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 | ||
48 | 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 | ||
139 | 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 | ||
9c234f6c | 178 | 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 | } | |
f267cdb4 | 183 | #if defined (CONFIG_PROC_FS) |
d7e09d03 PT |
184 | void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir, |
185 | char *name, struct proc_dir_entry **procroot_ret, | |
186 | struct lprocfs_stats **stats_ret) | |
187 | { | |
188 | struct proc_dir_entry *svc_procroot; | |
189 | struct lprocfs_stats *svc_stats; | |
190 | int i, rc; | |
191 | unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX | | |
192 | LPROCFS_CNTR_STDDEV; | |
193 | ||
194 | LASSERT(*procroot_ret == NULL); | |
195 | LASSERT(*stats_ret == NULL); | |
196 | ||
1d8cb70c GD |
197 | svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES, |
198 | 0); | |
d7e09d03 PT |
199 | if (svc_stats == NULL) |
200 | return; | |
201 | ||
202 | if (dir) { | |
203 | svc_procroot = lprocfs_register(dir, root, NULL, NULL); | |
204 | if (IS_ERR(svc_procroot)) { | |
205 | lprocfs_free_stats(&svc_stats); | |
206 | return; | |
207 | } | |
208 | } else { | |
209 | svc_procroot = root; | |
210 | } | |
211 | ||
212 | lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR, | |
213 | svc_counter_config, "req_waittime", "usec"); | |
214 | lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR, | |
215 | svc_counter_config, "req_qdepth", "reqs"); | |
216 | lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR, | |
217 | svc_counter_config, "req_active", "reqs"); | |
218 | lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT, | |
219 | svc_counter_config, "req_timeout", "sec"); | |
220 | lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR, | |
221 | svc_counter_config, "reqbuf_avail", "bufs"); | |
222 | for (i = 0; i < EXTRA_LAST_OPC; i++) { | |
223 | char *units; | |
224 | ||
3949015e | 225 | switch (i) { |
d7e09d03 PT |
226 | case BRW_WRITE_BYTES: |
227 | case BRW_READ_BYTES: | |
228 | units = "bytes"; | |
229 | break; | |
230 | default: | |
231 | units = "reqs"; | |
232 | break; | |
233 | } | |
234 | lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i, | |
235 | svc_counter_config, | |
236 | ll_eopcode2str(i), units); | |
237 | } | |
238 | for (i = 0; i < LUSTRE_MAX_OPCODES; i++) { | |
239 | __u32 opcode = ll_rpc_opcode_table[i].opcode; | |
240 | lprocfs_counter_init(svc_stats, | |
241 | EXTRA_MAX_OPCODES + i, svc_counter_config, | |
242 | ll_opcode2str(opcode), "usec"); | |
243 | } | |
244 | ||
245 | rc = lprocfs_register_stats(svc_procroot, name, svc_stats); | |
246 | if (rc < 0) { | |
247 | if (dir) | |
248 | lprocfs_remove(&svc_procroot); | |
249 | lprocfs_free_stats(&svc_stats); | |
250 | } else { | |
251 | if (dir) | |
252 | *procroot_ret = svc_procroot; | |
253 | *stats_ret = svc_stats; | |
254 | } | |
255 | } | |
256 | ||
257 | static int | |
73bb1da6 | 258 | ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v) |
d7e09d03 | 259 | { |
73bb1da6 | 260 | struct ptlrpc_service *svc = m->private; |
d7e09d03 PT |
261 | struct ptlrpc_service_part *svcpt; |
262 | int total = 0; | |
263 | int i; | |
264 | ||
d7e09d03 PT |
265 | ptlrpc_service_for_each_part(svcpt, i, svc) |
266 | total += svcpt->scp_hist_nrqbds; | |
267 | ||
73bb1da6 | 268 | return seq_printf(m, "%d\n", total); |
d7e09d03 | 269 | } |
73bb1da6 | 270 | LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len); |
d7e09d03 PT |
271 | |
272 | static int | |
73bb1da6 | 273 | ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 274 | { |
73bb1da6 | 275 | struct ptlrpc_service *svc = m->private; |
d7e09d03 PT |
276 | struct ptlrpc_service_part *svcpt; |
277 | int total = 0; | |
278 | int i; | |
279 | ||
d7e09d03 PT |
280 | ptlrpc_service_for_each_part(svcpt, i, svc) |
281 | total += svc->srv_hist_nrqbds_cpt_max; | |
282 | ||
73bb1da6 | 283 | return seq_printf(m, "%d\n", total); |
d7e09d03 PT |
284 | } |
285 | ||
73bb1da6 PT |
286 | static ssize_t |
287 | ptlrpc_lprocfs_req_history_max_seq_write(struct file *file, const char *buffer, | |
288 | size_t count, loff_t *off) | |
d7e09d03 | 289 | { |
73bb1da6 | 290 | struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
291 | int bufpages; |
292 | int val; | |
293 | int rc; | |
294 | ||
295 | rc = lprocfs_write_helper(buffer, count, &val); | |
296 | if (rc < 0) | |
297 | return rc; | |
298 | ||
299 | if (val < 0) | |
300 | return -ERANGE; | |
301 | ||
302 | /* This sanity check is more of an insanity check; we can still | |
303 | * hose a kernel by allowing the request history to grow too | |
