xref: /freebsd/sys/compat/linuxkpi/common/src/linux_skbuff.c (revision b3e7694832e81d7a904a10f525f8797b753bf0d3)
1 /*-
2  * Copyright (c) 2020-2022 The FreeBSD Foundation
3  * Copyright (c) 2021-2022 Bjoern A. Zeeb
4  *
5  * This software was developed by Björn Zeeb under sponsorship from
6  * the FreeBSD Foundation.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * NOTE: this socket buffer compatibility code is highly EXPERIMENTAL.
32  *       Do not rely on the internals of this implementation.  They are highly
33  *       likely to change as we will improve the integration to FreeBSD mbufs.
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include "opt_ddb.h"
40 
41 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/sysctl.h>
46 
47 #ifdef DDB
48 #include <ddb/ddb.h>
49 #endif
50 
51 #include <linux/skbuff.h>
52 #include <linux/slab.h>
53 #include <linux/gfp.h>
54 #ifdef __LP64__
55 #include <linux/log2.h>
56 #endif
57 
58 SYSCTL_DECL(_compat_linuxkpi);
59 SYSCTL_NODE(_compat_linuxkpi, OID_AUTO, skb, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
60     "LinuxKPI skbuff");
61 
62 #ifdef SKB_DEBUG
63 int linuxkpi_debug_skb;
64 SYSCTL_INT(_compat_linuxkpi_skb, OID_AUTO, debug, CTLFLAG_RWTUN,
65     &linuxkpi_debug_skb, 0, "SKB debug level");
66 #endif
67 
68 #ifdef __LP64__
69 /*
70  * Realtek wireless drivers (e.g., rtw88) require 32bit DMA in a single segment.
71  * busdma(9) has a hard time providing this currently for 3-ish pages at large
72  * quantities (see lkpi_pci_nseg1_fail in linux_pci.c).
73  * Work around this for now by allowing a tunable to enforce physical addresses
74  * allocation limits on 64bit platforms using "old-school" contigmalloc(9) to
75  * avoid bouncing.
76  */
77 static int linuxkpi_skb_memlimit;
78 SYSCTL_INT(_compat_linuxkpi_skb, OID_AUTO, mem_limit, CTLFLAG_RDTUN,
79     &linuxkpi_skb_memlimit, 0, "SKB memory limit: 0=no limit, "
80     "1=32bit, 2=36bit, other=undef (currently 32bit)");
81 #endif
82 
83 static MALLOC_DEFINE(M_LKPISKB, "lkpiskb", "Linux KPI skbuff compat");
84 
85 struct sk_buff *
86 linuxkpi_alloc_skb(size_t size, gfp_t gfp)
87 {
88 	struct sk_buff *skb;
89 	size_t len;
90 
91 	len = sizeof(*skb) + size + sizeof(struct skb_shared_info);
92 	/*
93 	 * Using our own type here not backing my kmalloc.
94 	 * We assume no one calls kfree directly on the skb.
