]>
Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
30 | * Copyright (c) 2011, 2012, Intel Corporation. | |
31 | */ | |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | */ | |
36 | #define DEBUG_SUBSYSTEM S_CLASS | |
37 | ||
38 | ||
e27db149 GKH |
39 | #include "../include/obd_support.h" |
40 | #include "../include/obd.h" | |
41 | #include "../include/lprocfs_status.h" | |
42 | #include "../include/lustre/lustre_idl.h" | |
43 | #include "../include/lustre_net.h" | |
44 | #include "../include/obd_class.h" | |
d7e09d03 PT |
45 | #include "ptlrpc_internal.h" |
46 | ||
47 | ||
b3c986d3 | 48 | static struct ll_rpc_opcode { |
2a763282 KM |
49 | __u32 opcode; |
50 | const char *opname; | |
d7e09d03 PT |
51 | } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = { |
52 | { OST_REPLY, "ost_reply" }, | |
53 | { OST_GETATTR, "ost_getattr" }, | |
54 | { OST_SETATTR, "ost_setattr" }, | |
55 | { OST_READ, "ost_read" }, | |
56 | { OST_WRITE, "ost_write" }, | |
57 | { OST_CREATE , "ost_create" }, | |
58 | { OST_DESTROY, "ost_destroy" }, | |
59 | { OST_GET_INFO, "ost_get_info" }, | |
60 | { OST_CONNECT, "ost_connect" }, | |
61 | { OST_DISCONNECT, "ost_disconnect" }, | |
62 | { OST_PUNCH, "ost_punch" }, | |
63 | { OST_OPEN, "ost_open" }, | |
64 | { OST_CLOSE, "ost_close" }, | |
65 | { OST_STATFS, "ost_statfs" }, | |
66 | { 14, NULL }, /* formerly OST_SAN_READ */ | |
67 | { 15, NULL }, /* formerly OST_SAN_WRITE */ | |
68 | { OST_SYNC, "ost_sync" }, | |
69 | { OST_SET_INFO, "ost_set_info" }, | |
70 | { OST_QUOTACHECK, "ost_quotacheck" }, | |
71 | { OST_QUOTACTL, "ost_quotactl" }, | |
72 | { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" }, | |
73 | { MDS_GETATTR, "mds_getattr" }, | |
74 | { MDS_GETATTR_NAME, "mds_getattr_lock" }, | |
75 | { MDS_CLOSE, "mds_close" }, | |
76 | { MDS_REINT, "mds_reint" }, | |
77 | { MDS_READPAGE, "mds_readpage" }, | |
78 | { MDS_CONNECT, "mds_connect" }, | |
79 | { MDS_DISCONNECT, "mds_disconnect" }, | |
80 | { MDS_GETSTATUS, "mds_getstatus" }, | |
81 | { MDS_STATFS, "mds_statfs" }, | |
82 | { MDS_PIN, "mds_pin" }, | |
83 | { MDS_UNPIN, "mds_unpin" }, | |
84 | { MDS_SYNC, "mds_sync" }, | |
85 | { MDS_DONE_WRITING, "mds_done_writing" }, | |
86 | { MDS_SET_INFO, "mds_set_info" }, | |
87 | { MDS_QUOTACHECK, "mds_quotacheck" }, | |
88 | { MDS_QUOTACTL, "mds_quotactl" }, | |
89 | { MDS_GETXATTR, "mds_getxattr" }, | |
90 | { MDS_SETXATTR, "mds_setxattr" }, | |
91 | { MDS_WRITEPAGE, "mds_writepage" }, | |
92 | { MDS_IS_SUBDIR, "mds_is_subdir" }, | |
93 | { MDS_GET_INFO, "mds_get_info" }, | |
94 | { MDS_HSM_STATE_GET, "mds_hsm_state_get" }, | |
95 | { MDS_HSM_STATE_SET, "mds_hsm_state_set" }, | |
96 | { MDS_HSM_ACTION, "mds_hsm_action" }, | |
97 | { MDS_HSM_PROGRESS, "mds_hsm_progress" }, | |
98 | { MDS_HSM_REQUEST, "mds_hsm_request" }, | |
99 | { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" }, | |
100 | { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" }, | |
101 | { MDS_SWAP_LAYOUTS, "mds_swap_layouts" }, | |
102 | { LDLM_ENQUEUE, "ldlm_enqueue" }, | |
103 | { LDLM_CONVERT, "ldlm_convert" }, | |
104 | { LDLM_CANCEL, "ldlm_cancel" }, | |
105 | { LDLM_BL_CALLBACK, "ldlm_bl_callback" }, | |
106 | { LDLM_CP_CALLBACK, "ldlm_cp_callback" }, | |
107 | { LDLM_GL_CALLBACK, "ldlm_gl_callback" }, | |
108 | { LDLM_SET_INFO, "ldlm_set_info" }, | |
109 | { MGS_CONNECT, "mgs_connect" }, | |
110 | { MGS_DISCONNECT, "mgs_disconnect" }, | |
111 | { MGS_EXCEPTION, "mgs_exception" }, | |
112 | { MGS_TARGET_REG, "mgs_target_reg" }, | |
113 | { MGS_TARGET_DEL, "mgs_target_del" }, | |
114 | { MGS_SET_INFO, "mgs_set_info" }, | |
115 | { MGS_CONFIG_READ, "mgs_config_read" }, | |
116 | { OBD_PING, "obd_ping" }, | |
65f1c781 | 117 | { OBD_LOG_CANCEL, "llog_cancel" }, |
d7e09d03 PT |
118 | { OBD_QC_CALLBACK, "obd_quota_callback" }, |
119 | { OBD_IDX_READ, "dt_index_read" }, | |
65f1c781 | 120 | { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" }, |
d7e09d03 | 121 | { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" }, |
1d8cb70c | 122 | { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" }, |
d7e09d03 PT |
123 | { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" }, |
124 | { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" }, | |
125 | { LLOG_ORIGIN_CONNECT, "llog_origin_connect" }, | |
126 | { LLOG_CATINFO, "llog_catinfo" }, | |
127 | { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" }, | |
128 | { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" }, | |
129 | { QUOTA_DQACQ, "quota_acquire" }, | |
130 | { QUOTA_DQREL, "quota_release" }, | |
131 | { SEQ_QUERY, "seq_query" }, | |
132 | { SEC_CTX_INIT, "sec_ctx_init" }, | |
1d8cb70c | 133 | { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" }, |
d7e09d03 PT |
134 | { SEC_CTX_FINI, "sec_ctx_fini" }, |
135 | { FLD_QUERY, "fld_query" }, | |
136 | { UPDATE_OBJ, "update_obj" }, | |
137 | }; | |
138 | ||
b3c986d3 | 139 | static struct ll_eopcode { |
2a763282 KM |
140 | __u32 opcode; |
141 | const char *opname; | |
d7e09d03 PT |
142 | } ll_eopcode_table[EXTRA_LAST_OPC] = { |
143 | { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" }, | |
144 | { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" }, | |
145 | { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" }, | |
146 | { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" }, | |
147 | { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" }, | |
148 | { MDS_REINT_SETATTR, "mds_reint_setattr" }, | |
149 | { MDS_REINT_CREATE, "mds_reint_create" }, | |
150 | { MDS_REINT_LINK, "mds_reint_link" }, | |
151 | { MDS_REINT_UNLINK, "mds_reint_unlink" }, | |
152 | { MDS_REINT_RENAME, "mds_reint_rename" }, | |
153 | { MDS_REINT_OPEN, "mds_reint_open" }, | |
154 | { MDS_REINT_SETXATTR, "mds_reint_setxattr" }, | |
155 | { BRW_READ_BYTES, "read_bytes" }, | |
156 | { BRW_WRITE_BYTES, "write_bytes" }, | |
157 | }; | |
158 | ||
159 | const char *ll_opcode2str(__u32 opcode) | |
160 | { | |
161 | /* When one of the assertions below fail, chances are that: | |
162 | * 1) A new opcode was added in include/lustre/lustre_idl.h, | |
163 | * but is missing from the table above. | |
164 | * or 2) The opcode space was renumbered or rearranged, | |
165 | * and the opcode_offset() function in | |
166 | * ptlrpc_internal.h needs to be modified. | |
167 | */ | |
