1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/net/sunrpc/socklib.c
4 *
5 * Common socket helper routines for RPC client and server
6 *
7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 */
9
10 #include <linux/compiler.h>
11 #include <linux/netdevice.h>
12 #include <linux/gfp.h>
13 #include <linux/skbuff.h>
14 #include <linux/types.h>
15 #include <linux/pagemap.h>
16 #include <linux/udp.h>
17 #include <linux/sunrpc/msg_prot.h>
18 #include <linux/sunrpc/sched.h>
19 #include <linux/sunrpc/xdr.h>
20 #include <linux/export.h>
21
22 #include "socklib.h"
23
24 /*
25 * Helper structure for copying from an sk_buff.
26 */
27 struct xdr_skb_reader {
28 struct sk_buff *skb;
29 unsigned int offset;
30 size_t count;
31 __wsum csum;
32 };
33
34 typedef size_t (*xdr_skb_read_actor)(struct xdr_skb_reader *desc, void *to,
35 size_t len);
36
37 /**
38 * xdr_skb_read_bits - copy some data bits from skb to internal buffer
39 * @desc: sk_buff copy helper
40 * @to: copy destination
41 * @len: number of bytes to copy
42 *
43 * Possibly called several times to iterate over an sk_buff and copy
44 * data out of it.
45 */
46 static size_t
47 xdr_skb_read_bits(struct xdr_skb_reader *desc, void *to, size_t len)
48 {
49 if (len > desc->count)
50 len = desc->count;
51 if (unlikely(skb_copy_bits(desc->skb, desc->offset, to, len)))
52 return 0;
53 desc->count -= len;
54 desc->offset += len;
55 return len;
56 }
57
58 /**
59 * xdr_skb_read_and_csum_bits - copy and checksum from skb to buffer
60 * @desc: sk_buff copy helper
61 * @to: copy destination
62 * @len: number of bytes to copy
63 *
64 * Same as skb_read_bits, but calculate a checksum at the same time.
65 */
66 static size_t xdr_skb_read_and_csum_bits(struct xdr_skb_reader *desc, void *to, size_t len)
67 {
68 unsigned int pos;
69 __wsum csum2;
70
71 if (len > desc->count)
72 len = desc->count;
73 pos = desc->offset;
74 csum2 = skb_copy_and_csum_bits(desc->skb, pos, to, len);
75 desc->csum = csum_block_add(desc->csum, csum2, pos);
76 desc->count -= len;
77 desc->offset += len;
78 return len;
79 }
80
81 /**
82 * xdr_partial_copy_from_skb - copy data out of an skb
83 * @xdr: target XDR buffer
84 * @base: starting offset
85 * @desc: sk_buff copy helper
86 * @copy_actor: virtual method for copying data
87 *
88 */
89 static ssize_t
90 xdr_partial_copy_from_skb(struct xdr_buf *xdr, unsigned int base, struct xdr_skb_reader *desc, xdr_skb_read_actor copy_actor)
91 {
92 struct page **ppage = xdr->pages;
93 unsigned int len, pglen = xdr->page_len;
94 ssize_t copied = 0;
95 size_t ret;
96
97 len = xdr->head[0].iov_len;
98 if (base < len) {
99 len -= base;
100 ret = copy_actor(desc, (char *)xdr->head[0].iov_base + base, len);
101 copied += ret;
102 if (ret != len || !desc->count)
103 goto out;
104 base = 0;
105 } else
106 base -= len;
107
108 if (unlikely(pglen == 0))
109 goto copy_tail;
110 if (unlikely(base >= pglen)) {
111 base -= pglen;
112 goto copy_tail;
113 }
114 if (base || xdr->page_base) {
115 pglen -= base;
116 base += xdr->page_base;
117 ppage += base >> PAGE_SHIFT;
118 base &= ~PAGE_MASK;
119 }
120 do {
121 char *kaddr;
122
123 /* ACL likes to be lazy in allocating pages - ACLs
124 * are small by default but can get huge. */
125 if ((xdr->flags & XDRBUF_SPARSE_PAGES) && *ppage == NULL) {
126 *ppage = alloc_page(GFP_NOWAIT | __GFP_NOWARN);
127 if (unlikely(*ppage == NULL)) {
128 if (copied == 0)
129 copied = -ENOMEM;
130 goto out;
131 }
132 }
133
134 len = PAGE_SIZE;
135 kaddr = kmap_atomic(*ppage);
136 if (base) {
137 len -= base;
138 if (pglen < len)
139 len = pglen;
140 ret = copy_actor(desc, kaddr + base, len);
141 base = 0;
142 } else {
143 if (pglen < len)
144 len = pglen;
145 ret = copy_actor(desc, kaddr, len);
146 }
147 flush_dcache_page(*ppage);
148 kunmap_atomic(kaddr);
149 copied += ret;
150 if (ret != len || !desc->count)
151 goto out;
152 ppage++;
153 } while ((pglen -= len) != 0);
154 copy_tail:
155 len = xdr->tail[0].iov_len;
156 if (base < len)
157 copied += copy_actor(desc, (char *)xdr->tail[0].iov_base + base, len - base);
158 out:
159 return copied;
160 }
161
162 /**
163 * csum_partial_copy_to_xdr - checksum and copy data
164 * @xdr: target XDR buffer
165 * @skb: source skb
166 *
167 * We have set things up such that we perform the checksum of the UDP
168 * packet in parallel with the copies into the RPC client iovec. -DaveM
169 */
