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 { MDS_GETATTR, "mds_getattr" },
70 { MDS_GETATTR_NAME, "mds_getattr_lock" },
71 { MDS_CLOSE, "mds_close" },
72 { MDS_REINT, "mds_reint" },
73 { MDS_READPAGE, "mds_readpage" },
74 { MDS_CONNECT, "mds_connect" },
75 { MDS_DISCONNECT, "mds_disconnect" },
76 { MDS_GET_ROOT, "mds_get_root" },
77 { MDS_STATFS, "mds_statfs" },
78 { MDS_PIN, "mds_pin" },
79 { MDS_UNPIN, "mds_unpin" },
80 { MDS_SYNC, "mds_sync" },
81 { MDS_DONE_WRITING, "mds_done_writing" },
82 { MDS_SET_INFO, "mds_set_info" },
83 { MDS_QUOTACHECK, "mds_quotacheck" },
84 { MDS_QUOTACTL, "mds_quotactl" },
85 { MDS_GETXATTR, "mds_getxattr" },
86 { MDS_SETXATTR, "mds_setxattr" },
87 { MDS_WRITEPAGE, "mds_writepage" },
88 { MDS_IS_SUBDIR, "mds_is_subdir" },
89 { MDS_GET_INFO, "mds_get_info" },
90 { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
91 { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
92 { MDS_HSM_ACTION, "mds_hsm_action" },
93 { MDS_HSM_PROGRESS, "mds_hsm_progress" },
94 { MDS_HSM_REQUEST, "mds_hsm_request" },
95 { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
96 { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
97 { MDS_SWAP_LAYOUTS, "mds_swap_layouts" },
98 { MDS_RMFID, "mds_rmfid" },
99 { LDLM_ENQUEUE, "ldlm_enqueue" },
100 { LDLM_CONVERT, "ldlm_convert" },
101 { LDLM_CANCEL, "ldlm_cancel" },
102 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
103 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
104 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
105 { LDLM_SET_INFO, "ldlm_set_info" },
106 { MGS_CONNECT, "mgs_connect" },
107 { MGS_DISCONNECT, "mgs_disconnect" },
108 { MGS_EXCEPTION, "mgs_exception" },
109 { MGS_TARGET_REG, "mgs_target_reg" },
110 { MGS_TARGET_DEL, "mgs_target_del" },
111 { MGS_SET_INFO, "mgs_set_info" },
112 { MGS_CONFIG_READ, "mgs_config_read" },
113 { OBD_PING, "obd_ping" },
114 { 401, /* was OBD_LOG_CANCEL */ "llog_cancel" },
115 { 402, /* was OBD_QC_CALLBACK */ "obd_quota_callback" },
116 { OBD_IDX_READ, "dt_index_read" },
117 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
118 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
119 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
120 { 504, /*LLOG_ORIGIN_HANDLE_WRITE_REC*/"llog_origin_handle_write_rec" },
121 { 505, /* was LLOG_ORIGIN_HANDLE_CLOSE */ "llog_origin_handle_close" },
122 { 506, /* was LLOG_ORIGIN_CONNECT */ "llog_origin_connect" },
123 { 507, /* was LLOG_CATINFO */ "llog_catinfo" },
124 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
125 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
126 { QUOTA_DQACQ, "quota_acquire" },
127 { QUOTA_DQREL, "quota_release" },
128 { SEQ_QUERY, "seq_query" },
129 { SEC_CTX_INIT, "sec_ctx_init" },
130 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
131 { SEC_CTX_FINI, "sec_ctx_fini" },
132 { FLD_QUERY, "fld_query" },
133 { FLD_READ, "fld_read" },
134 { OUT_UPDATE, "out_update" },
135 { LFSCK_NOTIFY, "lfsck_notify" },
136 { LFSCK_QUERY, "lfsck_query" },
139 static struct ll_eopcode {
142 } ll_eopcode_table[EXTRA_LAST_OPC] = {
143 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
144 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
145 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
146 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
147 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
148 { MDS_REINT_SETATTR, "mds_reint_setattr" },
149 { MDS_REINT_CREATE, "mds_reint_create" },
150 { MDS_REINT_LINK, "mds_reint_link" },
151 { MDS_REINT_UNLINK, "mds_reint_unlink" },
152 { MDS_REINT_RENAME, "mds_reint_rename" },
153 { MDS_REINT_OPEN, "mds_reint_open" },
154 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
155 { MDS_REINT_RESYNC, "mds_reint_resync" },
156 { BRW_READ_BYTES, "read_bytes" },
157 { BRW_WRITE_BYTES, "write_bytes" },
160 const char *ll_opcode2str(__u32 opcode)
162 /* When one of the assertions below fail, chances are that:
163 * 1) A new opcode was added in include/lustre/lustre_idl.h,
164 * but is missing from the table above.
