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;
211 unsigned int svc_counter_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 svc_counter_config, "req_waittime", "usec");
229 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
230 svc_counter_config, "req_qdepth", "reqs");
231 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
232 svc_counter_config, "req_active", "reqs");
233 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
234 svc_counter_config, "req_timeout", "sec");
235 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
236 svc_counter_config, "reqbuf_avail", "bufs");
237 for (i = 0; i < EXTRA_LAST_OPC; i++) {
241 case BRW_WRITE_BYTES:
249 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
251 ll_eopcode2str(i), units);
253 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
254 __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 debugfs_create_file(name, 0644, svc_debugfs_entry, svc_stats,
262 &ldebugfs_stats_seq_fops);
265 *debugfs_root_ret = svc_debugfs_entry;
266 *stats_ret = svc_stats;
270 ptlrpc_lprocfs_req_buffer_history_len_seq_show(struct seq_file *m, void *v)
272 struct ptlrpc_service *svc = m->private;
273 struct ptlrpc_service_part *svcpt;
277 ptlrpc_service_for_each_part(svcpt, i, svc)
278 total += svcpt->scp_hist_nrqbds;
280 seq_printf(m, "%d\n", total);
285 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_buffer_history_len);
288 ptlrpc_lprocfs_req_buffer_history_max_seq_show(struct seq_file *m, void *n)
290 struct ptlrpc_service *svc = m->private;
291 struct ptlrpc_service_part *svcpt;
295 ptlrpc_service_for_each_part(svcpt, i, svc)
296 total += svc->srv_hist_nrqbds_cpt_max;
298 seq_printf(m, "%d\n", total);
303 ptlrpc_lprocfs_req_buffer_history_max_seq_write(struct file *file,
304 const char __user *buffer,
305 size_t count, loff_t *off)
307 struct seq_file *m = file->private_data;
308 struct ptlrpc_service *svc = m->private;
309 unsigned long long val;
310 unsigned long long limit;
314 rc = kstrtoull_from_user(buffer, count, 0, &val);
318 if (val < 0 || val > INT_MAX)
321 /* This sanity check is more of an insanity check; we can still
322 * hose a kernel by allowing the request history to grow too
323 * far. The roundup to the next power of two is an empirical way
324 * to take care that request buffer is allocated in Slab and thus
325 * will be upgraded */
326 bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
328 limit = cfs_totalram_pages() / (2 * bufpages);
329 /* do not allow history to consume more than half max number of rqbds */
330 if ((svc->srv_nrqbds_max == 0 && val > limit) ||
331 (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
334 spin_lock(&svc->srv_lock);
337 svc->srv_hist_nrqbds_cpt_max = 0;
339 svc->srv_hist_nrqbds_cpt_max =
340 max(1, ((int)val / svc->srv_ncpts));
342 spin_unlock(&svc->srv_lock);
347 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffer_history_max);
350 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
352 struct ptlrpc_service *svc = m->private;
354 seq_printf(m, "%d\n", svc->srv_nrqbds_max);
359 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
360 const char __user *buffer,
361 size_t count, loff_t *off)
363 struct seq_file *m = file->private_data;
364 struct ptlrpc_service *svc = m->private;
368 rc = kstrtoint_from_user(buffer, count, 0, &val);
372 if (val < svc->srv_nbuf_per_group && val != 0)
375 spin_lock(&svc->srv_lock);
377 svc->srv_nrqbds_max = (uint)val;
379 spin_unlock(&svc->srv_lock);
384 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
386 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
389 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
392 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
395 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
396 const char *buffer, size_t count)
398 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
403 rc = kstrtoul(buffer, 10, &val);
407 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
410 spin_lock(&svc->srv_lock);
411 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
412 spin_unlock(&svc->srv_lock);
416 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
418 spin_unlock(&svc->srv_lock);
422 LUSTRE_RW_ATTR(threads_min);
424 static ssize_t threads_started_show(struct kobject *kobj,
425 struct attribute *attr,
428 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
430 struct ptlrpc_service_part *svcpt;
434 ptlrpc_service_for_each_part(svcpt, i, svc)
435 total += svcpt->scp_nthrs_running;
437 return sprintf(buf, "%d\n", total);
439 LUSTRE_RO_ATTR(threads_started);
441 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
444 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
447 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
450 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
451 const char *buffer, size_t count)
453 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
458 rc = kstrtoul(buffer, 10, &val);
462 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
465 spin_lock(&svc->srv_lock);
466 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
467 spin_unlock(&svc->srv_lock);
471 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
473 spin_unlock(&svc->srv_lock);
477 LUSTRE_RW_ATTR(threads_max);
480 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
482 * \param[in] state The policy state
484 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
489 case NRS_POL_STATE_INVALID:
491 case NRS_POL_STATE_STOPPED:
493 case NRS_POL_STATE_STOPPING:
495 case NRS_POL_STATE_STARTING:
497 case NRS_POL_STATE_STARTED:
503 * Obtains status information for \a policy.