170 int csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
171 {
172 struct xdr_skb_reader desc;
173
174 desc.skb = skb;
175 desc.offset = 0;
176 desc.count = skb->len - desc.offset;
177
178 if (skb_csum_unnecessary(skb))
179 goto no_checksum;
180
181 desc.csum = csum_partial(skb->data, desc.offset, skb->csum);
182 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_and_csum_bits) < 0)
183 return -1;
184 if (desc.offset != skb->len) {
185 __wsum csum2;
186 csum2 = skb_checksum(skb, desc.offset, skb->len - desc.offset, 0);
187 desc.csum = csum_block_add(desc.csum, csum2, desc.offset);
188 }
189 if (desc.count)
190 return -1;
191 if (csum_fold(desc.csum))
192 return -1;
193 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
194 !skb->csum_complete_sw)
195 netdev_rx_csum_fault(skb->dev, skb);
196 return 0;
197 no_checksum:
198 if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
199 return -1;
200 if (desc.count)
201 return -1;
202 return 0;
203 }
204 EXPORT_SYMBOL_GPL(csum_partial_copy_to_xdr);
205
206 static inline int xprt_sendmsg(struct socket *sock, struct msghdr *msg,
207 size_t seek)
208 {
209 if (seek)
210 iov_iter_advance(&msg->msg_iter, seek);
211 return sock_sendmsg(sock, msg);
212 }
213
214 static int xprt_send_kvec(struct socket *sock, struct msghdr *msg,
215 struct kvec *vec, size_t seek)
216 {
217 iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
218 return xprt_sendmsg(sock, msg, seek);
219 }
220
221 static int xprt_send_pagedata(struct socket *sock, struct msghdr *msg,
222 struct xdr_buf *xdr, size_t base)
223 {
224 iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec, xdr_buf_pagecount(xdr),
225 xdr->page_len + xdr->page_base);
226 return xprt_sendmsg(sock, msg, base + xdr->page_base);
227 }
228
229 /* Common case:
230 * - stream transport
231 * - sending from byte 0 of the message
232 * - the message is wholly contained in @xdr's head iovec
233 */
234 static int xprt_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
235 rpc_fraghdr marker, struct kvec *vec,
236 size_t base)
237 {
238 struct kvec iov[2] = {
239 [0] = {
240 .iov_base = &marker,
241 .iov_len = sizeof(marker)
242 },
243 [1] = *vec,
244 };
245 size_t len = iov[0].iov_len + iov[1].iov_len;
246
247 iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
248 return xprt_sendmsg(sock, msg, base);
249 }
250
251 /**
252 * xprt_sock_sendmsg - write an xdr_buf directly to a socket
253 * @sock: open socket to send on
254 * @msg: socket message metadata
255 * @xdr: xdr_buf containing this request
256 * @base: starting position in the buffer
257 * @marker: stream record marker field
258 * @sent_p: return the total number of bytes successfully queued for sending
259 *
260 * Return values:
261 * On success, returns zero and fills in @sent_p.
262 * %-ENOTSOCK if @sock is not a struct socket.
263 */
264 int xprt_sock_sendmsg(struct socket *sock, struct msghdr *msg,
265 struct xdr_buf *xdr, unsigned int base,
266 rpc_fraghdr marker, unsigned int *sent_p)
267 {
268 unsigned int rmsize = marker ? sizeof(marker) : 0;
269 unsigned int remainder = rmsize + xdr->len - base;
270 unsigned int want;
271 int err = 0;
272
273 *sent_p = 0;
274
275 if (unlikely(!sock))
276 return -ENOTSOCK;
277
278 msg->msg_flags |= MSG_MORE;
279 want = xdr->head[0].iov_len + rmsize;
280 if (base < want) {
281 unsigned int len = want - base;
282
283 remainder -= len;
284 if (remainder == 0)
285 msg->msg_flags &= ~MSG_MORE;
286 if (rmsize)
287 err = xprt_send_rm_and_kvec(sock, msg, marker,
288 &xdr->head[0], base);
289 else
290 err = xprt_send_kvec(sock, msg, &xdr->head[0], base);
291 if (remainder == 0 || err != len)
292 goto out;
293 *sent_p += err;
294 base = 0;
295 } else {
296 base -= want;
297 }
298
299 if (base < xdr->page_len) {
300 unsigned int len = xdr->page_len - base;
301
302 remainder -= len;
303 if (remainder == 0)
304 msg->msg_flags &= ~MSG_MORE;
305 err = xprt_send_pagedata(sock, msg, xdr, base);
306 if (remainder == 0 || err != len)
307 goto out;
308 *sent_p += err;
309 base = 0;
310 } else {
311 base -= xdr->page_len;
312 }
313
314 if (base >= xdr->tail[0].iov_len)
315 return 0;
316 msg->msg_flags &= ~MSG_MORE;
317 err = xprt_send_kvec(sock, msg, &xdr->tail[0], base);
318 out:
319 if (err > 0) {
320 *sent_p += err;
321 err = 0;
322 }
323 return err;
324 }
325
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.