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, 2017, 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 { OST_FALLOCATE, "ost_fallocate"},
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 { MDS_RMFID, "mds_rmfid" },
100 { LDLM_ENQUEUE, "ldlm_enqueue" },
101 { LDLM_CONVERT, "ldlm_convert" },
102 { LDLM_CANCEL, "ldlm_cancel" },
103 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
104 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
105 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
106 { LDLM_SET_INFO, "ldlm_set_info" },
107 { MGS_CONNECT, "mgs_connect" },
108 { MGS_DISCONNECT, "mgs_disconnect" },
109 { MGS_EXCEPTION, "mgs_exception" },
110 { MGS_TARGET_REG, "mgs_target_reg" },
111 { MGS_TARGET_DEL, "mgs_target_del" },
112 { MGS_SET_INFO, "mgs_set_info" },
113 { MGS_CONFIG_READ, "mgs_config_read" },
114 { OBD_PING, "obd_ping" },
115 { 401, /* was OBD_LOG_CANCEL */ "llog_cancel" },
116 { 402, /* was OBD_QC_CALLBACK */ "obd_quota_callback" },
117 { OBD_IDX_READ, "dt_index_read" },
118 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
119 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
120 { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
121 { 504, /*LLOG_ORIGIN_HANDLE_WRITE_REC*/"llog_origin_handle_write_rec" },
122 { 505, /* was LLOG_ORIGIN_HANDLE_CLOSE */ "llog_origin_handle_close" },
123 { 506, /* was LLOG_ORIGIN_CONNECT */ "llog_origin_connect" },
124 { 507, /* was LLOG_CATINFO */ "llog_catinfo" },
125 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
126 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
127 { QUOTA_DQACQ, "quota_acquire" },
128 { QUOTA_DQREL, "quota_release" },
129 { SEQ_QUERY, "seq_query" },
130 { SEC_CTX_INIT, "sec_ctx_init" },
131 { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
132 { SEC_CTX_FINI, "sec_ctx_fini" },
133 { FLD_QUERY, "fld_query" },
134 { FLD_READ, "fld_read" },
135 { OUT_UPDATE, "out_update" },
136 { LFSCK_NOTIFY, "lfsck_notify" },
137 { LFSCK_QUERY, "lfsck_query" },
140 static struct ll_eopcode {
143 } ll_eopcode_table[EXTRA_LAST_OPC] = {
144 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
145 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
146 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
147 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
148 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
149 { MDS_REINT_SETATTR, "mds_reint_setattr" },
150 { MDS_REINT_CREATE, "mds_reint_create" },
151 { MDS_REINT_LINK, "mds_reint_link" },
152 { MDS_REINT_UNLINK, "mds_reint_unlink" },
153 { MDS_REINT_RENAME, "mds_reint_rename" },
154 { MDS_REINT_OPEN, "mds_reint_open" },
155 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
156 { MDS_REINT_RESYNC, "mds_reint_resync" },
157 { BRW_READ_BYTES, "read_bytes" },
158 { BRW_WRITE_BYTES, "write_bytes" },
161 const char *ll_opcode2str(__u32 opcode)
163 /* When one of the assertions below fail, chances are that:
164 * 1) A new opcode was added in include/lustre/lustre_idl.h,
165 * but is missing from the table above.
166 * or 2) The opcode space was renumbered or rearranged,
167 * and the opcode_offset() function in
168 * ptlrpc_internal.h needs to be modified.
