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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 #define DEBUG_SUBSYSTEM S_CLASS
35 #include <obd_support.h>
37 #include <lprocfs_status.h>
38 #include <lustre/lustre_idl.h>
39 #include <lustre_net.h>
40 #include <obd_class.h>
41 #include "ptlrpc_internal.h"
44 static struct ll_rpc_opcode {
47 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
48 { OST_REPLY, "ost_reply" },
49 { OST_GETATTR, "ost_getattr" },
50 { OST_SETATTR, "ost_setattr" },
51 { OST_READ, "ost_read" },
52 { OST_WRITE, "ost_write" },
53 { OST_CREATE , "ost_create" },
54 { OST_DESTROY, "ost_destroy" },
55 { OST_GET_INFO, "ost_get_info" },
56 { OST_CONNECT, "ost_connect" },
57 { OST_DISCONNECT, "ost_disconnect" },
58 { OST_PUNCH, "ost_punch" },
59 { OST_OPEN, "ost_open" },
60 { OST_CLOSE, "ost_close" },
61 { OST_STATFS, "ost_statfs" },
62 { 14, NULL }, /* formerly OST_SAN_READ */
63 { 15, NULL }, /* formerly OST_SAN_WRITE */
64 { OST_SYNC, "ost_sync" },
65 { OST_SET_INFO, "ost_set_info" },
66 { OST_QUOTACHECK, "ost_quotacheck" },
67 { OST_QUOTACTL, "ost_quotactl" },
68 { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
69 { OST_LADVISE, "ost_ladvise" },
70 { MDS_GETATTR, "mds_getattr" },
71 { MDS_GETATTR_NAME, "mds_getattr_lock" },
72 { MDS_CLOSE, "mds_close" },
73 { MDS_REINT, "mds_reint" },
74 { MDS_READPAGE, "mds_readpage" },
75 { MDS_CONNECT, "mds_connect" },
76 { MDS_DISCONNECT, "mds_disconnect" },
77 { MDS_GET_ROOT, "mds_get_root" },
78 { MDS_STATFS, "mds_statfs" },
79 { MDS_PIN, "mds_pin" },
80 { MDS_UNPIN, "mds_unpin" },
81 { MDS_SYNC, "mds_sync" },
82 { MDS_DONE_WRITING, "mds_done_writing" },
83 { MDS_SET_INFO, "mds_set_info" },
84 { MDS_QUOTACHECK, "mds_quotacheck" },
85 { MDS_QUOTACTL, "mds_quotactl" },
86 { MDS_GETXATTR, "mds_getxattr" },
87 { MDS_SETXATTR, "mds_setxattr" },
88 { MDS_WRITEPAGE, "mds_writepage" },
89 { MDS_IS_SUBDIR, "mds_is_subdir" },
90 { MDS_GET_INFO, "mds_get_info" },
91 { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
92 { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
93 { MDS_HSM_ACTION, "mds_hsm_action" },
94 { MDS_HSM_PROGRESS, "mds_hsm_progress" },
95 { MDS_HSM_REQUEST, "mds_hsm_request" },
96 { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
97 { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
98 { MDS_SWAP_LAYOUTS, "mds_swap_layouts" },
99 { LDLM_ENQUEUE, "ldlm_enqueue" },
100 { LDLM_CONVERT, "ldlm_convert" },
101 { LDLM_CANCEL, "ldlm_cancel" },
102 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
103 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
104 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
105 { LDLM_SET_INFO, "ldlm_set_info" },
106 { MGS_CONNECT, "mgs_connect" },
107 { MGS_DISCONNECT, "mgs_disconnect" },
108 { MGS_EXCEPTION, "mgs_exception" },
109 { MGS_TARGET_REG, "mgs_target_reg" },
110 { MGS_TARGET_DEL, "mgs_target_del" },
111 { MGS_SET_INFO, "mgs_set_info" },
112 { MGS_CONFIG_READ, "mgs_config_read" },
113 { OBD_PING, "obd_ping" },
114 { OBD_LOG_CANCEL, "llog_cancel" },
115 { OBD_QC_CALLBACK, "obd_quota_callback" },
116 { OBD_IDX_READ, "dt_index_read" },
117 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
118 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
119 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
120 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
121 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
122 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
123 { LLOG_CATINFO, "llog_catinfo" },
124 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
125 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
126 { QUOTA_DQACQ, "quota_acquire" },
127 { QUOTA_DQREL, "quota_release" },
128 { SEQ_QUERY, "seq_query" },
129 { SEC_CTX_INIT, "sec_ctx_init" },
130 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
131 { SEC_CTX_FINI, "sec_ctx_fini" },
132 { FLD_QUERY, "fld_query" },
133 { FLD_READ, "fld_read" },
134 { OUT_UPDATE, "out_update" },
135 { LFSCK_NOTIFY, "lfsck_notify" },
136 { LFSCK_QUERY, "lfsck_query" },
139 static struct ll_eopcode {
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" },
159 const char *ll_opcode2str(__u32 opcode)
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.
