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, 2016, 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/lustre_idl.h>
39 #include <lustre_net.h>
40 #include <obd_class.h>
41 #include "ptlrpc_internal.h"
44 static struct ll_rpc_opcode {
47 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
48 { OST_REPLY, "ost_reply" },
49 { OST_GETATTR, "ost_getattr" },
50 { OST_SETATTR, "ost_setattr" },
51 { OST_READ, "ost_read" },
52 { OST_WRITE, "ost_write" },
53 { OST_CREATE , "ost_create" },
54 { OST_DESTROY, "ost_destroy" },
55 { OST_GET_INFO, "ost_get_info" },
56 { OST_CONNECT, "ost_connect" },
57 { OST_DISCONNECT, "ost_disconnect" },
58 { OST_PUNCH, "ost_punch" },
59 { OST_OPEN, "ost_open" },
60 { OST_CLOSE, "ost_close" },
61 { OST_STATFS, "ost_statfs" },
62 { 14, NULL }, /* formerly OST_SAN_READ */
63 { 15, NULL }, /* formerly OST_SAN_WRITE */
64 { OST_SYNC, "ost_sync" },
65 { OST_SET_INFO, "ost_set_info" },
66 { OST_QUOTACHECK, "ost_quotacheck" },
67 { OST_QUOTACTL, "ost_quotactl" },
68 { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
69 { OST_LADVISE, "ost_ladvise" },
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 { 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 { OBD_LOG_CANCEL, "llog_cancel" },
115 { 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 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
121 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
122 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
123 { 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 { BRW_READ_BYTES, "read_bytes" },
156 { BRW_WRITE_BYTES, "write_bytes" },
159 const char *ll_opcode2str(__u32 opcode)
161 /* When one of the assertions below fail, chances are that:
162 * 1) A new opcode was added in include/lustre/lustre_idl.h,
163 * but is missing from the table above.
164 * or 2) The opcode space was renumbered or rearranged,
165 * and the opcode_offset() function in
166 * ptlrpc_internal.h needs to be modified.
168 __u32 offset = opcode_offset(opcode);
169 LASSERTF(offset < LUSTRE_MAX_OPCODES,
170 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
171 offset, LUSTRE_MAX_OPCODES);
172 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
173 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
174 offset, ll_rpc_opcode_table[offset].opcode, opcode);
175 return ll_rpc_opcode_table[offset].opname;
178 const int ll_str2opcode(const char *ops)
182 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
183 if (ll_rpc_opcode_table[i].opname != NULL &&
184 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
185 return ll_rpc_opcode_table[i].opcode;
191 static const char *ll_eopcode2str(__u32 opcode)
193 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
194 return ll_eopcode_table[opcode].opname;
197 #ifdef CONFIG_PROC_FS
198 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
199 char *name, struct proc_dir_entry **procroot_ret,
200 struct lprocfs_stats **stats_ret)
202 struct proc_dir_entry *svc_procroot;
203 struct lprocfs_stats *svc_stats;
205 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
208 LASSERT(*procroot_ret == NULL);
209 LASSERT(*stats_ret == NULL);
211 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
212 if (svc_stats == NULL)
216 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
217 if (IS_ERR(svc_procroot)) {
218 lprocfs_free_stats(&svc_stats);
225 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
226 svc_counter_config, "req_waittime", "usec");
227 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
228 svc_counter_config, "req_qdepth", "reqs");
229 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
230 svc_counter_config, "req_active", "reqs");
231 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
232 svc_counter_config, "req_timeout", "sec");
233 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
234 svc_counter_config, "reqbuf_avail", "bufs");
235 for (i = 0; i < EXTRA_LAST_OPC; i++) {
239 case BRW_WRITE_BYTES:
247 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
249 ll_eopcode2str(i), units);
251 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
252 __u32 opcode = ll_rpc_opcode_table[i].opcode;
253 lprocfs_counter_init(svc_stats,
254 EXTRA_MAX_OPCODES + i, svc_counter_config,
255 ll_opcode2str(opcode), "usec");