304 | * far. */ | |
305 | bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
4f6cc9ab | 306 | if (val > totalram_pages / (2 * bufpages)) |
d7e09d03 PT |
307 | return -ERANGE; |
308 | ||
309 | spin_lock(&svc->srv_lock); | |
310 | ||
311 | if (val == 0) | |
312 | svc->srv_hist_nrqbds_cpt_max = 0; | |
313 | else | |
314 | svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts)); | |
315 | ||
316 | spin_unlock(&svc->srv_lock); | |
317 | ||
318 | return count; | |
319 | } | |
73bb1da6 | 320 | LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max); |
d7e09d03 PT |
321 | |
322 | static int | |
73bb1da6 | 323 | ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 324 | { |
73bb1da6 | 325 | struct ptlrpc_service *svc = m->private; |
d7e09d03 | 326 | |
73bb1da6 | 327 | return seq_printf(m, "%d\n", |
d7e09d03 PT |
328 | svc->srv_nthrs_cpt_init * svc->srv_ncpts); |
329 | } | |
330 | ||
73bb1da6 PT |
331 | static ssize_t |
332 | ptlrpc_lprocfs_threads_min_seq_write(struct file *file, const char *buffer, | |
333 | size_t count, loff_t *off) | |
d7e09d03 | 334 | { |
73bb1da6 | 335 | struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
336 | int val; |
337 | int rc = lprocfs_write_helper(buffer, count, &val); | |
338 | ||
339 | if (rc < 0) | |
340 | return rc; | |
341 | ||
342 | if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT) | |
343 | return -ERANGE; | |
344 | ||
345 | spin_lock(&svc->srv_lock); | |
346 | if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) { | |
347 | spin_unlock(&svc->srv_lock); | |
348 | return -ERANGE; | |
349 | } | |
350 | ||
351 | svc->srv_nthrs_cpt_init = val / svc->srv_ncpts; | |
352 | ||
353 | spin_unlock(&svc->srv_lock); | |
354 | ||
355 | return count; | |
356 | } | |
73bb1da6 | 357 | LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min); |
d7e09d03 PT |
358 | |
359 | static int | |
73bb1da6 | 360 | ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 361 | { |
73bb1da6 | 362 | struct ptlrpc_service *svc = m->private; |
d7e09d03 PT |
363 | struct ptlrpc_service_part *svcpt; |
364 | int total = 0; | |
365 | int i; | |
366 | ||
367 | ptlrpc_service_for_each_part(svcpt, i, svc) | |
368 | total += svcpt->scp_nthrs_running; | |
369 | ||
73bb1da6 | 370 | return seq_printf(m, "%d\n", total); |
d7e09d03 | 371 | } |
73bb1da6 | 372 | LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started); |
d7e09d03 PT |
373 | |
374 | static int | |
73bb1da6 | 375 | ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 376 | { |
73bb1da6 | 377 | struct ptlrpc_service *svc = m->private; |
d7e09d03 | 378 | |
73bb1da6 | 379 | return seq_printf(m, "%d\n", |
d7e09d03 PT |
380 | svc->srv_nthrs_cpt_limit * svc->srv_ncpts); |
381 | } | |
382 | ||
73bb1da6 PT |
383 | static ssize_t |
384 | ptlrpc_lprocfs_threads_max_seq_write(struct file *file, const char *buffer, | |
385 | size_t count, loff_t *off) | |
d7e09d03 | 386 | { |
73bb1da6 | 387 | struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
388 | int val; |
389 | int rc = lprocfs_write_helper(buffer, count, &val); | |
390 | ||
391 | if (rc < 0) | |
392 | return rc; | |
393 | ||
394 | if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT) | |
395 | return -ERANGE; | |
396 | ||
397 | spin_lock(&svc->srv_lock); | |
398 | if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) { | |
399 | spin_unlock(&svc->srv_lock); | |
400 | return -ERANGE; | |
401 | } | |
402 | ||
403 | svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts; | |
404 | ||
405 | spin_unlock(&svc->srv_lock); | |
406 | ||
407 | return count; | |
408 | } | |
73bb1da6 | 409 | LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max); |
d7e09d03 PT |
410 | |
411 | /** | |
412 | * \addtogoup nrs | |
413 | * @{ | |
414 | */ | |
415 | extern struct nrs_core nrs_core; | |
416 | ||
417 | /** | |
418 | * Translates \e ptlrpc_nrs_pol_state values to human-readable strings. | |
419 | * | |
420 | * \param[in] state The policy state | |
421 | */ | |
422 | static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state) | |
423 | { | |
424 | switch (state) { | |
425 | default: | |
426 | LBUG(); | |
427 | case NRS_POL_STATE_INVALID: | |
428 | return "invalid"; | |
429 | case NRS_POL_STATE_STOPPED: | |
430 | return "stopped"; | |
431 | case NRS_POL_STATE_STOPPING: | |
432 | return "stopping"; | |
433 | case NRS_POL_STATE_STARTING: | |
434 | return "starting"; | |
435 | case NRS_POL_STATE_STARTED: | |
436 | return "started"; | |
437 | } | |
438 | } | |
439 | ||
440 | /** | |
441 | * Obtains status information for \a policy. | |
442 | * | |
443 | * Information is copied in \a info. | |
444 | * | |
