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/
31 #define DEBUG_SUBSYSTEM S_CLASS
34 #include <obd_support.h>
36 #include <lprocfs_status.h>
37 #include <lustre_net.h>
38 #include <obd_class.h>
39 #include "ptlrpc_internal.h"
42 static struct ll_rpc_opcode {
45 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
46 { OST_REPLY, "ost_reply" },
47 { OST_GETATTR, "ost_getattr" },
48 { OST_SETATTR, "ost_setattr" },
49 { OST_READ, "ost_read" },
50 { OST_WRITE, "ost_write" },
51 { OST_CREATE , "ost_create" },
52 { OST_DESTROY, "ost_destroy" },
53 { OST_GET_INFO, "ost_get_info" },
54 { OST_CONNECT, "ost_connect" },
55 { OST_DISCONNECT, "ost_disconnect" },
56 { OST_PUNCH, "ost_punch" },
57 { OST_OPEN, "ost_open" },
58 { OST_CLOSE, "ost_close" },
59 { OST_STATFS, "ost_statfs" },
60 { 14, NULL }, /* formerly OST_SAN_READ */
61 { 15, NULL }, /* formerly OST_SAN_WRITE */
62 { OST_SYNC, "ost_sync" },
63 { OST_SET_INFO, "ost_set_info" },
64 { OST_QUOTACHECK, "ost_quotacheck" },
65 { OST_QUOTACTL, "ost_quotactl" },
66 { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
67 { OST_LADVISE, "ost_ladvise" },
68 { OST_FALLOCATE, "ost_fallocate" },
69 { OST_SEEK, "ost_seek" },
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 { MDS_BATCH, "mds_batch" },
101 { LDLM_ENQUEUE, "ldlm_enqueue" },
102 { LDLM_CONVERT, "ldlm_convert" },
103 { LDLM_CANCEL, "ldlm_cancel" },
104 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
105 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
106 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
107 { LDLM_SET_INFO, "ldlm_set_info" },
108 { MGS_CONNECT, "mgs_connect" },
109 { MGS_DISCONNECT, "mgs_disconnect" },
110 { MGS_EXCEPTION, "mgs_exception" },
111 { MGS_TARGET_REG, "mgs_target_reg" },
112 { MGS_TARGET_DEL, "mgs_target_del" },
113 { MGS_SET_INFO, "mgs_set_info" },
114 { MGS_CONFIG_READ, "mgs_config_read" },
115 { OBD_PING, "obd_ping" },
116 { 401, /* was OBD_LOG_CANCEL */ "llog_cancel" },
117 { 402, /* was OBD_QC_CALLBACK */ "obd_quota_callback" },
118 { OBD_IDX_READ, "dt_index_read" },
119 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
120 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
121 { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
122 { 504, /*LLOG_ORIGIN_HANDLE_WRITE_REC*/"llog_origin_handle_write_rec" },
123 { 505, /* was LLOG_ORIGIN_HANDLE_CLOSE */ "llog_origin_handle_close" },
124 { 506, /* was LLOG_ORIGIN_CONNECT */ "llog_origin_connect" },
125 { 507, /* was LLOG_CATINFO */ "llog_catinfo" },
126 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
127 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
128 { QUOTA_DQACQ, "quota_acquire" },
129 { QUOTA_DQREL, "quota_release" },
130 { SEQ_QUERY, "seq_query" },
131 { SEC_CTX_INIT, "sec_ctx_init" },
132 { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
133 { SEC_CTX_FINI, "sec_ctx_fini" },
134 { FLD_QUERY, "fld_query" },
135 { FLD_READ, "fld_read" },
136 #ifdef HAVE_SERVER_SUPPORT
137 { OUT_UPDATE, "out_update" },
138 { LFSCK_NOTIFY, "lfsck_notify" },
139 { LFSCK_QUERY, "lfsck_query" },
143 static struct ll_eopcode {
146 } ll_eopcode_table[EXTRA_LAST_OPC] = {
147 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
148 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
149 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
150 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
151 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
152 { MDS_REINT_SETATTR, "mds_reint_setattr" },
153 { MDS_REINT_CREATE, "mds_reint_create" },
154 { MDS_REINT_LINK, "mds_reint_link" },
155 { MDS_REINT_UNLINK, "mds_reint_unlink" },
156 { MDS_REINT_RENAME, "mds_reint_rename" },
157 { MDS_REINT_OPEN, "mds_reint_open" },
158 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
159 { MDS_REINT_RESYNC, "mds_reint_resync" },
160 { BRW_READ_BYTES, "read_bytes" },
161 { BRW_WRITE_BYTES, "write_bytes" },
164 const char *ll_opcode2str(__u32 opcode)
166 __u32 offset = opcode_offset(opcode);
168 /* When one of the assertions below fail, chances are that:
169 * 1) A new opcode was added in include/lustre/lustre_idl.h,
170 * but is missing from the table above.