258 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
261 lprocfs_remove(&svc_procroot);
262 lprocfs_free_stats(&svc_stats);
265 *procroot_ret = svc_procroot;
266 *stats_ret = svc_stats;
271 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
273 struct ptlrpc_service *svc = m->private;
274 struct ptlrpc_service_part *svcpt;
278 ptlrpc_service_for_each_part(svcpt, i, svc)
279 total += svcpt->scp_hist_nrqbds;
281 seq_printf(m, "%d\n", total);
284 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
287 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
289 struct ptlrpc_service *svc = m->private;
290 struct ptlrpc_service_part *svcpt;
294 ptlrpc_service_for_each_part(svcpt, i, svc)
295 total += svc->srv_hist_nrqbds_cpt_max;
297 seq_printf(m, "%d\n", total);
302 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
303 const char __user *buffer,
304 size_t count, loff_t *off)
306 struct seq_file *m = file->private_data;
307 struct ptlrpc_service *svc = m->private;
312 rc = lprocfs_str_to_s64(buffer, count, &val);
316 if (val < 0 || val > INT_MAX)
319 /* This sanity check is more of an insanity check; we can still
320 * hose a kernel by allowing the request history to grow too
322 bufpages = (svc->srv_buf_size + PAGE_SIZE - 1) >>
324 if (val > totalram_pages/(2 * bufpages))
327 spin_lock(&svc->srv_lock);
330 svc->srv_hist_nrqbds_cpt_max = 0;
332 svc->srv_hist_nrqbds_cpt_max =
333 max(1, ((int)val / svc->srv_ncpts));
335 spin_unlock(&svc->srv_lock);
339 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
341 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
344 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
347 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
350 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
351 const char *buffer, size_t count)
353 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
358 rc = kstrtoul(buffer, 10, &val);
362 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
365 spin_lock(&svc->srv_lock);
366 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
367 spin_unlock(&svc->srv_lock);
371 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
373 spin_unlock(&svc->srv_lock);
377 LUSTRE_RW_ATTR(threads_min);
379 static ssize_t threads_started_show(struct kobject *kobj,
380 struct attribute *attr,
383 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
385 struct ptlrpc_service_part *svcpt;
389 ptlrpc_service_for_each_part(svcpt, i, svc)
390 total += svcpt->scp_nthrs_running;
392 return sprintf(buf, "%d\n", total);
394 LUSTRE_RO_ATTR(threads_started);
396 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
399 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
402 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
405 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
406 const char *buffer, size_t count)
408 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
413 rc = kstrtoul(buffer, 10, &val);
417 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
420 spin_lock(&svc->srv_lock);
421 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
422 spin_unlock(&svc->srv_lock);
426 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
428 spin_unlock(&svc->srv_lock);
432 LUSTRE_RW_ATTR(threads_max);
435 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
437 * \param[in] state The policy state
439 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
444 case NRS_POL_STATE_INVALID:
446 case NRS_POL_STATE_STOPPED:
448 case NRS_POL_STATE_STOPPING:
450 case NRS_POL_STATE_STARTING:
452 case NRS_POL_STATE_STARTED:
458 * Obtains status information for \a policy.
460 * Information is copied in \a info.
462 * \param[in] policy The policy
463 * \param[out] info Holds returned status information
465 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
466 struct ptlrpc_nrs_pol_info *info)
468 LASSERT(policy != NULL);
469 LASSERT(info != NULL);
470 assert_spin_locked(&policy->pol_nrs->nrs_lock);
472 LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
473 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
474 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
476 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
477 info->pi_state = policy->pol_state;
479 * XXX: These are accessed without holding
480 * ptlrpc_service_part::scp_req_lock.
