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 = ldebugfs_register(dir, root, NULL, NULL);
220 if (IS_ERR(svc_debugfs_entry)) {
221 lprocfs_free_stats(&svc_stats);
225 svc_debugfs_entry = root;
228 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
229 svc_counter_config, "req_waittime", "usec");
230 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
231 svc_counter_config, "req_qdepth", "reqs");
232 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
233 svc_counter_config, "req_active", "reqs");
234 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
235 svc_counter_config, "req_timeout", "sec");
236 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
237 svc_counter_config, "reqbuf_avail", "bufs");
238 for (i = 0; i < EXTRA_LAST_OPC; i++) {
242 case BRW_WRITE_BYTES:
250 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
252 ll_eopcode2str(i), units);
254 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
255 __u32 opcode = ll_rpc_opcode_table[i].opcode;
256 lprocfs_counter_init(svc_stats,
257 EXTRA_MAX_OPCODES + i, svc_counter_config,
258 ll_opcode2str(opcode), "usec");
261 rc = ldebugfs_register_stats(svc_debugfs_entry, name, svc_stats);
264 debugfs_remove_recursive(svc_debugfs_entry);
265 lprocfs_free_stats(&svc_stats);
268 *debugfs_root_ret = svc_debugfs_entry;
269 *stats_ret = svc_stats;
274 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
276 struct ptlrpc_service *svc = m->private;
277 struct ptlrpc_service_part *svcpt;
281 ptlrpc_service_for_each_part(svcpt, i, svc)
282 total += svcpt->scp_hist_nrqbds;
284 seq_printf(m, "%d\n", total);
289 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
292 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
294 struct ptlrpc_service *svc = m->private;
295 struct ptlrpc_service_part *svcpt;
299 ptlrpc_service_for_each_part(svcpt, i, svc)
300 total += svc->srv_hist_nrqbds_cpt_max;
302 seq_printf(m, "%d\n", total);
307 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
308 const char __user *buffer,
309 size_t count, loff_t *off)
311 struct seq_file *m = file->private_data;
312 struct ptlrpc_service *svc = m->private;
313 unsigned long long val;
314 unsigned long long limit;
318 rc = kstrtoull_from_user(buffer, count, 0, &val);
322 if (val < 0 || val > INT_MAX)
325 /* This sanity check is more of an insanity check; we can still
326 * hose a kernel by allowing the request history to grow too
327 * far. The roundup to the next power of two is an empirical way
328 * to take care that request buffer is allocated in Slab and thus
329 * will be upgraded */
330 bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
332 limit = cfs_totalram_pages() / (2 * bufpages);
333 /* do not allow history to consume more than half max number of rqbds */
334 if ((svc->srv_nrqbds_max == 0 && val > limit) ||
335 (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
338 spin_lock(&svc->srv_lock);
341 svc->srv_hist_nrqbds_cpt_max = 0;
343 svc->srv_hist_nrqbds_cpt_max =
344 max(1, ((int)val / svc->srv_ncpts));
346 spin_unlock(&svc->srv_lock);
351 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
354 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
356 struct ptlrpc_service *svc = m->private;
358 seq_printf(m, "%d\n", svc->srv_nrqbds_max);
363 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
364 const char __user *buffer,
365 size_t count, loff_t *off)
367 struct seq_file *m = file->private_data;
368 struct ptlrpc_service *svc = m->private;
372 rc = kstrtoint_from_user(buffer, count, 0, &val);
376 if (val < svc->srv_nbuf_per_group && val != 0)
379 spin_lock(&svc->srv_lock);
381 svc->srv_nrqbds_max = (uint)val;
383 spin_unlock(&svc->srv_lock);
388 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
390 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
393 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
396 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
399 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
400 const char *buffer, size_t count)
402 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
407 rc = kstrtoul(buffer, 10, &val);
411 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
414 spin_lock(&svc->srv_lock);
415 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