171 * or 2) The opcode space was renumbered or rearranged,
172 * and the opcode_offset() function in
173 * ptlrpc_internal.h needs to be modified.
175 LASSERTF(offset < LUSTRE_MAX_OPCODES,
176 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
177 offset, LUSTRE_MAX_OPCODES);
178 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
179 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
180 offset, ll_rpc_opcode_table[offset].opcode, opcode);
182 return ll_rpc_opcode_table[offset].opname;
185 const int ll_str2opcode(const char *ops)
189 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
190 if (ll_rpc_opcode_table[i].opname != NULL &&
191 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
192 return ll_rpc_opcode_table[i].opcode;
198 static const char *ll_eopcode2str(__u32 opcode)
200 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
201 return ll_eopcode_table[opcode].opname;
205 ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
206 struct dentry **debugfs_root_ret,
207 struct lprocfs_stats **stats_ret)
209 struct dentry *svc_debugfs_entry;
210 struct lprocfs_stats *svc_stats;
211 enum lprocfs_counter_config config = LPROCFS_CNTR_AVGMINMAX |
215 LASSERT(!*debugfs_root_ret);
216 LASSERT(!*stats_ret);
218 svc_stats = lprocfs_stats_alloc(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
224 svc_debugfs_entry = debugfs_create_dir(dir, root);
226 svc_debugfs_entry = root;
228 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
229 config | LPROCFS_TYPE_USECS, "req_waittime");
230 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
231 config | LPROCFS_TYPE_REQS, "req_qdepth");
232 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
233 config | LPROCFS_TYPE_REQS, "req_active");
234 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
235 config | LPROCFS_TYPE_SECS, "req_timeout");
236 lprocfs_counter_init_units(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
237 config, "reqbuf_avail", "bufs");
238 for (i = 0; i < EXTRA_LAST_OPC; i++) {
239 enum lprocfs_counter_config extra_type = LPROCFS_TYPE_REQS;
242 case BRW_WRITE_BYTES:
244 extra_type = LPROCFS_TYPE_BYTES;
247 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
248 config | extra_type, ll_eopcode2str(i));
250 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
251 __u32 opcode = ll_rpc_opcode_table[i].opcode;
253 lprocfs_counter_init(svc_stats, EXTRA_MAX_OPCODES + i,
254 config | LPROCFS_TYPE_USECS,
255 ll_opcode2str(opcode));
258 debugfs_create_file(name, 0644, svc_debugfs_entry, svc_stats,
259 &ldebugfs_stats_seq_fops);
262 *debugfs_root_ret = svc_debugfs_entry;
263 *stats_ret = svc_stats;
267 ptlrpc_lprocfs_req_buffer_history_len_seq_show(struct seq_file *m, void *v)
269 struct ptlrpc_service *svc = m->private;
270 struct ptlrpc_service_part *svcpt;
274 ptlrpc_service_for_each_part(svcpt, i, svc)
275 total += svcpt->scp_hist_nrqbds;
277 seq_printf(m, "%d\n", total);
282 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_buffer_history_len);
285 ptlrpc_lprocfs_req_buffer_history_max_seq_show(struct seq_file *m, void *n)
287 struct ptlrpc_service *svc = m->private;
288 struct ptlrpc_service_part *svcpt;
292 ptlrpc_service_for_each_part(svcpt, i, svc)
293 total += svc->srv_hist_nrqbds_cpt_max;
295 seq_printf(m, "%d\n", total);
300 ptlrpc_lprocfs_req_buffer_history_max_seq_write(struct file *file,
301 const char __user *buffer,
302 size_t count, loff_t *off)
304 struct seq_file *m = file->private_data;
305 struct ptlrpc_service *svc = m->private;
306 unsigned long long val;
307 unsigned long long limit;
311 rc = kstrtoull_from_user(buffer, count, 0, &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
320 * far. The roundup to the next power of two is an empirical way
321 * to take care that request buffer is allocated in Slab and thus
322 * will be upgraded */
323 bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
325 limit = cfs_totalram_pages() / (2 * bufpages);
326 /* do not allow history to consume more than half max number of rqbds */
327 if ((svc->srv_nrqbds_max == 0 && val > limit) ||
328 (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
331 spin_lock(&svc->srv_lock);
334 svc->srv_hist_nrqbds_cpt_max = 0;
336 svc->srv_hist_nrqbds_cpt_max =
337 max(1, ((int)val / svc->srv_ncpts));
339 spin_unlock(&svc->srv_lock);
