Merge tag 'platform-drivers-x86-v5.2-1' of git://git.infradead.org/linux-platform...
[linux-2.6-block.git] / ipc / mqueue.c
CommitLineData
1da177e4
LT
1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
f66e928b 5 * Michal Wronski (michal.wronski@gmail.com)
1da177e4
LT
6 *
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
239521f3 9 * Manfred Spraul (manfred@colorfullife.com)
1da177e4 10 *
20ca73bc
GW
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
12 *
1da177e4
LT
13 * This file is released under the GPL.
14 */
15
c59ede7b 16#include <linux/capability.h>
1da177e4
LT
17#include <linux/init.h>
18#include <linux/pagemap.h>
19#include <linux/file.h>
20#include <linux/mount.h>
935c6912 21#include <linux/fs_context.h>
1da177e4
LT
22#include <linux/namei.h>
23#include <linux/sysctl.h>
24#include <linux/poll.h>
25#include <linux/mqueue.h>
26#include <linux/msg.h>
27#include <linux/skbuff.h>
5b5c4d1a 28#include <linux/vmalloc.h>
1da177e4
LT
29#include <linux/netlink.h>
30#include <linux/syscalls.h>
20ca73bc 31#include <linux/audit.h>
7ed20e1a 32#include <linux/signal.h>
5f921ae9 33#include <linux/mutex.h>
b488893a
PE
34#include <linux/nsproxy.h>
35#include <linux/pid.h>
614b84cf 36#include <linux/ipc_namespace.h>
6b550f94 37#include <linux/user_namespace.h>
5a0e3ad6 38#include <linux/slab.h>
84f001e1 39#include <linux/sched/wake_q.h>
3f07c014 40#include <linux/sched/signal.h>
8703e8a4 41#include <linux/sched/user.h>
5f921ae9 42
1da177e4
LT
43#include <net/sock.h>
44#include "util.h"
45
935c6912
DH
46struct mqueue_fs_context {
47 struct ipc_namespace *ipc_ns;
48};
49
1da177e4
LT
50#define MQUEUE_MAGIC 0x19800202
51#define DIRENT_SIZE 20
52#define FILENT_SIZE 80
53
54#define SEND 0
55#define RECV 1
56
57#define STATE_NONE 0
fa6004ad 58#define STATE_READY 1
1da177e4 59
d6629859
DL
60struct posix_msg_tree_node {
61 struct rb_node rb_node;
62 struct list_head msg_list;
63 int priority;
64};
65
1da177e4
LT
66struct ext_wait_queue { /* queue of sleeping tasks */
67 struct task_struct *task;
68 struct list_head list;
69 struct msg_msg *msg; /* ptr of loaded message */
70 int state; /* one of STATE_* values */
71};
72
73struct mqueue_inode_info {
74 spinlock_t lock;
75 struct inode vfs_inode;
76 wait_queue_head_t wait_q;
77
d6629859 78 struct rb_root msg_tree;
ce2d52cc 79 struct posix_msg_tree_node *node_cache;
1da177e4
LT
80 struct mq_attr attr;
81
82 struct sigevent notify;
239521f3 83 struct pid *notify_owner;
6f9ac6d9 84 struct user_namespace *notify_user_ns;
338cec32 85 struct user_struct *user; /* user who created, for accounting */
1da177e4
LT
86 struct sock *notify_sock;
87 struct sk_buff *notify_cookie;
88
89 /* for tasks waiting for free space and messages, respectively */
90 struct ext_wait_queue e_wait_q[2];
91
92 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
93};
94
935c6912 95static struct file_system_type mqueue_fs_type;
92e1d5be 96static const struct inode_operations mqueue_dir_inode_operations;
9a32144e 97static const struct file_operations mqueue_file_operations;
b87221de 98static const struct super_operations mqueue_super_ops;
935c6912 99static const struct fs_context_operations mqueue_fs_context_ops;
1da177e4
LT
100static void remove_notification(struct mqueue_inode_info *info);
101
e18b890b 102static struct kmem_cache *mqueue_inode_cachep;
1da177e4 103
239521f3 104static struct ctl_table_header *mq_sysctl_table;
1da177e4
LT
105
106static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
107{
108 return container_of(inode, struct mqueue_inode_info, vfs_inode);
109}
110
7eafd7c7
SH
111/*
112 * This routine should be called with the mq_lock held.
113 */
114static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
614b84cf 115{
7eafd7c7 116 return get_ipc_ns(inode->i_sb->s_fs_info);
614b84cf
SH
117}
118
7eafd7c7 119static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
614b84cf 120{
7eafd7c7
SH
121 struct ipc_namespace *ns;
122
123 spin_lock(&mq_lock);
124 ns = __get_ns_from_inode(inode);
125 spin_unlock(&mq_lock);
126 return ns;
614b84cf
SH
127}
128
d6629859
DL
129/* Auxiliary functions to manipulate messages' list */
130static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
131{
132 struct rb_node **p, *parent = NULL;
133 struct posix_msg_tree_node *leaf;
134
135 p = &info->msg_tree.rb_node;
136 while (*p) {
137 parent = *p;
138 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
139
140 if (likely(leaf->priority == msg->m_type))
141 goto insert_msg;
142 else if (msg->m_type < leaf->priority)
143 p = &(*p)->rb_left;
144 else
145 p = &(*p)->rb_right;
146 }
ce2d52cc
DL
147 if (info->node_cache) {
148 leaf = info->node_cache;
149 info->node_cache = NULL;
150 } else {
151 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
152 if (!leaf)
153 return -ENOMEM;
ce2d52cc 154 INIT_LIST_HEAD(&leaf->msg_list);
ce2d52cc 155 }
d6629859
DL
156 leaf->priority = msg->m_type;
157 rb_link_node(&leaf->rb_node, parent, p);
158 rb_insert_color(&leaf->rb_node, &info->msg_tree);
d6629859
DL
159insert_msg:
160 info->attr.mq_curmsgs++;
161 info->qsize += msg->m_ts;
162 list_add_tail(&msg->m_list, &leaf->msg_list);
163 return 0;
164}
165
166static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
167{
168 struct rb_node **p, *parent = NULL;
169 struct posix_msg_tree_node *leaf;
170 struct msg_msg *msg;
171
172try_again:
173 p = &info->msg_tree.rb_node;
174 while (*p) {
175 parent = *p;
176 /*
177 * During insert, low priorities go to the left and high to the
178 * right. On receive, we want the highest priorities first, so
179 * walk all the way to the right.
