xref: /linux/net/rds/message.c (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
1 /*
2  * Copyright (c) 2006, 2020 Oracle and/or its affiliates.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 #include <linux/kernel.h>
34 #include <linux/slab.h>
35 #include <linux/export.h>
36 #include <linux/skbuff.h>
37 #include <linux/list.h>
38 #include <linux/errqueue.h>
39 
40 #include "rds.h"
41 
42 static unsigned int	rds_exthdr_size[__RDS_EXTHDR_MAX] = {
43 [RDS_EXTHDR_NONE]	= 0,
44 [RDS_EXTHDR_VERSION]	= sizeof(struct rds_ext_header_version),
45 [RDS_EXTHDR_RDMA]	= sizeof(struct rds_ext_header_rdma),
46 [RDS_EXTHDR_RDMA_DEST]	= sizeof(struct rds_ext_header_rdma_dest),
47 [RDS_EXTHDR_NPATHS]	= sizeof(u16),
48 [RDS_EXTHDR_GEN_NUM]	= sizeof(u32),
49 };
50 
51 void rds_message_addref(struct rds_message *rm)
52 {
53 	rdsdebug("addref rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
54 	refcount_inc(&rm->m_refcount);
55 }
56 EXPORT_SYMBOL_GPL(rds_message_addref);
57 
58 static inline bool rds_zcookie_add(struct rds_msg_zcopy_info *info, u32 cookie)
59 {
60 	struct rds_zcopy_cookies *ck = &info->zcookies;
61 	int ncookies = ck->num;
62 
63 	if (ncookies == RDS_MAX_ZCOOKIES)
64 		return false;
65 	ck->cookies[ncookies] = cookie;
66 	ck->num =  ++ncookies;
67 	return true;
68 }
69 
70 static struct rds_msg_zcopy_info *rds_info_from_znotifier(struct rds_znotifier *znotif)
71 {
72 	return container_of(znotif, struct rds_msg_zcopy_info, znotif);
73 }
74 
75 void rds_notify_msg_zcopy_purge(struct rds_msg_zcopy_queue *q)
76 {
77 	unsigned long flags;
78 	LIST_HEAD(copy);
79 	struct rds_msg_zcopy_info *info, *tmp;
80 
81 	spin_lock_irqsave(&q->lock, flags);
82 	list_splice(&q->zcookie_head, &copy);
83 	INIT_LIST_HEAD(&q->zcookie_head);
84 	spin_unlock_irqrestore(&q->lock, flags);
85 
86 	list_for_each_entry_safe(info, tmp, &copy, rs_zcookie_next) {
87 		list_del(&info->rs_zcookie_next);
88 		kfree(info);
89 	}
90 }
91 
92 static void rds_rm_zerocopy_callback(struct rds_sock *rs,
93 				     struct rds_znotifier *znotif)
94 {
95 	struct rds_msg_zcopy_info *info;
96 	struct rds_msg_zcopy_queue *q;
97 	u32 cookie = znotif->z_cookie;
98 	struct rds_zcopy_cookies *ck;
99 	struct list_head *head;
100 	unsigned long flags;
101 
102 	mm_unaccount_pinned_pages(&znotif->z_mmp);
103 	q = &rs->rs_zcookie_queue;
104 	spin_lock_irqsave(&q->lock, flags);
105 	head = &q->zcookie_head;
106 	if (!list_empty(head)) {
107 		info = list_entry(head, struct rds_msg_zcopy_info,
108 				  rs_zcookie_next);
109 		if (info && rds_zcookie_add(info, cookie)) {
110 			spin_unlock_irqrestore(&q->lock, flags);
111 			kfree(rds_info_from_znotifier(znotif));
112 			/* caller invokes rds_wake_sk_sleep() */
113 			return;
114 		}
115 	}
116 
117 	info = rds_info_from_znotifier(znotif);
118 	ck = &info->zcookies;
119 	memset(ck, 0, sizeof(*ck));
120 	WARN_ON(!rds_zcookie_add(info, cookie));
121 	list_add_tail(&q->zcookie_head, &info->rs_zcookie_next);
122 
123 	spin_unlock_irqrestore(&q->lock, flags);
124 	/* caller invokes rds_wake_sk_sleep() */
125 }
126 
127 /*
128  * This relies on dma_map_sg() not touching sg[].page during merging.