416 spin_unlock(&svc->srv_lock);
420 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
422 spin_unlock(&svc->srv_lock);
426 LUSTRE_RW_ATTR(threads_min);
428 static ssize_t threads_started_show(struct kobject *kobj,
429 struct attribute *attr,
432 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
434 struct ptlrpc_service_part *svcpt;
438 ptlrpc_service_for_each_part(svcpt, i, svc)
439 total += svcpt->scp_nthrs_running;
441 return sprintf(buf, "%d\n", total);
443 LUSTRE_RO_ATTR(threads_started);
445 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
448 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
451 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
454 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
455 const char *buffer, size_t count)
457 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
462 rc = kstrtoul(buffer, 10, &val);
466 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
469 spin_lock(&svc->srv_lock);
470 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
471 spin_unlock(&svc->srv_lock);
475 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
477 spin_unlock(&svc->srv_lock);
481 LUSTRE_RW_ATTR(threads_max);
484 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
486 * \param[in] state The policy state
488 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
493 case NRS_POL_STATE_INVALID:
495 case NRS_POL_STATE_STOPPED:
497 case NRS_POL_STATE_STOPPING:
499 case NRS_POL_STATE_STARTING:
501 case NRS_POL_STATE_STARTED:
507 * Obtains status information for \a policy.
509 * Information is copied in \a info.
511 * \param[in] policy The policy
512 * \param[out] info Holds returned status information
514 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
515 struct ptlrpc_nrs_pol_info *info)
517 LASSERT(policy != NULL);
518 LASSERT(info != NULL);
519 assert_spin_locked(&policy->pol_nrs->nrs_lock);
521 BUILD_BUG_ON(sizeof(info->pi_arg) != sizeof(policy->pol_arg));
522 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
523 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
525 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
526 info->pi_state = policy->pol_state;
528 * XXX: These are accessed without holding
529 * ptlrpc_service_part::scp_req_lock.
531 info->pi_req_queued = policy->pol_req_queued;
532 info->pi_req_started = policy->pol_req_started;
536 * Reads and prints policy status information for all policies of a PTLRPC
539 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
541 struct ptlrpc_service *svc = m->private;
542 struct ptlrpc_service_part *svcpt;
543 struct ptlrpc_nrs *nrs;
544 struct ptlrpc_nrs_policy *policy;
545 struct ptlrpc_nrs_pol_info *infos;
546 struct ptlrpc_nrs_pol_info tmp;
548 unsigned pol_idx = 0;
555 * Serialize NRS core lprocfs operations with policy registration/
558 mutex_lock(&nrs_core.nrs_mutex);
561 * Use the first service partition's regular NRS head in order to obtain
562 * the number of policies registered with NRS heads of this service. All
563 * service partitions will have the same number of policies.
565 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
567 spin_lock(&nrs->nrs_lock);
568 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
569 spin_unlock(&nrs->nrs_lock);
571 OBD_ALLOC(infos, num_pols * sizeof(*infos));
573 GOTO(out, rc = -ENOMEM);
576 ptlrpc_service_for_each_part(svcpt, i, svc) {
577 nrs = nrs_svcpt2nrs(svcpt, hp);
578 spin_lock(&nrs->nrs_lock);
582 list_for_each_entry(policy, &nrs->nrs_policy_list,
584 LASSERT(pol_idx < num_pols);
586 nrs_policy_get_info_locked(policy, &tmp);
588 * Copy values when handling the first service
592 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
594 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
596 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
597 sizeof(tmp.pi_state));
598 infos[pol_idx].pi_fallback = tmp.pi_fallback;
600 * For the rest of the service partitions
601 * sanity-check the values we get.
604 LASSERT(strncmp(infos[pol_idx].pi_name,
606 NRS_POL_NAME_MAX) == 0);
607 LASSERT(strncmp(infos[pol_idx].pi_arg,
609 sizeof(tmp.pi_arg)) == 0);
611 * Not asserting ptlrpc_nrs_pol_info::pi_state,
612 * because it may be different between
613 * instances of the same policy in different
614 * service partitions.
