Commit f6fb3f6f591b50fa4f51962ad06ee0d8782e1bc8
Committed by
J. Bruce Fields
1 parent
9d548b9c95
Exists in
master
and in
7 other branches
SUNRPC: Fix up svc_unregister()
With the new rpcbind code, a PMAP_UNSET will not have any effect on services registered via rpcbind v3 or v4. Implement a version of svc_unregister() that uses an RPCB_UNSET with an empty netid string to make sure we have cleared *all* entries for a kernel RPC service when shutting down, or before starting a fresh instance of the service. Use the new version only when CONFIG_SUNRPC_REGISTER_V4 is enabled; otherwise, the legacy PMAP version is used to ensure complete backwards-compatibility with the Linux portmapper daemon. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Showing 1 changed file with 38 additions and 20 deletions Inline Diff
net/sunrpc/svc.c
1 | /* | 1 | /* |
2 | * linux/net/sunrpc/svc.c | 2 | * linux/net/sunrpc/svc.c |
3 | * | 3 | * |
4 | * High-level RPC service routines | 4 | * High-level RPC service routines |
5 | * | 5 | * |
6 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> | 6 | * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> |
7 | * | 7 | * |
8 | * Multiple threads pools and NUMAisation | 8 | * Multiple threads pools and NUMAisation |
9 | * Copyright (c) 2006 Silicon Graphics, Inc. | 9 | * Copyright (c) 2006 Silicon Graphics, Inc. |
10 | * by Greg Banks <gnb@melbourne.sgi.com> | 10 | * by Greg Banks <gnb@melbourne.sgi.com> |
11 | */ | 11 | */ |
12 | 12 | ||
13 | #include <linux/linkage.h> | 13 | #include <linux/linkage.h> |
14 | #include <linux/sched.h> | 14 | #include <linux/sched.h> |
15 | #include <linux/errno.h> | 15 | #include <linux/errno.h> |
16 | #include <linux/net.h> | 16 | #include <linux/net.h> |
17 | #include <linux/in.h> | 17 | #include <linux/in.h> |
18 | #include <linux/mm.h> | 18 | #include <linux/mm.h> |
19 | #include <linux/interrupt.h> | 19 | #include <linux/interrupt.h> |
20 | #include <linux/module.h> | 20 | #include <linux/module.h> |
21 | #include <linux/kthread.h> | 21 | #include <linux/kthread.h> |
22 | 22 | ||
23 | #include <linux/sunrpc/types.h> | 23 | #include <linux/sunrpc/types.h> |
24 | #include <linux/sunrpc/xdr.h> | 24 | #include <linux/sunrpc/xdr.h> |
25 | #include <linux/sunrpc/stats.h> | 25 | #include <linux/sunrpc/stats.h> |
26 | #include <linux/sunrpc/svcsock.h> | 26 | #include <linux/sunrpc/svcsock.h> |
27 | #include <linux/sunrpc/clnt.h> | 27 | #include <linux/sunrpc/clnt.h> |
28 | 28 | ||
29 | #define RPCDBG_FACILITY RPCDBG_SVCDSP | 29 | #define RPCDBG_FACILITY RPCDBG_SVCDSP |
30 | 30 | ||
31 | static void svc_unregister(const struct svc_serv *serv); | 31 | static void svc_unregister(const struct svc_serv *serv); |
32 | 32 | ||
33 | #define svc_serv_is_pooled(serv) ((serv)->sv_function) | 33 | #define svc_serv_is_pooled(serv) ((serv)->sv_function) |
34 | 34 | ||
35 | /* | 35 | /* |
36 | * Mode for mapping cpus to pools. | 36 | * Mode for mapping cpus to pools. |
37 | */ | 37 | */ |
38 | enum { | 38 | enum { |
39 | SVC_POOL_AUTO = -1, /* choose one of the others */ | 39 | SVC_POOL_AUTO = -1, /* choose one of the others */ |
40 | SVC_POOL_GLOBAL, /* no mapping, just a single global pool | 40 | SVC_POOL_GLOBAL, /* no mapping, just a single global pool |
41 | * (legacy & UP mode) */ | 41 | * (legacy & UP mode) */ |
42 | SVC_POOL_PERCPU, /* one pool per cpu */ | 42 | SVC_POOL_PERCPU, /* one pool per cpu */ |
43 | SVC_POOL_PERNODE /* one pool per numa node */ | 43 | SVC_POOL_PERNODE /* one pool per numa node */ |
44 | }; | 44 | }; |
45 | #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL | 45 | #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL |
46 | 46 | ||
47 | /* | 47 | /* |
48 | * Structure for mapping cpus to pools and vice versa. | 48 | * Structure for mapping cpus to pools and vice versa. |
49 | * Setup once during sunrpc initialisation. | 49 | * Setup once during sunrpc initialisation. |
50 | */ | 50 | */ |
51 | static struct svc_pool_map { | 51 | static struct svc_pool_map { |
52 | int count; /* How many svc_servs use us */ | 52 | int count; /* How many svc_servs use us */ |
53 | int mode; /* Note: int not enum to avoid | 53 | int mode; /* Note: int not enum to avoid |
54 | * warnings about "enumeration value | 54 | * warnings about "enumeration value |
55 | * not handled in switch" */ | 55 | * not handled in switch" */ |
56 | unsigned int npools; | 56 | unsigned int npools; |
57 | unsigned int *pool_to; /* maps pool id to cpu or node */ | 57 | unsigned int *pool_to; /* maps pool id to cpu or node */ |
58 | unsigned int *to_pool; /* maps cpu or node to pool id */ | 58 | unsigned int *to_pool; /* maps cpu or node to pool id */ |
59 | } svc_pool_map = { | 59 | } svc_pool_map = { |
60 | .count = 0, | 60 | .count = 0, |
61 | .mode = SVC_POOL_DEFAULT | 61 | .mode = SVC_POOL_DEFAULT |
62 | }; | 62 | }; |
63 | static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ | 63 | static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ |
64 | 64 | ||
65 | static int | 65 | static int |
66 | param_set_pool_mode(const char *val, struct kernel_param *kp) | 66 | param_set_pool_mode(const char *val, struct kernel_param *kp) |
67 | { | 67 | { |
68 | int *ip = (int *)kp->arg; | 68 | int *ip = (int *)kp->arg; |
69 | struct svc_pool_map *m = &svc_pool_map; | 69 | struct svc_pool_map *m = &svc_pool_map; |
70 | int err; | 70 | int err; |
71 | 71 | ||
72 | mutex_lock(&svc_pool_map_mutex); | 72 | mutex_lock(&svc_pool_map_mutex); |
73 | 73 | ||
74 | err = -EBUSY; | 74 | err = -EBUSY; |
75 | if (m->count) | 75 | if (m->count) |
76 | goto out; | 76 | goto out; |
77 | 77 | ||
78 | err = 0; | 78 | err = 0; |
79 | if (!strncmp(val, "auto", 4)) | 79 | if (!strncmp(val, "auto", 4)) |
80 | *ip = SVC_POOL_AUTO; | 80 | *ip = SVC_POOL_AUTO; |
81 | else if (!strncmp(val, "global", 6)) | 81 | else if (!strncmp(val, "global", 6)) |
82 | *ip = SVC_POOL_GLOBAL; | 82 | *ip = SVC_POOL_GLOBAL; |
83 | else if (!strncmp(val, "percpu", 6)) | 83 | else if (!strncmp(val, "percpu", 6)) |
84 | *ip = SVC_POOL_PERCPU; | 84 | *ip = SVC_POOL_PERCPU; |
85 | else if (!strncmp(val, "pernode", 7)) | 85 | else if (!strncmp(val, "pernode", 7)) |
86 | *ip = SVC_POOL_PERNODE; | 86 | *ip = SVC_POOL_PERNODE; |
87 | else | 87 | else |
88 | err = -EINVAL; | 88 | err = -EINVAL; |
89 | 89 | ||
90 | out: | 90 | out: |
91 | mutex_unlock(&svc_pool_map_mutex); | 91 | mutex_unlock(&svc_pool_map_mutex); |
92 | return err; | 92 | return err; |
93 | } | 93 | } |
94 | 94 | ||
95 | static int | 95 | static int |
96 | param_get_pool_mode(char *buf, struct kernel_param *kp) | 96 | param_get_pool_mode(char *buf, struct kernel_param *kp) |
97 | { | 97 | { |
98 | int *ip = (int *)kp->arg; | 98 | int *ip = (int *)kp->arg; |
99 | 99 | ||
100 | switch (*ip) | 100 | switch (*ip) |
101 | { | 101 | { |
102 | case SVC_POOL_AUTO: | 102 | case SVC_POOL_AUTO: |
103 | return strlcpy(buf, "auto", 20); | 103 | return strlcpy(buf, "auto", 20); |
104 | case SVC_POOL_GLOBAL: | 104 | case SVC_POOL_GLOBAL: |
105 | return strlcpy(buf, "global", 20); | 105 | return strlcpy(buf, "global", 20); |
106 | case SVC_POOL_PERCPU: | 106 | case SVC_POOL_PERCPU: |
107 | return strlcpy(buf, "percpu", 20); | 107 | return strlcpy(buf, "percpu", 20); |
108 | case SVC_POOL_PERNODE: | 108 | case SVC_POOL_PERNODE: |
109 | return strlcpy(buf, "pernode", 20); | 109 | return strlcpy(buf, "pernode", 20); |
110 | default: | 110 | default: |
111 | return sprintf(buf, "%d", *ip); | 111 | return sprintf(buf, "%d", *ip); |
112 | } | 112 | } |
113 | } | 113 | } |
114 | 114 | ||
115 | module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, | 115 | module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, |
116 | &svc_pool_map.mode, 0644); | 116 | &svc_pool_map.mode, 0644); |
117 | 117 | ||
118 | /* | 118 | /* |
119 | * Detect best pool mapping mode heuristically, | 119 | * Detect best pool mapping mode heuristically, |
120 | * according to the machine's topology. | 120 | * according to the machine's topology. |
121 | */ | 121 | */ |
122 | static int | 122 | static int |
123 | svc_pool_map_choose_mode(void) | 123 | svc_pool_map_choose_mode(void) |
124 | { | 124 | { |
125 | unsigned int node; | 125 | unsigned int node; |
126 | 126 | ||
127 | if (num_online_nodes() > 1) { | 127 | if (num_online_nodes() > 1) { |
128 | /* | 128 | /* |
129 | * Actually have multiple NUMA nodes, | 129 | * Actually have multiple NUMA nodes, |
130 | * so split pools on NUMA node boundaries | 130 | * so split pools on NUMA node boundaries |
131 | */ | 131 | */ |
132 | return SVC_POOL_PERNODE; | 132 | return SVC_POOL_PERNODE; |
133 | } | 133 | } |
134 | 134 | ||
135 | node = any_online_node(node_online_map); | 135 | node = any_online_node(node_online_map); |
136 | if (nr_cpus_node(node) > 2) { | 136 | if (nr_cpus_node(node) > 2) { |
137 | /* | 137 | /* |
138 | * Non-trivial SMP, or CONFIG_NUMA on | 138 | * Non-trivial SMP, or CONFIG_NUMA on |
139 | * non-NUMA hardware, e.g. with a generic | 139 | * non-NUMA hardware, e.g. with a generic |
140 | * x86_64 kernel on Xeons. In this case we | 140 | * x86_64 kernel on Xeons. In this case we |
141 | * want to divide the pools on cpu boundaries. | 141 | * want to divide the pools on cpu boundaries. |
142 | */ | 142 | */ |
143 | return SVC_POOL_PERCPU; | 143 | return SVC_POOL_PERCPU; |
144 | } | 144 | } |