445 | * \param[in] policy The policy | |
446 | * \param[out] info Holds returned status information | |
447 | */ | |
448 | void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy, | |
449 | struct ptlrpc_nrs_pol_info *info) | |
450 | { | |
451 | LASSERT(policy != NULL); | |
452 | LASSERT(info != NULL); | |
5e42bc9d | 453 | assert_spin_locked(&policy->pol_nrs->nrs_lock); |
d7e09d03 PT |
454 | |
455 | memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX); | |
456 | ||
457 | info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK); | |
458 | info->pi_state = policy->pol_state; | |
459 | /** | |
460 | * XXX: These are accessed without holding | |
461 | * ptlrpc_service_part::scp_req_lock. | |
462 | */ | |
463 | info->pi_req_queued = policy->pol_req_queued; | |
464 | info->pi_req_started = policy->pol_req_started; | |
465 | } | |
466 | ||
467 | /** | |
468 | * Reads and prints policy status information for all policies of a PTLRPC | |
469 | * service. | |
470 | */ | |
73bb1da6 | 471 | static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n) |
d7e09d03 | 472 | { |
73bb1da6 | 473 | struct ptlrpc_service *svc = m->private; |
d7e09d03 PT |
474 | struct ptlrpc_service_part *svcpt; |
475 | struct ptlrpc_nrs *nrs; | |
476 | struct ptlrpc_nrs_policy *policy; | |
477 | struct ptlrpc_nrs_pol_info *infos; | |
478 | struct ptlrpc_nrs_pol_info tmp; | |
479 | unsigned num_pols; | |
480 | unsigned pol_idx = 0; | |
481 | bool hp = false; | |
482 | int i; | |
483 | int rc = 0; | |
d7e09d03 PT |
484 | |
485 | /** | |
486 | * Serialize NRS core lprocfs operations with policy registration/ | |
487 | * unregistration. | |
488 | */ | |
489 | mutex_lock(&nrs_core.nrs_mutex); | |
490 | ||
491 | /** | |
492 | * Use the first service partition's regular NRS head in order to obtain | |
493 | * the number of policies registered with NRS heads of this service. All | |
494 | * service partitions will have the same number of policies. | |
495 | */ | |
496 | nrs = nrs_svcpt2nrs(svc->srv_parts[0], false); | |
497 | ||
498 | spin_lock(&nrs->nrs_lock); | |
499 | num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols; | |
500 | spin_unlock(&nrs->nrs_lock); | |
501 | ||
502 | OBD_ALLOC(infos, num_pols * sizeof(*infos)); | |
a9b3e8f3 JL |
503 | if (infos == NULL) { |
504 | rc = -ENOMEM; | |
505 | goto out; | |
506 | } | |
d7e09d03 PT |
507 | again: |
508 | ||
509 | ptlrpc_service_for_each_part(svcpt, i, svc) { | |
510 | nrs = nrs_svcpt2nrs(svcpt, hp); | |
511 | spin_lock(&nrs->nrs_lock); | |
512 | ||
513 | pol_idx = 0; | |
514 | ||
515 | list_for_each_entry(policy, &nrs->nrs_policy_list, | |
516 | pol_list) { | |
517 | LASSERT(pol_idx < num_pols); | |
518 | ||
519 | nrs_policy_get_info_locked(policy, &tmp); | |
520 | /** | |
521 | * Copy values when handling the first service | |
522 | * partition. | |
523 | */ | |
524 | if (i == 0) { | |
525 | memcpy(infos[pol_idx].pi_name, tmp.pi_name, | |
526 | NRS_POL_NAME_MAX); | |
527 | memcpy(&infos[pol_idx].pi_state, &tmp.pi_state, | |
528 | sizeof(tmp.pi_state)); | |
529 | infos[pol_idx].pi_fallback = tmp.pi_fallback; | |
530 | /** | |
531 | * For the rest of the service partitions | |
532 | * sanity-check the values we get. | |
533 | */ | |
534 | } else { | |
535 | LASSERT(strncmp(infos[pol_idx].pi_name, | |
536 | tmp.pi_name, | |
537 | NRS_POL_NAME_MAX) == 0); | |
538 | /** | |
539 | * Not asserting ptlrpc_nrs_pol_info::pi_state, | |
540 | * because it may be different between | |
541 | * instances of the same policy in different | |
542 | * service partitions. | |
543 | */ | |
544 | LASSERT(infos[pol_idx].pi_fallback == | |
545 | tmp.pi_fallback); | |
546 | } | |
547 | ||
548 | infos[pol_idx].pi_req_queued += tmp.pi_req_queued; | |
549 | infos[pol_idx].pi_req_started += tmp.pi_req_started; | |
550 | ||
551 | pol_idx++; | |
552 | } | |
553 | spin_unlock(&nrs->nrs_lock); | |
554 | } | |
555 | ||
556 | /** | |
557 | * Policy status information output is in YAML format. | |
558 | * For example: | |
559 | * | |
560 | * regular_requests: | |
561 | * - name: fifo | |
562 | * state: started | |
563 | * fallback: yes | |
564 | * queued: 0 | |
565 | * active: 0 | |
566 | * | |
567 | * - name: crrn | |
568 | * state: started | |
569 | * fallback: no | |
570 | * queued: 2015 | |
571 | * active: 384 | |
572 | * | |
573 | * high_priority_requests: | |
574 | * - name: fifo | |
575 | * state: started | |
576 | * fallback: yes | |
577 | * queued: 0 | |
578 | * active: 2 | |
579 | * | |
580 | * - name: crrn | |
581 | * state: stopped | |
582 | * fallback: no | |
583 | * queued: 0 | |
584 | * active: 0 | |
585 | */ | |
