1 /*
2 * Copyright (c) 2006 Oracle. All rights reserved.
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/percpu.h>
34 #include <linux/seq_file.h>
35 #include <linux/srcu.h>
36 #include <linux/slab.h>
37 #include <linux/proc_fs.h>
38 #include <linux/export.h>
39 #include <linux/uio.h>
40
41 #include "rds.h"
42
43 /*
44 * This file implements a getsockopt() call which copies a set of fixed
45 * sized structs into a user-specified buffer as a means of providing
46 * read-only information about RDS.
47 *
48 * For a given information source there are a given number of fixed sized
49 * structs at a given time. The structs are only copied if the user-specified
50 * buffer is big enough. The destination pages that make up the buffer
51 * are pinned for the duration of the copy.
52 *
53 * This gives us the following benefits:
54 *
55 * - simple implementation, no copy "position" across multiple calls
56 * - consistent snapshot of an info source
57 * - atomic copy works well with whatever locking info source has
58 * - one portable tool to get rds info across implementations
59 * - long-lived tool can get info without allocating
60 *
61 * at the following costs:
62 *
63 * - info source copy must be pinned, may be "large"
64 */
65
66 struct rds_info_iterator {
67 struct page **pages;
68 void *addr;
69 unsigned long offset;
70 };
71
72 DEFINE_STATIC_SRCU(rds_info_srcu);
73 static DEFINE_SPINLOCK(rds_info_lock);
74 static rds_info_func rds_info_funcs[RDS_INFO_LAST - RDS_INFO_FIRST + 1];
75
rds_info_register_func(int optname,rds_info_func func)76 void rds_info_register_func(int optname, rds_info_func func)
77 {
78 int offset = optname - RDS_INFO_FIRST;
79
80 BUG_ON(optname < RDS_INFO_FIRST || optname > RDS_INFO_LAST);
81
82 spin_lock(&rds_info_lock);
83 if (WARN_ON_ONCE(rds_info_funcs[offset])) {
84 spin_unlock(&rds_info_lock);
85 return;
86 }
87 WRITE_ONCE(rds_info_funcs[offset], func);
88 spin_unlock(&rds_info_lock);
89 }
90 EXPORT_SYMBOL_GPL(rds_info_register_func);
91
rds_info_deregister_func(int optname,rds_info_func func)92 void rds_info_deregister_func(int optname, rds_info_func func)
93 {
94 int offset = optname - RDS_INFO_FIRST;
95
96 BUG_ON(optname < RDS_INFO_FIRST || optname > RDS_INFO_LAST);
97
98 spin_lock(&rds_info_lock);
99 if (WARN_ON_ONCE(rds_info_funcs[offset] != func)) {
100 spin_unlock(&rds_info_lock);
101 return;
102 }
103 WRITE_ONCE(rds_info_funcs[offset], NULL);
104 spin_unlock(&rds_info_lock);
105 synchronize_srcu(&rds_info_srcu);
106 }
107 EXPORT_SYMBOL_GPL(rds_info_deregister_func);
108
109 /*
110 * Typically we hold an atomic kmap across multiple rds_info_copy() calls
111 * because the kmap is so expensive. This must be called before using blocking
112 * operations while holding the mapping and as the iterator is torn down.
113 */
rds_info_iter_unmap(struct rds_info_iterator * iter)114 void rds_info_iter_unmap(struct rds_info_iterator *iter)
115 {
116 if (iter->addr) {
117 kunmap_atomic(iter->addr);
118 iter->addr = NULL;
119 }
120 }
121
122 /*
123 * get_user_pages() called flush_dcache_page() on the pages for us.
124 */
rds_info_copy(struct rds_info_iterator * iter,void * data,unsigned long bytes)125 void rds_info_copy(struct rds_info_iterator *iter, void *data,
126 unsigned long bytes)
127 {
128 unsigned long this;
129
130 while (bytes) {
131 if (!iter->addr)
132 iter->addr = kmap_atomic(*iter->pages);
133
134 this = min(bytes, PAGE_SIZE - iter->offset);
135
136 rdsdebug("page %p addr %p offset %lu this %lu data %p "
137 "bytes %lu\n", *iter->pages, iter->addr,
138 iter->offset, this, data, bytes);
139
140 memcpy(iter->addr + iter->offset, data, this);
141
142 data += this;
143 bytes -= this;
144 iter->offset += this;
145
146 if (iter->offset == PAGE_SIZE) {
147 kunmap_atomic(iter->addr);
148 iter->addr = NULL;
149 iter->offset = 0;
150 iter->pages++;
151 }
152 }
153 }
154 EXPORT_SYMBOL_GPL(rds_info_copy);
155
156 /*
157 * @opt->iter_out describes the buffer that the information snapshot will be
158 * copied into, and @opt->optlen is the size of that buffer on input. On
159 * output @opt->optlen is set to the size of the requested snapshot in bytes.