180 */
181 p = &(*p)->rb_right;
182 }
183 if (!parent) {
184 if (info->attr.mq_curmsgs) {
185 pr_warn_once("Inconsistency in POSIX message queue, "
186 "no tree element, but supposedly messages "
187 "should exist!\n");
188 info->attr.mq_curmsgs = 0;
189 }
190 return NULL;
191 }
192 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
ce2d52cc 193 if (unlikely(list_empty(&leaf->msg_list))) {
d6629859
DL
194 pr_warn_once("Inconsistency in POSIX message queue, "
195 "empty leaf node but we haven't implemented "
196 "lazy leaf delete!\n");
197 rb_erase(&leaf->rb_node, &info->msg_tree);
ce2d52cc 198 if (info->node_cache) {
ce2d52cc
DL
199 kfree(leaf);
200 } else {
201 info->node_cache = leaf;
202 }
d6629859
DL
203 goto try_again;
204 } else {
205 msg = list_first_entry(&leaf->msg_list,
206 struct msg_msg, m_list);
207 list_del(&msg->m_list);
208 if (list_empty(&leaf->msg_list)) {
209 rb_erase(&leaf->rb_node, &info->msg_tree);
ce2d52cc 210 if (info->node_cache) {
ce2d52cc
DL
211 kfree(leaf);
212 } else {
213 info->node_cache = leaf;
214 }
d6629859
DL
215 }
216 }
217 info->attr.mq_curmsgs--;
218 info->qsize -= msg->m_ts;
219 return msg;
220}
221
7eafd7c7 222static struct inode *mqueue_get_inode(struct super_block *sb,
1b9d5ff7 223 struct ipc_namespace *ipc_ns, umode_t mode,
7eafd7c7 224 struct mq_attr *attr)
1da177e4 225{
86a264ab 226 struct user_struct *u = current_user();
1da177e4 227 struct inode *inode;
d40dcdb0 228 int ret = -ENOMEM;
1da177e4
LT
229
230 inode = new_inode(sb);
04715206
JS
231 if (!inode)
232 goto err;
233
234 inode->i_ino = get_next_ino();
235 inode->i_mode = mode;
236 inode->i_uid = current_fsuid();
237 inode->i_gid = current_fsgid();
078cd827 238 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
04715206
JS
239
240 if (S_ISREG(mode)) {
241 struct mqueue_inode_info *info;
d6629859 242 unsigned long mq_bytes, mq_treesize;
04715206
JS
243
244 inode->i_fop = &mqueue_file_operations;
245 inode->i_size = FILENT_SIZE;
246 /* mqueue specific info */
247 info = MQUEUE_I(inode);
248 spin_lock_init(&info->lock);
249 init_waitqueue_head(&info->wait_q);
250 INIT_LIST_HEAD(&info->e_wait_q[0].list);
251 INIT_LIST_HEAD(&info->e_wait_q[1].list);
252 info->notify_owner = NULL;
6f9ac6d9 253 info->notify_user_ns = NULL;
04715206
JS
254 info->qsize = 0;
255 info->user = NULL; /* set when all is ok */
d6629859 256 info->msg_tree = RB_ROOT;
ce2d52cc 257 info->node_cache = NULL;
04715206 258 memset(&info->attr, 0, sizeof(info->attr));
cef0184c
KM
259 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
260 ipc_ns->mq_msg_default);
261 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
262 ipc_ns->mq_msgsize_default);
04715206
JS
263 if (attr) {
264 info->attr.mq_maxmsg = attr->mq_maxmsg;
265 info->attr.mq_msgsize = attr->mq_msgsize;
266 }
d6629859
DL
267 /*
268 * We used to allocate a static array of pointers and account
269 * the size of that array as well as one msg_msg struct per
270 * possible message into the queue size. That's no longer
271 * accurate as the queue is now an rbtree and will grow and
272 * shrink depending on usage patterns. We can, however, still
273 * account one msg_msg struct per message, but the nodes are
274 * allocated depending on priority usage, and most programs
275 * only use one, or a handful, of priorities. However, since
276 * this is pinned memory, we need to assume worst case, so
277 * that means the min(mq_maxmsg, max_priorities) * struct
278 * posix_msg_tree_node.
279 */
05c1b290
AV
280
281 ret = -EINVAL;
282 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
283 goto out_inode;
284 if (capable(CAP_SYS_RESOURCE)) {
285 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
286 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
287 goto out_inode;
288 } else {
289 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
290 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
291 goto out_inode;
292 }
293 ret = -EOVERFLOW;
294 /* check for overflow */
295 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
296 goto out_inode;
d6629859
DL
297 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
298 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
299 sizeof(struct posix_msg_tree_node);
05c1b290
AV
300 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
301 if (mq_bytes + mq_treesize < mq_bytes)
302 goto out_inode;
303 mq_bytes += mq_treesize;
04715206
JS
304 spin_lock(&mq_lock);
305 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
2a4e64b8 306 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
04715206
JS
307 spin_unlock(&mq_lock);
308 /* mqueue_evict_inode() releases info->messages */
d40dcdb0 309 ret = -EMFILE;
04715206 310 goto out_inode;
1da177e4 311 }
04715206
JS
312 u->mq_bytes += mq_bytes;
313 spin_unlock(&mq_lock);
314
315 /* all is ok */
316 info->user = get_uid(u);
317 } else if (S_ISDIR(mode)) {
318 inc_nlink(inode);
319 /* Some things misbehave if size == 0 on a directory */
320 inode->i_size = 2 * DIRENT_SIZE;
321 inode->i_op = &mqueue_dir_inode_operations;
322 inode->i_fop = &simple_dir_operations;
1da177e4 323 }
04715206 324
1da177e4
LT
325 return inode;
326out_inode:
1da177e4 327 iput(inode);
04715206 328err:
d40dcdb0 329 return ERR_PTR(ret);
1da177e4
LT
330}
331
935c6912 332static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
1da177e4
LT
333{
334 struct inode *inode;
cfb2f6f6 335 struct ipc_namespace *ns = sb->s_fs_info;
1da177e4 336
a2982cc9 337 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
09cbfeaf
KS
338 sb->s_blocksize = PAGE_SIZE;
339 sb->s_blocksize_bits = PAGE_SHIFT;
1da177e4
LT
340 sb->s_magic = MQUEUE_MAGIC;
341 sb->s_op = &mqueue_super_ops;
342
48fde701
AV
343 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
344 if (IS_ERR(inode))
345 return PTR_ERR(inode);
1da177e4 346
48fde701
AV
347 sb->s_root = d_make_root(inode);
348 if (!sb->s_root)
349 return -ENOMEM;
350 return 0;
1da177e4
LT
351}
352
935c6912 353static int mqueue_get_tree(struct fs_context *fc)
1da177e4 354{
935c6912
DH
355 struct mqueue_fs_context *ctx = fc->fs_private;
356
357 put_user_ns(fc->user_ns);
358 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
359 fc->s_fs_info = ctx->ipc_ns;
360 return vfs_get_super(fc, vfs_get_keyed_super, mqueue_fill_super);
361}
362
363static void mqueue_fs_context_free(struct fs_context *fc)
364{
365 struct mqueue_fs_context *ctx = fc->fs_private;
366
367 if (ctx->ipc_ns)
368 put_ipc_ns(ctx->ipc_ns);
369 kfree(ctx);
370}
371
372static int mqueue_init_fs_context(struct fs_context *fc)
373{
374 struct mqueue_fs_context *ctx;
375
376 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
377 if (!ctx)
378 return -ENOMEM;
379
380 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
381 fc->fs_private = ctx;
382 fc->ops = &mqueue_fs_context_ops;
383 return 0;
384}
385
386static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
387{
388 struct mqueue_fs_context *ctx;
389 struct fs_context *fc;
390 struct vfsmount *mnt;
391
392 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
393 if (IS_ERR(fc))
394 return ERR_CAST(fc);
395
396 ctx = fc->fs_private;
397 put_ipc_ns(ctx->ipc_ns);
398 ctx->ipc_ns = get_ipc_ns(ns);
399
400 mnt = fc_mount(fc);
401 put_fs_context(fc);
402 return mnt;
1da177e4
LT
403}
404
51cc5068 405static void init_once(void *foo)
1da177e4
LT
406{
407 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
408
a35afb83 409 inode_init_once(&p->vfs_inode);
1da177e4
LT
410}
411
412static struct inode *mqueue_alloc_inode(struct super_block *sb)
413{
414 struct mqueue_inode_info *ei;
415
e94b1766 416 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
1da177e4
LT
417 if (!ei)
418 return NULL;
419 return &ei->vfs_inode;
420}
421
015d7956 422static void mqueue_free_inode(struct inode *inode)
1da177e4
LT
423{
424 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
425}
426
6d8af64c 427static void mqueue_evict_inode(struct inode *inode)
1da177e4
LT
428{
429 struct mqueue_inode_info *info;
430 struct user_struct *user;
d6629859 431 unsigned long mq_bytes, mq_treesize;
7eafd7c7 432 struct ipc_namespace *ipc_ns;
d6629859 433 struct msg_msg *msg;
1da177e4 434
dbd5768f 435 clear_inode(inode);
6d8af64c
AV
436
437 if (S_ISDIR(inode->i_mode))
1da177e4 438 return;
6d8af64c 439
7eafd7c7 440 ipc_ns = get_ns_from_inode(inode);
1da177e4
LT
441 info = MQUEUE_I(inode);
442 spin_lock(&info->lock);
d6629859
DL
443 while ((msg = msg_get(info)) != NULL)
444 free_msg(msg);
ce2d52cc 445 kfree(info->node_cache);
1da177e4
LT
446 spin_unlock(&info->lock);
447
8834cf79 448 /* Total amount of bytes accounted for the mqueue */
d6629859
DL
449 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
450 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
451 sizeof(struct posix_msg_tree_node);
452
453 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
454 info->attr.mq_msgsize);
455
1da177e4
LT
456 user = info->user;
457 if (user) {
458 spin_lock(&mq_lock);
459 user->mq_bytes -= mq_bytes;
7eafd7c7
SH
460 /*
461 * get_ns_from_inode() ensures that the
462 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
463 * to which we now hold a reference, or it is NULL.