145 | 145 | ||
146 | /* default: one global pool */ | 146 | /* default: one global pool */ |
147 | return SVC_POOL_GLOBAL; | 147 | return SVC_POOL_GLOBAL; |
148 | } | 148 | } |
149 | 149 | ||
150 | /* | 150 | /* |
151 | * Allocate the to_pool[] and pool_to[] arrays. | 151 | * Allocate the to_pool[] and pool_to[] arrays. |
152 | * Returns 0 on success or an errno. | 152 | * Returns 0 on success or an errno. |
153 | */ | 153 | */ |
154 | static int | 154 | static int |
155 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) | 155 | svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) |
156 | { | 156 | { |
157 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | 157 | m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); |
158 | if (!m->to_pool) | 158 | if (!m->to_pool) |
159 | goto fail; | 159 | goto fail; |
160 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); | 160 | m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); |
161 | if (!m->pool_to) | 161 | if (!m->pool_to) |
162 | goto fail_free; | 162 | goto fail_free; |
163 | 163 | ||
164 | return 0; | 164 | return 0; |
165 | 165 | ||
166 | fail_free: | 166 | fail_free: |
167 | kfree(m->to_pool); | 167 | kfree(m->to_pool); |
168 | fail: | 168 | fail: |
169 | return -ENOMEM; | 169 | return -ENOMEM; |
170 | } | 170 | } |
171 | 171 | ||
172 | /* | 172 | /* |
173 | * Initialise the pool map for SVC_POOL_PERCPU mode. | 173 | * Initialise the pool map for SVC_POOL_PERCPU mode. |
174 | * Returns number of pools or <0 on error. | 174 | * Returns number of pools or <0 on error. |
175 | */ | 175 | */ |
176 | static int | 176 | static int |
177 | svc_pool_map_init_percpu(struct svc_pool_map *m) | 177 | svc_pool_map_init_percpu(struct svc_pool_map *m) |
178 | { | 178 | { |
179 | unsigned int maxpools = nr_cpu_ids; | 179 | unsigned int maxpools = nr_cpu_ids; |
180 | unsigned int pidx = 0; | 180 | unsigned int pidx = 0; |
181 | unsigned int cpu; | 181 | unsigned int cpu; |
182 | int err; | 182 | int err; |
183 | 183 | ||
184 | err = svc_pool_map_alloc_arrays(m, maxpools); | 184 | err = svc_pool_map_alloc_arrays(m, maxpools); |
185 | if (err) | 185 | if (err) |
186 | return err; | 186 | return err; |
187 | 187 | ||
188 | for_each_online_cpu(cpu) { | 188 | for_each_online_cpu(cpu) { |
189 | BUG_ON(pidx > maxpools); | 189 | BUG_ON(pidx > maxpools); |
190 | m->to_pool[cpu] = pidx; | 190 | m->to_pool[cpu] = pidx; |
191 | m->pool_to[pidx] = cpu; | 191 | m->pool_to[pidx] = cpu; |
192 | pidx++; | 192 | pidx++; |
193 | } | 193 | } |
194 | /* cpus brought online later all get mapped to pool0, sorry */ | 194 | /* cpus brought online later all get mapped to pool0, sorry */ |
195 | 195 | ||
196 | return pidx; | 196 | return pidx; |
197 | }; | 197 | }; |
198 | 198 | ||
199 | 199 | ||
200 | /* | 200 | /* |
201 | * Initialise the pool map for SVC_POOL_PERNODE mode. | 201 | * Initialise the pool map for SVC_POOL_PERNODE mode. |
202 | * Returns number of pools or <0 on error. | 202 | * Returns number of pools or <0 on error. |
203 | */ | 203 | */ |
204 | static int | 204 | static int |
205 | svc_pool_map_init_pernode(struct svc_pool_map *m) | 205 | svc_pool_map_init_pernode(struct svc_pool_map *m) |
206 | { | 206 | { |
207 | unsigned int maxpools = nr_node_ids; | 207 | unsigned int maxpools = nr_node_ids; |
208 | unsigned int pidx = 0; | 208 | unsigned int pidx = 0; |
209 | unsigned int node; | 209 | unsigned int node; |
210 | int err; | 210 | int err; |
211 | 211 | ||
212 | err = svc_pool_map_alloc_arrays(m, maxpools); | 212 | err = svc_pool_map_alloc_arrays(m, maxpools); |
213 | if (err) | 213 | if (err) |
214 | return err; | 214 | return err; |
215 | 215 | ||
216 | for_each_node_with_cpus(node) { | 216 | for_each_node_with_cpus(node) { |
217 | /* some architectures (e.g. SN2) have cpuless nodes */ | 217 | /* some architectures (e.g. SN2) have cpuless nodes */ |
218 | BUG_ON(pidx > maxpools); | 218 | BUG_ON(pidx > maxpools); |
219 | m->to_pool[node] = pidx; | 219 | m->to_pool[node] = pidx; |
220 | m->pool_to[pidx] = node; | 220 | m->pool_to[pidx] = node; |
221 | pidx++; | 221 | pidx++; |
222 | } | 222 | } |
223 | /* nodes brought online later all get mapped to pool0, sorry */ | 223 | /* nodes brought online later all get mapped to pool0, sorry */ |
224 | 224 | ||
225 | return pidx; | 225 | return pidx; |
226 | } | 226 | } |
227 | 227 | ||
228 | 228 | ||
229 | /* | 229 | /* |
230 | * Add a reference to the global map of cpus to pools (and | 230 | * Add a reference to the global map of cpus to pools (and |
231 | * vice versa). Initialise the map if we're the first user. | 231 | * vice versa). Initialise the map if we're the first user. |
232 | * Returns the number of pools. | 232 | * Returns the number of pools. |
233 | */ | 233 | */ |
234 | static unsigned int | 234 | static unsigned int |
235 | svc_pool_map_get(void) | 235 | svc_pool_map_get(void) |
236 | { | 236 | { |
237 | struct svc_pool_map *m = &svc_pool_map; | 237 | struct svc_pool_map *m = &svc_pool_map; |
238 | int npools = -1; | 238 | int npools = -1; |
239 | 239 | ||
240 | mutex_lock(&svc_pool_map_mutex); | 240 | mutex_lock(&svc_pool_map_mutex); |
241 | 241 | ||
242 | if (m->count++) { | 242 | if (m->count++) { |
243 | mutex_unlock(&svc_pool_map_mutex); | 243 | mutex_unlock(&svc_pool_map_mutex); |
244 | return m->npools; | 244 | return m->npools; |
245 | } | 245 | } |
246 | 246 | ||
247 | if (m->mode == SVC_POOL_AUTO) | 247 | if (m->mode == SVC_POOL_AUTO) |
248 | m->mode = svc_pool_map_choose_mode(); | 248 | m->mode = svc_pool_map_choose_mode(); |
249 | 249 | ||
250 | switch (m->mode) { | 250 | switch (m->mode) { |
251 | case SVC_POOL_PERCPU: | 251 | case SVC_POOL_PERCPU: |
252 | npools = svc_pool_map_init_percpu(m); | 252 | npools = svc_pool_map_init_percpu(m); |
253 | break; | 253 | break; |
254 | case SVC_POOL_PERNODE: | 254 | case SVC_POOL_PERNODE: |
255 | npools = svc_pool_map_init_pernode(m); | 255 | npools = svc_pool_map_init_pernode(m); |
256 | break; | 256 | break; |
257 | } | 257 | } |
258 | 258 | ||
259 | if (npools < 0) { | 259 | if (npools < 0) { |
260 | /* default, or memory allocation failure */ | 260 | /* default, or memory allocation failure */ |
261 | npools = 1; | 261 | npools = 1; |
262 | m->mode = SVC_POOL_GLOBAL; | 262 | m->mode = SVC_POOL_GLOBAL; |
263 | } | 263 | } |
264 | m->npools = npools; | 264 | m->npools = npools; |
265 | 265 | ||
266 | mutex_unlock(&svc_pool_map_mutex); | 266 | mutex_unlock(&svc_pool_map_mutex); |
267 | return m->npools; | 267 | return m->npools; |
268 | } | 268 | } |
269 | 269 | ||
270 | 270 | ||
271 | /* | 271 | /* |
272 | * Drop a reference to the global map of cpus to pools. | 272 | * Drop a reference to the global map of cpus to pools. |
273 | * When the last reference is dropped, the map data is | 273 | * When the last reference is dropped, the map data is |
274 | * freed; this allows the sysadmin to change the pool | 274 | * freed; this allows the sysadmin to change the pool |
275 | * mode using the pool_mode module option without | 275 | * mode using the pool_mode module option without |
276 | * rebooting or re-loading sunrpc.ko. | 276 | * rebooting or re-loading sunrpc.ko. |
277 | */ | 277 | */ |
278 | static void | 278 | static void |
279 | svc_pool_map_put(void) | 279 | svc_pool_map_put(void) |
280 | { | 280 | { |
281 | struct svc_pool_map *m = &svc_pool_map; | 281 | struct svc_pool_map *m = &svc_pool_map; |
282 | 282 | ||
283 | mutex_lock(&svc_pool_map_mutex); | 283 | mutex_lock(&svc_pool_map_mutex); |
284 | 284 | ||
285 | if (!--m->count) { | 285 | if (!--m->count) { |
286 | m->mode = SVC_POOL_DEFAULT; | 286 | m->mode = SVC_POOL_DEFAULT; |
287 | kfree(m->to_pool); | 287 | kfree(m->to_pool); |
288 | kfree(m->pool_to); | 288 | kfree(m->pool_to); |
289 | m->npools = 0; | 289 | m->npools = 0; |
290 | } | 290 | } |
291 | 291 | ||
292 | mutex_unlock(&svc_pool_map_mutex); | 292 | mutex_unlock(&svc_pool_map_mutex); |
293 | } | 293 | } |
294 | 294 | ||
295 | 295 | ||
296 | /* | 296 | /* |
297 | * Set the given thread's cpus_allowed mask so that it | 297 | * Set the given thread's cpus_allowed mask so that it |
298 | * will only run on cpus in the given pool. | 298 | * will only run on cpus in the given pool. |
299 | */ | 299 | */ |
300 | static inline void | 300 | static inline void |
301 | svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) | 301 | svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) |
302 | { | 302 | { |
303 | struct svc_pool_map *m = &svc_pool_map; | 303 | struct svc_pool_map *m = &svc_pool_map; |
304 | unsigned int node = m->pool_to[pidx]; | 304 | unsigned int node = m->pool_to[pidx]; |
305 | 305 | ||
306 | /* | 306 | /* |
307 | * The caller checks for sv_nrpools > 1, which | 307 | * The caller checks for sv_nrpools > 1, which |
308 | * implies that we've been initialized. | 308 | * implies that we've been initialized. |
309 | */ | 309 | */ |
310 | BUG_ON(m->count == 0); | 310 | BUG_ON(m->count == 0); |
311 | 311 | ||
312 | switch (m->mode) { | 312 | switch (m->mode) { |
313 | case SVC_POOL_PERCPU: | 313 | case SVC_POOL_PERCPU: |
314 | { | 314 | { |
315 | set_cpus_allowed_ptr(task, &cpumask_of_cpu(node)); | 315 | set_cpus_allowed_ptr(task, &cpumask_of_cpu(node)); |
316 | break; | 316 | break; |
317 | } | 317 | } |
318 | case SVC_POOL_PERNODE: | 318 | case SVC_POOL_PERNODE: |