616 LASSERT(infos[pol_idx].pi_fallback ==
620 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
621 infos[pol_idx].pi_req_started += tmp.pi_req_started;
625 spin_unlock(&nrs->nrs_lock);
629 * Policy status information output is in YAML format.
645 * high_priority_requests:
658 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
659 "high_priority_requests:");
661 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
662 if (strlen(infos[pol_idx].pi_arg) > 0)
663 seq_printf(m, " - name: %s %s\n",
664 infos[pol_idx].pi_name,
665 infos[pol_idx].pi_arg);
667 seq_printf(m, " - name: %s\n",
668 infos[pol_idx].pi_name);
671 seq_printf(m, " state: %s\n"
674 " active: %-20d\n\n",
675 nrs_state2str(infos[pol_idx].pi_state),
676 infos[pol_idx].pi_fallback ? "yes" : "no",
677 (int)infos[pol_idx].pi_req_queued,
678 (int)infos[pol_idx].pi_req_started);
681 if (!hp && nrs_svc_has_hp(svc)) {
682 memset(infos, 0, num_pols * sizeof(*infos));
685 * Redo the processing for the service's HP NRS heads' policies.
693 OBD_FREE(infos, num_pols * sizeof(*infos));
695 mutex_unlock(&nrs_core.nrs_mutex);
701 #define LPROCFS_NRS_WR_MAX_ARG (1024)
703 * The longest valid command string is the maxium policy name size, plus the
704 * length of the " reg" substring, plus the lenght of argument
706 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
707 + LPROCFS_NRS_WR_MAX_ARG)
710 * Starts and stops a given policy on a PTLRPC service.
712 * Commands consist of the policy name, followed by an optional [reg|hp] token;
713 * if the optional token is omitted, the operation is performed on both the
714 * regular and high-priority (if the service has one) NRS head.
717 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
718 size_t count, loff_t *off)
720 struct seq_file *m = file->private_data;
721 struct ptlrpc_service *svc = m->private;
722 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
724 char *cmd_copy = NULL;
730 if (count >= LPROCFS_NRS_WR_MAX_CMD)
731 GOTO(out, rc = -EINVAL);
733 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
735 GOTO(out, rc = -ENOMEM);
737 * strsep() modifies its argument, so keep a copy
741 if (copy_from_user(cmd, buffer, count))
742 GOTO(out, rc = -EFAULT);
746 policy_name = strsep(&cmd, " ");
748 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
749 GOTO(out, rc = -EINVAL);
752 * No [reg|hp] token has been specified
757 queue_name = strsep(&cmd, " ");
759 * The second token is either an optional [reg|hp] string,
762 if (strcmp(queue_name, "reg") == 0)
763 queue = PTLRPC_NRS_QUEUE_REG;
764 else if (strcmp(queue_name, "hp") == 0)
765 queue = PTLRPC_NRS_QUEUE_HP;
774 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
775 GOTO(out, rc = -ENODEV);
776 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
777 queue = PTLRPC_NRS_QUEUE_REG;
780 * Serialize NRS core lprocfs operations with policy registration/
783 mutex_lock(&nrs_core.nrs_mutex);
785 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
786 PTLRPC_NRS_CTL_START,
789 mutex_unlock(&nrs_core.nrs_mutex);
792 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
794 RETURN(rc < 0 ? rc : count);
797 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs);
801 struct ptlrpc_srh_iterator {
804 struct ptlrpc_request *srhi_req;
808 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
809 struct ptlrpc_srh_iterator *srhi,
813 struct ptlrpc_request *req;
815 if (srhi->srhi_req != NULL &&
816 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
817 srhi->srhi_seq <= seq) {
818 /* If srhi_req was set previously, hasn't been culled and
819 * we're searching for a seq on or after it (i.e. more
820 * recent), search from it onwards.