464 * We can't put it here under mq_lock, though.
465 */
466 if (ipc_ns)
467 ipc_ns->mq_queues_count--;
1da177e4
LT
468 spin_unlock(&mq_lock);
469 free_uid(user);
470 }
7eafd7c7
SH
471 if (ipc_ns)
472 put_ipc_ns(ipc_ns);
1da177e4
LT
473}
474
eecec19d 475static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
1da177e4 476{
eecec19d 477 struct inode *dir = dentry->d_parent->d_inode;
1da177e4 478 struct inode *inode;
eecec19d 479 struct mq_attr *attr = arg;
1da177e4 480 int error;
7eafd7c7 481 struct ipc_namespace *ipc_ns;
1da177e4
LT
482
483 spin_lock(&mq_lock);
7eafd7c7
SH
484 ipc_ns = __get_ns_from_inode(dir);
485 if (!ipc_ns) {
486 error = -EACCES;
487 goto out_unlock;
488 }
f3713fd9
DB
489
490 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
491 !capable(CAP_SYS_RESOURCE)) {
1da177e4 492 error = -ENOSPC;
614b84cf 493 goto out_unlock;
1da177e4 494 }
614b84cf 495 ipc_ns->mq_queues_count++;
1da177e4
LT
496 spin_unlock(&mq_lock);
497
7eafd7c7 498 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
d40dcdb0
JS
499 if (IS_ERR(inode)) {
500 error = PTR_ERR(inode);
1da177e4 501 spin_lock(&mq_lock);
614b84cf
SH
502 ipc_ns->mq_queues_count--;
503 goto out_unlock;
1da177e4
LT
504 }
505
7eafd7c7 506 put_ipc_ns(ipc_ns);
1da177e4 507 dir->i_size += DIRENT_SIZE;
078cd827 508 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
1da177e4
LT
509
510 d_instantiate(dentry, inode);
511 dget(dentry);
512 return 0;
614b84cf 513out_unlock:
1da177e4 514 spin_unlock(&mq_lock);
7eafd7c7
SH
515 if (ipc_ns)
516 put_ipc_ns(ipc_ns);
1da177e4
LT
517 return error;
518}
519
eecec19d
AV
520static int mqueue_create(struct inode *dir, struct dentry *dentry,
521 umode_t mode, bool excl)
522{
523 return mqueue_create_attr(dentry, mode, NULL);
524}
525
1da177e4
LT
526static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
527{
75c3cfa8 528 struct inode *inode = d_inode(dentry);
1da177e4 529
078cd827 530 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
1da177e4 531 dir->i_size -= DIRENT_SIZE;
239521f3
MS
532 drop_nlink(inode);
533 dput(dentry);
534 return 0;
1da177e4
LT
535}
536
537/*
538* This is routine for system read from queue file.
539* To avoid mess with doing here some sort of mq_receive we allow
540* to read only queue size & notification info (the only values
541* that are interesting from user point of view and aren't accessible
542* through std routines)
543*/
544static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
f1a43f93 545 size_t count, loff_t *off)
1da177e4 546{
496ad9aa 547 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4 548 char buffer[FILENT_SIZE];
f1a43f93 549 ssize_t ret;
1da177e4
LT
550
551 spin_lock(&info->lock);
552 snprintf(buffer, sizeof(buffer),
553 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
554 info->qsize,
555 info->notify_owner ? info->notify.sigev_notify : 0,
556 (info->notify_owner &&
557 info->notify.sigev_notify == SIGEV_SIGNAL) ?
558 info->notify.sigev_signo : 0,
6c5f3e7b 559 pid_vnr(info->notify_owner));
1da177e4
LT
560 spin_unlock(&info->lock);
561 buffer[sizeof(buffer)-1] = '\0';
1da177e4 562
f1a43f93
AM
563 ret = simple_read_from_buffer(u_data, count, off, buffer,
564 strlen(buffer));
565 if (ret <= 0)
566 return ret;
1da177e4 567
078cd827 568 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
f1a43f93 569 return ret;
1da177e4
LT
570}
571
75e1fcc0 572static int mqueue_flush_file(struct file *filp, fl_owner_t id)
1da177e4 573{
496ad9aa 574 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4
LT
575
576 spin_lock(&info->lock);
a03fcb73 577 if (task_tgid(current) == info->notify_owner)
1da177e4
LT
578 remove_notification(info);
579
580 spin_unlock(&info->lock);
581 return 0;
582}
583
9dd95748 584static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
1da177e4 585{
496ad9aa 586 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
9dd95748 587 __poll_t retval = 0;
1da177e4
LT
588
589 poll_wait(filp, &info->wait_q, poll_tab);
590
591 spin_lock(&info->lock);
592 if (info->attr.mq_curmsgs)
a9a08845 593 retval = EPOLLIN | EPOLLRDNORM;
1da177e4
LT
594
595 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
a9a08845 596 retval |= EPOLLOUT | EPOLLWRNORM;
1da177e4
LT
597 spin_unlock(&info->lock);
598
599 return retval;
600}
601
602/* Adds current to info->e_wait_q[sr] before element with smaller prio */
603static void wq_add(struct mqueue_inode_info *info, int sr,
604 struct ext_wait_queue *ewp)
605{
606 struct ext_wait_queue *walk;
607
608 ewp->task = current;
609
610 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
68e34f4e 611 if (walk->task->prio <= current->prio) {
1da177e4
LT
612 list_add_tail(&ewp->list, &walk->list);
613 return;
614 }
615 }
616 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
617}
618
619/*
620 * Puts current task to sleep. Caller must hold queue lock. After return
621 * lock isn't held.