170 __u32 offset = opcode_offset(opcode);
171 LASSERTF(offset < LUSTRE_MAX_OPCODES,
172 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
173 offset, LUSTRE_MAX_OPCODES);
174 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
175 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
176 offset, ll_rpc_opcode_table[offset].opcode, opcode);
177 return ll_rpc_opcode_table[offset].opname;
180 const int ll_str2opcode(const char *ops)
184 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
185 if (ll_rpc_opcode_table[i].opname != NULL &&
186 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
187 return ll_rpc_opcode_table[i].opcode;
193 static const char *ll_eopcode2str(__u32 opcode)
195 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
196 return ll_eopcode_table[opcode].opname;
200 ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
201 struct dentry **debugfs_root_ret,
202 struct lprocfs_stats **stats_ret)
204 struct dentry *svc_debugfs_entry;
205 struct lprocfs_stats *svc_stats;
207 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
210 LASSERT(!*debugfs_root_ret);
211 LASSERT(!*stats_ret);
213 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
219 svc_debugfs_entry = debugfs_create_dir(dir, root);
221 svc_debugfs_entry = root;
223 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
224 svc_counter_config, "req_waittime", "usec");
225 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
226 svc_counter_config, "req_qdepth", "reqs");
227 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
228 svc_counter_config, "req_active", "reqs");
229 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
230 svc_counter_config, "req_timeout", "sec");
231 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
232 svc_counter_config, "reqbuf_avail", "bufs");
233 for (i = 0; i < EXTRA_LAST_OPC; i++) {
237 case BRW_WRITE_BYTES:
245 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
247 ll_eopcode2str(i), units);
249 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
250 __u32 opcode = ll_rpc_opcode_table[i].opcode;
251 lprocfs_counter_init(svc_stats,
252 EXTRA_MAX_OPCODES + i, svc_counter_config,
253 ll_opcode2str(opcode), "usec");
256 debugfs_create_file(name, 0644, svc_debugfs_entry, svc_stats,
257 &ldebugfs_stats_seq_fops);
260 *debugfs_root_ret = svc_debugfs_entry;
261 *stats_ret = svc_stats;
265 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
267 struct ptlrpc_service *svc = m->private;
268 struct ptlrpc_service_part *svcpt;
272 ptlrpc_service_for_each_part(svcpt, i, svc)
273 total += svcpt->scp_hist_nrqbds;
275 seq_printf(m, "%d\n", total);
280 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
283 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
285 struct ptlrpc_service *svc = m->private;
286 struct ptlrpc_service_part *svcpt;
290 ptlrpc_service_for_each_part(svcpt, i, svc)
291 total += svc->srv_hist_nrqbds_cpt_max;
293 seq_printf(m, "%d\n", total);
298 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
299 const char __user *buffer,
300 size_t count, loff_t *off)
302 struct seq_file *m = file->private_data;
303 struct ptlrpc_service *svc = m->private;
304 unsigned long long val;
305 unsigned long long limit;
309 rc = kstrtoull_from_user(buffer, count, 0, &val);
313 if (val < 0 || val > INT_MAX)
316 /* This sanity check is more of an insanity check; we can still
317 * hose a kernel by allowing the request history to grow too
318 * far. The roundup to the next power of two is an empirical way
319 * to take care that request buffer is allocated in Slab and thus
320 * will be upgraded */
321 bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
323 limit = cfs_totalram_pages() / (2 * bufpages);
324 /* do not allow history to consume more than half max number of rqbds */
325 if ((svc->srv_nrqbds_max == 0 && val > limit) ||
326 (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
329 spin_lock(&svc->srv_lock);
332 svc->srv_hist_nrqbds_cpt_max = 0;
334 svc->srv_hist_nrqbds_cpt_max =
335 max(1, ((int)val / svc->srv_ncpts));
337 spin_unlock(&svc->srv_lock);
342 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
345 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
347 struct ptlrpc_service *svc = m->private;
349 seq_printf(m, "%d\n", svc->srv_nrqbds_max);
354 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
355 const char __user *buffer,
356 size_t count, loff_t *off)
358 struct seq_file *m = file->private_data;
359 struct ptlrpc_service *svc = m->private;
363 rc = kstrtoint_from_user(buffer, count, 0, &val);
367 if (val < svc->srv_nbuf_per_group && val != 0)
