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_net.h>
39 #include <obd_class.h>
40 #include "ptlrpc_internal.h"
43 static struct ll_rpc_opcode {
46 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
47 { OST_REPLY, "ost_reply" },
48 { OST_GETATTR, "ost_getattr" },
49 { OST_SETATTR, "ost_setattr" },
50 { OST_READ, "ost_read" },
51 { OST_WRITE, "ost_write" },
52 { OST_CREATE , "ost_create" },
53 { OST_DESTROY, "ost_destroy" },
54 { OST_GET_INFO, "ost_get_info" },
55 { OST_CONNECT, "ost_connect" },
56 { OST_DISCONNECT, "ost_disconnect" },
57 { OST_PUNCH, "ost_punch" },
58 { OST_OPEN, "ost_open" },
59 { OST_CLOSE, "ost_close" },
60 { OST_STATFS, "ost_statfs" },
61 { 14, NULL }, /* formerly OST_SAN_READ */
62 { 15, NULL }, /* formerly OST_SAN_WRITE */
63 { OST_SYNC, "ost_sync" },
64 { OST_SET_INFO, "ost_set_info" },
65 { OST_QUOTACHECK, "ost_quotacheck" },
66 { OST_QUOTACTL, "ost_quotactl" },
67 { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
68 { OST_LADVISE, "ost_ladvise" },
69 { MDS_GETATTR, "mds_getattr" },
70 { MDS_GETATTR_NAME, "mds_getattr_lock" },
71 { MDS_CLOSE, "mds_close" },
72 { MDS_REINT, "mds_reint" },
73 { MDS_READPAGE, "mds_readpage" },
74 { MDS_CONNECT, "mds_connect" },
75 { MDS_DISCONNECT, "mds_disconnect" },
76 { MDS_GET_ROOT, "mds_get_root" },
77 { MDS_STATFS, "mds_statfs" },
78 { MDS_PIN, "mds_pin" },
79 { MDS_UNPIN, "mds_unpin" },
80 { MDS_SYNC, "mds_sync" },
81 { MDS_DONE_WRITING, "mds_done_writing" },
82 { MDS_SET_INFO, "mds_set_info" },
83 { MDS_QUOTACHECK, "mds_quotacheck" },
84 { MDS_QUOTACTL, "mds_quotactl" },
85 { MDS_GETXATTR, "mds_getxattr" },
86 { MDS_SETXATTR, "mds_setxattr" },
87 { MDS_WRITEPAGE, "mds_writepage" },
88 { MDS_IS_SUBDIR, "mds_is_subdir" },
89 { MDS_GET_INFO, "mds_get_info" },
90 { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
91 { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
92 { MDS_HSM_ACTION, "mds_hsm_action" },
93 { MDS_HSM_PROGRESS, "mds_hsm_progress" },
94 { MDS_HSM_REQUEST, "mds_hsm_request" },
95 { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
96 { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
97 { MDS_SWAP_LAYOUTS, "mds_swap_layouts" },
98 { LDLM_ENQUEUE, "ldlm_enqueue" },
99 { LDLM_CONVERT, "ldlm_convert" },
100 { LDLM_CANCEL, "ldlm_cancel" },
101 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
102 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
103 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
104 { LDLM_SET_INFO, "ldlm_set_info" },
105 { MGS_CONNECT, "mgs_connect" },
106 { MGS_DISCONNECT, "mgs_disconnect" },
107 { MGS_EXCEPTION, "mgs_exception" },
108 { MGS_TARGET_REG, "mgs_target_reg" },
109 { MGS_TARGET_DEL, "mgs_target_del" },
110 { MGS_SET_INFO, "mgs_set_info" },
111 { MGS_CONFIG_READ, "mgs_config_read" },
112 { OBD_PING, "obd_ping" },
113 { OBD_LOG_CANCEL, "llog_cancel" },
114 { OBD_QC_CALLBACK, "obd_quota_callback" },
115 { OBD_IDX_READ, "dt_index_read" },
116 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
117 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
118 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
119 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
120 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
121 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
122 { LLOG_CATINFO, "llog_catinfo" },
123 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
124 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
125 { QUOTA_DQACQ, "quota_acquire" },
126 { QUOTA_DQREL, "quota_release" },
127 { SEQ_QUERY, "seq_query" },
128 { SEC_CTX_INIT, "sec_ctx_init" },
129 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
130 { SEC_CTX_FINI, "sec_ctx_fini" },
131 { FLD_QUERY, "fld_query" },
132 { FLD_READ, "fld_read" },
133 { OUT_UPDATE, "out_update" },
134 { LFSCK_NOTIFY, "lfsck_notify" },
135 { LFSCK_QUERY, "lfsck_query" },
138 static struct ll_eopcode {
141 } ll_eopcode_table[EXTRA_LAST_OPC] = {
142 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
143 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
144 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
145 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
