4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 #define DEBUG_SUBSYSTEM S_CLASS
35 #include <obd_support.h>
37 #include <lprocfs_status.h>
38 #include <lustre_net.h>
39 #include <obd_class.h>
40 #include "ptlrpc_internal.h"
43 static struct ll_rpc_opcode {
46 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
47 { OST_REPLY, "ost_reply" },
48 { OST_GETATTR, "ost_getattr" },
49 { OST_SETATTR, "ost_setattr" },
50 { OST_READ, "ost_read" },
51 { OST_WRITE, "ost_write" },
52 { OST_CREATE , "ost_create" },
53 { OST_DESTROY, "ost_destroy" },
54 { OST_GET_INFO, "ost_get_info" },
55 { OST_CONNECT, "ost_connect" },
56 { OST_DISCONNECT, "ost_disconnect" },
57 { OST_PUNCH, "ost_punch" },
58 { OST_OPEN, "ost_open" },
59 { OST_CLOSE, "ost_close" },
60 { OST_STATFS, "ost_statfs" },
61 { 14, NULL }, /* formerly OST_SAN_READ */
62 { 15, NULL }, /* formerly OST_SAN_WRITE */
63 { OST_SYNC, "ost_sync" },
64 { OST_SET_INFO, "ost_set_info" },
65 { OST_QUOTACHECK, "ost_quotacheck" },
66 { OST_QUOTACTL, "ost_quotactl" },
67 { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
68 { OST_LADVISE, "ost_ladvise" },
69 { OST_FALLOCATE, "ost_fallocate"},
70 { MDS_GETATTR, "mds_getattr" },
71 { MDS_GETATTR_NAME, "mds_getattr_lock" },
72 { MDS_CLOSE, "mds_close" },
73 { MDS_REINT, "mds_reint" },
74 { MDS_READPAGE, "mds_readpage" },
75 { MDS_CONNECT, "mds_connect" },
76 { MDS_DISCONNECT, "mds_disconnect" },
77 { MDS_GET_ROOT, "mds_get_root" },
78 { MDS_STATFS, "mds_statfs" },
79 { MDS_PIN, "mds_pin" },
80 { MDS_UNPIN, "mds_unpin" },
81 { MDS_SYNC, "mds_sync" },
82 { MDS_DONE_WRITING, "mds_done_writing" },
83 { MDS_SET_INFO, "mds_set_info" },
84 { MDS_QUOTACHECK, "mds_quotacheck" },
85 { MDS_QUOTACTL, "mds_quotactl" },
86 { MDS_GETXATTR, "mds_getxattr" },
87 { MDS_SETXATTR, "mds_setxattr" },
88 { MDS_WRITEPAGE, "mds_writepage" },
89 { MDS_IS_SUBDIR, "mds_is_subdir" },
90 { MDS_GET_INFO, "mds_get_info" },
91 { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
92 { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
93 { MDS_HSM_ACTION, "mds_hsm_action" },
94 { MDS_HSM_PROGRESS, "mds_hsm_progress" },
95 { MDS_HSM_REQUEST, "mds_hsm_request" },
96 { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
97 { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
98 { MDS_SWAP_LAYOUTS, "mds_swap_layouts" },
99 { MDS_RMFID, "mds_rmfid" },
100 { LDLM_ENQUEUE, "ldlm_enqueue" },
101 { LDLM_CONVERT, "ldlm_convert" },
102 { LDLM_CANCEL, "ldlm_cancel" },
103 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
104 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
105 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
106 { LDLM_SET_INFO, "ldlm_set_info" },
107 { MGS_CONNECT, "mgs_connect" },
108 { MGS_DISCONNECT, "mgs_disconnect" },
109 { MGS_EXCEPTION, "mgs_exception" },
110 { MGS_TARGET_REG, "mgs_target_reg" },
111 { MGS_TARGET_DEL, "mgs_target_del" },
112 { MGS_SET_INFO, "mgs_set_info" },
113 { MGS_CONFIG_READ, "mgs_config_read" },
114 { OBD_PING, "obd_ping" },
115 { 401, /* was OBD_LOG_CANCEL */ "llog_cancel" },
116 { 402, /* was OBD_QC_CALLBACK */ "obd_quota_callback" },
117 { OBD_IDX_READ, "dt_index_read" },
118 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
119 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
120 { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
121 { 504, /*LLOG_ORIGIN_HANDLE_WRITE_REC*/"llog_origin_handle_write_rec" },
122 { 505, /* was LLOG_ORIGIN_HANDLE_CLOSE */ "llog_origin_handle_close" },
123 { 506, /* was LLOG_ORIGIN_CONNECT */ "llog_origin_connect" },
124 { 507, /* was LLOG_CATINFO */ "llog_catinfo" },
125 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
126 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
127 { QUOTA_DQACQ, "quota_acquire" },
128 { QUOTA_DQREL, "quota_release" },
129 { SEQ_QUERY, "seq_query" },
130 { SEC_CTX_INIT, "sec_ctx_init" },
131 { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
132 { SEC_CTX_FINI, "sec_ctx_fini" },
133 { FLD_QUERY, "fld_query" },
134 { FLD_READ, "fld_read" },
135 { OUT_UPDATE, "out_update" },
136 { LFSCK_NOTIFY, "lfsck_notify" },
137 { LFSCK_QUERY, "lfsck_query" },
140 static struct ll_eopcode {