344 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffer_history_max);
347 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
349 struct ptlrpc_service *svc = m->private;
351 seq_printf(m, "%d\n", svc->srv_nrqbds_max);
356 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
357 const char __user *buffer,
358 size_t count, loff_t *off)
360 struct seq_file *m = file->private_data;
361 struct ptlrpc_service *svc = m->private;
365 rc = kstrtoint_from_user(buffer, count, 0, &val);
369 if (val < svc->srv_nbuf_per_group && val != 0)
372 spin_lock(&svc->srv_lock);
374 svc->srv_nrqbds_max = (uint)val;
376 spin_unlock(&svc->srv_lock);
381 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
383 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
386 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
389 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
392 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
393 const char *buffer, size_t count)
395 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
400 rc = kstrtoul(buffer, 10, &val);
404 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
407 spin_lock(&svc->srv_lock);
408 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
409 spin_unlock(&svc->srv_lock);
413 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
415 spin_unlock(&svc->srv_lock);
419 LUSTRE_RW_ATTR(threads_min);
421 static ssize_t threads_started_show(struct kobject *kobj,
422 struct attribute *attr,
425 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
427 struct ptlrpc_service_part *svcpt;
431 ptlrpc_service_for_each_part(svcpt, i, svc)
432 total += svcpt->scp_nthrs_running;
434 return sprintf(buf, "%d\n", total);
436 LUSTRE_RO_ATTR(threads_started);
438 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
441 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
444 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
447 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
448 const char *buffer, size_t count)
450 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
455 rc = kstrtoul(buffer, 10, &val);
459 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
462 spin_lock(&svc->srv_lock);
463 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
464 spin_unlock(&svc->srv_lock);
468 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
470 spin_unlock(&svc->srv_lock);
474 LUSTRE_RW_ATTR(threads_max);
477 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
479 * \param[in] state The policy state
481 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
486 case NRS_POL_STATE_INVALID:
488 case NRS_POL_STATE_STOPPED:
490 case NRS_POL_STATE_STOPPING:
492 case NRS_POL_STATE_STARTING:
494 case NRS_POL_STATE_STARTED:
500 * Obtains status information for \a policy.
502 * Information is copied in \a info.
504 * \param[in] policy The policy
505 * \param[out] info Holds returned status information
507 static void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
508 struct ptlrpc_nrs_pol_info *info)
510 LASSERT(policy != NULL);
511 LASSERT(info != NULL);
512 assert_spin_locked(&policy->pol_nrs->nrs_lock);
514 BUILD_BUG_ON(sizeof(info->pi_arg) != sizeof(policy->pol_arg));
515 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
516 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
518 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
519 info->pi_state = policy->pol_state;
521 * XXX: These are accessed without holding
522 * ptlrpc_service_part::scp_req_lock.
524 info->pi_req_queued = policy->pol_req_queued;
525 info->pi_req_started = policy->pol_req_started;
529 * Reads and prints policy status information for all policies of a PTLRPC
532 static int ptlrpc_lprocfs_nrs_policies_seq_show(struct seq_file *m, void *n)
534 struct ptlrpc_service *svc = m->private;
535 struct ptlrpc_service_part *svcpt;
536 struct ptlrpc_nrs *nrs;
537 struct ptlrpc_nrs_policy *policy;
538 struct ptlrpc_nrs_pol_info *infos;
539 struct ptlrpc_nrs_pol_info tmp;
540 unsigned int num_pols;
541 unsigned int pol_idx = 0;
548 * Serialize NRS core lprocfs operations with policy registration/
551 mutex_lock(&nrs_core.nrs_mutex);
554 * Use the first service partition's regular NRS head in order to obtain
555 * the number of policies registered with NRS heads of this service. All
556 * service partitions will have the same number of policies.