505 * Information is copied in \a info.
507 * \param[in] policy The policy
508 * \param[out] info Holds returned status information
510 static void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
511 struct ptlrpc_nrs_pol_info *info)
513 LASSERT(policy != NULL);
514 LASSERT(info != NULL);
515 assert_spin_locked(&policy->pol_nrs->nrs_lock);
517 BUILD_BUG_ON(sizeof(info->pi_arg) != sizeof(policy->pol_arg));
518 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
519 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
521 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
522 info->pi_state = policy->pol_state;
524 * XXX: These are accessed without holding
525 * ptlrpc_service_part::scp_req_lock.
527 info->pi_req_queued = policy->pol_req_queued;
528 info->pi_req_started = policy->pol_req_started;
532 * Reads and prints policy status information for all policies of a PTLRPC
535 static int ptlrpc_lprocfs_nrs_policies_seq_show(struct seq_file *m, void *n)
537 struct ptlrpc_service *svc = m->private;
538 struct ptlrpc_service_part *svcpt;
539 struct ptlrpc_nrs *nrs;
540 struct ptlrpc_nrs_policy *policy;
541 struct ptlrpc_nrs_pol_info *infos;
542 struct ptlrpc_nrs_pol_info tmp;
543 unsigned int num_pols;
544 unsigned int pol_idx = 0;
551 * Serialize NRS core lprocfs operations with policy registration/
554 mutex_lock(&nrs_core.nrs_mutex);
557 * Use the first service partition's regular NRS head in order to obtain
558 * the number of policies registered with NRS heads of this service. All
559 * service partitions will have the same number of policies.
561 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
563 spin_lock(&nrs->nrs_lock);
564 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
565 spin_unlock(&nrs->nrs_lock);
567 OBD_ALLOC_PTR_ARRAY(infos, num_pols);
569 GOTO(out, rc = -ENOMEM);
572 ptlrpc_service_for_each_part(svcpt, i, svc) {
573 nrs = nrs_svcpt2nrs(svcpt, hp);
574 spin_lock(&nrs->nrs_lock);
578 list_for_each_entry(policy, &nrs->nrs_policy_list,
580 LASSERT(pol_idx < num_pols);
582 nrs_policy_get_info_locked(policy, &tmp);
584 * Copy values when handling the first service
588 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
590 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
592 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
593 sizeof(tmp.pi_state));
594 infos[pol_idx].pi_fallback = tmp.pi_fallback;
596 * For the rest of the service partitions
597 * sanity-check the values we get.
600 LASSERT(strncmp(infos[pol_idx].pi_name,
602 NRS_POL_NAME_MAX) == 0);
603 LASSERT(strncmp(infos[pol_idx].pi_arg,
605 sizeof(tmp.pi_arg)) == 0);
607 * Not asserting ptlrpc_nrs_pol_info::pi_state,
608 * because it may be different between
609 * instances of the same policy in different
610 * service partitions.
612 LASSERT(infos[pol_idx].pi_fallback ==
616 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
617 infos[pol_idx].pi_req_started += tmp.pi_req_started;
621 spin_unlock(&nrs->nrs_lock);
625 * Policy status information output is in YAML format.
641 * high_priority_requests:
654 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
655 "high_priority_requests:");
657 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
658 if (strlen(infos[pol_idx].pi_arg) > 0)
659 seq_printf(m, " - name: %s %s\n",
660 infos[pol_idx].pi_name,
661 infos[pol_idx].pi_arg);
663 seq_printf(m, " - name: %s\n",
664 infos[pol_idx].pi_name);
667 seq_printf(m, " state: %s\n"
670 " active: %-20d\n\n",
671 nrs_state2str(infos[pol_idx].pi_state),
672 infos[pol_idx].pi_fallback ? "yes" : "no",
673 (int)infos[pol_idx].pi_req_queued,
674 (int)infos[pol_idx].pi_req_started);
677 if (!hp && nrs_svc_has_hp(svc)) {
678 memset(infos, 0, num_pols * sizeof(*infos));
681 * Redo the processing for the service's HP NRS heads' policies.