143 } ll_eopcode_table[EXTRA_LAST_OPC] = {
144 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
145 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
146 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
147 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
148 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
149 { MDS_REINT_SETATTR, "mds_reint_setattr" },
150 { MDS_REINT_CREATE, "mds_reint_create" },
151 { MDS_REINT_LINK, "mds_reint_link" },
152 { MDS_REINT_UNLINK, "mds_reint_unlink" },
153 { MDS_REINT_RENAME, "mds_reint_rename" },
154 { MDS_REINT_OPEN, "mds_reint_open" },
155 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
156 { MDS_REINT_RESYNC, "mds_reint_resync" },
157 { BRW_READ_BYTES, "read_bytes" },
158 { BRW_WRITE_BYTES, "write_bytes" },
161 const char *ll_opcode2str(__u32 opcode)
163 /* When one of the assertions below fail, chances are that:
164 * 1) A new opcode was added in include/lustre/lustre_idl.h,
165 * but is missing from the table above.
166 * or 2) The opcode space was renumbered or rearranged,
167 * and the opcode_offset() function in
168 * ptlrpc_internal.h needs to be modified.
170 __u32 offset = opcode_offset(opcode);
171 LASSERTF(offset < LUSTRE_MAX_OPCODES,
172 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
173 offset, LUSTRE_MAX_OPCODES);
174 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
175 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
176 offset, ll_rpc_opcode_table[offset].opcode, opcode);
177 return ll_rpc_opcode_table[offset].opname;
180 const int ll_str2opcode(const char *ops)
184 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
185 if (ll_rpc_opcode_table[i].opname != NULL &&
186 strcmp(ll_rpc_opcode_table[i].opname, ops) == 0)
187 return ll_rpc_opcode_table[i].opcode;
193 static const char *ll_eopcode2str(__u32 opcode)
195 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
196 return ll_eopcode_table[opcode].opname;
200 ptlrpc_ldebugfs_register(struct dentry *root, char *dir, char *name,
201 struct dentry **debugfs_root_ret,
202 struct lprocfs_stats **stats_ret)
204 struct dentry *svc_debugfs_entry;
205 struct lprocfs_stats *svc_stats;
207 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
210 LASSERT(!*debugfs_root_ret);
211 LASSERT(!*stats_ret);
213 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES,
219 svc_debugfs_entry = debugfs_create_dir(dir, root);
221 svc_debugfs_entry = root;
223 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
224 svc_counter_config, "req_waittime", "usec");
225 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
226 svc_counter_config, "req_qdepth", "reqs");
227 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
228 svc_counter_config, "req_active", "reqs");
229 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
230 svc_counter_config, "req_timeout", "sec");
231 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
232 svc_counter_config, "reqbuf_avail", "bufs");
233 for (i = 0; i < EXTRA_LAST_OPC; i++) {
237 case BRW_WRITE_BYTES:
245 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
247 ll_eopcode2str(i), units);
249 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
250 __u32 opcode = ll_rpc_opcode_table[i].opcode;
251 lprocfs_counter_init(svc_stats,
252 EXTRA_MAX_OPCODES + i, svc_counter_config,
253 ll_opcode2str(opcode), "usec");
256 rc = ldebugfs_register_stats(svc_debugfs_entry, name, svc_stats);
259 debugfs_remove_recursive(svc_debugfs_entry);
260 lprocfs_free_stats(&svc_stats);
263 *debugfs_root_ret = svc_debugfs_entry;
264 *stats_ret = svc_stats;
269 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
271 struct ptlrpc_service *svc = m->private;
272 struct ptlrpc_service_part *svcpt;
276 ptlrpc_service_for_each_part(svcpt, i, svc)
277 total += svcpt->scp_hist_nrqbds;
279 seq_printf(m, "%d\n", total);
284 LDEBUGFS_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;
308 unsigned long long val;
309 unsigned long long limit;
313 rc = kstrtoull_from_user(buffer, count, 0, &val);
317 if (val < 0 || val > INT_MAX)
320 /* This sanity check is more of an insanity check; we can still
321 * hose a kernel by allowing the request history to grow too