73bb1da6 PT |
586 | seq_printf(m, "%s\n", |
587 | !hp ? "\nregular_requests:" : "high_priority_requests:"); | |
d7e09d03 PT |
588 | |
589 | for (pol_idx = 0; pol_idx < num_pols; pol_idx++) { | |
73bb1da6 | 590 | seq_printf(m, " - name: %s\n" |
d7e09d03 PT |
591 | " state: %s\n" |
592 | " fallback: %s\n" | |
593 | " queued: %-20d\n" | |
594 | " active: %-20d\n\n", | |
595 | infos[pol_idx].pi_name, | |
596 | nrs_state2str(infos[pol_idx].pi_state), | |
597 | infos[pol_idx].pi_fallback ? "yes" : "no", | |
598 | (int)infos[pol_idx].pi_req_queued, | |
599 | (int)infos[pol_idx].pi_req_started); | |
d7e09d03 PT |
600 | } |
601 | ||
602 | if (!hp && nrs_svc_has_hp(svc)) { | |
603 | memset(infos, 0, num_pols * sizeof(*infos)); | |
604 | ||
605 | /** | |
606 | * Redo the processing for the service's HP NRS heads' policies. | |
607 | */ | |
608 | hp = true; | |
609 | goto again; | |
610 | } | |
611 | ||
d7e09d03 PT |
612 | out: |
613 | if (infos) | |
614 | OBD_FREE(infos, num_pols * sizeof(*infos)); | |
615 | ||
616 | mutex_unlock(&nrs_core.nrs_mutex); | |
617 | ||
0a3bdb00 | 618 | return rc; |
d7e09d03 PT |
619 | } |
620 | ||
621 | /** | |
b6da17f3 | 622 | * The longest valid command string is the maximum policy name size, plus the |
d7e09d03 PT |
623 | * length of the " reg" substring |
624 | */ | |
625 | #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1) | |
626 | ||
627 | /** | |
628 | * Starts and stops a given policy on a PTLRPC service. | |
629 | * | |
630 | * Commands consist of the policy name, followed by an optional [reg|hp] token; | |
631 | * if the optional token is omitted, the operation is performed on both the | |
632 | * regular and high-priority (if the service has one) NRS head. | |
633 | */ | |
29dd4f9b AB |
634 | static ssize_t ptlrpc_lprocfs_nrs_seq_write(struct file *file, |
635 | const char __user *buffer, | |
73bb1da6 | 636 | size_t count, loff_t *off) |
d7e09d03 | 637 | { |
73bb1da6 | 638 | struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
639 | enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH; |
640 | char *cmd; | |
641 | char *cmd_copy = NULL; | |
642 | char *token; | |
643 | int rc = 0; | |
d7e09d03 | 644 | |
a9b3e8f3 JL |
645 | if (count >= LPROCFS_NRS_WR_MAX_CMD) { |
646 | rc = -EINVAL; | |
647 | goto out; | |
648 | } | |
d7e09d03 PT |
649 | |
650 | OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD); | |
a9b3e8f3 JL |
651 | if (cmd == NULL) { |
652 | rc = -ENOMEM; | |
653 | goto out; | |
654 | } | |
d7e09d03 PT |
655 | /** |
656 | * strsep() modifies its argument, so keep a copy | |
657 | */ | |
658 | cmd_copy = cmd; | |
659 | ||
a9b3e8f3 JL |
660 | if (copy_from_user(cmd, buffer, count)) { |
661 | rc = -EFAULT; | |
662 | goto out; | |
663 | } | |
d7e09d03 PT |
664 | |
665 | cmd[count] = '\0'; | |
666 | ||
667 | token = strsep(&cmd, " "); | |
668 | ||
a9b3e8f3 JL |
669 | if (strlen(token) > NRS_POL_NAME_MAX - 1) { |
670 | rc = -EINVAL; | |
671 | goto out; | |
672 | } | |
d7e09d03 PT |
673 | |
674 | /** | |
675 | * No [reg|hp] token has been specified | |
676 | */ | |
677 | if (cmd == NULL) | |
678 | goto default_queue; | |
679 | ||
680 | /** | |
681 | * The second token is either NULL, or an optional [reg|hp] string | |
682 | */ | |
683 | if (strcmp(cmd, "reg") == 0) | |
684 | queue = PTLRPC_NRS_QUEUE_REG; | |
685 | else if (strcmp(cmd, "hp") == 0) | |
686 | queue = PTLRPC_NRS_QUEUE_HP; | |
a9b3e8f3 JL |
687 | else { |
688 | rc = -EINVAL; | |
689 | goto out; | |
690 | } | |
d7e09d03 PT |
691 | |
692 | default_queue: | |
693 | ||
a9b3e8f3 JL |
694 | if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc)) { |
695 | rc = -ENODEV; | |
696 | goto out; | |
4d1d413a | 697 | } else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc)) |
d7e09d03 PT |
698 | queue = PTLRPC_NRS_QUEUE_REG; |
699 | ||
700 | /** | |
701 | * Serialize NRS core lprocfs operations with policy registration/ | |
702 | * unregistration. | |
703 | */ | |
704 | mutex_lock(&nrs_core.nrs_mutex); | |
705 | ||
706 | rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START, | |
707 | false, NULL); | |
708 | ||
709 | mutex_unlock(&nrs_core.nrs_mutex); | |
710 | out: | |
711 | if (cmd_copy) | |
712 | OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD); | |
713 | ||
0a3bdb00 | 714 | return rc < 0 ? rc : count; |
d7e09d03 | 715 | } |
73bb1da6 | 716 | LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs); |
d7e09d03 PT |
717 | |
718 | /** @} nrs */ | |
719 | ||
720 | struct ptlrpc_srh_iterator { | |
721 | int srhi_idx; | |
722 | __u64 srhi_seq; | |
723 | struct ptlrpc_request *srhi_req; | |
724 | }; | |
725 | ||
726 | int | |