146 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
147 { MDS_REINT_SETATTR, "mds_reint_setattr" },
148 { MDS_REINT_CREATE, "mds_reint_create" },
149 { MDS_REINT_LINK, "mds_reint_link" },
150 { MDS_REINT_UNLINK, "mds_reint_unlink" },
151 { MDS_REINT_RENAME, "mds_reint_rename" },
152 { MDS_REINT_OPEN, "mds_reint_open" },
153 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
154 { BRW_READ_BYTES, "read_bytes" },
155 { BRW_WRITE_BYTES, "write_bytes" },
158 const char *ll_opcode2str(__u32 opcode)
160 /* When one of the assertions below fail, chances are that:
161 * 1) A new opcode was added in include/lustre/lustre_idl.h,
162 * but is missing from the table above.
163 * or 2) The opcode space was renumbered or rearranged,
164 * and the opcode_offset() function in
165 * ptlrpc_internal.h needs to be modified.
167 __u32 offset = opcode_offset(opcode);
168 LASSERTF(offset < LUSTRE_MAX_OPCODES,
169 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
170 offset, LUSTRE_MAX_OPCODES);
171 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
172 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
173 offset, ll_rpc_opcode_table[offset].opcode, opcode);
174 return ll_rpc_opcode_table[offset].opname;
177 const int ll_str2opcode(const char *ops)
181 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
182 if (ll_rpc_opcode_table[i].opname != NULL &&
183 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
184 return ll_rpc_opcode_table[i].opcode;
190 static const char *ll_eopcode2str(__u32 opcode)
192 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
193 return ll_eopcode_table[opcode].opname;
196 #ifdef CONFIG_PROC_FS
197 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
198 char *name, struct proc_dir_entry **procroot_ret,
199 struct lprocfs_stats **stats_ret)
201 struct proc_dir_entry *svc_procroot;
202 struct lprocfs_stats *svc_stats;
204 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
207 LASSERT(*procroot_ret == NULL);
208 LASSERT(*stats_ret == NULL);
210 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
211 if (svc_stats == NULL)
215 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
216 if (IS_ERR(svc_procroot)) {
217 lprocfs_free_stats(&svc_stats);
224 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
225 svc_counter_config, "req_waittime", "usec");
226 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
227 svc_counter_config, "req_qdepth", "reqs");
228 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
229 svc_counter_config, "req_active", "reqs");
230 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
231 svc_counter_config, "req_timeout", "sec");
232 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
233 svc_counter_config, "reqbuf_avail", "bufs");
234 for (i = 0; i < EXTRA_LAST_OPC; i++) {
238 case BRW_WRITE_BYTES:
246 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
248 ll_eopcode2str(i), units);
250 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
251 __u32 opcode = ll_rpc_opcode_table[i].opcode;
252 lprocfs_counter_init(svc_stats,
253 EXTRA_MAX_OPCODES + i, svc_counter_config,
254 ll_opcode2str(opcode), "usec");
257 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
260 lprocfs_remove(&svc_procroot);
261 lprocfs_free_stats(&svc_stats);
264 *procroot_ret = svc_procroot;
265 *stats_ret = svc_stats;
270 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
272 struct ptlrpc_service *svc = m->private;
273 struct ptlrpc_service_part *svcpt;
277 ptlrpc_service_for_each_part(svcpt, i, svc)
278 total += svcpt->scp_hist_nrqbds;
280 seq_printf(m, "%d\n", total);
283 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
286 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
288 struct ptlrpc_service *svc = m->private;
289 struct ptlrpc_service_part *svcpt;
293 ptlrpc_service_for_each_part(svcpt, i, svc)
294 total += svc->srv_hist_nrqbds_cpt_max;
296 seq_printf(m, "%d\n", total);
301 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
302 const char __user *buffer,
303 size_t count, loff_t *off)
305 struct seq_file *m = file->private_data;
306 struct ptlrpc_service *svc = m->private;