622 * sr: SEND or RECV
623 */
624static int wq_sleep(struct mqueue_inode_info *info, int sr,
9ca7d8e6 625 ktime_t *timeout, struct ext_wait_queue *ewp)
eac0b1c3 626 __releases(&info->lock)
1da177e4
LT
627{
628 int retval;
629 signed long time;
630
631 wq_add(info, sr, ewp);
632
633 for (;;) {
fa6004ad 634 __set_current_state(TASK_INTERRUPTIBLE);
1da177e4
LT
635
636 spin_unlock(&info->lock);
32ea845d
WG
637 time = schedule_hrtimeout_range_clock(timeout, 0,
638 HRTIMER_MODE_ABS, CLOCK_REALTIME);
1da177e4 639
1da177e4
LT
640 if (ewp->state == STATE_READY) {
641 retval = 0;
642 goto out;
643 }
644 spin_lock(&info->lock);
645 if (ewp->state == STATE_READY) {
646 retval = 0;
647 goto out_unlock;
648 }
649 if (signal_pending(current)) {
650 retval = -ERESTARTSYS;
651 break;
652 }
653 if (time == 0) {
654 retval = -ETIMEDOUT;
655 break;
656 }
657 }
658 list_del(&ewp->list);
659out_unlock:
660 spin_unlock(&info->lock);
661out:
662 return retval;
663}
664
665/*
666 * Returns waiting task that should be serviced first or NULL if none exists
667 */
668static struct ext_wait_queue *wq_get_first_waiter(
669 struct mqueue_inode_info *info, int sr)
670{
671 struct list_head *ptr;
672
673 ptr = info->e_wait_q[sr].list.prev;
674 if (ptr == &info->e_wait_q[sr].list)
675 return NULL;
676 return list_entry(ptr, struct ext_wait_queue, list);
677}
678
1da177e4
LT
679
680static inline void set_cookie(struct sk_buff *skb, char code)
681{
239521f3 682 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
1da177e4
LT
683}
684
685/*
686 * The next function is only to split too long sys_mq_timedsend
687 */
688static void __do_notify(struct mqueue_inode_info *info)
689{
690 /* notification
691 * invoked when there is registered process and there isn't process
692 * waiting synchronously for message AND state of queue changed from
693 * empty to not empty. Here we are sure that no one is waiting
694 * synchronously. */
695 if (info->notify_owner &&
696 info->attr.mq_curmsgs == 1) {
ae7795bc 697 struct kernel_siginfo sig_i;
1da177e4
LT
698 switch (info->notify.sigev_notify) {
699 case SIGEV_NONE:
700 break;
701 case SIGEV_SIGNAL:
702 /* sends signal */
703
faf1f22b 704 clear_siginfo(&sig_i);
1da177e4
LT
705 sig_i.si_signo = info->notify.sigev_signo;
706 sig_i.si_errno = 0;
707 sig_i.si_code = SI_MESGQ;
708 sig_i.si_value = info->notify.sigev_value;
6b550f94
SH
709 /* map current pid/uid into info->owner's namespaces */
710 rcu_read_lock();
a6684999
SB
711 sig_i.si_pid = task_tgid_nr_ns(current,
712 ns_of_pid(info->notify_owner));
76b6db01 713 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
6b550f94 714 rcu_read_unlock();
1da177e4 715
a03fcb73
CLG
716 kill_pid_info(info->notify.sigev_signo,
717 &sig_i, info->notify_owner);
1da177e4
LT
718 break;
719 case SIGEV_THREAD:
720 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
7ee015e0 721 netlink_sendskb(info->notify_sock, info->notify_cookie);
1da177e4
LT
722 break;
723 }
724 /* after notification unregisters process */
a03fcb73 725 put_pid(info->notify_owner);
6f9ac6d9 726 put_user_ns(info->notify_user_ns);
a03fcb73 727 info->notify_owner = NULL;
6f9ac6d9 728 info->notify_user_ns = NULL;
1da177e4
LT
729 }
730 wake_up(&info->wait_q);
731}
732
21fc538d 733static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
b9047726 734 struct timespec64 *ts)
1da177e4 735{
b9047726 736 if (get_timespec64(ts, u_abs_timeout))
9ca7d8e6 737 return -EFAULT;
b9047726 738 if (!timespec64_valid(ts))
9ca7d8e6 739 return -EINVAL;
9ca7d8e6 740 return 0;
1da177e4
LT
741}
742
743static void remove_notification(struct mqueue_inode_info *info)
744{
a03fcb73 745 if (info->notify_owner != NULL &&
1da177e4
LT
746 info->notify.sigev_notify == SIGEV_THREAD) {
747 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
7ee015e0 748 netlink_sendskb(info->notify_sock, info->notify_cookie);
1da177e4 749 }
a03fcb73 750 put_pid(info->notify_owner);
6f9ac6d9 751 put_user_ns(info->notify_user_ns);
a03fcb73 752 info->notify_owner = NULL;
6f9ac6d9 753 info->notify_user_ns = NULL;
1da177e4
LT
754}
755
066cc813
AV
756static int prepare_open(struct dentry *dentry, int oflag, int ro,
757 umode_t mode, struct filename *name,
614b84cf 758 struct mq_attr *attr)
1da177e4 759{
745ca247
DH
760 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
761 MAY_READ | MAY_WRITE };
765927b2 762 int acc;
066cc813 763
9b20d7fc
AV
764 if (d_really_is_negative(dentry)) {
765 if (!(oflag & O_CREAT))
066cc813 766 return -ENOENT;
9b20d7fc
AV
767 if (ro)
768 return ro;
769 audit_inode_parent_hidden(name, dentry->d_parent);
770 return vfs_mkobj(dentry, mode & ~current_umask(),
771 mqueue_create_attr, attr);
066cc813 772 }
9b20d7fc
AV
773 /* it already existed */
774 audit_inode(name, dentry, 0);
775 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
776 return -EEXIST;
765927b2 777 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
af4a5372 778 return -EINVAL;
765927b2 779 acc = oflag2acc[oflag & O_ACCMODE];
066cc813 780 return inode_permission(d_inode(dentry), acc);
1da177e4
LT
781}
782
0d060606
AV
783static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
784 struct mq_attr *attr)
1da177e4 785{
cfb2f6f6
EB
786 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
787 struct dentry *root = mnt->mnt_root;
91a27b2a 788 struct filename *name;
a713fd7f 789 struct path path;
1da177e4 790 int fd, error;
312b90fb 791 int ro;
1da177e4 792
0d060606 793 audit_mq_open(oflag, mode, attr);
20ca73bc 794
1da177e4
LT
795 if (IS_ERR(name = getname(u_name)))
796 return PTR_ERR(name);
797
269f2134 798 fd = get_unused_fd_flags(O_CLOEXEC);
1da177e4
LT
799 if (fd < 0)
800 goto out_putname;
801
312b90fb 802 ro = mnt_want_write(mnt); /* we'll drop it in any case */
5955102c 803 inode_lock(d_inode(root));
91a27b2a 804 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
765927b2
AV
805 if (IS_ERR(path.dentry)) {
806 error = PTR_ERR(path.dentry);
4294a8ee 807 goto out_putfd;
1da177e4 808 }
312b90fb 809 path.mnt = mntget(mnt);
066cc813
AV
810 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
811 if (!error) {
812 struct file *file = dentry_open(&path, oflag, current_cred());
813 if (!IS_ERR(file))
814 fd_install(fd, file);
815 else
816 error = PTR_ERR(file);
7c7dce92 817 }
765927b2 818 path_put(&path);
7c7dce92 819out_putfd:
765927b2
AV
820 if (error) {
821 put_unused_fd(fd);
822 fd = error;
823 }
5955102c 824 inode_unlock(d_inode(root));
38d78e58
VD
825 if (!ro)
826 mnt_drop_write(mnt);
1da177e4
LT
827out_putname:
828 putname(name);
829 return fd;
830}
831
0d060606
AV
832SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
833 struct mq_attr __user *, u_attr)
834{
835 struct mq_attr attr;
836 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
837 return -EFAULT;
838
839 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
840}
841
d5460c99 842SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
1da177e4
LT
843{
844 int err;
91a27b2a 845 struct filename *name;
1da177e4
LT
846 struct dentry *dentry;
847 struct inode *inode = NULL;
7eafd7c7 848 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
312b90fb 849 struct vfsmount *mnt = ipc_ns->mq_mnt;
1da177e4
LT
850
851 name = getname(u_name);
852 if (IS_ERR(name))
853 return PTR_ERR(name);
854
79f6530c 855 audit_inode_parent_hidden(name, mnt->mnt_root);
312b90fb
AV
856 err = mnt_want_write(mnt);
857 if (err)
858 goto out_name;
5955102c 859 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
91a27b2a
JL
860 dentry = lookup_one_len(name->name, mnt->mnt_root,
861 strlen(name->name));
1da177e4
LT
862 if (IS_ERR(dentry)) {
863 err = PTR_ERR(dentry);
864 goto out_unlock;
865 }
866
75c3cfa8 867 inode = d_inode(dentry);
312b90fb
AV
868 if (!inode) {
869 err = -ENOENT;
870 } else {
7de9c6ee 871 ihold(inode);
75c3cfa8 872 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
312b90fb 873 }
1da177e4
LT
874 dput(dentry);
875
876out_unlock:
5955102c 877 inode_unlock(d_inode(mnt->mnt_root));
1da177e4
LT
878 if (inode)
879 iput(inode);
312b90fb
AV
880 mnt_drop_write(mnt);
881out_name:
882 putname(name);
1da177e4
LT
883
884 return err;
885}
886
887/* Pipelined send and receive functions.