689 OBD_FREE_PTR_ARRAY(infos, num_pols);
691 mutex_unlock(&nrs_core.nrs_mutex);
696 #define LPROCFS_NRS_WR_MAX_ARG (1024)
698 * The longest valid command string is the maxium policy name size, plus the
699 * length of the " reg" substring, plus the lenght of argument
701 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 + \
702 LPROCFS_NRS_WR_MAX_ARG)
705 * Starts and stops a given policy on a PTLRPC service.
707 * Commands consist of the policy name, followed by an optional [reg|hp] token;
708 * if the optional token is omitted, the operation is performed on both the
709 * regular and high-priority (if the service has one) NRS head.
712 ptlrpc_lprocfs_nrs_policies_seq_write(struct file *file,
713 const char __user *buffer,
714 size_t count, loff_t *off)
716 struct seq_file *m = file->private_data;
717 struct ptlrpc_service *svc = m->private;
718 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
720 char *cmd_copy = NULL;
726 if (count >= LPROCFS_NRS_WR_MAX_CMD)
727 GOTO(out, rc = -EINVAL);
729 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
731 GOTO(out, rc = -ENOMEM);
733 * strsep() modifies its argument, so keep a copy
737 if (copy_from_user(cmd, buffer, count))
738 GOTO(out, rc = -EFAULT);
742 policy_name = strsep(&cmd, " ");
744 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
745 GOTO(out, rc = -EINVAL);
748 * No [reg|hp] token has been specified
753 queue_name = strsep(&cmd, " ");
755 * The second token is either an optional [reg|hp] string,
758 if (strcmp(queue_name, "reg") == 0)
759 queue = PTLRPC_NRS_QUEUE_REG;
760 else if (strcmp(queue_name, "hp") == 0)
761 queue = PTLRPC_NRS_QUEUE_HP;
770 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
771 GOTO(out, rc = -ENODEV);
772 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
773 queue = PTLRPC_NRS_QUEUE_REG;
776 * Serialize NRS core lprocfs operations with policy registration/
779 mutex_lock(&nrs_core.nrs_mutex);
781 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
782 PTLRPC_NRS_CTL_START,
785 mutex_unlock(&nrs_core.nrs_mutex);
788 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
790 RETURN(rc < 0 ? rc : count);
793 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs_policies);
797 struct ptlrpc_srh_iterator {
800 struct ptlrpc_request *srhi_req;
804 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
805 struct ptlrpc_srh_iterator *srhi,
809 struct ptlrpc_request *req;
811 if (srhi->srhi_req != NULL &&
812 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
813 srhi->srhi_seq <= seq) {
814 /* If srhi_req was set previously, hasn't been culled and
815 * we're searching for a seq on or after it (i.e. more
816 * recent), search from it onwards.
817 * Since the service history is LRU (i.e. culled reqs will
818 * be near the head), we shouldn't have to do long re-scans.
820 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
821 "%s:%d: seek seq %llu, request seq %llu\n",
822 svcpt->scp_service->srv_name, svcpt->scp_cpt,
823 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
824 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
825 "%s:%d: seek offset %llu, request seq %llu, "
826 "last culled %llu\n",
827 svcpt->scp_service->srv_name, svcpt->scp_cpt,
828 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
829 e = &srhi->srhi_req->rq_history_list;
831 /* search from start */
832 e = svcpt->scp_hist_reqs.next;
835 while (e != &svcpt->scp_hist_reqs) {
836 req = list_entry(e, struct ptlrpc_request, rq_history_list);
838 if (req->rq_history_seq >= seq) {
839 srhi->srhi_seq = req->rq_history_seq;
840 srhi->srhi_req = req;
850 * ptlrpc history sequence is used as "position" of seq_file, in some case,
851 * seq_read() will increase "position" to indicate reading the next
852 * element, however, low bits of history sequence are reserved for CPT id
853 * (check the details from comments before ptlrpc_req_add_history), which
854 * means seq_read() might change CPT id of history sequence and never
855 * finish reading of requests on a CPT. To make it work, we have to shift
856 * CPT id to high bits and timestamp to low bits, so seq_read() will only
857 * increase timestamp which can correctly indicate the next position.
