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 const int ll_str2opcode(const char *ops)
182 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
183 if (ll_rpc_opcode_table[i].opname != NULL &&
184 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
185 return ll_rpc_opcode_table[i].opcode;
191 static const char *ll_eopcode2str(__u32 opcode)
193 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
194 return ll_eopcode_table[opcode].opname;
197 #ifdef CONFIG_PROC_FS
198 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
199 char *name, struct proc_dir_entry **procroot_ret,
200 struct lprocfs_stats **stats_ret)
202 struct proc_dir_entry *svc_procroot;
203 struct lprocfs_stats *svc_stats;
205 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
208 LASSERT(*procroot_ret == NULL);
209 LASSERT(*stats_ret == NULL);
211 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
212 if (svc_stats == NULL)
216 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
217 if (IS_ERR(svc_procroot)) {
218 lprocfs_free_stats(&svc_stats);
225 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
226 svc_counter_config, "req_waittime", "usec");
227 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
228 svc_counter_config, "req_qdepth", "reqs");
229 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
230 svc_counter_config, "req_active", "reqs");
231 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
232 svc_counter_config, "req_timeout", "sec");
233 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
234 svc_counter_config, "reqbuf_avail", "bufs");
235 for (i = 0; i < EXTRA_LAST_OPC; i++) {
239 case BRW_WRITE_BYTES:
247 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
249 ll_eopcode2str(i), units);
251 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
252 __u32 opcode = ll_rpc_opcode_table[i].opcode;
253 lprocfs_counter_init(svc_stats,
254 EXTRA_MAX_OPCODES + i, svc_counter_config,
255 ll_opcode2str(opcode), "usec");
258 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
261 lprocfs_remove(&svc_procroot);
262 lprocfs_free_stats(&svc_stats);
265 *procroot_ret = svc_procroot;
266 *stats_ret = svc_stats;
271 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
273 struct ptlrpc_service *svc = m->private;
274 struct ptlrpc_service_part *svcpt;
278 ptlrpc_service_for_each_part(svcpt, i, svc)
279 total += svcpt->scp_hist_nrqbds;
281 seq_printf(m, "%d\n", total);
284 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
287 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
289 struct ptlrpc_service *svc = m->private;
290 struct ptlrpc_service_part *svcpt;
294 ptlrpc_service_for_each_part(svcpt, i, svc)
295 total += svc->srv_hist_nrqbds_cpt_max;
297 seq_printf(m, "%d\n", total);
302 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
303 const char __user *buffer,
304 size_t count, loff_t *off)
306 struct seq_file *m = file->private_data;
307 struct ptlrpc_service *svc = m->private;
312 rc = lprocfs_str_to_s64(buffer, count, &val);
316 if (val < 0 || val > INT_MAX)
319 /* This sanity check is more of an insanity check; we can still
320 * hose a kernel by allowing the request history to grow too
322 bufpages = (svc->srv_buf_size + PAGE_SIZE - 1) >>
324 if (val > totalram_pages/(2 * bufpages))
327 spin_lock(&svc->srv_lock);
330 svc->srv_hist_nrqbds_cpt_max = 0;
332 svc->srv_hist_nrqbds_cpt_max =
333 max(1, ((int)val / svc->srv_ncpts));
335 spin_unlock(&svc->srv_lock);
339 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
342 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
344 struct ptlrpc_service *svc = m->private;
346 seq_printf(m, "%d\n",
347 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
352 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
353 const char __user *buffer,
354 size_t count, loff_t *off)
356 struct seq_file *m = file->private_data;
357 struct ptlrpc_service *svc = m->private;
359 int rc = lprocfs_str_to_s64(buffer, count, &val);
364 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
367 spin_lock(&svc->srv_lock);
368 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
369 spin_unlock(&svc->srv_lock);
373 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
375 spin_unlock(&svc->srv_lock);
379 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
382 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
384 struct ptlrpc_service *svc = m->private;
385 struct ptlrpc_service_part *svcpt;
389 ptlrpc_service_for_each_part(svcpt, i, svc)
390 total += svcpt->scp_nthrs_running;
392 seq_printf(m, "%d\n", total);
395 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
398 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
400 struct ptlrpc_service *svc = m->private;
402 seq_printf(m, "%d\n",
403 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
408 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
409 const char __user *buffer,
410 size_t count, loff_t *off)
412 struct seq_file *m = file->private_data;
413 struct ptlrpc_service *svc = m->private;
415 int rc = lprocfs_str_to_s64(buffer, count, &val);
420 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
423 spin_lock(&svc->srv_lock);
424 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
425 spin_unlock(&svc->srv_lock);
429 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
431 spin_unlock(&svc->srv_lock);
435 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
438 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
440 * \param[in] state The policy state
442 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
447 case NRS_POL_STATE_INVALID:
449 case NRS_POL_STATE_STOPPED:
451 case NRS_POL_STATE_STOPPING:
453 case NRS_POL_STATE_STARTING:
455 case NRS_POL_STATE_STARTED:
461 * Obtains status information for \a policy.