311 rc = lprocfs_str_to_s64(buffer, count, &val);
315 if (val < 0 || val > INT_MAX)
318 /* This sanity check is more of an insanity check; we can still
319 * hose a kernel by allowing the request history to grow too
321 bufpages = (svc->srv_buf_size + PAGE_SIZE - 1) >>
323 if (val > totalram_pages/(2 * bufpages))
326 spin_lock(&svc->srv_lock);
329 svc->srv_hist_nrqbds_cpt_max = 0;
331 svc->srv_hist_nrqbds_cpt_max =
332 max(1, ((int)val / svc->srv_ncpts));
334 spin_unlock(&svc->srv_lock);
338 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
340 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
343 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
346 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
349 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
350 const char *buffer, size_t count)
352 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
357 rc = kstrtoul(buffer, 10, &val);
361 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
364 spin_lock(&svc->srv_lock);
365 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
366 spin_unlock(&svc->srv_lock);
370 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
372 spin_unlock(&svc->srv_lock);
376 LUSTRE_RW_ATTR(threads_min);
378 static ssize_t threads_started_show(struct kobject *kobj,
379 struct attribute *attr,
382 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
384 struct ptlrpc_service_part *svcpt;
388 ptlrpc_service_for_each_part(svcpt, i, svc)
389 total += svcpt->scp_nthrs_running;
391 return sprintf(buf, "%d\n", total);
393 LUSTRE_RO_ATTR(threads_started);
395 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
398 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
401 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
404 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
405 const char *buffer, size_t count)
407 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
412 rc = kstrtoul(buffer, 10, &val);
416 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
419 spin_lock(&svc->srv_lock);
420 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
421 spin_unlock(&svc->srv_lock);
425 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
427 spin_unlock(&svc->srv_lock);
431 LUSTRE_RW_ATTR(threads_max);
434 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
436 * \param[in] state The policy state
438 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
443 case NRS_POL_STATE_INVALID:
445 case NRS_POL_STATE_STOPPED:
447 case NRS_POL_STATE_STOPPING:
449 case NRS_POL_STATE_STARTING:
451 case NRS_POL_STATE_STARTED:
457 * Obtains status information for \a policy.
459 * Information is copied in \a info.
461 * \param[in] policy The policy
462 * \param[out] info Holds returned status information
464 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
465 struct ptlrpc_nrs_pol_info *info)
467 LASSERT(policy != NULL);
468 LASSERT(info != NULL);
469 assert_spin_locked(&policy->pol_nrs->nrs_lock);
471 CLASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
472 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
473 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
475 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
476 info->pi_state = policy->pol_state;
478 * XXX: These are accessed without holding
479 * ptlrpc_service_part::scp_req_lock.
481 info->pi_req_queued = policy->pol_req_queued;
482 info->pi_req_started = policy->pol_req_started;
486 * Reads and prints policy status information for all policies of a PTLRPC
489 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
491 struct ptlrpc_service *svc = m->private;
492 struct ptlrpc_service_part *svcpt;
493 struct ptlrpc_nrs *nrs;
494 struct ptlrpc_nrs_policy *policy;
495 struct ptlrpc_nrs_pol_info *infos;
496 struct ptlrpc_nrs_pol_info tmp;
498 unsigned pol_idx = 0;
505 * Serialize NRS core lprocfs operations with policy registration/
508 mutex_lock(&nrs_core.nrs_mutex);
511 * Use the first service partition's regular NRS head in order to obtain
512 * the number of policies registered with NRS heads of this service. All
513 * service partitions will have the same number of policies.