821 * Since the service history is LRU (i.e. culled reqs will
822 * be near the head), we shouldn't have to do long
824 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
825 "%s:%d: seek seq %llu, request seq %llu\n",
826 svcpt->scp_service->srv_name, svcpt->scp_cpt,
827 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
828 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
829 "%s:%d: seek offset %llu, request seq %llu, "
830 "last culled %llu\n",
831 svcpt->scp_service->srv_name, svcpt->scp_cpt,
832 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
833 e = &srhi->srhi_req->rq_history_list;
835 /* search from start */
836 e = svcpt->scp_hist_reqs.next;
839 while (e != &svcpt->scp_hist_reqs) {
840 req = list_entry(e, struct ptlrpc_request, rq_history_list);
842 if (req->rq_history_seq >= seq) {
843 srhi->srhi_seq = req->rq_history_seq;
844 srhi->srhi_req = req;
854 * ptlrpc history sequence is used as "position" of seq_file, in some case,
855 * seq_read() will increase "position" to indicate reading the next
856 * element, however, low bits of history sequence are reserved for CPT id
857 * (check the details from comments before ptlrpc_req_add_history), which
858 * means seq_read() might change CPT id of history sequence and never
859 * finish reading of requests on a CPT. To make it work, we have to shift
860 * CPT id to high bits and timestamp to low bits, so seq_read() will only
861 * increase timestamp which can correctly indicate the next position.
864 /* convert seq_file pos to cpt */
865 #define PTLRPC_REQ_POS2CPT(svc, pos) \
866 ((svc)->srv_cpt_bits == 0 ? 0 : \
867 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
869 /* make up seq_file pos from cpt */
870 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
871 ((svc)->srv_cpt_bits == 0 ? 0 : \
872 (cpt) << (64 - (svc)->srv_cpt_bits))
874 /* convert sequence to position */
875 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
876 ((svc)->srv_cpt_bits == 0 ? (seq) : \
877 ((seq) >> (svc)->srv_cpt_bits) | \
878 ((seq) << (64 - (svc)->srv_cpt_bits)))
880 /* convert position to sequence */
881 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
882 ((svc)->srv_cpt_bits == 0 ? (pos) : \
883 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
884 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
887 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
889 struct ptlrpc_service *svc = s->private;
890 struct ptlrpc_service_part *svcpt;
891 struct ptlrpc_srh_iterator *srhi;
896 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
897 CWARN("Failed to read request history because size of loff_t "
898 "%d can't match size of u64\n", (int)sizeof(loff_t));
902 OBD_ALLOC(srhi, sizeof(*srhi));
907 srhi->srhi_req = NULL;
909 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
911 ptlrpc_service_for_each_part(svcpt, i, svc) {
912 if (i < cpt) /* skip */
914 if (i > cpt) /* make up the lowest position for this CPT */
915 *pos = PTLRPC_REQ_CPT2POS(svc, i);
917 mutex_lock(&svcpt->scp_mutex);
918 spin_lock(&svcpt->scp_lock);
919 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
920 PTLRPC_REQ_POS2SEQ(svc, *pos));
921 spin_unlock(&svcpt->scp_lock);
922 mutex_unlock(&svcpt->scp_mutex);
924 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
930 OBD_FREE(srhi, sizeof(*srhi));
935 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
937 struct ptlrpc_srh_iterator *srhi = iter;
940 OBD_FREE(srhi, sizeof(*srhi));
944 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
945 void *iter, loff_t *pos)
947 struct ptlrpc_service *svc = s->private;
948 struct ptlrpc_srh_iterator *srhi = iter;
949 struct ptlrpc_service_part *svcpt;
954 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
955 svcpt = svc->srv_parts[i];
957 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
958 srhi->srhi_req = NULL;
959 seq = srhi->srhi_seq = 0;
960 } else { /* the next sequence */
961 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
964 mutex_lock(&svcpt->scp_mutex);
965 spin_lock(&svcpt->scp_lock);
966 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
967 spin_unlock(&svcpt->scp_lock);
968 mutex_unlock(&svcpt->scp_mutex);
970 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
976 OBD_FREE(srhi, sizeof(*srhi));
980 /* common ost/mdt so_req_printer */
981 void target_print_req(void *seq_file, struct ptlrpc_request *req)
983 /* Called holding srv_lock with irqs disabled.