482 info->pi_req_queued = policy->pol_req_queued;
483 info->pi_req_started = policy->pol_req_started;
487 * Reads and prints policy status information for all policies of a PTLRPC
490 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
492 struct ptlrpc_service *svc = m->private;
493 struct ptlrpc_service_part *svcpt;
494 struct ptlrpc_nrs *nrs;
495 struct ptlrpc_nrs_policy *policy;
496 struct ptlrpc_nrs_pol_info *infos;
497 struct ptlrpc_nrs_pol_info tmp;
499 unsigned pol_idx = 0;
506 * Serialize NRS core lprocfs operations with policy registration/
509 mutex_lock(&nrs_core.nrs_mutex);
512 * Use the first service partition's regular NRS head in order to obtain
513 * the number of policies registered with NRS heads of this service. All
514 * service partitions will have the same number of policies.
516 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
518 spin_lock(&nrs->nrs_lock);
519 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
520 spin_unlock(&nrs->nrs_lock);
522 OBD_ALLOC(infos, num_pols * sizeof(*infos));
524 GOTO(out, rc = -ENOMEM);
527 ptlrpc_service_for_each_part(svcpt, i, svc) {
528 nrs = nrs_svcpt2nrs(svcpt, hp);
529 spin_lock(&nrs->nrs_lock);
533 list_for_each_entry(policy, &nrs->nrs_policy_list,
535 LASSERT(pol_idx < num_pols);
537 nrs_policy_get_info_locked(policy, &tmp);
539 * Copy values when handling the first service
543 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
545 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
547 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
548 sizeof(tmp.pi_state));
549 infos[pol_idx].pi_fallback = tmp.pi_fallback;
551 * For the rest of the service partitions
552 * sanity-check the values we get.
555 LASSERT(strncmp(infos[pol_idx].pi_name,
557 NRS_POL_NAME_MAX) == 0);
558 LASSERT(strncmp(infos[pol_idx].pi_arg,
560 sizeof(tmp.pi_arg)) == 0);
562 * Not asserting ptlrpc_nrs_pol_info::pi_state,
563 * because it may be different between
564 * instances of the same policy in different
565 * service partitions.
567 LASSERT(infos[pol_idx].pi_fallback ==
571 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
572 infos[pol_idx].pi_req_started += tmp.pi_req_started;
576 spin_unlock(&nrs->nrs_lock);
580 * Policy status information output is in YAML format.
596 * high_priority_requests:
609 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
610 "high_priority_requests:");
612 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
613 if (strlen(infos[pol_idx].pi_arg) > 0)
614 seq_printf(m, " - name: %s %s\n",
615 infos[pol_idx].pi_name,
616 infos[pol_idx].pi_arg);
618 seq_printf(m, " - name: %s\n",
619 infos[pol_idx].pi_name);
622 seq_printf(m, " state: %s\n"
625 " active: %-20d\n\n",
626 nrs_state2str(infos[pol_idx].pi_state),
627 infos[pol_idx].pi_fallback ? "yes" : "no",
628 (int)infos[pol_idx].pi_req_queued,
629 (int)infos[pol_idx].pi_req_started);
632 if (!hp && nrs_svc_has_hp(svc)) {
633 memset(infos, 0, num_pols * sizeof(*infos));
636 * Redo the processing for the service's HP NRS heads' policies.
644 OBD_FREE(infos, num_pols * sizeof(*infos));
646 mutex_unlock(&nrs_core.nrs_mutex);
652 #define LPROCFS_NRS_WR_MAX_ARG (1024)
654 * The longest valid command string is the maxium policy name size, plus the
655 * length of the " reg" substring, plus the lenght of argument
657 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
658 + LPROCFS_NRS_WR_MAX_ARG)
661 * Starts and stops a given policy on a PTLRPC service.
663 * Commands consist of the policy name, followed by an optional [reg|hp] token;
664 * if the optional token is omitted, the operation is performed on both the
665 * regular and high-priority (if the service has one) NRS head.