558 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
560 spin_lock(&nrs->nrs_lock);
561 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
562 spin_unlock(&nrs->nrs_lock);
564 OBD_ALLOC_PTR_ARRAY(infos, num_pols);
566 GOTO(out, rc = -ENOMEM);
569 ptlrpc_service_for_each_part(svcpt, i, svc) {
570 nrs = nrs_svcpt2nrs(svcpt, hp);
571 spin_lock(&nrs->nrs_lock);
575 list_for_each_entry(policy, &nrs->nrs_policy_list,
577 LASSERT(pol_idx < num_pols);
579 nrs_policy_get_info_locked(policy, &tmp);
581 * Copy values when handling the first service
585 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
587 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
589 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
590 sizeof(tmp.pi_state));
591 infos[pol_idx].pi_fallback = tmp.pi_fallback;
593 * For the rest of the service partitions
594 * sanity-check the values we get.
597 if (strncmp(infos[pol_idx].pi_name,
599 NRS_POL_NAME_MAX) != 0) {
600 spin_unlock(&nrs->nrs_lock);
602 CERROR("%s: failed to check pi_name: rc = %d\n",
603 svc->srv_thread_name, rc);
606 if (strncmp(infos[pol_idx].pi_arg,
608 sizeof(tmp.pi_arg)) != 0) {
609 spin_unlock(&nrs->nrs_lock);
611 CERROR("%s: failed to check pi_arg: rc = %d\n",
612 svc->srv_thread_name, rc);
616 * Not checking ptlrpc_nrs_pol_info::pi_state,
617 * because it may be different between
618 * instances of the same policy in different
619 * service partitions.
622 if (infos[pol_idx].pi_fallback !=
624 spin_unlock(&nrs->nrs_lock);
626 CERROR("%s: failed to check pi_fallback: rc = %d\n",
627 svc->srv_thread_name, rc);
632 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
633 infos[pol_idx].pi_req_started += tmp.pi_req_started;
637 spin_unlock(&nrs->nrs_lock);
641 * Policy status information output is in YAML format.
657 * high_priority_requests:
670 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
671 "high_priority_requests:");
673 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
674 if (strlen(infos[pol_idx].pi_arg) > 0)
675 seq_printf(m, " - name: %s %s\n",
676 infos[pol_idx].pi_name,
677 infos[pol_idx].pi_arg);
679 seq_printf(m, " - name: %s\n",
680 infos[pol_idx].pi_name);
683 seq_printf(m, " state: %s\n"
686 " active: %-20d\n\n",
687 nrs_state2str(infos[pol_idx].pi_state),
688 infos[pol_idx].pi_fallback ? "yes" : "no",
689 (int)infos[pol_idx].pi_req_queued,
690 (int)infos[pol_idx].pi_req_started);
693 if (!hp && nrs_svc_has_hp(svc)) {
694 memset(infos, 0, num_pols * sizeof(*infos));
697 * Redo the processing for the service's HP NRS heads' policies.
705 OBD_FREE_PTR_ARRAY(infos, num_pols);
707 mutex_unlock(&nrs_core.nrs_mutex);
712 #define LPROCFS_NRS_WR_MAX_ARG (1024)
714 * The longest valid command string is the maxium policy name size, plus the
715 * length of the " reg" substring, plus the lenght of argument
717 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 + \
718 LPROCFS_NRS_WR_MAX_ARG)
721 * Starts and stops a given policy on a PTLRPC service.
723 * Commands consist of the policy name, followed by an optional [reg|hp] token;
724 * if the optional token is omitted, the operation is performed on both the
725 * regular and high-priority (if the service has one) NRS head.