515 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
517 spin_lock(&nrs->nrs_lock);
518 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
519 spin_unlock(&nrs->nrs_lock);
521 OBD_ALLOC(infos, num_pols * sizeof(*infos));
523 GOTO(out, rc = -ENOMEM);
526 ptlrpc_service_for_each_part(svcpt, i, svc) {
527 nrs = nrs_svcpt2nrs(svcpt, hp);
528 spin_lock(&nrs->nrs_lock);
532 list_for_each_entry(policy, &nrs->nrs_policy_list,
534 LASSERT(pol_idx < num_pols);
536 nrs_policy_get_info_locked(policy, &tmp);
538 * Copy values when handling the first service
542 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
544 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
546 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
547 sizeof(tmp.pi_state));
548 infos[pol_idx].pi_fallback = tmp.pi_fallback;
550 * For the rest of the service partitions
551 * sanity-check the values we get.
554 LASSERT(strncmp(infos[pol_idx].pi_name,
556 NRS_POL_NAME_MAX) == 0);
557 LASSERT(strncmp(infos[pol_idx].pi_arg,
559 sizeof(tmp.pi_arg)) == 0);
561 * Not asserting ptlrpc_nrs_pol_info::pi_state,
562 * because it may be different between
563 * instances of the same policy in different
564 * service partitions.
566 LASSERT(infos[pol_idx].pi_fallback ==
570 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
571 infos[pol_idx].pi_req_started += tmp.pi_req_started;
575 spin_unlock(&nrs->nrs_lock);
579 * Policy status information output is in YAML format.
595 * high_priority_requests:
608 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
609 "high_priority_requests:");
611 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
612 if (strlen(infos[pol_idx].pi_arg) > 0)
613 seq_printf(m, " - name: %s %s\n",
614 infos[pol_idx].pi_name,
615 infos[pol_idx].pi_arg);
617 seq_printf(m, " - name: %s\n",
618 infos[pol_idx].pi_name);
621 seq_printf(m, " state: %s\n"
624 " active: %-20d\n\n",
625 nrs_state2str(infos[pol_idx].pi_state),
626 infos[pol_idx].pi_fallback ? "yes" : "no",
627 (int)infos[pol_idx].pi_req_queued,
628 (int)infos[pol_idx].pi_req_started);
631 if (!hp && nrs_svc_has_hp(svc)) {
632 memset(infos, 0, num_pols * sizeof(*infos));
635 * Redo the processing for the service's HP NRS heads' policies.
643 OBD_FREE(infos, num_pols * sizeof(*infos));
645 mutex_unlock(&nrs_core.nrs_mutex);
651 #define LPROCFS_NRS_WR_MAX_ARG (1024)
653 * The longest valid command string is the maxium policy name size, plus the
654 * length of the " reg" substring, plus the lenght of argument
656 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
657 + LPROCFS_NRS_WR_MAX_ARG)
660 * Starts and stops a given policy on a PTLRPC service.
662 * Commands consist of the policy name, followed by an optional [reg|hp] token;
663 * if the optional token is omitted, the operation is performed on both the
664 * regular and high-priority (if the service has one) NRS head.
