current; 1094 wait.state = STATE_NONE; 1095 ret = wq_sleep(info, RECV, timeout, &wait); 1096 msg_ptr = wait.msg; 1097 } 1098 } else { 1099 msg_ptr = msg_get(info); 1100 1101 inode->i_atime = inode->i_mtime = inode->i_ctime = 1102 CURRENT_TIME; 1103 1104 /* There is now free space in queue. */ 1105 pipelined_receive(info); 1106 spin_unlock(&info->lock); 1107 ret = 0; 1108 } 1109 if (ret == 0) { 1110 ret = msg_ptr->m_ts; 1111 1112 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) || 1113 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) { 1114 ret = -EFAULT; 1115 } 1116 free_msg(msg_ptr); 1117 } 1118out_fput: 1119 fput(filp); 1120out: 1121 return ret; 1122} 1123 1124/* 1125 * Notes: the case when user wants us to deregister (with NULL as pointer) 1126 * and he isn't currently owner of notification, will be silently discarded. 1127 * It isn't explicitly defined in the POSIX. 1128 */ 1129SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes, 1130 const struct sigevent __user *, u_notification) 1131{ 1132 int ret; 1133 struct file *filp; 1134 struct sock *sock; 1135 struct inode *inode; 1136 struct sigevent notification; 1137 struct mqueue_inode_info *info; 1138 struct sk_buff *nc; 1139 1140 if (u_notification) { 1141 if (copy_from_user(¬ification, u_notification, 1142 sizeof(struct sigevent))) 1143 return -EFAULT; 1144 } 1145 1146 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL); 1147 1148 nc = NULL; 1149 sock = NULL; 1150 if (u_notification != NULL) { 1151 if (unlikely(notification.sigev_notify != SIGEV_NONE && 1152 notification.sigev_notify != SIGEV_SIGNAL && 1153 notification.sigev_notify != SIGEV_THREAD)) 1154 return -EINVAL; 1155 if (notification.sigev_notify == SIGEV_SIGNAL && 1156 !valid_signal(notification.sigev_signo)) { 1157 return -EINVAL; 1158 } 1159 if (notification.sigev_notify == SIGEV_THREAD) { 1160 long timeo; 1161 1162 /* create the notify skb */ 1163 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL); 1164 if (!nc) { 1165 ret = -ENOMEM; 1166 goto out; 1167 } 1168 if (copy_from_user(nc->data, 1169 notification.sigev_value.sival_ptr, 1170 NOTIFY_COOKIE_LEN)) { 1171 ret = -EFAULT; 1172 goto out; 1173 } 1174 1175 /* TODO: add a header? */ 1176 skb_put(nc, NOTIFY_COOKIE_LEN); 1177 /* and attach it to the socket */ 1178retry: 1179 filp = fget(notification.sigev_signo); 1180 if (!filp) { 1181 ret = -EBADF; 1182 goto out; 1183 } 1184 sock = netlink_getsockbyfilp(filp); 1185 fput(filp); 1186 if (IS_ERR(sock)) { 1187 ret = PTR_ERR(sock); 1188 sock = NULL; 1189 goto out; 1190 } 1191 1192 timeo = MAX_SCHEDULE_TIMEOUT; 1193 ret = netlink_attachskb(sock, nc, &timeo, NULL); 1194 if (ret == 1) 1195 goto retry; 1196 if (ret) { 1197 sock = NULL; 1198 nc = NULL; 1199 goto out; 1200 } 1201 } 1202 } 1203 1204 filp = fget(mqdes); 1205 if (!filp) { 1206 ret = -EBADF; 1207 goto out; 1208 } 1209 1210 inode = filp->f_path.dentry->d_inode; 1211 if (unlikely(filp->f_op != &mqueue_file_operations)) { 1212 ret = -EBADF; 1213 goto out_fput; 1214 } 1215 info = MQUEUE_I(inode); 1216 1217 ret = 0; 1218 spin_lock(&info->lock); 1219 if (u_notification == NULL) { 1220 if (info->notify_owner == task_tgid(current)) { 1221 remove_notification(info); 1222 inode->i_atime = inode->i_ctime = CURRENT_TIME; 1223 } 1224 } else if (info->notify_owner != NULL) { 1225 ret = -EBUSY; 1226 } else { 1227 switch (notification.sigev_notify) { 1228 case SIGEV_NONE: 1229 info->notify.sigev_notify = SIGEV_NONE; 1230 break; 1231 case SIGEV_THREAD: 1232 info->notify_sock = sock; 1233 info->notify_cookie = nc; 1234 sock = NULL; 1235 nc = NULL; 1236 info->notify.sigev_notify = SIGEV_THREAD; 1237 break; 1238 case SIGEV_SIGNAL: 1239 info->notify.sigev_signo = notification.sigev_signo; 1240 info->notify.sigev_value = notification.sigev_value; 1241 info->notify.sigev_notify = SIGEV_SIGNAL; 1242 break; 1243 } 1244 1245 info->notify_owner = get_pid(task_tgid(current)); 1246 info->notify_user_ns = get_user_ns(current_user_ns()); 1247 inode->i_atime = inode->i_ctime = CURRENT_TIME; 1248 } 1249 spin_unlock(&info->lock); 1250out_fput: 1251 fput(filp); 1252out: 1253 if (sock) { 1254 netlink_detachskb(sock, nc); 1255 } else if (nc) { 1256 dev_kfree_skb(nc); 1257 } 1258 return ret; 1259} 1260 1261SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes, 1262 const struct mq_attr __user *, u_mqstat, 1263 struct mq_attr __user *, u_omqstat) 1264{ 1265 int ret; 1266 struct mq_attr mqstat, omqstat; 1267 struct file *filp; 1268 struct inode *inode; 1269 struct mqueue_inode_info *info; 1270 1271 if (u_mqstat != NULL) { 1272 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr))) 1273 return -EFAULT; 1274 if (mqstat.mq_flags & (~O_NONBLOCK)) 1275 return -EINVAL; 1276 } 1277 1278 filp = fget(mqdes); 1279 if (!filp) { 1280 ret = -EBADF; 1281 goto out; 1282 } 1283 1284 inode = filp->f_path.dentry->d_inode; 1285 if (unlikely(filp->f_op != &mqueue_file_operations)) { 1286 ret = -EBADF; 1287 goto out_fput; 1288 } 1289 info = MQUEUE_I(inode); 1290 1291 spin_lock(&info->lock); 1292 1293 omqstat = info->attr; 1294 omqstat.mq_flags = filp->f_flags & O_NONBLOCK; 1295 if (u_mqstat) { 1296 audit_mq_getsetattr(mqdes, &mqstat); 1297 spin_lock(&filp->f_lock); 1298 if (mqstat.mq_flags & O_NONBLOCK) 1299 filp->f_flags |= O_NONBLOCK; 1300 else 1301 filp->f_flags &= ~O_NONBLOCK; 1302 spin_unlock(&filp->f_lock); 1303 1304 inode->i_atime = inode->i_ctime = CURRENT_TIME; 1305 } 1306 1307 spin_unlock(&info->lock); 1308 1309 ret = 0; 1310 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat, 1311 sizeof(struct mq_attr))) 1312 ret = -EFAULT; 1313 1314out_fput: 1315 fput(filp); 1316out: 1317 return ret; 1318} 1319 1320static const struct inode_operations mqueue_dir_inode_operations = { 1321 .lookup = simple_lookup, 1322 .create = mqueue_create, 1323 .unlink = mqueue_unlink, 1324}; 1325 1326static const struct file_operations mqueue_file_operations = { 1327 .flush = mqueue_flush_file, 1328 .poll = mqueue_poll_file, 1329 .read = mqueue_read_file, 1330 .llseek = default_llseek, 1331}; 1332 1333static const struct super_operations mqueue_super_ops = { 1334 .alloc_inode = mqueue_alloc_inode, 1335 .destroy_inode = mqueue_destroy_inode, 1336 .evict_inode = mqueue_evict_inode, 1337 .statfs = simple_statfs, 1338}; 1339 1340static struct file_system_type mqueue_fs_type = { 1341 .name = "mqueue", 1342 .mount = mqueue_mount, 1343 .kill_sb = kill_litter_super, 1344}; 1345 1346int mq_init_ns(struct ipc_namespace *ns) 1347{ 1348 ns->mq_queues_count = 0; 1349 ns->mq_queues_max = DFLT_QUEUESMAX; 1350 ns->mq_msg_max = DFLT_MSGMAX; 1351 ns->mq_msgsize_max = DFLT_MSGSIZEMAX; 1352 ns->mq_msg_default = DFLT_MSG; 1353 ns->mq_msgsize_default = DFLT_MSGSIZE; 1354 1355 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns); 1356 if (IS_ERR(ns->mq_mnt)) { 1357 int err = PTR_ERR(ns->mq_mnt); 1358 ns->mq_mnt = NULL; 1359 return err; 1360 } 1361 return 0; 1362} 1363 1364void mq_clear_sbinfo(struct ipc_namespace *ns) 1365{ 1366 ns->mq_mnt->mnt_sb->s_fs_info = NULL; 1367} 1368 1369void mq_put_mnt(struct ipc_namespace *ns) 1370{ 1371 kern_unmount(ns->mq_mnt); 1372} 1373 1374static int __init init_mqueue_fs(void) 1375{ 1376 int error; 1377 1378 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache", 1379 sizeof(struct mqueue_inode_info), 0, 1380 SLAB_HWCACHE_ALIGN, init_once); 1381 if (mqueue_inode_cachep == NULL) 1382 return -ENOMEM; 1383 1384 /* ignore failures - they are not fatal */ 1385 mq_sysctl_table = mq_register_sysctl_table(); 1386 1387 error = register_filesystem(&mqueue_fs_type); 1388 if (error) 1389 goto out_sysctl; 1390 1391 spin_lock_init(&mq_lock); 1392 1393 error = mq_init_ns(&init_ipc_ns); 1394 if (error) 1395 goto out_filesystem; 1396 1397 return 0; 1398 1399out_filesystem: 1400 unregister_filesystem(&mqueue_fs_type); 1401out_sysctl: 1402 if (mq_sysctl_table) 1403 unregister_sysctl_table(mq_sysctl_table); 1404 kmem_cache_destroy(mqueue_inode_cachep); 1405 return error; 1406} 1407 1408__initcall(init_mqueue_fs); 1409