4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 #define DEBUG_SUBSYSTEM S_CLASS
39 #include <obd_support.h>
41 #include <lprocfs_status.h>
42 #include <lustre/lustre_idl.h>
43 #include <lustre_net.h>
44 #include <obd_class.h>
45 #include "ptlrpc_internal.h"
48 struct ll_rpc_opcode {
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" },
117 { OBD_LOG_CANCEL, "llog_cancel" },
118 { OBD_QC_CALLBACK, "obd_quota_callback" },
119 { OBD_IDX_READ, "dt_index_read" },
120 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
121 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
122 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
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" },
133 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
134 { SEC_CTX_FINI, "sec_ctx_fini" },
135 { FLD_QUERY, "fld_query" },
136 { FLD_READ, "fld_read" },
137 { OUT_UPDATE, "out_update" },
138 { LFSCK_NOTIFY, "lfsck_notify" },
139 { LFSCK_QUERY, "lfsck_query" },
145 } ll_eopcode_table[EXTRA_LAST_OPC] = {
146 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
147 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
148 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
149 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
150 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
151 { MDS_REINT_SETATTR, "mds_reint_setattr" },
152 { MDS_REINT_CREATE, "mds_reint_create" },
153 { MDS_REINT_LINK, "mds_reint_link" },
154 { MDS_REINT_UNLINK, "mds_reint_unlink" },
155 { MDS_REINT_RENAME, "mds_reint_rename" },
156 { MDS_REINT_OPEN, "mds_reint_open" },
157 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
158 { BRW_READ_BYTES, "read_bytes" },
159 { BRW_WRITE_BYTES, "write_bytes" },
162 const char *ll_opcode2str(__u32 opcode)
164 /* When one of the assertions below fail, chances are that:
165 * 1) A new opcode was added in include/lustre/lustre_idl.h,
166 * but is missing from the table above.
167 * or 2) The opcode space was renumbered or rearranged,
168 * and the opcode_offset() function in
169 * ptlrpc_internal.h needs to be modified.
171 __u32 offset = opcode_offset(opcode);
172 LASSERTF(offset < LUSTRE_MAX_OPCODES,
173 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
174 offset, LUSTRE_MAX_OPCODES);
175 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
176 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
177 offset, ll_rpc_opcode_table[offset].opcode, opcode);
178 return ll_rpc_opcode_table[offset].opname;
181 const char* ll_eopcode2str(__u32 opcode)
183 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
184 return ll_eopcode_table[opcode].opname;
187 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
188 char *name, struct proc_dir_entry **procroot_ret,
189 struct lprocfs_stats **stats_ret)
191 struct proc_dir_entry *svc_procroot;
192 struct lprocfs_stats *svc_stats;
194 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
197 LASSERT(*procroot_ret == NULL);
198 LASSERT(*stats_ret == NULL);
200 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
201 if (svc_stats == NULL)
205 svc_procroot = lprocfs_seq_register(dir, root, NULL, NULL);
206 if (IS_ERR(svc_procroot)) {
207 lprocfs_free_stats(&svc_stats);
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++) {
228 case BRW_WRITE_BYTES:
236 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
238 ll_eopcode2str(i), units);
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");
247 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
250 lprocfs_remove(&svc_procroot);
251 lprocfs_free_stats(&svc_stats);
254 *procroot_ret = svc_procroot;
255 *stats_ret = svc_stats;
260 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
262 struct ptlrpc_service *svc = m->private;
263 struct ptlrpc_service_part *svcpt;
267 ptlrpc_service_for_each_part(svcpt, i, svc)
268 total += svcpt->scp_hist_nrqbds;
270 return seq_printf(m, "%d\n", total);
272 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
275 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
277 struct ptlrpc_service *svc = m->private;
278 struct ptlrpc_service_part *svcpt;
282 ptlrpc_service_for_each_part(svcpt, i, svc)
283 total += svc->srv_hist_nrqbds_cpt_max;
285 return seq_printf(m, "%d\n", total);
289 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
290 const char __user *buffer,
291 size_t count, loff_t *off)
293 struct seq_file *m = file->private_data;
294 struct ptlrpc_service *svc = m->private;
299 rc = lprocfs_write_helper(buffer, count, &val);
306 /* This sanity check is more of an insanity check; we can still
307 * hose a kernel by allowing the request history to grow too
309 bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
311 if (val > totalram_pages/(2 * bufpages))
314 spin_lock(&svc->srv_lock);
