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 { 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 #ifdef HAVE_SERVER_SUPPORT
136 { OUT_UPDATE, "out_update" },
137 { LFSCK_NOTIFY, "lfsck_notify" },
138 { LFSCK_QUERY, "lfsck_query" },
142 static struct ll_eopcode {
145 } ll_eopcode_table[EXTRA_LAST_OPC] = {
146 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
147 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
148 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
149 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
150 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
151 { MDS_REINT_SETATTR, "mds_reint_setattr" },
152 { MDS_REINT_CREATE, "mds_reint_create" },
153 { MDS_REINT_LINK, "mds_reint_link" },
154 { MDS_REINT_UNLINK, "mds_reint_unlink" },
155 { MDS_REINT_RENAME, "mds_reint_rename" },
156 { MDS_REINT_OPEN, "mds_reint_open" },
157 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
158 { MDS_REINT_RESYNC, "mds_reint_resync" },
159 { BRW_READ_BYTES, "read_bytes" },
160 { BRW_WRITE_BYTES, "write_bytes" },
163 const char *ll_opcode2str(__u32 opcode)
165 __u32 offset = opcode_offset(opcode);
167 /* When one of the assertions below fail, chances are that:
168 * 1) A new opcode was added in include/lustre/lustre_idl.h,
169 * but is missing from the table above.
170 * or 2) The opcode space was renumbered or rearranged,
171 * and the opcode_offset() function in
172 * ptlrpc_internal.h needs to be modified.
174 LASSERTF(offset < LUSTRE_MAX_OPCODES,
175 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
176 offset, LUSTRE_MAX_OPCODES);
177 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
178 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
179 offset, ll_rpc_opcode_table[offset].opcode, opcode);
181 return ll_rpc_opcode_table[offset].opname;
184 const int ll_str2opcode(const char *ops)
188 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
189 if (ll_rpc_opcode_table[i].opname != NULL &&
190 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
191 return ll_rpc_opcode_table[i].opcode;
197 static const char *ll_eopcode2str(__u32 opcode)
199 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
200 return ll_eopcode_table[opcode].opname;
204 ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
205 struct dentry **debugfs_root_ret,
206 struct lprocfs_stats **stats_ret)
208 struct dentry *svc_debugfs_entry;
209 struct lprocfs_stats *svc_stats;
210 enum lprocfs_counter_config config = LPROCFS_CNTR_AVGMINMAX |
214 LASSERT(!*debugfs_root_ret);
215 LASSERT(!*stats_ret);
217 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
223 svc_debugfs_entry = debugfs_create_dir(dir, root);
225 svc_debugfs_entry = root;
227 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
228 config | LPROCFS_TYPE_USECS, "req_waittime");
229 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
230 config | LPROCFS_TYPE_REQS, "req_qdepth");
231 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
232 config | LPROCFS_TYPE_REQS, "req_active");
233 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
234 config | LPROCFS_TYPE_SECS, "req_timeout");
235 lprocfs_counter_init_units(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
236 config, "reqbuf_avail", "bufs");
237 for (i = 0; i < EXTRA_LAST_OPC; i++) {
238 enum lprocfs_counter_config extra_type = LPROCFS_TYPE_REQS;
241 case BRW_WRITE_BYTES:
243 extra_type = LPROCFS_TYPE_BYTES;
246 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
247 config | extra_type, ll_eopcode2str(i));
249 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
250 __u32 opcode = ll_rpc_opcode_table[i].opcode;
252 lprocfs_counter_init(svc_stats, EXTRA_MAX_OPCODES + i,
253 config | LPROCFS_TYPE_USECS,
254 ll_opcode2str(opcode));
257 debugfs_create_file(name, 0644, svc_debugfs_entry, svc_stats,
258 &ldebugfs_stats_seq_fops);
261 *debugfs_root_ret = svc_debugfs_entry;
262 *stats_ret = svc_stats;
266 ptlrpc_lprocfs_req_buffer_history_len_seq_show(struct seq_file *m, void *v)
268 struct ptlrpc_service *svc = m->private;
269 struct ptlrpc_service_part *svcpt;
273 ptlrpc_service_for_each_part(svcpt, i, svc)
274 total += svcpt->scp_hist_nrqbds;
276 seq_printf(m, "%d\n", total);