168 | __u32 offset = opcode_offset(opcode); | |
169 | LASSERTF(offset < LUSTRE_MAX_OPCODES, | |
170 | "offset %u >= LUSTRE_MAX_OPCODES %u\n", | |
171 | offset, LUSTRE_MAX_OPCODES); | |
172 | LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode, | |
173 | "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n", | |
174 | offset, ll_rpc_opcode_table[offset].opcode, opcode); | |
175 | return ll_rpc_opcode_table[offset].opname; | |
176 | } | |
177 | ||
b3c986d3 | 178 | static const char *ll_eopcode2str(__u32 opcode) |
d7e09d03 PT |
179 | { |
180 | LASSERT(ll_eopcode_table[opcode].opcode == opcode); | |
181 | return ll_eopcode_table[opcode].opname; | |
182 | } | |
b3c986d3 | 183 | |
700815d4 DE |
184 | static void |
185 | ptlrpc_ldebugfs_register(struct dentry *root, char *dir, | |
186 | char *name, | |
187 | struct dentry **debugfs_root_ret, | |
188 | struct lprocfs_stats **stats_ret) | |
d7e09d03 | 189 | { |
700815d4 | 190 | struct dentry *svc_debugfs_entry; |
d7e09d03 PT |
191 | struct lprocfs_stats *svc_stats; |
192 | int i, rc; | |
193 | unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX | | |
194 | LPROCFS_CNTR_STDDEV; | |
195 | ||
700815d4 | 196 | LASSERT(*debugfs_root_ret == NULL); |
d7e09d03 PT |
197 | LASSERT(*stats_ret == NULL); |
198 | ||
1d8cb70c GD |
199 | svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES, |
200 | 0); | |
d7e09d03 PT |
201 | if (svc_stats == NULL) |
202 | return; | |
203 | ||
700815d4 DE |
204 | if (dir != NULL) { |
205 | svc_debugfs_entry = ldebugfs_register(dir, root, NULL, NULL); | |
206 | if (IS_ERR(svc_debugfs_entry)) { | |
d7e09d03 PT |
207 | lprocfs_free_stats(&svc_stats); |
208 | return; | |
209 | } | |
210 | } else { | |
700815d4 | 211 | svc_debugfs_entry = root; |
d7e09d03 PT |
212 | } |
213 | ||
214 | lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR, | |
215 | svc_counter_config, "req_waittime", "usec"); | |
216 | lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR, | |
217 | svc_counter_config, "req_qdepth", "reqs"); | |
218 | lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR, | |
219 | svc_counter_config, "req_active", "reqs"); | |
220 | lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT, | |
221 | svc_counter_config, "req_timeout", "sec"); | |
222 | lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR, | |
223 | svc_counter_config, "reqbuf_avail", "bufs"); | |
224 | for (i = 0; i < EXTRA_LAST_OPC; i++) { | |
225 | char *units; | |
226 | ||
3949015e | 227 | switch (i) { |
d7e09d03 PT |
228 | case BRW_WRITE_BYTES: |
229 | case BRW_READ_BYTES: | |
230 | units = "bytes"; | |
231 | break; | |
232 | default: | |
233 | units = "reqs"; | |
234 | break; | |
235 | } | |
236 | lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i, | |
237 | svc_counter_config, | |
238 | ll_eopcode2str(i), units); | |
239 | } | |
240 | for (i = 0; i < LUSTRE_MAX_OPCODES; i++) { | |
241 | __u32 opcode = ll_rpc_opcode_table[i].opcode; | |
242 | lprocfs_counter_init(svc_stats, | |
243 | EXTRA_MAX_OPCODES + i, svc_counter_config, | |
244 | ll_opcode2str(opcode), "usec"); | |
245 | } | |
246 | ||
700815d4 | 247 | rc = ldebugfs_register_stats(svc_debugfs_entry, name, svc_stats); |
d7e09d03 | 248 | if (rc < 0) { |
700815d4 DE |
249 | if (dir != NULL) |
250 | ldebugfs_remove(&svc_debugfs_entry); | |
d7e09d03 PT |
251 | lprocfs_free_stats(&svc_stats); |
252 | } else { | |
700815d4 DE |
253 | if (dir != NULL) |
254 | *debugfs_root_ret = svc_debugfs_entry; | |
d7e09d03 PT |
255 | *stats_ret = svc_stats; |
256 | } | |
257 | } | |
258 | ||
259 | static int | |
73bb1da6 | 260 | ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v) |
d7e09d03 | 261 | { |
73bb1da6 | 262 | struct ptlrpc_service *svc = m->private; |
d7e09d03 | 263 | struct ptlrpc_service_part *svcpt; |
d0bfef31 CH |
264 | int total = 0; |
265 | int i; | |
d7e09d03 | 266 | |
d7e09d03 PT |
267 | ptlrpc_service_for_each_part(svcpt, i, svc) |
268 | total += svcpt->scp_hist_nrqbds; | |
269 | ||
8faeebdf JP |
270 | seq_printf(m, "%d\n", total); |
271 | return 0; | |
d7e09d03 | 272 | } |
73bb1da6 | 273 | LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len); |
d7e09d03 PT |
274 | |
275 | static int | |
73bb1da6 | 276 | ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 277 | { |
73bb1da6 | 278 | struct ptlrpc_service *svc = m->private; |
d7e09d03 | 279 | struct ptlrpc_service_part *svcpt; |
d0bfef31 CH |
280 | int total = 0; |
281 | int i; | |
d7e09d03 | 282 | |
d7e09d03 PT |
283 | ptlrpc_service_for_each_part(svcpt, i, svc) |
284 | total += svc->srv_hist_nrqbds_cpt_max; | |
285 | ||
8faeebdf JP |
286 | seq_printf(m, "%d\n", total); |
287 | return 0; | |
d7e09d03 PT |
288 | } |
289 | ||
73bb1da6 | 290 | static ssize_t |
e84962e3 | 291 | ptlrpc_lprocfs_req_history_max_seq_write(struct file *file, |
d0bfef31 CH |
292 | const char __user *buffer, |
293 | size_t count, loff_t *off) | |
d7e09d03 | 294 | { |
73bb1da6 | 295 | struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private; |
d0bfef31 CH |
296 | int bufpages; |
297 | int val; | |
298 | int rc; | |
d7e09d03 PT |
299 | |
300 | rc = lprocfs_write_helper(buffer, count, &val); | |
301 | if (rc < 0) | |
302 | return rc; | |
303 | ||
304 | if (val < 0) | |
305 | return -ERANGE; | |
306 | ||
307 | /* This sanity check is more of an insanity check; we can still | |
308 | * hose a kernel by allowing the request history to grow too | |
309 | * far. */ | |
310 | bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
4f6cc9ab | 311 | if (val > totalram_pages / (2 * bufpages)) |
d7e09d03 PT |
312 | return -ERANGE; |
313 | ||
314 | spin_lock(&svc->srv_lock); | |
315 | ||
316 | if (val == 0) | |
317 | svc->srv_hist_nrqbds_cpt_max = 0; | |
318 | else | |
319 | svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts)); | |
320 | ||
321 | spin_unlock(&svc->srv_lock); | |
322 | ||
323 | return count; | |
324 | } | |
73bb1da6 | 325 | LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max); |
d7e09d03 | 326 | |
673a6796 OD |
327 | |
328 | static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr, | |
329 | char *buf) | |
d7e09d03 | 330 | { |
673a6796 OD |
331 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, |
332 | srv_kobj); | |
d7e09d03 | 333 | |
673a6796 | 334 | return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts); |
d7e09d03 PT |
335 | } |
336 | ||
673a6796 OD |