727 | ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt, | |
728 | struct ptlrpc_srh_iterator *srhi, | |
729 | __u64 seq) | |
730 | { | |
731 | struct list_head *e; | |
732 | struct ptlrpc_request *req; | |
733 | ||
734 | if (srhi->srhi_req != NULL && | |
735 | srhi->srhi_seq > svcpt->scp_hist_seq_culled && | |
736 | srhi->srhi_seq <= seq) { | |
737 | /* If srhi_req was set previously, hasn't been culled and | |
738 | * we're searching for a seq on or after it (i.e. more | |
739 | * recent), search from it onwards. | |
740 | * Since the service history is LRU (i.e. culled reqs will | |
741 | * be near the head), we shouldn't have to do long | |
742 | * re-scans */ | |
743 | LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq, | |
b0f5aad5 | 744 | "%s:%d: seek seq %llu, request seq %llu\n", |
d7e09d03 PT |
745 | svcpt->scp_service->srv_name, svcpt->scp_cpt, |
746 | srhi->srhi_seq, srhi->srhi_req->rq_history_seq); | |
747 | LASSERTF(!list_empty(&svcpt->scp_hist_reqs), | |
2d00bd17 | 748 | "%s:%d: seek offset %llu, request seq %llu, last culled %llu\n", |
d7e09d03 PT |
749 | svcpt->scp_service->srv_name, svcpt->scp_cpt, |
750 | seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled); | |
751 | e = &srhi->srhi_req->rq_history_list; | |
752 | } else { | |
753 | /* search from start */ | |
754 | e = svcpt->scp_hist_reqs.next; | |
755 | } | |
756 | ||
757 | while (e != &svcpt->scp_hist_reqs) { | |
758 | req = list_entry(e, struct ptlrpc_request, rq_history_list); | |
759 | ||
760 | if (req->rq_history_seq >= seq) { | |
761 | srhi->srhi_seq = req->rq_history_seq; | |
762 | srhi->srhi_req = req; | |
763 | return 0; | |
764 | } | |
765 | e = e->next; | |
766 | } | |
767 | ||
768 | return -ENOENT; | |
769 | } | |
770 | ||
771 | /* | |
772 | * ptlrpc history sequence is used as "position" of seq_file, in some case, | |
773 | * seq_read() will increase "position" to indicate reading the next | |
774 | * element, however, low bits of history sequence are reserved for CPT id | |
775 | * (check the details from comments before ptlrpc_req_add_history), which | |
776 | * means seq_read() might change CPT id of history sequence and never | |
777 | * finish reading of requests on a CPT. To make it work, we have to shift | |
778 | * CPT id to high bits and timestamp to low bits, so seq_read() will only | |
779 | * increase timestamp which can correctly indicate the next position. | |
780 | */ | |
781 | ||
782 | /* convert seq_file pos to cpt */ | |
783 | #define PTLRPC_REQ_POS2CPT(svc, pos) \ | |
784 | ((svc)->srv_cpt_bits == 0 ? 0 : \ | |
785 | (__u64)(pos) >> (64 - (svc)->srv_cpt_bits)) | |
786 | ||
787 | /* make up seq_file pos from cpt */ | |
788 | #define PTLRPC_REQ_CPT2POS(svc, cpt) \ | |
789 | ((svc)->srv_cpt_bits == 0 ? 0 : \ | |
790 | (cpt) << (64 - (svc)->srv_cpt_bits)) | |
791 | ||
792 | /* convert sequence to position */ | |
793 | #define PTLRPC_REQ_SEQ2POS(svc, seq) \ | |
794 | ((svc)->srv_cpt_bits == 0 ? (seq) : \ | |
795 | ((seq) >> (svc)->srv_cpt_bits) | \ | |
796 | ((seq) << (64 - (svc)->srv_cpt_bits))) | |
797 | ||
798 | /* convert position to sequence */ | |
799 | #define PTLRPC_REQ_POS2SEQ(svc, pos) \ | |
800 | ((svc)->srv_cpt_bits == 0 ? (pos) : \ | |
801 | ((__u64)(pos) << (svc)->srv_cpt_bits) | \ | |
802 | ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits))) | |
803 | ||
804 | static void * | |
805 | ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos) | |
806 | { | |
807 | struct ptlrpc_service *svc = s->private; | |
808 | struct ptlrpc_service_part *svcpt; | |
809 | struct ptlrpc_srh_iterator *srhi; | |
810 | unsigned int cpt; | |
811 | int rc; | |
812 | int i; | |
813 | ||
814 | if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */ | |
2d00bd17 JP |
815 | CWARN("Failed to read request history because size of loff_t %d can't match size of u64\n", |
816 | (int)sizeof(loff_t)); | |
d7e09d03 PT |
817 | return NULL; |
818 | } | |
819 | ||
820 | OBD_ALLOC(srhi, sizeof(*srhi)); | |
821 | if (srhi == NULL) | |
822 | return NULL; | |
823 | ||
824 | srhi->srhi_seq = 0; | |
825 | srhi->srhi_req = NULL; | |
826 | ||
827 | cpt = PTLRPC_REQ_POS2CPT(svc, *pos); | |
828 | ||
829 | ptlrpc_service_for_each_part(svcpt, i, svc) { | |
830 | if (i < cpt) /* skip */ | |
831 | continue; | |
832 | if (i > cpt) /* make up the lowest position for this CPT */ | |
833 | *pos = PTLRPC_REQ_CPT2POS(svc, i); | |
834 | ||
835 | spin_lock(&svcpt->scp_lock); | |
836 | rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, | |
837 | PTLRPC_REQ_POS2SEQ(svc, *pos)); | |
838 | spin_unlock(&svcpt->scp_lock); | |
839 | if (rc == 0) { | |