322 * far. The roundup to the next power of two is an empirical way
323 * to take care that request buffer is allocated in Slab and thus
324 * will be upgraded */
325 bufpages = (roundup_pow_of_two(svc->srv_buf_size) + PAGE_SIZE - 1) >>
327 limit = cfs_totalram_pages() / (2 * bufpages);
328 /* do not allow history to consume more than half max number of rqbds */
329 if ((svc->srv_nrqbds_max == 0 && val > limit) ||
330 (svc->srv_nrqbds_max != 0 && val > svc->srv_nrqbds_max / 2))
333 spin_lock(&svc->srv_lock);
336 svc->srv_hist_nrqbds_cpt_max = 0;
338 svc->srv_hist_nrqbds_cpt_max =
339 max(1, ((int)val / svc->srv_ncpts));
341 spin_unlock(&svc->srv_lock);
346 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
349 ptlrpc_lprocfs_req_buffers_max_seq_show(struct seq_file *m, void *n)
351 struct ptlrpc_service *svc = m->private;
353 seq_printf(m, "%d\n", svc->srv_nrqbds_max);
358 ptlrpc_lprocfs_req_buffers_max_seq_write(struct file *file,
359 const char __user *buffer,
360 size_t count, loff_t *off)
362 struct seq_file *m = file->private_data;
363 struct ptlrpc_service *svc = m->private;
367 rc = kstrtoint_from_user(buffer, count, 0, &val);
371 if (val < svc->srv_nbuf_per_group && val != 0)
374 spin_lock(&svc->srv_lock);
376 svc->srv_nrqbds_max = (uint)val;
378 spin_unlock(&svc->srv_lock);
383 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_req_buffers_max);
385 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
388 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
391 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
394 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
395 const char *buffer, size_t count)
397 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
402 rc = kstrtoul(buffer, 10, &val);
406 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
409 spin_lock(&svc->srv_lock);
410 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
411 spin_unlock(&svc->srv_lock);
415 svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
417 spin_unlock(&svc->srv_lock);
421 LUSTRE_RW_ATTR(threads_min);
423 static ssize_t threads_started_show(struct kobject *kobj,
424 struct attribute *attr,
427 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
429 struct ptlrpc_service_part *svcpt;
433 ptlrpc_service_for_each_part(svcpt, i, svc)
434 total += svcpt->scp_nthrs_running;
436 return sprintf(buf, "%d\n", total);
438 LUSTRE_RO_ATTR(threads_started);
440 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
443 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
446 return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
449 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
450 const char *buffer, size_t count)
452 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
457 rc = kstrtoul(buffer, 10, &val);
461 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
464 spin_lock(&svc->srv_lock);
465 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
466 spin_unlock(&svc->srv_lock);
470 svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
472 spin_unlock(&svc->srv_lock);
476 LUSTRE_RW_ATTR(threads_max);
479 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
481 * \param[in] state The policy state
483 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
488 case NRS_POL_STATE_INVALID:
490 case NRS_POL_STATE_STOPPED:
492 case NRS_POL_STATE_STOPPING:
494 case NRS_POL_STATE_STARTING:
496 case NRS_POL_STATE_STARTED:
502 * Obtains status information for \a policy.
504 * Information is copied in \a info.
506 * \param[in] policy The policy
507 * \param[out] info Holds returned status information
509 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
510 struct ptlrpc_nrs_pol_info *info)
512 LASSERT(policy != NULL);
513 LASSERT(info != NULL);
514 assert_spin_locked(&policy->pol_nrs->nrs_lock);
516 BUILD_BUG_ON(sizeof(info->pi_arg) != sizeof(policy->pol_arg));
517 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
518 memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
520 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
521 info->pi_state = policy->pol_state;
523 * XXX: These are accessed without holding
524 * ptlrpc_service_part::scp_req_lock.