888 *
889 * If a receiver finds no waiting message, then it registers itself in the
890 * list of waiting receivers. A sender checks that list before adding the new
891 * message into the message array. If there is a waiting receiver, then it
892 * bypasses the message array and directly hands the message over to the
fa6004ad
DB
893 * receiver. The receiver accepts the message and returns without grabbing the
894 * queue spinlock:
895 *
896 * - Set pointer to message.
897 * - Queue the receiver task for later wakeup (without the info->lock).
898 * - Update its state to STATE_READY. Now the receiver can continue.
899 * - Wake up the process after the lock is dropped. Should the process wake up
900 * before this wakeup (due to a timeout or a signal) it will either see
901 * STATE_READY and continue or acquire the lock to check the state again.
1da177e4
LT
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 */
fa6004ad
DB
909static inline void pipelined_send(struct wake_q_head *wake_q,
910 struct mqueue_inode_info *info,
1da177e4
LT
911 struct msg_msg *message,
912 struct ext_wait_queue *receiver)
913{
914 receiver->msg = message;
915 list_del(&receiver->list);
fa6004ad
DB
916 wake_q_add(wake_q, receiver->task);
917 /*
918 * Rely on the implicit cmpxchg barrier from wake_q_add such
919 * that we can ensure that updating receiver->state is the last
920 * write operation: As once set, the receiver can continue,
921 * and if we don't have the reference count from the wake_q,
922 * yet, at that point we can later have a use-after-free
923 * condition and bogus wakeup.
924 */
1da177e4
LT
925 receiver->state = STATE_READY;
926}
927
928/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
929 * gets its message and put to the queue (we have one free place for sure). */
fa6004ad
DB
930static inline void pipelined_receive(struct wake_q_head *wake_q,
931 struct mqueue_inode_info *info)
1da177e4
LT
932{
933 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
934
935 if (!sender) {
936 /* for poll */
937 wake_up_interruptible(&info->wait_q);
938 return;
939 }
d6629859
DL
940 if (msg_insert(sender->msg, info))
941 return;
fa6004ad 942
1da177e4 943 list_del(&sender->list);
fa6004ad 944 wake_q_add(wake_q, sender->task);
1da177e4
LT
945 sender->state = STATE_READY;
946}
947
0d060606
AV
948static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
949 size_t msg_len, unsigned int msg_prio,
b9047726 950 struct timespec64 *ts)
1da177e4 951{
2903ff01 952 struct fd f;
1da177e4
LT
953 struct inode *inode;
954 struct ext_wait_queue wait;
955 struct ext_wait_queue *receiver;
956 struct msg_msg *msg_ptr;
957 struct mqueue_inode_info *info;
9ca7d8e6 958 ktime_t expires, *timeout = NULL;
ce2d52cc 959 struct posix_msg_tree_node *new_leaf = NULL;
2903ff01 960 int ret = 0;
194a6b5b 961 DEFINE_WAKE_Q(wake_q);
1da177e4
LT
962
963 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
964 return -EINVAL;
965
0d060606 966 if (ts) {
b9047726 967 expires = timespec64_to_ktime(*ts);
0d060606
AV
968 timeout = &expires;
969 }
970
971 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1da177e4 972
2903ff01
AV
973 f = fdget(mqdes);
974 if (unlikely(!f.file)) {
8d8ffefa 975 ret = -EBADF;
1da177e4 976 goto out;
8d8ffefa 977 }
1da177e4 978
496ad9aa 979 inode = file_inode(f.file);
2903ff01 980 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 981 ret = -EBADF;
1da177e4 982 goto out_fput;
8d8ffefa 983 }
1da177e4 984 info = MQUEUE_I(inode);
9f45f5bf 985 audit_file(f.file);
1da177e4 986
2903ff01 987 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
8d8ffefa 988 ret = -EBADF;
1da177e4 989 goto out_fput;
8d8ffefa 990 }
1da177e4
LT
991
992 if (unlikely(msg_len > info->attr.mq_msgsize)) {
993 ret = -EMSGSIZE;
994 goto out_fput;
995 }
996
997 /* First try to allocate memory, before doing anything with
998 * existing queues. */
999 msg_ptr = load_msg(u_msg_ptr, msg_len);
1000 if (IS_ERR(msg_ptr)) {
1001 ret = PTR_ERR(msg_ptr);
1002 goto out_fput;
1003 }
1004 msg_ptr->m_ts = msg_len;
1005 msg_ptr->m_type = msg_prio;
1006
ce2d52cc
DL
1007 /*
1008 * msg_insert really wants us to have a valid, spare node struct so
1009 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1010 * fall back to that if necessary.