165 * or 2) The opcode space was renumbered or rearranged,
166 * and the opcode_offset() function in
167 * ptlrpc_internal.h needs to be modified.
169 __u32 offset = opcode_offset(opcode);
170 LASSERTF(offset < LUSTRE_MAX_OPCODES,
171 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
172 offset, LUSTRE_MAX_OPCODES);
173 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
174 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
175 offset, ll_rpc_opcode_table[offset].opcode, opcode);
176 return ll_rpc_opcode_table[offset].opname;
179 const int ll_str2opcode(const char *ops)
183 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
184 if (ll_rpc_opcode_table[i].opname != NULL &&
185 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
186 return ll_rpc_opcode_table[i].opcode;
192 static const char *ll_eopcode2str(__u32 opcode)
194 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
195 return ll_eopcode_table[opcode].opname;
199 ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
200 struct dentry **debugfs_root_ret,
201 struct lprocfs_stats **stats_ret)
203 struct dentry *svc_debugfs_entry;
204 struct lprocfs_stats *svc_stats;
206 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
209 LASSERT(!*debugfs_root_ret);
210 LASSERT(!*stats_ret);
212 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
218 svc_debugfs_entry = ldebugfs_register(dir, root, NULL, NULL);
219 if (IS_ERR(svc_debugfs_entry)) {
220 lprocfs_free_stats(&svc_stats);
224 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;
255 lprocfs_counter_init(svc_stats,
256 EXTRA_MAX_OPCODES + i, svc_counter_config,
257 ll_opcode2str(opcode), "usec");
260 rc = ldebugfs_register_stats(svc_debugfs_entry, name, svc_stats);
263 ldebugfs_remove(&svc_debugfs_entry);
264 lprocfs_free_stats(&svc_stats);
267 *debugfs_root_ret = svc_debugfs_entry;
268 *stats_ret = svc_stats;
273 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
275 struct ptlrpc_service *svc = m->private;
276 struct ptlrpc_service_part *svcpt;
280 ptlrpc_service_for_each_part(svcpt, i, svc)
281 total += svcpt->scp_hist_nrqbds;
283 seq_printf(m, "%d\n", total);
288 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
291 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
293 struct ptlrpc_service *svc = m->private;
294 struct ptlrpc_service_part *svcpt;
298 ptlrpc_service_for_each_part(svcpt, i, svc)
299 total += svc->srv_hist_nrqbds_cpt_max;
301 seq_printf(m, "%d\n", total);
306 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
307 const char __user *buffer,
308 size_t count, loff_t *off)
310 struct seq_file *m = file->private_data;
311 struct ptlrpc_service *svc = m->private;
312 unsigned long long val;
316 rc = kstrtoull_from_user(buffer, count, 0, &val);
320 if (val < 0 || val > INT_MAX)
323 /* This sanity check is more of an insanity check; we can still
324 * hose a kernel by allowing the request history to grow too
325 * far. The roundup to the next power of two is an empirical way
326 * to take care that request buffer is allocated in Slab and thus
327 * will be upgraded */
328 bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
330 /* do not allow history to consume more than half max number of rqbds */
331 if ((svc->srv_nrqbds_max == 0 && val > totalram_pages / (2 * bufpages)) ||
332 (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
335 spin_lock(&svc->srv_lock);
338 svc->srv_hist_nrqbds_cpt_max = 0;
340 svc->srv_hist_nrqbds_cpt_max =
341 max(1, ((int)val / svc->srv_ncpts));
343 spin_unlock(&svc->srv_lock);
348 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
351 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
353 struct ptlrpc_service *svc = m->private;
355 seq_printf(m, "%d\n", svc->srv_nrqbds_max);
360 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
361 const char __user *buffer,
362 size_t count, loff_t *off)
364 struct seq_file *m = file->private_data;