370 spin_lock(&svc->srv_lock);
372 svc->srv_nrqbds_max = (uint)val;
374 spin_unlock(&svc->srv_lock);
379 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
381 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
384 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
387 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
390 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
391 const char *buffer, size_t count)
393 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
398 rc = kstrtoul(buffer, 10, &val);
402 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
405 spin_lock(&svc->srv_lock);
406 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
407 spin_unlock(&svc->srv_lock);
411 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
413 spin_unlock(&svc->srv_lock);
417 LUSTRE_RW_ATTR(threads_min);
419 static ssize_t threads_started_show(struct kobject *kobj,
420 struct attribute *attr,
423 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
425 struct ptlrpc_service_part *svcpt;
429 ptlrpc_service_for_each_part(svcpt, i, svc)
430 total += svcpt->scp_nthrs_running;
432 return sprintf(buf, "%d\n", total);
434 LUSTRE_RO_ATTR(threads_started);
436 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
439 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
442 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
445 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
446 const char *buffer, size_t count)
448 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
453 rc = kstrtoul(buffer, 10, &val);
457 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
460 spin_lock(&svc->srv_lock);
461 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
462 spin_unlock(&svc->srv_lock);
466 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
468 spin_unlock(&svc->srv_lock);
472 LUSTRE_RW_ATTR(threads_max);
475 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
477 * \param[in] state The policy state
479 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
484 case NRS_POL_STATE_INVALID:
486 case NRS_POL_STATE_STOPPED:
488 case NRS_POL_STATE_STOPPING:
490 case NRS_POL_STATE_STARTING:
492 case NRS_POL_STATE_STARTED:
498 * Obtains status information for \a policy.
500 * Information is copied in \a info.
502 * \param[in] policy The policy
503 * \param[out] info Holds returned status information
505 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
506 struct ptlrpc_nrs_pol_info *info)
508 LASSERT(policy != NULL);
509 LASSERT(info != NULL);
510 assert_spin_locked(&policy->pol_nrs->nrs_lock);
512 BUILD_BUG_ON(sizeof(info->pi_arg) != sizeof(policy->pol_arg));
513 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
514 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
516 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
517 info->pi_state = policy->pol_state;
519 * XXX: These are accessed without holding
520 * ptlrpc_service_part::scp_req_lock.
522 info->pi_req_queued = policy->pol_req_queued;
523 info->pi_req_started = policy->pol_req_started;
527 * Reads and prints policy status information for all policies of a PTLRPC
530 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
532 struct ptlrpc_service *svc = m->private;
533 struct ptlrpc_service_part *svcpt;
534 struct ptlrpc_nrs *nrs;
535 struct ptlrpc_nrs_policy *policy;
536 struct ptlrpc_nrs_pol_info *infos;
537 struct ptlrpc_nrs_pol_info tmp;
539 unsigned pol_idx = 0;
546 * Serialize NRS core lprocfs operations with policy registration/
549 mutex_lock(&nrs_core.nrs_mutex);
552 * Use the first service partition's regular NRS head in order to obtain
553 * the number of policies registered with NRS heads of this service. All
554 * service partitions will have the same number of policies.
556 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
558 spin_lock(&nrs->nrs_lock);
559 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
560 spin_unlock(&nrs->nrs_lock);
562 OBD_ALLOC_PTR_ARRAY(infos, num_pols);
564 GOTO(out, rc = -ENOMEM);
567 ptlrpc_service_for_each_part(svcpt, i, svc) {
568 nrs = nrs_svcpt2nrs(svcpt, hp);
569 spin_lock(&nrs->nrs_lock);
573 list_for_each_entry(policy, &nrs->nrs_policy_list,
575 LASSERT(pol_idx < num_pols);
577 nrs_policy_get_info_locked(policy, &tmp);
579 * Copy values when handling the first service
583 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
585 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
587 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
588 sizeof(tmp.pi_state));
589 infos[pol_idx].pi_fallback = tmp.pi_fallback;
591 * For the rest of the service partitions
592 * sanity-check the values we get.