1011 */
1012 if (!info->node_cache)
1013 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1014
1da177e4
LT
1015 spin_lock(&info->lock);
1016
ce2d52cc
DL
1017 if (!info->node_cache && new_leaf) {
1018 /* Save our speculative allocation into the cache */
ce2d52cc
DL
1019 INIT_LIST_HEAD(&new_leaf->msg_list);
1020 info->node_cache = new_leaf;
ce2d52cc
DL
1021 new_leaf = NULL;
1022 } else {
1023 kfree(new_leaf);
1024 }
1025
1da177e4 1026 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
2903ff01 1027 if (f.file->f_flags & O_NONBLOCK) {
1da177e4 1028 ret = -EAGAIN;
1da177e4
LT
1029 } else {
1030 wait.task = current;
1031 wait.msg = (void *) msg_ptr;
1032 wait.state = STATE_NONE;
1033 ret = wq_sleep(info, SEND, timeout, &wait);
ce2d52cc
DL
1034 /*
1035 * wq_sleep must be called with info->lock held, and
1036 * returns with the lock released
1037 */
1038 goto out_free;
1da177e4 1039 }
1da177e4
LT
1040 } else {
1041 receiver = wq_get_first_waiter(info, RECV);
1042 if (receiver) {
fa6004ad 1043 pipelined_send(&wake_q, info, msg_ptr, receiver);
1da177e4
LT
1044 } else {
1045 /* adds message to the queue */
ce2d52cc
DL
1046 ret = msg_insert(msg_ptr, info);
1047 if (ret)
1048 goto out_unlock;
1da177e4
LT
1049 __do_notify(info);
1050 }
1051 inode->i_atime = inode->i_mtime = inode->i_ctime =
078cd827 1052 current_time(inode);
1da177e4 1053 }
ce2d52cc
DL
1054out_unlock:
1055 spin_unlock(&info->lock);
fa6004ad 1056 wake_up_q(&wake_q);
ce2d52cc
DL
1057out_free:
1058 if (ret)
1059 free_msg(msg_ptr);
1da177e4 1060out_fput:
2903ff01 1061 fdput(f);
1da177e4
LT
1062out:
1063 return ret;
1064}
1065
0d060606
AV
1066static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1067 size_t msg_len, unsigned int __user *u_msg_prio,
b9047726 1068 struct timespec64 *ts)
1da177e4 1069{
1da177e4
LT
1070 ssize_t ret;
1071 struct msg_msg *msg_ptr;
2903ff01 1072 struct fd f;
1da177e4
LT
1073 struct inode *inode;
1074 struct mqueue_inode_info *info;
1075 struct ext_wait_queue wait;
9ca7d8e6 1076 ktime_t expires, *timeout = NULL;
ce2d52cc 1077 struct posix_msg_tree_node *new_leaf = NULL;
1da177e4 1078
0d060606 1079 if (ts) {
b9047726 1080 expires = timespec64_to_ktime(*ts);
9ca7d8e6 1081 timeout = &expires;
c32c8af4 1082 }
20ca73bc 1083
0d060606 1084 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1da177e4 1085
2903ff01
AV
1086 f = fdget(mqdes);
1087 if (unlikely(!f.file)) {
8d8ffefa 1088 ret = -EBADF;
1da177e4 1089 goto out;
8d8ffefa 1090 }
1da177e4 1091
496ad9aa 1092 inode = file_inode(f.file);
2903ff01 1093 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1094 ret = -EBADF;
1da177e4 1095 goto out_fput;
8d8ffefa 1096 }
1da177e4 1097 info = MQUEUE_I(inode);
9f45f5bf 1098 audit_file(f.file);
1da177e4 1099
2903ff01 1100 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
8d8ffefa 1101 ret = -EBADF;
1da177e4 1102 goto out_fput;
8d8ffefa 1103 }
1da177e4
LT
1104
1105 /* checks if buffer is big enough */
1106 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1107 ret = -EMSGSIZE;
1108 goto out_fput;
1109 }
1110
ce2d52cc
DL
1111 /*
1112 * msg_insert really wants us to have a valid, spare node struct so
1113 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1114 * fall back to that if necessary.
1115 */
1116 if (!info->node_cache)
1117 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1118
1da177e4 1119 spin_lock(&info->lock);
ce2d52cc
DL
1120
1121 if (!info->node_cache && new_leaf) {
1122 /* Save our speculative allocation into the cache */
ce2d52cc
DL
1123 INIT_LIST_HEAD(&new_leaf->msg_list);
1124 info->node_cache = new_leaf;
ce2d52cc
DL
1125 } else {
1126 kfree(new_leaf);
1127 }
1128
1da177e4 1129 if (info->attr.mq_curmsgs == 0) {
2903ff01 1130 if (f.file->f_flags & O_NONBLOCK) {
1da177e4
LT
1131 spin_unlock(&info->lock);
1132 ret = -EAGAIN;
1da177e4
LT
1133 } else {
1134 wait.task = current;
1135 wait.state = STATE_NONE;
1136 ret = wq_sleep(info, RECV, timeout, &wait);
1137 msg_ptr = wait.msg;
1138 }
1139 } else {
194a6b5b 1140 DEFINE_WAKE_Q(wake_q);
fa6004ad 1141
1da177e4
LT
1142 msg_ptr = msg_get(info);
1143
1144 inode->i_atime = inode->i_mtime = inode->i_ctime =
078cd827 1145 current_time(inode);
1da177e4
LT
1146
1147 /* There is now free space in queue. */
fa6004ad 1148 pipelined_receive(&wake_q, info);
1da177e4 1149 spin_unlock(&info->lock);
fa6004ad 1150 wake_up_q(&wake_q);
1da177e4
LT
1151 ret = 0;
1152 }
1153 if (ret == 0) {
1154 ret = msg_ptr->m_ts;
1155
1156 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1157 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1158 ret = -EFAULT;
1159 }
1160 free_msg(msg_ptr);
1161 }
1162out_fput:
2903ff01 1163 fdput(f);
1da177e4
LT
1164out:
1165 return ret;
1166}
1167
0d060606
AV
1168SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1169 size_t, msg_len, unsigned int, msg_prio,
21fc538d 1170 const struct __kernel_timespec __user *, u_abs_timeout)
0d060606 1171{
b9047726 1172 struct timespec64 ts, *p = NULL;
0d060606
AV
1173 if (u_abs_timeout) {
1174 int res = prepare_timeout(u_abs_timeout, &ts);
1175 if (res)
1176 return res;
1177 p = &ts;
1178 }
1179 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1180}
1181
1182SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1183 size_t, msg_len, unsigned int __user *, u_msg_prio,
21fc538d 1184 const struct __kernel_timespec __user *, u_abs_timeout)
0d060606 1185{
b9047726 1186 struct timespec64 ts, *p = NULL;
0d060606
AV
1187 if (u_abs_timeout) {
1188 int res = prepare_timeout(u_abs_timeout, &ts);
1189 if (res)
1190 return res;
1191 p = &ts;
1192 }
1193 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1194}
1195
1da177e4
LT
1196/*
1197 * Notes: the case when user wants us to deregister (with NULL as pointer)
1198 * and he isn't currently owner of notification, will be silently discarded.
1199 * It isn't explicitly defined in the POSIX.