317 svc->srv_hist_nrqbds_cpt_max = 0;
319 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
321 spin_unlock(&svc->srv_lock);
325 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
328 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
330 struct ptlrpc_service *svc = m->private;
332 return seq_printf(m, "%d\n",
333 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
337 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
338 const char __user *buffer,
339 size_t count, loff_t *off)
341 struct seq_file *m = file->private_data;
342 struct ptlrpc_service *svc = m->private;
344 int rc = lprocfs_write_helper(buffer, count, &val);
349 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
352 spin_lock(&svc->srv_lock);
353 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
354 spin_unlock(&svc->srv_lock);
358 svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
360 spin_unlock(&svc->srv_lock);
364 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
367 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
369 struct ptlrpc_service *svc = m->private;
370 struct ptlrpc_service_part *svcpt;
374 ptlrpc_service_for_each_part(svcpt, i, svc)
375 total += svcpt->scp_nthrs_running;
377 return seq_printf(m, "%d\n", total);
379 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
382 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
384 struct ptlrpc_service *svc = m->private;
386 return seq_printf(m, "%d\n",
387 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
391 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
392 const char __user *buffer,
393 size_t count, loff_t *off)
395 struct seq_file *m = file->private_data;
396 struct ptlrpc_service *svc = m->private;
398 int rc = lprocfs_write_helper(buffer, count, &val);
403 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
406 spin_lock(&svc->srv_lock);
407 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
408 spin_unlock(&svc->srv_lock);
412 svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
414 spin_unlock(&svc->srv_lock);
418 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
424 extern struct nrs_core nrs_core;
427 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
429 * \param[in] state The policy state
431 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
436 case NRS_POL_STATE_INVALID:
438 case NRS_POL_STATE_STOPPED:
440 case NRS_POL_STATE_STOPPING:
442 case NRS_POL_STATE_STARTING:
444 case NRS_POL_STATE_STARTED:
450 * Obtains status information for \a policy.
452 * Information is copied in \a info.
454 * \param[in] policy The policy
455 * \param[out] info Holds returned status information
457 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
458 struct ptlrpc_nrs_pol_info *info)
460 LASSERT(policy != NULL);
461 LASSERT(info != NULL);
462 assert_spin_locked(&policy->pol_nrs->nrs_lock);
464 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
466 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
467 info->pi_state = policy->pol_state;
469 * XXX: These are accessed without holding
470 * ptlrpc_service_part::scp_req_lock.
472 info->pi_req_queued = policy->pol_req_queued;
473 info->pi_req_started = policy->pol_req_started;
477 * Reads and prints policy status information for all policies of a PTLRPC
480 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
482 struct ptlrpc_service *svc = m->private;
483 struct ptlrpc_service_part *svcpt;
484 struct ptlrpc_nrs *nrs;
485 struct ptlrpc_nrs_policy *policy;
486 struct ptlrpc_nrs_pol_info *infos;
487 struct ptlrpc_nrs_pol_info tmp;
489 unsigned pol_idx = 0;
496 * Serialize NRS core lprocfs operations with policy registration/
499 mutex_lock(&nrs_core.nrs_mutex);
502 * Use the first service partition's regular NRS head in order to obtain
503 * the number of policies registered with NRS heads of this service. All
504 * service partitions will have the same number of policies.
506 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
508 spin_lock(&nrs->nrs_lock);
509 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
510 spin_unlock(&nrs->nrs_lock);
512 OBD_ALLOC(infos, num_pols * sizeof(*infos));
514 GOTO(out, rc = -ENOMEM);
517 ptlrpc_service_for_each_part(svcpt, i, svc) {
518 nrs = nrs_svcpt2nrs(svcpt, hp);
519 spin_lock(&nrs->nrs_lock);
523 list_for_each_entry(policy, &nrs->nrs_policy_list,
525 LASSERT(pol_idx < num_pols);
527 nrs_policy_get_info_locked(policy, &tmp);
529 * Copy values when handling the first service
533 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
535 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
536 sizeof(tmp.pi_state));
537 infos[pol_idx].pi_fallback = tmp.pi_fallback;
539 * For the rest of the service partitions
540 * sanity-check the values we get.