984 * Print specific req contents and a newline.
985 * CAVEAT EMPTOR: check request message length before printing!!!
986 * You might have received any old crap so you must be just as
987 * careful here as the service's request parser!!! */
988 struct seq_file *sf = seq_file;
990 switch (req->rq_phase) {
992 /* still awaiting a service thread's attention, or rejected
993 * because the generic request message didn't unpack */
994 seq_printf(sf, "<not swabbed>\n");
996 case RQ_PHASE_INTERPRET:
997 /* being handled, so basic msg swabbed, and opc is valid
998 * but racing with mds_handle() */
999 case RQ_PHASE_COMPLETE:
1000 /* been handled by mds_handle() reply state possibly still
1002 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
1005 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
1008 EXPORT_SYMBOL(target_print_req);
1010 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1012 struct ptlrpc_service *svc = s->private;
1013 struct ptlrpc_srh_iterator *srhi = iter;
1014 struct ptlrpc_service_part *svcpt;
1015 struct ptlrpc_request *req;
1018 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1020 svcpt = svc->srv_parts[srhi->srhi_idx];
1022 mutex_lock(&svcpt->scp_mutex);
1023 spin_lock(&svcpt->scp_lock);
1025 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1028 struct timespec64 arrival, sent, arrivaldiff;
1029 char nidstr[LNET_NIDSTR_SIZE];
1031 req = srhi->srhi_req;
1033 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1034 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1035 sent.tv_sec = req->rq_sent;
1037 arrivaldiff = timespec64_sub(sent, arrival);
1039 /* Print common req fields.
1040 * CAVEAT EMPTOR: we're racing with the service handler
1041 * here. The request could contain any old crap, so you
1042 * must be just as careful as the service's request
1043 * parser. Currently I only print stuff here I know is OK
1044 * to look at coz it was set up in request_in_callback()!!!
1047 "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1048 req->rq_history_seq,
1049 req->rq_export && req->rq_export->exp_obd ?
1050 req->rq_export->exp_obd->obd_name :
1051 libcfs_nid2str_r(req->rq_self, nidstr,
1053 libcfs_id2str(req->rq_peer), req->rq_xid,
1054 req->rq_reqlen, ptlrpc_rqphase2str(req),
1055 (s64)req->rq_arrival_time.tv_sec,
1056 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1057 (s64)arrivaldiff.tv_sec,
1058 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1059 (s64)(req->rq_sent - req->rq_deadline));
1060 if (svc->srv_ops.so_req_printer == NULL)
1061 seq_printf(s, "\n");
1063 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1066 spin_unlock(&svcpt->scp_lock);
1067 mutex_unlock(&svcpt->scp_mutex);
1073 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1075 static struct seq_operations sops = {
1076 .start = ptlrpc_lprocfs_svc_req_history_start,
1077 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1078 .next = ptlrpc_lprocfs_svc_req_history_next,
1079 .show = ptlrpc_lprocfs_svc_req_history_show,
1081 struct seq_file *seqf;
1084 rc = LPROCFS_ENTRY_CHECK(inode);
1088 rc = seq_open(file, &sops);
1092 seqf = file->private_data;
1093 seqf->private = inode->i_private;
1097 /* See also lprocfs_rd_timeouts */
1098 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1100 struct ptlrpc_service *svc = m->private;
1101 struct ptlrpc_service_part *svcpt;
1108 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1113 ptlrpc_service_for_each_part(svcpt, i, svc) {
1114 cur = at_get(&svcpt->scp_at_estimate);
1115 worst = svcpt->scp_at_estimate.at_worst_ever;
1116 worstt = svcpt->scp_at_estimate.at_worst_time;
1118 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1119 "service", cur, worst, (s64)worstt,
1120 (s64)(ktime_get_real_seconds() - worstt));