365 struct ptlrpc_service *svc = m->private;
369 rc = kstrtoint_from_user(buffer, count, 0, &val);
373 if (val < svc->srv_nbuf_per_group && val != 0)
376 spin_lock(&svc->srv_lock);
378 svc->srv_nrqbds_max = (uint)val;
380 spin_unlock(&svc->srv_lock);
385 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
387 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
390 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
393 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
396 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
397 const char *buffer, size_t count)
399 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
404 rc = kstrtoul(buffer, 10, &val);
408 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
411 spin_lock(&svc->srv_lock);
412 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
413 spin_unlock(&svc->srv_lock);
417 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
419 spin_unlock(&svc->srv_lock);
423 LUSTRE_RW_ATTR(threads_min);
425 static ssize_t threads_started_show(struct kobject *kobj,
426 struct attribute *attr,
429 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
431 struct ptlrpc_service_part *svcpt;
435 ptlrpc_service_for_each_part(svcpt, i, svc)
436 total += svcpt->scp_nthrs_running;
438 return sprintf(buf, "%d\n", total);
440 LUSTRE_RO_ATTR(threads_started);
442 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
445 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
448 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
451 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
452 const char *buffer, size_t count)
454 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
459 rc = kstrtoul(buffer, 10, &val);
463 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
466 spin_lock(&svc->srv_lock);
467 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
468 spin_unlock(&svc->srv_lock);
472 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
474 spin_unlock(&svc->srv_lock);
478 LUSTRE_RW_ATTR(threads_max);
481 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
483 * \param[in] state The policy state
485 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
490 case NRS_POL_STATE_INVALID:
492 case NRS_POL_STATE_STOPPED:
494 case NRS_POL_STATE_STOPPING:
496 case NRS_POL_STATE_STARTING:
498 case NRS_POL_STATE_STARTED:
504 * Obtains status information for \a policy.
506 * Information is copied in \a info.
508 * \param[in] policy The policy
509 * \param[out] info Holds returned status information
511 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
512 struct ptlrpc_nrs_pol_info *info)
514 LASSERT(policy != NULL);
515 LASSERT(info != NULL);
516 assert_spin_locked(&policy->pol_nrs->nrs_lock);
518 CLASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
519 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
520 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
522 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
523 info->pi_state = policy->pol_state;
525 * XXX: These are accessed without holding
526 * ptlrpc_service_part::scp_req_lock.
528 info->pi_req_queued = policy->pol_req_queued;
529 info->pi_req_started = policy->pol_req_started;
533 * Reads and prints policy status information for all policies of a PTLRPC
536 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
538 struct ptlrpc_service *svc = m->private;
539 struct ptlrpc_service_part *svcpt;
540 struct ptlrpc_nrs *nrs;
541 struct ptlrpc_nrs_policy *policy;
542 struct ptlrpc_nrs_pol_info *infos;
543 struct ptlrpc_nrs_pol_info tmp;
545 unsigned pol_idx = 0;
552 * Serialize NRS core lprocfs operations with policy registration/
555 mutex_lock(&nrs_core.nrs_mutex);
558 * Use the first service partition's regular NRS head in order to obtain
559 * the number of policies registered with NRS heads of this service. All
560 * service partitions will have the same number of policies.