281 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_buffer_history_len);
284 ptlrpc_lprocfs_req_buffer_history_max_seq_show(struct seq_file *m, void *n)
286 struct ptlrpc_service *svc = m->private;
287 struct ptlrpc_service_part *svcpt;
291 ptlrpc_service_for_each_part(svcpt, i, svc)
292 total += svc->srv_hist_nrqbds_cpt_max;
294 seq_printf(m, "%d\n", total);
299 ptlrpc_lprocfs_req_buffer_history_max_seq_write(struct file *file,
300 const char __user *buffer,
301 size_t count, loff_t *off)
303 struct seq_file *m = file->private_data;
304 struct ptlrpc_service *svc = m->private;
305 unsigned long long val;
306 unsigned long long limit;
310 rc = kstrtoull_from_user(buffer, count, 0, &val);
314 if (val < 0 || val > INT_MAX)
317 /* This sanity check is more of an insanity check; we can still
318 * hose a kernel by allowing the request history to grow too
319 * far. The roundup to the next power of two is an empirical way
320 * to take care that request buffer is allocated in Slab and thus
321 * will be upgraded */
322 bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
324 limit = cfs_totalram_pages() / (2 * bufpages);
325 /* do not allow history to consume more than half max number of rqbds */
326 if ((svc->srv_nrqbds_max == 0 && val > limit) ||
327 (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
330 spin_lock(&svc->srv_lock);
333 svc->srv_hist_nrqbds_cpt_max = 0;
335 svc->srv_hist_nrqbds_cpt_max =
336 max(1, ((int)val / svc->srv_ncpts));
338 spin_unlock(&svc->srv_lock);
343 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffer_history_max);
346 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
348 struct ptlrpc_service *svc = m->private;
350 seq_printf(m, "%d\n", svc->srv_nrqbds_max);
355 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
356 const char __user *buffer,
357 size_t count, loff_t *off)
359 struct seq_file *m = file->private_data;
360 struct ptlrpc_service *svc = m->private;
364 rc = kstrtoint_from_user(buffer, count, 0, &val);
368 if (val < svc->srv_nbuf_per_group && val != 0)
371 spin_lock(&svc->srv_lock);
373 svc->srv_nrqbds_max = (uint)val;
375 spin_unlock(&svc->srv_lock);
380 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
382 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
385 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
388 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
391 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
392 const char *buffer, size_t count)
394 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
399 rc = kstrtoul(buffer, 10, &val);
403 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
406 spin_lock(&svc->srv_lock);
407 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
408 spin_unlock(&svc->srv_lock);
412 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
414 spin_unlock(&svc->srv_lock);
418 LUSTRE_RW_ATTR(threads_min);
420 static ssize_t threads_started_show(struct kobject *kobj,
421 struct attribute *attr,
424 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
426 struct ptlrpc_service_part *svcpt;
430 ptlrpc_service_for_each_part(svcpt, i, svc)
431 total += svcpt->scp_nthrs_running;
433 return sprintf(buf, "%d\n", total);
435 LUSTRE_RO_ATTR(threads_started);
437 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
440 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
443 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
446 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
447 const char *buffer, size_t count)
449 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
454 rc = kstrtoul(buffer, 10, &val);
458 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
461 spin_lock(&svc->srv_lock);
462 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
463 spin_unlock(&svc->srv_lock);
467 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
469 spin_unlock(&svc->srv_lock);
473 LUSTRE_RW_ATTR(threads_max);
476 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
478 * \param[in] state The policy state
480 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
485 case NRS_POL_STATE_INVALID:
487 case NRS_POL_STATE_STOPPED:
489 case NRS_POL_STATE_STOPPING:
491 case NRS_POL_STATE_STARTING:
493 case NRS_POL_STATE_STARTED:
499 * Obtains status information for \a policy.