667 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
668 size_t count, loff_t *off)
670 struct seq_file *m = file->private_data;
671 struct ptlrpc_service *svc = m->private;
672 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
674 char *cmd_copy = NULL;
680 if (count >= LPROCFS_NRS_WR_MAX_CMD)
681 GOTO(out, rc = -EINVAL);
683 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
685 GOTO(out, rc = -ENOMEM);
687 * strsep() modifies its argument, so keep a copy
691 if (copy_from_user(cmd, buffer, count))
692 GOTO(out, rc = -EFAULT);
696 policy_name = strsep(&cmd, " ");
698 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
699 GOTO(out, rc = -EINVAL);
702 * No [reg|hp] token has been specified
707 queue_name = strsep(&cmd, " ");
709 * The second token is either an optional [reg|hp] string,
712 if (strcmp(queue_name, "reg") == 0)
713 queue = PTLRPC_NRS_QUEUE_REG;
714 else if (strcmp(queue_name, "hp") == 0)
715 queue = PTLRPC_NRS_QUEUE_HP;
724 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
725 GOTO(out, rc = -ENODEV);
726 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
727 queue = PTLRPC_NRS_QUEUE_REG;
730 * Serialize NRS core lprocfs operations with policy registration/
733 mutex_lock(&nrs_core.nrs_mutex);
735 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
736 PTLRPC_NRS_CTL_START,
739 mutex_unlock(&nrs_core.nrs_mutex);
742 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
744 RETURN(rc < 0 ? rc : count);
746 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
750 struct ptlrpc_srh_iterator {
753 struct ptlrpc_request *srhi_req;
757 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
758 struct ptlrpc_srh_iterator *srhi,
762 struct ptlrpc_request *req;
764 if (srhi->srhi_req != NULL &&
765 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
766 srhi->srhi_seq <= seq) {
767 /* If srhi_req was set previously, hasn't been culled and
768 * we're searching for a seq on or after it (i.e. more
769 * recent), search from it onwards.
770 * Since the service history is LRU (i.e. culled reqs will
771 * be near the head), we shouldn't have to do long
773 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
774 "%s:%d: seek seq %llu, request seq %llu\n",
775 svcpt->scp_service->srv_name, svcpt->scp_cpt,
776 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
777 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
778 "%s:%d: seek offset %llu, request seq %llu, "
779 "last culled %llu\n",
780 svcpt->scp_service->srv_name, svcpt->scp_cpt,
781 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
782 e = &srhi->srhi_req->rq_history_list;
784 /* search from start */
785 e = svcpt->scp_hist_reqs.next;
788 while (e != &svcpt->scp_hist_reqs) {
789 req = list_entry(e, struct ptlrpc_request, rq_history_list);
791 if (req->rq_history_seq >= seq) {
792 srhi->srhi_seq = req->rq_history_seq;
793 srhi->srhi_req = req;
803 * ptlrpc history sequence is used as "position" of seq_file, in some case,
804 * seq_read() will increase "position" to indicate reading the next
805 * element, however, low bits of history sequence are reserved for CPT id
806 * (check the details from comments before ptlrpc_req_add_history), which
807 * means seq_read() might change CPT id of history sequence and never
808 * finish reading of requests on a CPT. To make it work, we have to shift
809 * CPT id to high bits and timestamp to low bits, so seq_read() will only
810 * increase timestamp which can correctly indicate the next position.