337 | static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr, |
338 | const char *buffer, size_t count) | |
d7e09d03 | 339 | { |
673a6796 OD |
340 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, |
341 | srv_kobj); | |
342 | unsigned long val; | |
343 | int rc = kstrtoul(buffer, 10, &val); | |
d7e09d03 PT |
344 | |
345 | if (rc < 0) | |
346 | return rc; | |
347 | ||
348 | if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT) | |
349 | return -ERANGE; | |
350 | ||
351 | spin_lock(&svc->srv_lock); | |
352 | if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) { | |
353 | spin_unlock(&svc->srv_lock); | |
354 | return -ERANGE; | |
355 | } | |
356 | ||
357 | svc->srv_nthrs_cpt_init = val / svc->srv_ncpts; | |
358 | ||
359 | spin_unlock(&svc->srv_lock); | |
360 | ||
361 | return count; | |
362 | } | |
673a6796 | 363 | LUSTRE_RW_ATTR(threads_min); |
d7e09d03 | 364 | |
673a6796 OD |
365 | static ssize_t threads_started_show(struct kobject *kobj, |
366 | struct attribute *attr, | |
367 | char *buf) | |
d7e09d03 | 368 | { |
673a6796 OD |
369 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, |
370 | srv_kobj); | |
d7e09d03 | 371 | struct ptlrpc_service_part *svcpt; |
673a6796 OD |
372 | int total = 0; |
373 | int i; | |
d7e09d03 PT |
374 | |
375 | ptlrpc_service_for_each_part(svcpt, i, svc) | |
376 | total += svcpt->scp_nthrs_running; | |
377 | ||
673a6796 | 378 | return sprintf(buf, "%d\n", total); |
d7e09d03 | 379 | } |
673a6796 | 380 | LUSTRE_RO_ATTR(threads_started); |
d7e09d03 | 381 | |
673a6796 OD |
382 | static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr, |
383 | char *buf) | |
d7e09d03 | 384 | { |
673a6796 OD |
385 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, |
386 | srv_kobj); | |
d7e09d03 | 387 | |
673a6796 | 388 | return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts); |
d7e09d03 PT |
389 | } |
390 | ||
673a6796 OD |
391 | static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr, |
392 | const char *buffer, size_t count) | |
d7e09d03 | 393 | { |
673a6796 OD |
394 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, |
395 | srv_kobj); | |
396 | unsigned long val; | |
397 | int rc = kstrtoul(buffer, 10, &val); | |
d7e09d03 PT |
398 | |
399 | if (rc < 0) | |
400 | return rc; | |
401 | ||
402 | if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT) | |
403 | return -ERANGE; | |
404 | ||
405 | spin_lock(&svc->srv_lock); | |
406 | if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) { | |
407 | spin_unlock(&svc->srv_lock); | |
408 | return -ERANGE; | |
409 | } | |
410 | ||
411 | svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts; | |
412 | ||
413 | spin_unlock(&svc->srv_lock); | |
414 | ||
415 | return count; | |
416 | } | |
673a6796 | 417 | LUSTRE_RW_ATTR(threads_max); |
d7e09d03 PT |
418 | |
419 | /** | |
420 | * \addtogoup nrs | |
421 | * @{ | |
422 | */ | |
423 | extern struct nrs_core nrs_core; | |
424 | ||
425 | /** | |
426 | * Translates \e ptlrpc_nrs_pol_state values to human-readable strings. | |
427 | * | |
428 | * \param[in] state The policy state | |
429 | */ | |
430 | static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state) | |
431 | { | |
432 | switch (state) { | |
433 | default: | |
434 | LBUG(); | |
435 | case NRS_POL_STATE_INVALID: | |
436 | return "invalid"; | |
437 | case NRS_POL_STATE_STOPPED: | |
438 | return "stopped"; | |
439 | case NRS_POL_STATE_STOPPING: | |
440 | return "stopping"; | |
441 | case NRS_POL_STATE_STARTING: | |
442 | return "starting"; | |
443 | case NRS_POL_STATE_STARTED: | |
444 | return "started"; | |
445 | } | |
446 | } | |
447 | ||
448 | /** | |
449 | * Obtains status information for \a policy. | |
450 | * | |
451 | * Information is copied in \a info. | |
452 | * | |
453 | * \param[in] policy The policy | |
454 | * \param[out] info Holds returned status information | |
455 | */ | |
456 | void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy, | |
457 | struct ptlrpc_nrs_pol_info *info) | |
458 | { | |
459 | LASSERT(policy != NULL); | |
460 | LASSERT(info != NULL); | |
5e42bc9d | 461 | assert_spin_locked(&policy->pol_nrs->nrs_lock); |
d7e09d03 PT |
462 | |
463 | memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX); | |
464 | ||
465 | info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK); | |
466 | info->pi_state = policy->pol_state; | |
467 | /** | |
468 | * XXX: These are accessed without holding | |
469 | * ptlrpc_service_part::scp_req_lock. | |
470 | */ | |
471 | info->pi_req_queued = policy->pol_req_queued; | |
472 | info->pi_req_started = policy->pol_req_started; | |
473 | } | |
474 | ||
475 | /** | |
476 | * Reads and prints policy status information for all policies of a PTLRPC | |
477 | * service. | |
478 | */ | |
73bb1da6 | 479 | static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 480 | { |
d0bfef31 CH |
481 | struct ptlrpc_service *svc = m->private; |
482 | struct ptlrpc_service_part *svcpt; | |
483 | struct ptlrpc_nrs *nrs; | |
484 | struct ptlrpc_nrs_policy *policy; | |
485 | struct ptlrpc_nrs_pol_info *infos; | |
486 | struct ptlrpc_nrs_pol_info tmp; | |
487 | unsigned num_pols; | |
488 | unsigned pol_idx = 0; | |
489 | bool hp = false; | |
490 | int i; | |
491 | int rc = 0; | |
d7e09d03 PT |
492 | |
493 | /** | |
494 | * Serialize NRS core lprocfs operations with policy registration/ | |
495 | * unregistration. | |
496 | */ | |
497 | mutex_lock(&nrs_core.nrs_mutex); | |
498 | ||
499 | /** | |
500 | * Use the first service partition's regular NRS head in order to obtain | |
501 | * the number of policies registered with NRS heads of this service. All | |
502 | * service partitions will have the same number of policies. | |
503 | */ | |
504 | nrs = nrs_svcpt2nrs(svc->srv_parts[0], false); | |
505 | ||
506 | spin_lock(&nrs->nrs_lock); | |
507 | num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols; | |
508 | spin_unlock(&nrs->nrs_lock); | |
509 | ||
9ae10597 | 510 | infos = kcalloc(num_pols, sizeof(*infos), GFP_NOFS); |
a9b3e8f3 JL |
511 | if (infos == NULL) { |
512 | rc = -ENOMEM; | |
848e90c3 | 513 | goto unlock; |
a9b3e8f3 | 514 | } |
d7e09d03 PT |
515 | again: |
516 | ||
517 | ptlrpc_service_for_each_part(svcpt, i, svc) { | |
518 | nrs = nrs_svcpt2nrs(svcpt, hp); | |
519 | spin_lock(&nrs->nrs_lock); | |
520 | ||
521 | pol_idx = 0; | |
522 | ||