501 * Information is copied in \a info.
503 * \param[in] policy The policy
504 * \param[out] info Holds returned status information
506 static void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
507 struct ptlrpc_nrs_pol_info *info)
509 LASSERT(policy != NULL);
510 LASSERT(info != NULL);
511 assert_spin_locked(&policy->pol_nrs->nrs_lock);
513 BUILD_BUG_ON(sizeof(info->pi_arg) != sizeof(policy->pol_arg));
514 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
515 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
517 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
518 info->pi_state = policy->pol_state;
520 * XXX: These are accessed without holding
521 * ptlrpc_service_part::scp_req_lock.
523 info->pi_req_queued = policy->pol_req_queued;
524 info->pi_req_started = policy->pol_req_started;
528 * Reads and prints policy status information for all policies of a PTLRPC
531 static int ptlrpc_lprocfs_nrs_policies_seq_show(struct seq_file *m, void *n)
533 struct ptlrpc_service *svc = m->private;
534 struct ptlrpc_service_part *svcpt;
535 struct ptlrpc_nrs *nrs;
536 struct ptlrpc_nrs_policy *policy;
537 struct ptlrpc_nrs_pol_info *infos;
538 struct ptlrpc_nrs_pol_info tmp;
539 unsigned int num_pols;
540 unsigned int pol_idx = 0;
547 * Serialize NRS core lprocfs operations with policy registration/
550 mutex_lock(&nrs_core.nrs_mutex);
553 * Use the first service partition's regular NRS head in order to obtain
554 * the number of policies registered with NRS heads of this service. All
555 * service partitions will have the same number of policies.
557 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
559 spin_lock(&nrs->nrs_lock);
560 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
561 spin_unlock(&nrs->nrs_lock);
563 OBD_ALLOC_PTR_ARRAY(infos, num_pols);
565 GOTO(out, rc = -ENOMEM);
568 ptlrpc_service_for_each_part(svcpt, i, svc) {
569 nrs = nrs_svcpt2nrs(svcpt, hp);
570 spin_lock(&nrs->nrs_lock);
574 list_for_each_entry(policy, &nrs->nrs_policy_list,
576 LASSERT(pol_idx < num_pols);
578 nrs_policy_get_info_locked(policy, &tmp);
580 * Copy values when handling the first service
584 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
586 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
588 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
589 sizeof(tmp.pi_state));
590 infos[pol_idx].pi_fallback = tmp.pi_fallback;
592 * For the rest of the service partitions
593 * sanity-check the values we get.
596 if (strncmp(infos[pol_idx].pi_name,
598 NRS_POL_NAME_MAX) != 0) {
599 spin_unlock(&nrs->nrs_lock);
601 CERROR("%s: failed to check pi_name: rc = %d\n",
602 svc->srv_thread_name, rc);
605 if (strncmp(infos[pol_idx].pi_arg,
607 sizeof(tmp.pi_arg)) != 0) {
608 spin_unlock(&nrs->nrs_lock);
610 CERROR("%s: failed to check pi_arg: rc = %d\n",
611 svc->srv_thread_name, rc);
615 * Not checking ptlrpc_nrs_pol_info::pi_state,
616 * because it may be different between
617 * instances of the same policy in different
618 * service partitions.
621 if (infos[pol_idx].pi_fallback !=
623 spin_unlock(&nrs->nrs_lock);
625 CERROR("%s: failed to check pi_fallback: rc = %d\n",
626 svc->srv_thread_name, rc);
631 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
632 infos[pol_idx].pi_req_started += tmp.pi_req_started;
636 spin_unlock(&nrs->nrs_lock);
640 * Policy status information output is in YAML format.