668 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
669 size_t count, loff_t *off)
671 struct seq_file *m = file->private_data;
672 struct ptlrpc_service *svc = m->private;
673 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
675 char *cmd_copy = NULL;
681 if (count >= LPROCFS_NRS_WR_MAX_CMD)
682 GOTO(out, rc = -EINVAL);
684 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
686 GOTO(out, rc = -ENOMEM);
688 * strsep() modifies its argument, so keep a copy
692 if (copy_from_user(cmd, buffer, count))
693 GOTO(out, rc = -EFAULT);
697 policy_name = strsep(&cmd, " ");
699 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
700 GOTO(out, rc = -EINVAL);
703 * No [reg|hp] token has been specified
708 queue_name = strsep(&cmd, " ");
710 * The second token is either an optional [reg|hp] string,
713 if (strcmp(queue_name, "reg") == 0)
714 queue = PTLRPC_NRS_QUEUE_REG;
715 else if (strcmp(queue_name, "hp") == 0)
716 queue = PTLRPC_NRS_QUEUE_HP;
725 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
726 GOTO(out, rc = -ENODEV);
727 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
728 queue = PTLRPC_NRS_QUEUE_REG;
731 * Serialize NRS core lprocfs operations with policy registration/
734 mutex_lock(&nrs_core.nrs_mutex);
736 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
737 PTLRPC_NRS_CTL_START,
740 mutex_unlock(&nrs_core.nrs_mutex);
743 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
745 RETURN(rc < 0 ? rc : count);
747 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
751 struct ptlrpc_srh_iterator {
754 struct ptlrpc_request *srhi_req;
758 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
759 struct ptlrpc_srh_iterator *srhi,
763 struct ptlrpc_request *req;
765 if (srhi->srhi_req != NULL &&
766 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
767 srhi->srhi_seq <= seq) {
768 /* If srhi_req was set previously, hasn't been culled and
769 * we're searching for a seq on or after it (i.e. more
770 * recent), search from it onwards.
771 * Since the service history is LRU (i.e. culled reqs will
772 * be near the head), we shouldn't have to do long
774 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
775 "%s:%d: seek seq %llu, request seq %llu\n",
776 svcpt->scp_service->srv_name, svcpt->scp_cpt,
777 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
778 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
779 "%s:%d: seek offset %llu, request seq %llu, "
780 "last culled %llu\n",
781 svcpt->scp_service->srv_name, svcpt->scp_cpt,
782 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
783 e = &srhi->srhi_req->rq_history_list;
785 /* search from start */
786 e = svcpt->scp_hist_reqs.next;
789 while (e != &svcpt->scp_hist_reqs) {
790 req = list_entry(e, struct ptlrpc_request, rq_history_list);
792 if (req->rq_history_seq >= seq) {
793 srhi->srhi_seq = req->rq_history_seq;
794 srhi->srhi_req = req;
804 * ptlrpc history sequence is used as "position" of seq_file, in some case,
805 * seq_read() will increase "position" to indicate reading the next
806 * element, however, low bits of history sequence are reserved for CPT id
807 * (check the details from comments before ptlrpc_req_add_history), which
808 * means seq_read() might change CPT id of history sequence and never
809 * finish reading of requests on a CPT. To make it work, we have to shift
810 * CPT id to high bits and timestamp to low bits, so seq_read() will only
811 * increase timestamp which can correctly indicate the next position.