840 | *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq); | |
841 | srhi->srhi_idx = i; | |
842 | return srhi; | |
843 | } | |
844 | } | |
845 | ||
846 | OBD_FREE(srhi, sizeof(*srhi)); | |
847 | return NULL; | |
848 | } | |
849 | ||
850 | static void | |
851 | ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter) | |
852 | { | |
853 | struct ptlrpc_srh_iterator *srhi = iter; | |
854 | ||
855 | if (srhi != NULL) | |
856 | OBD_FREE(srhi, sizeof(*srhi)); | |
857 | } | |
858 | ||
859 | static void * | |
860 | ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s, | |
861 | void *iter, loff_t *pos) | |
862 | { | |
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; | |
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 | ||
890 | OBD_FREE(srhi, sizeof(*srhi)); | |
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 | { | |
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; | |
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 | }; | |
d7e09d03 PT |
975 | struct seq_file *seqf; |
976 | int rc; | |
977 | ||
d7e09d03 | 978 | rc = seq_open(file, &sops); |
73bb1da6 | 979 | if (rc) |
d7e09d03 | 980 | return rc; |
d7e09d03 PT |
981 | |
982 | seqf = file->private_data; | |
73bb1da6 | 983 | seqf->private = PDE_DATA(inode); |
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 | { |
73bb1da6 | 990 | struct ptlrpc_service *svc = m->private; |
d7e09d03 PT |
991 | struct ptlrpc_service_part *svcpt; |
992 | struct dhms ts; | |
993 | time_t worstt; | |
994 | unsigned int cur; | |
995 | unsigned int worst; | |
d7e09d03 PT |
996 | int i; |
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 | |
73bb1da6 | 1021 | static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v) |
d7e09d03 | 1022 | { |
73bb1da6 PT |
1023 | struct ptlrpc_service *svc = m->private; |
1024 | return seq_printf(m, "%d", svc->srv_hpreq_ratio); | |
d7e09d03 PT |
1025 | } |
1026 | ||
73bb1da6 PT |
1027 | static ssize_t ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, |
1028 | const char *buffer, | |
1029 | size_t count, | |
1030 | loff_t *off) | |
d7e09d03 | 1031 | { |
73bb1da6 | 1032 | struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
1033 | int rc; |
1034 | int val; | |
1035 | ||
1036 | rc = lprocfs_write_helper(buffer, count, &val); | |
1037 | if (rc < 0) | |
1038 | return rc; | |
1039 | ||
1040 | if (val < 0) | |
1041 | return -ERANGE; | |
1042 | ||
1043 | spin_lock(&svc->srv_lock); | |
1044 | svc->srv_hpreq_ratio = val; | |
1045 | spin_unlock(&svc->srv_lock); | |
1046 | ||
1047 | return count; | |
1048 | } | |
73bb1da6 | 1049 | LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio); |
d7e09d03 PT |
1050 | |
1051 | void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry, | |
1052 | struct ptlrpc_service *svc) | |
1053 | { | |
1054 | struct lprocfs_vars lproc_vars[] = { | |
1055 | {.name = "high_priority_ratio", | |
73bb1da6 | 1056 | .fops = &ptlrpc_lprocfs_hp_ratio_fops, |
d7e09d03 PT |
1057 | .data = svc}, |
1058 | {.name = "req_buffer_history_len", | |
73bb1da6 | 1059 | .fops = &ptlrpc_lprocfs_req_history_len_fops, |
d7e09d03 PT |
1060 | .data = svc}, |
1061 | {.name = "req_buffer_history_max", | |
73bb1da6 | 1062 | .fops = &ptlrpc_lprocfs_req_history_max_fops, |
d7e09d03 PT |
1063 | .data = svc}, |
1064 | {.name = "threads_min", | |
73bb1da6 | 1065 | .fops = &ptlrpc_lprocfs_threads_min_fops, |
d7e09d03 PT |
1066 | .data = svc}, |
1067 | {.name = "threads_max", | |
73bb1da6 | 1068 | .fops = &ptlrpc_lprocfs_threads_max_fops, |
d7e09d03 PT |
1069 | .data = svc}, |
1070 | {.name = "threads_started", | |
73bb1da6 | 1071 | .fops = &ptlrpc_lprocfs_threads_started_fops, |
d7e09d03 PT |
1072 | .data = svc}, |
1073 | {.name = "timeouts", | |
73bb1da6 | 1074 | .fops = &ptlrpc_lprocfs_timeouts_fops, |
d7e09d03 PT |
1075 | .data = svc}, |
1076 | {.name = "nrs_policies", | |
73bb1da6 | 1077 | .fops = &ptlrpc_lprocfs_nrs_fops, |
d7e09d03 PT |
1078 | .data = svc}, |
1079 | {NULL} | |
1080 | }; | |
1081 | static struct file_operations req_history_fops = { | |
1082 | .owner = THIS_MODULE, | |
1083 | .open = ptlrpc_lprocfs_svc_req_history_open, | |
1084 | .read = seq_read, | |
1085 | .llseek = seq_lseek, | |
1086 | .release = lprocfs_seq_release, | |
1087 | }; | |
1088 | ||
1089 | int rc; | |
1090 | ||
1091 | ptlrpc_lprocfs_register(entry, svc->srv_name, | |
1092 | "stats", &svc->srv_procroot, | |
1093 | &svc->srv_stats); | |
1094 | ||
1095 | if (svc->srv_procroot == NULL) | |
1096 | return; | |
1097 | ||
1098 | lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL); | |
1099 | ||
1100 | rc = lprocfs_seq_create(svc->srv_procroot, "req_history", | |