656 * high_priority_requests:
669 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
670 "high_priority_requests:");
672 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
673 if (strlen(infos[pol_idx].pi_arg) > 0)
674 seq_printf(m, " - name: %s %s\n",
675 infos[pol_idx].pi_name,
676 infos[pol_idx].pi_arg);
678 seq_printf(m, " - name: %s\n",
679 infos[pol_idx].pi_name);
682 seq_printf(m, " state: %s\n"
685 " active: %-20d\n\n",
686 nrs_state2str(infos[pol_idx].pi_state),
687 infos[pol_idx].pi_fallback ? "yes" : "no",
688 (int)infos[pol_idx].pi_req_queued,
689 (int)infos[pol_idx].pi_req_started);
692 if (!hp && nrs_svc_has_hp(svc)) {
693 memset(infos, 0, num_pols * sizeof(*infos));
696 * Redo the processing for the service's HP NRS heads' policies.
704 OBD_FREE_PTR_ARRAY(infos, num_pols);
706 mutex_unlock(&nrs_core.nrs_mutex);
711 #define LPROCFS_NRS_WR_MAX_ARG (1024)
713 * The longest valid command string is the maxium policy name size, plus the
714 * length of the " reg" substring, plus the lenght of argument
716 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 + \
717 LPROCFS_NRS_WR_MAX_ARG)
720 * Starts and stops a given policy on a PTLRPC service.
722 * Commands consist of the policy name, followed by an optional [reg|hp] token;
723 * if the optional token is omitted, the operation is performed on both the
724 * regular and high-priority (if the service has one) NRS head.
727 ptlrpc_lprocfs_nrs_policies_seq_write(struct file *file,
728 const char __user *buffer,
729 size_t count, loff_t *off)
731 struct seq_file *m = file->private_data;
732 struct ptlrpc_service *svc = m->private;
733 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
735 char *cmd_copy = NULL;
741 if (count >= LPROCFS_NRS_WR_MAX_CMD)
742 GOTO(out, rc = -EINVAL);
744 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
746 GOTO(out, rc = -ENOMEM);
748 * strsep() modifies its argument, so keep a copy
752 if (copy_from_user(cmd, buffer, count))
753 GOTO(out, rc = -EFAULT);
757 policy_name = strsep(&cmd, " ");
759 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
760 GOTO(out, rc = -EINVAL);
763 * No [reg|hp] token has been specified
768 queue_name = strsep(&cmd, " ");
770 * The second token is either an optional [reg|hp] string,
773 if (strcmp(queue_name, "reg") == 0)
774 queue = PTLRPC_NRS_QUEUE_REG;
775 else if (strcmp(queue_name, "hp") == 0)
776 queue = PTLRPC_NRS_QUEUE_HP;
785 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
786 GOTO(out, rc = -ENODEV);
787 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
788 queue = PTLRPC_NRS_QUEUE_REG;
791 * Serialize NRS core lprocfs operations with policy registration/
794 mutex_lock(&nrs_core.nrs_mutex);
796 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
797 PTLRPC_NRS_CTL_START,
800 mutex_unlock(&nrs_core.nrs_mutex);
803 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
805 RETURN(rc < 0 ? rc : count);
808 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs_policies);
812 struct ptlrpc_srh_iterator {
815 struct ptlrpc_request *srhi_req;
819 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
820 struct ptlrpc_srh_iterator *srhi,
824 struct ptlrpc_request *req;
826 if (srhi->srhi_req != NULL &&
827 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
828 srhi->srhi_seq <= seq) {
829 /* If srhi_req was set previously, hasn't been culled and
830 * we're searching for a seq on or after it (i.e. more
831 * recent), search from it onwards.
832 * Since the service history is LRU (i.e. culled reqs will
833 * be near the head), we shouldn't have to do long re-scans.
