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/slab.h>
36 #include <linux/proc_fs.h>
37 
38 #include "rds.h"
39 
40 /*
41  * This file implements a getsockopt() call which copies a set of fixed
42  * sized structs into a user-specified buffer as a means of providing
43  * read-only information about RDS.
44  *
45  * For a given information source there are a given number of fixed sized
46  * structs at a given time.  The structs are only copied if the user-specified
47  * buffer is big enough.  The destination pages that make up the buffer
48  * are pinned for the duration of the copy.
49  *
50  * This gives us the following benefits:
51  *
52  * - simple implementation, no copy "position" across multiple calls
53  * - consistent snapshot of an info source
54  * - atomic copy works well with whatever locking info source has
55  * - one portable tool to get rds info across implementations
56  * - long-lived tool can get info without allocating
57  *
58  * at the following costs:
59  *
60  * - info source copy must be pinned, may be "large"
61  */
62 
63 struct rds_info_iterator {
64 	struct page **pages;
65 	void *addr;
66 	unsigned long offset;
67 };
68 
69 static DEFINE_SPINLOCK(rds_info_lock);
70 static rds_info_func rds_info_funcs[RDS_INFO_LAST - RDS_INFO_FIRST + 1];
71 
rds_info_register_func(int optname,rds_info_func func)72 void rds_info_register_func(int optname, rds_info_func func)
73 {
74 	int offset = optname - RDS_INFO_FIRST;
75 
76 	BUG_ON(optname < RDS_INFO_FIRST || optname > RDS_INFO_LAST);
77 
78 	spin_lock(&rds_info_lock);
79 	BUG_ON(rds_info_funcs[offset]);
80 	rds_info_funcs[offset] = func;
81 	spin_unlock(&rds_info_lock);
82 }
83 EXPORT_SYMBOL_GPL(rds_info_register_func);
84 
rds_info_deregister_func(int optname,rds_info_func func)85 void rds_info_deregister_func(int optname, rds_info_func func)
86 {
87 	int offset = optname - RDS_INFO_FIRST;
88 
89 	BUG_ON(optname < RDS_INFO_FIRST || optname > RDS_INFO_LAST);
90 
91 	spin_lock(&rds_info_lock);
92 	BUG_ON(rds_info_funcs[offset] != func);
93 	rds_info_funcs[offset] = NULL;
94 	spin_unlock(&rds_info_lock);
95 }
96 EXPORT_SYMBOL_GPL(rds_info_deregister_func);
97 
98 /*
99  * Typically we hold an atomic kmap across multiple rds_info_copy() calls
100  * because the kmap is so expensive.  This must be called before using blocking
101  * operations while holding the mapping and as the iterator is torn down.
102  */
rds_info_iter_unmap(struct rds_info_iterator * iter)103 void rds_info_iter_unmap(struct rds_info_iterator *iter)
104 {
105 	if (iter->addr) {
106 		kunmap_atomic(iter->addr, KM_USER0);
107 		iter->addr = NULL;
108 	}
109 }
110 
111 /*
112  * get_user_pages() called flush_dcache_page() on the pages for us.
113  */
rds_info_copy(struct rds_info_iterator * iter,void * data,unsigned long bytes)114 void rds_info_copy(struct rds_info_iterator *iter, void *data,
115 		   unsigned long bytes)
116 {
117 	unsigned long this;
118 
119 	while (bytes) {
120 		if (!iter->addr)
121 			iter->addr = kmap_atomic(*iter->pages, KM_USER0);
122 
123 		this = min(bytes, PAGE_SIZE - iter->offset);
124 
125 		rdsdebug("page %p addr %p offset %lu this %lu data %p "
126 			  "bytes %lu\n", *iter->pages, iter->addr,
127 			  iter->offset, this, data, bytes);
128 
129 		memcpy(iter->addr + iter->offset, data, this);
130 
131 		data += this;
132 		bytes -= this;
133 		iter->offset += this;
134 
135 		if (iter->offset == PAGE_SIZE) {
136 			kunmap_atomic(iter->addr, KM_USER0);
137 			iter->addr = NULL;
138 			iter->offset = 0;
139 			iter->pages++;
140 		}
141 	}
142 }
143 EXPORT_SYMBOL_GPL(rds_info_copy);
144 
145 /*
146  * @optval points to the userspace buffer that the information snapshot
147  * will be copied into.
148  *
149  * @optlen on input is the size of the buffer in userspace.  @optlen
150  * on output is the size of the requested snapshot in bytes.
151  *
152  * This function returns -errno if there is a failure, particularly -ENOSPC
153  * if the given userspace buffer was not large enough to fit the snapshot.
154  * On success it returns the positive number of bytes of each array element
155  * in the snapshot.
156  */
rds_info_getsockopt(struct socket * sock,int optname,char __user * optval,int __user * optlen)157 int rds_info_getsockopt(struct socket *sock, int optname, char __user *optval,
158 			int __user *optlen)
159 {
160 	struct rds_info_iterator iter;
161 	struct rds_info_lengths lens;
162 	unsigned long nr_pages = 0;
163 	unsigned long start;
164 	unsigned long i;
165 	rds_info_func func;
166 	struct page **pages = NULL;
167 	int ret;
168 	int len;
169 	int total;
170 
171 	if (get_user(len, optlen)) {
172 		ret = -EFAULT;
173 		goto out;
174 	}
175 
176 	/* check for all kinds of wrapping and the like */
177 	start = (unsigned long)optval;
178 	if (len < 0 || len + PAGE_SIZE - 1 < len || start + len < start) {
179 		ret = -EINVAL;
180 		goto out;
181 	}
182 
183 	/* a 0 len call is just trying to probe its length */
184 	if (len == 0)
185 		goto call_func;
186 
187 	nr_pages = (PAGE_ALIGN(start + len) - (start & PAGE_MASK))
188 			>> PAGE_SHIFT;
189 
190 	pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
191 	if (!pages) {
192 		ret = -ENOMEM;
193 		goto out;
194 	}
195 	ret = get_user_pages_fast(start, nr_pages, 1, pages);
196 	if (ret != nr_pages) {
197 		if (ret > 0)
198 			nr_pages = ret;
199 		else
200 			nr_pages = 0;
201 		ret = -EAGAIN; /* XXX ? */
202 		goto out;
203 	}
204 
205 	rdsdebug("len %d nr_pages %lu\n", len, nr_pages);
206 
207 call_func:
208 	func = rds_info_funcs[optname - RDS_INFO_FIRST];
209 	if (!func) {
210 		ret = -ENOPROTOOPT;
211 		goto out;
212 	}
213 
214 	iter.pages = pages;
215 	iter.addr = NULL;
216 	iter.offset = start & (PAGE_SIZE - 1);
217 
218 	func(sock, len, &iter, &lens);
219 	BUG_ON(lens.each == 0);
220 
221 	total = lens.nr * lens.each;
222 
223 	rds_info_iter_unmap(&iter);
224 
225 	if (total > len) {
226 		len = total;
227 		ret = -ENOSPC;
228 	} else {
229 		len = total;
230 		ret = lens.each;
231 	}
232 
233 	if (put_user(len, optlen))
234 		ret = -EFAULT;
235 
236 out:
237 	for (i = 0; pages && i < nr_pages; i++)
238 		put_page(pages[i]);
239 	kfree(pages);
240 
241 	return ret;
242 }
243