129  */
130 static void rds_message_purge(struct rds_message *rm)
131 {
132 	unsigned long i, flags;
133 	bool zcopy = false;
134 
135 	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
136 		return;
137 
138 	spin_lock_irqsave(&rm->m_rs_lock, flags);
139 	if (rm->m_rs) {
140 		struct rds_sock *rs = rm->m_rs;
141 
142 		if (rm->data.op_mmp_znotifier) {
143 			zcopy = true;
144 			rds_rm_zerocopy_callback(rs, rm->data.op_mmp_znotifier);
145 			rds_wake_sk_sleep(rs);
146 			rm->data.op_mmp_znotifier = NULL;
147 		}
148 		sock_put(rds_rs_to_sk(rs));
149 		rm->m_rs = NULL;
150 	}
151 	spin_unlock_irqrestore(&rm->m_rs_lock, flags);
152 
153 	for (i = 0; i < rm->data.op_nents; i++) {
154 		/* XXX will have to put_page for page refs */
155 		if (!zcopy)
156 			__free_page(sg_page(&rm->data.op_sg[i]));
157 		else
158 			put_page(sg_page(&rm->data.op_sg[i]));
159 	}
160 	rm->data.op_nents = 0;
161 
162 	if (rm->rdma.op_active)
163 		rds_rdma_free_op(&rm->rdma);
164 	if (rm->rdma.op_rdma_mr)
165 		kref_put(&rm->rdma.op_rdma_mr->r_kref, __rds_put_mr_final);
166 
167 	if (rm->atomic.op_active)
168 		rds_atomic_free_op(&rm->atomic);
169 	if (rm->atomic.op_rdma_mr)
170 		kref_put(&rm->atomic.op_rdma_mr->r_kref, __rds_put_mr_final);
171 }
172 
173 void rds_message_put(struct rds_message *rm)
174 {
175 	rdsdebug("put rm %p ref %d\n", rm, refcount_read(&rm->m_refcount));
176 	WARN(!refcount_read(&rm->m_refcount), "danger refcount zero on %p\n", rm);
177 	if (refcount_dec_and_test(&rm->m_refcount)) {
178 		BUG_ON(!list_empty(&rm->m_sock_item));
179 		BUG_ON(!list_empty(&rm->m_conn_item));
180 		rds_message_purge(rm);
181 
182 		kfree(rm);
183 	}
184 }
185 EXPORT_SYMBOL_GPL(rds_message_put);
186 
187 void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
188 				 __be16 dport, u64 seq)
189 {
190 	hdr->h_flags = 0;
191 	hdr->h_sport = sport;
192 	hdr->h_dport = dport;
193 	hdr->h_sequence = cpu_to_be64(seq);
194 	hdr->h_exthdr[0] = RDS_EXTHDR_NONE;
195 }
196 EXPORT_SYMBOL_GPL(rds_message_populate_header);
197 
198 int rds_message_add_extension(struct rds_header *hdr, unsigned int type,
199 			      const void *data, unsigned int len)
200 {
201 	unsigned int ext_len = sizeof(u8) + len;
202 	unsigned char *dst;
203 
204 	/* For now, refuse to add more than one extension header */
205 	if (hdr->h_exthdr[0] != RDS_EXTHDR_NONE)
206 		return 0;
207 
208 	if (type >= __RDS_EXTHDR_MAX || len != rds_exthdr_size[type])
209 		return 0;
210 
211 	if (ext_len >= RDS_HEADER_EXT_SPACE)
212 		return 0;
213 	dst = hdr->h_exthdr;
214 
215 	*dst++ = type;
216 	memcpy(dst, data, len);
217 
218 	dst[len] = RDS_EXTHDR_NONE;
219 	return 1;
220 }
221 EXPORT_SYMBOL_GPL(rds_message_add_extension);
222 
223 /*
224  * If a message has extension headers, retrieve them here.
225  * Call like this:
226  *
227  * unsigned int pos = 0;
228  *
229  * while (1) {
230  *	buflen = sizeof(buffer);
231  *	type = rds_message_next_extension(hdr, &pos, buffer, &buflen);
232  *	if (type == RDS_EXTHDR_NONE)
233  *		break;
234  *	...