1200 */
0d060606 1201static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1da177e4 1202{
2903ff01
AV
1203 int ret;
1204 struct fd f;
1da177e4
LT
1205 struct sock *sock;
1206 struct inode *inode;
1da177e4
LT
1207 struct mqueue_inode_info *info;
1208 struct sk_buff *nc;
1209
0d060606 1210 audit_mq_notify(mqdes, notification);
1da177e4 1211
20114f71
AV
1212 nc = NULL;
1213 sock = NULL;
0d060606
AV
1214 if (notification != NULL) {
1215 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1216 notification->sigev_notify != SIGEV_SIGNAL &&
1217 notification->sigev_notify != SIGEV_THREAD))
1da177e4 1218 return -EINVAL;
0d060606
AV
1219 if (notification->sigev_notify == SIGEV_SIGNAL &&
1220 !valid_signal(notification->sigev_signo)) {
1da177e4
LT
1221 return -EINVAL;
1222 }
0d060606 1223 if (notification->sigev_notify == SIGEV_THREAD) {
c3d8d1e3
PM
1224 long timeo;
1225
1da177e4
LT
1226 /* create the notify skb */
1227 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
8d8ffefa
AGR
1228 if (!nc) {
1229 ret = -ENOMEM;
1da177e4 1230 goto out;
8d8ffefa 1231 }
1da177e4 1232 if (copy_from_user(nc->data,
0d060606 1233 notification->sigev_value.sival_ptr,
1da177e4 1234 NOTIFY_COOKIE_LEN)) {
8d8ffefa 1235 ret = -EFAULT;
1da177e4
LT
1236 goto out;
1237 }
1238
1239 /* TODO: add a header? */
1240 skb_put(nc, NOTIFY_COOKIE_LEN);
1241 /* and attach it to the socket */
1242retry:
0d060606 1243 f = fdget(notification->sigev_signo);
2903ff01 1244 if (!f.file) {
8d8ffefa 1245 ret = -EBADF;
1da177e4 1246 goto out;
8d8ffefa 1247 }
2903ff01
AV
1248 sock = netlink_getsockbyfilp(f.file);
1249 fdput(f);
1da177e4
LT
1250 if (IS_ERR(sock)) {
1251 ret = PTR_ERR(sock);
1252 sock = NULL;
1253 goto out;
1254 }
1255
c3d8d1e3 1256 timeo = MAX_SCHEDULE_TIMEOUT;
9457afee 1257 ret = netlink_attachskb(sock, nc, &timeo, NULL);
f991af3d
CW
1258 if (ret == 1) {
1259 sock = NULL;
8d8ffefa 1260 goto retry;
f991af3d 1261 }
1da177e4
LT
1262 if (ret) {
1263 sock = NULL;
1264 nc = NULL;
1265 goto out;
1266 }
1267 }
1268 }
1269
2903ff01
AV
1270 f = fdget(mqdes);
1271 if (!f.file) {
8d8ffefa 1272 ret = -EBADF;
1da177e4 1273 goto out;
8d8ffefa 1274 }
1da177e4 1275
496ad9aa 1276 inode = file_inode(f.file);
2903ff01 1277 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1278 ret = -EBADF;
1da177e4 1279 goto out_fput;
8d8ffefa 1280 }
1da177e4
LT
1281 info = MQUEUE_I(inode);
1282
1283 ret = 0;
1284 spin_lock(&info->lock);
0d060606 1285 if (notification == NULL) {
a03fcb73 1286 if (info->notify_owner == task_tgid(current)) {
1da177e4 1287 remove_notification(info);
078cd827 1288 inode->i_atime = inode->i_ctime = current_time(inode);
1da177e4 1289 }
a03fcb73 1290 } else if (info->notify_owner != NULL) {
1da177e4
LT
1291 ret = -EBUSY;
1292 } else {
0d060606 1293 switch (notification->sigev_notify) {
1da177e4
LT
1294 case SIGEV_NONE:
1295 info->notify.sigev_notify = SIGEV_NONE;
1296 break;
1297 case SIGEV_THREAD:
1298 info->notify_sock = sock;
1299 info->notify_cookie = nc;
1300 sock = NULL;
1301 nc = NULL;
1302 info->notify.sigev_notify = SIGEV_THREAD;
1303 break;
1304 case SIGEV_SIGNAL:
0d060606
AV
1305 info->notify.sigev_signo = notification->sigev_signo;
1306 info->notify.sigev_value = notification->sigev_value;
1da177e4
LT
1307 info->notify.sigev_notify = SIGEV_SIGNAL;
1308 break;
1309 }
a03fcb73
CLG
1310
1311 info->notify_owner = get_pid(task_tgid(current));
6f9ac6d9 1312 info->notify_user_ns = get_user_ns(current_user_ns());
078cd827 1313 inode->i_atime = inode->i_ctime = current_time(inode);
1da177e4
LT
1314 }
1315 spin_unlock(&info->lock);
1316out_fput:
2903ff01 1317 fdput(f);
1da177e4 1318out:
3ab08fe2 1319 if (sock)
1da177e4 1320 netlink_detachskb(sock, nc);
3ab08fe2 1321 else if (nc)
1da177e4 1322 dev_kfree_skb(nc);
3ab08fe2 1323
1da177e4
LT
1324 return ret;
1325}
1326
0d060606
AV
1327SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1328 const struct sigevent __user *, u_notification)
1329{
1330 struct sigevent n, *p = NULL;
1331 if (u_notification) {
1332 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1333 return -EFAULT;
1334 p = &n;
1335 }
1336 return do_mq_notify(mqdes, p);
1337}
1338
1339static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1da177e4 1340{
2903ff01 1341 struct fd f;
1da177e4
LT
1342 struct inode *inode;
1343 struct mqueue_inode_info *info;
1344
0d060606
AV
1345 if (new && (new->mq_flags & (~O_NONBLOCK)))
1346 return -EINVAL;
1da177e4 1347
2903ff01 1348 f = fdget(mqdes);
0d060606
AV
1349 if (!f.file)
1350 return -EBADF;
1da177e4 1351
2903ff01 1352 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
0d060606
AV
1353 fdput(f);
1354 return -EBADF;
8d8ffefa 1355 }
0d060606
AV
1356
1357 inode = file_inode(f.file);
1da177e4
LT
1358 info = MQUEUE_I(inode);
1359
1360 spin_lock(&info->lock);
1361
0d060606
AV
1362 if (old) {
1363 *old = info->attr;
1364 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1365 }
1366 if (new) {
1367 audit_mq_getsetattr(mqdes, new);
2903ff01 1368 spin_lock(&f.file->f_lock);
0d060606 1369 if (new->mq_flags & O_NONBLOCK)
2903ff01 1370 f.file->f_flags |= O_NONBLOCK;
1da177e4 1371 else
2903ff01
AV
1372 f.file->f_flags &= ~O_NONBLOCK;
1373 spin_unlock(&f.file->f_lock);
1da177e4 1374
078cd827 1375 inode->i_atime = inode->i_ctime = current_time(inode);
1da177e4
LT
1376 }
1377
1378 spin_unlock(&info->lock);
0d060606
AV
1379 fdput(f);
1380 return 0;
1381}
1da177e4 1382
0d060606
AV
1383SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1384 const struct mq_attr __user *, u_mqstat,
1385 struct mq_attr __user *, u_omqstat)
1386{
1387 int ret;
1388 struct mq_attr mqstat, omqstat;
1389 struct mq_attr *new = NULL, *old = NULL;
1da177e4 1390
0d060606
AV
1391 if (u_mqstat) {
1392 new = &mqstat;
1393 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1394 return -EFAULT;
1395 }
1396 if (u_omqstat)
1397 old = &omqstat;
1398
1399 ret = do_mq_getsetattr(mqdes, new, old);
1400 if (ret || !old)
1401 return ret;
1402
1403 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1404 return -EFAULT;
1405 return 0;
1406}
1407
1408#ifdef CONFIG_COMPAT
1409
1410struct compat_mq_attr {
1411 compat_long_t mq_flags; /* message queue flags */
1412 compat_long_t mq_maxmsg; /* maximum number of messages */
1413 compat_long_t mq_msgsize; /* maximum message size */
1414 compat_long_t mq_curmsgs; /* number of messages currently queued */
1415 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1416};
1417
1418static inline int get_compat_mq_attr(struct mq_attr *attr,
1419 const struct compat_mq_attr __user *uattr)
1420{