814 /* convert seq_file pos to cpt */
815 #define PTLRPC_REQ_POS2CPT(svc, pos) \
816 ((svc)->srv_cpt_bits == 0 ? 0 : \
817 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
819 /* make up seq_file pos from cpt */
820 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
821 ((svc)->srv_cpt_bits == 0 ? 0 : \
822 (cpt) << (64 - (svc)->srv_cpt_bits))
824 /* convert sequence to position */
825 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
826 ((svc)->srv_cpt_bits == 0 ? (seq) : \
827 ((seq) >> (svc)->srv_cpt_bits) | \
828 ((seq) << (64 - (svc)->srv_cpt_bits)))
830 /* convert position to sequence */
831 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
832 ((svc)->srv_cpt_bits == 0 ? (pos) : \
833 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
834 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
837 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
839 struct ptlrpc_service *svc = s->private;
840 struct ptlrpc_service_part *svcpt;
841 struct ptlrpc_srh_iterator *srhi;
846 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
847 CWARN("Failed to read request history because size of loff_t "
848 "%d can't match size of u64\n", (int)sizeof(loff_t));
852 OBD_ALLOC(srhi, sizeof(*srhi));
857 srhi->srhi_req = NULL;
859 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
861 ptlrpc_service_for_each_part(svcpt, i, svc) {
862 if (i < cpt) /* skip */
864 if (i > cpt) /* make up the lowest position for this CPT */
865 *pos = PTLRPC_REQ_CPT2POS(svc, i);
867 spin_lock(&svcpt->scp_lock);
868 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
869 PTLRPC_REQ_POS2SEQ(svc, *pos));
870 spin_unlock(&svcpt->scp_lock);
872 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
878 OBD_FREE(srhi, sizeof(*srhi));
883 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
885 struct ptlrpc_srh_iterator *srhi = iter;
888 OBD_FREE(srhi, sizeof(*srhi));
892 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
893 void *iter, loff_t *pos)
895 struct ptlrpc_service *svc = s->private;
896 struct ptlrpc_srh_iterator *srhi = iter;
897 struct ptlrpc_service_part *svcpt;
902 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
903 svcpt = svc->srv_parts[i];
905 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
906 srhi->srhi_req = NULL;
907 seq = srhi->srhi_seq = 0;
908 } else { /* the next sequence */
909 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
912 spin_lock(&svcpt->scp_lock);
913 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
914 spin_unlock(&svcpt->scp_lock);
916 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
922 OBD_FREE(srhi, sizeof(*srhi));
926 /* common ost/mdt so_req_printer */
927 void target_print_req(void *seq_file, struct ptlrpc_request *req)
929 /* Called holding srv_lock with irqs disabled.
930 * Print specific req contents and a newline.
931 * CAVEAT EMPTOR: check request message length before printing!!!
932 * You might have received any old crap so you must be just as
933 * careful here as the service's request parser!!! */
934 struct seq_file *sf = seq_file;
936 switch (req->rq_phase) {
938 /* still awaiting a service thread's attention, or rejected
939 * because the generic request message didn't unpack */
940 seq_printf(sf, "<not swabbed>\n");
942 case RQ_PHASE_INTERPRET:
943 /* being handled, so basic msg swabbed, and opc is valid
944 * but racing with mds_handle() */
945 case RQ_PHASE_COMPLETE:
946 /* been handled by mds_handle() reply state possibly still
948 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
951 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
954 EXPORT_SYMBOL(target_print_req);
956 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
958 struct ptlrpc_service *svc = s->private;
959 struct ptlrpc_srh_iterator *srhi = iter;
960 struct ptlrpc_service_part *svcpt;
961 struct ptlrpc_request *req;
964 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
966 svcpt = svc->srv_parts[srhi->srhi_idx];
968 spin_lock(&svcpt->scp_lock);
970 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
973 struct timespec64 arrival, sent, arrivaldiff;
974 char nidstr[LNET_NIDSTR_SIZE];
976 req = srhi->srhi_req;
978 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
979 arrival.tv_sec = req->rq_arrival_time.tv_sec;
980 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
981 sent.tv_sec = req->rq_sent;
983 arrivaldiff = timespec64_sub(sent, arrival);
985 /* Print common req fields.
986 * CAVEAT EMPTOR: we're racing with the service handler
987 * here. The request could contain any old crap, so you
988 * must be just as careful as the service's request
989 * parser. Currently I only print stuff here I know is OK
990 * to look at coz it was set up in request_in_callback()!!!
