xref: /linux/net/sunrpc/socklib.c (revision c5d3cdad688ed75fb311a3a671eb30ba7106d7d3)
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/xdr.h>
19 #include <linux/export.h>
20 
21 #include "socklib.h"
22 
23 /*
24  * Helper structure for copying from an sk_buff.
25  */
26 struct xdr_skb_reader {
27 	struct sk_buff	*skb;
28 	unsigned int	offset;
29 	size_t		count;
30 	__wsum		csum;
31 };
32 
33 typedef size_t (*xdr_skb_read_actor)(struct xdr_skb_reader *desc, void *to,
34 				     size_t len);
35 
36 /**
37  * xdr_skb_read_bits - copy some data bits from skb to internal buffer
38  * @desc: sk_buff copy helper
39  * @to: copy destination
40  * @len: number of bytes to copy
41  *
42  * Possibly called several times to iterate over an sk_buff and copy
43  * data out of it.
44  */
45 static size_t
46 xdr_skb_read_bits(struct xdr_skb_reader *desc, void *to, size_t len)
47 {
48 	if (len > desc->count)
49 		len = desc->count;
50 	if (unlikely(skb_copy_bits(desc->skb, desc->offset, to, len)))
51 		return 0;
52 	desc->count -= len;
53 	desc->offset += len;
54 	return len;
55 }
56 
57 /**
58  * xdr_skb_read_and_csum_bits - copy and checksum from skb to buffer
59  * @desc: sk_buff copy helper
60  * @to: copy destination
61  * @len: number of bytes to copy
62  *
63  * Same as skb_read_bits, but calculate a checksum at the same time.
64  */
65 static size_t xdr_skb_read_and_csum_bits(struct xdr_skb_reader *desc, void *to, size_t len)
66 {
67 	unsigned int pos;
68 	__wsum csum2;
69 
70 	if (len > desc->count)
71 		len = desc->count;
72 	pos = desc->offset;
73 	csum2 = skb_copy_and_csum_bits(desc->skb, pos, to, len, 0);
74 	desc->csum = csum_block_add(desc->csum, csum2, pos);
75 	desc->count -= len;
76 	desc->offset += len;
77 	return len;
78 }
79 
80 /**
81  * xdr_partial_copy_from_skb - copy data out of an skb
82  * @xdr: target XDR buffer
83  * @base: starting offset
84  * @desc: sk_buff copy helper
85  * @copy_actor: virtual method for copying data
86  *
87  */
88 static ssize_t
89 xdr_partial_copy_from_skb(struct xdr_buf *xdr, unsigned int base, struct xdr_skb_reader *desc, xdr_skb_read_actor copy_actor)
90 {
91 	struct page	**ppage = xdr->pages;
92 	unsigned int	len, pglen = xdr->page_len;
93 	ssize_t		copied = 0;
94 	size_t		ret;
95 
96 	len = xdr->head[0].iov_len;
97 	if (base < len) {
98 		len -= base;
99 		ret = copy_actor(desc, (char *)xdr->head[0].iov_base + base, len);
100 		copied += ret;
101 		if (ret != len || !desc->count)
102 			goto out;
103 		base = 0;
104 	} else
105 		base -= len;
106 
107 	if (unlikely(pglen == 0))
108 		goto copy_tail;
109 	if (unlikely(base >= pglen)) {
110 		base -= pglen;
111 		goto copy_tail;
112 	}
113 	if (base || xdr->page_base) {
114 		pglen -= base;
115 		base += xdr->page_base;
116 		ppage += base >> PAGE_SHIFT;
117 		base &= ~PAGE_MASK;
118 	}
119 	do {
120 		char *kaddr;
121 
122 		/* ACL likes to be lazy in allocating pages - ACLs
123 		 * are small by default but can get huge. */
124 		if ((xdr->flags & XDRBUF_SPARSE_PAGES) && *ppage == NULL) {
125 			*ppage = alloc_page(GFP_NOWAIT | __GFP_NOWARN);
126 			if (unlikely(*ppage == NULL)) {
127 				if (copied == 0)
128 					copied = -ENOMEM;
129 				goto out;
130 			}
131 		}
132 
133 		len = PAGE_SIZE;
134 		kaddr = kmap_atomic(*ppage);
135 		if (base) {
136 			len -= base;
137 			if (pglen < len)
138 				len = pglen;
139 			ret = copy_actor(desc, kaddr + base, len);
140 			base = 0;
141 		} else {
142 			if (pglen < len)
143 				len = pglen;
144 			ret = copy_actor(desc, kaddr, len);
145 		}
146 		flush_dcache_page(*ppage);
147 		kunmap_atomic(kaddr);
148 		copied += ret;
149 		if (ret != len || !desc->count)
150 			goto out;
151 		ppage++;
152 	} while ((pglen -= len) != 0);
153 copy_tail:
154 	len = xdr->tail[0].iov_len;
155 	if (base < len)
156 		copied += copy_actor(desc, (char *)xdr->tail[0].iov_base + base, len - base);
157 out:
158 	return copied;
159 }
160 
161 /**
162  * csum_partial_copy_to_xdr - checksum and copy data
163  * @xdr: target XDR buffer
164  * @skb: source skb
165  *
166  * We have set things up such that we perform the checksum of the UDP
167  * packet in parallel with the copies into the RPC client iovec.  -DaveM
168  */
169 int csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
170 {
171 	struct xdr_skb_reader	desc;
172 
173 	desc.skb = skb;
174 	desc.offset = 0;
175 	desc.count = skb->len - desc.offset;
176 
177 	if (skb_csum_unnecessary(skb))