562 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
564 spin_lock(&nrs->nrs_lock);
565 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
566 spin_unlock(&nrs->nrs_lock);
568 OBD_ALLOC(infos, num_pols * sizeof(*infos));
570 GOTO(out, rc = -ENOMEM);
573 ptlrpc_service_for_each_part(svcpt, i, svc) {
574 nrs = nrs_svcpt2nrs(svcpt, hp);
575 spin_lock(&nrs->nrs_lock);
579 list_for_each_entry(policy, &nrs->nrs_policy_list,
581 LASSERT(pol_idx < num_pols);
583 nrs_policy_get_info_locked(policy, &tmp);
585 * Copy values when handling the first service
589 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
591 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
593 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
594 sizeof(tmp.pi_state));
595 infos[pol_idx].pi_fallback = tmp.pi_fallback;
597 * For the rest of the service partitions
598 * sanity-check the values we get.
601 LASSERT(strncmp(infos[pol_idx].pi_name,
603 NRS_POL_NAME_MAX) == 0);
604 LASSERT(strncmp(infos[pol_idx].pi_arg,
606 sizeof(tmp.pi_arg)) == 0);
608 * Not asserting ptlrpc_nrs_pol_info::pi_state,
609 * because it may be different between
610 * instances of the same policy in different
611 * service partitions.
613 LASSERT(infos[pol_idx].pi_fallback ==
617 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
618 infos[pol_idx].pi_req_started += tmp.pi_req_started;
622 spin_unlock(&nrs->nrs_lock);
626 * Policy status information output is in YAML format.
642 * high_priority_requests:
655 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
656 "high_priority_requests:");
658 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
659 if (strlen(infos[pol_idx].pi_arg) > 0)
660 seq_printf(m, " - name: %s %s\n",
661 infos[pol_idx].pi_name,
662 infos[pol_idx].pi_arg);
664 seq_printf(m, " - name: %s\n",
665 infos[pol_idx].pi_name);
668 seq_printf(m, " state: %s\n"
671 " active: %-20d\n\n",
672 nrs_state2str(infos[pol_idx].pi_state),
673 infos[pol_idx].pi_fallback ? "yes" : "no",
674 (int)infos[pol_idx].pi_req_queued,
675 (int)infos[pol_idx].pi_req_started);
678 if (!hp && nrs_svc_has_hp(svc)) {
679 memset(infos, 0, num_pols * sizeof(*infos));
682 * Redo the processing for the service's HP NRS heads' policies.
690 OBD_FREE(infos, num_pols * sizeof(*infos));
692 mutex_unlock(&nrs_core.nrs_mutex);
698 #define LPROCFS_NRS_WR_MAX_ARG (1024)
700 * The longest valid command string is the maxium policy name size, plus the
701 * length of the " reg" substring, plus the lenght of argument
703 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
704 + LPROCFS_NRS_WR_MAX_ARG)
707 * Starts and stops a given policy on a PTLRPC service.
709 * Commands consist of the policy name, followed by an optional [reg|hp] token;
710 * if the optional token is omitted, the operation is performed on both the
711 * regular and high-priority (if the service has one) NRS head.