1122 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1128 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1130 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1131 struct attribute *attr,
1134 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1137 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1140 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1141 struct attribute *attr,
1145 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1150 rc = kstrtoul(buffer, 10, &val);
1154 spin_lock(&svc->srv_lock);
1155 svc->srv_hpreq_ratio = val;
1156 spin_unlock(&svc->srv_lock);
1160 LUSTRE_RW_ATTR(high_priority_ratio);
1162 static struct attribute *ptlrpc_svc_attrs[] = {
1163 &lustre_attr_threads_min.attr,
1164 &lustre_attr_threads_started.attr,
1165 &lustre_attr_threads_max.attr,
1166 &lustre_attr_high_priority_ratio.attr,
1170 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1172 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1175 complete(&svc->srv_kobj_unregister);
1178 static struct kobj_type ptlrpc_svc_ktype = {
1179 .default_attrs = ptlrpc_svc_attrs,
1180 .sysfs_ops = &lustre_sysfs_ops,
1181 .release = ptlrpc_sysfs_svc_release,
1184 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1186 /* Let's see if we had a chance at initialization first */
1187 if (svc->srv_kobj.kset) {
1188 kobject_put(&svc->srv_kobj);
1189 wait_for_completion(&svc->srv_kobj_unregister);
1193 int ptlrpc_sysfs_register_service(struct kset *parent,
1194 struct ptlrpc_service *svc)
1196 svc->srv_kobj.kset = parent;
1197 init_completion(&svc->srv_kobj_unregister);
1198 return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1199 &parent->kobj, "%s", svc->srv_name);
1202 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1203 struct ptlrpc_service *svc)
1205 struct lprocfs_vars lproc_vars[] = {
1206 { .name = "req_buffer_history_len",
1207 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1209 { .name = "req_buffer_history_max",
1210 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1212 { .name = "timeouts",
1213 .fops = &ptlrpc_lprocfs_timeouts_fops,
1215 { .name = "nrs_policies",
1216 .fops = &ptlrpc_lprocfs_nrs_fops,
1218 { .name = "req_buffers_max",
1219 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1223 static struct file_operations req_history_fops = {
1224 .owner = THIS_MODULE,
1225 .open = ptlrpc_lprocfs_svc_req_history_open,
1227 .llseek = seq_lseek,
1228 .release = lprocfs_seq_release,
1233 ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1234 &svc->srv_debugfs_entry, &svc->srv_stats);
1235 if (!svc->srv_debugfs_entry)
1238 ldebugfs_add_vars(svc->srv_debugfs_entry, lproc_vars, NULL);
1240 rc = ldebugfs_seq_create(svc->srv_debugfs_entry, "req_history",
1241 0400, &req_history_fops, svc);
1243 CWARN("Error adding the req_history file\n");
1246 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1248 ptlrpc_ldebugfs_register(obddev->obd_debugfs_entry, NULL, "stats",
1249 &obddev->obd_svc_debugfs_entry,
1250 &obddev->obd_svc_stats);
1252 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1254 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1256 struct lprocfs_stats *svc_stats;
1257 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1258 int opc = opcode_offset(op);
1260 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1261 if (svc_stats == NULL || opc <= 0)
1263 LASSERT(opc < LUSTRE_MAX_OPCODES);
1264 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1265 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1268 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1270 struct lprocfs_stats *svc_stats;
1273 if (!req->rq_import)
1275 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1278 idx = lustre_msg_get_opc(req->rq_reqmsg);