523 | list_for_each_entry(policy, &nrs->nrs_policy_list, | |
524 | pol_list) { | |
525 | LASSERT(pol_idx < num_pols); | |
526 | ||
527 | nrs_policy_get_info_locked(policy, &tmp); | |
528 | /** | |
529 | * Copy values when handling the first service | |
530 | * partition. | |
531 | */ | |
532 | if (i == 0) { | |
533 | memcpy(infos[pol_idx].pi_name, tmp.pi_name, | |
534 | NRS_POL_NAME_MAX); | |
535 | memcpy(&infos[pol_idx].pi_state, &tmp.pi_state, | |
536 | sizeof(tmp.pi_state)); | |
537 | infos[pol_idx].pi_fallback = tmp.pi_fallback; | |
538 | /** | |
539 | * For the rest of the service partitions | |
540 | * sanity-check the values we get. | |
541 | */ | |
542 | } else { | |
543 | LASSERT(strncmp(infos[pol_idx].pi_name, | |
544 | tmp.pi_name, | |
545 | NRS_POL_NAME_MAX) == 0); | |
546 | /** | |
547 | * Not asserting ptlrpc_nrs_pol_info::pi_state, | |
548 | * because it may be different between | |
549 | * instances of the same policy in different | |
550 | * service partitions. | |
551 | */ | |
552 | LASSERT(infos[pol_idx].pi_fallback == | |
553 | tmp.pi_fallback); | |
554 | } | |
555 | ||
556 | infos[pol_idx].pi_req_queued += tmp.pi_req_queued; | |
557 | infos[pol_idx].pi_req_started += tmp.pi_req_started; | |
558 | ||
559 | pol_idx++; | |
560 | } | |
561 | spin_unlock(&nrs->nrs_lock); | |
562 | } | |
563 | ||
564 | /** | |
565 | * Policy status information output is in YAML format. | |
566 | * For example: | |
567 | * | |
568 | * regular_requests: | |
569 | * - name: fifo | |
570 | * state: started | |
571 | * fallback: yes | |
572 | * queued: 0 | |
573 | * active: 0 | |
574 | * | |
575 | * - name: crrn | |
576 | * state: started | |
577 | * fallback: no | |
578 | * queued: 2015 | |
579 | * active: 384 | |
580 | * | |
581 | * high_priority_requests: | |
582 | * - name: fifo | |
583 | * state: started | |
584 | * fallback: yes | |
585 | * queued: 0 | |
586 | * active: 2 | |
587 | * | |
588 | * - name: crrn | |
589 | * state: stopped | |
590 | * fallback: no | |
591 | * queued: 0 | |
592 | * active: 0 | |
593 | */ | |
73bb1da6 PT |
594 | seq_printf(m, "%s\n", |
595 | !hp ? "\nregular_requests:" : "high_priority_requests:"); | |
d7e09d03 PT |
596 | |
597 | for (pol_idx = 0; pol_idx < num_pols; pol_idx++) { | |
73bb1da6 | 598 | seq_printf(m, " - name: %s\n" |
d7e09d03 PT |
599 | " state: %s\n" |
600 | " fallback: %s\n" | |
601 | " queued: %-20d\n" | |
602 | " active: %-20d\n\n", | |
603 | infos[pol_idx].pi_name, | |
604 | nrs_state2str(infos[pol_idx].pi_state), | |
605 | infos[pol_idx].pi_fallback ? "yes" : "no", | |
606 | (int)infos[pol_idx].pi_req_queued, | |
607 | (int)infos[pol_idx].pi_req_started); | |
d7e09d03 PT |
608 | } |
609 | ||
610 | if (!hp && nrs_svc_has_hp(svc)) { | |
611 | memset(infos, 0, num_pols * sizeof(*infos)); | |
612 | ||
613 | /** | |
614 | * Redo the processing for the service's HP NRS heads' policies. | |
615 | */ | |
616 | hp = true; | |
617 | goto again; | |
618 | } | |
619 | ||
848e90c3 JL |
620 | kfree(infos); |
621 | unlock: | |
d7e09d03 PT |
622 | mutex_unlock(&nrs_core.nrs_mutex); |
623 | ||
0a3bdb00 | 624 | return rc; |
d7e09d03 PT |
625 | } |
626 | ||
627 | /** | |
b6da17f3 | 628 | * The longest valid command string is the maximum policy name size, plus the |
d7e09d03 PT |
629 | * length of the " reg" substring |
630 | */ | |
631 | #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1) | |
632 | ||
633 | /** | |
634 | * Starts and stops a given policy on a PTLRPC service. | |
635 | * | |
636 | * Commands consist of the policy name, followed by an optional [reg|hp] token; | |
637 | * if the optional token is omitted, the operation is performed on both the | |
638 | * regular and high-priority (if the service has one) NRS head. | |
639 | */ | |
29dd4f9b | 640 | static ssize_t ptlrpc_lprocfs_nrs_seq_write(struct file *file, |
d0bfef31 CH |
641 | const char __user *buffer, |
642 | size_t count, loff_t *off) | |
d7e09d03 | 643 | { |
73bb1da6 | 644 | struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private; |
d0bfef31 CH |
645 | enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH; |
646 | char *cmd; | |
647 | char *cmd_copy = NULL; | |
648 | char *token; | |
649 | int rc = 0; | |
d7e09d03 | 650 | |
848e90c3 JL |
651 | if (count >= LPROCFS_NRS_WR_MAX_CMD) |
652 | return -EINVAL; | |
d7e09d03 | 653 | |
9ae10597 | 654 | cmd = kzalloc(LPROCFS_NRS_WR_MAX_CMD, GFP_NOFS); |
597851ac | 655 | if (!cmd) |
848e90c3 | 656 | return -ENOMEM; |
d7e09d03 PT |
657 | /** |
658 | * strsep() modifies its argument, so keep a copy | |
659 | */ | |
660 | cmd_copy = cmd; | |
661 | ||
a9b3e8f3 JL |
662 | if (copy_from_user(cmd, buffer, count)) { |
663 | rc = -EFAULT; | |
664 | goto out; | |
665 | } | |
d7e09d03 PT |
666 | |
667 | cmd[count] = '\0'; | |
668 | ||
669 | token = strsep(&cmd, " "); | |
670 | ||
a9b3e8f3 JL |
671 | if (strlen(token) > NRS_POL_NAME_MAX - 1) { |
672 | rc = -EINVAL; | |
673 | goto out; | |
674 | } | |
d7e09d03 PT |
675 | |
676 | /** | |
677 | * No [reg|hp] token has been specified | |
678 | */ | |
679 | if (cmd == NULL) | |
680 | goto default_queue; | |
681 | ||
682 | /** | |
683 | * The second token is either NULL, or an optional [reg|hp] string | |
684 | */ | |
685 | if (strcmp(cmd, "reg") == 0) | |
686 | queue = PTLRPC_NRS_QUEUE_REG; | |
687 | else if (strcmp(cmd, "hp") == 0) | |
688 | queue = PTLRPC_NRS_QUEUE_HP; | |
a9b3e8f3 JL |
689 | else { |
690 | rc = -EINVAL; | |
691 | goto out; | |
692 | } | |
d7e09d03 PT |
693 | |
694 | default_queue: | |
695 | ||
a9b3e8f3 JL |
696 | if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc)) { |
697 | rc = -ENODEV; | |
698 | goto out; | |
4d1d413a | 699 | } else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc)) |
d7e09d03 PT |
700 | queue = PTLRPC_NRS_QUEUE_REG; |
701 | ||
702 | /** | |
703 | * Serialize NRS core lprocfs operations with policy registration/ | |
704 | * unregistration. | |
705 | */ | |
706 | mutex_lock(&nrs_core.nrs_mutex); | |
707 | ||
708 | rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START, | |
709 | false, NULL); | |
710 | ||
711 | mutex_unlock(&nrs_core.nrs_mutex); | |
712 | out: | |
848e90c3 | 713 | kfree(cmd_copy); |
d7e09d03 | 714 | |
0a3bdb00 | 715 | return rc < 0 ? rc : count; |
d7e09d03 | 716 | } |
73bb1da6 | 717 | LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs); |
d7e09d03 PT |
718 | |