526 info->pi_req_queued = policy->pol_req_queued;
527 info->pi_req_started = policy->pol_req_started;
531 * Reads and prints policy status information for all policies of a PTLRPC
534 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
536 struct ptlrpc_service *svc = m->private;
537 struct ptlrpc_service_part *svcpt;
538 struct ptlrpc_nrs *nrs;
539 struct ptlrpc_nrs_policy *policy;
540 struct ptlrpc_nrs_pol_info *infos;
541 struct ptlrpc_nrs_pol_info tmp;
543 unsigned pol_idx = 0;
550 * Serialize NRS core lprocfs operations with policy registration/
553 mutex_lock(&nrs_core.nrs_mutex);
556 * Use the first service partition's regular NRS head in order to obtain
557 * the number of policies registered with NRS heads of this service. All
558 * service partitions will have the same number of policies.
560 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
562 spin_lock(&nrs->nrs_lock);
563 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
564 spin_unlock(&nrs->nrs_lock);
566 OBD_ALLOC(infos, num_pols * sizeof(*infos));
568 GOTO(out, rc = -ENOMEM);
571 ptlrpc_service_for_each_part(svcpt, i, svc) {
572 nrs = nrs_svcpt2nrs(svcpt, hp);
573 spin_lock(&nrs->nrs_lock);
577 list_for_each_entry(policy, &nrs->nrs_policy_list,
579 LASSERT(pol_idx < num_pols);
581 nrs_policy_get_info_locked(policy, &tmp);
583 * Copy values when handling the first service
587 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
589 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
591 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
592 sizeof(tmp.pi_state));
593 infos[pol_idx].pi_fallback = tmp.pi_fallback;
595 * For the rest of the service partitions
596 * sanity-check the values we get.
599 LASSERT(strncmp(infos[pol_idx].pi_name,
601 NRS_POL_NAME_MAX) == 0);
602 LASSERT(strncmp(infos[pol_idx].pi_arg,
604 sizeof(tmp.pi_arg)) == 0);
606 * Not asserting ptlrpc_nrs_pol_info::pi_state,
607 * because it may be different between
608 * instances of the same policy in different
609 * service partitions.
611 LASSERT(infos[pol_idx].pi_fallback ==
615 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
616 infos[pol_idx].pi_req_started += tmp.pi_req_started;
620 spin_unlock(&nrs->nrs_lock);
624 * Policy status information output is in YAML format.
640 * high_priority_requests:
653 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
654 "high_priority_requests:");
656 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
657 if (strlen(infos[pol_idx].pi_arg) > 0)
658 seq_printf(m, " - name: %s %s\n",
659 infos[pol_idx].pi_name,
660 infos[pol_idx].pi_arg);
662 seq_printf(m, " - name: %s\n",
663 infos[pol_idx].pi_name);
666 seq_printf(m, " state: %s\n"
669 " active: %-20d\n\n",
670 nrs_state2str(infos[pol_idx].pi_state),
671 infos[pol_idx].pi_fallback ? "yes" : "no",
672 (int)infos[pol_idx].pi_req_queued,
673 (int)infos[pol_idx].pi_req_started);
676 if (!hp && nrs_svc_has_hp(svc)) {
677 memset(infos, 0, num_pols * sizeof(*infos));
680 * Redo the processing for the service's HP NRS heads' policies.