813 /* convert seq_file pos to cpt */
814 #define PTLRPC_REQ_POS2CPT(svc, pos) \
815 ((svc)->srv_cpt_bits == 0 ? 0 : \
816 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
818 /* make up seq_file pos from cpt */
819 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
820 ((svc)->srv_cpt_bits == 0 ? 0 : \
821 (cpt) << (64 - (svc)->srv_cpt_bits))
823 /* convert sequence to position */
824 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
825 ((svc)->srv_cpt_bits == 0 ? (seq) : \
826 ((seq) >> (svc)->srv_cpt_bits) | \
827 ((seq) << (64 - (svc)->srv_cpt_bits)))
829 /* convert position to sequence */
830 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
831 ((svc)->srv_cpt_bits == 0 ? (pos) : \
832 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
833 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
836 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
838 struct ptlrpc_service *svc = s->private;
839 struct ptlrpc_service_part *svcpt;
840 struct ptlrpc_srh_iterator *srhi;
845 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
846 CWARN("Failed to read request history because size of loff_t "
847 "%d can't match size of u64\n", (int)sizeof(loff_t));
851 OBD_ALLOC(srhi, sizeof(*srhi));
856 srhi->srhi_req = NULL;
858 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
860 ptlrpc_service_for_each_part(svcpt, i, svc) {
861 if (i < cpt) /* skip */
863 if (i > cpt) /* make up the lowest position for this CPT */
864 *pos = PTLRPC_REQ_CPT2POS(svc, i);
866 spin_lock(&svcpt->scp_lock);
867 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
868 PTLRPC_REQ_POS2SEQ(svc, *pos));
869 spin_unlock(&svcpt->scp_lock);
871 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
877 OBD_FREE(srhi, sizeof(*srhi));
882 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
884 struct ptlrpc_srh_iterator *srhi = iter;
887 OBD_FREE(srhi, sizeof(*srhi));
891 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
892 void *iter, loff_t *pos)
894 struct ptlrpc_service *svc = s->private;
895 struct ptlrpc_srh_iterator *srhi = iter;
896 struct ptlrpc_service_part *svcpt;
901 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
902 svcpt = svc->srv_parts[i];
904 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
905 srhi->srhi_req = NULL;
906 seq = srhi->srhi_seq = 0;
907 } else { /* the next sequence */
908 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
911 spin_lock(&svcpt->scp_lock);
912 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
913 spin_unlock(&svcpt->scp_lock);
915 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
921 OBD_FREE(srhi, sizeof(*srhi));
925 /* common ost/mdt so_req_printer */
926 void target_print_req(void *seq_file, struct ptlrpc_request *req)
928 /* Called holding srv_lock with irqs disabled.
929 * Print specific req contents and a newline.
930 * CAVEAT EMPTOR: check request message length before printing!!!
931 * You might have received any old crap so you must be just as
932 * careful here as the service's request parser!!! */
933 struct seq_file *sf = seq_file;
935 switch (req->rq_phase) {
937 /* still awaiting a service thread's attention, or rejected
938 * because the generic request message didn't unpack */
939 seq_printf(sf, "<not swabbed>\n");
941 case RQ_PHASE_INTERPRET:
942 /* being handled, so basic msg swabbed, and opc is valid
943 * but racing with mds_handle() */
944 case RQ_PHASE_COMPLETE:
945 /* been handled by mds_handle() reply state possibly still
947 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
950 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
953 EXPORT_SYMBOL(target_print_req);
955 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
957 struct ptlrpc_service *svc = s->private;
958 struct ptlrpc_srh_iterator *srhi = iter;
959 struct ptlrpc_service_part *svcpt;
960 struct ptlrpc_request *req;
963 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
965 svcpt = svc->srv_parts[srhi->srhi_idx];
967 spin_lock(&svcpt->scp_lock);
969 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
972 struct timespec64 arrival, sent, arrivaldiff;
973 char nidstr[LNET_NIDSTR_SIZE];
975 req = srhi->srhi_req;
977 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
978 arrival.tv_sec = req->rq_arrival_time.tv_sec;
979 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
980 sent.tv_sec = req->rq_sent;
982 arrivaldiff = timespec64_sub(sent, arrival);
984 /* Print common req fields.
985 * CAVEAT EMPTOR: we're racing with the service handler
986 * here. The request could contain any old crap, so you
987 * must be just as careful as the service's request
988 * parser. Currently I only print stuff here I know is OK
989 * to look at coz it was set up in request_in_callback()!!!
