]>
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 | ||
b24b8a24 PE |
66 | setup_timer(&f->secret_timer, inet_frag_secret_rebuild, |
67 | (unsigned long)f); | |
321a3a99 PE |
68 | f->secret_timer.expires = jiffies + f->ctl->secret_interval; |
69 | add_timer(&f->secret_timer); | |
7eb95156 PE |
70 | } |
71 | EXPORT_SYMBOL(inet_frags_init); | |
72 | ||
e5a2bb84 PE |
73 | void inet_frags_init_net(struct netns_frags *nf) |
74 | { | |
75 | nf->nqueues = 0; | |
6ddc0822 | 76 | atomic_set(&nf->mem, 0); |
e5a2bb84 PE |
77 | } |
78 | EXPORT_SYMBOL(inet_frags_init_net); | |
79 | ||
7eb95156 PE |
80 | void inet_frags_fini(struct inet_frags *f) |
81 | { | |
321a3a99 | 82 | del_timer(&f->secret_timer); |
7eb95156 PE |
83 | } |
84 | EXPORT_SYMBOL(inet_frags_fini); | |
277e650d PE |
85 | |
86 | static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) | |
87 | { | |
88 | write_lock(&f->lock); | |
89 | hlist_del(&fq->list); | |
90 | list_del(&fq->lru_list); | |
e5a2bb84 | 91 | fq->net->nqueues--; |
277e650d PE |
92 | write_unlock(&f->lock); |
93 | } | |
94 | ||
95 | void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) | |
96 | { | |
97 | if (del_timer(&fq->timer)) | |
98 | atomic_dec(&fq->refcnt); | |
99 | ||
100 | if (!(fq->last_in & COMPLETE)) { | |
101 | fq_unlink(fq, f); | |
102 | atomic_dec(&fq->refcnt); | |
103 | fq->last_in |= COMPLETE; | |
104 | } | |
105 | } | |
106 | ||
107 | EXPORT_SYMBOL(inet_frag_kill); | |
1e4b8287 | 108 | |
6ddc0822 PE |
109 | static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, |
110 | struct sk_buff *skb, int *work) | |
1e4b8287 PE |
111 | { |
112 | if (work) | |
113 | *work -= skb->truesize; | |
114 | ||
6ddc0822 | 115 | atomic_sub(skb->truesize, &nf->mem); |
1e4b8287 PE |
116 | if (f->skb_free) |
117 | f->skb_free(skb); | |
118 | kfree_skb(skb); | |
119 | } | |
120 | ||
121 | void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, | |
122 | int *work) | |
123 | { | |
124 | struct sk_buff *fp; | |
6ddc0822 | 125 | struct netns_frags *nf; |
1e4b8287 PE |
126 | |
127 | BUG_TRAP(q->last_in & COMPLETE); | |
128 | BUG_TRAP(del_timer(&q->timer) == 0); | |
129 | ||
130 | /* Release all fragment data. */ | |
131 | fp = q->fragments; | |
6ddc0822 | 132 | nf = q->net; |
1e4b8287 PE |
133 | while (fp) { |
134 | struct sk_buff *xp = fp->next; | |
135 | ||
6ddc0822 | 136 | frag_kfree_skb(nf, f, fp, work); |
1e4b8287 PE |
137 | fp = xp; |
138 | } | |
139 | ||
140 | if (work) | |
141 | *work -= f->qsize; | |
6ddc0822 | 142 | atomic_sub(f->qsize, &nf->mem); |
1e4b8287 | 143 | |
c9547709 PE |
144 | if (f->destructor) |
145 | f->destructor(q); | |
146 | kfree(q); | |
1e4b8287 PE |
147 | |
148 | } | |
149 | EXPORT_SYMBOL(inet_frag_destroy); | |
8e7999c4 | 150 | |
6ddc0822 | 151 | int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f) |
8e7999c4 PE |
152 | { |
153 | struct inet_frag_queue *q; | |
154 | int work, evicted = 0; | |
155 | ||
6ddc0822 | 156 | work = atomic_read(&nf->mem) - f->ctl->low_thresh; |
8e7999c4 PE |
157 | while (work > 0) { |
158 | read_lock(&f->lock); | |
159 | if (list_empty(&f->lru_list)) { | |
160 | read_unlock(&f->lock); | |
161 | break; | |
162 | } | |
163 | ||
164 | q = list_first_entry(&f->lru_list, | |
165 | struct inet_frag_queue, lru_list); | |
166 | atomic_inc(&q->refcnt); | |
167 | read_unlock(&f->lock); | |
168 | ||
169 | spin_lock(&q->lock); | |
170 | if (!(q->last_in & COMPLETE)) | |