319 | { | 319 | { |
320 | node_to_cpumask_ptr(nodecpumask, node); | 320 | node_to_cpumask_ptr(nodecpumask, node); |
321 | set_cpus_allowed_ptr(task, nodecpumask); | 321 | set_cpus_allowed_ptr(task, nodecpumask); |
322 | break; | 322 | break; |
323 | } | 323 | } |
324 | } | 324 | } |
325 | } | 325 | } |
326 | 326 | ||
327 | /* | 327 | /* |
328 | * Use the mapping mode to choose a pool for a given CPU. | 328 | * Use the mapping mode to choose a pool for a given CPU. |
329 | * Used when enqueueing an incoming RPC. Always returns | 329 | * Used when enqueueing an incoming RPC. Always returns |
330 | * a non-NULL pool pointer. | 330 | * a non-NULL pool pointer. |
331 | */ | 331 | */ |
332 | struct svc_pool * | 332 | struct svc_pool * |
333 | svc_pool_for_cpu(struct svc_serv *serv, int cpu) | 333 | svc_pool_for_cpu(struct svc_serv *serv, int cpu) |
334 | { | 334 | { |
335 | struct svc_pool_map *m = &svc_pool_map; | 335 | struct svc_pool_map *m = &svc_pool_map; |
336 | unsigned int pidx = 0; | 336 | unsigned int pidx = 0; |
337 | 337 | ||
338 | /* | 338 | /* |
339 | * An uninitialised map happens in a pure client when | 339 | * An uninitialised map happens in a pure client when |
340 | * lockd is brought up, so silently treat it the | 340 | * lockd is brought up, so silently treat it the |
341 | * same as SVC_POOL_GLOBAL. | 341 | * same as SVC_POOL_GLOBAL. |
342 | */ | 342 | */ |
343 | if (svc_serv_is_pooled(serv)) { | 343 | if (svc_serv_is_pooled(serv)) { |
344 | switch (m->mode) { | 344 | switch (m->mode) { |
345 | case SVC_POOL_PERCPU: | 345 | case SVC_POOL_PERCPU: |
346 | pidx = m->to_pool[cpu]; | 346 | pidx = m->to_pool[cpu]; |
347 | break; | 347 | break; |
348 | case SVC_POOL_PERNODE: | 348 | case SVC_POOL_PERNODE: |
349 | pidx = m->to_pool[cpu_to_node(cpu)]; | 349 | pidx = m->to_pool[cpu_to_node(cpu)]; |
350 | break; | 350 | break; |
351 | } | 351 | } |
352 | } | 352 | } |
353 | return &serv->sv_pools[pidx % serv->sv_nrpools]; | 353 | return &serv->sv_pools[pidx % serv->sv_nrpools]; |
354 | } | 354 | } |
355 | 355 | ||
356 | 356 | ||
357 | /* | 357 | /* |
358 | * Create an RPC service | 358 | * Create an RPC service |
359 | */ | 359 | */ |
360 | static struct svc_serv * | 360 | static struct svc_serv * |
361 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, | 361 | __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, |
362 | sa_family_t family, void (*shutdown)(struct svc_serv *serv)) | 362 | sa_family_t family, void (*shutdown)(struct svc_serv *serv)) |
363 | { | 363 | { |
364 | struct svc_serv *serv; | 364 | struct svc_serv *serv; |
365 | unsigned int vers; | 365 | unsigned int vers; |
366 | unsigned int xdrsize; | 366 | unsigned int xdrsize; |
367 | unsigned int i; | 367 | unsigned int i; |
368 | 368 | ||
369 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) | 369 | if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) |
370 | return NULL; | 370 | return NULL; |
371 | serv->sv_family = family; | 371 | serv->sv_family = family; |
372 | serv->sv_name = prog->pg_name; | 372 | serv->sv_name = prog->pg_name; |
373 | serv->sv_program = prog; | 373 | serv->sv_program = prog; |
374 | serv->sv_nrthreads = 1; | 374 | serv->sv_nrthreads = 1; |
375 | serv->sv_stats = prog->pg_stats; | 375 | serv->sv_stats = prog->pg_stats; |
376 | if (bufsize > RPCSVC_MAXPAYLOAD) | 376 | if (bufsize > RPCSVC_MAXPAYLOAD) |
377 | bufsize = RPCSVC_MAXPAYLOAD; | 377 | bufsize = RPCSVC_MAXPAYLOAD; |
378 | serv->sv_max_payload = bufsize? bufsize : 4096; | 378 | serv->sv_max_payload = bufsize? bufsize : 4096; |
379 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); | 379 | serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); |
380 | serv->sv_shutdown = shutdown; | 380 | serv->sv_shutdown = shutdown; |
381 | xdrsize = 0; | 381 | xdrsize = 0; |
382 | while (prog) { | 382 | while (prog) { |
383 | prog->pg_lovers = prog->pg_nvers-1; | 383 | prog->pg_lovers = prog->pg_nvers-1; |
384 | for (vers=0; vers<prog->pg_nvers ; vers++) | 384 | for (vers=0; vers<prog->pg_nvers ; vers++) |
385 | if (prog->pg_vers[vers]) { | 385 | if (prog->pg_vers[vers]) { |
386 | prog->pg_hivers = vers; | 386 | prog->pg_hivers = vers; |
387 | if (prog->pg_lovers > vers) | 387 | if (prog->pg_lovers > vers) |
388 | prog->pg_lovers = vers; | 388 | prog->pg_lovers = vers; |
389 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) | 389 | if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) |
390 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; | 390 | xdrsize = prog->pg_vers[vers]->vs_xdrsize; |
391 | } | 391 | } |
392 | prog = prog->pg_next; | 392 | prog = prog->pg_next; |
393 | } | 393 | } |
394 | serv->sv_xdrsize = xdrsize; | 394 | serv->sv_xdrsize = xdrsize; |
395 | INIT_LIST_HEAD(&serv->sv_tempsocks); | 395 | INIT_LIST_HEAD(&serv->sv_tempsocks); |
396 | INIT_LIST_HEAD(&serv->sv_permsocks); | 396 | INIT_LIST_HEAD(&serv->sv_permsocks); |
397 | init_timer(&serv->sv_temptimer); | 397 | init_timer(&serv->sv_temptimer); |
398 | spin_lock_init(&serv->sv_lock); | 398 | spin_lock_init(&serv->sv_lock); |
399 | 399 | ||
400 | serv->sv_nrpools = npools; | 400 | serv->sv_nrpools = npools; |
401 | serv->sv_pools = | 401 | serv->sv_pools = |
402 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), | 402 | kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), |
403 | GFP_KERNEL); | 403 | GFP_KERNEL); |
404 | if (!serv->sv_pools) { | 404 | if (!serv->sv_pools) { |
405 | kfree(serv); | 405 | kfree(serv); |
406 | return NULL; | 406 | return NULL; |
407 | } | 407 | } |
408 | 408 | ||
409 | for (i = 0; i < serv->sv_nrpools; i++) { | 409 | for (i = 0; i < serv->sv_nrpools; i++) { |
410 | struct svc_pool *pool = &serv->sv_pools[i]; | 410 | struct svc_pool *pool = &serv->sv_pools[i]; |
411 | 411 | ||
412 | dprintk("svc: initialising pool %u for %s\n", | 412 | dprintk("svc: initialising pool %u for %s\n", |
413 | i, serv->sv_name); | 413 | i, serv->sv_name); |
414 | 414 | ||
415 | pool->sp_id = i; | 415 | pool->sp_id = i; |
416 | INIT_LIST_HEAD(&pool->sp_threads); | 416 | INIT_LIST_HEAD(&pool->sp_threads); |
417 | INIT_LIST_HEAD(&pool->sp_sockets); | 417 | INIT_LIST_HEAD(&pool->sp_sockets); |
418 | INIT_LIST_HEAD(&pool->sp_all_threads); | 418 | INIT_LIST_HEAD(&pool->sp_all_threads); |
419 | spin_lock_init(&pool->sp_lock); | 419 | spin_lock_init(&pool->sp_lock); |
420 | } | 420 | } |
421 | 421 | ||
422 | /* Remove any stale portmap registrations */ | 422 | /* Remove any stale portmap registrations */ |
423 | svc_unregister(serv); | 423 | svc_unregister(serv); |
424 | 424 | ||
425 | return serv; | 425 | return serv; |
426 | } | 426 | } |
427 | 427 | ||
428 | struct svc_serv * | 428 | struct svc_serv * |
429 | svc_create(struct svc_program *prog, unsigned int bufsize, | 429 | svc_create(struct svc_program *prog, unsigned int bufsize, |
430 | sa_family_t family, void (*shutdown)(struct svc_serv *serv)) | 430 | sa_family_t family, void (*shutdown)(struct svc_serv *serv)) |
431 | { | 431 | { |
432 | return __svc_create(prog, bufsize, /*npools*/1, family, shutdown); | 432 | return __svc_create(prog, bufsize, /*npools*/1, family, shutdown); |
433 | } | 433 | } |
434 | EXPORT_SYMBOL(svc_create); | 434 | EXPORT_SYMBOL(svc_create); |
435 | 435 | ||
436 | struct svc_serv * | 436 | struct svc_serv * |
437 | svc_create_pooled(struct svc_program *prog, unsigned int bufsize, | 437 | svc_create_pooled(struct svc_program *prog, unsigned int bufsize, |
438 | sa_family_t family, void (*shutdown)(struct svc_serv *serv), | 438 | sa_family_t family, void (*shutdown)(struct svc_serv *serv), |
439 | svc_thread_fn func, struct module *mod) | 439 | svc_thread_fn func, struct module *mod) |
440 | { | 440 | { |
441 | struct svc_serv *serv; | 441 | struct svc_serv *serv; |
442 | unsigned int npools = svc_pool_map_get(); | 442 | unsigned int npools = svc_pool_map_get(); |
443 | 443 | ||
444 | serv = __svc_create(prog, bufsize, npools, family, shutdown); | 444 | serv = __svc_create(prog, bufsize, npools, family, shutdown); |
445 | 445 | ||
446 | if (serv != NULL) { | 446 | if (serv != NULL) { |
447 | serv->sv_function = func; | 447 | serv->sv_function = func; |
448 | serv->sv_module = mod; | 448 | serv->sv_module = mod; |
449 | } | 449 | } |
450 | 450 | ||
451 | return serv; | 451 | return serv; |
452 | } | 452 | } |
453 | EXPORT_SYMBOL(svc_create_pooled); | 453 | EXPORT_SYMBOL(svc_create_pooled); |
454 | 454 | ||
455 | /* | 455 | /* |
456 | * Destroy an RPC service. Should be called with appropriate locking to | 456 | * Destroy an RPC service. Should be called with appropriate locking to |
457 | * protect the sv_nrthreads, sv_permsocks and sv_tempsocks. | 457 | * protect the sv_nrthreads, sv_permsocks and sv_tempsocks. |
458 | */ | 458 | */ |
459 | void | 459 | void |
460 | svc_destroy(struct svc_serv *serv) | 460 | svc_destroy(struct svc_serv *serv) |
461 | { | 461 | { |
462 | dprintk("svc: svc_destroy(%s, %d)\n", | 462 | dprintk("svc: svc_destroy(%s, %d)\n", |
463 | serv->sv_program->pg_name, | 463 | serv->sv_program->pg_name, |
464 | serv->sv_nrthreads); | 464 | serv->sv_nrthreads); |
465 | 465 | ||
466 | if (serv->sv_nrthreads) { | 466 | if (serv->sv_nrthreads) { |
467 | if (--(serv->sv_nrthreads) != 0) { | 467 | if (--(serv->sv_nrthreads) != 0) { |
468 | svc_sock_update_bufs(serv); | 468 | svc_sock_update_bufs(serv); |