1281 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1284 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1287 LASSERTF(0, "unsupported opcode %u\n", idx);
1291 lprocfs_counter_add(svc_stats, idx, bytes);
1294 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1296 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1298 debugfs_remove_recursive(svc->srv_debugfs_entry);
1301 lprocfs_free_stats(&svc->srv_stats);
1304 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1306 /* cleanup first to allow concurrent access to device's
1307 * stats via debugfs to complete safely
1309 lprocfs_obd_cleanup(obd);
1311 debugfs_remove_recursive(obd->obd_svc_debugfs_entry);
1313 if (obd->obd_svc_stats)
1314 lprocfs_free_stats(&obd->obd_svc_stats);
1316 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1318 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1321 struct obd_device *obd = container_of(kobj, struct obd_device,
1323 struct ptlrpc_request *req;
1327 LPROCFS_CLIMP_CHECK(obd);
1328 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1329 LPROCFS_CLIMP_EXIT(obd);
1333 req->rq_send_state = LUSTRE_IMP_FULL;
1335 rc = ptlrpc_queue_wait(req);
1336 ptlrpc_req_finished(req);
1340 EXPORT_SYMBOL(ping_show);
1342 /* kept for older verison of tools. */
1343 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1344 const char *buffer, size_t count)
1346 return ping_show(kobj, attr, (char *)buffer);
1348 EXPORT_SYMBOL(ping_store);
1350 /* Write the connection UUID to this file to attempt to connect to that node.
1351 * The connection UUID is a node's primary NID. For example,
1352 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1355 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1356 size_t count, loff_t *off)
1358 struct seq_file *m = file->private_data;
1359 struct obd_device *obd = m->private;
1360 struct obd_import *imp = obd->u.cli.cl_import;
1365 const char prefix[] = "connection=";
1366 const int prefix_len = sizeof(prefix) - 1;
1368 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1371 OBD_ALLOC(kbuf, count + 1);
1375 if (copy_from_user(kbuf, buffer, count))
1376 GOTO(out, count = -EFAULT);
1380 /* only support connection=uuid::instance now */
1381 if (strncmp(prefix, kbuf, prefix_len) != 0)
1382 GOTO(out, count = -EINVAL);
1384 LPROCFS_CLIMP_CHECK(obd);
1386 uuid = kbuf + prefix_len;
1387 ptr = strstr(uuid, "::");
1394 ptr += 2; /* Skip :: */
1395 rc = kstrtouint(ptr, 10, &inst);
1397 CERROR("config: wrong instance # %s\n", ptr);
1398 } else if (inst != imp->imp_connect_data.ocd_instance) {
1399 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1400 "target(%u/%u), reconnecting...\n",
1401 imp->imp_obd->obd_name,
1402 imp->imp_connect_data.ocd_instance, inst);
1405 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1407 imp->imp_obd->obd_name, inst);
1411 LPROCFS_CLIMP_EXIT(obd);
1413 ptlrpc_recover_import(imp, uuid, 1);
1416 OBD_FREE(kbuf, count + 1);
1419 EXPORT_SYMBOL(lprocfs_import_seq_write);
1421 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1423 struct obd_device *obd = m->private;
1424 struct obd_import *imp = obd->u.cli.cl_import;
1426 LPROCFS_CLIMP_CHECK(obd);
1427 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1428 LPROCFS_CLIMP_EXIT(obd);
1431 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1434 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1435 size_t count, loff_t *off)
1437 struct seq_file *m = file->private_data;
1438 struct obd_device *obd = m->private;
1439 struct client_obd *cli = &obd->u.cli;
1440 struct obd_import *imp = cli->cl_import;
1444 rc = kstrtobool_from_user(buffer, count, &val);
1448 LPROCFS_CLIMP_CHECK(obd);
1449 spin_lock(&imp->imp_lock);
1450 imp->imp_no_pinger_recover = !val;
1451 spin_unlock(&imp->imp_lock);
1452 LPROCFS_CLIMP_EXIT(obd);
1455 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);