595 LASSERT(strncmp(infos[pol_idx].pi_name,
597 NRS_POL_NAME_MAX) == 0);
598 LASSERT(strncmp(infos[pol_idx].pi_arg,
600 sizeof(tmp.pi_arg)) == 0);
602 * Not asserting ptlrpc_nrs_pol_info::pi_state,
603 * because it may be different between
604 * instances of the same policy in different
605 * service partitions.
607 LASSERT(infos[pol_idx].pi_fallback ==
611 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
612 infos[pol_idx].pi_req_started += tmp.pi_req_started;
616 spin_unlock(&nrs->nrs_lock);
620 * Policy status information output is in YAML format.
636 * high_priority_requests:
649 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
650 "high_priority_requests:");
652 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
653 if (strlen(infos[pol_idx].pi_arg) > 0)
654 seq_printf(m, " - name: %s %s\n",
655 infos[pol_idx].pi_name,
656 infos[pol_idx].pi_arg);
658 seq_printf(m, " - name: %s\n",
659 infos[pol_idx].pi_name);
662 seq_printf(m, " state: %s\n"
665 " active: %-20d\n\n",
666 nrs_state2str(infos[pol_idx].pi_state),
667 infos[pol_idx].pi_fallback ? "yes" : "no",
668 (int)infos[pol_idx].pi_req_queued,
669 (int)infos[pol_idx].pi_req_started);
672 if (!hp && nrs_svc_has_hp(svc)) {
673 memset(infos, 0, num_pols * sizeof(*infos));
676 * Redo the processing for the service's HP NRS heads' policies.
684 OBD_FREE_PTR_ARRAY(infos, num_pols);
686 mutex_unlock(&nrs_core.nrs_mutex);
692 #define LPROCFS_NRS_WR_MAX_ARG (1024)
694 * The longest valid command string is the maxium policy name size, plus the
695 * length of the " reg" substring, plus the lenght of argument
697 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
698 + LPROCFS_NRS_WR_MAX_ARG)
701 * Starts and stops a given policy on a PTLRPC service.
703 * Commands consist of the policy name, followed by an optional [reg|hp] token;
704 * if the optional token is omitted, the operation is performed on both the
705 * regular and high-priority (if the service has one) NRS head.
708 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
709 size_t count, loff_t *off)
711 struct seq_file *m = file->private_data;
712 struct ptlrpc_service *svc = m->private;
713 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
715 char *cmd_copy = NULL;
721 if (count >= LPROCFS_NRS_WR_MAX_CMD)
722 GOTO(out, rc = -EINVAL);
724 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
726 GOTO(out, rc = -ENOMEM);
728 * strsep() modifies its argument, so keep a copy
732 if (copy_from_user(cmd, buffer, count))
733 GOTO(out, rc = -EFAULT);
737 policy_name = strsep(&cmd, " ");
739 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
740 GOTO(out, rc = -EINVAL);
743 * No [reg|hp] token has been specified
748 queue_name = strsep(&cmd, " ");
750 * The second token is either an optional [reg|hp] string,
753 if (strcmp(queue_name, "reg") == 0)
754 queue = PTLRPC_NRS_QUEUE_REG;
755 else if (strcmp(queue_name, "hp") == 0)
756 queue = PTLRPC_NRS_QUEUE_HP;
765 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
766 GOTO(out, rc = -ENODEV);
767 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
768 queue = PTLRPC_NRS_QUEUE_REG;
771 * Serialize NRS core lprocfs operations with policy registration/
774 mutex_lock(&nrs_core.nrs_mutex);
776 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
777 PTLRPC_NRS_CTL_START,
780 mutex_unlock(&nrs_core.nrs_mutex);
783 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
785 RETURN(rc < 0 ? rc : count);
788 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs);
792 struct ptlrpc_srh_iterator {
795 struct ptlrpc_request *srhi_req;
799 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
800 struct ptlrpc_srh_iterator *srhi,
804 struct ptlrpc_request *req;
806 if (srhi->srhi_req != NULL &&
807 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
808 srhi->srhi_seq <= seq) {
809 /* If srhi_req was set previously, hasn't been culled and
810 * we're searching for a seq on or after it (i.e. more
811 * recent), search from it onwards.
