]>
Commit | Line | Data |
---|---|---|
1 | diff -ur ../vger3-011229/linux/net/unix/af_unix.c linux/net/unix/af_unix.c | |
2 | --- ../vger3-011229/linux/net/unix/af_unix.c Mon Dec 3 20:24:03 2001 | |
3 | +++ linux/net/unix/af_unix.c Sat Jan 5 04:30:19 2002 | |
4 | @@ -112,6 +112,7 @@ | |
5 | #include <asm/checksum.h> | |
6 | ||
7 | int sysctl_unix_max_dgram_qlen = 10; | |
8 | +int sysctl_unix_stream_pages = MAX_SKB_FRAGS; | |
9 | ||
10 | unix_socket *unix_socket_table[UNIX_HASH_SIZE+1]; | |
11 | rwlock_t unix_table_lock = RW_LOCK_UNLOCKED; | |
12 | @@ -1123,9 +1124,6 @@ | |
13 | struct scm_cookie scm; | |
14 | memset(&scm, 0, sizeof(scm)); | |
15 | unix_detach_fds(&scm, skb); | |
16 | - | |
17 | - /* Alas, it calls VFS */ | |
18 | - /* So fscking what? fput() had been SMP-safe since the last Summer */ | |
19 | scm_destroy(&scm); | |
20 | sock_wfree(skb); | |
21 | } | |
22 | @@ -1140,6 +1138,67 @@ | |
23 | scm->fp = NULL; | |
24 | } | |
25 | ||
26 | +int datagram_copy_fromiovec(struct iovec *iov, struct sk_buff *skb, int size) | |
27 | +{ | |
28 | + struct sock *sk; | |
29 | + struct sk_buff **tail, *skb1; | |
30 | + int copy = min_t(int, size, skb_tailroom(skb)); | |
31 | + | |
32 | + if (memcpy_fromiovec(skb_put(skb, copy), iov, copy)) | |
33 | + goto do_fault; | |
34 | + | |
35 | + if ((size -= copy) == 0) | |
36 | + return 0; | |
37 | + | |
38 | + sk = skb->sk; | |
39 | + skb1 = skb; | |
40 | + tail = &skb_shinfo(skb)->frag_list; | |
41 | + | |
42 | + do { | |
43 | + struct page *page; | |
44 | + int i = skb_shinfo(skb1)->nr_frags; | |
45 | + | |
46 | + if (i == MAX_SKB_FRAGS) { | |
47 | + skb1 = alloc_skb(0, sk->allocation); | |
48 | + if (skb1 == NULL) | |
49 | + goto do_oom; | |
50 | + *tail = skb1; | |
51 | + tail = &skb1->next; | |
52 | + i = 0; | |
53 | + skb->truesize += skb1->truesize; | |
54 | + atomic_add(skb1->truesize, &sk->wmem_alloc); | |
55 | + } | |
56 | + | |
57 | + page = alloc_pages(sk->allocation, 0); | |
58 | + if (page == NULL) | |
59 | + goto do_oom; | |
60 | + | |
61 | + copy = min_t(int, size, PAGE_SIZE); | |
62 | + skb_shinfo(skb1)->nr_frags=i+1; | |
63 | + skb_shinfo(skb1)->frags[i].page = page; | |
64 | + skb_shinfo(skb1)->frags[i].page_offset = 0; | |
65 | + skb_shinfo(skb1)->frags[i].size = copy; | |
66 | + | |
67 | + skb1->len += copy; | |
68 | + skb1->data_len += copy; | |
69 | + if (skb != skb1) { | |
70 | + skb->len += copy; | |
71 | + skb->data_len += copy; | |
72 | + } | |
73 | + skb->truesize += PAGE_SIZE; | |
74 | + atomic_add(PAGE_SIZE, &sk->wmem_alloc); | |
75 | + if (memcpy_fromiovec(page_address(page), iov, copy)) | |
76 | + goto do_fault; | |
77 | + } while ((size -= copy) > 0); | |
78 | + return 0; | |
79 | + | |
80 | +do_oom: | |
81 | + return -ENOMEM; | |
82 | + | |
83 | +do_fault: | |
84 | + return -EFAULT; | |
85 | +} | |
86 | + | |
87 | /* | |
88 | * Send AF_UNIX data. | |
89 | */ | |
90 | @@ -1155,6 +1214,7 @@ | |
91 | unsigned hash; | |
92 | struct sk_buff *skb; | |
93 | long timeo; | |
94 | + int alloc; | |
95 | ||
96 | err = -EOPNOTSUPP; | |
97 | if (msg->msg_flags&MSG_OOB) | |
98 | @@ -1178,10 +1238,14 @@ | |
99 | goto out; | |
100 | ||
101 | err = -EMSGSIZE; | |
102 | - if ((unsigned)len > sk->sndbuf - 32) | |