463 * Information is copied in \a info.
465 * \param[in] policy The policy
466 * \param[out] info Holds returned status information
468 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
469 struct ptlrpc_nrs_pol_info *info)
471 LASSERT(policy != NULL);
472 LASSERT(info != NULL);
473 assert_spin_locked(&policy->pol_nrs->nrs_lock);
475 LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
476 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
477 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
479 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
480 info->pi_state = policy->pol_state;
482 * XXX: These are accessed without holding
483 * ptlrpc_service_part::scp_req_lock.
485 info->pi_req_queued = policy->pol_req_queued;
486 info->pi_req_started = policy->pol_req_started;
490 * Reads and prints policy status information for all policies of a PTLRPC
493 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
495 struct ptlrpc_service *svc = m->private;
496 struct ptlrpc_service_part *svcpt;
497 struct ptlrpc_nrs *nrs;
498 struct ptlrpc_nrs_policy *policy;
499 struct ptlrpc_nrs_pol_info *infos;
500 struct ptlrpc_nrs_pol_info tmp;
502 unsigned pol_idx = 0;
509 * Serialize NRS core lprocfs operations with policy registration/
512 mutex_lock(&nrs_core.nrs_mutex);
515 * Use the first service partition's regular NRS head in order to obtain
516 * the number of policies registered with NRS heads of this service. All
517 * service partitions will have the same number of policies.
519 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
521 spin_lock(&nrs->nrs_lock);
522 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
523 spin_unlock(&nrs->nrs_lock);
525 OBD_ALLOC(infos, num_pols * sizeof(*infos));
527 GOTO(out, rc = -ENOMEM);
530 ptlrpc_service_for_each_part(svcpt, i, svc) {
531 nrs = nrs_svcpt2nrs(svcpt, hp);
532 spin_lock(&nrs->nrs_lock);
536 list_for_each_entry(policy, &nrs->nrs_policy_list,
538 LASSERT(pol_idx < num_pols);
540 nrs_policy_get_info_locked(policy, &tmp);
542 * Copy values when handling the first service
546 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
548 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
550 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
551 sizeof(tmp.pi_state));
552 infos[pol_idx].pi_fallback = tmp.pi_fallback;
554 * For the rest of the service partitions
555 * sanity-check the values we get.
558 LASSERT(strncmp(infos[pol_idx].pi_name,
560 NRS_POL_NAME_MAX) == 0);
561 LASSERT(strncmp(infos[pol_idx].pi_arg,
563 sizeof(tmp.pi_arg)) == 0);
565 * Not asserting ptlrpc_nrs_pol_info::pi_state,
566 * because it may be different between
567 * instances of the same policy in different
568 * service partitions.
570 LASSERT(infos[pol_idx].pi_fallback ==
574 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
575 infos[pol_idx].pi_req_started += tmp.pi_req_started;
579 spin_unlock(&nrs->nrs_lock);
583 * Policy status information output is in YAML format.
599 * high_priority_requests:
612 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
613 "high_priority_requests:");
615 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
616 if (strlen(infos[pol_idx].pi_arg) > 0)
617 seq_printf(m, " - name: %s %s\n",
618 infos[pol_idx].pi_name,
619 infos[pol_idx].pi_arg);
621 seq_printf(m, " - name: %s\n",
622 infos[pol_idx].pi_name);
625 seq_printf(m, " state: %s\n"
628 " active: %-20d\n\n",
629 nrs_state2str(infos[pol_idx].pi_state),
630 infos[pol_idx].pi_fallback ? "yes" : "no",
631 (int)infos[pol_idx].pi_req_queued,
632 (int)infos[pol_idx].pi_req_started);
635 if (!hp && nrs_svc_has_hp(svc)) {
636 memset(infos, 0, num_pols * sizeof(*infos));
639 * Redo the processing for the service's HP NRS heads' policies.