812 * Since the service history is LRU (i.e. culled reqs will
813 * be near the head), we shouldn't have to do long
815 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
816 "%s:%d: seek seq %llu, request seq %llu\n",
817 svcpt->scp_service->srv_name, svcpt->scp_cpt,
818 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
819 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
820 "%s:%d: seek offset %llu, request seq %llu, "
821 "last culled %llu\n",
822 svcpt->scp_service->srv_name, svcpt->scp_cpt,
823 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
824 e = &srhi->srhi_req->rq_history_list;
826 /* search from start */
827 e = svcpt->scp_hist_reqs.next;
830 while (e != &svcpt->scp_hist_reqs) {
831 req = list_entry(e, struct ptlrpc_request, rq_history_list);
833 if (req->rq_history_seq >= seq) {
834 srhi->srhi_seq = req->rq_history_seq;
835 srhi->srhi_req = req;
845 * ptlrpc history sequence is used as "position" of seq_file, in some case,
846 * seq_read() will increase "position" to indicate reading the next
847 * element, however, low bits of history sequence are reserved for CPT id
848 * (check the details from comments before ptlrpc_req_add_history), which
849 * means seq_read() might change CPT id of history sequence and never
850 * finish reading of requests on a CPT. To make it work, we have to shift
851 * CPT id to high bits and timestamp to low bits, so seq_read() will only
852 * increase timestamp which can correctly indicate the next position.
855 /* convert seq_file pos to cpt */
856 #define PTLRPC_REQ_POS2CPT(svc, pos) \
857 ((svc)->srv_cpt_bits == 0 ? 0 : \
858 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
860 /* make up seq_file pos from cpt */
861 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
862 ((svc)->srv_cpt_bits == 0 ? 0 : \
863 (cpt) << (64 - (svc)->srv_cpt_bits))
865 /* convert sequence to position */
866 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
867 ((svc)->srv_cpt_bits == 0 ? (seq) : \
868 ((seq) >> (svc)->srv_cpt_bits) | \
869 ((seq) << (64 - (svc)->srv_cpt_bits)))
871 /* convert position to sequence */
872 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
873 ((svc)->srv_cpt_bits == 0 ? (pos) : \
874 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
875 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
878 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
880 struct ptlrpc_service *svc = s->private;
881 struct ptlrpc_service_part *svcpt;
882 struct ptlrpc_srh_iterator *srhi;
887 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
888 CWARN("Failed to read request history because size of loff_t "
889 "%d can't match size of u64\n", (int)sizeof(loff_t));
893 OBD_ALLOC(srhi, sizeof(*srhi));
898 srhi->srhi_req = NULL;
900 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
902 ptlrpc_service_for_each_part(svcpt, i, svc) {
903 if (i < cpt) /* skip */
905 if (i > cpt) /* make up the lowest position for this CPT */
906 *pos = PTLRPC_REQ_CPT2POS(svc, i);
908 mutex_lock(&svcpt->scp_mutex);
909 spin_lock(&svcpt->scp_lock);
910 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
911 PTLRPC_REQ_POS2SEQ(svc, *pos));
912 spin_unlock(&svcpt->scp_lock);
913 mutex_unlock(&svcpt->scp_mutex);
915 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
921 OBD_FREE(srhi, sizeof(*srhi));
926 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
928 struct ptlrpc_srh_iterator *srhi = iter;
931 OBD_FREE(srhi, sizeof(*srhi));
935 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
936 void *iter, loff_t *pos)
938 struct ptlrpc_service *svc = s->private;
939 struct ptlrpc_srh_iterator *srhi = iter;
940 struct ptlrpc_service_part *svcpt;
945 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
946 svcpt = svc->srv_parts[i];
948 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
949 srhi->srhi_req = NULL;
950 seq = srhi->srhi_seq = 0;
951 } else { /* the next sequence */
952 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
955 mutex_lock(&svcpt->scp_mutex);
956 spin_lock(&svcpt->scp_lock);
957 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
958 spin_unlock(&svcpt->scp_lock);
959 mutex_unlock(&svcpt->scp_mutex);
961 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
967 OBD_FREE(srhi, sizeof(*srhi));
971 /* common ost/mdt so_req_printer */
972 void target_print_req(void *seq_file, struct ptlrpc_request *req)
974 /* Called holding srv_lock with irqs disabled.