103 | + if ((unsigned)len > sk->sndbuf) | |
104 | goto out; | |
105 | ||
106 | - skb = sock_alloc_send_skb(sk, len, msg->msg_flags&MSG_DONTWAIT, &err); | |
107 | + alloc = len; | |
108 | + if (alloc > SKB_MAX_HEAD(0)) | |
109 | + alloc = SKB_MAX_HEAD(0); | |
110 | + | |
111 | + skb = sock_alloc_send_skb(sk, alloc, msg->msg_flags&MSG_DONTWAIT, &err); | |
112 | if (skb==NULL) | |
113 | goto out; | |
114 | ||
115 | @@ -1190,7 +1254,7 @@ | |
116 | unix_attach_fds(scm, skb); | |
117 | ||
118 | skb->h.raw = skb->data; | |
119 | - err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len); | |
120 | + err = datagram_copy_fromiovec(msg->msg_iov, skb, len); | |
121 | if (err) | |
122 | goto out_free; | |
123 | ||
124 | @@ -1275,74 +1339,57 @@ | |
125 | return err; | |
126 | } | |
127 | ||
128 | - | |
129 | static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg, int len, | |
130 | struct scm_cookie *scm) | |
131 | { | |
132 | struct sock *sk = sock->sk; | |
133 | unix_socket *other = NULL; | |
134 | - struct sockaddr_un *sunaddr=msg->msg_name; | |
135 | - int err,size; | |
136 | struct sk_buff *skb; | |
137 | + int err; | |
138 | int sent=0; | |
139 | ||
140 | err = -EOPNOTSUPP; | |
141 | if (msg->msg_flags&MSG_OOB) | |
142 | goto out_err; | |
143 | ||
144 | - if (msg->msg_namelen) { | |
145 | - err = (sk->state==TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP); | |
146 | + err = -ENOTCONN; | |
147 | + other = unix_peer_get(sk); | |
148 | + if (!other) | |
149 | goto out_err; | |
150 | - } else { | |
151 | - sunaddr = NULL; | |
152 | - err = -ENOTCONN; | |
153 | - other = unix_peer_get(sk); | |
154 | - if (!other) | |
155 | - goto out_err; | |
156 | - } | |
157 | ||
158 | if (sk->shutdown&SEND_SHUTDOWN) | |
159 | goto pipe_err; | |
160 | ||
161 | - while(sent < len) | |
162 | - { | |
163 | - /* | |
164 | - * Optimisation for the fact that under 0.01% of X messages typically | |
165 | - * need breaking up. | |
166 | - */ | |
167 | + while(sent < len) { | |
168 | + int size, alloc; | |
169 | ||
170 | - size=len-sent; | |
171 | + size = len-sent; | |
172 | ||
173 | /* Keep two messages in the pipe so it schedules better */ | |
174 | - if (size > sk->sndbuf/2 - 64) | |
175 | - size = sk->sndbuf/2 - 64; | |
176 | + if (size > sk->sndbuf/2) | |
177 | + size = sk->sndbuf/2; | |
178 | ||
179 | - if (size > SKB_MAX_ALLOC) | |
180 | - size = SKB_MAX_ALLOC; | |
181 | - | |
182 | /* | |
183 | * Grab a buffer | |
184 | */ | |
185 | - | |
186 | - skb=sock_alloc_send_skb(sk,size,msg->msg_flags&MSG_DONTWAIT, &err); | |
187 | + alloc = size; | |
188 | + | |
189 | + if (size > SKB_MAX_HEAD(0)) { | |
190 | + alloc = SKB_MAX_HEAD(0); | |
191 | + if (size > alloc + sysctl_unix_stream_pages*PAGE_SIZE) | |
192 | + size = alloc + sysctl_unix_stream_pages*PAGE_SIZE; | |
193 | + } | |
194 | + | |
195 | + skb=sock_alloc_send_skb(sk,alloc,msg->msg_flags&MSG_DONTWAIT, &err); | |
196 | ||
197 | if (skb==NULL) | |
198 | goto out_err; | |
199 | ||
200 | - /* | |
201 | - * If you pass two values to the sock_alloc_send_skb | |
202 | - * it tries to grab the large buffer with GFP_NOFS | |
203 | - * (which can fail easily), and if it fails grab the | |
204 | - * fallback size buffer which is under a page and will | |