714 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
715 size_t count, loff_t *off)
717 struct seq_file *m = file->private_data;
718 struct ptlrpc_service *svc = m->private;
719 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
721 char *cmd_copy = NULL;
727 if (count >= LPROCFS_NRS_WR_MAX_CMD)
728 GOTO(out, rc = -EINVAL);
730 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
732 GOTO(out, rc = -ENOMEM);
734 * strsep() modifies its argument, so keep a copy
738 if (copy_from_user(cmd, buffer, count))
739 GOTO(out, rc = -EFAULT);
743 policy_name = strsep(&cmd, " ");
745 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
746 GOTO(out, rc = -EINVAL);
749 * No [reg|hp] token has been specified
754 queue_name = strsep(&cmd, " ");
756 * The second token is either an optional [reg|hp] string,
759 if (strcmp(queue_name, "reg") == 0)
760 queue = PTLRPC_NRS_QUEUE_REG;
761 else if (strcmp(queue_name, "hp") == 0)
762 queue = PTLRPC_NRS_QUEUE_HP;
771 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
772 GOTO(out, rc = -ENODEV);
773 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
774 queue = PTLRPC_NRS_QUEUE_REG;
777 * Serialize NRS core lprocfs operations with policy registration/
780 mutex_lock(&nrs_core.nrs_mutex);
782 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
783 PTLRPC_NRS_CTL_START,
786 mutex_unlock(&nrs_core.nrs_mutex);
789 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
791 RETURN(rc < 0 ? rc : count);
794 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs);
798 struct ptlrpc_srh_iterator {
801 struct ptlrpc_request *srhi_req;
805 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
806 struct ptlrpc_srh_iterator *srhi,
810 struct ptlrpc_request *req;
812 if (srhi->srhi_req != NULL &&
813 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
814 srhi->srhi_seq <= seq) {
815 /* If srhi_req was set previously, hasn't been culled and
816 * we're searching for a seq on or after it (i.e. more
817 * recent), search from it onwards.
818 * Since the service history is LRU (i.e. culled reqs will
819 * be near the head), we shouldn't have to do long
821 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
822 "%s:%d: seek seq %llu, request seq %llu\n",
823 svcpt->scp_service->srv_name, svcpt->scp_cpt,
824 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
825 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
826 "%s:%d: seek offset %llu, request seq %llu, "
827 "last culled %llu\n",
828 svcpt->scp_service->srv_name, svcpt->scp_cpt,
829 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
830 e = &srhi->srhi_req->rq_history_list;
832 /* search from start */
833 e = svcpt->scp_hist_reqs.next;
836 while (e != &svcpt->scp_hist_reqs) {
837 req = list_entry(e, struct ptlrpc_request, rq_history_list);
839 if (req->rq_history_seq >= seq) {
840 srhi->srhi_seq = req->rq_history_seq;
841 srhi->srhi_req = req;
851 * ptlrpc history sequence is used as "position" of seq_file, in some case,
852 * seq_read() will increase "position" to indicate reading the next
853 * element, however, low bits of history sequence are reserved for CPT id
854 * (check the details from comments before ptlrpc_req_add_history), which
855 * means seq_read() might change CPT id of history sequence and never
856 * finish reading of requests on a CPT. To make it work, we have to shift
857 * CPT id to high bits and timestamp to low bits, so seq_read() will only
858 * increase timestamp which can correctly indicate the next position.
861 /* convert seq_file pos to cpt */
862 #define PTLRPC_REQ_POS2CPT(svc, pos) \
863 ((svc)->srv_cpt_bits == 0 ? 0 : \
864 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
866 /* make up seq_file pos from cpt */
867 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
868 ((svc)->srv_cpt_bits == 0 ? 0 : \
869 (cpt) << (64 - (svc)->srv_cpt_bits))
871 /* convert sequence to position */
872 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
873 ((svc)->srv_cpt_bits == 0 ? (seq) : \
874 ((seq) >> (svc)->srv_cpt_bits) | \