178 		goto no_checksum;
179 
180 	desc.csum = csum_partial(skb->data, desc.offset, skb->csum);
181 	if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_and_csum_bits) < 0)
182 		return -1;
183 	if (desc.offset != skb->len) {
184 		__wsum csum2;
185 		csum2 = skb_checksum(skb, desc.offset, skb->len - desc.offset, 0);
186 		desc.csum = csum_block_add(desc.csum, csum2, desc.offset);
187 	}
188 	if (desc.count)
189 		return -1;
190 	if (csum_fold(desc.csum))
191 		return -1;
192 	if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
193 	    !skb->csum_complete_sw)
194 		netdev_rx_csum_fault(skb->dev, skb);
195 	return 0;
196 no_checksum:
197 	if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
198 		return -1;
199 	if (desc.count)
200 		return -1;
201 	return 0;
202 }
203 EXPORT_SYMBOL_GPL(csum_partial_copy_to_xdr);
204 
205 static inline int xprt_sendmsg(struct socket *sock, struct msghdr *msg,
206 			       size_t seek)
207 {
208 	if (seek)
209 		iov_iter_advance(&msg->msg_iter, seek);
210 	return sock_sendmsg(sock, msg);
211 }
212 
213 static int xprt_send_kvec(struct socket *sock, struct msghdr *msg,
214 			  struct kvec *vec, size_t seek)
215 {
216 	iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
217 	return xprt_sendmsg(sock, msg, seek);
218 }
219 
220 static int xprt_send_pagedata(struct socket *sock, struct msghdr *msg,
221 			      struct xdr_buf *xdr, size_t base)
222 {
223 	int err;
224 
225 	err = xdr_alloc_bvec(xdr, GFP_KERNEL);
226 	if (err < 0)
227 		return err;
228 
229 	iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec, xdr_buf_pagecount(xdr),
230 		      xdr->page_len + xdr->page_base);
231 	return xprt_sendmsg(sock, msg, base + xdr->page_base);
232 }
233 
234 /* Common case:
235  *  - stream transport
236  *  - sending from byte 0 of the message
237  *  - the message is wholly contained in @xdr's head iovec
238  */
239 static int xprt_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
240 				 rpc_fraghdr marker, struct kvec *vec,
241 				 size_t base)
242 {
243 	struct kvec iov[2] = {
244 		[0] = {
245 			.iov_base	= &marker,
246 			.iov_len	= sizeof(marker)
247 		},
248 		[1] = *vec,
249 	};
250 	size_t len = iov[0].iov_len + iov[1].iov_len;
251 
252 	iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
253 	return xprt_sendmsg(sock, msg, base);
254 }
255 
256 /**
257  * xprt_sock_sendmsg - write an xdr_buf directly to a socket
258  * @sock: open socket to send on
259  * @msg: socket message metadata
260  * @xdr: xdr_buf containing this request
261  * @base: starting position in the buffer
262  * @marker: stream record marker field
263  * @sent_p: return the total number of bytes successfully queued for sending
264  *
265  * Return values:
266  *   On success, returns zero and fills in @sent_p.
267  *   %-ENOTSOCK if  @sock is not a struct socket.
268  */
269 int xprt_sock_sendmsg(struct socket *sock, struct msghdr *msg,
270 		      struct xdr_buf *xdr, unsigned int base,
271 		      rpc_fraghdr marker, unsigned int *sent_p)
272 {
273 	unsigned int rmsize = marker ? sizeof(marker) : 0;
274 	unsigned int remainder = rmsize + xdr->len - base;
275 	unsigned int want;
276 	int err = 0;
277 
278 	*sent_p = 0;
279 
280 	if (unlikely(!sock))
281 		return -ENOTSOCK;
282 
283 	msg->msg_flags |= MSG_MORE;
284 	want = xdr->head[0].iov_len + rmsize;
285 	if (base < want) {
286 		unsigned int len = want - base;
287 
288 		remainder -= len;
289 		if (remainder == 0)
290 			msg->msg_flags &= ~MSG_MORE;
291 		if (rmsize)
292 			err = xprt_send_rm_and_kvec(sock, msg, marker,
293 						    &xdr->head[0], base);
294 		else
295 			err = xprt_send_kvec(sock, msg, &xdr->head[0], base);
296 		if (remainder == 0 || err != len)
297 			goto out;
298 		*sent_p += err;
299 		base = 0;
300 	} else {
301 		base -= want;
302 	}
303 
304 	if (base < xdr->page_len) {
305 		unsigned int len = xdr->page_len - base;
306 
307 		remainder -= len;
308 		if (remainder == 0)
309 			msg->msg_flags &= ~MSG_MORE;
310 		err = xprt_send_pagedata(sock, msg, xdr, base);
311 		if (remainder == 0 || err != len)
312 			goto out;
313 		*sent_p += err;
314 		base = 0;
315 	} else {
316 		base -= xdr->page_len;
317 	}
318 
319 	if (base >= xdr->tail[0].iov_len)
320 		return 0;
321 	msg->msg_flags &= ~MSG_MORE;
322 	err = xprt_send_kvec(sock, msg, &xdr->tail[0], base);
323 out:
324 	if (err > 0) {
325 		*sent_p += err;
326 		err = 0;
327 	}
328 	return err;
329 }
330