socklib.c

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