647 OBD_FREE(infos, num_pols * sizeof(*infos));
649 mutex_unlock(&nrs_core.nrs_mutex);
655 #define LPROCFS_NRS_WR_MAX_ARG (1024)
657 * The longest valid command string is the maxium policy name size, plus the
658 * length of the " reg" substring, plus the lenght of argument
660 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
661 + LPROCFS_NRS_WR_MAX_ARG)
664 * Starts and stops a given policy on a PTLRPC service.
666 * Commands consist of the policy name, followed by an optional [reg|hp] token;
667 * if the optional token is omitted, the operation is performed on both the
668 * regular and high-priority (if the service has one) NRS head.
671 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
672 size_t count, loff_t *off)
674 struct seq_file *m = file->private_data;
675 struct ptlrpc_service *svc = m->private;
676 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
678 char *cmd_copy = NULL;
684 if (count >= LPROCFS_NRS_WR_MAX_CMD)
685 GOTO(out, rc = -EINVAL);
687 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
689 GOTO(out, rc = -ENOMEM);
691 * strsep() modifies its argument, so keep a copy
695 if (copy_from_user(cmd, buffer, count))
696 GOTO(out, rc = -EFAULT);
700 policy_name = strsep(&cmd, " ");
702 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
703 GOTO(out, rc = -EINVAL);
706 * No [reg|hp] token has been specified
711 queue_name = strsep(&cmd, " ");
713 * The second token is either an optional [reg|hp] string,
716 if (strcmp(queue_name, "reg") == 0)
717 queue = PTLRPC_NRS_QUEUE_REG;
718 else if (strcmp(queue_name, "hp") == 0)
719 queue = PTLRPC_NRS_QUEUE_HP;
728 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
729 GOTO(out, rc = -ENODEV);
730 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
731 queue = PTLRPC_NRS_QUEUE_REG;
734 * Serialize NRS core lprocfs operations with policy registration/
737 mutex_lock(&nrs_core.nrs_mutex);
739 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
740 PTLRPC_NRS_CTL_START,
743 mutex_unlock(&nrs_core.nrs_mutex);
746 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
748 RETURN(rc < 0 ? rc : count);
750 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
754 struct ptlrpc_srh_iterator {
757 struct ptlrpc_request *srhi_req;
761 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
762 struct ptlrpc_srh_iterator *srhi,
766 struct ptlrpc_request *req;
768 if (srhi->srhi_req != NULL &&
769 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
770 srhi->srhi_seq <= seq) {
771 /* If srhi_req was set previously, hasn't been culled and
772 * we're searching for a seq on or after it (i.e. more
773 * recent), search from it onwards.
774 * Since the service history is LRU (i.e. culled reqs will
775 * be near the head), we shouldn't have to do long
777 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
778 "%s:%d: seek seq %llu, request seq %llu\n",
779 svcpt->scp_service->srv_name, svcpt->scp_cpt,
780 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
781 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
782 "%s:%d: seek offset %llu, request seq %llu, "
783 "last culled %llu\n",
784 svcpt->scp_service->srv_name, svcpt->scp_cpt,
785 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
786 e = &srhi->srhi_req->rq_history_list;
788 /* search from start */
789 e = svcpt->scp_hist_reqs.next;
792 while (e != &svcpt->scp_hist_reqs) {
793 req = list_entry(e, struct ptlrpc_request, rq_history_list);
795 if (req->rq_history_seq >= seq) {
796 srhi->srhi_seq = req->rq_history_seq;
797 srhi->srhi_req = req;
807 * ptlrpc history sequence is used as "position" of seq_file, in some case,
808 * seq_read() will increase "position" to indicate reading the next
809 * element, however, low bits of history sequence are reserved for CPT id
810 * (check the details from comments before ptlrpc_req_add_history), which
811 * means seq_read() might change CPT id of history sequence and never
812 * finish reading of requests on a CPT. To make it work, we have to shift
813 * CPT id to high bits and timestamp to low bits, so seq_read() will only
814 * increase timestamp which can correctly indicate the next position.