543 LASSERT(strncmp(infos[pol_idx].pi_name,
545 NRS_POL_NAME_MAX) == 0);
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.
552 LASSERT(infos[pol_idx].pi_fallback ==
556 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
557 infos[pol_idx].pi_req_started += tmp.pi_req_started;
561 spin_unlock(&nrs->nrs_lock);
565 * Policy status information output is in YAML format.
581 * high_priority_requests:
594 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
595 "high_priority_requests:");
597 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
598 seq_printf(m, " - name: %s\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);
610 if (!hp && nrs_svc_has_hp(svc)) {
611 memset(infos, 0, num_pols * sizeof(*infos));
614 * Redo the processing for the service's HP NRS heads' policies.
622 OBD_FREE(infos, num_pols * sizeof(*infos));
624 mutex_unlock(&nrs_core.nrs_mutex);
630 #define LPROCFS_NRS_WR_MAX_ARG (1024)
632 * The longest valid command string is the maxium policy name size, plus the
633 * length of the " reg" substring, plus the lenght of argument
635 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
636 + LPROCFS_NRS_WR_MAX_ARG)
639 * Starts and stops a given policy on a PTLRPC service.
641 * Commands consist of the policy name, followed by an optional [reg|hp] token;
642 * if the optional token is omitted, the operation is performed on both the
643 * regular and high-priority (if the service has one) NRS head.
646 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
647 size_t count, loff_t *off)
649 struct seq_file *m = file->private_data;
650 struct ptlrpc_service *svc = m->private;
651 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
653 char *cmd_copy = NULL;
659 if (count >= LPROCFS_NRS_WR_MAX_CMD)
660 GOTO(out, rc = -EINVAL);
662 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
664 GOTO(out, rc = -ENOMEM);
666 * strsep() modifies its argument, so keep a copy
670 if (copy_from_user(cmd, buffer, count))
671 GOTO(out, rc = -EFAULT);
675 policy_name = strsep(&cmd, " ");
677 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
678 GOTO(out, rc = -EINVAL);
681 * No [reg|hp] token has been specified
686 queue_name = strsep(&cmd, " ");
688 * The second token is either an optional [reg|hp] string,
691 if (strcmp(queue_name, "reg") == 0)
692 queue = PTLRPC_NRS_QUEUE_REG;
693 else if (strcmp(queue_name, "hp") == 0)
694 queue = PTLRPC_NRS_QUEUE_HP;
703 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
704 GOTO(out, rc = -ENODEV);
705 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
706 queue = PTLRPC_NRS_QUEUE_REG;
709 * Serialize NRS core lprocfs operations with policy registration/
712 mutex_lock(&nrs_core.nrs_mutex);
714 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
715 PTLRPC_NRS_CTL_START,
718 mutex_unlock(&nrs_core.nrs_mutex);
721 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
723 RETURN(rc < 0 ? rc : count);
725 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
729 struct ptlrpc_srh_iterator {
732 struct ptlrpc_request *srhi_req;
736 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
737 struct ptlrpc_srh_iterator *srhi,
741 struct ptlrpc_request *req;
743 if (srhi->srhi_req != NULL &&
744 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
745 srhi->srhi_seq <= seq) {
746 /* If srhi_req was set previously, hasn't been culled and
747 * we're searching for a seq on or after it (i.e. more
748 * recent), search from it onwards.
749 * Since the service history is LRU (i.e. culled reqs will
750 * be near the head), we shouldn't have to do long
752 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
753 "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
754 svcpt->scp_service->srv_name, svcpt->scp_cpt,
755 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
756 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
757 "%s:%d: seek offset "LPU64", request seq "LPU64", "
758 "last culled "LPU64"\n",
759 svcpt->scp_service->srv_name, svcpt->scp_cpt,
760 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
761 e = &srhi->srhi_req->rq_history_list;
763 /* search from start */
764 e = svcpt->scp_hist_reqs.next;
767 while (e != &svcpt->scp_hist_reqs) {
768 req = list_entry(e, struct ptlrpc_request, rq_history_list);
770 if (req->rq_history_seq >= seq) {
771 srhi->srhi_seq = req->rq_history_seq;
772 srhi->srhi_req = req;
782 * ptlrpc history sequence is used as "position" of seq_file, in some case,
783 * seq_read() will increase "position" to indicate reading the next
784 * element, however, low bits of history sequence are reserved for CPT id
785 * (check the details from comments before ptlrpc_req_add_history), which
786 * means seq_read() might change CPT id of history sequence and never
787 * finish reading of requests on a CPT. To make it work, we have to shift
788 * CPT id to high bits and timestamp to low bits, so seq_read() will only
789 * increase timestamp which can correctly indicate the next position.