835 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
836 "%s:%d: seek seq %llu, request seq %llu\n",
837 svcpt->scp_service->srv_name, svcpt->scp_cpt,
838 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
839 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
840 "%s:%d: seek offset %llu, request seq %llu, "
841 "last culled %llu\n",
842 svcpt->scp_service->srv_name, svcpt->scp_cpt,
843 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
844 e = &srhi->srhi_req->rq_history_list;
846 /* search from start */
847 e = svcpt->scp_hist_reqs.next;
850 while (e != &svcpt->scp_hist_reqs) {
851 req = list_entry(e, struct ptlrpc_request, rq_history_list);
853 if (req->rq_history_seq >= seq) {
854 srhi->srhi_seq = req->rq_history_seq;
855 srhi->srhi_req = req;
865 * ptlrpc history sequence is used as "position" of seq_file, in some case,
866 * seq_read() will increase "position" to indicate reading the next
867 * element, however, low bits of history sequence are reserved for CPT id
868 * (check the details from comments before ptlrpc_req_add_history), which
869 * means seq_read() might change CPT id of history sequence and never
870 * finish reading of requests on a CPT. To make it work, we have to shift
871 * CPT id to high bits and timestamp to low bits, so seq_read() will only
872 * increase timestamp which can correctly indicate the next position.
875 /* convert seq_file pos to cpt */
876 #define PTLRPC_REQ_POS2CPT(svc, pos) \
877 ((svc)->srv_cpt_bits == 0 ? 0 : \
878 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
880 /* make up seq_file pos from cpt */
881 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
882 ((svc)->srv_cpt_bits == 0 ? 0 : \
883 (cpt) << (64 - (svc)->srv_cpt_bits))
885 /* convert sequence to position */
886 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
887 ((svc)->srv_cpt_bits == 0 ? (seq) : \
888 ((seq) >> (svc)->srv_cpt_bits) | \
889 ((seq) << (64 - (svc)->srv_cpt_bits)))
891 /* convert position to sequence */
892 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
893 ((svc)->srv_cpt_bits == 0 ? (pos) : \
894 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
895 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
898 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
900 struct ptlrpc_service *svc = s->private;
901 struct ptlrpc_service_part *svcpt;
902 struct ptlrpc_srh_iterator *srhi;
907 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
908 CWARN("Failed to read request history because size of loff_t "
909 "%d can't match size of u64\n", (int)sizeof(loff_t));
913 OBD_ALLOC(srhi, sizeof(*srhi));
918 srhi->srhi_req = NULL;
920 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
922 ptlrpc_service_for_each_part(svcpt, i, svc) {
923 if (i < cpt) /* skip */
925 if (i > cpt) /* make up the lowest position for this CPT */
926 *pos = PTLRPC_REQ_CPT2POS(svc, i);
928 mutex_lock(&svcpt->scp_mutex);
929 spin_lock(&svcpt->scp_lock);
930 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
931 PTLRPC_REQ_POS2SEQ(svc, *pos));
932 spin_unlock(&svcpt->scp_lock);
933 mutex_unlock(&svcpt->scp_mutex);
935 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
941 OBD_FREE(srhi, sizeof(*srhi));
946 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
948 struct ptlrpc_srh_iterator *srhi = iter;
951 OBD_FREE(srhi, sizeof(*srhi));
955 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
956 void *iter, loff_t *pos)
958 struct ptlrpc_service *svc = s->private;
959 struct ptlrpc_srh_iterator *srhi = iter;
960 struct ptlrpc_service_part *svcpt;
965 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
966 svcpt = svc->srv_parts[i];
968 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
969 srhi->srhi_req = NULL;
970 seq = srhi->srhi_seq = 0;
971 } else { /* the next sequence */
972 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
975 mutex_lock(&svcpt->scp_mutex);
976 spin_lock(&svcpt->scp_lock);
977 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
978 spin_unlock(&svcpt->scp_lock);
979 mutex_unlock(&svcpt->scp_mutex);
981 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
987 OBD_FREE(srhi, sizeof(*srhi));
992 /* common ost/mdt so_req_printer */
993 void target_print_req(void *seq_file, struct ptlrpc_request *req)
995 /* Called holding srv_lock with irqs disabled.