860 /* convert seq_file pos to cpt */
861 #define PTLRPC_REQ_POS2CPT(svc, pos) \
862 ((svc)->srv_cpt_bits == 0 ? 0 : \
863 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
865 /* make up seq_file pos from cpt */
866 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
867 ((svc)->srv_cpt_bits == 0 ? 0 : \
868 (cpt) << (64 - (svc)->srv_cpt_bits))
870 /* convert sequence to position */
871 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
872 ((svc)->srv_cpt_bits == 0 ? (seq) : \
873 ((seq) >> (svc)->srv_cpt_bits) | \
874 ((seq) << (64 - (svc)->srv_cpt_bits)))
876 /* convert position to sequence */
877 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
878 ((svc)->srv_cpt_bits == 0 ? (pos) : \
879 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
880 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
883 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
885 struct ptlrpc_service *svc = s->private;
886 struct ptlrpc_service_part *svcpt;
887 struct ptlrpc_srh_iterator *srhi;
892 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
893 CWARN("Failed to read request history because size of loff_t "
894 "%d can't match size of u64\n", (int)sizeof(loff_t));
898 OBD_ALLOC(srhi, sizeof(*srhi));
903 srhi->srhi_req = NULL;
905 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
907 ptlrpc_service_for_each_part(svcpt, i, svc) {
908 if (i < cpt) /* skip */
910 if (i > cpt) /* make up the lowest position for this CPT */
911 *pos = PTLRPC_REQ_CPT2POS(svc, i);
913 mutex_lock(&svcpt->scp_mutex);
914 spin_lock(&svcpt->scp_lock);
915 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
916 PTLRPC_REQ_POS2SEQ(svc, *pos));
917 spin_unlock(&svcpt->scp_lock);
918 mutex_unlock(&svcpt->scp_mutex);
920 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
926 OBD_FREE(srhi, sizeof(*srhi));
931 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
933 struct ptlrpc_srh_iterator *srhi = iter;
936 OBD_FREE(srhi, sizeof(*srhi));
940 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
941 void *iter, loff_t *pos)
943 struct ptlrpc_service *svc = s->private;
944 struct ptlrpc_srh_iterator *srhi = iter;
945 struct ptlrpc_service_part *svcpt;
950 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
951 svcpt = svc->srv_parts[i];
953 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
954 srhi->srhi_req = NULL;
955 seq = srhi->srhi_seq = 0;
956 } else { /* the next sequence */
957 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
960 mutex_lock(&svcpt->scp_mutex);
961 spin_lock(&svcpt->scp_lock);
962 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
963 spin_unlock(&svcpt->scp_lock);
964 mutex_unlock(&svcpt->scp_mutex);
966 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
972 OBD_FREE(srhi, sizeof(*srhi));
977 /* common ost/mdt so_req_printer */
978 void target_print_req(void *seq_file, struct ptlrpc_request *req)
980 /* Called holding srv_lock with irqs disabled.
981 * Print specific req contents and a newline.
982 * CAVEAT EMPTOR: check request message length before printing!!!
983 * You might have received any old crap so you must be just as
984 * careful here as the service's request parser!!!
986 struct seq_file *sf = seq_file;
988 switch (req->rq_phase) {
990 /* still awaiting a service thread's attention, or rejected
991 * because the generic request message didn't unpack
993 seq_printf(sf, "<not swabbed>\n");
995 case RQ_PHASE_INTERPRET:
996 /* being handled, so basic msg swabbed, and opc is valid
997 * but racing with mds_handle(). fallthrough.
999 case RQ_PHASE_COMPLETE:
1000 /* been handled by mds_handle(), reply state may be volatile */
1001 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
1004 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
1007 EXPORT_SYMBOL(target_print_req);
1009 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1011 struct ptlrpc_service *svc = s->private;
1012 struct ptlrpc_srh_iterator *srhi = iter;
1013 struct ptlrpc_service_part *svcpt;
1014 struct ptlrpc_request *req;
1017 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1019 svcpt = svc->srv_parts[srhi->srhi_idx];
1021 mutex_lock(&svcpt->scp_mutex);
1022 spin_lock(&svcpt->scp_lock);
1024 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1027 struct timespec64 arrival, sent, arrivaldiff;
1028 char nidstr[LNET_NIDSTR_SIZE];
1030 req = srhi->srhi_req;
1032 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1033 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1034 sent.tv_sec = req->rq_sent;
1036 arrivaldiff = timespec64_sub(sent, arrival);
1038 /* Print common req fields.