728 ptlrpc_lprocfs_nrs_policies_seq_write(struct file *file,
729 const char __user *buffer,
730 size_t count, loff_t *off)
732 struct seq_file *m = file->private_data;
733 struct ptlrpc_service *svc = m->private;
734 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
736 char *cmd_copy = NULL;
742 if (count >= LPROCFS_NRS_WR_MAX_CMD)
743 GOTO(out, rc = -EINVAL);
745 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
747 GOTO(out, rc = -ENOMEM);
749 * strsep() modifies its argument, so keep a copy
753 if (copy_from_user(cmd, buffer, count))
754 GOTO(out, rc = -EFAULT);
758 policy_name = strsep(&cmd, " ");
760 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
761 GOTO(out, rc = -EINVAL);
764 * No [reg|hp] token has been specified
769 queue_name = strsep(&cmd, " ");
771 * The second token is either an optional [reg|hp] string,
774 if (strcmp(queue_name, "reg") == 0)
775 queue = PTLRPC_NRS_QUEUE_REG;
776 else if (strcmp(queue_name, "hp") == 0)
777 queue = PTLRPC_NRS_QUEUE_HP;
786 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
787 GOTO(out, rc = -ENODEV);
788 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
789 queue = PTLRPC_NRS_QUEUE_REG;
792 * Serialize NRS core lprocfs operations with policy registration/
795 mutex_lock(&nrs_core.nrs_mutex);
797 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
798 PTLRPC_NRS_CTL_START,
801 mutex_unlock(&nrs_core.nrs_mutex);
804 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
806 RETURN(rc < 0 ? rc : count);
809 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs_policies);
813 struct ptlrpc_srh_iterator {
816 struct ptlrpc_request *srhi_req;
820 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
821 struct ptlrpc_srh_iterator *srhi,
825 struct ptlrpc_request *req;
827 if (srhi->srhi_req != NULL &&
828 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
829 srhi->srhi_seq <= seq) {
830 /* If srhi_req was set previously, hasn't been culled and
831 * we're searching for a seq on or after it (i.e. more
832 * recent), search from it onwards.
833 * Since the service history is LRU (i.e. culled reqs will
834 * be near the head), we shouldn't have to do long re-scans.
836 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
837 "%s:%d: seek seq %llu, request seq %llu\n",
838 svcpt->scp_service->srv_name, svcpt->scp_cpt,
839 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
840 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
841 "%s:%d: seek offset %llu, request seq %llu, "
842 "last culled %llu\n",
843 svcpt->scp_service->srv_name, svcpt->scp_cpt,
844 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
845 e = &srhi->srhi_req->rq_history_list;
847 /* search from start */
848 e = svcpt->scp_hist_reqs.next;
851 while (e != &svcpt->scp_hist_reqs) {
852 req = list_entry(e, struct ptlrpc_request, rq_history_list);
854 if (req->rq_history_seq >= seq) {
855 srhi->srhi_seq = req->rq_history_seq;
856 srhi->srhi_req = req;
866 * ptlrpc history sequence is used as "position" of seq_file, in some case,
867 * seq_read() will increase "position" to indicate reading the next
868 * element, however, low bits of history sequence are reserved for CPT id
869 * (check the details from comments before ptlrpc_req_add_history), which
870 * means seq_read() might change CPT id of history sequence and never
871 * finish reading of requests on a CPT. To make it work, we have to shift
872 * CPT id to high bits and timestamp to low bits, so seq_read() will only
873 * increase timestamp which can correctly indicate the next position.
876 /* convert seq_file pos to cpt */
877 #define PTLRPC_REQ_POS2CPT(svc, pos) \
878 ((svc)->srv_cpt_bits == 0 ? 0 : \
879 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
881 /* make up seq_file pos from cpt */
882 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
883 ((svc)->srv_cpt_bits == 0 ? 0 : \
884 (cpt) << (64 - (svc)->srv_cpt_bits))
886 /* convert sequence to position */
887 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
888 ((svc)->srv_cpt_bits == 0 ? (seq) : \
889 ((seq) >> (svc)->srv_cpt_bits) | \
890 ((seq) << (64 - (svc)->srv_cpt_bits)))
892 /* convert position to sequence */
893 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
894 ((svc)->srv_cpt_bits == 0 ? (pos) : \