875 ((seq) << (64 - (svc)->srv_cpt_bits)))
877 /* convert position to sequence */
878 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
879 ((svc)->srv_cpt_bits == 0 ? (pos) : \
880 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
881 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
884 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
886 struct ptlrpc_service *svc = s->private;
887 struct ptlrpc_service_part *svcpt;
888 struct ptlrpc_srh_iterator *srhi;
893 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
894 CWARN("Failed to read request history because size of loff_t "
895 "%d can't match size of u64\n", (int)sizeof(loff_t));
899 OBD_ALLOC(srhi, sizeof(*srhi));
904 srhi->srhi_req = NULL;
906 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
908 ptlrpc_service_for_each_part(svcpt, i, svc) {
909 if (i < cpt) /* skip */
911 if (i > cpt) /* make up the lowest position for this CPT */
912 *pos = PTLRPC_REQ_CPT2POS(svc, i);
914 mutex_lock(&svcpt->scp_mutex);
915 spin_lock(&svcpt->scp_lock);
916 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
917 PTLRPC_REQ_POS2SEQ(svc, *pos));
918 spin_unlock(&svcpt->scp_lock);
919 mutex_unlock(&svcpt->scp_mutex);
921 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
927 OBD_FREE(srhi, sizeof(*srhi));
932 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
934 struct ptlrpc_srh_iterator *srhi = iter;
937 OBD_FREE(srhi, sizeof(*srhi));
941 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
942 void *iter, loff_t *pos)
944 struct ptlrpc_service *svc = s->private;
945 struct ptlrpc_srh_iterator *srhi = iter;
946 struct ptlrpc_service_part *svcpt;
951 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
952 svcpt = svc->srv_parts[i];
954 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
955 srhi->srhi_req = NULL;
956 seq = srhi->srhi_seq = 0;
957 } else { /* the next sequence */
958 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
961 mutex_lock(&svcpt->scp_mutex);
962 spin_lock(&svcpt->scp_lock);
963 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
964 spin_unlock(&svcpt->scp_lock);
965 mutex_unlock(&svcpt->scp_mutex);
967 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
973 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!!! */
985 struct seq_file *sf = seq_file;
987 switch (req->rq_phase) {
989 /* still awaiting a service thread's attention, or rejected
990 * because the generic request message didn't unpack */
991 seq_printf(sf, "<not swabbed>\n");
993 case RQ_PHASE_INTERPRET:
994 /* being handled, so basic msg swabbed, and opc is valid
995 * but racing with mds_handle() */
996 case RQ_PHASE_COMPLETE:
997 /* been handled by mds_handle() reply state possibly still
999 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
1002 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
1005 EXPORT_SYMBOL(target_print_req);
1007 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1009 struct ptlrpc_service *svc = s->private;
1010 struct ptlrpc_srh_iterator *srhi = iter;
1011 struct ptlrpc_service_part *svcpt;
1012 struct ptlrpc_request *req;
1015 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1017 svcpt = svc->srv_parts[srhi->srhi_idx];
1019 mutex_lock(&svcpt->scp_mutex);
1020 spin_lock(&svcpt->scp_lock);
1022 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1025 struct timespec64 arrival, sent, arrivaldiff;
1026 char nidstr[LNET_NIDSTR_SIZE];
1028 req = srhi->srhi_req;
1030 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
1031 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1032 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1033 sent.tv_sec = req->rq_sent;
1035 arrivaldiff = timespec64_sub(sent, arrival);
1037 /* Print common req fields.
1038 * CAVEAT EMPTOR: we're racing with the service handler
1039 * here. The request could contain any old crap, so you
1040 * must be just as careful as the service's request
1041 * parser. Currently I only print stuff here I know is OK
1042 * to look at coz it was set up in request_in_callback()!!!