171 | inet_frag_kill(q, f); | |
172 | spin_unlock(&q->lock); | |
173 | ||
174 | if (atomic_dec_and_test(&q->refcnt)) | |
175 | inet_frag_destroy(q, f, &work); | |
176 | evicted++; | |
177 | } | |
178 | ||
179 | return evicted; | |
180 | } | |
181 | EXPORT_SYMBOL(inet_frag_evictor); | |
2588fe1d | 182 | |
ac18e750 PE |
183 | static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, |
184 | struct inet_frag_queue *qp_in, struct inet_frags *f, | |
185 | unsigned int hash, void *arg) | |
2588fe1d PE |
186 | { |
187 | struct inet_frag_queue *qp; | |
188 | #ifdef CONFIG_SMP | |
189 | struct hlist_node *n; | |
190 | #endif | |
191 | ||
192 | write_lock(&f->lock); | |
193 | #ifdef CONFIG_SMP | |
194 | /* With SMP race we have to recheck hash table, because | |
195 | * such entry could be created on other cpu, while we | |
196 | * promoted read lock to write lock. | |
197 | */ | |
198 | hlist_for_each_entry(qp, n, &f->hash[hash], list) { | |
ac18e750 | 199 | if (qp->net == nf && f->match(qp, arg)) { |
2588fe1d PE |
200 | atomic_inc(&qp->refcnt); |
201 | write_unlock(&f->lock); | |
202 | qp_in->last_in |= COMPLETE; | |
203 | inet_frag_put(qp_in, f); | |
204 | return qp; | |
205 | } | |
206 | } | |
207 | #endif | |
208 | qp = qp_in; | |
b2fd5321 | 209 | if (!mod_timer(&qp->timer, jiffies + nf->timeout)) |
2588fe1d PE |
210 | atomic_inc(&qp->refcnt); |
211 | ||
212 | atomic_inc(&qp->refcnt); | |
213 | hlist_add_head(&qp->list, &f->hash[hash]); | |
214 | list_add_tail(&qp->lru_list, &f->lru_list); | |
e5a2bb84 | 215 | nf->nqueues++; |
2588fe1d PE |
216 | write_unlock(&f->lock); |
217 | return qp; | |
218 | } | |
e521db9d | 219 | |
ac18e750 PE |
220 | static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, |
221 | struct inet_frags *f, void *arg) | |
e521db9d PE |
222 | { |
223 | struct inet_frag_queue *q; | |
224 | ||
225 | q = kzalloc(f->qsize, GFP_ATOMIC); | |
226 | if (q == NULL) | |
227 | return NULL; | |
228 | ||
c6fda282 | 229 | f->constructor(q, arg); |
6ddc0822 | 230 | atomic_add(f->qsize, &nf->mem); |
e521db9d PE |
231 | setup_timer(&q->timer, f->frag_expire, (unsigned long)q); |
232 | spin_lock_init(&q->lock); | |
233 | atomic_set(&q->refcnt, 1); | |
ac18e750 | 234 | q->net = nf; |
e521db9d PE |
235 | |
236 | return q; | |
237 | } | |
c6fda282 | 238 | |
ac18e750 PE |
239 | static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, |
240 | struct inet_frags *f, void *arg, unsigned int hash) | |
c6fda282 PE |
241 | { |
242 | struct inet_frag_queue *q; | |
243 | ||
ac18e750 | 244 | q = inet_frag_alloc(nf, f, arg); |
c6fda282 PE |
245 | if (q == NULL) |
246 | return NULL; | |
247 | ||
ac18e750 | 248 | return inet_frag_intern(nf, q, f, hash, arg); |
c6fda282 | 249 | } |
abd6523d | 250 | |
ac18e750 PE |
251 | struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, |
252 | struct inet_frags *f, void *key, unsigned int hash) | |
abd6523d PE |
253 | { |
254 | struct inet_frag_queue *q; | |
255 | struct hlist_node *n; | |
256 | ||
257 | read_lock(&f->lock); | |
258 | hlist_for_each_entry(q, n, &f->hash[hash], list) { | |
ac18e750 | 259 | if (q->net == nf && f->match(q, key)) { |
abd6523d PE |
260 | atomic_inc(&q->refcnt); |
261 | read_unlock(&f->lock); | |
262 | return q; | |
263 | } | |
264 | } | |
265 | read_unlock(&f->lock); | |
266 | ||
ac18e750 | 267 | return inet_frag_create(nf, f, key, hash); |
abd6523d PE |
268 | } |
269 | EXPORT_SYMBOL(inet_frag_find); |