992 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
993 req->rq_history_seq, nidstr,
994 libcfs_id2str(req->rq_peer), req->rq_xid,
995 req->rq_reqlen, ptlrpc_rqphase2str(req),
996 (s64)req->rq_arrival_time.tv_sec,
997 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
998 (s64)arrivaldiff.tv_sec,
999 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1000 (s64)(req->rq_sent - req->rq_deadline));
1001 if (svc->srv_ops.so_req_printer == NULL)
1002 seq_printf(s, "\n");
1004 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1007 spin_unlock(&svcpt->scp_lock);
1012 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1014 static struct seq_operations sops = {
1015 .start = ptlrpc_lprocfs_svc_req_history_start,
1016 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1017 .next = ptlrpc_lprocfs_svc_req_history_next,
1018 .show = ptlrpc_lprocfs_svc_req_history_show,
1020 struct seq_file *seqf;
1023 rc = LPROCFS_ENTRY_CHECK(inode);
1027 rc = seq_open(file, &sops);
1031 seqf = file->private_data;
1032 seqf->private = PDE_DATA(inode);
1036 /* See also lprocfs_rd_timeouts */
1037 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1039 struct ptlrpc_service *svc = m->private;
1040 struct ptlrpc_service_part *svcpt;
1047 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1052 ptlrpc_service_for_each_part(svcpt, i, svc) {
1053 cur = at_get(&svcpt->scp_at_estimate);
1054 worst = svcpt->scp_at_estimate.at_worst_ever;
1055 worstt = svcpt->scp_at_estimate.at_worst_time;
1057 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1058 "service", cur, worst, (s64)worstt,
1059 (s64)(ktime_get_real_seconds() - worstt));
1061 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1066 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1068 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1069 struct attribute *attr,
1072 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1075 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1078 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1079 struct attribute *attr,
1083 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1088 rc = kstrtoul(buffer, 10, &val);
1092 spin_lock(&svc->srv_lock);
1093 svc->srv_hpreq_ratio = val;
1094 spin_unlock(&svc->srv_lock);
1098 LUSTRE_RW_ATTR(high_priority_ratio);
1100 static struct attribute *ptlrpc_svc_attrs[] = {
1101 &lustre_attr_threads_min.attr,
1102 &lustre_attr_threads_started.attr,
1103 &lustre_attr_threads_max.attr,
1104 &lustre_attr_high_priority_ratio.attr,
1108 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1110 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1113 complete(&svc->srv_kobj_unregister);
1116 static struct kobj_type ptlrpc_svc_ktype = {
1117 .default_attrs = ptlrpc_svc_attrs,
1118 .sysfs_ops = &lustre_sysfs_ops,
1119 .release = ptlrpc_sysfs_svc_release,
1122 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1124 /* Let's see if we had a chance at initialization first */
1125 if (svc->srv_kobj.kset) {
1126 kobject_put(&svc->srv_kobj);
1127 wait_for_completion(&svc->srv_kobj_unregister);
1131 int ptlrpc_sysfs_register_service(struct kset *parent,
1132 struct ptlrpc_service *svc)
1136 svc->srv_kobj.kset = parent;
1137 init_completion(&svc->srv_kobj_unregister);
1138 rc = kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype, NULL,
1139 "%s", svc->srv_name);
1144 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1145 struct ptlrpc_service *svc)
1147 struct lprocfs_vars lproc_vars[] = {
1148 { .name = "req_buffer_history_len",
1149 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1151 { .name = "req_buffer_history_max",
1152 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1154 { .name = "timeouts",
1155 .fops = &ptlrpc_lprocfs_timeouts_fops,
1157 { .name = "nrs_policies",
1158 .fops = &ptlrpc_lprocfs_nrs_fops,
1162 static struct file_operations req_history_fops = {
1163 .owner = THIS_MODULE,
1164 .open = ptlrpc_lprocfs_svc_req_history_open,
1166 .llseek = seq_lseek,
1167 .release = lprocfs_seq_release,
1172 ptlrpc_lprocfs_register(entry, svc->srv_name,
1173 "stats", &svc->srv_procroot,
1175 if (svc->srv_procroot == NULL)
1178 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1180 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1181 0400, &req_history_fops, svc);
1183 CWARN("Error adding the req_history file\n");
1186 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1188 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1189 &obddev->obd_svc_procroot,