1044 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1045 req->rq_history_seq, nidstr,
1046 libcfs_id2str(req->rq_peer), req->rq_xid,
1047 req->rq_reqlen, ptlrpc_rqphase2str(req),
1048 (s64)req->rq_arrival_time.tv_sec,
1049 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1050 (s64)arrivaldiff.tv_sec,
1051 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1052 (s64)(req->rq_sent - req->rq_deadline));
1053 if (svc->srv_ops.so_req_printer == NULL)
1054 seq_printf(s, "\n");
1056 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1059 spin_unlock(&svcpt->scp_lock);
1060 mutex_unlock(&svcpt->scp_mutex);
1066 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1068 static struct seq_operations sops = {
1069 .start = ptlrpc_lprocfs_svc_req_history_start,
1070 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1071 .next = ptlrpc_lprocfs_svc_req_history_next,
1072 .show = ptlrpc_lprocfs_svc_req_history_show,
1074 struct seq_file *seqf;
1077 rc = LPROCFS_ENTRY_CHECK(inode);
1081 rc = seq_open(file, &sops);
1085 seqf = file->private_data;
1086 seqf->private = inode->i_private;
1090 /* See also lprocfs_rd_timeouts */
1091 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1093 struct ptlrpc_service *svc = m->private;
1094 struct ptlrpc_service_part *svcpt;
1101 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1106 ptlrpc_service_for_each_part(svcpt, i, svc) {
1107 cur = at_get(&svcpt->scp_at_estimate);
1108 worst = svcpt->scp_at_estimate.at_worst_ever;
1109 worstt = svcpt->scp_at_estimate.at_worst_time;
1111 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1112 "service", cur, worst, (s64)worstt,
1113 (s64)(ktime_get_real_seconds() - worstt));
1115 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1121 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1123 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1124 struct attribute *attr,
1127 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1130 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1133 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1134 struct attribute *attr,
1138 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1143 rc = kstrtoul(buffer, 10, &val);
1147 spin_lock(&svc->srv_lock);
1148 svc->srv_hpreq_ratio = val;
1149 spin_unlock(&svc->srv_lock);
1153 LUSTRE_RW_ATTR(high_priority_ratio);
1155 static struct attribute *ptlrpc_svc_attrs[] = {
1156 &lustre_attr_threads_min.attr,
1157 &lustre_attr_threads_started.attr,
1158 &lustre_attr_threads_max.attr,
1159 &lustre_attr_high_priority_ratio.attr,
1163 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1165 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1168 complete(&svc->srv_kobj_unregister);
1171 static struct kobj_type ptlrpc_svc_ktype = {
1172 .default_attrs = ptlrpc_svc_attrs,
1173 .sysfs_ops = &lustre_sysfs_ops,
1174 .release = ptlrpc_sysfs_svc_release,
1177 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1179 /* Let's see if we had a chance at initialization first */
1180 if (svc->srv_kobj.kset) {
1181 kobject_put(&svc->srv_kobj);
1182 wait_for_completion(&svc->srv_kobj_unregister);
1186 int ptlrpc_sysfs_register_service(struct kset *parent,
1187 struct ptlrpc_service *svc)
1189 svc->srv_kobj.kset = parent;
1190 init_completion(&svc->srv_kobj_unregister);
1191 return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1192 &parent->kobj, "%s", svc->srv_name);
1195 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1196 struct ptlrpc_service *svc)
1198 struct lprocfs_vars lproc_vars[] = {
1199 { .name = "req_buffer_history_len",
1200 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1202 { .name = "req_buffer_history_max",
1203 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1205 { .name = "timeouts",
1206 .fops = &ptlrpc_lprocfs_timeouts_fops,
1208 { .name = "nrs_policies",
1209 .fops = &ptlrpc_lprocfs_nrs_fops,
1211 { .name = "req_buffers_max",
1212 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1216 static struct file_operations req_history_fops = {
1217 .owner = THIS_MODULE,
1218 .open = ptlrpc_lprocfs_svc_req_history_open,
1220 .llseek = seq_lseek,
1221 .release = lprocfs_seq_release,
1226 ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1227 &svc->srv_debugfs_entry, &svc->srv_stats);
1228 if (IS_ERR_OR_NULL(svc->srv_debugfs_entry))
1231 ldebugfs_add_vars(svc->srv_debugfs_entry, lproc_vars, NULL);
1233 rc = ldebugfs_seq_create(svc->srv_debugfs_entry, "req_history",