1101 | 0400, &req_history_fops, svc); | |
1102 | if (rc) | |
1103 | CWARN("Error adding the req_history file\n"); | |
1104 | } | |
1105 | ||
1106 | void ptlrpc_lprocfs_register_obd(struct obd_device *obddev) | |
1107 | { | |
1108 | ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats", | |
1109 | &obddev->obd_svc_procroot, | |
1110 | &obddev->obd_svc_stats); | |
1111 | } | |
1112 | EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd); | |
1113 | ||
1114 | void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount) | |
1115 | { | |
1116 | struct lprocfs_stats *svc_stats; | |
1117 | __u32 op = lustre_msg_get_opc(req->rq_reqmsg); | |
1118 | int opc = opcode_offset(op); | |
1119 | ||
1120 | svc_stats = req->rq_import->imp_obd->obd_svc_stats; | |
1121 | if (svc_stats == NULL || opc <= 0) | |
1122 | return; | |
1123 | LASSERT(opc < LUSTRE_MAX_OPCODES); | |
1124 | if (!(op == LDLM_ENQUEUE || op == MDS_REINT)) | |
1125 | lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount); | |
1126 | } | |
1127 | ||
1128 | void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes) | |
1129 | { | |
1130 | struct lprocfs_stats *svc_stats; | |
1131 | int idx; | |
1132 | ||
1133 | if (!req->rq_import) | |
1134 | return; | |
1135 | svc_stats = req->rq_import->imp_obd->obd_svc_stats; | |
1136 | if (!svc_stats) | |
1137 | return; | |
1138 | idx = lustre_msg_get_opc(req->rq_reqmsg); | |
1139 | switch (idx) { | |
1140 | case OST_READ: | |
1141 | idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR; | |
1142 | break; | |
1143 | case OST_WRITE: | |
1144 | idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR; | |
1145 | break; | |
1146 | default: | |
1147 | LASSERTF(0, "unsupported opcode %u\n", idx); | |
1148 | break; | |
1149 | } | |
1150 | ||
1151 | lprocfs_counter_add(svc_stats, idx, bytes); | |
1152 | } | |
1153 | ||
1154 | EXPORT_SYMBOL(ptlrpc_lprocfs_brw); | |
1155 | ||
1156 | void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc) | |
1157 | { | |
1158 | if (svc->srv_procroot != NULL) | |
1159 | lprocfs_remove(&svc->srv_procroot); | |
1160 | ||
1161 | if (svc->srv_stats) | |
1162 | lprocfs_free_stats(&svc->srv_stats); | |
1163 | } | |
1164 | ||
1165 | void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd) | |
1166 | { | |
1167 | if (obd->obd_svc_procroot) | |
1168 | lprocfs_remove(&obd->obd_svc_procroot); | |
1169 | ||
1170 | if (obd->obd_svc_stats) | |
1171 | lprocfs_free_stats(&obd->obd_svc_stats); | |
1172 | } | |
1173 | EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd); | |
1174 | ||
1175 | ||
1176 | #define BUFLEN (UUID_MAX + 5) | |
1177 | ||
1178 | int lprocfs_wr_evict_client(struct file *file, const char *buffer, | |
73bb1da6 | 1179 | size_t count, loff_t *off) |
d7e09d03 | 1180 | { |
73bb1da6 | 1181 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
1182 | char *kbuf; |
1183 | char *tmpbuf; | |
1184 | ||
1185 | OBD_ALLOC(kbuf, BUFLEN); | |
1186 | if (kbuf == NULL) | |
1187 | return -ENOMEM; | |
1188 | ||
1189 | /* | |
1190 | * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1 | |
1191 | * bytes into kbuf, to ensure that the string is NUL-terminated. | |
1192 | * UUID_MAX should include a trailing NUL already. | |
1193 | */ | |
1194 | if (copy_from_user(kbuf, buffer, | |
1195 | min_t(unsigned long, BUFLEN - 1, count))) { | |
1196 | count = -EFAULT; | |
1197 | goto out; | |
1198 | } | |
1199 | tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count)); | |
1200 | /* Kludge code(deadlock situation): the lprocfs lock has been held | |
b6da17f3 | 1201 | * since the client is evicted by writing client's |
d7e09d03 PT |
1202 | * uuid/nid to procfs "evict_client" entry. However, |
1203 | * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy | |
1204 | * the proc entries under the being destroyed export{}, so I have | |
1205 | * to drop the lock at first here. | |
1206 | * - jay, jxiong@clusterfs.com */ | |
f9bd9c1a | 1207 | class_incref(obd, __func__, current); |
d7e09d03 PT |
1208 | |
1209 | if (strncmp(tmpbuf, "nid:", 4) == 0) | |
1210 | obd_export_evict_by_nid(obd, tmpbuf + 4); | |
1211 | else if (strncmp(tmpbuf, "uuid:", 5) == 0) | |
1212 | obd_export_evict_by_uuid(obd, tmpbuf + 5); | |
1213 | else | |
1214 | obd_export_evict_by_uuid(obd, tmpbuf); | |
1215 | ||
f9bd9c1a | 1216 | class_decref(obd, __func__, current); |
d7e09d03 PT |
1217 | |
1218 | out: | |
1219 | OBD_FREE(kbuf, BUFLEN); | |
1220 | return count; | |
1221 | } | |
1222 | EXPORT_SYMBOL(lprocfs_wr_evict_client); | |
1223 | ||
1224 | #undef BUFLEN | |
1225 | ||
1226 | int lprocfs_wr_ping(struct file *file, const char *buffer, | |