205 | - * succeed. [Alan] | |
206 | - */ | |
207 | - size = min_t(int, size, skb_tailroom(skb)); | |
208 | - | |
209 | memcpy(UNIXCREDS(skb), &scm->creds, sizeof(struct ucred)); | |
210 | if (scm->fp) | |
211 | unix_attach_fds(scm, skb); | |
212 | ||
213 | - if ((err = memcpy_fromiovec(skb_put(skb,size), msg->msg_iov, size)) != 0) { | |
214 | + if ((err = datagram_copy_fromiovec(msg->msg_iov, skb, size)) != 0) { | |
215 | kfree_skb(skb); | |
216 | goto out_err; | |
217 | } | |
218 | @@ -1418,13 +1465,10 @@ | |
219 | ||
220 | scm->creds = *UNIXCREDS(skb); | |
221 | ||
222 | - if (!(flags & MSG_PEEK)) | |
223 | - { | |
224 | + if (!(flags & MSG_PEEK)) { | |
225 | if (UNIXCB(skb).fp) | |
226 | unix_detach_fds(scm, skb); | |
227 | - } | |
228 | - else | |
229 | - { | |
230 | + } else { | |
231 | /* It is questionable: on PEEK we could: | |
232 | - do not return fds - good, but too simple 8) | |
233 | - return fds, and do not return them on read (old strategy, | |
234 | @@ -1483,13 +1527,10 @@ | |
235 | return timeo; | |
236 | } | |
237 | ||
238 | - | |
239 | - | |
240 | static int unix_stream_recvmsg(struct socket *sock, struct msghdr *msg, int size, | |
241 | int flags, struct scm_cookie *scm) | |
242 | { | |
243 | struct sock *sk = sock->sk; | |
244 | - struct sockaddr_un *sunaddr=msg->msg_name; | |
245 | int copied = 0; | |
246 | int check_creds = 0; | |
247 | int target; | |
248 | @@ -1515,21 +1556,18 @@ | |
249 | ||
250 | down(&sk->protinfo.af_unix.readsem); | |
251 | ||
252 | - do | |
253 | - { | |
254 | + do { | |
255 | int chunk; | |
256 | struct sk_buff *skb; | |
257 | ||
258 | skb=skb_dequeue(&sk->receive_queue); | |
259 | - if (skb==NULL) | |
260 | - { | |
261 | + if (skb==NULL) { | |
262 | if (copied >= target) | |
263 | break; | |
264 | ||
265 | /* | |
266 | * POSIX 1003.1g mandates this order. | |
267 | */ | |
268 | - | |
269 | if ((err = sock_error(sk)) != 0) | |
270 | break; | |
271 | if (sk->shutdown & RCV_SHUTDOWN) | |
272 | @@ -1551,60 +1589,44 @@ | |
273 | ||
274 | if (check_creds) { | |
275 | /* Never glue messages from different writers */ | |
276 | - if (memcmp(UNIXCREDS(skb), &scm->creds, sizeof(scm->creds)) != 0) { | |
277 | - skb_queue_head(&sk->receive_queue, skb); | |
278 | - break; | |
279 | - } | |
280 | + if (memcmp(UNIXCREDS(skb), &scm->creds, sizeof(scm->creds)) != 0) | |
281 | + goto out_put_back; | |
282 | } else { | |
283 | /* Copy credentials */ | |
284 | scm->creds = *UNIXCREDS(skb); | |
285 | check_creds = 1; | |
286 | } | |
287 | ||
288 | - /* Copy address just once */ | |
289 | - if (sunaddr) | |
290 | - { | |
291 | - unix_copy_addr(msg, skb->sk); | |
292 | - sunaddr = NULL; | |
293 | - } | |
294 | + chunk = min_t(int, skb->len - sk->protinfo.af_unix.copied, size); | |
295 | + err = skb_copy_datagram_iovec(skb, sk->protinfo.af_unix.copied, msg->msg_iov, chunk); | |
296 | + if (err) | |
297 | + goto out_put_back; | |
298 | ||
299 | - chunk = min_t(unsigned int, skb->len, size); | |
300 | - if (memcpy_toiovec(msg->msg_iov, skb->data, chunk)) { | |
301 | - skb_queue_head(&sk->receive_queue, skb); | |
302 | - if (copied == 0) | |
303 | - copied = -EFAULT; | |
304 | - break; | |
305 | - } | |
