Commit | Line | Data |
---|---|---|
7eb95156 PE |
1 | /* |
2 | * inet fragments management | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Authors: Pavel Emelyanov <xemul@openvz.org> | |
10 | * Started as consolidation of ipv4/ip_fragment.c, | |
11 | * ipv6/reassembly. and ipv6 nf conntrack reassembly | |
12 | */ | |
13 | ||
14 | #include <linux/list.h> | |
15 | #include <linux/spinlock.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/timer.h> | |
18 | #include <linux/mm.h> | |
321a3a99 | 19 | #include <linux/random.h> |
1e4b8287 PE |
20 | #include <linux/skbuff.h> |
21 | #include <linux/rtnetlink.h> | |
7eb95156 PE |
22 | |
23 | #include <net/inet_frag.h> | |
24 | ||
321a3a99 PE |
25 | static void inet_frag_secret_rebuild(unsigned long dummy) |
26 | { | |
27 | struct inet_frags *f = (struct inet_frags *)dummy; | |
28 | unsigned long now = jiffies; | |
29 | int i; | |
30 | ||
31 | write_lock(&f->lock); | |
32 | get_random_bytes(&f->rnd, sizeof(u32)); | |
33 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { | |
34 | struct inet_frag_queue *q; | |
35 | struct hlist_node *p, *n; | |
36 | ||
37 | hlist_for_each_entry_safe(q, p, n, &f->hash[i], list) { | |
38 | unsigned int hval = f->hashfn(q); | |
39 | ||
40 | if (hval != i) { | |
41 | hlist_del(&q->list); | |
42 | ||
43 | /* Relink to new hash chain. */ | |
44 | hlist_add_head(&q->list, &f->hash[hval]); | |
45 | } | |
46 | } | |
47 | } | |
48 | write_unlock(&f->lock); | |
49 | ||
50 | mod_timer(&f->secret_timer, now + f->ctl->secret_interval); | |
51 | } | |
52 | ||
7eb95156 PE |
53 | void inet_frags_init(struct inet_frags *f) |
54 | { | |
55 | int i; | |
56 | ||
57 | for (i = 0; i < INETFRAGS_HASHSZ; i++) | |
58 | INIT_HLIST_HEAD(&f->hash[i]); | |
59 | ||
60 | INIT_LIST_HEAD(&f->lru_list); | |
61 | rwlock_init(&f->lock); | |
62 | ||
63 | f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ | |
64 | (jiffies ^ (jiffies >> 6))); | |
65 | ||
66 | f->nqueues = 0; | |
67 | atomic_set(&f->mem, 0); | |
68 | ||
b24b8a24 PE |
69 | setup_timer(&f->secret_timer, inet_frag_secret_rebuild, |
70 | (unsigned long)f); | |
321a3a99 PE |
71 | f->secret_timer.expires = jiffies + f->ctl->secret_interval; |
72 | add_timer(&f->secret_timer); | |
7eb95156 PE |
73 | } |
74 | EXPORT_SYMBOL(inet_frags_init); | |
75 | ||
76 | void inet_frags_fini(struct inet_frags *f) | |
77 | { | |
321a3a99 | 78 | del_timer(&f->secret_timer); |
7eb95156 PE |
79 | } |
80 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d PE |
81 | |
82 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) | |
83 | { | |
84 | write_lock(&f->lock); | |
85 | hlist_del(&fq->list); | |
86 | list_del(&fq->lru_list); | |
87 | f->nqueues--; | |
88 | write_unlock(&f->lock); | |
89 | } | |
90 | ||
91 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
92 | { | |
93 | if (del_timer(&fq->timer)) | |
94 | atomic_dec(&fq->refcnt); | |
95 | ||
96 | if (!(fq->last_in & COMPLETE)) { | |
97 | fq_unlink(fq, f); | |
98 | atomic_dec(&fq->refcnt); | |
99 | fq->last_in |= COMPLETE; | |
100 | } | |
101 | } | |
102 | ||
103 | EXPORT_SYMBOL(inet_frag_kill); | |
1e4b8287 PE |
104 | |
105 | static inline void frag_kfree_skb(struct inet_frags *f, struct sk_buff *skb, | |
106 | int *work) | |
107 | { | |
108 | if (work) | |
109 | *work -= skb->truesize; | |
110 | ||
111 | atomic_sub(skb->truesize, &f->mem); | |
112 | if (f->skb_free) | |
113 | f->skb_free(skb); | |
114 | kfree_skb(skb); | |
115 | } | |
116 | ||
117 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, | |
118 | int *work) | |
119 | { | |
120 | struct sk_buff *fp; | |
121 | ||
122 | BUG_TRAP(q->last_in & COMPLETE); | |
123 | BUG_TRAP(del_timer(&q->timer) == 0); | |
124 | ||
125 | /* Release all fragment data. */ | |
126 | fp = q->fragments; | |
127 | while (fp) { | |
128 | struct sk_buff *xp = fp->next; | |
129 | ||
130 | frag_kfree_skb(f, fp, work); | |
131 | fp = xp; | |
132 | } | |
133 | ||
134 | if (work) | |
135 | *work -= f->qsize; | |