991 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
992 req->rq_history_seq, nidstr,
993 libcfs_id2str(req->rq_peer), req->rq_xid,
994 req->rq_reqlen, ptlrpc_rqphase2str(req),
995 (s64)req->rq_arrival_time.tv_sec,
996 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
997 (s64)arrivaldiff.tv_sec,
998 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
999 (s64)(req->rq_sent - req->rq_deadline));
1000 if (svc->srv_ops.so_req_printer == NULL)
1001 seq_printf(s, "\n");
1003 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1006 spin_unlock(&svcpt->scp_lock);
1011 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1013 static struct seq_operations sops = {
1014 .start = ptlrpc_lprocfs_svc_req_history_start,
1015 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1016 .next = ptlrpc_lprocfs_svc_req_history_next,
1017 .show = ptlrpc_lprocfs_svc_req_history_show,
1019 struct seq_file *seqf;
1022 rc = LPROCFS_ENTRY_CHECK(inode);
1026 rc = seq_open(file, &sops);
1030 seqf = file->private_data;
1031 seqf->private = PDE_DATA(inode);
1035 /* See also lprocfs_rd_timeouts */
1036 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1038 struct ptlrpc_service *svc = m->private;
1039 struct ptlrpc_service_part *svcpt;
1046 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1051 ptlrpc_service_for_each_part(svcpt, i, svc) {
1052 cur = at_get(&svcpt->scp_at_estimate);
1053 worst = svcpt->scp_at_estimate.at_worst_ever;
1054 worstt = svcpt->scp_at_estimate.at_worst_time;
1056 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1057 "service", cur, worst, (s64)worstt,
1058 (s64)(ktime_get_real_seconds() - worstt));
1060 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1065 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1067 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1068 struct attribute *attr,
1071 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1074 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1077 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1078 struct attribute *attr,
1082 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1087 rc = kstrtoul(buffer, 10, &val);
1091 spin_lock(&svc->srv_lock);
1092 svc->srv_hpreq_ratio = val;
1093 spin_unlock(&svc->srv_lock);
1097 LUSTRE_RW_ATTR(high_priority_ratio);
1099 static struct attribute *ptlrpc_svc_attrs[] = {
1100 &lustre_attr_threads_min.attr,
1101 &lustre_attr_threads_started.attr,
1102 &lustre_attr_threads_max.attr,
1103 &lustre_attr_high_priority_ratio.attr,
1107 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1109 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1112 complete(&svc->srv_kobj_unregister);
1115 static struct kobj_type ptlrpc_svc_ktype = {
1116 .default_attrs = ptlrpc_svc_attrs,
1117 .sysfs_ops = &lustre_sysfs_ops,
1118 .release = ptlrpc_sysfs_svc_release,
1121 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1123 /* Let's see if we had a chance at initialization first */
1124 if (svc->srv_kobj.kset) {
1125 kobject_put(&svc->srv_kobj);
1126 wait_for_completion(&svc->srv_kobj_unregister);
1130 int ptlrpc_sysfs_register_service(struct kset *parent,
1131 struct ptlrpc_service *svc)
1135 svc->srv_kobj.kset = parent;
1136 init_completion(&svc->srv_kobj_unregister);
1137 rc = kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype, NULL,
1138 "%s", svc->srv_name);
1143 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1144 struct ptlrpc_service *svc)
1146 struct lprocfs_vars lproc_vars[] = {
1147 { .name = "req_buffer_history_len",
1148 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1150 { .name = "req_buffer_history_max",
1151 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1153 { .name = "timeouts",
1154 .fops = &ptlrpc_lprocfs_timeouts_fops,
1156 { .name = "nrs_policies",
1157 .fops = &ptlrpc_lprocfs_nrs_fops,
1161 static struct file_operations req_history_fops = {
1162 .owner = THIS_MODULE,
1163 .open = ptlrpc_lprocfs_svc_req_history_open,
1165 .llseek = seq_lseek,
1166 .release = lprocfs_seq_release,
1171 ptlrpc_lprocfs_register(entry, svc->srv_name,
1172 "stats", &svc->srv_procroot,
1174 if (svc->srv_procroot == NULL)
1177 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1179 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1180 0400, &req_history_fops, svc);
1182 CWARN("Error adding the req_history file\n");
1185 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1187 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1188 &obddev->obd_svc_procroot,