688 OBD_FREE(infos, num_pols * sizeof(*infos));
690 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_seq_write(struct file *file, const char __user *buffer,
713 size_t count, loff_t *off)
715 struct seq_file *m = file->private_data;
716 struct ptlrpc_service *svc = m->private;
717 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
719 char *cmd_copy = NULL;
725 if (count >= LPROCFS_NRS_WR_MAX_CMD)
726 GOTO(out, rc = -EINVAL);
728 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
730 GOTO(out, rc = -ENOMEM);
732 * strsep() modifies its argument, so keep a copy
736 if (copy_from_user(cmd, buffer, count))
737 GOTO(out, rc = -EFAULT);
741 policy_name = strsep(&cmd, " ");
743 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
744 GOTO(out, rc = -EINVAL);
747 * No [reg|hp] token has been specified
752 queue_name = strsep(&cmd, " ");
754 * The second token is either an optional [reg|hp] string,
757 if (strcmp(queue_name, "reg") == 0)
758 queue = PTLRPC_NRS_QUEUE_REG;
759 else if (strcmp(queue_name, "hp") == 0)
760 queue = PTLRPC_NRS_QUEUE_HP;
769 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
770 GOTO(out, rc = -ENODEV);
771 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
772 queue = PTLRPC_NRS_QUEUE_REG;
775 * Serialize NRS core lprocfs operations with policy registration/
778 mutex_lock(&nrs_core.nrs_mutex);
780 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
781 PTLRPC_NRS_CTL_START,
784 mutex_unlock(&nrs_core.nrs_mutex);
787 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
789 RETURN(rc < 0 ? rc : count);
792 LDEBUGFS_SEQ_FOPS(ptlrpc_lprocfs_nrs);
796 struct ptlrpc_srh_iterator {
799 struct ptlrpc_request *srhi_req;
803 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
804 struct ptlrpc_srh_iterator *srhi,
808 struct ptlrpc_request *req;
810 if (srhi->srhi_req != NULL &&
811 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
812 srhi->srhi_seq <= seq) {
813 /* If srhi_req was set previously, hasn't been culled and
814 * we're searching for a seq on or after it (i.e. more
815 * recent), search from it onwards.
816 * Since the service history is LRU (i.e. culled reqs will
817 * be near the head), we shouldn't have to do long
819 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
820 "%s:%d: seek seq %llu, request seq %llu\n",
821 svcpt->scp_service->srv_name, svcpt->scp_cpt,
822 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
823 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
824 "%s:%d: seek offset %llu, request seq %llu, "
825 "last culled %llu\n",
826 svcpt->scp_service->srv_name, svcpt->scp_cpt,
827 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
828 e = &srhi->srhi_req->rq_history_list;
830 /* search from start */
831 e = svcpt->scp_hist_reqs.next;
834 while (e != &svcpt->scp_hist_reqs) {
835 req = list_entry(e, struct ptlrpc_request, rq_history_list);
837 if (req->rq_history_seq >= seq) {
838 srhi->srhi_seq = req->rq_history_seq;
839 srhi->srhi_req = req;
849 * ptlrpc history sequence is used as "position" of seq_file, in some case,
850 * seq_read() will increase "position" to indicate reading the next
851 * element, however, low bits of history sequence are reserved for CPT id
852 * (check the details from comments before ptlrpc_req_add_history), which
853 * means seq_read() might change CPT id of history sequence and never
854 * finish reading of requests on a CPT. To make it work, we have to shift
855 * CPT id to high bits and timestamp to low bits, so seq_read() will only
856 * increase timestamp which can correctly indicate the next position.
859 /* convert seq_file pos to cpt */
860 #define PTLRPC_REQ_POS2CPT(svc, pos) \
861 ((svc)->srv_cpt_bits == 0 ? 0 : \
862 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
864 /* make up seq_file pos from cpt */
865 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
866 ((svc)->srv_cpt_bits == 0 ? 0 : \
867 (cpt) << (64 - (svc)->srv_cpt_bits))
869 /* convert sequence to position */
870 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
871 ((svc)->srv_cpt_bits == 0 ? (seq) : \
872 ((seq) >> (svc)->srv_cpt_bits) | \
873 ((seq) << (64 - (svc)->srv_cpt_bits)))
875 /* convert position to sequence */
876 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
877 ((svc)->srv_cpt_bits == 0 ? (pos) : \