996 * Print specific req contents and a newline.
997 * CAVEAT EMPTOR: check request message length before printing!!!
998 * You might have received any old crap so you must be just as
999 * careful here as the service's request parser!!!
1001 struct seq_file *sf = seq_file;
1003 switch (req->rq_phase) {
1005 /* still awaiting a service thread's attention, or rejected
1006 * because the generic request message didn't unpack
1008 seq_printf(sf, "<not swabbed>\n");
1010 case RQ_PHASE_INTERPRET:
1011 /* being handled, so basic msg swabbed, and opc is valid
1012 * but racing with mds_handle(). fallthrough.
1015 case RQ_PHASE_COMPLETE:
1016 /* been handled by mds_handle(), reply state may be volatile */
1017 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
1020 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
1023 EXPORT_SYMBOL(target_print_req);
1025 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1027 struct ptlrpc_service *svc = s->private;
1028 struct ptlrpc_srh_iterator *srhi = iter;
1029 struct ptlrpc_service_part *svcpt;
1030 struct ptlrpc_request *req;
1033 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1035 svcpt = svc->srv_parts[srhi->srhi_idx];
1037 mutex_lock(&svcpt->scp_mutex);
1038 spin_lock(&svcpt->scp_lock);
1040 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1043 struct timespec64 arrival, sent, arrivaldiff;
1044 char nidstr[LNET_NIDSTR_SIZE];
1046 req = srhi->srhi_req;
1048 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1049 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1050 sent.tv_sec = req->rq_sent;
1052 arrivaldiff = timespec64_sub(sent, arrival);
1054 /* Print common req fields.
1055 * CAVEAT EMPTOR: we're racing with the service handler
1056 * here. The request could contain any old crap, so you
1057 * must be just as careful as the service's request
1058 * parser. Currently I only print stuff here I know is OK
1059 * to look at coz it was set up in request_in_callback()!!!
1062 "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1063 req->rq_history_seq,
1064 req->rq_export && req->rq_export->exp_obd ?
1065 req->rq_export->exp_obd->obd_name :
1066 libcfs_nidstr_r(&req->rq_self, nidstr,
1068 libcfs_idstr(&req->rq_peer), req->rq_xid,
1069 req->rq_reqlen, ptlrpc_rqphase2str(req),
1070 (s64)req->rq_arrival_time.tv_sec,
1071 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1072 (s64)arrivaldiff.tv_sec,
1073 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1074 (s64)(req->rq_sent - req->rq_deadline));
1075 if (svc->srv_ops.so_req_printer == NULL)
1076 seq_printf(s, "\n");
1078 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1081 spin_unlock(&svcpt->scp_lock);
1082 mutex_unlock(&svcpt->scp_mutex);
1088 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1090 static const struct seq_operations sops = {
1091 .start = ptlrpc_lprocfs_svc_req_history_start,
1092 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1093 .next = ptlrpc_lprocfs_svc_req_history_next,
1094 .show = ptlrpc_lprocfs_svc_req_history_show,
1096 struct seq_file *seqf;
1099 rc = seq_open(file, &sops);
1103 seqf = file->private_data;
1104 seqf->private = inode->i_private;
1108 /* See also lprocfs_rd_timeouts */
1109 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1111 struct ptlrpc_service *svc = m->private;
1112 struct ptlrpc_service_part *svcpt;
1113 time64_t worst_timestamp;
1114 timeout_t cur_timeout;
1115 timeout_t worst_timeout;
1119 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1124 ptlrpc_service_for_each_part(svcpt, i, svc) {
1125 cur_timeout = at_get(&svcpt->scp_at_estimate);
1126 worst_timeout = svcpt->scp_at_estimate.at_worst_timeout_ever;
1127 worst_timestamp = svcpt->scp_at_estimate.at_worst_timestamp;