1189 &obddev->obd_svc_stats);
1191 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1193 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1195 struct lprocfs_stats *svc_stats;
1196 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1197 int opc = opcode_offset(op);
1199 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1200 if (svc_stats == NULL || opc <= 0)
1202 LASSERT(opc < LUSTRE_MAX_OPCODES);
1203 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1204 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1207 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1209 struct lprocfs_stats *svc_stats;
1212 if (!req->rq_import)
1214 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1217 idx = lustre_msg_get_opc(req->rq_reqmsg);
1220 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1223 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1226 LASSERTF(0, "unsupported opcode %u\n", idx);
1230 lprocfs_counter_add(svc_stats, idx, bytes);
1233 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1235 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1237 if (svc->srv_procroot != NULL)
1238 lprocfs_remove(&svc->srv_procroot);
1241 lprocfs_free_stats(&svc->srv_stats);
1244 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1246 if (obd->obd_svc_procroot)
1247 lprocfs_remove(&obd->obd_svc_procroot);
1249 if (obd->obd_svc_stats)
1250 lprocfs_free_stats(&obd->obd_svc_stats);
1252 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1255 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1256 size_t count, loff_t *off)
1258 struct seq_file *m = file->private_data;
1259 struct obd_device *obd = m->private;
1260 struct ptlrpc_request *req;
1264 LPROCFS_CLIMP_CHECK(obd);
1265 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1266 LPROCFS_CLIMP_EXIT(obd);
1270 req->rq_send_state = LUSTRE_IMP_FULL;
1272 rc = ptlrpc_queue_wait(req);
1274 ptlrpc_req_finished(req);
1279 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1281 /* Write the connection UUID to this file to attempt to connect to that node.
1282 * The connection UUID is a node's primary NID. For example,
1283 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1286 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1287 size_t count, loff_t *off)
1289 struct seq_file *m = file->private_data;
1290 struct obd_device *obd = m->private;
1291 struct obd_import *imp = obd->u.cli.cl_import;
1296 const char prefix[] = "connection=";
1297 const int prefix_len = sizeof(prefix) - 1;
1299 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1302 OBD_ALLOC(kbuf, count + 1);
1306 if (copy_from_user(kbuf, buffer, count))
1307 GOTO(out, count = -EFAULT);
1311 /* only support connection=uuid::instance now */
1312 if (strncmp(prefix, kbuf, prefix_len) != 0)
1313 GOTO(out, count = -EINVAL);
1315 uuid = kbuf + prefix_len;
1316 ptr = strstr(uuid, "::");
1323 ptr += 2; /* Skip :: */
1324 inst = simple_strtol(ptr, &endptr, 10);
1326 CERROR("config: wrong instance # %s\n", ptr);
1327 } else if (inst != imp->imp_connect_data.ocd_instance) {
1328 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1329 "target(%u/%u), reconnecting...\n",
1330 imp->imp_obd->obd_name,
1331 imp->imp_connect_data.ocd_instance, inst);
1334 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1336 imp->imp_obd->obd_name, inst);
1341 ptlrpc_recover_import(imp, uuid, 1);
1344 OBD_FREE(kbuf, count + 1);
1347 EXPORT_SYMBOL(lprocfs_import_seq_write);
1349 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1351 struct obd_device *obd = m->private;
1352 struct obd_import *imp = obd->u.cli.cl_import;
1354 LPROCFS_CLIMP_CHECK(obd);
1355 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1356 LPROCFS_CLIMP_EXIT(obd);
1359 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1362 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1363 size_t count, loff_t *off)
1365 struct seq_file *m = file->private_data;
1366 struct obd_device *obd = m->private;
1367 struct client_obd *cli = &obd->u.cli;
1368 struct obd_import *imp = cli->cl_import;
1372 rc = lprocfs_str_to_s64(buffer, count, &val);
1376 if (val != 0 && val != 1)
1379 LPROCFS_CLIMP_CHECK(obd);
1380 spin_lock(&imp->imp_lock);
1381 imp->imp_no_pinger_recover = !val;
1382 spin_unlock(&imp->imp_lock);
1383 LPROCFS_CLIMP_EXIT(obd);
1386 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1388 #endif /* CONFIG_PROC_FS */