469 | return; | 469 | return; |
470 | } | 470 | } |
471 | } else | 471 | } else |
472 | printk("svc_destroy: no threads for serv=%p!\n", serv); | 472 | printk("svc_destroy: no threads for serv=%p!\n", serv); |
473 | 473 | ||
474 | del_timer_sync(&serv->sv_temptimer); | 474 | del_timer_sync(&serv->sv_temptimer); |
475 | 475 | ||
476 | svc_close_all(&serv->sv_tempsocks); | 476 | svc_close_all(&serv->sv_tempsocks); |
477 | 477 | ||
478 | if (serv->sv_shutdown) | 478 | if (serv->sv_shutdown) |
479 | serv->sv_shutdown(serv); | 479 | serv->sv_shutdown(serv); |
480 | 480 | ||
481 | svc_close_all(&serv->sv_permsocks); | 481 | svc_close_all(&serv->sv_permsocks); |
482 | 482 | ||
483 | BUG_ON(!list_empty(&serv->sv_permsocks)); | 483 | BUG_ON(!list_empty(&serv->sv_permsocks)); |
484 | BUG_ON(!list_empty(&serv->sv_tempsocks)); | 484 | BUG_ON(!list_empty(&serv->sv_tempsocks)); |
485 | 485 | ||
486 | cache_clean_deferred(serv); | 486 | cache_clean_deferred(serv); |
487 | 487 | ||
488 | if (svc_serv_is_pooled(serv)) | 488 | if (svc_serv_is_pooled(serv)) |
489 | svc_pool_map_put(); | 489 | svc_pool_map_put(); |
490 | 490 | ||
491 | svc_unregister(serv); | 491 | svc_unregister(serv); |
492 | kfree(serv->sv_pools); | 492 | kfree(serv->sv_pools); |
493 | kfree(serv); | 493 | kfree(serv); |
494 | } | 494 | } |
495 | EXPORT_SYMBOL(svc_destroy); | 495 | EXPORT_SYMBOL(svc_destroy); |
496 | 496 | ||
497 | /* | 497 | /* |
498 | * Allocate an RPC server's buffer space. | 498 | * Allocate an RPC server's buffer space. |
499 | * We allocate pages and place them in rq_argpages. | 499 | * We allocate pages and place them in rq_argpages. |
500 | */ | 500 | */ |
501 | static int | 501 | static int |
502 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size) | 502 | svc_init_buffer(struct svc_rqst *rqstp, unsigned int size) |
503 | { | 503 | { |
504 | unsigned int pages, arghi; | 504 | unsigned int pages, arghi; |
505 | 505 | ||
506 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. | 506 | pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. |
507 | * We assume one is at most one page | 507 | * We assume one is at most one page |
508 | */ | 508 | */ |
509 | arghi = 0; | 509 | arghi = 0; |
510 | BUG_ON(pages > RPCSVC_MAXPAGES); | 510 | BUG_ON(pages > RPCSVC_MAXPAGES); |
511 | while (pages) { | 511 | while (pages) { |
512 | struct page *p = alloc_page(GFP_KERNEL); | 512 | struct page *p = alloc_page(GFP_KERNEL); |
513 | if (!p) | 513 | if (!p) |
514 | break; | 514 | break; |
515 | rqstp->rq_pages[arghi++] = p; | 515 | rqstp->rq_pages[arghi++] = p; |
516 | pages--; | 516 | pages--; |
517 | } | 517 | } |
518 | return pages == 0; | 518 | return pages == 0; |
519 | } | 519 | } |
520 | 520 | ||
521 | /* | 521 | /* |
522 | * Release an RPC server buffer | 522 | * Release an RPC server buffer |
523 | */ | 523 | */ |
524 | static void | 524 | static void |
525 | svc_release_buffer(struct svc_rqst *rqstp) | 525 | svc_release_buffer(struct svc_rqst *rqstp) |
526 | { | 526 | { |
527 | unsigned int i; | 527 | unsigned int i; |
528 | 528 | ||
529 | for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++) | 529 | for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++) |
530 | if (rqstp->rq_pages[i]) | 530 | if (rqstp->rq_pages[i]) |
531 | put_page(rqstp->rq_pages[i]); | 531 | put_page(rqstp->rq_pages[i]); |
532 | } | 532 | } |
533 | 533 | ||
534 | struct svc_rqst * | 534 | struct svc_rqst * |
535 | svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool) | 535 | svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool) |
536 | { | 536 | { |
537 | struct svc_rqst *rqstp; | 537 | struct svc_rqst *rqstp; |
538 | 538 | ||
539 | rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL); | 539 | rqstp = kzalloc(sizeof(*rqstp), GFP_KERNEL); |
540 | if (!rqstp) | 540 | if (!rqstp) |
541 | goto out_enomem; | 541 | goto out_enomem; |
542 | 542 | ||
543 | init_waitqueue_head(&rqstp->rq_wait); | 543 | init_waitqueue_head(&rqstp->rq_wait); |
544 | 544 | ||
545 | serv->sv_nrthreads++; | 545 | serv->sv_nrthreads++; |
546 | spin_lock_bh(&pool->sp_lock); | 546 | spin_lock_bh(&pool->sp_lock); |
547 | pool->sp_nrthreads++; | 547 | pool->sp_nrthreads++; |
548 | list_add(&rqstp->rq_all, &pool->sp_all_threads); | 548 | list_add(&rqstp->rq_all, &pool->sp_all_threads); |
549 | spin_unlock_bh(&pool->sp_lock); | 549 | spin_unlock_bh(&pool->sp_lock); |
550 | rqstp->rq_server = serv; | 550 | rqstp->rq_server = serv; |
551 | rqstp->rq_pool = pool; | 551 | rqstp->rq_pool = pool; |
552 | 552 | ||
553 | rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL); | 553 | rqstp->rq_argp = kmalloc(serv->sv_xdrsize, GFP_KERNEL); |
554 | if (!rqstp->rq_argp) | 554 | if (!rqstp->rq_argp) |
555 | goto out_thread; | 555 | goto out_thread; |
556 | 556 | ||
557 | rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL); | 557 | rqstp->rq_resp = kmalloc(serv->sv_xdrsize, GFP_KERNEL); |
558 | if (!rqstp->rq_resp) | 558 | if (!rqstp->rq_resp) |
559 | goto out_thread; | 559 | goto out_thread; |
560 | 560 | ||
561 | if (!svc_init_buffer(rqstp, serv->sv_max_mesg)) | 561 | if (!svc_init_buffer(rqstp, serv->sv_max_mesg)) |
562 | goto out_thread; | 562 | goto out_thread; |
563 | 563 | ||
564 | return rqstp; | 564 | return rqstp; |
565 | out_thread: | 565 | out_thread: |
566 | svc_exit_thread(rqstp); | 566 | svc_exit_thread(rqstp); |
567 | out_enomem: | 567 | out_enomem: |
568 | return ERR_PTR(-ENOMEM); | 568 | return ERR_PTR(-ENOMEM); |
569 | } | 569 | } |
570 | EXPORT_SYMBOL(svc_prepare_thread); | 570 | EXPORT_SYMBOL(svc_prepare_thread); |
571 | 571 | ||
572 | /* | 572 | /* |
573 | * Choose a pool in which to create a new thread, for svc_set_num_threads | 573 | * Choose a pool in which to create a new thread, for svc_set_num_threads |
574 | */ | 574 | */ |
575 | static inline struct svc_pool * | 575 | static inline struct svc_pool * |
576 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | 576 | choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) |
577 | { | 577 | { |
578 | if (pool != NULL) | 578 | if (pool != NULL) |
579 | return pool; | 579 | return pool; |
580 | 580 | ||
581 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; | 581 | return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; |
582 | } | 582 | } |
583 | 583 | ||
584 | /* | 584 | /* |
585 | * Choose a thread to kill, for svc_set_num_threads | 585 | * Choose a thread to kill, for svc_set_num_threads |
586 | */ | 586 | */ |
587 | static inline struct task_struct * | 587 | static inline struct task_struct * |
588 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) | 588 | choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) |
589 | { | 589 | { |
590 | unsigned int i; | 590 | unsigned int i; |
591 | struct task_struct *task = NULL; | 591 | struct task_struct *task = NULL; |
592 | 592 | ||
593 | if (pool != NULL) { | 593 | if (pool != NULL) { |
594 | spin_lock_bh(&pool->sp_lock); | 594 | spin_lock_bh(&pool->sp_lock); |
595 | } else { | 595 | } else { |
596 | /* choose a pool in round-robin fashion */ | 596 | /* choose a pool in round-robin fashion */ |
597 | for (i = 0; i < serv->sv_nrpools; i++) { | 597 | for (i = 0; i < serv->sv_nrpools; i++) { |
598 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; | 598 | pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; |
599 | spin_lock_bh(&pool->sp_lock); | 599 | spin_lock_bh(&pool->sp_lock); |
600 | if (!list_empty(&pool->sp_all_threads)) | 600 | if (!list_empty(&pool->sp_all_threads)) |
601 | goto found_pool; | 601 | goto found_pool; |
602 | spin_unlock_bh(&pool->sp_lock); | 602 | spin_unlock_bh(&pool->sp_lock); |
603 | } | 603 | } |
604 | return NULL; | 604 | return NULL; |
605 | } | 605 | } |
606 | 606 | ||
607 | found_pool: | 607 | found_pool: |
608 | if (!list_empty(&pool->sp_all_threads)) { | 608 | if (!list_empty(&pool->sp_all_threads)) { |
609 | struct svc_rqst *rqstp; | 609 | struct svc_rqst *rqstp; |
610 | 610 | ||
611 | /* | 611 | /* |
612 | * Remove from the pool->sp_all_threads list | 612 | * Remove from the pool->sp_all_threads list |
613 | * so we don't try to kill it again. | 613 | * so we don't try to kill it again. |
614 | */ | 614 | */ |
615 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); | 615 | rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); |
616 | list_del_init(&rqstp->rq_all); | 616 | list_del_init(&rqstp->rq_all); |
617 | task = rqstp->rq_task; | 617 | task = rqstp->rq_task; |
618 | } | 618 | } |
619 | spin_unlock_bh(&pool->sp_lock); | 619 | spin_unlock_bh(&pool->sp_lock); |
620 | 620 | ||
621 | return task; | 621 | return task; |
622 | } | 622 | } |
623 | 623 | ||
624 | /* | 624 | /* |
625 | * Create or destroy enough new threads to make the number | 625 | * Create or destroy enough new threads to make the number |
626 | * of threads the given number. If `pool' is non-NULL, applies | 626 | * of threads the given number. If `pool' is non-NULL, applies |
627 | * only to threads in that pool, otherwise round-robins between | 627 | * only to threads in that pool, otherwise round-robins between |
628 | * all pools. Must be called with a svc_get() reference and | 628 | * all pools. Must be called with a svc_get() reference and |
629 | * the BKL or another lock to protect access to svc_serv fields. | 629 | * the BKL or another lock to protect access to svc_serv fields. |