160 *
161 * This function returns -errno if there is a failure, particularly -ENOSPC
162 * if the given buffer was not large enough to fit the snapshot. On success
163 * it returns the positive number of bytes of each array element in the
164 * snapshot.
165 */
rds_info_getsockopt(struct socket * sock,int optname,sockopt_t * opt)166 int rds_info_getsockopt(struct socket *sock, int optname, sockopt_t *opt)
167 {
168 struct rds_info_iterator iter;
169 struct rds_info_lengths lens;
170 unsigned long nr_pages = 0;
171 rds_info_func func;
172 struct page **pages = NULL;
173 size_t offset0 = 0;
174 int srcu_idx;
175 int npages = 0;
176 int ret;
177 int len;
178 int total;
179
180 len = opt->optlen;
181
182 /* check for all kinds of wrapping and the like */
183 if (len < 0 || len > INT_MAX - PAGE_SIZE + 1) {
184 ret = -EINVAL;
185 goto out;
186 }
187
188 /* The info producers write into the pages with kmap_atomic() while
189 * holding a spinlock, so they need a genuine page-backed user buffer.
190 */
191 if (!user_backed_iter(&opt->iter_out)) {
192 ret = -EOPNOTSUPP;
193 goto out;
194 }
195
196 /* a 0 len call is just trying to probe its length */
197 if (len == 0)
198 goto call_func;
199
200 /*
201 * Preallocate the page array and pass it in so that
202 * iov_iter_extract_pages() fills it in place rather than allocating
203 * one for us. Handing it a non-NULL array keeps ownership of the
204 * array with us on every return path, instead of depending on the
205 * iterator code to allocate and hand it back.
206 */
207 npages = iov_iter_npages(&opt->iter_out, INT_MAX);
208 pages = kvmalloc_objs(*pages, npages);
209 if (!pages) {
210 ret = -ENOMEM;
211 goto out;
212 }
213
214 ret = iov_iter_extract_pages(&opt->iter_out, &pages, len, npages,
215 0, &offset0);
216 if (ret < 0)
217 goto out;
218 nr_pages = DIV_ROUND_UP(offset0 + ret, PAGE_SIZE);
219 if (ret != len) {
220 ret = -EAGAIN; /* XXX ? */
221 goto out;
222 }
223
224 rdsdebug("len %d nr_pages %lu\n", len, nr_pages);
225
226 call_func:
227 srcu_idx = srcu_read_lock(&rds_info_srcu);
228 func = READ_ONCE(rds_info_funcs[optname - RDS_INFO_FIRST]);
229 if (!func) {
230 srcu_read_unlock(&rds_info_srcu, srcu_idx);
231 ret = -ENOPROTOOPT;
232 goto out;
233 }
234
235 iter.pages = pages;
236 iter.addr = NULL;
237 iter.offset = offset0;
238
239 func(sock, len, &iter, &lens);
240 srcu_read_unlock(&rds_info_srcu, srcu_idx);
241 BUG_ON(lens.each == 0);
242
243 total = lens.nr * lens.each;
244
245 rds_info_iter_unmap(&iter);
246
247 if (total > len) {
248 len = total;
249 ret = -ENOSPC;
250 } else {
251 len = total;
252 ret = lens.each;
253 }
254
255 opt->optlen = len;
256
257 out:
258 /*
259 * iov_iter_extract_pages() pins only user-backed (ubuf) iters;
260 * iov_iter_extract_will_pin() reports whether an unpin is owed here.
261 */
262 if (pages && iov_iter_extract_will_pin(&opt->iter_out))
263 unpin_user_pages_dirty_lock(pages, nr_pages, true);
264 kvfree(pages);
265
266 return ret;
267 }
268