975 * Print specific req contents and a newline.
976 * CAVEAT EMPTOR: check request message length before printing!!!
977 * You might have received any old crap so you must be just as
978 * careful here as the service's request parser!!! */
979 struct seq_file *sf = seq_file;
981 switch (req->rq_phase) {
983 /* still awaiting a service thread's attention, or rejected
984 * because the generic request message didn't unpack */
985 seq_printf(sf, "<not swabbed>\n");
987 case RQ_PHASE_INTERPRET:
988 /* being handled, so basic msg swabbed, and opc is valid
989 * but racing with mds_handle() */
990 case RQ_PHASE_COMPLETE:
991 /* been handled by mds_handle() reply state possibly still
993 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
996 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
999 EXPORT_SYMBOL(target_print_req);
1001 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1003 struct ptlrpc_service *svc = s->private;
1004 struct ptlrpc_srh_iterator *srhi = iter;
1005 struct ptlrpc_service_part *svcpt;
1006 struct ptlrpc_request *req;
1009 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1011 svcpt = svc->srv_parts[srhi->srhi_idx];
1013 mutex_lock(&svcpt->scp_mutex);
1014 spin_lock(&svcpt->scp_lock);
1016 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1019 struct timespec64 arrival, sent, arrivaldiff;
1020 char nidstr[LNET_NIDSTR_SIZE];
1022 req = srhi->srhi_req;
1024 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1025 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1026 sent.tv_sec = req->rq_sent;
1028 arrivaldiff = timespec64_sub(sent, arrival);
1030 /* Print common req fields.
1031 * CAVEAT EMPTOR: we're racing with the service handler
1032 * here. The request could contain any old crap, so you
1033 * must be just as careful as the service's request
1034 * parser. Currently I only print stuff here I know is OK
1035 * to look at coz it was set up in request_in_callback()!!!
1038 "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1039 req->rq_history_seq,
1040 req->rq_export && req->rq_export->exp_obd ?
1041 req->rq_export->exp_obd->obd_name :
1042 libcfs_nid2str_r(req->rq_self, nidstr,
1044 libcfs_id2str(req->rq_peer), req->rq_xid,
1045 req->rq_reqlen, ptlrpc_rqphase2str(req),
1046 (s64)req->rq_arrival_time.tv_sec,
1047 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1048 (s64)arrivaldiff.tv_sec,
1049 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1050 (s64)(req->rq_sent - req->rq_deadline));
1051 if (svc->srv_ops.so_req_printer == NULL)
1052 seq_printf(s, "\n");
1054 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1057 spin_unlock(&svcpt->scp_lock);
1058 mutex_unlock(&svcpt->scp_mutex);
1064 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1066 static struct seq_operations sops = {
1067 .start = ptlrpc_lprocfs_svc_req_history_start,
1068 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1069 .next = ptlrpc_lprocfs_svc_req_history_next,
1070 .show = ptlrpc_lprocfs_svc_req_history_show,
1072 struct seq_file *seqf;
1075 rc = seq_open(file, &sops);
1079 seqf = file->private_data;
1080 seqf->private = inode->i_private;
1084 /* See also lprocfs_rd_timeouts */
1085 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1087 struct ptlrpc_service *svc = m->private;
1088 struct ptlrpc_service_part *svcpt;
1089 time64_t worst_timestamp;
1090 timeout_t cur_timeout;
1091 timeout_t worst_timeout;
1095 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1100 ptlrpc_service_for_each_part(svcpt, i, svc) {
1101 cur_timeout = at_get(&svcpt->scp_at_estimate);
1102 worst_timeout = svcpt->scp_at_estimate.at_worst_timeout_ever;
1103 worst_timestamp = svcpt->scp_at_estimate.at_worst_timestamp;