630 | * | 630 | * |
631 | * Destroying threads relies on the service threads filling in | 631 | * Destroying threads relies on the service threads filling in |
632 | * rqstp->rq_task, which only the nfs ones do. Assumes the serv | 632 | * rqstp->rq_task, which only the nfs ones do. Assumes the serv |
633 | * has been created using svc_create_pooled(). | 633 | * has been created using svc_create_pooled(). |
634 | * | 634 | * |
635 | * Based on code that used to be in nfsd_svc() but tweaked | 635 | * Based on code that used to be in nfsd_svc() but tweaked |
636 | * to be pool-aware. | 636 | * to be pool-aware. |
637 | */ | 637 | */ |
638 | int | 638 | int |
639 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) | 639 | svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) |
640 | { | 640 | { |
641 | struct svc_rqst *rqstp; | 641 | struct svc_rqst *rqstp; |
642 | struct task_struct *task; | 642 | struct task_struct *task; |
643 | struct svc_pool *chosen_pool; | 643 | struct svc_pool *chosen_pool; |
644 | int error = 0; | 644 | int error = 0; |
645 | unsigned int state = serv->sv_nrthreads-1; | 645 | unsigned int state = serv->sv_nrthreads-1; |
646 | 646 | ||
647 | if (pool == NULL) { | 647 | if (pool == NULL) { |
648 | /* The -1 assumes caller has done a svc_get() */ | 648 | /* The -1 assumes caller has done a svc_get() */ |
649 | nrservs -= (serv->sv_nrthreads-1); | 649 | nrservs -= (serv->sv_nrthreads-1); |
650 | } else { | 650 | } else { |
651 | spin_lock_bh(&pool->sp_lock); | 651 | spin_lock_bh(&pool->sp_lock); |
652 | nrservs -= pool->sp_nrthreads; | 652 | nrservs -= pool->sp_nrthreads; |
653 | spin_unlock_bh(&pool->sp_lock); | 653 | spin_unlock_bh(&pool->sp_lock); |
654 | } | 654 | } |
655 | 655 | ||
656 | /* create new threads */ | 656 | /* create new threads */ |
657 | while (nrservs > 0) { | 657 | while (nrservs > 0) { |
658 | nrservs--; | 658 | nrservs--; |
659 | chosen_pool = choose_pool(serv, pool, &state); | 659 | chosen_pool = choose_pool(serv, pool, &state); |
660 | 660 | ||
661 | rqstp = svc_prepare_thread(serv, chosen_pool); | 661 | rqstp = svc_prepare_thread(serv, chosen_pool); |
662 | if (IS_ERR(rqstp)) { | 662 | if (IS_ERR(rqstp)) { |
663 | error = PTR_ERR(rqstp); | 663 | error = PTR_ERR(rqstp); |
664 | break; | 664 | break; |
665 | } | 665 | } |
666 | 666 | ||
667 | __module_get(serv->sv_module); | 667 | __module_get(serv->sv_module); |
668 | task = kthread_create(serv->sv_function, rqstp, serv->sv_name); | 668 | task = kthread_create(serv->sv_function, rqstp, serv->sv_name); |
669 | if (IS_ERR(task)) { | 669 | if (IS_ERR(task)) { |
670 | error = PTR_ERR(task); | 670 | error = PTR_ERR(task); |
671 | module_put(serv->sv_module); | 671 | module_put(serv->sv_module); |
672 | svc_exit_thread(rqstp); | 672 | svc_exit_thread(rqstp); |
673 | break; | 673 | break; |
674 | } | 674 | } |
675 | 675 | ||
676 | rqstp->rq_task = task; | 676 | rqstp->rq_task = task; |
677 | if (serv->sv_nrpools > 1) | 677 | if (serv->sv_nrpools > 1) |
678 | svc_pool_map_set_cpumask(task, chosen_pool->sp_id); | 678 | svc_pool_map_set_cpumask(task, chosen_pool->sp_id); |
679 | 679 | ||
680 | svc_sock_update_bufs(serv); | 680 | svc_sock_update_bufs(serv); |
681 | wake_up_process(task); | 681 | wake_up_process(task); |
682 | } | 682 | } |
683 | /* destroy old threads */ | 683 | /* destroy old threads */ |
684 | while (nrservs < 0 && | 684 | while (nrservs < 0 && |
685 | (task = choose_victim(serv, pool, &state)) != NULL) { | 685 | (task = choose_victim(serv, pool, &state)) != NULL) { |
686 | send_sig(SIGINT, task, 1); | 686 | send_sig(SIGINT, task, 1); |
687 | nrservs++; | 687 | nrservs++; |
688 | } | 688 | } |
689 | 689 | ||
690 | return error; | 690 | return error; |
691 | } | 691 | } |
692 | EXPORT_SYMBOL(svc_set_num_threads); | 692 | EXPORT_SYMBOL(svc_set_num_threads); |
693 | 693 | ||
694 | /* | 694 | /* |
695 | * Called from a server thread as it's exiting. Caller must hold the BKL or | 695 | * Called from a server thread as it's exiting. Caller must hold the BKL or |
696 | * the "service mutex", whichever is appropriate for the service. | 696 | * the "service mutex", whichever is appropriate for the service. |
697 | */ | 697 | */ |
698 | void | 698 | void |
699 | svc_exit_thread(struct svc_rqst *rqstp) | 699 | svc_exit_thread(struct svc_rqst *rqstp) |
700 | { | 700 | { |
701 | struct svc_serv *serv = rqstp->rq_server; | 701 | struct svc_serv *serv = rqstp->rq_server; |
702 | struct svc_pool *pool = rqstp->rq_pool; | 702 | struct svc_pool *pool = rqstp->rq_pool; |
703 | 703 | ||
704 | svc_release_buffer(rqstp); | 704 | svc_release_buffer(rqstp); |
705 | kfree(rqstp->rq_resp); | 705 | kfree(rqstp->rq_resp); |
706 | kfree(rqstp->rq_argp); | 706 | kfree(rqstp->rq_argp); |
707 | kfree(rqstp->rq_auth_data); | 707 | kfree(rqstp->rq_auth_data); |
708 | 708 | ||
709 | spin_lock_bh(&pool->sp_lock); | 709 | spin_lock_bh(&pool->sp_lock); |
710 | pool->sp_nrthreads--; | 710 | pool->sp_nrthreads--; |
711 | list_del(&rqstp->rq_all); | 711 | list_del(&rqstp->rq_all); |
712 | spin_unlock_bh(&pool->sp_lock); | 712 | spin_unlock_bh(&pool->sp_lock); |
713 | 713 | ||
714 | kfree(rqstp); | 714 | kfree(rqstp); |
715 | 715 | ||
716 | /* Release the server */ | 716 | /* Release the server */ |
717 | if (serv) | 717 | if (serv) |
718 | svc_destroy(serv); | 718 | svc_destroy(serv); |
719 | } | 719 | } |
720 | EXPORT_SYMBOL(svc_exit_thread); | 720 | EXPORT_SYMBOL(svc_exit_thread); |
721 | 721 | ||
722 | #ifdef CONFIG_SUNRPC_REGISTER_V4 | 722 | #ifdef CONFIG_SUNRPC_REGISTER_V4 |
723 | 723 | ||
724 | /* | 724 | /* |
725 | * Register an "inet" protocol family netid with the local | 725 | * Register an "inet" protocol family netid with the local |
726 | * rpcbind daemon via an rpcbind v4 SET request. | 726 | * rpcbind daemon via an rpcbind v4 SET request. |
727 | * | 727 | * |
728 | * No netconfig infrastructure is available in the kernel, so | 728 | * No netconfig infrastructure is available in the kernel, so |
729 | * we map IP_ protocol numbers to netids by hand. | 729 | * we map IP_ protocol numbers to netids by hand. |
730 | * | 730 | * |
731 | * Returns zero on success; a negative errno value is returned | 731 | * Returns zero on success; a negative errno value is returned |
732 | * if any error occurs. | 732 | * if any error occurs. |
733 | */ | 733 | */ |
734 | static int __svc_rpcb_register4(const u32 program, const u32 version, | 734 | static int __svc_rpcb_register4(const u32 program, const u32 version, |
735 | const unsigned short protocol, | 735 | const unsigned short protocol, |
736 | const unsigned short port) | 736 | const unsigned short port) |
737 | { | 737 | { |
738 | struct sockaddr_in sin = { | 738 | struct sockaddr_in sin = { |
739 | .sin_family = AF_INET, | 739 | .sin_family = AF_INET, |
740 | .sin_addr.s_addr = htonl(INADDR_ANY), | 740 | .sin_addr.s_addr = htonl(INADDR_ANY), |
741 | .sin_port = htons(port), | 741 | .sin_port = htons(port), |
742 | }; | 742 | }; |
743 | char *netid; | 743 | char *netid; |
744 | 744 | ||
745 | switch (protocol) { | 745 | switch (protocol) { |
746 | case IPPROTO_UDP: | 746 | case IPPROTO_UDP: |
747 | netid = RPCBIND_NETID_UDP; | 747 | netid = RPCBIND_NETID_UDP; |
748 | break; | 748 | break; |
749 | case IPPROTO_TCP: | 749 | case IPPROTO_TCP: |
750 | netid = RPCBIND_NETID_TCP; | 750 | netid = RPCBIND_NETID_TCP; |
751 | break; | 751 | break; |
752 | default: | 752 | default: |
753 | return -EPROTONOSUPPORT; | 753 | return -EPROTONOSUPPORT; |
754 | } | 754 | } |
755 | 755 | ||
756 | return rpcb_v4_register(program, version, | 756 | return rpcb_v4_register(program, version, |
757 | (struct sockaddr *)&sin, netid); | 757 | (struct sockaddr *)&sin, netid); |
758 | } | 758 | } |
759 | 759 | ||
760 | /* | 760 | /* |
761 | * Register an "inet6" protocol family netid with the local | 761 | * Register an "inet6" protocol family netid with the local |
762 | * rpcbind daemon via an rpcbind v4 SET request. | 762 | * rpcbind daemon via an rpcbind v4 SET request. |
763 | * | 763 | * |
764 | * No netconfig infrastructure is available in the kernel, so | 764 | * No netconfig infrastructure is available in the kernel, so |
765 | * we map IP_ protocol numbers to netids by hand. | 765 | * we map IP_ protocol numbers to netids by hand. |
766 | * | 766 | * |
767 | * Returns zero on success; a negative errno value is returned | 767 | * Returns zero on success; a negative errno value is returned |
768 | * if any error occurs. | 768 | * if any error occurs. |
769 | */ | 769 | */ |
770 | static int __svc_rpcb_register6(const u32 program, const u32 version, | 770 | static int __svc_rpcb_register6(const u32 program, const u32 version, |
771 | const unsigned short protocol, | 771 | const unsigned short protocol, |
772 | const unsigned short port) | 772 | const unsigned short port) |
773 | { | 773 | { |
774 | struct sockaddr_in6 sin6 = { | 774 | struct sockaddr_in6 sin6 = { |
775 | .sin6_family = AF_INET6, | 775 | .sin6_family = AF_INET6, |
776 | .sin6_addr = IN6ADDR_ANY_INIT, | 776 | .sin6_addr = IN6ADDR_ANY_INIT, |
777 | .sin6_port = htons(port), | 777 | .sin6_port = htons(port), |
778 | }; | 778 | }; |
779 | char *netid; | 779 | char *netid; |
780 | 780 | ||
781 | switch (protocol) { | 781 | switch (protocol) { |
782 | case IPPROTO_UDP: | 782 | case IPPROTO_UDP: |