817 /* convert seq_file pos to cpt */
818 #define PTLRPC_REQ_POS2CPT(svc, pos) \
819 ((svc)->srv_cpt_bits == 0 ? 0 : \
820 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
822 /* make up seq_file pos from cpt */
823 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
824 ((svc)->srv_cpt_bits == 0 ? 0 : \
825 (cpt) << (64 - (svc)->srv_cpt_bits))
827 /* convert sequence to position */
828 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
829 ((svc)->srv_cpt_bits == 0 ? (seq) : \
830 ((seq) >> (svc)->srv_cpt_bits) | \
831 ((seq) << (64 - (svc)->srv_cpt_bits)))
833 /* convert position to sequence */
834 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
835 ((svc)->srv_cpt_bits == 0 ? (pos) : \
836 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
837 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
840 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
842 struct ptlrpc_service *svc = s->private;
843 struct ptlrpc_service_part *svcpt;
844 struct ptlrpc_srh_iterator *srhi;
849 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
850 CWARN("Failed to read request history because size of loff_t "
851 "%d can't match size of u64\n", (int)sizeof(loff_t));
855 OBD_ALLOC(srhi, sizeof(*srhi));
860 srhi->srhi_req = NULL;
862 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
864 ptlrpc_service_for_each_part(svcpt, i, svc) {
865 if (i < cpt) /* skip */
867 if (i > cpt) /* make up the lowest position for this CPT */
868 *pos = PTLRPC_REQ_CPT2POS(svc, i);
870 spin_lock(&svcpt->scp_lock);
871 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
872 PTLRPC_REQ_POS2SEQ(svc, *pos));
873 spin_unlock(&svcpt->scp_lock);
875 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
881 OBD_FREE(srhi, sizeof(*srhi));
886 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
888 struct ptlrpc_srh_iterator *srhi = iter;
891 OBD_FREE(srhi, sizeof(*srhi));
895 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
896 void *iter, loff_t *pos)
898 struct ptlrpc_service *svc = s->private;
899 struct ptlrpc_srh_iterator *srhi = iter;
900 struct ptlrpc_service_part *svcpt;
905 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
906 svcpt = svc->srv_parts[i];
908 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
909 srhi->srhi_req = NULL;
910 seq = srhi->srhi_seq = 0;
911 } else { /* the next sequence */
912 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
915 spin_lock(&svcpt->scp_lock);
916 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
917 spin_unlock(&svcpt->scp_lock);
919 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
925 OBD_FREE(srhi, sizeof(*srhi));
929 /* common ost/mdt so_req_printer */
930 void target_print_req(void *seq_file, struct ptlrpc_request *req)
932 /* Called holding srv_lock with irqs disabled.
933 * Print specific req contents and a newline.
934 * CAVEAT EMPTOR: check request message length before printing!!!
935 * You might have received any old crap so you must be just as
936 * careful here as the service's request parser!!! */
937 struct seq_file *sf = seq_file;
939 switch (req->rq_phase) {
941 /* still awaiting a service thread's attention, or rejected
942 * because the generic request message didn't unpack */
943 seq_printf(sf, "<not swabbed>\n");
945 case RQ_PHASE_INTERPRET:
946 /* being handled, so basic msg swabbed, and opc is valid
947 * but racing with mds_handle() */
948 case RQ_PHASE_COMPLETE:
949 /* been handled by mds_handle() reply state possibly still
951 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
954 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
957 EXPORT_SYMBOL(target_print_req);
959 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
961 struct ptlrpc_service *svc = s->private;
962 struct ptlrpc_srh_iterator *srhi = iter;
963 struct ptlrpc_service_part *svcpt;
964 struct ptlrpc_request *req;
967 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
969 svcpt = svc->srv_parts[srhi->srhi_idx];
971 spin_lock(&svcpt->scp_lock);
973 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
976 struct timespec64 arrival, sent, arrivaldiff;
977 char nidstr[LNET_NIDSTR_SIZE];
979 req = srhi->srhi_req;
981 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
982 arrival.tv_sec = req->rq_arrival_time.tv_sec;
983 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
984 sent.tv_sec = req->rq_sent;
986 arrivaldiff = timespec64_sub(sent, arrival);
988 /* Print common req fields.