73bb1da6 | 1227 | size_t count, loff_t *off) |
d7e09d03 | 1228 | { |
73bb1da6 | 1229 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
1230 | struct ptlrpc_request *req; |
1231 | int rc; | |
d7e09d03 PT |
1232 | |
1233 | LPROCFS_CLIMP_CHECK(obd); | |
1234 | req = ptlrpc_prep_ping(obd->u.cli.cl_import); | |
1235 | LPROCFS_CLIMP_EXIT(obd); | |
1236 | if (req == NULL) | |
0a3bdb00 | 1237 | return -ENOMEM; |
d7e09d03 PT |
1238 | |
1239 | req->rq_send_state = LUSTRE_IMP_FULL; | |
1240 | ||
1241 | rc = ptlrpc_queue_wait(req); | |
1242 | ||
1243 | ptlrpc_req_finished(req); | |
1244 | if (rc >= 0) | |
0a3bdb00 GKH |
1245 | return count; |
1246 | return rc; | |
d7e09d03 PT |
1247 | } |
1248 | EXPORT_SYMBOL(lprocfs_wr_ping); | |
1249 | ||
1250 | /* Write the connection UUID to this file to attempt to connect to that node. | |
1251 | * The connection UUID is a node's primary NID. For example, | |
1252 | * "echo connection=192.168.0.1@tcp0::instance > .../import". | |
1253 | */ | |
1254 | int lprocfs_wr_import(struct file *file, const char *buffer, | |
73bb1da6 | 1255 | size_t count, loff_t *off) |
d7e09d03 | 1256 | { |
73bb1da6 | 1257 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
1258 | struct obd_import *imp = obd->u.cli.cl_import; |
1259 | char *kbuf = NULL; | |
1260 | char *uuid; | |
1261 | char *ptr; | |
1262 | int do_reconn = 1; | |
1263 | const char prefix[] = "connection="; | |
1264 | const int prefix_len = sizeof(prefix) - 1; | |
1265 | ||
1266 | if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len) | |
1267 | return -EINVAL; | |
1268 | ||
1269 | OBD_ALLOC(kbuf, count + 1); | |
1270 | if (kbuf == NULL) | |
1271 | return -ENOMEM; | |
1272 | ||
a9b3e8f3 JL |
1273 | if (copy_from_user(kbuf, buffer, count)) { |
1274 | count = -EFAULT; | |
1275 | goto out; | |
1276 | } | |
d7e09d03 PT |
1277 | |
1278 | kbuf[count] = 0; | |
1279 | ||
1280 | /* only support connection=uuid::instance now */ | |
a9b3e8f3 JL |
1281 | if (strncmp(prefix, kbuf, prefix_len) != 0) { |
1282 | count = -EINVAL; | |
1283 | goto out; | |
1284 | } | |
d7e09d03 PT |
1285 | |
1286 | uuid = kbuf + prefix_len; | |
1287 | ptr = strstr(uuid, "::"); | |
1288 | if (ptr) { | |
1289 | __u32 inst; | |
1290 | char *endptr; | |
1291 | ||
1292 | *ptr = 0; | |
1293 | do_reconn = 0; | |
1294 | ptr += strlen("::"); | |
1295 | inst = simple_strtol(ptr, &endptr, 10); | |
1296 | if (*endptr) { | |
1297 | CERROR("config: wrong instance # %s\n", ptr); | |
1298 | } else if (inst != imp->imp_connect_data.ocd_instance) { | |
2d00bd17 | 1299 | CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n", |
d7e09d03 PT |
1300 | imp->imp_obd->obd_name, |
1301 | imp->imp_connect_data.ocd_instance, inst); | |
1302 | do_reconn = 1; | |
1303 | } else { | |
2d00bd17 | 1304 | CDEBUG(D_INFO, "IR: %s has already been connecting to new target(%u)\n", |
d7e09d03 PT |
1305 | imp->imp_obd->obd_name, inst); |
1306 | } | |
1307 | } | |
1308 | ||
1309 | if (do_reconn) | |
1310 | ptlrpc_recover_import(imp, uuid, 1); | |
1311 | ||
1312 | out: | |
1313 | OBD_FREE(kbuf, count + 1); | |
1314 | return count; | |
1315 | } | |
1316 | EXPORT_SYMBOL(lprocfs_wr_import); | |
1317 | ||
73bb1da6 | 1318 | int lprocfs_rd_pinger_recov(struct seq_file *m, void *n) |
d7e09d03 | 1319 | { |
73bb1da6 | 1320 | struct obd_device *obd = m->private; |
d7e09d03 PT |
1321 | struct obd_import *imp = obd->u.cli.cl_import; |
1322 | int rc; | |
1323 | ||
1324 | LPROCFS_CLIMP_CHECK(obd); | |
73bb1da6 | 1325 | rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover); |
d7e09d03 PT |
1326 | LPROCFS_CLIMP_EXIT(obd); |
1327 | ||
1328 | return rc; | |
1329 | } | |
1330 | EXPORT_SYMBOL(lprocfs_rd_pinger_recov); | |
1331 | ||
1332 | int lprocfs_wr_pinger_recov(struct file *file, const char *buffer, | |
73bb1da6 | 1333 | size_t count, loff_t *off) |
d7e09d03 | 1334 | { |
73bb1da6 | 1335 | struct obd_device *obd = ((struct seq_file *)file->private_data)->private; |
d7e09d03 PT |
1336 | struct client_obd *cli = &obd->u.cli; |
1337 | struct obd_import *imp = cli->cl_import; | |
1338 | int rc, val; | |
1339 | ||
1340 | rc = lprocfs_write_helper(buffer, count, &val); | |
1341 | if (rc < 0) | |
1342 | return rc; | |
1343 | ||
1344 | if (val != 0 && val != 1) | |
1345 | return -ERANGE; | |
1346 | ||
1347 | LPROCFS_CLIMP_CHECK(obd); | |
1348 | spin_lock(&imp->imp_lock); | |
1349 | imp->imp_no_pinger_recover = !val; | |
1350 | spin_unlock(&imp->imp_lock); | |
1351 | LPROCFS_CLIMP_EXIT(obd); | |
1352 | ||
1353 | return count; | |
1354 | ||
1355 | } | |
1356 | EXPORT_SYMBOL(lprocfs_wr_pinger_recov); | |
1357 | ||
f267cdb4 | 1358 | #endif /* CONFIG_PROC_FS */ |