895 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
896 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
899 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
901 struct ptlrpc_service *svc = s->private;
902 struct ptlrpc_service_part *svcpt;
903 struct ptlrpc_srh_iterator *srhi;
908 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
909 CWARN("Failed to read request history because size of loff_t "
910 "%d can't match size of u64\n", (int)sizeof(loff_t));
914 OBD_ALLOC(srhi, sizeof(*srhi));
919 srhi->srhi_req = NULL;
921 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
923 ptlrpc_service_for_each_part(svcpt, i, svc) {
924 if (i < cpt) /* skip */
926 if (i > cpt) /* make up the lowest position for this CPT */
927 *pos = PTLRPC_REQ_CPT2POS(svc, i);
929 mutex_lock(&svcpt->scp_mutex);
930 spin_lock(&svcpt->scp_lock);
931 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
932 PTLRPC_REQ_POS2SEQ(svc, *pos));
933 spin_unlock(&svcpt->scp_lock);
934 mutex_unlock(&svcpt->scp_mutex);
936 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
942 OBD_FREE(srhi, sizeof(*srhi));
947 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
949 struct ptlrpc_srh_iterator *srhi = iter;
952 OBD_FREE(srhi, sizeof(*srhi));
956 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
957 void *iter, loff_t *pos)
959 struct ptlrpc_service *svc = s->private;
960 struct ptlrpc_srh_iterator *srhi = iter;
961 struct ptlrpc_service_part *svcpt;
966 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
967 svcpt = svc->srv_parts[i];
969 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
970 srhi->srhi_req = NULL;
971 seq = srhi->srhi_seq = 0;
972 } else { /* the next sequence */
973 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
976 mutex_lock(&svcpt->scp_mutex);
977 spin_lock(&svcpt->scp_lock);
978 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
979 spin_unlock(&svcpt->scp_lock);
980 mutex_unlock(&svcpt->scp_mutex);
982 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
988 OBD_FREE(srhi, sizeof(*srhi));
993 /* common ost/mdt so_req_printer */
994 void target_print_req(void *seq_file, struct ptlrpc_request *req)
996 /* Called holding srv_lock with irqs disabled.
997 * Print specific req contents and a newline.
998 * CAVEAT EMPTOR: check request message length before printing!!!
999 * You might have received any old crap so you must be just as
1000 * careful here as the service's request parser!!!
1002 struct seq_file *sf = seq_file;
1004 switch (req->rq_phase) {
1006 /* still awaiting a service thread's attention, or rejected
1007 * because the generic request message didn't unpack
1009 seq_printf(sf, "<not swabbed>\n");
1011 case RQ_PHASE_INTERPRET:
1012 /* being handled, so basic msg swabbed, and opc is valid
1013 * but racing with mds_handle(). fallthrough.
1016 case RQ_PHASE_COMPLETE:
1017 /* been handled by mds_handle(), reply state may be volatile */
1018 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
1021 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
1024 EXPORT_SYMBOL(target_print_req);
1026 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1028 struct ptlrpc_service *svc = s->private;
1029 struct ptlrpc_srh_iterator *srhi = iter;
1030 struct ptlrpc_service_part *svcpt;
1031 struct ptlrpc_request *req;
1034 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1036 svcpt = svc->srv_parts[srhi->srhi_idx];
1038 mutex_lock(&svcpt->scp_mutex);
1039 spin_lock(&svcpt->scp_lock);
1041 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1044 struct timespec64 arrival, sent, arrivaldiff;
1045 char nidstr[LNET_NIDSTR_SIZE];
1047 req = srhi->srhi_req;
1049 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1050 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1051 sent.tv_sec = req->rq_sent;
1053 arrivaldiff = timespec64_sub(sent, arrival);
1055 /* Print common req fields.
1056 * CAVEAT EMPTOR: we're racing with the service handler
1057 * here. The request could contain any old crap, so you
1058 * must be just as careful as the service's request
1059 * parser. Currently I only print stuff here I know is OK
1060 * to look at coz it was set up in request_in_callback()!!!
1063 "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1064 req->rq_history_seq,
1065 req->rq_export && req->rq_export->exp_obd ?