878 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
879 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
882 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
884 struct ptlrpc_service *svc = s->private;
885 struct ptlrpc_service_part *svcpt;
886 struct ptlrpc_srh_iterator *srhi;
891 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
892 CWARN("Failed to read request history because size of loff_t "
893 "%d can't match size of u64\n", (int)sizeof(loff_t));
897 OBD_ALLOC(srhi, sizeof(*srhi));
902 srhi->srhi_req = NULL;
904 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
906 ptlrpc_service_for_each_part(svcpt, i, svc) {
907 if (i < cpt) /* skip */
909 if (i > cpt) /* make up the lowest position for this CPT */
910 *pos = PTLRPC_REQ_CPT2POS(svc, i);
912 mutex_lock(&svcpt->scp_mutex);
913 spin_lock(&svcpt->scp_lock);
914 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
915 PTLRPC_REQ_POS2SEQ(svc, *pos));
916 spin_unlock(&svcpt->scp_lock);
917 mutex_unlock(&svcpt->scp_mutex);
919 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
925 OBD_FREE(srhi, sizeof(*srhi));
930 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
932 struct ptlrpc_srh_iterator *srhi = iter;
935 OBD_FREE(srhi, sizeof(*srhi));
939 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
940 void *iter, loff_t *pos)
942 struct ptlrpc_service *svc = s->private;
943 struct ptlrpc_srh_iterator *srhi = iter;
944 struct ptlrpc_service_part *svcpt;
949 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
950 svcpt = svc->srv_parts[i];
952 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
953 srhi->srhi_req = NULL;
954 seq = srhi->srhi_seq = 0;
955 } else { /* the next sequence */
956 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
959 mutex_lock(&svcpt->scp_mutex);
960 spin_lock(&svcpt->scp_lock);
961 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
962 spin_unlock(&svcpt->scp_lock);
963 mutex_unlock(&svcpt->scp_mutex);
965 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
971 OBD_FREE(srhi, sizeof(*srhi));
975 /* common ost/mdt so_req_printer */
976 void target_print_req(void *seq_file, struct ptlrpc_request *req)
978 /* Called holding srv_lock with irqs disabled.
979 * Print specific req contents and a newline.
980 * CAVEAT EMPTOR: check request message length before printing!!!
981 * You might have received any old crap so you must be just as
982 * careful here as the service's request parser!!! */
983 struct seq_file *sf = seq_file;
985 switch (req->rq_phase) {
987 /* still awaiting a service thread's attention, or rejected
988 * because the generic request message didn't unpack */
989 seq_printf(sf, "<not swabbed>\n");
991 case RQ_PHASE_INTERPRET:
992 /* being handled, so basic msg swabbed, and opc is valid
993 * but racing with mds_handle() */
994 case RQ_PHASE_COMPLETE:
995 /* been handled by mds_handle() reply state possibly still
997 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
1000 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
1003 EXPORT_SYMBOL(target_print_req);
1005 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
1007 struct ptlrpc_service *svc = s->private;
1008 struct ptlrpc_srh_iterator *srhi = iter;
1009 struct ptlrpc_service_part *svcpt;
1010 struct ptlrpc_request *req;
1013 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
1015 svcpt = svc->srv_parts[srhi->srhi_idx];
1017 mutex_lock(&svcpt->scp_mutex);
1018 spin_lock(&svcpt->scp_lock);
1020 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
1023 struct timespec64 arrival, sent, arrivaldiff;
1024 char nidstr[LNET_NIDSTR_SIZE];
1026 req = srhi->srhi_req;
1028 arrival.tv_sec = req->rq_arrival_time.tv_sec;
1029 arrival.tv_nsec = req->rq_arrival_time.tv_nsec;
1030 sent.tv_sec = req->rq_sent;
1032 arrivaldiff = timespec64_sub(sent, arrival);
1034 /* Print common req fields.
1035 * CAVEAT EMPTOR: we're racing with the service handler
1036 * here. The request could contain any old crap, so you
1037 * must be just as careful as the service's request
1038 * parser. Currently I only print stuff here I know is OK
1039 * to look at coz it was set up in request_in_callback()!!!
1042 "%lld:%s:%s:x%llu:%d:%s:%lld.%06lld:%lld.%06llds(%+lld.0s) ",
1043 req->rq_history_seq,
1044 req->rq_export && req->rq_export->exp_obd ?