1105 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1106 "service", cur_timeout, worst_timeout,
1108 ktime_get_real_seconds() - worst_timestamp);
1110 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1116 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1118 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1119 struct attribute *attr,
1122 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1125 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1128 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1129 struct attribute *attr,
1133 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1138 rc = kstrtoul(buffer, 10, &val);
1142 spin_lock(&svc->srv_lock);
1143 svc->srv_hpreq_ratio = val;
1144 spin_unlock(&svc->srv_lock);
1148 LUSTRE_RW_ATTR(high_priority_ratio);
1150 static struct attribute *ptlrpc_svc_attrs[] = {
1151 &lustre_attr_threads_min.attr,
1152 &lustre_attr_threads_started.attr,
1153 &lustre_attr_threads_max.attr,
1154 &lustre_attr_high_priority_ratio.attr,
1158 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1160 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1163 complete(&svc->srv_kobj_unregister);
1166 static struct kobj_type ptlrpc_svc_ktype = {
1167 .default_attrs = ptlrpc_svc_attrs,
1168 .sysfs_ops = &lustre_sysfs_ops,
1169 .release = ptlrpc_sysfs_svc_release,
1172 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1174 /* Let's see if we had a chance at initialization first */
1175 if (svc->srv_kobj.kset) {
1176 kobject_put(&svc->srv_kobj);
1177 wait_for_completion(&svc->srv_kobj_unregister);
1181 int ptlrpc_sysfs_register_service(struct kset *parent,
1182 struct ptlrpc_service *svc)
1184 svc->srv_kobj.kset = parent;
1185 init_completion(&svc->srv_kobj_unregister);
1186 return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1187 &parent->kobj, "%s", svc->srv_name);
1190 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1191 struct ptlrpc_service *svc)
1193 struct ldebugfs_vars ldebugfs_vars[] = {
1194 { .name = "req_buffer_history_len",
1195 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1197 { .name = "req_buffer_history_max",
1198 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1200 { .name = "timeouts",
1201 .fops = &ptlrpc_lprocfs_timeouts_fops,
1203 { .name = "nrs_policies",
1204 .fops = &ptlrpc_lprocfs_nrs_fops,
1206 { .name = "req_buffers_max",
1207 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1211 static struct file_operations req_history_fops = {
1212 .owner = THIS_MODULE,
1213 .open = ptlrpc_lprocfs_svc_req_history_open,
1215 .llseek = seq_lseek,
1216 .release = lprocfs_seq_release,
1219 ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1220 &svc->srv_debugfs_entry, &svc->srv_stats);
1221 if (!svc->srv_debugfs_entry)
1224 ldebugfs_add_vars(svc->srv_debugfs_entry, ldebugfs_vars, NULL);
1226 debugfs_create_file("req_history", 0400, svc->srv_debugfs_entry, svc,
1230 void ptlrpc_lprocfs_register_obd(struct obd_device *obd)
1232 ptlrpc_ldebugfs_register(obd->obd_debugfs_entry, NULL, "stats",
1233 &obd->obd_svc_debugfs_entry,
1234 &obd->obd_svc_stats);
1236 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1238 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1240 struct lprocfs_stats *svc_stats;
1241 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1242 int opc = opcode_offset(op);
1244 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1245 if (svc_stats == NULL || opc <= 0)
1247 LASSERT(opc < LUSTRE_MAX_OPCODES);
1248 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1249 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1252 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1254 struct lprocfs_stats *svc_stats;
1257 if (!req->rq_import)
