1/*-
2 * Copyright (c) 2015 The FreeBSD Foundation
3 *
4 * This software was developed by Konstantin Belousov
5 * under sponsorship from the FreeBSD Foundation.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <sys/types.h>
30#include <sys/mman.h>
31#include <sys/queue.h>
32#include "namespace.h"
33#include <stdlib.h>
34#include "un-namespace.h"
35
36#include "thr_private.h"
37
38struct psh {
39	LIST_ENTRY(psh) link;
40	void *key;
41	void *val;
42};
43
44LIST_HEAD(pshared_hash_head, psh);
45#define	HASH_SIZE	128
46static struct pshared_hash_head pshared_hash[HASH_SIZE];
47#define	PSHARED_KEY_HASH(key)	(((unsigned long)(key) >> 8) % HASH_SIZE)
48/* XXXKIB: lock could be split to per-hash chain, if appears contested */
49static struct urwlock pshared_lock = DEFAULT_URWLOCK;
50static int page_size;
51
52void
53__thr_pshared_init(void)
54{
55	int i;
56
57	page_size = getpagesize();
58	THR_ASSERT(page_size >= THR_PAGE_SIZE_MIN,
59	    "THR_PAGE_SIZE_MIN is too large");
60
61	_thr_urwlock_init(&pshared_lock);
62	for (i = 0; i < HASH_SIZE; i++)
63		LIST_INIT(&pshared_hash[i]);
64}
65
66static void
67pshared_rlock(struct pthread *curthread)
68{
69
70	curthread->locklevel++;
71	_thr_rwl_rdlock(&pshared_lock);
72}
73
74static void
75pshared_wlock(struct pthread *curthread)
76{
77
78	curthread->locklevel++;
79	_thr_rwl_wrlock(&pshared_lock);
80}
81
82static void
83pshared_unlock(struct pthread *curthread)
84{
85
86	_thr_rwl_unlock(&pshared_lock);
87	curthread->locklevel--;
88	_thr_ast(curthread);
89}
90
91/*
92 * Among all processes sharing a lock only one executes
93 * pthread_lock_destroy().  Other processes still have the hash and
94 * mapped off-page.
95 *
96 * Mitigate the problem by checking the liveness of all hashed keys
97 * periodically.  Right now this is executed on each
98 * pthread_lock_destroy(), but may be done less often if found to be
99 * too time-consuming.
100 */
101static void
102pshared_gc(struct pthread *curthread)
103{
104	struct pshared_hash_head *hd;
105	struct psh *h, *h1;
106	int error, i;
107
108	pshared_wlock(curthread);
109	for (i = 0; i < HASH_SIZE; i++) {
110		hd = &pshared_hash[i];
111		LIST_FOREACH_SAFE(h, hd, link, h1) {
112			error = _umtx_op(NULL, UMTX_OP_SHM, UMTX_SHM_ALIVE,
113			    h->val, NULL);
114			if (error == 0)
115				continue;
116			LIST_REMOVE(h, link);
117			munmap(h->val, page_size);
118			free(h);
119		}
120	}
121	pshared_unlock(curthread);
122}
123
124static void *
125pshared_lookup(void *key)
126{
127	struct pshared_hash_head *hd;
128	struct psh *h;
129
130	hd = &pshared_hash[PSHARED_KEY_HASH(key)];
131	LIST_FOREACH(h, hd, link) {
132		if (h->key == key)
133			return (h->val);
134	}
135	return (NULL);
136}
137
138static int
139pshared_insert(void *key, void **val)
140{
141	struct pshared_hash_head *hd;
142	struct psh *h;
143
144	hd = &pshared_hash[PSHARED_KEY_HASH(key)];
145	LIST_FOREACH(h, hd, link) {
146		/*
147		 * When the key already exists in the hash, we should
148		 * return either the new (just mapped) or old (hashed)
149		 * val, and the other val should be unmapped to avoid
150		 * address space leak.
151		 *
152		 * If two threads perform lock of the same object
153		 * which is not yet stored in the pshared_hash, then
154		 * the val already inserted by the first thread should
155		 * be returned, and the second val freed (order is by
156		 * the pshared_lock()).  Otherwise, if we unmap the
157		 * value obtained from the hash, the first thread
158		 * might operate on an unmapped off-page object.
159		 *
160		 * There is still an issue: if hashed key was unmapped
161		 * and then other page is mapped at the same key
162		 * address, the hash would return the old val.  I
163		 * decided to handle the race of simultaneous hash
164		 * insertion, leaving the unlikely remap problem
165		 * unaddressed.
166		 */
167		if (h->key == key) {
168			if (h->val != *val) {
169				munmap(*val, page_size);
170				*val = h->val;
171			}
172			return (1);
173		}
174	}
175
176	h = malloc(sizeof(*h));
177	if (h == NULL)
178		return (0);
179	h->key = key;
180	h->val = *val;
181	LIST_INSERT_HEAD(hd, h, link);
182	return (1);
183}
184
185static void *
186pshared_remove(void *key)
187{
188	struct pshared_hash_head *hd;
189	struct psh *h;
190	void *val;
191
192	hd = &pshared_hash[PSHARED_KEY_HASH(key)];
193	LIST_FOREACH(h, hd, link) {
194		if (h->key == key) {
195			LIST_REMOVE(h, link);
196			val = h->val;
197			free(h);
198			return (val);
199		}
200	}
201	return (NULL);
202}
203
204static void
205pshared_clean(void *key, void *val)
206{
207
208	if (val != NULL)
209		munmap(val, page_size);
210	_umtx_op(NULL, UMTX_OP_SHM, UMTX_SHM_DESTROY, key, NULL);
211}
212
213static void
214pshared_destroy(struct pthread *curthread, void *key)
215{
216	void *val;
217
218	pshared_wlock(curthread);
219	val = pshared_remove(key);
220	pshared_unlock(curthread);
221	pshared_clean(key, val);
222}
223
224void *
225__thr_pshared_offpage(void *key, int doalloc)
226{
227	struct pthread *curthread;
228	void *res;
229	int fd, ins_done;
230
231	curthread = _get_curthread();
232	if (doalloc) {
233		pshared_destroy(curthread, key);
234		res = NULL;
235	} else {
236		pshared_rlock(curthread);
237		res = pshared_lookup(key);
238		pshared_unlock(curthread);
239		if (res != NULL)
240			return (res);
241	}
242	fd = _umtx_op(NULL, UMTX_OP_SHM, doalloc ? UMTX_SHM_CREAT :
243	    UMTX_SHM_LOOKUP, key, NULL);
244	if (fd == -1)
245		return (NULL);
246	res = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
247	close(fd);
248	if (res == MAP_FAILED)
249		return (NULL);
250	pshared_wlock(curthread);
251	ins_done = pshared_insert(key, &res);
252	pshared_unlock(curthread);
253	if (!ins_done) {
254		pshared_clean(key, res);
255		res = NULL;
256	}
257	return (res);
258}
259
260void
261__thr_pshared_destroy(void *key)
262{
263	struct pthread *curthread;
264
265	curthread = _get_curthread();
266	pshared_destroy(curthread, key);
267	pshared_gc(curthread);
268}
269
270void
271__thr_pshared_atfork_pre(void)
272{
273
274	_thr_rwl_rdlock(&pshared_lock);
275}
276
277void
278__thr_pshared_atfork_post(void)
279{
280
281	_thr_rwl_unlock(&pshared_lock);
282}
283