792 /* convert seq_file pos to cpt */
793 #define PTLRPC_REQ_POS2CPT(svc, pos) \
794 ((svc)->srv_cpt_bits == 0 ? 0 : \
795 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
797 /* make up seq_file pos from cpt */
798 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
799 ((svc)->srv_cpt_bits == 0 ? 0 : \
800 (cpt) << (64 - (svc)->srv_cpt_bits))
802 /* convert sequence to position */
803 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
804 ((svc)->srv_cpt_bits == 0 ? (seq) : \
805 ((seq) >> (svc)->srv_cpt_bits) | \
806 ((seq) << (64 - (svc)->srv_cpt_bits)))
808 /* convert position to sequence */
809 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
810 ((svc)->srv_cpt_bits == 0 ? (pos) : \
811 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
812 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
815 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
817 struct ptlrpc_service *svc = s->private;
818 struct ptlrpc_service_part *svcpt;
819 struct ptlrpc_srh_iterator *srhi;
824 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
825 CWARN("Failed to read request history because size of loff_t "
826 "%d can't match size of u64\n", (int)sizeof(loff_t));
830 OBD_ALLOC(srhi, sizeof(*srhi));
835 srhi->srhi_req = NULL;
837 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
839 ptlrpc_service_for_each_part(svcpt, i, svc) {
840 if (i < cpt) /* skip */
842 if (i > cpt) /* make up the lowest position for this CPT */
843 *pos = PTLRPC_REQ_CPT2POS(svc, i);
845 spin_lock(&svcpt->scp_lock);
846 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
847 PTLRPC_REQ_POS2SEQ(svc, *pos));
848 spin_unlock(&svcpt->scp_lock);
850 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
856 OBD_FREE(srhi, sizeof(*srhi));
861 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
863 struct ptlrpc_srh_iterator *srhi = iter;
866 OBD_FREE(srhi, sizeof(*srhi));
870 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
871 void *iter, loff_t *pos)
873 struct ptlrpc_service *svc = s->private;
874 struct ptlrpc_srh_iterator *srhi = iter;
875 struct ptlrpc_service_part *svcpt;
880 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
881 svcpt = svc->srv_parts[i];
883 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
884 srhi->srhi_req = NULL;
885 seq = srhi->srhi_seq = 0;
886 } else { /* the next sequence */
887 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
890 spin_lock(&svcpt->scp_lock);
891 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
892 spin_unlock(&svcpt->scp_lock);
894 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
900 OBD_FREE(srhi, sizeof(*srhi));
904 /* common ost/mdt so_req_printer */
905 void target_print_req(void *seq_file, struct ptlrpc_request *req)
907 /* Called holding srv_lock with irqs disabled.
908 * Print specific req contents and a newline.
909 * CAVEAT EMPTOR: check request message length before printing!!!
910 * You might have received any old crap so you must be just as
911 * careful here as the service's request parser!!! */
912 struct seq_file *sf = seq_file;
914 switch (req->rq_phase) {
916 /* still awaiting a service thread's attention, or rejected
917 * because the generic request message didn't unpack */
918 seq_printf(sf, "<not swabbed>\n");
920 case RQ_PHASE_INTERPRET:
921 /* being handled, so basic msg swabbed, and opc is valid
922 * but racing with mds_handle() */
923 case RQ_PHASE_COMPLETE:
924 /* been handled by mds_handle() reply state possibly still
926 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
929 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
932 EXPORT_SYMBOL(target_print_req);
934 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
936 struct ptlrpc_service *svc = s->private;
937 struct ptlrpc_srh_iterator *srhi = iter;
938 struct ptlrpc_service_part *svcpt;
939 struct ptlrpc_request *req;
942 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
944 svcpt = svc->srv_parts[srhi->srhi_idx];
946 spin_lock(&svcpt->scp_lock);
948 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
951 req = srhi->srhi_req;
953 /* Print common req fields.