235  * }
236  */
237 int rds_message_next_extension(struct rds_header *hdr,
238 		unsigned int *pos, void *buf, unsigned int *buflen)
239 {
240 	unsigned int offset, ext_type, ext_len;
241 	u8 *src = hdr->h_exthdr;
242 
243 	offset = *pos;
244 	if (offset >= RDS_HEADER_EXT_SPACE)
245 		goto none;
246 
247 	/* Get the extension type and length. For now, the
248 	 * length is implied by the extension type. */
249 	ext_type = src[offset++];
250 
251 	if (ext_type == RDS_EXTHDR_NONE || ext_type >= __RDS_EXTHDR_MAX)
252 		goto none;
253 	ext_len = rds_exthdr_size[ext_type];
254 	if (offset + ext_len > RDS_HEADER_EXT_SPACE)
255 		goto none;
256 
257 	*pos = offset + ext_len;
258 	if (ext_len < *buflen)
259 		*buflen = ext_len;
260 	memcpy(buf, src + offset, *buflen);
261 	return ext_type;
262 
263 none:
264 	*pos = RDS_HEADER_EXT_SPACE;
265 	*buflen = 0;
266 	return RDS_EXTHDR_NONE;
267 }
268 
269 int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset)
270 {
271 	struct rds_ext_header_rdma_dest ext_hdr;
272 
273 	ext_hdr.h_rdma_rkey = cpu_to_be32(r_key);
274 	ext_hdr.h_rdma_offset = cpu_to_be32(offset);
275 	return rds_message_add_extension(hdr, RDS_EXTHDR_RDMA_DEST, &ext_hdr, sizeof(ext_hdr));
276 }
277 EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension);
278 
279 /*
280  * Each rds_message is allocated with extra space for the scatterlist entries
281  * rds ops will need. This is to minimize memory allocation count. Then, each rds op
282  * can grab SGs when initializing its part of the rds_message.
283  */
284 struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp)
285 {
286 	struct rds_message *rm;
287 
288 	if (extra_len > KMALLOC_MAX_SIZE - sizeof(struct rds_message))
289 		return NULL;
290 
291 	rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp);
292 	if (!rm)
293 		goto out;
294 
295 	rm->m_used_sgs = 0;
296 	rm->m_total_sgs = extra_len / sizeof(struct scatterlist);
297 
298 	refcount_set(&rm->m_refcount, 1);
299 	INIT_LIST_HEAD(&rm->m_sock_item);
300 	INIT_LIST_HEAD(&rm->m_conn_item);
301 	spin_lock_init(&rm->m_rs_lock);
302 	init_waitqueue_head(&rm->m_flush_wait);
303 
304 out:
305 	return rm;
306 }
307 
308 /*
309  * RDS ops use this to grab SG entries from the rm's sg pool.
310  */
311 struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents)
312 {
313 	struct scatterlist *sg_first = (struct scatterlist *) &rm[1];
314 	struct scatterlist *sg_ret;
315 
316 	if (nents <= 0) {
317 		pr_warn("rds: alloc sgs failed! nents <= 0\n");
318 		return ERR_PTR(-EINVAL);
319 	}
320 
321 	if (rm->m_used_sgs + nents > rm->m_total_sgs) {
322 		pr_warn("rds: alloc sgs failed! total %d used %d nents %d\n",
323 			rm->m_total_sgs, rm->m_used_sgs, nents);
324 		return ERR_PTR(-ENOMEM);
325 	}
326 
327 	sg_ret = &sg_first[rm->m_used_sgs];
328 	sg_init_table(sg_ret, nents);
329 	rm->m_used_sgs += nents;
330 
331 	return sg_ret;
332 }
333 
334 struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len)
335 {
336 	struct rds_message *rm;
337 	unsigned int i;
338 	int num_sgs = DIV_ROUND_UP(total_len, PAGE_SIZE);
339 	int extra_bytes = num_sgs * sizeof(struct scatterlist);
340 
341 	rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
342 	if (!rm)
343 		return ERR_PTR(-ENOMEM);
344 
345 	set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
346 	rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
347 	rm->data.op_nents = DIV_ROUND_UP(total_len, PAGE_SIZE);
348 	rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
349 	if (IS_ERR(rm->data.op_sg)) {
350 		void *err = ERR_CAST(rm->data.op_sg);
351 		rds_message_put(rm);
352 		return err;
353 	}
354 
355 	for (i = 0; i < rm->data.op_nents; ++i) {
356 		sg_set_page(&rm->data.op_sg[i],
357 				virt_to_page(page_addrs[i]),
358 				PAGE_SIZE, 0);
359 	}
360 
361 	return rm;
362 }
363 
364 static int rds_message_zcopy_from_user(struct rds_message *rm, struct iov_iter *from)
365 {
366 	struct scatterlist *sg;
367 	int ret = 0;
368 	int length = iov_iter_count(from);
369 	int total_copied = 0;
370 	struct rds_msg_zcopy_info *info;
371 
372 	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));
373 
374 	/*
375 	 * now allocate and copy in the data payload.