1066 req->rq_export->exp_obd->obd_name :
1067 libcfs_nidstr_r(&req->rq_self, nidstr,
1069 libcfs_idstr(&req->rq_peer), req->rq_xid,
1070 req->rq_reqlen, ptlrpc_rqphase2str(req),
1071 (s64)req->rq_arrival_time.tv_sec,
1072 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1073 (s64)arrivaldiff.tv_sec,
1074 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1075 (s64)(req->rq_sent - req->rq_deadline));
1076 if (svc->srv_ops.so_req_printer == NULL)
1077 seq_printf(s, "\n");
1079 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1082 spin_unlock(&svcpt->scp_lock);
1083 mutex_unlock(&svcpt->scp_mutex);
1089 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1091 static const struct seq_operations sops = {
1092 .start = ptlrpc_lprocfs_svc_req_history_start,
1093 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1094 .next = ptlrpc_lprocfs_svc_req_history_next,
1095 .show = ptlrpc_lprocfs_svc_req_history_show,
1097 struct seq_file *seqf;
1100 rc = seq_open(file, &sops);
1104 seqf = file->private_data;
1105 seqf->private = inode->i_private;
1109 /* See also lprocfs_rd_timeouts */
1110 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1112 struct ptlrpc_service *svc = m->private;
1113 struct ptlrpc_service_part *svcpt;
1114 time64_t worst_timestamp;
1115 timeout_t cur_timeout;
1116 timeout_t worst_timeout;
1120 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1125 ptlrpc_service_for_each_part(svcpt, i, svc) {
1126 cur_timeout = at_get(&svcpt->scp_at_estimate);
1127 worst_timeout = svcpt->scp_at_estimate.at_worst_timeout_ever;
1128 worst_timestamp = svcpt->scp_at_estimate.at_worst_timestamp;
1130 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1131 "service", cur_timeout, worst_timeout,
1133 ktime_get_real_seconds() - worst_timestamp);
1135 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1141 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1143 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1144 struct attribute *attr,
1147 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1150 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1153 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1154 struct attribute *attr,
1158 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1163 rc = kstrtoul(buffer, 10, &val);
1167 spin_lock(&svc->srv_lock);
1168 svc->srv_hpreq_ratio = val;
1169 spin_unlock(&svc->srv_lock);
1173 LUSTRE_RW_ATTR(high_priority_ratio);
1175 static struct attribute *ptlrpc_svc_attrs[] = {
1176 &lustre_attr_threads_min.attr,
1177 &lustre_attr_threads_started.attr,
1178 &lustre_attr_threads_max.attr,
1179 &lustre_attr_high_priority_ratio.attr,
1183 KOBJ_ATTRIBUTE_GROUPS(ptlrpc_svc); /* creates ptlrpc_svc_groups */
1185 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1187 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1190 complete(&svc->srv_kobj_unregister);
1193 static struct kobj_type ptlrpc_svc_ktype = {
1194 .default_groups = KOBJ_ATTR_GROUPS(ptlrpc_svc),
1195 .sysfs_ops = &lustre_sysfs_ops,
1196 .release = ptlrpc_sysfs_svc_release,
1199 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1201 /* Let's see if we had a chance at initialization first */
1202 if (svc->srv_kobj.kset) {
1203 kobject_put(&svc->srv_kobj);
1204 wait_for_completion(&svc->srv_kobj_unregister);
1208 int ptlrpc_sysfs_register_service(struct kset *parent,
1209 struct ptlrpc_service *svc)
1211 svc->srv_kobj.kset = parent;
1212 init_completion(&svc->srv_kobj_unregister);
1213 return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1214 &parent->kobj, "%s", svc->srv_name);
1217 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1218 struct ptlrpc_service *svc)
1220 struct ldebugfs_vars ldebugfs_vars[] = {
1221 { .name = "req_buffer_history_len",
1222 .fops = &ptlrpc_lprocfs_req_buffer_history_len_fops,
1224 { .name = "req_buffer_history_max",
1225 .fops = &ptlrpc_lprocfs_req_buffer_history_max_fops,
1227 { .name = "timeouts",
1228 .fops = &ptlrpc_lprocfs_timeouts_fops,
1230 { .name = "nrs_policies",
1231 .fops = &ptlrpc_lprocfs_nrs_policies_fops,
1233 { .name = "req_buffers_max",
1234 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1238 static const struct file_operations req_history_fops = {
1239 .owner = THIS_MODULE,
1240 .open = ptlrpc_lprocfs_svc_req_history_open,
1242 .llseek = seq_lseek,
1243 .release = lprocfs_seq_release,
1246 ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1247 &svc->srv_debugfs_entry, &svc->srv_stats);
1248 if (!svc->srv_debugfs_entry)
1251 ldebugfs_add_vars(svc->srv_debugfs_entry, ldebugfs_vars, NULL);
1253 debugfs_create_file("req_history", 0400, svc->srv_debugfs_entry, svc,