719 | /** @} nrs */ | |
720 | ||
721 | struct ptlrpc_srh_iterator { | |
722 | int srhi_idx; | |
723 | __u64 srhi_seq; | |
724 | struct ptlrpc_request *srhi_req; | |
725 | }; | |
726 | ||
b3c986d3 | 727 | static int |
d7e09d03 PT |
728 | ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt, |
729 | struct ptlrpc_srh_iterator *srhi, | |
730 | __u64 seq) | |
731 | { | |
d0bfef31 CH |
732 | struct list_head *e; |
733 | struct ptlrpc_request *req; | |
d7e09d03 PT |
734 | |
735 | if (srhi->srhi_req != NULL && | |
736 | srhi->srhi_seq > svcpt->scp_hist_seq_culled && | |
737 | srhi->srhi_seq <= seq) { | |
738 | /* If srhi_req was set previously, hasn't been culled and | |
739 | * we're searching for a seq on or after it (i.e. more | |
740 | * recent), search from it onwards. | |
741 | * Since the service history is LRU (i.e. culled reqs will | |
742 | * be near the head), we shouldn't have to do long | |
743 | * re-scans */ | |
744 | LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq, | |
b0f5aad5 | 745 | "%s:%d: seek seq %llu, request seq %llu\n", |
d7e09d03 PT |
746 | svcpt->scp_service->srv_name, svcpt->scp_cpt, |
747 | srhi->srhi_seq, srhi->srhi_req->rq_history_seq); | |
748 | LASSERTF(!list_empty(&svcpt->scp_hist_reqs), | |
2d00bd17 | 749 | "%s:%d: seek offset %llu, request seq %llu, last culled %llu\n", |
d7e09d03 PT |
750 | svcpt->scp_service->srv_name, svcpt->scp_cpt, |
751 | seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled); | |
752 | e = &srhi->srhi_req->rq_history_list; | |
753 | } else { | |
754 | /* search from start */ | |
755 | e = svcpt->scp_hist_reqs.next; | |
756 | } | |
757 | ||
758 | while (e != &svcpt->scp_hist_reqs) { | |
759 | req = list_entry(e, struct ptlrpc_request, rq_history_list); | |
760 | ||
761 | if (req->rq_history_seq >= seq) { | |
762 | srhi->srhi_seq = req->rq_history_seq; | |
763 | srhi->srhi_req = req; | |
764 | return 0; | |
765 | } | |
766 | e = e->next; | |
767 | } | |
768 | ||
769 | return -ENOENT; | |
770 | } | |
771 | ||
772 | /* | |
773 | * ptlrpc history sequence is used as "position" of seq_file, in some case, | |
774 | * seq_read() will increase "position" to indicate reading the next | |
775 | * element, however, low bits of history sequence are reserved for CPT id | |
776 | * (check the details from comments before ptlrpc_req_add_history), which | |
777 | * means seq_read() might change CPT id of history sequence and never | |
778 | * finish reading of requests on a CPT. To make it work, we have to shift | |
779 | * CPT id to high bits and timestamp to low bits, so seq_read() will only | |
780 | * increase timestamp which can correctly indicate the next position. | |
781 | */ | |
782 | ||
783 | /* convert seq_file pos to cpt */ | |
784 | #define PTLRPC_REQ_POS2CPT(svc, pos) \ | |
785 | ((svc)->srv_cpt_bits == 0 ? 0 : \ | |
786 | (__u64)(pos) >> (64 - (svc)->srv_cpt_bits)) | |
787 | ||
788 | /* make up seq_file pos from cpt */ | |
789 | #define PTLRPC_REQ_CPT2POS(svc, cpt) \ | |
790 | ((svc)->srv_cpt_bits == 0 ? 0 : \ | |
791 | (cpt) << (64 - (svc)->srv_cpt_bits)) | |
792 | ||
793 | /* convert sequence to position */ | |
794 | #define PTLRPC_REQ_SEQ2POS(svc, seq) \ | |
795 | ((svc)->srv_cpt_bits == 0 ? (seq) : \ | |
796 | ((seq) >> (svc)->srv_cpt_bits) | \ | |
797 | ((seq) << (64 - (svc)->srv_cpt_bits))) | |
798 | ||
799 | /* convert position to sequence */ | |
800 | #define PTLRPC_REQ_POS2SEQ(svc, pos) \ | |
801 | ((svc)->srv_cpt_bits == 0 ? (pos) : \ | |
802 | ((__u64)(pos) << (svc)->srv_cpt_bits) | \ | |
803 | ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits))) | |
804 | ||
805 | static void * | |
806 | ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos) | |
807 | { | |
808 | struct ptlrpc_service *svc = s->private; | |
809 | struct ptlrpc_service_part *svcpt; | |
810 | struct ptlrpc_srh_iterator *srhi; | |
811 | unsigned int cpt; | |
812 | int rc; | |
813 | int i; | |
814 | ||
815 | if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */ | |
2d00bd17 JP |
816 | CWARN("Failed to read request history because size of loff_t %d can't match size of u64\n", |
817 | (int)sizeof(loff_t)); | |
d7e09d03 PT |
818 | return NULL; |
819 | } | |
820 | ||
9ae10597 | 821 | srhi = kzalloc(sizeof(*srhi), GFP_NOFS); |
597851ac | 822 | if (!srhi) |
d7e09d03 PT |
823 | return NULL; |
824 | ||
825 | srhi->srhi_seq = 0; | |
826 | srhi->srhi_req = NULL; | |
827 | ||
828 | cpt = PTLRPC_REQ_POS2CPT(svc, *pos); | |
829 | ||
830 | ptlrpc_service_for_each_part(svcpt, i, svc) { | |
831 | if (i < cpt) /* skip */ | |
832 | continue; | |
833 | if (i > cpt) /* make up the lowest position for this CPT */ | |
834 | *pos = PTLRPC_REQ_CPT2POS(svc, i); | |
835 | ||
836 | spin_lock(&svcpt->scp_lock); | |
837 | rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, | |
838 | PTLRPC_REQ_POS2SEQ(svc, *pos)); | |
839 | spin_unlock(&svcpt->scp_lock); | |
840 | if (rc == 0) { | |
841 | *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq); | |
842 | srhi->srhi_idx = i; | |
843 | return srhi; | |
844 | } | |
845 | } | |
846 | ||
9ae10597 | 847 | kfree(srhi); |
d7e09d03 PT |
848 | return NULL; |
849 | } | |
850 | ||
851 | static void | |
852 | ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter) | |
853 | { | |
854 | struct ptlrpc_srh_iterator *srhi = iter; | |
855 | ||
848e90c3 | 856 | kfree(srhi); |
d7e09d03 PT |
857 | } |
858 | ||
859 | static void * | |
860 | ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s, | |
861 | void *iter, loff_t *pos) | |
862 | { | |
d0bfef31 CH |
863 | struct ptlrpc_service *svc = s->private; |
864 | struct ptlrpc_srh_iterator *srhi = iter; | |
865 | struct ptlrpc_service_part *svcpt; | |
866 | __u64 seq; | |
867 | int rc; | |
868 | int i; | |
d7e09d03 PT |
869 | |
870 | for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) { | |
871 | svcpt = svc->srv_parts[i]; | |
872 | ||
873 | if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */ | |
874 | srhi->srhi_req = NULL; | |
875 | seq = srhi->srhi_seq = 0; | |
876 | } else { /* the next sequence */ | |
877 | seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits); | |
878 | } | |
879 | ||
880 | spin_lock(&svcpt->scp_lock); | |
881 | rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq); | |
882 | spin_unlock(&svcpt->scp_lock); | |
883 | if (rc == 0) { | |