1129 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1130 "service", cur_timeout, worst_timeout,
1132 ktime_get_real_seconds() - worst_timestamp);
1134 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1140 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1142 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1143 struct attribute *attr,
1146 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1149 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1152 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1153 struct attribute *attr,
1157 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1162 rc = kstrtoul(buffer, 10, &val);
1166 spin_lock(&svc->srv_lock);
1167 svc->srv_hpreq_ratio = val;
1168 spin_unlock(&svc->srv_lock);
1172 LUSTRE_RW_ATTR(high_priority_ratio);
1174 static struct attribute *ptlrpc_svc_attrs[] = {
1175 &lustre_attr_threads_min.attr,
1176 &lustre_attr_threads_started.attr,
1177 &lustre_attr_threads_max.attr,
1178 &lustre_attr_high_priority_ratio.attr,
1182 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1184 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1187 complete(&svc->srv_kobj_unregister);
1190 static struct kobj_type ptlrpc_svc_ktype = {
1191 .default_attrs = ptlrpc_svc_attrs,
1192 .sysfs_ops = &lustre_sysfs_ops,
1193 .release = ptlrpc_sysfs_svc_release,
1196 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1198 /* Let's see if we had a chance at initialization first */
1199 if (svc->srv_kobj.kset) {
1200 kobject_put(&svc->srv_kobj);
1201 wait_for_completion(&svc->srv_kobj_unregister);
1205 int ptlrpc_sysfs_register_service(struct kset *parent,
1206 struct ptlrpc_service *svc)
1208 svc->srv_kobj.kset = parent;
1209 init_completion(&svc->srv_kobj_unregister);
1210 return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1211 &parent->kobj, "%s", svc->srv_name);
1214 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1215 struct ptlrpc_service *svc)
1217 struct ldebugfs_vars ldebugfs_vars[] = {
1218 { .name = "req_buffer_history_len",
1219 .fops = &ptlrpc_lprocfs_req_buffer_history_len_fops,
1221 { .name = "req_buffer_history_max",
1222 .fops = &ptlrpc_lprocfs_req_buffer_history_max_fops,
1224 { .name = "timeouts",
1225 .fops = &ptlrpc_lprocfs_timeouts_fops,
1227 { .name = "nrs_policies",
1228 .fops = &ptlrpc_lprocfs_nrs_policies_fops,
1230 { .name = "req_buffers_max",
1231 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1235 static const struct file_operations req_history_fops = {
1236 .owner = THIS_MODULE,
1237 .open = ptlrpc_lprocfs_svc_req_history_open,
1239 .llseek = seq_lseek,
1240 .release = lprocfs_seq_release,
1243 ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1244 &svc->srv_debugfs_entry, &svc->srv_stats);
1245 if (!svc->srv_debugfs_entry)
1248 ldebugfs_add_vars(svc->srv_debugfs_entry, ldebugfs_vars, NULL);
1250 debugfs_create_file("req_history", 0400, svc->srv_debugfs_entry, svc,
1254 void ptlrpc_lprocfs_register_obd(struct obd_device *obd)
1256 ptlrpc_ldebugfs_register(obd->obd_debugfs_entry, NULL, "stats",
1257 &obd->obd_svc_debugfs_entry,
1258 &obd->obd_svc_stats);
1260 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1262 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1264 struct lprocfs_stats *svc_stats;
1265 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1266 int opc = opcode_offset(op);
1268 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1269 if (svc_stats == NULL || opc <= 0)
1272 LASSERT(opc < LUSTRE_MAX_OPCODES);
1273 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1274 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1277 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1279 struct lprocfs_stats *svc_stats;
1282 if (!req->rq_import)