95 	 */
96 #ifdef __LP64__
97 	if (__predict_true(linuxkpi_skb_memlimit == 0)) {
98 		skb = malloc(len, M_LKPISKB, linux_check_m_flags(gfp) | M_ZERO);
99 	} else {
100 		vm_paddr_t high;
101 
102 		switch (linuxkpi_skb_memlimit) {
103 		case 2:
104 			high = (0xfffffffff);	/* 1<<36 really. */
105 			break;
106 		case 1:
107 		default:
108 			high = (0xffffffff);	/* 1<<32 really. */
109 			break;
110 		}
111 		len = roundup_pow_of_two(len);
112 		skb = contigmalloc(len, M_LKPISKB,
113 		    linux_check_m_flags(gfp) | M_ZERO, 0, high, PAGE_SIZE, 0);
114 	}
115 #else
116 	skb = malloc(len, M_LKPISKB, linux_check_m_flags(gfp) | M_ZERO);
117 #endif
118 	if (skb == NULL)
119 		return (skb);
120 	skb->_alloc_len = len;
121 	skb->truesize = size;
122 
123 	skb->head = skb->data = skb->tail = (uint8_t *)(skb+1);
124 	skb->end = skb->head + size;
125 
126 	skb->prev = skb->next = skb;
127 
128 	skb->shinfo = (struct skb_shared_info *)(skb->end);
129 
130 	SKB_TRACE_FMT(skb, "data %p size %zu", (skb) ? skb->data : NULL, size);
131 	return (skb);
132 }
133 
134 struct sk_buff *
135 linuxkpi_dev_alloc_skb(size_t size, gfp_t gfp)
136 {
137 	struct sk_buff *skb;
138 	size_t len;
139 
140 	len = size + NET_SKB_PAD;
141 	skb = linuxkpi_alloc_skb(len, gfp);
142 
143 	if (skb != NULL)
144 		skb_reserve(skb, NET_SKB_PAD);
145 
146 	SKB_TRACE_FMT(skb, "data %p size %zu len %zu",
147 	    (skb) ? skb->data : NULL, size, len);
148 	return (skb);
149 }
150 
151 struct sk_buff *
152 linuxkpi_build_skb(void *data, size_t fragsz)
153 {
154 	struct sk_buff *skb;
155 
156 	if (data == NULL || fragsz == 0)
157 		return (NULL);
158 
159 	/* Just allocate a skb without data area. */
160 	skb = linuxkpi_alloc_skb(0, GFP_KERNEL);
161 	if (skb == NULL)
162 		return (NULL);
163 
164 	skb->_flags |= _SKB_FLAGS_SKBEXTFRAG;
165 	skb->truesize = fragsz;
166 	skb->head = skb->data = data;
167 	skb_reset_tail_pointer(skb);	/* XXX is that correct? */
168 	skb->end = (void *)((uintptr_t)skb->head + fragsz);
169 
170 	return (skb);
171 }
172 
173 struct sk_buff *
174 linuxkpi_skb_copy(struct sk_buff *skb, gfp_t gfp)
175 {
176 	struct sk_buff *new;
177 	struct skb_shared_info *shinfo;
178 	size_t len;
179 	unsigned int headroom;
180 
181 	/* Full buffer size + any fragments. */
182 	len = skb->end - skb->head + skb->data_len;
183 
184 	new = linuxkpi_alloc_skb(len, gfp);
185 	if (new == NULL)
186 		return (NULL);
187 
188 	headroom = skb_headroom(skb);
189 	/* Fixup head and end. */
190 	skb_reserve(new, headroom);	/* data and tail move headroom forward. */
191 	skb_put(new, skb->len);		/* tail and len get adjusted */
192 
193 	/* Copy data. */
194 	memcpy(new->head, skb->data - headroom, headroom + skb->len);
195 
196 	/* Deal with fragments. */
197 	shinfo = skb->shinfo;
198 	if (shinfo->nr_frags > 0) {
199 		printf("%s:%d: NOT YET SUPPORTED; missing %d frags\n",
200 		    __func__, __LINE__, shinfo->nr_frags);
201 		SKB_TODO();
202 	}
203 
204 	/* Deal with header fields. */
205 	memcpy(new->cb, skb->cb, sizeof(skb->cb));
206 	SKB_IMPROVE("more header fields to copy?");
207 
208 	return (new);
209 }
210 
211 void
212 linuxkpi_kfree_skb(struct sk_buff *skb)
213 {
214 	struct skb_shared_info *shinfo;
215 	uint16_t fragno, count;
216 
217 	SKB_TRACE(skb);
218 	if (skb == NULL)
219 		return;
220 
221 	/*
222 	 * XXX TODO this will go away once we have skb backed by mbuf.
223 	 * currently we allow the mbuf to stay around and use a private
224 	 * free function to allow secondary resources to be freed along.