783 | netid = RPCBIND_NETID_UDP6; | 783 | netid = RPCBIND_NETID_UDP6; |
784 | break; | 784 | break; |
785 | case IPPROTO_TCP: | 785 | case IPPROTO_TCP: |
786 | netid = RPCBIND_NETID_TCP6; | 786 | netid = RPCBIND_NETID_TCP6; |
787 | break; | 787 | break; |
788 | default: | 788 | default: |
789 | return -EPROTONOSUPPORT; | 789 | return -EPROTONOSUPPORT; |
790 | } | 790 | } |
791 | 791 | ||
792 | return rpcb_v4_register(program, version, | 792 | return rpcb_v4_register(program, version, |
793 | (struct sockaddr *)&sin6, netid); | 793 | (struct sockaddr *)&sin6, netid); |
794 | } | 794 | } |
795 | 795 | ||
796 | /* | 796 | /* |
797 | * Register a kernel RPC service via rpcbind version 4. | 797 | * Register a kernel RPC service via rpcbind version 4. |
798 | * | 798 | * |
799 | * Returns zero on success; a negative errno value is returned | 799 | * Returns zero on success; a negative errno value is returned |
800 | * if any error occurs. | 800 | * if any error occurs. |
801 | */ | 801 | */ |
802 | static int __svc_register(const u32 program, const u32 version, | 802 | static int __svc_register(const u32 program, const u32 version, |
803 | const sa_family_t family, | 803 | const sa_family_t family, |
804 | const unsigned short protocol, | 804 | const unsigned short protocol, |
805 | const unsigned short port) | 805 | const unsigned short port) |
806 | { | 806 | { |
807 | int error; | 807 | int error; |
808 | 808 | ||
809 | switch (family) { | 809 | switch (family) { |
810 | case AF_INET: | 810 | case AF_INET: |
811 | return __svc_rpcb_register4(program, version, | 811 | return __svc_rpcb_register4(program, version, |
812 | protocol, port); | 812 | protocol, port); |
813 | case AF_INET6: | 813 | case AF_INET6: |
814 | error = __svc_rpcb_register6(program, version, | 814 | error = __svc_rpcb_register6(program, version, |
815 | protocol, port); | 815 | protocol, port); |
816 | if (error < 0) | 816 | if (error < 0) |
817 | return error; | 817 | return error; |
818 | 818 | ||
819 | /* | 819 | /* |
820 | * Work around bug in some versions of Linux rpcbind | 820 | * Work around bug in some versions of Linux rpcbind |
821 | * which don't allow registration of both inet and | 821 | * which don't allow registration of both inet and |
822 | * inet6 netids. | 822 | * inet6 netids. |
823 | * | 823 | * |
824 | * Error return ignored for now. | 824 | * Error return ignored for now. |
825 | */ | 825 | */ |
826 | __svc_rpcb_register4(program, version, | 826 | __svc_rpcb_register4(program, version, |
827 | protocol, port); | 827 | protocol, port); |
828 | return 0; | 828 | return 0; |
829 | } | 829 | } |
830 | 830 | ||
831 | return -EAFNOSUPPORT; | 831 | return -EAFNOSUPPORT; |
832 | } | 832 | } |
833 | 833 | ||
834 | #else /* CONFIG_SUNRPC_REGISTER_V4 */ | 834 | #else /* CONFIG_SUNRPC_REGISTER_V4 */ |
835 | 835 | ||
836 | /* | 836 | /* |
837 | * Register a kernel RPC service via rpcbind version 2. | 837 | * Register a kernel RPC service via rpcbind version 2. |
838 | * | 838 | * |
839 | * Returns zero on success; a negative errno value is returned | 839 | * Returns zero on success; a negative errno value is returned |
840 | * if any error occurs. | 840 | * if any error occurs. |
841 | */ | 841 | */ |
842 | static int __svc_register(const u32 program, const u32 version, | 842 | static int __svc_register(const u32 program, const u32 version, |
843 | sa_family_t family, | 843 | sa_family_t family, |
844 | const unsigned short protocol, | 844 | const unsigned short protocol, |
845 | const unsigned short port) | 845 | const unsigned short port) |
846 | { | 846 | { |
847 | if (family != AF_INET) | 847 | if (family != AF_INET) |
848 | return -EAFNOSUPPORT; | 848 | return -EAFNOSUPPORT; |
849 | 849 | ||
850 | return rpcb_register(program, version, protocol, port); | 850 | return rpcb_register(program, version, protocol, port); |
851 | } | 851 | } |
852 | 852 | ||
853 | #endif /* CONFIG_SUNRPC_REGISTER_V4 */ | 853 | #endif /* CONFIG_SUNRPC_REGISTER_V4 */ |
854 | 854 | ||
855 | /** | 855 | /** |
856 | * svc_register - register an RPC service with the local portmapper | 856 | * svc_register - register an RPC service with the local portmapper |
857 | * @serv: svc_serv struct for the service to register | 857 | * @serv: svc_serv struct for the service to register |
858 | * @proto: transport protocol number to advertise | 858 | * @proto: transport protocol number to advertise |
859 | * @port: port to advertise | 859 | * @port: port to advertise |
860 | * | 860 | * |
861 | * Service is registered for any address in serv's address family | 861 | * Service is registered for any address in serv's address family |
862 | */ | 862 | */ |
863 | int svc_register(const struct svc_serv *serv, const unsigned short proto, | 863 | int svc_register(const struct svc_serv *serv, const unsigned short proto, |
864 | const unsigned short port) | 864 | const unsigned short port) |
865 | { | 865 | { |
866 | struct svc_program *progp; | 866 | struct svc_program *progp; |
867 | unsigned int i; | 867 | unsigned int i; |
868 | int error = 0; | 868 | int error = 0; |
869 | 869 | ||
870 | BUG_ON(proto == 0 && port == 0); | 870 | BUG_ON(proto == 0 && port == 0); |
871 | 871 | ||
872 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | 872 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
873 | for (i = 0; i < progp->pg_nvers; i++) { | 873 | for (i = 0; i < progp->pg_nvers; i++) { |
874 | if (progp->pg_vers[i] == NULL) | 874 | if (progp->pg_vers[i] == NULL) |
875 | continue; | 875 | continue; |
876 | 876 | ||
877 | dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n", | 877 | dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n", |
878 | progp->pg_name, | 878 | progp->pg_name, |
879 | i, | 879 | i, |
880 | proto == IPPROTO_UDP? "udp" : "tcp", | 880 | proto == IPPROTO_UDP? "udp" : "tcp", |
881 | port, | 881 | port, |
882 | serv->sv_family, | 882 | serv->sv_family, |
883 | progp->pg_vers[i]->vs_hidden? | 883 | progp->pg_vers[i]->vs_hidden? |
884 | " (but not telling portmap)" : ""); | 884 | " (but not telling portmap)" : ""); |
885 | 885 | ||
886 | if (progp->pg_vers[i]->vs_hidden) | 886 | if (progp->pg_vers[i]->vs_hidden) |
887 | continue; | 887 | continue; |
888 | 888 | ||
889 | error = __svc_register(progp->pg_prog, i, | 889 | error = __svc_register(progp->pg_prog, i, |
890 | serv->sv_family, proto, port); | 890 | serv->sv_family, proto, port); |
891 | if (error < 0) | 891 | if (error < 0) |
892 | break; | 892 | break; |
893 | } | 893 | } |
894 | } | 894 | } |
895 | 895 | ||
896 | return error; | 896 | return error; |
897 | } | 897 | } |
898 | 898 | ||
899 | #ifdef CONFIG_SUNRPC_REGISTER_V4 | ||
900 | |||
901 | static void __svc_unregister(const u32 program, const u32 version, | ||
902 | const char *progname) | ||
903 | { | ||
904 | struct sockaddr_in6 sin6 = { | ||
905 | .sin6_family = AF_INET6, | ||
906 | .sin6_addr = IN6ADDR_ANY_INIT, | ||
907 | .sin6_port = 0, | ||
908 | }; | ||
909 | int error; | ||
910 | |||
911 | error = rpcb_v4_register(program, version, | ||
912 | (struct sockaddr *)&sin6, ""); | ||
913 | dprintk("svc: %s(%sv%u), error %d\n", | ||
914 | __func__, progname, version, error); | ||
915 | } | ||
916 | |||
917 | #else /* CONFIG_SUNRPC_REGISTER_V4 */ | ||
918 | |||
919 | static void __svc_unregister(const u32 program, const u32 version, | ||
920 | const char *progname) | ||
921 | { | ||
922 | int error; | ||
923 | |||
924 | error = rpcb_register(program, version, 0, 0); | ||
925 | dprintk("svc: %s(%sv%u), error %d\n", | ||
926 | __func__, progname, version, error); | ||
927 | } | ||
928 | |||
929 | #endif /* CONFIG_SUNRPC_REGISTER_V4 */ | ||
930 | |||
899 | /* | 931 | /* |
900 | * All transport protocols and ports for this service are removed | 932 | * All netids, bind addresses and ports registered for [program, version] |
901 | * from the local rpcbind database if the service is not hidden. | 933 | * are removed from the local rpcbind database (if the service is not |
934 | * hidden) to make way for a new instance of the service. | ||
902 | * | 935 | * |
903 | * The result of unregistration is reported via dprintk for those | 936 | * The result of unregistration is reported via dprintk for those who want |
904 | * who want verification of the result, but is otherwise not | 937 | * verification of the result, but is otherwise not important. |
905 | * important. | ||
906 | * | ||
907 | * The local rpcbind daemon listens on either only IPv6 or only | ||
908 | * IPv4. The kernel can't tell how it's configured. However, | ||
909 | * AF_INET addresses are mapped to AF_INET6 in IPv6-only config- | ||
910 | * urations, so even an unregistration request on AF_INET will | ||
911 | * get to a local rpcbind daemon listening only on AF_INET6. So | ||
912 | * we always unregister via AF_INET. | ||
913 | * | ||
914 | * At this point we don't need rpcbind version 4 for unregis- | ||
915 | * tration: A v2 UNSET request will clear all transports (netids), | ||
916 | * addresses, and address families for [program, version]. | ||
917 | */ | 938 | */ |
918 | static void svc_unregister(const struct svc_serv *serv) | 939 | static void svc_unregister(const struct svc_serv *serv) |
919 | { | 940 | { |
920 | struct svc_program *progp; | 941 | struct svc_program *progp; |
921 | unsigned long flags; | 942 | unsigned long flags; |