1039 * CAVEAT EMPTOR: we're racing with the service handler
1040 * here. The request could contain any old crap, so you
1041 * must be just as careful as the service's request
1042 * parser. Currently I only print stuff here I know is OK
1043 * to look at coz it was set up in request_in_callback()!!!
1046 "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1047 req->rq_history_seq,
1048 req->rq_export && req->rq_export->exp_obd ?
1049 req->rq_export->exp_obd->obd_name :
1050 libcfs_nid2str_r(req->rq_self, nidstr,
1052 libcfs_id2str(req->rq_peer), req->rq_xid,
1053 req->rq_reqlen, ptlrpc_rqphase2str(req),
1054 (s64)req->rq_arrival_time.tv_sec,
1055 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1056 (s64)arrivaldiff.tv_sec,
1057 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1058 (s64)(req->rq_sent - req->rq_deadline));
1059 if (svc->srv_ops.so_req_printer == NULL)
1060 seq_printf(s, "\n");
1062 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1065 spin_unlock(&svcpt->scp_lock);
1066 mutex_unlock(&svcpt->scp_mutex);
1072 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1074 static const struct seq_operations sops = {
1075 .start = ptlrpc_lprocfs_svc_req_history_start,
1076 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1077 .next = ptlrpc_lprocfs_svc_req_history_next,
1078 .show = ptlrpc_lprocfs_svc_req_history_show,
1080 struct seq_file *seqf;
1083 rc = seq_open(file, &sops);
1087 seqf = file->private_data;
1088 seqf->private = inode->i_private;
1092 /* See also lprocfs_rd_timeouts */
1093 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1095 struct ptlrpc_service *svc = m->private;
1096 struct ptlrpc_service_part *svcpt;
1097 time64_t worst_timestamp;
1098 timeout_t cur_timeout;
1099 timeout_t worst_timeout;
1103 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1108 ptlrpc_service_for_each_part(svcpt, i, svc) {
1109 cur_timeout = at_get(&svcpt->scp_at_estimate);
1110 worst_timeout = svcpt->scp_at_estimate.at_worst_timeout_ever;
1111 worst_timestamp = svcpt->scp_at_estimate.at_worst_timestamp;
1113 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1114 "service", cur_timeout, worst_timeout,
1116 ktime_get_real_seconds() - worst_timestamp);
1118 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1124 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1126 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1127 struct attribute *attr,
1130 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1133 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1136 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1137 struct attribute *attr,
1141 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1146 rc = kstrtoul(buffer, 10, &val);
1150 spin_lock(&svc->srv_lock);
1151 svc->srv_hpreq_ratio = val;
1152 spin_unlock(&svc->srv_lock);
1156 LUSTRE_RW_ATTR(high_priority_ratio);
1158 static struct attribute *ptlrpc_svc_attrs[] = {
1159 &lustre_attr_threads_min.attr,
1160 &lustre_attr_threads_started.attr,
1161 &lustre_attr_threads_max.attr,
1162 &lustre_attr_high_priority_ratio.attr,
1166 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1168 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1171 complete(&svc->srv_kobj_unregister);
1174 static struct kobj_type ptlrpc_svc_ktype = {
1175 .default_attrs = ptlrpc_svc_attrs,
1176 .sysfs_ops = &lustre_sysfs_ops,
1177 .release = ptlrpc_sysfs_svc_release,
1180 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1182 /* Let's see if we had a chance at initialization first */
1183 if (svc->srv_kobj.kset) {
1184 kobject_put(&svc->srv_kobj);
1185 wait_for_completion(&svc->srv_kobj_unregister);
1189 int ptlrpc_sysfs_register_service(struct kset *parent,
1190 struct ptlrpc_service *svc)
1192 svc->srv_kobj.kset = parent;
1193 init_completion(&svc->srv_kobj_unregister);
1194 return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1195 &parent->kobj, "%s", svc->srv_name);
1198 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1199 struct ptlrpc_service *svc)
1201 struct ldebugfs_vars ldebugfs_vars[] = {
1202 { .name = "req_buffer_history_len",
1203 .fops = &ptlrpc_lprocfs_req_buffer_history_len_fops,
1205 { .name = "req_buffer_history_max",
1206 .fops = &ptlrpc_lprocfs_req_buffer_history_max_fops,
1208 { .name = "timeouts",
1209 .fops = &ptlrpc_lprocfs_timeouts_fops,
1211 { .name = "nrs_policies",
1212 .fops = &ptlrpc_lprocfs_nrs_policies_fops,
1214 { .name = "req_buffers_max",
1215 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1219 static const struct file_operations req_history_fops = {