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;
178 static const char *ll_eopcode2str(__u32 opcode)
180 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
181 return ll_eopcode_table[opcode].opname;
184 #ifdef CONFIG_PROC_FS
185 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
186 char *name, struct proc_dir_entry **procroot_ret,
187 struct lprocfs_stats **stats_ret)
189 struct proc_dir_entry *svc_procroot;
190 struct lprocfs_stats *svc_stats;
192 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
195 LASSERT(*procroot_ret == NULL);
196 LASSERT(*stats_ret == NULL);
198 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
199 if (svc_stats == NULL)
203 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
204 if (IS_ERR(svc_procroot)) {
205 lprocfs_free_stats(&svc_stats);
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++) {
226 case BRW_WRITE_BYTES:
234 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
236 ll_eopcode2str(i), units);
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");
245 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
248 lprocfs_remove(&svc_procroot);
249 lprocfs_free_stats(&svc_stats);
252 *procroot_ret = svc_procroot;
253 *stats_ret = svc_stats;
258 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
260 struct ptlrpc_service *svc = m->private;
261 struct ptlrpc_service_part *svcpt;
265 ptlrpc_service_for_each_part(svcpt, i, svc)
266 total += svcpt->scp_hist_nrqbds;
268 seq_printf(m, "%d\n", total);
271 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
274 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
276 struct ptlrpc_service *svc = m->private;
277 struct ptlrpc_service_part *svcpt;
281 ptlrpc_service_for_each_part(svcpt, i, svc)
282 total += svc->srv_hist_nrqbds_cpt_max;
284 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_str_to_s64(buffer, count, &val);
303 if (val < 0 || val > INT_MAX)
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_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 =
320 max(1, ((int)val / svc->srv_ncpts));
322 spin_unlock(&svc->srv_lock);
326 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
329 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
331 struct ptlrpc_service *svc = m->private;
333 seq_printf(m, "%d\n",
334 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
339 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
340 const char __user *buffer,
341 size_t count, loff_t *off)
343 struct seq_file *m = file->private_data;
344 struct ptlrpc_service *svc = m->private;
346 int rc = lprocfs_str_to_s64(buffer, count, &val);
351 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
354 spin_lock(&svc->srv_lock);
355 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
356 spin_unlock(&svc->srv_lock);
360 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
362 spin_unlock(&svc->srv_lock);
366 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
369 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
371 struct ptlrpc_service *svc = m->private;
372 struct ptlrpc_service_part *svcpt;
376 ptlrpc_service_for_each_part(svcpt, i, svc)
377 total += svcpt->scp_nthrs_running;
379 seq_printf(m, "%d\n", total);
382 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
385 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
387 struct ptlrpc_service *svc = m->private;
389 seq_printf(m, "%d\n",
390 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
395 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
396 const char __user *buffer,
397 size_t count, loff_t *off)
399 struct seq_file *m = file->private_data;
400 struct ptlrpc_service *svc = m->private;
402 int rc = lprocfs_str_to_s64(buffer, count, &val);
407 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
410 spin_lock(&svc->srv_lock);
411 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
412 spin_unlock(&svc->srv_lock);
416 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
418 spin_unlock(&svc->srv_lock);
422 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
425 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
427 * \param[in] state The policy state
429 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
434 case NRS_POL_STATE_INVALID:
436 case NRS_POL_STATE_STOPPED:
438 case NRS_POL_STATE_STOPPING:
440 case NRS_POL_STATE_STARTING:
442 case NRS_POL_STATE_STARTED:
448 * Obtains status information for \a policy.