922 | unsigned int i; | 943 | unsigned int i; |
923 | int error; | ||
924 | 944 | ||
925 | clear_thread_flag(TIF_SIGPENDING); | 945 | clear_thread_flag(TIF_SIGPENDING); |
926 | 946 | ||
927 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { | 947 | for (progp = serv->sv_program; progp; progp = progp->pg_next) { |
928 | for (i = 0; i < progp->pg_nvers; i++) { | 948 | for (i = 0; i < progp->pg_nvers; i++) { |
929 | if (progp->pg_vers[i] == NULL) | 949 | if (progp->pg_vers[i] == NULL) |
930 | continue; | 950 | continue; |
931 | if (progp->pg_vers[i]->vs_hidden) | 951 | if (progp->pg_vers[i]->vs_hidden) |
932 | continue; | 952 | continue; |
933 | 953 | ||
934 | error = rpcb_register(progp->pg_prog, i, 0, 0); | 954 | __svc_unregister(progp->pg_prog, i, progp->pg_name); |
935 | dprintk("svc: svc_unregister(%sv%u), error %d\n", | ||
936 | progp->pg_name, i, error); | ||
937 | } | 955 | } |
938 | } | 956 | } |
939 | 957 | ||
940 | spin_lock_irqsave(¤t->sighand->siglock, flags); | 958 | spin_lock_irqsave(¤t->sighand->siglock, flags); |
941 | recalc_sigpending(); | 959 | recalc_sigpending(); |
942 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); | 960 | spin_unlock_irqrestore(¤t->sighand->siglock, flags); |
943 | } | 961 | } |
944 | 962 | ||
945 | /* | 963 | /* |
946 | * Printk the given error with the address of the client that caused it. | 964 | * Printk the given error with the address of the client that caused it. |
947 | */ | 965 | */ |
948 | static int | 966 | static int |
949 | __attribute__ ((format (printf, 2, 3))) | 967 | __attribute__ ((format (printf, 2, 3))) |
950 | svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) | 968 | svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) |
951 | { | 969 | { |
952 | va_list args; | 970 | va_list args; |
953 | int r; | 971 | int r; |
954 | char buf[RPC_MAX_ADDRBUFLEN]; | 972 | char buf[RPC_MAX_ADDRBUFLEN]; |
955 | 973 | ||
956 | if (!net_ratelimit()) | 974 | if (!net_ratelimit()) |
957 | return 0; | 975 | return 0; |
958 | 976 | ||
959 | printk(KERN_WARNING "svc: %s: ", | 977 | printk(KERN_WARNING "svc: %s: ", |
960 | svc_print_addr(rqstp, buf, sizeof(buf))); | 978 | svc_print_addr(rqstp, buf, sizeof(buf))); |
961 | 979 | ||
962 | va_start(args, fmt); | 980 | va_start(args, fmt); |
963 | r = vprintk(fmt, args); | 981 | r = vprintk(fmt, args); |
964 | va_end(args); | 982 | va_end(args); |
965 | 983 | ||
966 | return r; | 984 | return r; |
967 | } | 985 | } |
968 | 986 | ||
969 | /* | 987 | /* |
970 | * Process the RPC request. | 988 | * Process the RPC request. |
971 | */ | 989 | */ |
972 | int | 990 | int |
973 | svc_process(struct svc_rqst *rqstp) | 991 | svc_process(struct svc_rqst *rqstp) |
974 | { | 992 | { |
975 | struct svc_program *progp; | 993 | struct svc_program *progp; |
976 | struct svc_version *versp = NULL; /* compiler food */ | 994 | struct svc_version *versp = NULL; /* compiler food */ |
977 | struct svc_procedure *procp = NULL; | 995 | struct svc_procedure *procp = NULL; |
978 | struct kvec * argv = &rqstp->rq_arg.head[0]; | 996 | struct kvec * argv = &rqstp->rq_arg.head[0]; |
979 | struct kvec * resv = &rqstp->rq_res.head[0]; | 997 | struct kvec * resv = &rqstp->rq_res.head[0]; |
980 | struct svc_serv *serv = rqstp->rq_server; | 998 | struct svc_serv *serv = rqstp->rq_server; |
981 | kxdrproc_t xdr; | 999 | kxdrproc_t xdr; |
982 | __be32 *statp; | 1000 | __be32 *statp; |
983 | u32 dir, prog, vers, proc; | 1001 | u32 dir, prog, vers, proc; |
984 | __be32 auth_stat, rpc_stat; | 1002 | __be32 auth_stat, rpc_stat; |
985 | int auth_res; | 1003 | int auth_res; |
986 | __be32 *reply_statp; | 1004 | __be32 *reply_statp; |
987 | 1005 | ||
988 | rpc_stat = rpc_success; | 1006 | rpc_stat = rpc_success; |
989 | 1007 | ||
990 | if (argv->iov_len < 6*4) | 1008 | if (argv->iov_len < 6*4) |
991 | goto err_short_len; | 1009 | goto err_short_len; |
992 | 1010 | ||
993 | /* setup response xdr_buf. | 1011 | /* setup response xdr_buf. |
994 | * Initially it has just one page | 1012 | * Initially it has just one page |
995 | */ | 1013 | */ |
996 | rqstp->rq_resused = 1; | 1014 | rqstp->rq_resused = 1; |
997 | resv->iov_base = page_address(rqstp->rq_respages[0]); | 1015 | resv->iov_base = page_address(rqstp->rq_respages[0]); |
998 | resv->iov_len = 0; | 1016 | resv->iov_len = 0; |
999 | rqstp->rq_res.pages = rqstp->rq_respages + 1; | 1017 | rqstp->rq_res.pages = rqstp->rq_respages + 1; |
1000 | rqstp->rq_res.len = 0; | 1018 | rqstp->rq_res.len = 0; |
1001 | rqstp->rq_res.page_base = 0; | 1019 | rqstp->rq_res.page_base = 0; |
1002 | rqstp->rq_res.page_len = 0; | 1020 | rqstp->rq_res.page_len = 0; |
1003 | rqstp->rq_res.buflen = PAGE_SIZE; | 1021 | rqstp->rq_res.buflen = PAGE_SIZE; |
1004 | rqstp->rq_res.tail[0].iov_base = NULL; | 1022 | rqstp->rq_res.tail[0].iov_base = NULL; |
1005 | rqstp->rq_res.tail[0].iov_len = 0; | 1023 | rqstp->rq_res.tail[0].iov_len = 0; |
1006 | /* Will be turned off only in gss privacy case: */ | 1024 | /* Will be turned off only in gss privacy case: */ |
1007 | rqstp->rq_splice_ok = 1; | 1025 | rqstp->rq_splice_ok = 1; |
1008 | 1026 | ||
1009 | /* Setup reply header */ | 1027 | /* Setup reply header */ |
1010 | rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); | 1028 | rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); |
1011 | 1029 | ||
1012 | rqstp->rq_xid = svc_getu32(argv); | 1030 | rqstp->rq_xid = svc_getu32(argv); |
1013 | svc_putu32(resv, rqstp->rq_xid); | 1031 | svc_putu32(resv, rqstp->rq_xid); |
1014 | 1032 | ||
1015 | dir = svc_getnl(argv); | 1033 | dir = svc_getnl(argv); |
1016 | vers = svc_getnl(argv); | 1034 | vers = svc_getnl(argv); |
1017 | 1035 | ||
1018 | /* First words of reply: */ | 1036 | /* First words of reply: */ |
1019 | svc_putnl(resv, 1); /* REPLY */ | 1037 | svc_putnl(resv, 1); /* REPLY */ |
1020 | 1038 | ||
1021 | if (dir != 0) /* direction != CALL */ | 1039 | if (dir != 0) /* direction != CALL */ |
1022 | goto err_bad_dir; | 1040 | goto err_bad_dir; |
1023 | if (vers != 2) /* RPC version number */ | 1041 | if (vers != 2) /* RPC version number */ |
1024 | goto err_bad_rpc; | 1042 | goto err_bad_rpc; |
1025 | 1043 | ||
1026 | /* Save position in case we later decide to reject: */ | 1044 | /* Save position in case we later decide to reject: */ |
1027 | reply_statp = resv->iov_base + resv->iov_len; | 1045 | reply_statp = resv->iov_base + resv->iov_len; |
1028 | 1046 | ||
1029 | svc_putnl(resv, 0); /* ACCEPT */ | 1047 | svc_putnl(resv, 0); /* ACCEPT */ |
1030 | 1048 | ||
1031 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ | 1049 | rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ |
1032 | rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ | 1050 | rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ |
1033 | rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ | 1051 | rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ |
1034 | 1052 | ||
1035 | progp = serv->sv_program; | 1053 | progp = serv->sv_program; |
1036 | 1054 | ||
1037 | for (progp = serv->sv_program; progp; progp = progp->pg_next) | 1055 | for (progp = serv->sv_program; progp; progp = progp->pg_next) |
1038 | if (prog == progp->pg_prog) | 1056 | if (prog == progp->pg_prog) |
1039 | break; | 1057 | break; |
1040 | 1058 | ||
1041 | /* | 1059 | /* |
1042 | * Decode auth data, and add verifier to reply buffer. | 1060 | * Decode auth data, and add verifier to reply buffer. |
1043 | * We do this before anything else in order to get a decent | 1061 | * We do this before anything else in order to get a decent |
1044 | * auth verifier. | 1062 | * auth verifier. |
1045 | */ | 1063 | */ |
1046 | auth_res = svc_authenticate(rqstp, &auth_stat); | 1064 | auth_res = svc_authenticate(rqstp, &auth_stat); |
1047 | /* Also give the program a chance to reject this call: */ | 1065 | /* Also give the program a chance to reject this call: */ |
1048 | if (auth_res == SVC_OK && progp) { | 1066 | if (auth_res == SVC_OK && progp) { |
1049 | auth_stat = rpc_autherr_badcred; | 1067 | auth_stat = rpc_autherr_badcred; |
1050 | auth_res = progp->pg_authenticate(rqstp); | 1068 | auth_res = progp->pg_authenticate(rqstp); |
1051 | } | 1069 | } |
1052 | switch (auth_res) { | 1070 | switch (auth_res) { |
1053 | case SVC_OK: | 1071 | case SVC_OK: |
1054 | break; | 1072 | break; |
1055 | case SVC_GARBAGE: | 1073 | case SVC_GARBAGE: |
1056 | goto err_garbage; | 1074 | goto err_garbage; |
1057 | case SVC_SYSERR: | 1075 | case SVC_SYSERR: |
1058 | rpc_stat = rpc_system_err; | 1076 | rpc_stat = rpc_system_err; |
1059 | goto err_bad; | 1077 | goto err_bad; |
1060 | case SVC_DENIED: | 1078 | case SVC_DENIED: |
1061 | goto err_bad_auth; | 1079 | goto err_bad_auth; |
1062 | case SVC_DROP: | 1080 | case SVC_DROP: |
1063 | goto dropit; | 1081 | goto dropit; |
1064 | case SVC_COMPLETE: | 1082 | case SVC_COMPLETE: |
1065 | goto sendit; | 1083 | goto sendit; |
1066 | } | 1084 | } |
1067 | 1085 | ||
1068 | if (progp == NULL) | 1086 | if (progp == NULL) |
1069 | goto err_bad_prog; | 1087 | goto err_bad_prog; |
1070 | 1088 | ||
1071 | if (vers >= progp->pg_nvers || | 1089 | if (vers >= progp->pg_nvers || |
1072 | !(versp = progp->pg_vers[vers])) | 1090 | !(versp = progp->pg_vers[vers])) |
1073 | goto err_bad_vers; | 1091 | goto err_bad_vers; |
1074 | 1092 | ||