1259 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1262 idx = lustre_msg_get_opc(req->rq_reqmsg);
1265 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1268 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1271 LASSERTF(0, "unsupported opcode %u\n", idx);
1275 lprocfs_counter_add(svc_stats, idx, bytes);
1278 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1280 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1282 debugfs_remove_recursive(svc->srv_debugfs_entry);
1285 lprocfs_free_stats(&svc->srv_stats);
1288 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1290 /* cleanup first to allow concurrent access to device's
1291 * stats via debugfs to complete safely
1293 lprocfs_obd_cleanup(obd);
1295 debugfs_remove_recursive(obd->obd_svc_debugfs_entry);
1297 if (obd->obd_svc_stats)
1298 lprocfs_free_stats(&obd->obd_svc_stats);
1300 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1302 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1305 struct obd_device *obd = container_of(kobj, struct obd_device,
1307 struct obd_import *imp;
1308 struct ptlrpc_request *req;
1312 with_imp_locked(obd, imp, rc)
1313 req = ptlrpc_prep_ping(imp);
1320 req->rq_send_state = LUSTRE_IMP_FULL;
1322 rc = ptlrpc_queue_wait(req);
1323 ptlrpc_req_finished(req);
1327 EXPORT_SYMBOL(ping_show);
1329 /* kept for older verison of tools. */
1330 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1331 const char *buffer, size_t count)
1333 int rc = ping_show(kobj, attr, (char *)buffer);
1335 return (rc < 0) ? rc : count;
1337 EXPORT_SYMBOL(ping_store);
1339 /* Write the connection UUID to this file to attempt to connect to that node.
1340 * The connection UUID is a node's primary NID. For example,
1341 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1344 ldebugfs_import_seq_write(struct file *file, const char __user *buffer,
1345 size_t count, loff_t *off)
1347 struct seq_file *m = file->private_data;
1348 struct obd_device *obd = m->private;
1349 struct obd_import *imp;
1354 const char prefix[] = "connection=";
1355 const int prefix_len = sizeof(prefix) - 1;
1358 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1361 OBD_ALLOC(kbuf, count + 1);
1365 if (copy_from_user(kbuf, buffer, count))
1366 GOTO(out, rc = -EFAULT);
1370 /* only support connection=uuid::instance now */
1371 if (strncmp(prefix, kbuf, prefix_len) != 0)
1372 GOTO(out, rc = -EINVAL);
1374 with_imp_locked(obd, imp, rc) {
1375 uuid = kbuf + prefix_len;
1376 ptr = strstr(uuid, "::");
1383 ptr += 2; /* Skip :: */
1384 rc = kstrtouint(ptr, 10, &inst);
1386 CERROR("config: wrong instance # %s\n", ptr);
1387 } else if (inst != imp->imp_connect_data.ocd_instance) {
1389 "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1390 imp->imp_obd->obd_name,
1391 imp->imp_connect_data.ocd_instance,
1396 "IR: %s has already been connecting to "
1398 imp->imp_obd->obd_name, inst);
1403 ptlrpc_recover_import(imp, uuid, 1);
1407 OBD_FREE(kbuf, count + 1);
1410 EXPORT_SYMBOL(ldebugfs_import_seq_write);
1412 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1414 struct obd_device *obd = m->private;
1415 struct obd_import *imp;
1418 with_imp_locked(obd, imp, rc)
1419 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1423 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1426 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1427 size_t count, loff_t *off)
1429 struct seq_file *m = file->private_data;
1430 struct obd_device *obd = m->private;
1431 struct obd_import *imp;
1435 rc = kstrtobool_from_user(buffer, count, &val);
1439 with_imp_locked(obd, imp, rc) {
1440 spin_lock(&imp->imp_lock);
1441 imp->imp_no_pinger_recover = !val;
1442 spin_unlock(&imp->imp_lock);
1447 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);