1190 &obddev->obd_svc_stats);
1192 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1194 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1196 struct lprocfs_stats *svc_stats;
1197 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1198 int opc = opcode_offset(op);
1200 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1201 if (svc_stats == NULL || opc <= 0)
1203 LASSERT(opc < LUSTRE_MAX_OPCODES);
1204 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1205 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1208 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1210 struct lprocfs_stats *svc_stats;
1213 if (!req->rq_import)
1215 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1218 idx = lustre_msg_get_opc(req->rq_reqmsg);
1221 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1224 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1227 LASSERTF(0, "unsupported opcode %u\n", idx);
1231 lprocfs_counter_add(svc_stats, idx, bytes);
1234 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1236 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1238 if (svc->srv_procroot != NULL)
1239 lprocfs_remove(&svc->srv_procroot);
1242 lprocfs_free_stats(&svc->srv_stats);
1245 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1247 if (obd->obd_svc_procroot)
1248 lprocfs_remove(&obd->obd_svc_procroot);
1250 if (obd->obd_svc_stats)
1251 lprocfs_free_stats(&obd->obd_svc_stats);
1253 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1256 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1257 size_t count, loff_t *off)
1259 struct seq_file *m = file->private_data;
1260 struct obd_device *obd = m->private;
1261 struct ptlrpc_request *req;
1265 LPROCFS_CLIMP_CHECK(obd);
1266 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1267 LPROCFS_CLIMP_EXIT(obd);
1271 req->rq_send_state = LUSTRE_IMP_FULL;
1273 rc = ptlrpc_queue_wait(req);
1275 ptlrpc_req_finished(req);
1280 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1282 /* Write the connection UUID to this file to attempt to connect to that node.
1283 * The connection UUID is a node's primary NID. For example,
1284 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1287 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1288 size_t count, loff_t *off)
1290 struct seq_file *m = file->private_data;
1291 struct obd_device *obd = m->private;
1292 struct obd_import *imp = obd->u.cli.cl_import;
1297 const char prefix[] = "connection=";
1298 const int prefix_len = sizeof(prefix) - 1;
1300 if (count > PAGE_SIZE - 1 || count <= prefix_len)
1303 OBD_ALLOC(kbuf, count + 1);
1307 if (copy_from_user(kbuf, buffer, count))
1308 GOTO(out, count = -EFAULT);
1312 /* only support connection=uuid::instance now */
1313 if (strncmp(prefix, kbuf, prefix_len) != 0)
1314 GOTO(out, count = -EINVAL);
1316 uuid = kbuf + prefix_len;
1317 ptr = strstr(uuid, "::");
1324 ptr += 2; /* Skip :: */
1325 inst = simple_strtol(ptr, &endptr, 10);
1327 CERROR("config: wrong instance # %s\n", ptr);
1328 } else if (inst != imp->imp_connect_data.ocd_instance) {
1329 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1330 "target(%u/%u), reconnecting...\n",
1331 imp->imp_obd->obd_name,
1332 imp->imp_connect_data.ocd_instance, inst);
1335 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1337 imp->imp_obd->obd_name, inst);
1342 ptlrpc_recover_import(imp, uuid, 1);
1345 OBD_FREE(kbuf, count + 1);
1348 EXPORT_SYMBOL(lprocfs_import_seq_write);
1350 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1352 struct obd_device *obd = m->private;
1353 struct obd_import *imp = obd->u.cli.cl_import;
1355 LPROCFS_CLIMP_CHECK(obd);
1356 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1357 LPROCFS_CLIMP_EXIT(obd);
1360 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1363 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1364 size_t count, loff_t *off)
1366 struct seq_file *m = file->private_data;
1367 struct obd_device *obd = m->private;
1368 struct client_obd *cli = &obd->u.cli;
1369 struct obd_import *imp = cli->cl_import;
1373 rc = lprocfs_str_to_s64(buffer, count, &val);
1377 if (val != 0 && val != 1)
1380 LPROCFS_CLIMP_CHECK(obd);
1381 spin_lock(&imp->imp_lock);
1382 imp->imp_no_pinger_recover = !val;
1383 spin_unlock(&imp->imp_lock);
1384 LPROCFS_CLIMP_EXIT(obd);
1387 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1389 #endif /* CONFIG_PROC_FS */