1075 | procp = versp->vs_proc + proc; | 1093 | procp = versp->vs_proc + proc; |
1076 | if (proc >= versp->vs_nproc || !procp->pc_func) | 1094 | if (proc >= versp->vs_nproc || !procp->pc_func) |
1077 | goto err_bad_proc; | 1095 | goto err_bad_proc; |
1078 | rqstp->rq_server = serv; | 1096 | rqstp->rq_server = serv; |
1079 | rqstp->rq_procinfo = procp; | 1097 | rqstp->rq_procinfo = procp; |
1080 | 1098 | ||
1081 | /* Syntactic check complete */ | 1099 | /* Syntactic check complete */ |
1082 | serv->sv_stats->rpccnt++; | 1100 | serv->sv_stats->rpccnt++; |
1083 | 1101 | ||
1084 | /* Build the reply header. */ | 1102 | /* Build the reply header. */ |
1085 | statp = resv->iov_base +resv->iov_len; | 1103 | statp = resv->iov_base +resv->iov_len; |
1086 | svc_putnl(resv, RPC_SUCCESS); | 1104 | svc_putnl(resv, RPC_SUCCESS); |
1087 | 1105 | ||
1088 | /* Bump per-procedure stats counter */ | 1106 | /* Bump per-procedure stats counter */ |
1089 | procp->pc_count++; | 1107 | procp->pc_count++; |
1090 | 1108 | ||
1091 | /* Initialize storage for argp and resp */ | 1109 | /* Initialize storage for argp and resp */ |
1092 | memset(rqstp->rq_argp, 0, procp->pc_argsize); | 1110 | memset(rqstp->rq_argp, 0, procp->pc_argsize); |
1093 | memset(rqstp->rq_resp, 0, procp->pc_ressize); | 1111 | memset(rqstp->rq_resp, 0, procp->pc_ressize); |
1094 | 1112 | ||
1095 | /* un-reserve some of the out-queue now that we have a | 1113 | /* un-reserve some of the out-queue now that we have a |
1096 | * better idea of reply size | 1114 | * better idea of reply size |
1097 | */ | 1115 | */ |
1098 | if (procp->pc_xdrressize) | 1116 | if (procp->pc_xdrressize) |
1099 | svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); | 1117 | svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); |
1100 | 1118 | ||
1101 | /* Call the function that processes the request. */ | 1119 | /* Call the function that processes the request. */ |
1102 | if (!versp->vs_dispatch) { | 1120 | if (!versp->vs_dispatch) { |
1103 | /* Decode arguments */ | 1121 | /* Decode arguments */ |
1104 | xdr = procp->pc_decode; | 1122 | xdr = procp->pc_decode; |
1105 | if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) | 1123 | if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) |
1106 | goto err_garbage; | 1124 | goto err_garbage; |
1107 | 1125 | ||
1108 | *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); | 1126 | *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); |
1109 | 1127 | ||
1110 | /* Encode reply */ | 1128 | /* Encode reply */ |
1111 | if (*statp == rpc_drop_reply) { | 1129 | if (*statp == rpc_drop_reply) { |
1112 | if (procp->pc_release) | 1130 | if (procp->pc_release) |
1113 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | 1131 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); |
1114 | goto dropit; | 1132 | goto dropit; |
1115 | } | 1133 | } |
1116 | if (*statp == rpc_success && (xdr = procp->pc_encode) | 1134 | if (*statp == rpc_success && (xdr = procp->pc_encode) |
1117 | && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { | 1135 | && !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { |
1118 | dprintk("svc: failed to encode reply\n"); | 1136 | dprintk("svc: failed to encode reply\n"); |
1119 | /* serv->sv_stats->rpcsystemerr++; */ | 1137 | /* serv->sv_stats->rpcsystemerr++; */ |
1120 | *statp = rpc_system_err; | 1138 | *statp = rpc_system_err; |
1121 | } | 1139 | } |
1122 | } else { | 1140 | } else { |
1123 | dprintk("svc: calling dispatcher\n"); | 1141 | dprintk("svc: calling dispatcher\n"); |
1124 | if (!versp->vs_dispatch(rqstp, statp)) { | 1142 | if (!versp->vs_dispatch(rqstp, statp)) { |
1125 | /* Release reply info */ | 1143 | /* Release reply info */ |
1126 | if (procp->pc_release) | 1144 | if (procp->pc_release) |
1127 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | 1145 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); |
1128 | goto dropit; | 1146 | goto dropit; |
1129 | } | 1147 | } |
1130 | } | 1148 | } |
1131 | 1149 | ||
1132 | /* Check RPC status result */ | 1150 | /* Check RPC status result */ |
1133 | if (*statp != rpc_success) | 1151 | if (*statp != rpc_success) |
1134 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; | 1152 | resv->iov_len = ((void*)statp) - resv->iov_base + 4; |
1135 | 1153 | ||
1136 | /* Release reply info */ | 1154 | /* Release reply info */ |
1137 | if (procp->pc_release) | 1155 | if (procp->pc_release) |
1138 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); | 1156 | procp->pc_release(rqstp, NULL, rqstp->rq_resp); |
1139 | 1157 | ||
1140 | if (procp->pc_encode == NULL) | 1158 | if (procp->pc_encode == NULL) |
1141 | goto dropit; | 1159 | goto dropit; |
1142 | 1160 | ||
1143 | sendit: | 1161 | sendit: |
1144 | if (svc_authorise(rqstp)) | 1162 | if (svc_authorise(rqstp)) |
1145 | goto dropit; | 1163 | goto dropit; |
1146 | return svc_send(rqstp); | 1164 | return svc_send(rqstp); |
1147 | 1165 | ||
1148 | dropit: | 1166 | dropit: |
1149 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ | 1167 | svc_authorise(rqstp); /* doesn't hurt to call this twice */ |
1150 | dprintk("svc: svc_process dropit\n"); | 1168 | dprintk("svc: svc_process dropit\n"); |
1151 | svc_drop(rqstp); | 1169 | svc_drop(rqstp); |
1152 | return 0; | 1170 | return 0; |
1153 | 1171 | ||
1154 | err_short_len: | 1172 | err_short_len: |
1155 | svc_printk(rqstp, "short len %Zd, dropping request\n", | 1173 | svc_printk(rqstp, "short len %Zd, dropping request\n", |
1156 | argv->iov_len); | 1174 | argv->iov_len); |
1157 | 1175 | ||
1158 | goto dropit; /* drop request */ | 1176 | goto dropit; /* drop request */ |
1159 | 1177 | ||
1160 | err_bad_dir: | 1178 | err_bad_dir: |
1161 | svc_printk(rqstp, "bad direction %d, dropping request\n", dir); | 1179 | svc_printk(rqstp, "bad direction %d, dropping request\n", dir); |
1162 | 1180 | ||
1163 | serv->sv_stats->rpcbadfmt++; | 1181 | serv->sv_stats->rpcbadfmt++; |
1164 | goto dropit; /* drop request */ | 1182 | goto dropit; /* drop request */ |
1165 | 1183 | ||
1166 | err_bad_rpc: | 1184 | err_bad_rpc: |
1167 | serv->sv_stats->rpcbadfmt++; | 1185 | serv->sv_stats->rpcbadfmt++; |
1168 | svc_putnl(resv, 1); /* REJECT */ | 1186 | svc_putnl(resv, 1); /* REJECT */ |
1169 | svc_putnl(resv, 0); /* RPC_MISMATCH */ | 1187 | svc_putnl(resv, 0); /* RPC_MISMATCH */ |
1170 | svc_putnl(resv, 2); /* Only RPCv2 supported */ | 1188 | svc_putnl(resv, 2); /* Only RPCv2 supported */ |
1171 | svc_putnl(resv, 2); | 1189 | svc_putnl(resv, 2); |
1172 | goto sendit; | 1190 | goto sendit; |
1173 | 1191 | ||
1174 | err_bad_auth: | 1192 | err_bad_auth: |
1175 | dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); | 1193 | dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); |
1176 | serv->sv_stats->rpcbadauth++; | 1194 | serv->sv_stats->rpcbadauth++; |
1177 | /* Restore write pointer to location of accept status: */ | 1195 | /* Restore write pointer to location of accept status: */ |
1178 | xdr_ressize_check(rqstp, reply_statp); | 1196 | xdr_ressize_check(rqstp, reply_statp); |
1179 | svc_putnl(resv, 1); /* REJECT */ | 1197 | svc_putnl(resv, 1); /* REJECT */ |
1180 | svc_putnl(resv, 1); /* AUTH_ERROR */ | 1198 | svc_putnl(resv, 1); /* AUTH_ERROR */ |
1181 | svc_putnl(resv, ntohl(auth_stat)); /* status */ | 1199 | svc_putnl(resv, ntohl(auth_stat)); /* status */ |
1182 | goto sendit; | 1200 | goto sendit; |
1183 | 1201 | ||
1184 | err_bad_prog: | 1202 | err_bad_prog: |
1185 | dprintk("svc: unknown program %d\n", prog); | 1203 | dprintk("svc: unknown program %d\n", prog); |
1186 | serv->sv_stats->rpcbadfmt++; | 1204 | serv->sv_stats->rpcbadfmt++; |
1187 | svc_putnl(resv, RPC_PROG_UNAVAIL); | 1205 | svc_putnl(resv, RPC_PROG_UNAVAIL); |
1188 | goto sendit; | 1206 | goto sendit; |
1189 | 1207 | ||
1190 | err_bad_vers: | 1208 | err_bad_vers: |
1191 | svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", | 1209 | svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", |
1192 | vers, prog, progp->pg_name); | 1210 | vers, prog, progp->pg_name); |
1193 | 1211 | ||
1194 | serv->sv_stats->rpcbadfmt++; | 1212 | serv->sv_stats->rpcbadfmt++; |
1195 | svc_putnl(resv, RPC_PROG_MISMATCH); | 1213 | svc_putnl(resv, RPC_PROG_MISMATCH); |
1196 | svc_putnl(resv, progp->pg_lovers); | 1214 | svc_putnl(resv, progp->pg_lovers); |
1197 | svc_putnl(resv, progp->pg_hivers); | 1215 | svc_putnl(resv, progp->pg_hivers); |
1198 | goto sendit; | 1216 | goto sendit; |
1199 | 1217 | ||
1200 | err_bad_proc: | 1218 | err_bad_proc: |
1201 | svc_printk(rqstp, "unknown procedure (%d)\n", proc); | 1219 | svc_printk(rqstp, "unknown procedure (%d)\n", proc); |
1202 | 1220 | ||
1203 | serv->sv_stats->rpcbadfmt++; | 1221 | serv->sv_stats->rpcbadfmt++; |
1204 | svc_putnl(resv, RPC_PROC_UNAVAIL); | 1222 | svc_putnl(resv, RPC_PROC_UNAVAIL); |
1205 | goto sendit; | 1223 | goto sendit; |
1206 | 1224 | ||
1207 | err_garbage: | 1225 | err_garbage: |
1208 | svc_printk(rqstp, "failed to decode args\n"); | 1226 | svc_printk(rqstp, "failed to decode args\n"); |
1209 | 1227 | ||
1210 | rpc_stat = rpc_garbage_args; | 1228 | rpc_stat = rpc_garbage_args; |
1211 | err_bad: | 1229 | err_bad: |
1212 | serv->sv_stats->rpcbadfmt++; | 1230 | serv->sv_stats->rpcbadfmt++; |
1213 | svc_putnl(resv, ntohl(rpc_stat)); | 1231 | svc_putnl(resv, ntohl(rpc_stat)); |
1214 | goto sendit; | 1232 | goto sendit; |
1215 | } | 1233 | } |