376 	 */
377 	sg = rm->data.op_sg;
378 
379 	info = kzalloc(sizeof(*info), GFP_KERNEL);
380 	if (!info)
381 		return -ENOMEM;
382 	INIT_LIST_HEAD(&info->rs_zcookie_next);
383 	rm->data.op_mmp_znotifier = &info->znotif;
384 	if (mm_account_pinned_pages(&rm->data.op_mmp_znotifier->z_mmp,
385 				    length)) {
386 		ret = -ENOMEM;
387 		goto err;
388 	}
389 	while (iov_iter_count(from)) {
390 		struct page *pages;
391 		size_t start;
392 		ssize_t copied;
393 
394 		copied = iov_iter_get_pages2(from, &pages, PAGE_SIZE,
395 					    1, &start);
396 		if (copied < 0) {
397 			struct mmpin *mmp;
398 			int i;
399 
400 			for (i = 0; i < rm->data.op_nents; i++)
401 				put_page(sg_page(&rm->data.op_sg[i]));
402 			mmp = &rm->data.op_mmp_znotifier->z_mmp;
403 			mm_unaccount_pinned_pages(mmp);
404 			ret = -EFAULT;
405 			goto err;
406 		}
407 		total_copied += copied;
408 		length -= copied;
409 		sg_set_page(sg, pages, copied, start);
410 		rm->data.op_nents++;
411 		sg++;
412 	}
413 	WARN_ON_ONCE(length != 0);
414 	return ret;
415 err:
416 	kfree(info);
417 	rm->data.op_mmp_znotifier = NULL;
418 	return ret;
419 }
420 
421 int rds_message_copy_from_user(struct rds_message *rm, struct iov_iter *from,
422 			       bool zcopy)
423 {
424 	unsigned long to_copy, nbytes;
425 	unsigned long sg_off;
426 	struct scatterlist *sg;
427 	int ret = 0;
428 
429 	rm->m_inc.i_hdr.h_len = cpu_to_be32(iov_iter_count(from));
430 
431 	/* now allocate and copy in the data payload.  */
432 	sg = rm->data.op_sg;
433 	sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
434 
435 	if (zcopy)
436 		return rds_message_zcopy_from_user(rm, from);
437 
438 	while (iov_iter_count(from)) {
439 		if (!sg_page(sg)) {
440 			ret = rds_page_remainder_alloc(sg, iov_iter_count(from),
441 						       GFP_HIGHUSER);
442 			if (ret)
443 				return ret;
444 			rm->data.op_nents++;
445 			sg_off = 0;
446 		}
447 
448 		to_copy = min_t(unsigned long, iov_iter_count(from),
449 				sg->length - sg_off);
450 
451 		rds_stats_add(s_copy_from_user, to_copy);
452 		nbytes = copy_page_from_iter(sg_page(sg), sg->offset + sg_off,
453 					     to_copy, from);
454 		if (nbytes != to_copy)
455 			return -EFAULT;
456 
457 		sg_off += to_copy;
458 
459 		if (sg_off == sg->length)
460 			sg++;
461 	}
462 
463 	return ret;
464 }
465 
466 int rds_message_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to)
467 {
468 	struct rds_message *rm;
469 	struct scatterlist *sg;
470 	unsigned long to_copy;
471 	unsigned long vec_off;
472 	int copied;
473 	int ret;
474 	u32 len;
475 
476 	rm = container_of(inc, struct rds_message, m_inc);
477 	len = be32_to_cpu(rm->m_inc.i_hdr.h_len);
478 
479 	sg = rm->data.op_sg;
480 	vec_off = 0;
481 	copied = 0;
482 
483 	while (iov_iter_count(to) && copied < len) {
484 		to_copy = min_t(unsigned long, iov_iter_count(to),
485 				sg->length - vec_off);
486 		to_copy = min_t(unsigned long, to_copy, len - copied);
487 
488 		rds_stats_add(s_copy_to_user, to_copy);
489 		ret = copy_page_to_iter(sg_page(sg), sg->offset + vec_off,
490 					to_copy, to);
491 		if (ret != to_copy)
492 			return -EFAULT;
493 
494 		vec_off += to_copy;
495 		copied += to_copy;
496 
497 		if (vec_off == sg->length) {
498 			vec_off = 0;
499 			sg++;
500 		}
501 	}
502 
503 	return copied;
504 }
505 
506 /*
507  * If the message is still on the send queue, wait until the transport
508  * is done with it. This is particularly important for RDMA operations.
509  */
510 void rds_message_wait(struct rds_message *rm)
511 {
512 	wait_event_interruptible(rm->m_flush_wait,
513 			!test_bit(RDS_MSG_MAPPED, &rm->m_flags));
514 }
515 
516 void rds_message_unmapped(struct rds_message *rm)
517 {
518 	clear_bit(RDS_MSG_MAPPED, &rm->m_flags);
519 	wake_up_interruptible(&rm->m_flush_wait);
520 }
521 EXPORT_SYMBOL_GPL(rds_message_unmapped);
522