450 * Information is copied in \a info.
452 * \param[in] policy The policy
453 * \param[out] info Holds returned status information
455 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
456 struct ptlrpc_nrs_pol_info *info)
458 LASSERT(policy != NULL);
459 LASSERT(info != NULL);
460 assert_spin_locked(&policy->pol_nrs->nrs_lock);
462 LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
463 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
464 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
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_arg, tmp.pi_arg,
537 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
538 sizeof(tmp.pi_state));
539 infos[pol_idx].pi_fallback = tmp.pi_fallback;
541 * For the rest of the service partitions
542 * sanity-check the values we get.
545 LASSERT(strncmp(infos[pol_idx].pi_name,
547 NRS_POL_NAME_MAX) == 0);
548 LASSERT(strncmp(infos[pol_idx].pi_arg,
550 sizeof(tmp.pi_arg)) == 0);
552 * Not asserting ptlrpc_nrs_pol_info::pi_state,
553 * because it may be different between
554 * instances of the same policy in different
555 * service partitions.
557 LASSERT(infos[pol_idx].pi_fallback ==
561 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
562 infos[pol_idx].pi_req_started += tmp.pi_req_started;
566 spin_unlock(&nrs->nrs_lock);
570 * Policy status information output is in YAML format.
586 * high_priority_requests:
599 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
600 "high_priority_requests:");
602 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
603 if (strlen(infos[pol_idx].pi_arg) > 0)
604 seq_printf(m, " - name: %s %s\n",
605 infos[pol_idx].pi_name,
606 infos[pol_idx].pi_arg);
608 seq_printf(m, " - name: %s\n",
609 infos[pol_idx].pi_name);
612 seq_printf(m, " state: %s\n"
615 " active: %-20d\n\n",
616 nrs_state2str(infos[pol_idx].pi_state),
617 infos[pol_idx].pi_fallback ? "yes" : "no",
618 (int)infos[pol_idx].pi_req_queued,
619 (int)infos[pol_idx].pi_req_started);
622 if (!hp && nrs_svc_has_hp(svc)) {
623 memset(infos, 0, num_pols * sizeof(*infos));
626 * Redo the processing for the service's HP NRS heads' policies.
634 OBD_FREE(infos, num_pols * sizeof(*infos));
636 mutex_unlock(&nrs_core.nrs_mutex);
642 #define LPROCFS_NRS_WR_MAX_ARG (1024)
644 * The longest valid command string is the maxium policy name size, plus the
645 * length of the " reg" substring, plus the lenght of argument
647 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
648 + LPROCFS_NRS_WR_MAX_ARG)
651 * Starts and stops a given policy on a PTLRPC service.
653 * Commands consist of the policy name, followed by an optional [reg|hp] token;
654 * if the optional token is omitted, the operation is performed on both the
655 * regular and high-priority (if the service has one) NRS head.