884 | *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq); | |
885 | srhi->srhi_idx = i; | |
886 | return srhi; | |
887 | } | |
888 | } | |
889 | ||
9ae10597 | 890 | kfree(srhi); |
d7e09d03 PT |
891 | return NULL; |
892 | } | |
893 | ||
894 | /* common ost/mdt so_req_printer */ | |
895 | void target_print_req(void *seq_file, struct ptlrpc_request *req) | |
896 | { | |
897 | /* Called holding srv_lock with irqs disabled. | |
898 | * Print specific req contents and a newline. | |
899 | * CAVEAT EMPTOR: check request message length before printing!!! | |
900 | * You might have received any old crap so you must be just as | |
901 | * careful here as the service's request parser!!! */ | |
902 | struct seq_file *sf = seq_file; | |
903 | ||
904 | switch (req->rq_phase) { | |
905 | case RQ_PHASE_NEW: | |
906 | /* still awaiting a service thread's attention, or rejected | |
907 | * because the generic request message didn't unpack */ | |
908 | seq_printf(sf, "<not swabbed>\n"); | |
909 | break; | |
910 | case RQ_PHASE_INTERPRET: | |
911 | /* being handled, so basic msg swabbed, and opc is valid | |
912 | * but racing with mds_handle() */ | |
913 | case RQ_PHASE_COMPLETE: | |
914 | /* been handled by mds_handle() reply state possibly still | |
915 | * volatile */ | |
916 | seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg)); | |
917 | break; | |
918 | default: | |
919 | DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase); | |
920 | } | |
921 | } | |
922 | EXPORT_SYMBOL(target_print_req); | |
923 | ||
924 | static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter) | |
925 | { | |
d0bfef31 CH |
926 | struct ptlrpc_service *svc = s->private; |
927 | struct ptlrpc_srh_iterator *srhi = iter; | |
928 | struct ptlrpc_service_part *svcpt; | |
929 | struct ptlrpc_request *req; | |
930 | int rc; | |
d7e09d03 PT |
931 | |
932 | LASSERT(srhi->srhi_idx < svc->srv_ncpts); | |
933 | ||
934 | svcpt = svc->srv_parts[srhi->srhi_idx]; | |
935 | ||
936 | spin_lock(&svcpt->scp_lock); | |
937 | ||
938 | rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq); | |
939 | ||
940 | if (rc == 0) { | |
941 | req = srhi->srhi_req; | |
942 | ||
943 | /* Print common req fields. | |
944 | * CAVEAT EMPTOR: we're racing with the service handler | |
945 | * here. The request could contain any old crap, so you | |
946 | * must be just as careful as the service's request | |
947 | * parser. Currently I only print stuff here I know is OK | |
948 | * to look at coz it was set up in request_in_callback()!!! */ | |
b0f5aad5 | 949 | seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%ld:%lds(%+lds) ", |
d7e09d03 PT |
950 | req->rq_history_seq, libcfs_nid2str(req->rq_self), |
951 | libcfs_id2str(req->rq_peer), req->rq_xid, | |
952 | req->rq_reqlen, ptlrpc_rqphase2str(req), | |
953 | req->rq_arrival_time.tv_sec, | |
954 | req->rq_sent - req->rq_arrival_time.tv_sec, | |
955 | req->rq_sent - req->rq_deadline); | |
956 | if (svc->srv_ops.so_req_printer == NULL) | |
957 | seq_printf(s, "\n"); | |
958 | else | |
959 | svc->srv_ops.so_req_printer(s, srhi->srhi_req); | |
960 | } | |
961 | ||
962 | spin_unlock(&svcpt->scp_lock); | |
963 | return rc; | |
964 | } | |
965 | ||
966 | static int | |
967 | ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file) | |
968 | { | |
969 | static struct seq_operations sops = { | |
970 | .start = ptlrpc_lprocfs_svc_req_history_start, | |
971 | .stop = ptlrpc_lprocfs_svc_req_history_stop, | |
972 | .next = ptlrpc_lprocfs_svc_req_history_next, | |
973 | .show = ptlrpc_lprocfs_svc_req_history_show, | |
974 | }; | |
d0bfef31 CH |
975 | struct seq_file *seqf; |
976 | int rc; | |
d7e09d03 | 977 | |
d7e09d03 | 978 | rc = seq_open(file, &sops); |
73bb1da6 | 979 | if (rc) |
d7e09d03 | 980 | return rc; |
d7e09d03 PT |
981 | |
982 | seqf = file->private_data; | |
700815d4 | 983 | seqf->private = inode->i_private; |
d7e09d03 PT |
984 | return 0; |
985 | } | |
986 | ||
987 | /* See also lprocfs_rd_timeouts */ | |
73bb1da6 | 988 | static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 989 | { |
d0bfef31 CH |
990 | struct ptlrpc_service *svc = m->private; |
991 | struct ptlrpc_service_part *svcpt; | |
992 | struct dhms ts; | |
993 | time_t worstt; | |
994 | unsigned int cur; | |
995 | unsigned int worst; | |
996 | int i; | |
d7e09d03 PT |
997 | |
998 | if (AT_OFF) { | |
73bb1da6 | 999 | seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n", |
d7e09d03 | 1000 | obd_timeout); |
73bb1da6 | 1001 | return 0; |
d7e09d03 PT |
1002 | } |
1003 | ||
1004 | ptlrpc_service_for_each_part(svcpt, i, svc) { | |
1005 | cur = at_get(&svcpt->scp_at_estimate); | |
1006 | worst = svcpt->scp_at_estimate.at_worst_ever; | |
1007 | worstt = svcpt->scp_at_estimate.at_worst_time; | |
7264b8a5 | 1008 | s2dhms(&ts, get_seconds() - worstt); |
d7e09d03 | 1009 | |
73bb1da6 PT |
1010 | seq_printf(m, "%10s : cur %3u worst %3u (at %ld, " |
1011 | DHMS_FMT" ago) ", "service", | |
1012 | cur, worst, worstt, DHMS_VARS(&ts)); | |
1013 | ||
1014 | lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate); | |
d7e09d03 PT |
1015 | } |
1016 | ||
73bb1da6 | 1017 | return 0; |
d7e09d03 | 1018 | } |
73bb1da6 | 1019 | LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts); |
d7e09d03 | 1020 | |
b40881e5 OD |
1021 | static ssize_t high_priority_ratio_show(struct kobject *kobj, |
1022 | struct attribute *attr, | |
1023 | char *buf) | |
d7e09d03 | 1024 | { |
b40881e5 OD |
1025 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, |
1026 | srv_kobj); | |
1027 | return sprintf(buf, "%d\n", svc->srv_hpreq_ratio); | |
d7e09d03 PT |
1028 | } |
1029 | ||
b40881e5 OD |
1030 | static ssize_t high_priority_ratio_store(struct kobject *kobj, |
1031 | struct attribute *attr, | |
1032 | const char *buffer, | |
1033 | size_t count) | |
d7e09d03 | 1034 | { |
b40881e5 OD |
1035 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, |
1036 | srv_kobj); | |
1037 | int rc; | |
b815555d | 1038 | int val; |
d7e09d03 | 1039 | |
b815555d | 1040 | rc = kstrtoint(buffer, 10, &val); |
d7e09d03 PT |
1041 | if (rc < 0) |
1042 | return rc; | |
1043 | ||
1044 | if (val < 0) | |
1045 | return -ERANGE; | |
1046 | ||
1047 | spin_lock(&svc->srv_lock); | |
1048 | svc->srv_hpreq_ratio = val; | |
1049 | spin_unlock(&svc->srv_lock); | |
1050 | ||
1051 | return count; | |
1052 | } | |
b40881e5 | 1053 | LUSTRE_RW_ATTR(high_priority_ratio); |