1045 req->rq_export->exp_obd->obd_name :
1046 libcfs_nid2str_r(req->rq_self, nidstr,
1048 libcfs_id2str(req->rq_peer), req->rq_xid,
1049 req->rq_reqlen, ptlrpc_rqphase2str(req),
1050 (s64)req->rq_arrival_time.tv_sec,
1051 (s64)(req->rq_arrival_time.tv_nsec / NSEC_PER_USEC),
1052 (s64)arrivaldiff.tv_sec,
1053 (s64)(arrivaldiff.tv_nsec / NSEC_PER_USEC),
1054 (s64)(req->rq_sent - req->rq_deadline));
1055 if (svc->srv_ops.so_req_printer == NULL)
1056 seq_printf(s, "\n");
1058 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
1061 spin_unlock(&svcpt->scp_lock);
1062 mutex_unlock(&svcpt->scp_mutex);
1068 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1070 static struct seq_operations sops = {
1071 .start = ptlrpc_lprocfs_svc_req_history_start,
1072 .stop = ptlrpc_lprocfs_svc_req_history_stop,
1073 .next = ptlrpc_lprocfs_svc_req_history_next,
1074 .show = ptlrpc_lprocfs_svc_req_history_show,
1076 struct seq_file *seqf;
1079 rc = seq_open(file, &sops);
1083 seqf = file->private_data;
1084 seqf->private = inode->i_private;
1088 /* See also lprocfs_rd_timeouts */
1089 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1091 struct ptlrpc_service *svc = m->private;
1092 struct ptlrpc_service_part *svcpt;
1099 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1104 ptlrpc_service_for_each_part(svcpt, i, svc) {
1105 cur = at_get(&svcpt->scp_at_estimate);
1106 worst = svcpt->scp_at_estimate.at_worst_ever;
1107 worstt = svcpt->scp_at_estimate.at_worst_time;
1109 seq_printf(m, "%10s : cur %3u worst %3u (at %lld, %llds ago) ",
1110 "service", cur, worst, (s64)worstt,
1111 (s64)(ktime_get_real_seconds() - worstt));
1113 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1119 LDEBUGFS_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1121 static ssize_t high_priority_ratio_show(struct kobject *kobj,
1122 struct attribute *attr,
1125 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1128 return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1131 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1132 struct attribute *attr,
1136 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1141 rc = kstrtoul(buffer, 10, &val);
1145 spin_lock(&svc->srv_lock);
1146 svc->srv_hpreq_ratio = val;
1147 spin_unlock(&svc->srv_lock);
1151 LUSTRE_RW_ATTR(high_priority_ratio);
1153 static struct attribute *ptlrpc_svc_attrs[] = {
1154 &lustre_attr_threads_min.attr,
1155 &lustre_attr_threads_started.attr,
1156 &lustre_attr_threads_max.attr,
1157 &lustre_attr_high_priority_ratio.attr,
1161 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1163 struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1166 complete(&svc->srv_kobj_unregister);
1169 static struct kobj_type ptlrpc_svc_ktype = {
1170 .default_attrs = ptlrpc_svc_attrs,
1171 .sysfs_ops = &lustre_sysfs_ops,
1172 .release = ptlrpc_sysfs_svc_release,
1175 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1177 /* Let's see if we had a chance at initialization first */
1178 if (svc->srv_kobj.kset) {
1179 kobject_put(&svc->srv_kobj);
1180 wait_for_completion(&svc->srv_kobj_unregister);
1184 int ptlrpc_sysfs_register_service(struct kset *parent,
1185 struct ptlrpc_service *svc)
1187 svc->srv_kobj.kset = parent;
1188 init_completion(&svc->srv_kobj_unregister);
1189 return kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype,
1190 &parent->kobj, "%s", svc->srv_name);
1193 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1194 struct ptlrpc_service *svc)
1196 struct lprocfs_vars lproc_vars[] = {
1197 { .name = "req_buffer_history_len",
1198 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1200 { .name = "req_buffer_history_max",
1201 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1203 { .name = "timeouts",
1204 .fops = &ptlrpc_lprocfs_timeouts_fops,
1206 { .name = "nrs_policies",
1207 .fops = &ptlrpc_lprocfs_nrs_fops,
1209 { .name = "req_buffers_max",
1210 .fops = &ptlrpc_lprocfs_req_buffers_max_fops,
1214 static struct file_operations req_history_fops = {
1215 .owner = THIS_MODULE,
1216 .open = ptlrpc_lprocfs_svc_req_history_open,
1218 .llseek = seq_lseek,
1219 .release = lprocfs_seq_release,
1224 ptlrpc_ldebugfs_register(entry, svc->srv_name, "stats",
1225 &svc->srv_debugfs_entry, &svc->srv_stats);
1226 if (!svc->srv_debugfs_entry)
1229 ldebugfs_add_vars(svc->srv_debugfs_entry, lproc_vars, NULL);
1231 rc = ldebugfs_seq_create(svc->srv_debugfs_entry, "req_history",
1232 0400, &req_history_fops, svc);