954 * CAVEAT EMPTOR: we're racing with the service handler
955 * here. The request could contain any old crap, so you
956 * must be just as careful as the service's request
957 * parser. Currently I only print stuff here I know is OK
958 * to look at coz it was set up in request_in_callback()!!! */
959 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
960 req->rq_history_seq, libcfs_nid2str(req->rq_self),
961 libcfs_id2str(req->rq_peer), req->rq_xid,
962 req->rq_reqlen, ptlrpc_rqphase2str(req),
963 req->rq_arrival_time.tv_sec,
964 req->rq_sent - req->rq_arrival_time.tv_sec,
965 req->rq_sent - req->rq_deadline);
966 if (svc->srv_ops.so_req_printer == NULL)
969 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
972 spin_unlock(&svcpt->scp_lock);
977 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
979 static struct seq_operations sops = {
980 .start = ptlrpc_lprocfs_svc_req_history_start,
981 .stop = ptlrpc_lprocfs_svc_req_history_stop,
982 .next = ptlrpc_lprocfs_svc_req_history_next,
983 .show = ptlrpc_lprocfs_svc_req_history_show,
985 struct seq_file *seqf;
988 LPROCFS_ENTRY_CHECK(PDE(inode));
989 rc = seq_open(file, &sops);
993 seqf = file->private_data;
994 seqf->private = PDE_DATA(inode);
998 /* See also lprocfs_rd_timeouts */
999 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1001 struct ptlrpc_service *svc = m->private;
1002 struct ptlrpc_service_part *svcpt;
1010 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1015 ptlrpc_service_for_each_part(svcpt, i, svc) {
1016 cur = at_get(&svcpt->scp_at_estimate);
1017 worst = svcpt->scp_at_estimate.at_worst_ever;
1018 worstt = svcpt->scp_at_estimate.at_worst_time;
1019 s2dhms(&ts, cfs_time_current_sec() - worstt);
1021 seq_printf(m, "%10s : cur %3u worst %3u (at %ld, "
1022 DHMS_FMT" ago) ", "service",
1023 cur, worst, worstt, DHMS_VARS(&ts));
1025 lprocfs_seq_at_hist_helper(m, &svcpt->scp_at_estimate);
1030 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1032 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1034 struct ptlrpc_service *svc = m->private;
1035 return seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1039 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1040 size_t count, loff_t *off)
1042 struct seq_file *m = file->private_data;
1043 struct ptlrpc_service *svc = m->private;
1047 rc = lprocfs_write_helper(buffer, count, &val);
1054 spin_lock(&svc->srv_lock);
1055 svc->srv_hpreq_ratio = val;
1056 spin_unlock(&svc->srv_lock);
1060 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1062 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1063 struct ptlrpc_service *svc)
1065 struct lprocfs_seq_vars lproc_vars[] = {
1066 { .name = "high_priority_ratio",
1067 .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1069 { .name = "req_buffer_history_len",
1070 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1072 { .name = "req_buffer_history_max",
1073 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1075 { .name = "threads_min",
1076 .fops = &ptlrpc_lprocfs_threads_min_fops,
1078 { .name = "threads_max",
1079 .fops = &ptlrpc_lprocfs_threads_max_fops,
1081 { .name = "threads_started",
1082 .fops = &ptlrpc_lprocfs_threads_started_fops,
1084 { .name = "timeouts",
1085 .fops = &ptlrpc_lprocfs_timeouts_fops,
1087 { .name = "nrs_policies",
1088 .fops = &ptlrpc_lprocfs_nrs_fops,
1092 static struct file_operations req_history_fops = {
1093 .owner = THIS_MODULE,
1094 .open = ptlrpc_lprocfs_svc_req_history_open,
1096 .llseek = seq_lseek,
1097 .release = lprocfs_seq_release,
1102 ptlrpc_lprocfs_register(entry, svc->srv_name,
1103 "stats", &svc->srv_procroot,
1105 if (svc->srv_procroot == NULL)
1108 lprocfs_seq_add_vars(svc->srv_procroot, lproc_vars, NULL);
1110 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1111 0400, &req_history_fops, svc);
1113 CWARN("Error adding the req_history file\n");
1116 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1118 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1119 &obddev->obd_svc_procroot,
1120 &obddev->obd_svc_stats);
1122 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1124 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1126 struct lprocfs_stats *svc_stats;