d7e09d03 | 1054 | |
328676f8 | 1055 | static struct attribute *ptlrpc_svc_attrs[] = { |
673a6796 OD |
1056 | &lustre_attr_threads_min.attr, |
1057 | &lustre_attr_threads_started.attr, | |
1058 | &lustre_attr_threads_max.attr, | |
b40881e5 | 1059 | &lustre_attr_high_priority_ratio.attr, |
328676f8 OD |
1060 | NULL, |
1061 | }; | |
1062 | ||
1063 | static void ptlrpc_sysfs_svc_release(struct kobject *kobj) | |
1064 | { | |
1065 | struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service, | |
1066 | srv_kobj); | |
1067 | ||
1068 | complete(&svc->srv_kobj_unregister); | |
1069 | } | |
1070 | ||
1071 | static struct kobj_type ptlrpc_svc_ktype = { | |
1072 | .default_attrs = ptlrpc_svc_attrs, | |
1073 | .sysfs_ops = &lustre_sysfs_ops, | |
1074 | .release = ptlrpc_sysfs_svc_release, | |
1075 | }; | |
1076 | ||
1077 | void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc) | |
1078 | { | |
1079 | /* Let's see if we had a chance at initialization first */ | |
1080 | if (svc->srv_kobj.kset) { | |
1081 | kobject_put(&svc->srv_kobj); | |
1082 | wait_for_completion(&svc->srv_kobj_unregister); | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | int ptlrpc_sysfs_register_service(struct kset *parent, | |
1087 | struct ptlrpc_service *svc) | |
1088 | { | |
1089 | int rc; | |
1090 | ||
1091 | svc->srv_kobj.kset = parent; | |
1092 | init_completion(&svc->srv_kobj_unregister); | |
1093 | rc = kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype, NULL, | |
1094 | "%s", svc->srv_name); | |
1095 | ||
1096 | return rc; | |
1097 | } | |
1098 | ||
700815d4 DE |
1099 | void ptlrpc_ldebugfs_register_service(struct dentry *entry, |
1100 | struct ptlrpc_service *svc) | |
d7e09d03 PT |
1101 | { |
1102 | struct lprocfs_vars lproc_vars[] = { | |
d7e09d03 | 1103 | {.name = "req_buffer_history_len", |
73bb1da6 | 1104 | .fops = &ptlrpc_lprocfs_req_history_len_fops, |
d7e09d03 PT |
1105 | .data = svc}, |
1106 | {.name = "req_buffer_history_max", | |
73bb1da6 | 1107 | .fops = &ptlrpc_lprocfs_req_history_max_fops, |
d7e09d03 | 1108 | .data = svc}, |
d7e09d03 | 1109 | {.name = "timeouts", |
73bb1da6 | 1110 | .fops = &ptlrpc_lprocfs_timeouts_fops, |
d7e09d03 PT |
1111 | .data = svc}, |
1112 | {.name = "nrs_policies", | |
73bb1da6 | 1113 | .fops = &ptlrpc_lprocfs_nrs_fops, |
d7e09d03 PT |
1114 | .data = svc}, |
1115 | {NULL} | |
1116 | }; | |
dddf4c23 | 1117 | static const struct file_operations req_history_fops = { |
d7e09d03 PT |
1118 | .owner = THIS_MODULE, |
1119 | .open = ptlrpc_lprocfs_svc_req_history_open, | |
1120 | .read = seq_read, | |
1121 | .llseek = seq_lseek, | |
1122 | .release = lprocfs_seq_release, | |
1123 | }; | |
1124 | ||
1125 | int rc; | |
1126 | ||
700815d4 DE |
1127 | ptlrpc_ldebugfs_register(entry, svc->srv_name, |
1128 | "stats", &svc->srv_debugfs_entry, | |
1129 | &svc->srv_stats); | |
d7e09d03 | 1130 | |
700815d4 | 1131 | if (svc->srv_debugfs_entry == NULL) |
d7e09d03 PT |
1132 | return; |
1133 | ||
700815d4 | 1134 | ldebugfs_add_vars(svc->srv_debugfs_entry, lproc_vars, NULL); |
d7e09d03 | 1135 | |
700815d4 DE |
1136 | rc = ldebugfs_seq_create(svc->srv_debugfs_entry, "req_history", |
1137 | 0400, &req_history_fops, svc); | |
d7e09d03 PT |
1138 | if (rc) |
1139 | CWARN("Error adding the req_history file\n"); | |
1140 | } | |
1141 | ||
1142 | void ptlrpc_lprocfs_register_obd(struct obd_device *obddev) | |
1143 | { | |
61e87ab0 DE |
1144 | ptlrpc_ldebugfs_register(obddev->obd_debugfs_entry, NULL, "stats", |
1145 | &obddev->obd_svc_debugfs_entry, | |
1146 | &obddev->obd_svc_stats); | |
d7e09d03 PT |
1147 | } |
1148 | EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd); | |
1149 | ||
1150 | void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount) | |
1151 | { | |
1152 | struct lprocfs_stats *svc_stats; | |
1153 | __u32 op = lustre_msg_get_opc(req->rq_reqmsg); | |
1154 | int opc = opcode_offset(op); | |
1155 | ||
1156 | svc_stats = req->rq_import->imp_obd->obd_svc_stats; | |
1157 | if (svc_stats == NULL || opc <= 0) | |
1158 | return; | |
1159 | LASSERT(opc < LUSTRE_MAX_OPCODES); | |
1160 | if (!(op == LDLM_ENQUEUE || op == MDS_REINT)) | |
1161 | lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount); | |
1162 | } | |
1163 | ||
1164 | void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) | |
1165 | { | |
1166 | struct lprocfs_stats *svc_stats; | |
1167 | int idx; | |
1168 | ||
1169 | if (!req->rq_import) | |
1170 | return; | |
1171 | svc_stats = req->rq_import->imp_obd->obd_svc_stats; | |
1172 | if (!svc_stats) | |
1173 | return; | |
1174 | idx = lustre_msg_get_opc(req->rq_reqmsg); | |
1175 | switch (idx) { | |
1176 | case OST_READ: | |
1177 | idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR; | |
1178 | break; | |
1179 | case OST_WRITE: | |
1180 | idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR; | |
1181 | break; | |
1182 | default: | |
1183 | LASSERTF(0, "unsupported opcode %u\n", idx); | |
1184 | break; | |
1185 | } | |
1186 | ||
1187 | lprocfs_counter_add(svc_stats, idx, bytes); | |
1188 | } | |
1189 | ||
1190 | EXPORT_SYMBOL(ptlrpc_lprocfs_brw); | |
1191 | ||
1192 | void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc) | |
1193 | { | |
700815d4 DE |
1194 | if (svc->srv_debugfs_entry != NULL) |
1195 | ldebugfs_remove(&svc->srv_debugfs_entry); | |
d7e09d03 PT |
1196 | |
1197 | if (svc->srv_stats) | |
1198 | lprocfs_free_stats(&svc->srv_stats); | |
1199 | } | |
1200 | ||
1201 | void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) | |
1202 | { | |
61e87ab0 DE |
1203 | if (!IS_ERR_OR_NULL(obd->obd_svc_debugfs_entry)) |
1204 | ldebugfs_remove(&obd->obd_svc_debugfs_entry); | |
d7e09d03 PT |
1205 | |
1206 | if (obd->obd_svc_stats) | |
1207 | lprocfs_free_stats(&obd->obd_svc_stats); | |
1208 | } | |
1209 | EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd); | |
1210 | ||
1211 | ||
1212 | #define BUFLEN (UUID_MAX + 5) | |
1213 | ||
e84962e3 | 1214 | int lprocfs_wr_evict_client(struct file *file, const char __user *buffer, |
73bb1da6 | 1215 | size_t count, loff_t *off) |
d7e09d03 | 1216 | { |
73bb1da6 | 1217 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d0bfef31 CH |
1218 | char *kbuf; |
1219 | char *tmpbuf; | |
d7e09d03 | 1220 | |
9ae10597 | 1221 | kbuf = kzalloc(BUFLEN, GFP_NOFS); |
597851ac | 1222 | if (!kbuf) |
d7e09d03 PT |