658 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
659 size_t count, loff_t *off)
661 struct seq_file *m = file->private_data;
662 struct ptlrpc_service *svc = m->private;
663 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
665 char *cmd_copy = NULL;
671 if (count >= LPROCFS_NRS_WR_MAX_CMD)
672 GOTO(out, rc = -EINVAL);
674 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
676 GOTO(out, rc = -ENOMEM);
678 * strsep() modifies its argument, so keep a copy
682 if (copy_from_user(cmd, buffer, count))
683 GOTO(out, rc = -EFAULT);
687 policy_name = strsep(&cmd, " ");
689 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
690 GOTO(out, rc = -EINVAL);
693 * No [reg|hp] token has been specified
698 queue_name = strsep(&cmd, " ");
700 * The second token is either an optional [reg|hp] string,
703 if (strcmp(queue_name, "reg") == 0)
704 queue = PTLRPC_NRS_QUEUE_REG;
705 else if (strcmp(queue_name, "hp") == 0)
706 queue = PTLRPC_NRS_QUEUE_HP;
715 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
716 GOTO(out, rc = -ENODEV);
717 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
718 queue = PTLRPC_NRS_QUEUE_REG;
721 * Serialize NRS core lprocfs operations with policy registration/
724 mutex_lock(&nrs_core.nrs_mutex);
726 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
727 PTLRPC_NRS_CTL_START,
730 mutex_unlock(&nrs_core.nrs_mutex);
733 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
735 RETURN(rc < 0 ? rc : count);
737 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
741 struct ptlrpc_srh_iterator {
744 struct ptlrpc_request *srhi_req;
748 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
749 struct ptlrpc_srh_iterator *srhi,
753 struct ptlrpc_request *req;
755 if (srhi->srhi_req != NULL &&
756 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
757 srhi->srhi_seq <= seq) {
758 /* If srhi_req was set previously, hasn't been culled and
759 * we're searching for a seq on or after it (i.e. more
760 * recent), search from it onwards.
761 * Since the service history is LRU (i.e. culled reqs will
762 * be near the head), we shouldn't have to do long
764 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
765 "%s:%d: seek seq %llu, request seq %llu\n",
766 svcpt->scp_service->srv_name, svcpt->scp_cpt,
767 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
768 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
769 "%s:%d: seek offset %llu, request seq %llu, "
770 "last culled %llu\n",
771 svcpt->scp_service->srv_name, svcpt->scp_cpt,
772 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
773 e = &srhi->srhi_req->rq_history_list;
775 /* search from start */
776 e = svcpt->scp_hist_reqs.next;
779 while (e != &svcpt->scp_hist_reqs) {
780 req = list_entry(e, struct ptlrpc_request, rq_history_list);
782 if (req->rq_history_seq >= seq) {
783 srhi->srhi_seq = req->rq_history_seq;
784 srhi->srhi_req = req;
794 * ptlrpc history sequence is used as "position" of seq_file, in some case,
795 * seq_read() will increase "position" to indicate reading the next
796 * element, however, low bits of history sequence are reserved for CPT id
797 * (check the details from comments before ptlrpc_req_add_history), which
798 * means seq_read() might change CPT id of history sequence and never
799 * finish reading of requests on a CPT. To make it work, we have to shift
800 * CPT id to high bits and timestamp to low bits, so seq_read() will only
801 * increase timestamp which can correctly indicate the next position.