1421 struct compat_mq_attr v;
1422
1423 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1424 return -EFAULT;
1425
1426 memset(attr, 0, sizeof(*attr));
1427 attr->mq_flags = v.mq_flags;
1428 attr->mq_maxmsg = v.mq_maxmsg;
1429 attr->mq_msgsize = v.mq_msgsize;
1430 attr->mq_curmsgs = v.mq_curmsgs;
1431 return 0;
1432}
1433
1434static inline int put_compat_mq_attr(const struct mq_attr *attr,
1435 struct compat_mq_attr __user *uattr)
1436{
1437 struct compat_mq_attr v;
1438
1439 memset(&v, 0, sizeof(v));
1440 v.mq_flags = attr->mq_flags;
1441 v.mq_maxmsg = attr->mq_maxmsg;
1442 v.mq_msgsize = attr->mq_msgsize;
1443 v.mq_curmsgs = attr->mq_curmsgs;
1444 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1445 return -EFAULT;
1446 return 0;
1447}
1448
1449COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1450 int, oflag, compat_mode_t, mode,
1451 struct compat_mq_attr __user *, u_attr)
1452{
1453 struct mq_attr attr, *p = NULL;
1454 if (u_attr && oflag & O_CREAT) {
1455 p = &attr;
1456 if (get_compat_mq_attr(&attr, u_attr))
1457 return -EFAULT;
1458 }
1459 return do_mq_open(u_name, oflag, mode, p);
1460}
1461
b0d17578
AB
1462COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1463 const struct compat_sigevent __user *, u_notification)
1464{
1465 struct sigevent n, *p = NULL;
1466 if (u_notification) {
1467 if (get_compat_sigevent(&n, u_notification))
1468 return -EFAULT;
1469 if (n.sigev_notify == SIGEV_THREAD)
1470 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1471 p = &n;
1472 }
1473 return do_mq_notify(mqdes, p);
1474}
1475
1476COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1477 const struct compat_mq_attr __user *, u_mqstat,
1478 struct compat_mq_attr __user *, u_omqstat)
1479{
1480 int ret;
1481 struct mq_attr mqstat, omqstat;
1482 struct mq_attr *new = NULL, *old = NULL;
1483
1484 if (u_mqstat) {
1485 new = &mqstat;
1486 if (get_compat_mq_attr(new, u_mqstat))
1487 return -EFAULT;
1488 }
1489 if (u_omqstat)
1490 old = &omqstat;
1491
1492 ret = do_mq_getsetattr(mqdes, new, old);
1493 if (ret || !old)
1494 return ret;
1495
1496 if (put_compat_mq_attr(old, u_omqstat))
1497 return -EFAULT;
1498 return 0;
1499}
1500#endif
1501
1502#ifdef CONFIG_COMPAT_32BIT_TIME
9afc5eee 1503static int compat_prepare_timeout(const struct old_timespec32 __user *p,
b9047726 1504 struct timespec64 *ts)
0d060606 1505{
9afc5eee 1506 if (get_old_timespec32(ts, p))
0d060606 1507 return -EFAULT;
b9047726 1508 if (!timespec64_valid(ts))
0d060606
AV
1509 return -EINVAL;
1510 return 0;
1511}
1512
8dabe724
AB
1513SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1514 const char __user *, u_msg_ptr,
1515 unsigned int, msg_len, unsigned int, msg_prio,
1516 const struct old_timespec32 __user *, u_abs_timeout)
0d060606 1517{
b9047726 1518 struct timespec64 ts, *p = NULL;
0d060606
AV
1519 if (u_abs_timeout) {
1520 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1521 if (res)
1522 return res;
1523 p = &ts;
1524 }
1525 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1526}
1527
8dabe724
AB
1528SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1529 char __user *, u_msg_ptr,
1530 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1531 const struct old_timespec32 __user *, u_abs_timeout)
0d060606 1532{
b9047726 1533 struct timespec64 ts, *p = NULL;
0d060606
AV
1534 if (u_abs_timeout) {
1535 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1536 if (res)
1537 return res;
1538 p = &ts;
1539 }
1540 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1541}
0d060606 1542#endif
1da177e4 1543
92e1d5be 1544static const struct inode_operations mqueue_dir_inode_operations = {
1da177e4
LT
1545 .lookup = simple_lookup,
1546 .create = mqueue_create,
1547 .unlink = mqueue_unlink,
1548};
1549
9a32144e 1550static const struct file_operations mqueue_file_operations = {
1da177e4
LT
1551 .flush = mqueue_flush_file,
1552 .poll = mqueue_poll_file,
1553 .read = mqueue_read_file,
6038f373 1554 .llseek = default_llseek,
1da177e4
LT
1555};
1556
b87221de 1557static const struct super_operations mqueue_super_ops = {
1da177e4 1558 .alloc_inode = mqueue_alloc_inode,
015d7956 1559 .free_inode = mqueue_free_inode,
6d8af64c 1560 .evict_inode = mqueue_evict_inode,
1da177e4 1561 .statfs = simple_statfs,
1da177e4
LT
1562};
1563
935c6912
DH
1564static const struct fs_context_operations mqueue_fs_context_ops = {
1565 .free = mqueue_fs_context_free,
1566 .get_tree = mqueue_get_tree,
1567};
1568
1da177e4 1569static struct file_system_type mqueue_fs_type = {
935c6912
DH
1570 .name = "mqueue",
1571 .init_fs_context = mqueue_init_fs_context,
1572 .kill_sb = kill_litter_super,
1573 .fs_flags = FS_USERNS_MOUNT,
1da177e4
LT
1574};
1575
7eafd7c7
SH
1576int mq_init_ns(struct ipc_namespace *ns)
1577{
935c6912
DH
1578 struct vfsmount *m;
1579
7eafd7c7
SH
1580 ns->mq_queues_count = 0;
1581 ns->mq_queues_max = DFLT_QUEUESMAX;
1582 ns->mq_msg_max = DFLT_MSGMAX;
1583 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
cef0184c
KM
1584 ns->mq_msg_default = DFLT_MSG;
1585 ns->mq_msgsize_default = DFLT_MSGSIZE;
7eafd7c7 1586
935c6912
DH
1587 m = mq_create_mount(ns);
1588 if (IS_ERR(m))
1589 return PTR_ERR(m);
1590 ns->mq_mnt = m;
7eafd7c7
SH
1591 return 0;
1592}
1593
1594void mq_clear_sbinfo(struct ipc_namespace *ns)
1595{
cfb2f6f6 1596 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
7eafd7c7
SH
1597}
1598
1599void mq_put_mnt(struct ipc_namespace *ns)
1600{
cfb2f6f6 1601 kern_unmount(ns->mq_mnt);
7eafd7c7
SH
1602}
1603
1da177e4
LT
1604static int __init init_mqueue_fs(void)
1605{
1606 int error;
1607
1608 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1609 sizeof(struct mqueue_inode_info), 0,
5d097056 1610 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1da177e4
LT
1611 if (mqueue_inode_cachep == NULL)
1612 return -ENOMEM;
1613
2329e392 1614 /* ignore failures - they are not fatal */
bdc8e5f8 1615 mq_sysctl_table = mq_register_sysctl_table();
1da177e4
LT
1616
1617 error = register_filesystem(&mqueue_fs_type);
1618 if (error)
1619 goto out_sysctl;
1620
7eafd7c7
SH
1621 spin_lock_init(&mq_lock);
1622
6f686574
AV
1623 error = mq_init_ns(&init_ipc_ns);
1624 if (error)
1da177e4 1625 goto out_filesystem;
1da177e4 1626
1da177e4
LT
1627 return 0;
1628
1629out_filesystem:
1630 unregister_filesystem(&mqueue_fs_type);
1631out_sysctl:
1632 if (mq_sysctl_table)
1633 unregister_sysctl_table(mq_sysctl_table);
1a1d92c1 1634 kmem_cache_destroy(mqueue_inode_cachep);
1da177e4
LT
1635 return error;
1636}
1637
6d08a256 1638device_initcall(init_mqueue_fs);