225 	 */
226 	if (skb->m != NULL) {
227 		void *m;
228 
229 		m = skb->m;
230 		skb->m = NULL;
231 
232 		KASSERT(skb->m_free_func != NULL, ("%s: skb %p has m %p but no "
233 		    "m_free_func %p\n", __func__, skb, m, skb->m_free_func));
234 		skb->m_free_func(m);
235 	}
236 	KASSERT(skb->m == NULL,
237 	    ("%s: skb %p m %p != NULL\n", __func__, skb, skb->m));
238 
239 	shinfo = skb->shinfo;
240 	for (count = fragno = 0;
241 	    count < shinfo->nr_frags && fragno < nitems(shinfo->frags);
242 	    fragno++) {
243 
244 		if (shinfo->frags[fragno].page != NULL) {
245 			struct page *p;
246 
247 			p = shinfo->frags[fragno].page;
248 			shinfo->frags[fragno].size = 0;
249 			shinfo->frags[fragno].offset = 0;
250 			shinfo->frags[fragno].page = NULL;
251 			__free_page(p);
252 			count++;
253 		}
254 	}
255 
256 	if ((skb->_flags & _SKB_FLAGS_SKBEXTFRAG) != 0) {
257 		void *p;
258 
259 		p = skb->head;
260 		skb_free_frag(p);
261 	}
262 
263 #ifdef __LP64__
264 	if (__predict_true(linuxkpi_skb_memlimit == 0))
265 		free(skb, M_LKPISKB);
266 	else
267 		contigfree(skb, skb->_alloc_len, M_LKPISKB);
268 #else
269 	free(skb, M_LKPISKB);
270 #endif
271 }
272 
273 #ifdef DDB
274 DB_SHOW_COMMAND(skb, db_show_skb)
275 {
276 	struct sk_buff *skb;
277 	int i;
278 
279 	if (!have_addr) {
280 		db_printf("usage: show skb <addr>\n");
281 			return;
282 	}
283 
284 	skb = (struct sk_buff *)addr;
285 
286 	db_printf("skb %p\n", skb);
287 	db_printf("\tnext %p prev %p\n", skb->next, skb->prev);
288 	db_printf("\tlist %p\n", &skb->list);
289 	db_printf("\t_alloc_len %u len %u data_len %u truesize %u mac_len %u\n",
290 	    skb->_alloc_len, skb->len, skb->data_len, skb->truesize,
291 	    skb->mac_len);
292 	db_printf("\tcsum %#06x l3hdroff %u l4hdroff %u priority %u qmap %u\n",
293 	    skb->csum, skb->l3hdroff, skb->l4hdroff, skb->priority, skb->qmap);
294 	db_printf("\tpkt_type %d dev %p sk %p\n",
295 	    skb->pkt_type, skb->dev, skb->sk);
296 	db_printf("\tcsum_offset %d csum_start %d ip_summed %d protocol %d\n",
297 	    skb->csum_offset, skb->csum_start, skb->ip_summed, skb->protocol);
298 	db_printf("\t_flags %#06x\n", skb->_flags);		/* XXX-BZ print names? */
299 	db_printf("\thead %p data %p tail %p end %p\n",
300 	    skb->head, skb->data, skb->tail, skb->end);
301 	db_printf("\tshinfo %p m %p m_free_func %p\n",
302 	    skb->shinfo, skb->m, skb->m_free_func);
303 
304 	if (skb->shinfo != NULL) {
305 		struct skb_shared_info *shinfo;
306 
307 		shinfo = skb->shinfo;
308 		db_printf("\t\tgso_type %d gso_size %u nr_frags %u\n",
309 		    shinfo->gso_type, shinfo->gso_size, shinfo->nr_frags);
310 		for (i = 0; i < nitems(shinfo->frags); i++) {
311 			struct skb_frag *frag;
312 
313 			frag = &shinfo->frags[i];
314 			if (frag == NULL || frag->page == NULL)
315 				continue;
316 			db_printf("\t\t\tfrag %p fragno %d page %p %p "
317 			    "offset %ju size %zu\n",
318 			    frag, i, frag->page, linux_page_address(frag->page),
319 			    (uintmax_t)frag->offset, frag->size);
320 		}
321 	}
322 	db_printf("\tcb[] %p {", skb->cb);
323 	for (i = 0; i < nitems(skb->cb); i++) {
324 		db_printf("%#04x%s",
325 		    skb->cb[i], (i < (nitems(skb->cb)-1)) ? ", " : "");
326 	}
327 	db_printf("}\n");
328 
329 	db_printf("\t__scratch[0] %p\n", skb->__scratch);
330 };
331 #endif
332