989 * CAVEAT EMPTOR: we're racing with the service handler
990 * here. The request could contain any old crap, so you
991 * must be just as careful as the service's request
992 * parser. Currently I only print stuff here I know is OK
993 * to look at coz it was set up in request_in_callback()!!!
995 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
996 req->rq_history_seq, nidstr,
997 libcfs_id2str(req->rq_peer), req->rq_xid,
998 req->rq_reqlen, ptlrpc_rqphase2str(req),
999 (s64)req->rq_arrival_time.tv_sec,
1000 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1001 (s64)arrivaldiff.tv_sec,
1002 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1003 (s64)(req->rq_sent - req->rq_deadline));
1004 if (svc->srv_ops.so_req_printer == NULL)
1005 seq_printf(s, "\n");
1007 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1010 spin_unlock(&svcpt->scp_lock);
1015 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1017 static struct seq_operations sops = {
1018 .start = ptlrpc_lprocfs_svc_req_history_start,
1019 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1020 .next = ptlrpc_lprocfs_svc_req_history_next,
1021 .show = ptlrpc_lprocfs_svc_req_history_show,
1023 struct seq_file *seqf;
1026 rc = LPROCFS_ENTRY_CHECK(inode);
1030 rc = seq_open(file, &sops);
1034 seqf = file->private_data;
1035 seqf->private = PDE_DATA(inode);
1039 /* See also lprocfs_rd_timeouts */
1040 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1042 struct ptlrpc_service *svc = m->private;
1043 struct ptlrpc_service_part *svcpt;
1050 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1055 ptlrpc_service_for_each_part(svcpt, i, svc) {
1056 cur = at_get(&svcpt->scp_at_estimate);
1057 worst = svcpt->scp_at_estimate.at_worst_ever;
1058 worstt = svcpt->scp_at_estimate.at_worst_time;
1060 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1061 "service", cur, worst, (s64)worstt,
1062 (s64)(ktime_get_real_seconds() - worstt));
1064 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1069 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1071 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1073 struct ptlrpc_service *svc = m->private;
1074 seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1079 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1080 size_t count, loff_t *off)
1082 struct seq_file *m = file->private_data;
1083 struct ptlrpc_service *svc = m->private;
1087 rc = lprocfs_str_to_s64(buffer, count, &val);
1091 if (val < 0 || val > INT_MAX)
1094 spin_lock(&svc->srv_lock);
1095 svc->srv_hpreq_ratio = val;
1096 spin_unlock(&svc->srv_lock);
1100 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1102 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1103 struct ptlrpc_service *svc)
1105 struct lprocfs_vars lproc_vars[] = {
1106 { .name = "high_priority_ratio",
1107 .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1109 { .name = "req_buffer_history_len",
1110 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1112 { .name = "req_buffer_history_max",
1113 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1115 { .name = "threads_min",
1116 .fops = &ptlrpc_lprocfs_threads_min_fops,
1118 { .name = "threads_max",
1119 .fops = &ptlrpc_lprocfs_threads_max_fops,
1121 { .name = "threads_started",
1122 .fops = &ptlrpc_lprocfs_threads_started_fops,
1124 { .name = "timeouts",
1125 .fops = &ptlrpc_lprocfs_timeouts_fops,
1127 { .name = "nrs_policies",
1128 .fops = &ptlrpc_lprocfs_nrs_fops,
1132 static struct file_operations req_history_fops = {
1133 .owner = THIS_MODULE,
1134 .open = ptlrpc_lprocfs_svc_req_history_open,
1136 .llseek = seq_lseek,
1137 .release = lprocfs_seq_release,
1142 ptlrpc_lprocfs_register(entry, svc->srv_name,
1143 "stats", &svc->srv_procroot,
1145 if (svc->srv_procroot == NULL)
1148 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1150 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1151 0400, &req_history_fops, svc);
1153 CWARN("Error adding the req_history file\n");
1156 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1158 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1159 &obddev->obd_svc_procroot,
1160 &obddev->obd_svc_stats);
1162 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1164 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1166 struct lprocfs_stats *svc_stats;