804 /* convert seq_file pos to cpt */
805 #define PTLRPC_REQ_POS2CPT(svc, pos) \
806 ((svc)->srv_cpt_bits == 0 ? 0 : \
807 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
809 /* make up seq_file pos from cpt */
810 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
811 ((svc)->srv_cpt_bits == 0 ? 0 : \
812 (cpt) << (64 - (svc)->srv_cpt_bits))
814 /* convert sequence to position */
815 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
816 ((svc)->srv_cpt_bits == 0 ? (seq) : \
817 ((seq) >> (svc)->srv_cpt_bits) | \
818 ((seq) << (64 - (svc)->srv_cpt_bits)))
820 /* convert position to sequence */
821 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
822 ((svc)->srv_cpt_bits == 0 ? (pos) : \
823 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
824 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
827 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
829 struct ptlrpc_service *svc = s->private;
830 struct ptlrpc_service_part *svcpt;
831 struct ptlrpc_srh_iterator *srhi;
836 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
837 CWARN("Failed to read request history because size of loff_t "
838 "%d can't match size of u64\n", (int)sizeof(loff_t));
842 OBD_ALLOC(srhi, sizeof(*srhi));
847 srhi->srhi_req = NULL;
849 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
851 ptlrpc_service_for_each_part(svcpt, i, svc) {
852 if (i < cpt) /* skip */
854 if (i > cpt) /* make up the lowest position for this CPT */
855 *pos = PTLRPC_REQ_CPT2POS(svc, i);
857 spin_lock(&svcpt->scp_lock);
858 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
859 PTLRPC_REQ_POS2SEQ(svc, *pos));
860 spin_unlock(&svcpt->scp_lock);
862 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
868 OBD_FREE(srhi, sizeof(*srhi));
873 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
875 struct ptlrpc_srh_iterator *srhi = iter;
878 OBD_FREE(srhi, sizeof(*srhi));
882 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
883 void *iter, loff_t *pos)
885 struct ptlrpc_service *svc = s->private;
886 struct ptlrpc_srh_iterator *srhi = iter;
887 struct ptlrpc_service_part *svcpt;
892 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
893 svcpt = svc->srv_parts[i];
895 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
896 srhi->srhi_req = NULL;
897 seq = srhi->srhi_seq = 0;
898 } else { /* the next sequence */
899 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
902 spin_lock(&svcpt->scp_lock);
903 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
904 spin_unlock(&svcpt->scp_lock);
906 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
912 OBD_FREE(srhi, sizeof(*srhi));
916 /* common ost/mdt so_req_printer */
917 void target_print_req(void *seq_file, struct ptlrpc_request *req)
919 /* Called holding srv_lock with irqs disabled.
920 * Print specific req contents and a newline.
921 * CAVEAT EMPTOR: check request message length before printing!!!
922 * You might have received any old crap so you must be just as
923 * careful here as the service's request parser!!! */
924 struct seq_file *sf = seq_file;
926 switch (req->rq_phase) {
928 /* still awaiting a service thread's attention, or rejected
929 * because the generic request message didn't unpack */
930 seq_printf(sf, "<not swabbed>\n");
932 case RQ_PHASE_INTERPRET:
933 /* being handled, so basic msg swabbed, and opc is valid
934 * but racing with mds_handle() */
935 case RQ_PHASE_COMPLETE:
936 /* been handled by mds_handle() reply state possibly still
938 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
941 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
944 EXPORT_SYMBOL(target_print_req);
946 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
948 struct ptlrpc_service *svc = s->private;
949 struct ptlrpc_srh_iterator *srhi = iter;
950 struct ptlrpc_service_part *svcpt;
951 struct ptlrpc_request *req;
954 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
956 svcpt = svc->srv_parts[srhi->srhi_idx];
958 spin_lock(&svcpt->scp_lock);
960 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
963 struct timespec arrival, sent, arrivaldiff;
964 char nidstr[LNET_NIDSTR_SIZE];
966 req = srhi->srhi_req;
968 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
969 arrival.tv_sec = req->rq_arrival_time.tv_sec;
971 req->rq_arrival_time.tv_usec * NSEC_PER_USEC;
972 sent.tv_sec = req->rq_sent;
974 arrivaldiff = timespec_sub(sent, arrival);
976 /* Print common req fields.