136 | atomic_sub(f->qsize, &f->mem); | |
137 | ||
c9547709 PE |
138 | if (f->destructor) |
139 | f->destructor(q); | |
140 | kfree(q); | |
1e4b8287 PE |
141 | |
142 | } | |
143 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 PE |
144 | |
145 | int inet_frag_evictor(struct inet_frags *f) | |
146 | { | |
147 | struct inet_frag_queue *q; | |
148 | int work, evicted = 0; | |
149 | ||
150 | work = atomic_read(&f->mem) - f->ctl->low_thresh; | |
151 | while (work > 0) { | |
152 | read_lock(&f->lock); | |
153 | if (list_empty(&f->lru_list)) { | |
154 | read_unlock(&f->lock); | |
155 | break; | |
156 | } | |
157 | ||
158 | q = list_first_entry(&f->lru_list, | |
159 | struct inet_frag_queue, lru_list); | |
160 | atomic_inc(&q->refcnt); | |
161 | read_unlock(&f->lock); | |
162 | ||
163 | spin_lock(&q->lock); | |
164 | if (!(q->last_in & COMPLETE)) | |
165 | inet_frag_kill(q, f); | |
166 | spin_unlock(&q->lock); | |
167 | ||
168 | if (atomic_dec_and_test(&q->refcnt)) | |
169 | inet_frag_destroy(q, f, &work); | |
170 | evicted++; | |
171 | } | |
172 | ||
173 | return evicted; | |
174 | } | |
175 | EXPORT_SYMBOL(inet_frag_evictor); | |
2588fe1d | 176 | |
c6fda282 | 177 | static struct inet_frag_queue *inet_frag_intern(struct inet_frag_queue *qp_in, |
48d60056 | 178 | struct inet_frags *f, unsigned int hash, void *arg) |
2588fe1d PE |
179 | { |
180 | struct inet_frag_queue *qp; | |
181 | #ifdef CONFIG_SMP | |
182 | struct hlist_node *n; | |
183 | #endif | |
184 | ||
185 | write_lock(&f->lock); | |
186 | #ifdef CONFIG_SMP | |
187 | /* With SMP race we have to recheck hash table, because | |
188 | * such entry could be created on other cpu, while we | |
189 | * promoted read lock to write lock. | |
190 | */ | |
191 | hlist_for_each_entry(qp, n, &f->hash[hash], list) { | |
48d60056 | 192 | if (f->match(qp, arg)) { |
2588fe1d PE |
193 | atomic_inc(&qp->refcnt); |
194 | write_unlock(&f->lock); | |
195 | qp_in->last_in |= COMPLETE; | |
196 | inet_frag_put(qp_in, f); | |
197 | return qp; | |
198 | } | |
199 | } | |
200 | #endif | |
201 | qp = qp_in; | |
202 | if (!mod_timer(&qp->timer, jiffies + f->ctl->timeout)) | |
203 | atomic_inc(&qp->refcnt); | |
204 | ||
205 | atomic_inc(&qp->refcnt); | |
206 | hlist_add_head(&qp->list, &f->hash[hash]); | |
207 | list_add_tail(&qp->lru_list, &f->lru_list); | |
208 | f->nqueues++; | |
209 | write_unlock(&f->lock); | |
210 | return qp; | |
211 | } | |
e521db9d | 212 | |
c6fda282 | 213 | static struct inet_frag_queue *inet_frag_alloc(struct inet_frags *f, void *arg) |
e521db9d PE |
214 | { |
215 | struct inet_frag_queue *q; | |
216 | ||
217 | q = kzalloc(f->qsize, GFP_ATOMIC); | |
218 | if (q == NULL) | |
219 | return NULL; | |
220 | ||
c6fda282 | 221 | f->constructor(q, arg); |
e521db9d PE |
222 | atomic_add(f->qsize, &f->mem); |
223 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); | |
224 | spin_lock_init(&q->lock); | |
225 | atomic_set(&q->refcnt, 1); | |
226 | ||
227 | return q; | |
228 | } | |
c6fda282 | 229 | |
abd6523d PE |
230 | static struct inet_frag_queue *inet_frag_create(struct inet_frags *f, |
231 | void *arg, unsigned int hash) | |
c6fda282 PE |
232 | { |
233 | struct inet_frag_queue *q; | |
234 | ||
235 | q = inet_frag_alloc(f, arg); | |
236 | if (q == NULL) | |
237 | return NULL; | |
238 | ||
48d60056 | 239 | return inet_frag_intern(q, f, hash, arg); |
c6fda282 | 240 | } |
abd6523d PE |
241 | |
242 | struct inet_frag_queue *inet_frag_find(struct inet_frags *f, void *key, | |
243 | unsigned int hash) | |
244 | { | |
245 | struct inet_frag_queue *q; | |
246 | struct hlist_node *n; | |
247 | ||
248 | read_lock(&f->lock); | |
249 | hlist_for_each_entry(q, n, &f->hash[hash], list) { | |
250 | if (f->match(q, key)) { | |
251 | atomic_inc(&q->refcnt); | |
252 | read_unlock(&f->lock); | |
253 | return q; | |
254 | } | |
255 | } | |
256 | read_unlock(&f->lock); | |
257 | ||
258 | return inet_frag_create(f, key, hash); | |
259 | } | |
260 | EXPORT_SYMBOL(inet_frag_find); |