1223 | return -ENOMEM; |
1224 | ||
1225 | /* | |
1226 | * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1 | |
1227 | * bytes into kbuf, to ensure that the string is NUL-terminated. | |
1228 | * UUID_MAX should include a trailing NUL already. | |
1229 | */ | |
1230 | if (copy_from_user(kbuf, buffer, | |
1231 | min_t(unsigned long, BUFLEN - 1, count))) { | |
1232 | count = -EFAULT; | |
1233 | goto out; | |
1234 | } | |
1235 | tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count)); | |
1236 | /* Kludge code(deadlock situation): the lprocfs lock has been held | |
b6da17f3 | 1237 | * since the client is evicted by writing client's |
d7e09d03 | 1238 | * uuid/nid to procfs "evict_client" entry. However, |
e4ba525e | 1239 | * obd_export_evict_by_uuid() will call ldebugfs_remove() to destroy |
d7e09d03 PT |
1240 | * the proc entries under the being destroyed export{}, so I have |
1241 | * to drop the lock at first here. | |
1242 | * - jay, jxiong@clusterfs.com */ | |
f9bd9c1a | 1243 | class_incref(obd, __func__, current); |
d7e09d03 PT |
1244 | |
1245 | if (strncmp(tmpbuf, "nid:", 4) == 0) | |
1246 | obd_export_evict_by_nid(obd, tmpbuf + 4); | |
1247 | else if (strncmp(tmpbuf, "uuid:", 5) == 0) | |
1248 | obd_export_evict_by_uuid(obd, tmpbuf + 5); | |
1249 | else | |
1250 | obd_export_evict_by_uuid(obd, tmpbuf); | |
1251 | ||
f9bd9c1a | 1252 | class_decref(obd, __func__, current); |
d7e09d03 PT |
1253 | |
1254 | out: | |
9ae10597 | 1255 | kfree(kbuf); |
d7e09d03 PT |
1256 | return count; |
1257 | } | |
1258 | EXPORT_SYMBOL(lprocfs_wr_evict_client); | |
1259 | ||
1260 | #undef BUFLEN | |
1261 | ||
e84962e3 | 1262 | int lprocfs_wr_ping(struct file *file, const char __user *buffer, |
73bb1da6 | 1263 | size_t count, loff_t *off) |
d7e09d03 | 1264 | { |
73bb1da6 | 1265 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d7e09d03 | 1266 | struct ptlrpc_request *req; |
d0bfef31 | 1267 | int rc; |
d7e09d03 PT |
1268 | |
1269 | LPROCFS_CLIMP_CHECK(obd); | |
1270 | req = ptlrpc_prep_ping(obd->u.cli.cl_import); | |
1271 | LPROCFS_CLIMP_EXIT(obd); | |
1272 | if (req == NULL) | |
0a3bdb00 | 1273 | return -ENOMEM; |
d7e09d03 PT |
1274 | |
1275 | req->rq_send_state = LUSTRE_IMP_FULL; | |
1276 | ||
1277 | rc = ptlrpc_queue_wait(req); | |
1278 | ||
1279 | ptlrpc_req_finished(req); | |
1280 | if (rc >= 0) | |
0a3bdb00 GKH |
1281 | return count; |
1282 | return rc; | |
d7e09d03 PT |
1283 | } |
1284 | EXPORT_SYMBOL(lprocfs_wr_ping); | |
1285 | ||
1286 | /* Write the connection UUID to this file to attempt to connect to that node. | |
1287 | * The connection UUID is a node's primary NID. For example, | |
1288 | * "echo connection=192.168.0.1@tcp0::instance > .../import". | |
1289 | */ | |
e84962e3 | 1290 | int lprocfs_wr_import(struct file *file, const char __user *buffer, |
73bb1da6 | 1291 | size_t count, loff_t *off) |
d7e09d03 | 1292 | { |
73bb1da6 | 1293 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
1294 | struct obd_import *imp = obd->u.cli.cl_import; |
1295 | char *kbuf = NULL; | |
1296 | char *uuid; | |
1297 | char *ptr; | |
1298 | int do_reconn = 1; | |
1299 | const char prefix[] = "connection="; | |
1300 | const int prefix_len = sizeof(prefix) - 1; | |
1301 | ||
1302 | if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len) | |
1303 | return -EINVAL; | |
1304 | ||
9ae10597 | 1305 | kbuf = kzalloc(count + 1, GFP_NOFS); |
597851ac | 1306 | if (!kbuf) |
d7e09d03 PT |
1307 | return -ENOMEM; |
1308 | ||
a9b3e8f3 JL |
1309 | if (copy_from_user(kbuf, buffer, count)) { |
1310 | count = -EFAULT; | |
1311 | goto out; | |
1312 | } | |
d7e09d03 PT |
1313 | |
1314 | kbuf[count] = 0; | |
1315 | ||
1316 | /* only support connection=uuid::instance now */ | |
a9b3e8f3 JL |
1317 | if (strncmp(prefix, kbuf, prefix_len) != 0) { |
1318 | count = -EINVAL; | |
1319 | goto out; | |
1320 | } | |
d7e09d03 PT |
1321 | |
1322 | uuid = kbuf + prefix_len; | |
1323 | ptr = strstr(uuid, "::"); | |
1324 | if (ptr) { | |
1325 | __u32 inst; | |
1326 | char *endptr; | |
1327 | ||
1328 | *ptr = 0; | |
1329 | do_reconn = 0; | |
1330 | ptr += strlen("::"); | |
1331 | inst = simple_strtol(ptr, &endptr, 10); | |
1332 | if (*endptr) { | |
1333 | CERROR("config: wrong instance # %s\n", ptr); | |
1334 | } else if (inst != imp->imp_connect_data.ocd_instance) { | |
2d00bd17 | 1335 | CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n", |
d7e09d03 PT |
1336 | imp->imp_obd->obd_name, |
1337 | imp->imp_connect_data.ocd_instance, inst); | |
1338 | do_reconn = 1; | |
1339 | } else { | |
2d00bd17 | 1340 | CDEBUG(D_INFO, "IR: %s has already been connecting to new target(%u)\n", |
d7e09d03 PT |
1341 | imp->imp_obd->obd_name, inst); |
1342 | } | |
1343 | } | |
1344 | ||
1345 | if (do_reconn) | |
1346 | ptlrpc_recover_import(imp, uuid, 1); | |
1347 | ||
1348 | out: | |
9ae10597 | 1349 | kfree(kbuf); |
d7e09d03 PT |
1350 | return count; |
1351 | } | |
1352 | EXPORT_SYMBOL(lprocfs_wr_import); | |
1353 | ||
73bb1da6 | 1354 | int lprocfs_rd_pinger_recov(struct seq_file *m, void *n) |
d7e09d03 | 1355 | { |
73bb1da6 | 1356 | struct obd_device *obd = m->private; |
d7e09d03 | 1357 | struct obd_import *imp = obd->u.cli.cl_import; |
d7e09d03 PT |
1358 | |
1359 | LPROCFS_CLIMP_CHECK(obd); | |
91b3a685 | 1360 | seq_printf(m, "%d\n", !imp->imp_no_pinger_recover); |
d7e09d03 PT |
1361 | LPROCFS_CLIMP_EXIT(obd); |
1362 | ||
91b3a685 | 1363 | return 0; |
d7e09d03 PT |
1364 | } |
1365 | EXPORT_SYMBOL(lprocfs_rd_pinger_recov); | |
1366 | ||
e84962e3 | 1367 | int lprocfs_wr_pinger_recov(struct file *file, const char __user *buffer, |
73bb1da6 | 1368 | size_t count, loff_t *off) |
d7e09d03 | 1369 | { |
73bb1da6 | 1370 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
1371 | struct client_obd *cli = &obd->u.cli; |
1372 | struct obd_import *imp = cli->cl_import; | |
1373 | int rc, val; | |
1374 | ||
1375 | rc = lprocfs_write_helper(buffer, count, &val); | |
1376 | if (rc < 0) | |
1377 | return rc; | |
1378 | ||
1379 | if (val != 0 && val != 1) | |
1380 | return -ERANGE; | |
1381 | ||
1382 | LPROCFS_CLIMP_CHECK(obd); | |
1383 | spin_lock(&imp->imp_lock); | |
1384 | imp->imp_no_pinger_recover = !val; | |
1385 | spin_unlock(&imp->imp_lock); | |
1386 | LPROCFS_CLIMP_EXIT(obd); | |
1387 | ||
1388 | return count; | |
1389 | ||
1390 | } | |
1391 | EXPORT_SYMBOL(lprocfs_wr_pinger_recov); |