1167 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1168 int opc = opcode_offset(op);
1170 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1171 if (svc_stats == NULL || opc <= 0)
1173 LASSERT(opc < LUSTRE_MAX_OPCODES);
1174 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1175 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1178 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1180 struct lprocfs_stats *svc_stats;
1183 if (!req->rq_import)
1185 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1188 idx = lustre_msg_get_opc(req->rq_reqmsg);
1191 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1194 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1197 LASSERTF(0, "unsupported opcode %u\n", idx);
1201 lprocfs_counter_add(svc_stats, idx, bytes);
1204 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1206 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1208 if (svc->srv_procroot != NULL)
1209 lprocfs_remove(&svc->srv_procroot);
1212 lprocfs_free_stats(&svc->srv_stats);
1215 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1217 if (obd->obd_svc_procroot)
1218 lprocfs_remove(&obd->obd_svc_procroot);
1220 if (obd->obd_svc_stats)
1221 lprocfs_free_stats(&obd->obd_svc_stats);
1223 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1226 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1227 size_t count, loff_t *off)
1229 struct seq_file *m = file->private_data;
1230 struct obd_device *obd = m->private;
1231 struct ptlrpc_request *req;
1235 LPROCFS_CLIMP_CHECK(obd);
1236 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1237 LPROCFS_CLIMP_EXIT(obd);
1241 req->rq_send_state = LUSTRE_IMP_FULL;
1243 rc = ptlrpc_queue_wait(req);
1245 ptlrpc_req_finished(req);
1250 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1252 /* Write the connection UUID to this file to attempt to connect to that node.
1253 * The connection UUID is a node's primary NID. For example,
1254 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1257 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1258 size_t count, loff_t *off)
1260 struct seq_file *m = file->private_data;
1261 struct obd_device *obd = m->private;
1262 struct obd_import *imp = obd->u.cli.cl_import;
1267 const char prefix[] = "connection=";
1268 const int prefix_len = sizeof(prefix) - 1;
1270 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1273 OBD_ALLOC(kbuf, count + 1);
1277 if (copy_from_user(kbuf, buffer, count))
1278 GOTO(out, count = -EFAULT);
1282 /* only support connection=uuid::instance now */
1283 if (strncmp(prefix, kbuf, prefix_len) != 0)
1284 GOTO(out, count = -EINVAL);
1286 uuid = kbuf + prefix_len;
1287 ptr = strstr(uuid, "::");
1294 ptr += 2; /* Skip :: */
1295 inst = simple_strtol(ptr, &endptr, 10);
1297 CERROR("config: wrong instance # %s\n", ptr);
1298 } else if (inst != imp->imp_connect_data.ocd_instance) {
1299 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1300 "target(%u/%u), reconnecting...\n",
1301 imp->imp_obd->obd_name,
1302 imp->imp_connect_data.ocd_instance, inst);
1305 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1307 imp->imp_obd->obd_name, inst);
1312 ptlrpc_recover_import(imp, uuid, 1);
1315 OBD_FREE(kbuf, count + 1);
1318 EXPORT_SYMBOL(lprocfs_import_seq_write);
1320 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1322 struct obd_device *obd = m->private;
1323 struct obd_import *imp = obd->u.cli.cl_import;
1325 LPROCFS_CLIMP_CHECK(obd);
1326 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1327 LPROCFS_CLIMP_EXIT(obd);
1330 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1333 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1334 size_t count, loff_t *off)
1336 struct seq_file *m = file->private_data;
1337 struct obd_device *obd = m->private;
1338 struct client_obd *cli = &obd->u.cli;
1339 struct obd_import *imp = cli->cl_import;
1343 rc = lprocfs_str_to_s64(buffer, count, &val);
1347 if (val != 0 && val != 1)
1350 LPROCFS_CLIMP_CHECK(obd);
1351 spin_lock(&imp->imp_lock);
1352 imp->imp_no_pinger_recover = !val;
1353 spin_unlock(&imp->imp_lock);
1354 LPROCFS_CLIMP_EXIT(obd);
1357 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1359 #endif /* CONFIG_PROC_FS */