306 | copied += chunk; | |
307 | size -= chunk; | |
308 | ||
309 | /* Mark read part of skb as used */ | |
310 | - if (!(flags & MSG_PEEK)) | |
311 | - { | |
312 | - skb_pull(skb, chunk); | |
313 | - | |
314 | + if (!(flags & MSG_PEEK)) { | |
315 | if (UNIXCB(skb).fp) | |
316 | unix_detach_fds(scm, skb); | |
317 | ||
318 | /* put the skb back if we didn't use it up.. */ | |
319 | - if (skb->len) | |
320 | - { | |
321 | - skb_queue_head(&sk->receive_queue, skb); | |
322 | - break; | |
323 | - } | |
324 | + if ((sk->protinfo.af_unix.copied += chunk) < skb->len) | |
325 | + goto out_put_back; | |
326 | + | |
327 | + sk->protinfo.af_unix.copied = 0; | |
328 | ||
329 | kfree_skb(skb); | |
330 | ||
331 | if (scm->fp) | |
332 | break; | |
333 | - } | |
334 | - else | |
335 | - { | |
336 | + } else { | |
337 | /* It is questionable, see note in unix_dgram_recvmsg. | |
338 | */ | |
339 | if (UNIXCB(skb).fp) | |
340 | scm->fp = scm_fp_dup(UNIXCB(skb).fp); | |
341 | ||
342 | +out_put_back: | |
343 | /* put message back and return */ | |
344 | skb_queue_head(&sk->receive_queue, skb); | |
345 | break; | |
346 | @@ -1676,10 +1698,12 @@ | |
347 | break; | |
348 | } | |
349 | ||
350 | + down(&sk->protinfo.af_unix.readsem); | |
351 | spin_lock(&sk->receive_queue.lock); | |
352 | if((skb=skb_peek(&sk->receive_queue))!=NULL) | |
353 | - amount=skb->len; | |
354 | + amount=skb->len - sk->protinfo.af_unix.copied; | |
355 | spin_unlock(&sk->receive_queue.lock); | |
356 | + up(&sk->protinfo.af_unix.readsem); | |
357 | err = put_user(amount, (int *)arg); | |
358 | break; | |
359 | } | |
360 | @@ -1734,7 +1758,7 @@ | |
361 | int i; | |
362 | unix_socket *s; | |
363 | ||
364 | - len+= sprintf(buffer,"Num RefCount Protocol Flags Type St " | |
365 | + len+= sprintf(buffer,"Peer RcvQueue WMem Flags Type St " | |
366 | "Inode Path\n"); | |
367 | ||
368 | read_lock(&unix_table_lock); | |
369 | @@ -1742,10 +1766,10 @@ | |
370 | { | |
371 | unix_state_rlock(s); | |
372 | ||
373 | - len+=sprintf(buffer+len,"%p: %08X %08X %08X %04X %02X %5ld", | |
374 | - s, | |
375 | - atomic_read(&s->refcnt), | |
376 | - 0, | |
377 | + len+=sprintf(buffer+len,"%08lX: %08X %08X %08X %04X %02X %5ld", | |
378 | + unix_peer(s) ? sock_i_ino(unix_peer(s)) : 0, | |
379 | + skb_queue_len(&s->receive_queue), | |
380 | + atomic_read(&s->wmem_alloc), | |
381 | s->state == TCP_LISTEN ? __SO_ACCEPTCON : 0, | |
382 | s->type, | |
383 | s->socket ? | |
384 | diff -ur ../vger3-011229/linux/net/unix/sysctl_net_unix.c linux/net/unix/sysctl_net_unix.c | |
385 | --- ../vger3-011229/linux/net/unix/sysctl_net_unix.c Tue Jan 30 21:20:16 2001 | |
386 | +++ linux/net/unix/sysctl_net_unix.c Sat Jan 5 04:10:58 2002 | |
387 | @@ -13,10 +13,14 @@ | |
388 | #include <linux/sysctl.h> | |
389 | ||
390 | extern int sysctl_unix_max_dgram_qlen; | |
391 | +extern int sysctl_unix_stream_pages; | |
392 | ||
393 | ctl_table unix_table[] = { | |
394 | {NET_UNIX_MAX_DGRAM_QLEN, "max_dgram_qlen", | |
395 | &sysctl_unix_max_dgram_qlen, sizeof(int), 0600, NULL, | |
396 | + &proc_dointvec }, | |
397 | + {NET_UNIX_STREAM_PAGES, "stream_pages", | |
398 | + &sysctl_unix_stream_pages, sizeof(int), 0600, NULL, | |
399 | &proc_dointvec }, | |
400 | {0} | |
401 | }; |