977 * CAVEAT EMPTOR: we're racing with the service handler
978 * here. The request could contain any old crap, so you
979 * must be just as careful as the service's request
980 * parser. Currently I only print stuff here I know is OK
981 * to look at coz it was set up in request_in_callback()!!! */
982 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:"
983 "%ld.%06ld:%ld.%06lds(%+ld.0s) ",
984 req->rq_history_seq, nidstr,
985 libcfs_id2str(req->rq_peer), req->rq_xid,
986 req->rq_reqlen, ptlrpc_rqphase2str(req),
987 req->rq_arrival_time.tv_sec,
988 req->rq_arrival_time.tv_usec,
990 arrivaldiff.tv_nsec / NSEC_PER_USEC,
991 req->rq_sent - req->rq_deadline);
992 if (svc->srv_ops.so_req_printer == NULL)
995 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
998 spin_unlock(&svcpt->scp_lock);
1003 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1005 static struct seq_operations sops = {
1006 .start = ptlrpc_lprocfs_svc_req_history_start,
1007 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1008 .next = ptlrpc_lprocfs_svc_req_history_next,
1009 .show = ptlrpc_lprocfs_svc_req_history_show,
1011 struct seq_file *seqf;
1014 rc = LPROCFS_ENTRY_CHECK(inode);
1018 rc = seq_open(file, &sops);
1022 seqf = file->private_data;
1023 seqf->private = PDE_DATA(inode);
1027 /* See also lprocfs_rd_timeouts */
1028 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1030 struct ptlrpc_service *svc = m->private;
1031 struct ptlrpc_service_part *svcpt;
1039 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1044 ptlrpc_service_for_each_part(svcpt, i, svc) {
1045 cur = at_get(&svcpt->scp_at_estimate);
1046 worst = svcpt->scp_at_estimate.at_worst_ever;
1047 worstt = svcpt->scp_at_estimate.at_worst_time;
1048 s2dhms(&ts, cfs_time_current_sec() - worstt);
1050 seq_printf(m, "%10s : cur %3u worst %3u (at %ld, "
1051 DHMS_FMT" ago) ", "service",
1052 cur, worst, worstt, DHMS_VARS(&ts));
1054 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1059 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1061 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1063 struct ptlrpc_service *svc = m->private;
1064 seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1069 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1070 size_t count, loff_t *off)
1072 struct seq_file *m = file->private_data;
1073 struct ptlrpc_service *svc = m->private;
1077 rc = lprocfs_str_to_s64(buffer, count, &val);
1081 if (val < 0 || val > INT_MAX)
1084 spin_lock(&svc->srv_lock);
1085 svc->srv_hpreq_ratio = val;
1086 spin_unlock(&svc->srv_lock);
1090 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1092 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1093 struct ptlrpc_service *svc)
1095 struct lprocfs_vars lproc_vars[] = {
1096 { .name = "high_priority_ratio",
1097 .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1099 { .name = "req_buffer_history_len",
1100 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1102 { .name = "req_buffer_history_max",
1103 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1105 { .name = "threads_min",
1106 .fops = &ptlrpc_lprocfs_threads_min_fops,
1108 { .name = "threads_max",
1109 .fops = &ptlrpc_lprocfs_threads_max_fops,
1111 { .name = "threads_started",
1112 .fops = &ptlrpc_lprocfs_threads_started_fops,
1114 { .name = "timeouts",
1115 .fops = &ptlrpc_lprocfs_timeouts_fops,
1117 { .name = "nrs_policies",
1118 .fops = &ptlrpc_lprocfs_nrs_fops,
1122 static struct file_operations req_history_fops = {
1123 .owner = THIS_MODULE,
1124 .open = ptlrpc_lprocfs_svc_req_history_open,
1126 .llseek = seq_lseek,
1127 .release = lprocfs_seq_release,
1132 ptlrpc_lprocfs_register(entry, svc->srv_name,
1133 "stats", &svc->srv_procroot,
1135 if (svc->srv_procroot == NULL)
1138 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1140 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1141 0400, &req_history_fops, svc);
1143 CWARN("Error adding the req_history file\n");
1146 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1148 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1149 &obddev->obd_svc_procroot,
1150 &obddev->obd_svc_stats);
1152 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1154 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1156 struct lprocfs_stats *svc_stats;