1220 .owner = THIS_MODULE,
1221 .open = ptlrpc_lprocfs_svc_req_history_open,
1223 .llseek = seq_lseek,
1224 .release = lprocfs_seq_release,
1227 ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1228 &svc->srv_debugfs_entry, &svc->srv_stats);
1229 if (!svc->srv_debugfs_entry)
1232 ldebugfs_add_vars(svc->srv_debugfs_entry, ldebugfs_vars, NULL);
1234 debugfs_create_file("req_history", 0400, svc->srv_debugfs_entry, svc,
1238 void ptlrpc_lprocfs_register_obd(struct obd_device *obd)
1240 ptlrpc_ldebugfs_register(obd->obd_debugfs_entry, NULL, "stats",
1241 &obd->obd_svc_debugfs_entry,
1242 &obd->obd_svc_stats);
1244 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1246 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1248 struct lprocfs_stats *svc_stats;
1249 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1250 int opc = opcode_offset(op);
1252 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1253 if (svc_stats == NULL || opc <= 0)
1256 LASSERT(opc < LUSTRE_MAX_OPCODES);
1257 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1258 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1261 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1263 struct lprocfs_stats *svc_stats;
1266 if (!req->rq_import)
1269 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1273 idx = lustre_msg_get_opc(req->rq_reqmsg);
1276 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1279 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1282 LASSERTF(0, "unsupported opcode %u\n", idx);
1286 lprocfs_counter_add(svc_stats, idx, bytes);
1289 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1291 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1293 debugfs_remove_recursive(svc->srv_debugfs_entry);
1296 lprocfs_free_stats(&svc->srv_stats);
1299 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1301 /* cleanup first to allow concurrent access to device's
1302 * stats via debugfs to complete safely
1304 lprocfs_obd_cleanup(obd);
1306 debugfs_remove_recursive(obd->obd_svc_debugfs_entry);
1308 if (obd->obd_svc_stats)
1309 lprocfs_free_stats(&obd->obd_svc_stats);
1311 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1313 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1316 struct obd_device *obd = container_of(kobj, struct obd_device,
1318 struct obd_import *imp;
1319 struct ptlrpc_request *req;
1323 with_imp_locked(obd, imp, rc)
1324 req = ptlrpc_prep_ping(imp);
1331 req->rq_send_state = LUSTRE_IMP_FULL;
1333 rc = ptlrpc_queue_wait(req);
1334 ptlrpc_req_finished(req);
1338 EXPORT_SYMBOL(ping_show);
1340 /* kept for older verison of tools. */
1341 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1342 const char *buffer, size_t count)
1344 int rc = ping_show(kobj, attr, (char *)buffer);
1346 return (rc < 0) ? rc : count;
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 ldebugfs_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;
1365 const char prefix[] = "connection=";
1366 const int prefix_len = sizeof(prefix) - 1;
1369 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1372 OBD_ALLOC(kbuf, count + 1);
1376 if (copy_from_user(kbuf, buffer, count))
1377 GOTO(out, rc = -EFAULT);
1381 /* only support connection=uuid::instance now */
1382 if (strncmp(prefix, kbuf, prefix_len) != 0)
1383 GOTO(out, rc = -EINVAL);
1385 with_imp_locked(obd, imp, rc) {
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) {
1400 "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1401 imp->imp_obd->obd_name,
1402 imp->imp_connect_data.ocd_instance,
1407 "IR: %s has already been connecting to "
1409 imp->imp_obd->obd_name, inst);
1414 ptlrpc_recover_import(imp, uuid, 1);
1418 OBD_FREE(kbuf, count + 1);
1421 EXPORT_SYMBOL(ldebugfs_import_seq_write);
1423 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1425 struct obd_device *obd = m->private;
1426 struct obd_import *imp;
1429 with_imp_locked(obd, imp, rc)
1430 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1434 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1437 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1438 size_t count, loff_t *off)
1440 struct seq_file *m = file->private_data;
1441 struct obd_device *obd = m->private;
1442 struct obd_import *imp;
1446 rc = kstrtobool_from_user(buffer, count, &val);
1450 with_imp_locked(obd, imp, rc) {
1451 spin_lock(&imp->imp_lock);
1452 imp->imp_no_pinger_recover = !val;
1453 spin_unlock(&imp->imp_lock);
1458 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);