1234 CWARN("Error adding the req_history file\n");
1237 void ptlrpc_lprocfs_register_obd(struct obd_device *obd)
1239 ptlrpc_ldebugfs_register(obd->obd_debugfs_entry, NULL, "stats",
1240 &obd->obd_svc_debugfs_entry,
1241 &obd->obd_svc_stats);
1243 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1245 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1247 struct lprocfs_stats *svc_stats;
1248 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1249 int opc = opcode_offset(op);
1251 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1252 if (svc_stats == NULL || opc <= 0)
1254 LASSERT(opc < LUSTRE_MAX_OPCODES);
1255 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1256 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1259 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1261 struct lprocfs_stats *svc_stats;
1264 if (!req->rq_import)
1266 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1269 idx = lustre_msg_get_opc(req->rq_reqmsg);
1272 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1275 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1278 LASSERTF(0, "unsupported opcode %u\n", idx);
1282 lprocfs_counter_add(svc_stats, idx, bytes);
1285 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1287 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1289 debugfs_remove_recursive(svc->srv_debugfs_entry);
1292 lprocfs_free_stats(&svc->srv_stats);
1295 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1297 /* cleanup first to allow concurrent access to device's
1298 * stats via debugfs to complete safely
1300 lprocfs_obd_cleanup(obd);
1302 debugfs_remove_recursive(obd->obd_svc_debugfs_entry);
1304 if (obd->obd_svc_stats)
1305 lprocfs_free_stats(&obd->obd_svc_stats);
1307 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1309 ssize_t ping_show(struct kobject *kobj, struct attribute *attr,
1312 struct obd_device *obd = container_of(kobj, struct obd_device,
1314 struct obd_import *imp;
1315 struct ptlrpc_request *req;
1319 with_imp_locked(obd, imp, rc)
1320 req = ptlrpc_prep_ping(imp);
1327 req->rq_send_state = LUSTRE_IMP_FULL;
1329 rc = ptlrpc_queue_wait(req);
1330 ptlrpc_req_finished(req);
1334 EXPORT_SYMBOL(ping_show);
1336 /* kept for older verison of tools. */
1337 ssize_t ping_store(struct kobject *kobj, struct attribute *attr,
1338 const char *buffer, size_t count)
1340 return ping_show(kobj, attr, (char *)buffer);
1342 EXPORT_SYMBOL(ping_store);
1344 /* Write the connection UUID to this file to attempt to connect to that node.
1345 * The connection UUID is a node's primary NID. For example,
1346 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1349 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1350 size_t count, loff_t *off)
1352 struct seq_file *m = file->private_data;
1353 struct obd_device *obd = m->private;
1354 struct obd_import *imp;
1359 const char prefix[] = "connection=";
1360 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, rc = -EFAULT);
1375 /* only support connection=uuid::instance now */
1376 if (strncmp(prefix, kbuf, prefix_len) != 0)
1377 GOTO(out, rc = -EINVAL);
1379 with_imp_locked(obd, imp, rc) {
1380 uuid = kbuf + prefix_len;
1381 ptr = strstr(uuid, "::");
1388 ptr += 2; /* Skip :: */
1389 rc = kstrtouint(ptr, 10, &inst);
1391 CERROR("config: wrong instance # %s\n", ptr);
1392 } else if (inst != imp->imp_connect_data.ocd_instance) {
1394 "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1395 imp->imp_obd->obd_name,
1396 imp->imp_connect_data.ocd_instance,
1401 "IR: %s has already been connecting to "
1403 imp->imp_obd->obd_name, inst);
1408 ptlrpc_recover_import(imp, uuid, 1);
1412 OBD_FREE(kbuf, count + 1);
1415 EXPORT_SYMBOL(lprocfs_import_seq_write);
1417 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1419 struct obd_device *obd = m->private;
1420 struct obd_import *imp;
1423 with_imp_locked(obd, imp, rc)
1424 seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1428 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1431 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1432 size_t count, loff_t *off)
1434 struct seq_file *m = file->private_data;
1435 struct obd_device *obd = m->private;
1436 struct obd_import *imp;
1440 rc = kstrtobool_from_user(buffer, count, &val);
1444 with_imp_locked(obd, imp, rc) {
1445 spin_lock(&imp->imp_lock);
1446 imp->imp_no_pinger_recover = !val;
1447 spin_unlock(&imp->imp_lock);
1452 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);