1157 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1158 int opc = opcode_offset(op);
1160 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1161 if (svc_stats == NULL || opc <= 0)
1163 LASSERT(opc < LUSTRE_MAX_OPCODES);
1164 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1165 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1168 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1170 struct lprocfs_stats *svc_stats;
1173 if (!req->rq_import)
1175 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1178 idx = lustre_msg_get_opc(req->rq_reqmsg);
1181 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1184 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1187 LASSERTF(0, "unsupported opcode %u\n", idx);
1191 lprocfs_counter_add(svc_stats, idx, bytes);
1194 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1196 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1198 if (svc->srv_procroot != NULL)
1199 lprocfs_remove(&svc->srv_procroot);
1202 lprocfs_free_stats(&svc->srv_stats);
1205 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1207 if (obd->obd_svc_procroot)
1208 lprocfs_remove(&obd->obd_svc_procroot);
1210 if (obd->obd_svc_stats)
1211 lprocfs_free_stats(&obd->obd_svc_stats);
1213 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1216 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1217 size_t count, loff_t *off)
1219 struct seq_file *m = file->private_data;
1220 struct obd_device *obd = m->private;
1221 struct ptlrpc_request *req;
1225 LPROCFS_CLIMP_CHECK(obd);
1226 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1227 LPROCFS_CLIMP_EXIT(obd);
1231 req->rq_send_state = LUSTRE_IMP_FULL;
1233 rc = ptlrpc_queue_wait(req);
1235 ptlrpc_req_finished(req);
1240 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1242 /* Write the connection UUID to this file to attempt to connect to that node.
1243 * The connection UUID is a node's primary NID. For example,
1244 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1247 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1248 size_t count, loff_t *off)
1250 struct seq_file *m = file->private_data;
1251 struct obd_device *obd = m->private;
1252 struct obd_import *imp = obd->u.cli.cl_import;
1257 const char prefix[] = "connection=";
1258 const int prefix_len = sizeof(prefix) - 1;
1260 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1263 OBD_ALLOC(kbuf, count + 1);
1267 if (copy_from_user(kbuf, buffer, count))
1268 GOTO(out, count = -EFAULT);
1272 /* only support connection=uuid::instance now */
1273 if (strncmp(prefix, kbuf, prefix_len) != 0)
1274 GOTO(out, count = -EINVAL);
1276 uuid = kbuf + prefix_len;
1277 ptr = strstr(uuid, "::");
1284 ptr += 2; /* Skip :: */
1285 inst = simple_strtol(ptr, &endptr, 10);
1287 CERROR("config: wrong instance # %s\n", ptr);
1288 } else if (inst != imp->imp_connect_data.ocd_instance) {
1289 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1290 "target(%u/%u), reconnecting...\n",
1291 imp->imp_obd->obd_name,
1292 imp->imp_connect_data.ocd_instance, inst);
1295 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1297 imp->imp_obd->obd_name, inst);
1302 ptlrpc_recover_import(imp, uuid, 1);
1305 OBD_FREE(kbuf, count + 1);
1308 EXPORT_SYMBOL(lprocfs_import_seq_write);
1310 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1312 struct obd_device *obd = m->private;
1313 struct obd_import *imp = obd->u.cli.cl_import;
1315 LPROCFS_CLIMP_CHECK(obd);
1316 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1317 LPROCFS_CLIMP_EXIT(obd);
1320 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1323 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1324 size_t count, loff_t *off)
1326 struct seq_file *m = file->private_data;
1327 struct obd_device *obd = m->private;
1328 struct client_obd *cli = &obd->u.cli;
1329 struct obd_import *imp = cli->cl_import;
1333 rc = lprocfs_str_to_s64(buffer, count, &val);
1337 if (val != 0 && val != 1)
1340 LPROCFS_CLIMP_CHECK(obd);
1341 spin_lock(&imp->imp_lock);
1342 imp->imp_no_pinger_recover = !val;
1343 spin_unlock(&imp->imp_lock);
1344 LPROCFS_CLIMP_EXIT(obd);
1347 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1349 #endif /* CONFIG_PROC_FS */