1285 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1289 idx = lustre_msg_get_opc(req->rq_reqmsg);
1292 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1295 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1298 LASSERTF(0, "unsupported opcode %u\n", idx);
1302 lprocfs_counter_add(svc_stats, idx, bytes);
1305 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1307 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1309 debugfs_remove_recursive(svc->srv_debugfs_entry);
1312 lprocfs_free_stats(&svc->srv_stats);
1315 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1317 /* cleanup first to allow concurrent access to device's
1318 * stats via debugfs to complete safely
1320 lprocfs_obd_cleanup(obd);
1322 debugfs_remove_recursive(obd->obd_svc_debugfs_entry);
1324 if (obd->obd_svc_stats)
1325 lprocfs_free_stats(&obd->obd_svc_stats);
1327 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1329 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1332 struct obd_device *obd = container_of(kobj, struct obd_device,
1334 struct obd_import *imp;
1335 struct ptlrpc_request *req;
1339 with_imp_locked(obd, imp, rc)
1340 req = ptlrpc_prep_ping(imp);
1347 req->rq_send_state = LUSTRE_IMP_FULL;
1349 rc = ptlrpc_queue_wait(req);
1350 ptlrpc_req_finished(req);
1354 EXPORT_SYMBOL(ping_show);
1356 /* kept for older verison of tools. */
1357 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1358 const char *buffer, size_t count)
1360 int rc = ping_show(kobj, attr, (char *)buffer);
1362 return (rc < 0) ? rc : count;
1364 EXPORT_SYMBOL(ping_store);
1366 /* Write the connection UUID to this file to attempt to connect to that node.
1367 * The connection UUID is a node's primary NID. For example,
1368 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1371 ldebugfs_import_seq_write(struct file *file, const char __user *buffer,
1372 size_t count, loff_t *off)
1374 struct seq_file *m = file->private_data;
1375 struct obd_device *obd = m->private;
1376 struct obd_import *imp;
1381 const char prefix[] = "connection=";
1382 const int prefix_len = sizeof(prefix) - 1;
1385 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1388 OBD_ALLOC(kbuf, count + 1);
1392 if (copy_from_user(kbuf, buffer, count))
1393 GOTO(out, rc = -EFAULT);
1397 /* only support connection=uuid::instance now */
1398 if (strncmp(prefix, kbuf, prefix_len) != 0)
1399 GOTO(out, rc = -EINVAL);
1401 with_imp_locked(obd, imp, rc) {
1402 uuid = kbuf + prefix_len;
1403 ptr = strstr(uuid, "::");
1410 ptr += 2; /* Skip :: */
1411 rc = kstrtouint(ptr, 10, &inst);
1413 CERROR("config: wrong instance # %s\n", ptr);
1414 } else if (inst != imp->imp_connect_data.ocd_instance) {
1416 "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1417 imp->imp_obd->obd_name,
1418 imp->imp_connect_data.ocd_instance,
1423 "IR: %s has already been connecting to "
1425 imp->imp_obd->obd_name, inst);
1430 ptlrpc_recover_import(imp, uuid, 1);
1434 OBD_FREE(kbuf, count + 1);
1437 EXPORT_SYMBOL(ldebugfs_import_seq_write);
1439 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1441 struct obd_device *obd = m->private;
1442 struct obd_import *imp;
1445 with_imp_locked(obd, imp, rc)
1446 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1450 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1453 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1454 size_t count, loff_t *off)
1456 struct seq_file *m = file->private_data;
1457 struct obd_device *obd = m->private;
1458 struct obd_import *imp;
1462 rc = kstrtobool_from_user(buffer, count, &val);
1466 with_imp_locked(obd, imp, rc) {
1467 spin_lock(&imp->imp_lock);
1468 imp->imp_no_pinger_recover = !val;
1469 spin_unlock(&imp->imp_lock);
1474 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);