1257 void ptlrpc_lprocfs_register_obd(struct obd_device *obd)
1259 ptlrpc_ldebugfs_register(obd->obd_debugfs_entry, NULL, "stats",
1260 &obd->obd_svc_debugfs_entry,
1261 &obd->obd_svc_stats);
1263 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1265 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1267 struct lprocfs_stats *svc_stats;
1268 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1269 int opc = opcode_offset(op);
1271 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1272 if (svc_stats == NULL || opc <= 0)
1275 LASSERT(opc < LUSTRE_MAX_OPCODES);
1276 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1277 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1280 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1282 struct lprocfs_stats *svc_stats;
1285 if (!req->rq_import)
1288 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1292 idx = lustre_msg_get_opc(req->rq_reqmsg);
1295 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1298 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1301 LASSERTF(0, "unsupported opcode %u\n", idx);
1305 lprocfs_counter_add(svc_stats, idx, bytes);
1308 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1310 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1312 debugfs_remove_recursive(svc->srv_debugfs_entry);
1315 lprocfs_stats_free(&svc->srv_stats);
1318 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1320 /* cleanup first to allow concurrent access to device's
1321 * stats via debugfs to complete safely
1323 lprocfs_obd_cleanup(obd);
1325 debugfs_remove_recursive(obd->obd_svc_debugfs_entry);
1327 if (obd->obd_svc_stats)
1328 lprocfs_stats_free(&obd->obd_svc_stats);
1330 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1332 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1335 struct obd_device *obd = container_of(kobj, struct obd_device,
1337 struct obd_import *imp;
1338 struct ptlrpc_request *req;
1342 with_imp_locked(obd, imp, rc)
1343 req = ptlrpc_prep_ping(imp);
1350 req->rq_send_state = LUSTRE_IMP_FULL;
1352 rc = ptlrpc_queue_wait(req);
1353 ptlrpc_req_finished(req);
1357 EXPORT_SYMBOL(ping_show);
1359 /* kept for older verison of tools. */
1360 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1361 const char *buffer, size_t count)
1363 int rc = ping_show(kobj, attr, (char *)buffer);
1365 return (rc < 0) ? rc : count;
1367 EXPORT_SYMBOL(ping_store);
1369 /* Write the connection UUID to this file to attempt to connect to that node.
1370 * The connection UUID is a node's primary NID. For example,
1371 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1374 ldebugfs_import_seq_write(struct file *file, const char __user *buffer,
1375 size_t count, loff_t *off)
1377 struct seq_file *m = file->private_data;
1378 struct obd_device *obd = m->private;
1379 struct obd_import *imp;
1384 const char prefix[] = "connection=";
1385 const int prefix_len = sizeof(prefix) - 1;
1388 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1391 OBD_ALLOC(kbuf, count + 1);
1395 if (copy_from_user(kbuf, buffer, count))
1396 GOTO(out, rc = -EFAULT);
1400 /* only support connection=uuid::instance now */
1401 if (strncmp(prefix, kbuf, prefix_len) != 0)
1402 GOTO(out, rc = -EINVAL);
1404 with_imp_locked(obd, imp, rc) {
1405 uuid = kbuf + prefix_len;
1406 ptr = strstr(uuid, "::");
1413 ptr += 2; /* Skip :: */
1414 rc = kstrtouint(ptr, 10, &inst);
1416 CERROR("config: wrong instance # %s\n", ptr);
1417 } else if (inst != imp->imp_connect_data.ocd_instance) {
1419 "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1420 imp->imp_obd->obd_name,
1421 imp->imp_connect_data.ocd_instance,
1426 "IR: %s has already been connecting to "
1428 imp->imp_obd->obd_name, inst);
1433 ptlrpc_recover_import(imp, uuid, 1);
1437 OBD_FREE(kbuf, count + 1);
1440 EXPORT_SYMBOL(ldebugfs_import_seq_write);
1442 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1444 struct obd_device *obd = m->private;
1445 struct obd_import *imp;
1448 with_imp_locked(obd, imp, rc)
1449 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1453 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1456 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1457 size_t count, loff_t *off)
1459 struct seq_file *m = file->private_data;
1460 struct obd_device *obd = m->private;
1461 struct obd_import *imp;
1465 rc = kstrtobool_from_user(buffer, count, &val);
1469 with_imp_locked(obd, imp, rc) {
1470 spin_lock(&imp->imp_lock);
1471 imp->imp_no_pinger_recover = !val;
1472 spin_unlock(&imp->imp_lock);
1477 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);