1234 0400, &req_history_fops, svc);
1236 CWARN("Error adding the req_history file\n");
1239 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1241 ptlrpc_ldebugfs_register(obddev->obd_debugfs_entry, NULL, "stats",
1242 &obddev->obd_svc_debugfs_entry,
1243 &obddev->obd_svc_stats);
1245 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1247 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1249 struct lprocfs_stats *svc_stats;
1250 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1251 int opc = opcode_offset(op);
1253 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1254 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)
1268 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1271 idx = lustre_msg_get_opc(req->rq_reqmsg);
1274 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1277 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1280 LASSERTF(0, "unsupported opcode %u\n", idx);
1284 lprocfs_counter_add(svc_stats, idx, bytes);
1287 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1289 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1291 if (!IS_ERR_OR_NULL(svc->srv_debugfs_entry))
1292 ldebugfs_remove(&svc->srv_debugfs_entry);
1295 lprocfs_free_stats(&svc->srv_stats);
1298 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1300 /* cleanup first to allow concurrent access to device's
1301 * stats via debugfs to complete safely
1303 lprocfs_obd_cleanup(obd);
1305 if (!IS_ERR_OR_NULL(obd->obd_svc_debugfs_entry))
1306 ldebugfs_remove(&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 ptlrpc_request *req;
1322 LPROCFS_CLIMP_CHECK(obd);
1323 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1324 LPROCFS_CLIMP_EXIT(obd);
1328 req->rq_send_state = LUSTRE_IMP_FULL;
1330 rc = ptlrpc_queue_wait(req);
1331 ptlrpc_req_finished(req);
1335 EXPORT_SYMBOL(ping_show);
1337 /* kept for older verison of tools. */
1338 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1339 const char *buffer, size_t count)
1341 return ping_show(kobj, attr, (char *)buffer);
1343 EXPORT_SYMBOL(ping_store);
1345 /* Write the connection UUID to this file to attempt to connect to that node.
1346 * The connection UUID is a node's primary NID. For example,
1347 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1350 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1351 size_t count, loff_t *off)
1353 struct seq_file *m = file->private_data;
1354 struct obd_device *obd = m->private;
1355 struct obd_import *imp = obd->u.cli.cl_import;
1360 const char prefix[] = "connection=";
1361 const int prefix_len = sizeof(prefix) - 1;
1363 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1366 OBD_ALLOC(kbuf, count + 1);
1370 if (copy_from_user(kbuf, buffer, count))
1371 GOTO(out, count = -EFAULT);
1375 /* only support connection=uuid::instance now */
1376 if (strncmp(prefix, kbuf, prefix_len) != 0)
1377 GOTO(out, count = -EINVAL);
1379 uuid = kbuf + prefix_len;
1380 ptr = strstr(uuid, "::");
1387 ptr += 2; /* Skip :: */
1388 rc = kstrtouint(ptr, 10, &inst);
1390 CERROR("config: wrong instance # %s\n", ptr);
1391 } else if (inst != imp->imp_connect_data.ocd_instance) {
1392 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1393 "target(%u/%u), reconnecting...\n",
1394 imp->imp_obd->obd_name,
1395 imp->imp_connect_data.ocd_instance, inst);
1398 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1400 imp->imp_obd->obd_name, inst);
1405 ptlrpc_recover_import(imp, uuid, 1);
1408 OBD_FREE(kbuf, count + 1);
1411 EXPORT_SYMBOL(lprocfs_import_seq_write);
1413 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1415 struct obd_device *obd = m->private;
1416 struct obd_import *imp = obd->u.cli.cl_import;
1418 LPROCFS_CLIMP_CHECK(obd);
1419 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1420 LPROCFS_CLIMP_EXIT(obd);
1423 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1426 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1427 size_t count, loff_t *off)
1429 struct seq_file *m = file->private_data;
1430 struct obd_device *obd = m->private;
1431 struct client_obd *cli = &obd->u.cli;
1432 struct obd_import *imp = cli->cl_import;
1436 rc = kstrtobool_from_user(buffer, count, &val);
1440 LPROCFS_CLIMP_CHECK(obd);
1441 spin_lock(&imp->imp_lock);
1442 imp->imp_no_pinger_recover = !val;
1443 spin_unlock(&imp->imp_lock);
1444 LPROCFS_CLIMP_EXIT(obd);
1447 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);