1127 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1128 int opc = opcode_offset(op);
1130 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1131 if (svc_stats == NULL || opc <= 0)
1133 LASSERT(opc < LUSTRE_MAX_OPCODES);
1134 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1135 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1138 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1140 struct lprocfs_stats *svc_stats;
1143 if (!req->rq_import)
1145 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1148 idx = lustre_msg_get_opc(req->rq_reqmsg);
1151 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1154 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1157 LASSERTF(0, "unsupported opcode %u\n", idx);
1161 lprocfs_counter_add(svc_stats, idx, bytes);
1164 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1166 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1168 if (svc->srv_procroot != NULL)
1169 lprocfs_remove(&svc->srv_procroot);
1172 lprocfs_free_stats(&svc->srv_stats);
1175 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1177 if (obd->obd_svc_procroot)
1178 lprocfs_remove(&obd->obd_svc_procroot);
1180 if (obd->obd_svc_stats)
1181 lprocfs_free_stats(&obd->obd_svc_stats);
1183 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1186 lprocfs_ping_seq_write(struct file *file, const char *buffer,
1187 size_t count, loff_t *off)
1189 struct seq_file *m = file->private_data;
1190 struct obd_device *obd = m->private;
1191 struct ptlrpc_request *req;
1195 LPROCFS_CLIMP_CHECK(obd);
1196 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1197 LPROCFS_CLIMP_EXIT(obd);
1201 req->rq_send_state = LUSTRE_IMP_FULL;
1203 rc = ptlrpc_queue_wait(req);
1205 ptlrpc_req_finished(req);
1210 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1212 /* Write the connection UUID to this file to attempt to connect to that node.
1213 * The connection UUID is a node's primary NID. For example,
1214 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1217 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1218 size_t count, loff_t *off)
1220 struct seq_file *m = file->private_data;
1221 struct obd_device *obd = m->private;
1222 struct obd_import *imp = obd->u.cli.cl_import;
1227 const char prefix[] = "connection=";
1228 const int prefix_len = sizeof(prefix) - 1;
1230 if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1233 OBD_ALLOC(kbuf, count + 1);
1237 if (copy_from_user(kbuf, buffer, count))
1238 GOTO(out, count = -EFAULT);
1242 /* only support connection=uuid::instance now */
1243 if (strncmp(prefix, kbuf, prefix_len) != 0)
1244 GOTO(out, count = -EINVAL);
1246 uuid = kbuf + prefix_len;
1247 ptr = strstr(uuid, "::");
1254 ptr += 2; /* Skip :: */
1255 inst = simple_strtol(ptr, &endptr, 10);
1257 CERROR("config: wrong instance # %s\n", ptr);
1258 } else if (inst != imp->imp_connect_data.ocd_instance) {
1259 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1260 "target(%u/%u), reconnecting...\n",
1261 imp->imp_obd->obd_name,
1262 imp->imp_connect_data.ocd_instance, inst);
1265 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1267 imp->imp_obd->obd_name, inst);
1272 ptlrpc_recover_import(imp, uuid, 1);
1275 OBD_FREE(kbuf, count + 1);
1278 EXPORT_SYMBOL(lprocfs_import_seq_write);
1280 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1282 struct obd_device *obd = m->private;
1283 struct obd_import *imp = obd->u.cli.cl_import;
1286 LPROCFS_CLIMP_CHECK(obd);
1287 rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1288 LPROCFS_CLIMP_EXIT(obd);
1291 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1294 lprocfs_pinger_recov_seq_write(struct file *file, const char *buffer,
1295 size_t count, loff_t *off)
1297 struct seq_file *m = file->private_data;
1298 struct obd_device *obd = m->private;
1299 struct client_obd *cli = &obd->u.cli;
1300 struct obd_import *imp = cli->cl_import;
1303 rc = lprocfs_write_helper(buffer, count, &val);
1307 if (val != 0 && val != 1)
1310 LPROCFS_CLIMP_CHECK(obd);
1311 spin_lock(&imp->imp_lock);
1312 imp->imp_no_pinger_recover = !val;
1313 spin_unlock(&imp->imp_lock);
1314 LPROCFS_CLIMP_EXIT(obd);
1317 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1319 #endif /* LPROCFS */