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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 #define DEBUG_SUBSYSTEM S_CLASS
39 # include <liblustre.h>
42 #include <obd_support.h>
44 #include <lprocfs_status.h>
45 #include <lustre/lustre_idl.h>
46 #include <lustre_net.h>
47 #include <obd_class.h>
48 #include "ptlrpc_internal.h"
51 struct ll_rpc_opcode {
54 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
55 { OST_REPLY, "ost_reply" },
56 { OST_GETATTR, "ost_getattr" },
57 { OST_SETATTR, "ost_setattr" },
58 { OST_READ, "ost_read" },
59 { OST_WRITE, "ost_write" },
60 { OST_CREATE , "ost_create" },
61 { OST_DESTROY, "ost_destroy" },
62 { OST_GET_INFO, "ost_get_info" },
63 { OST_CONNECT, "ost_connect" },
64 { OST_DISCONNECT, "ost_disconnect" },
65 { OST_PUNCH, "ost_punch" },
66 { OST_OPEN, "ost_open" },
67 { OST_CLOSE, "ost_close" },
68 { OST_STATFS, "ost_statfs" },
69 { 14, NULL }, /* formerly OST_SAN_READ */
70 { 15, NULL }, /* formerly OST_SAN_WRITE */
71 { OST_SYNC, "ost_sync" },
72 { OST_SET_INFO, "ost_set_info" },
73 { OST_QUOTACHECK, "ost_quotacheck" },
74 { OST_QUOTACTL, "ost_quotactl" },
75 { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
76 { MDS_GETATTR, "mds_getattr" },
77 { MDS_GETATTR_NAME, "mds_getattr_lock" },
78 { MDS_CLOSE, "mds_close" },
79 { MDS_REINT, "mds_reint" },
80 { MDS_READPAGE, "mds_readpage" },
81 { MDS_CONNECT, "mds_connect" },
82 { MDS_DISCONNECT, "mds_disconnect" },
83 { MDS_GETSTATUS, "mds_getstatus" },
84 { MDS_STATFS, "mds_statfs" },
85 { MDS_PIN, "mds_pin" },
86 { MDS_UNPIN, "mds_unpin" },
87 { MDS_SYNC, "mds_sync" },
88 { MDS_DONE_WRITING, "mds_done_writing" },
89 { MDS_SET_INFO, "mds_set_info" },
90 { MDS_QUOTACHECK, "mds_quotacheck" },
91 { MDS_QUOTACTL, "mds_quotactl" },
92 { MDS_GETXATTR, "mds_getxattr" },
93 { MDS_SETXATTR, "mds_setxattr" },
94 { MDS_WRITEPAGE, "mds_writepage" },
95 { MDS_IS_SUBDIR, "mds_is_subdir" },
96 { MDS_GET_INFO, "mds_get_info" },
97 { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
98 { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
99 { MDS_HSM_ACTION, "mds_hsm_action" },
100 { MDS_HSM_PROGRESS, "mds_hsm_progress" },
101 { MDS_HSM_REQUEST, "mds_hsm_request" },
102 { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
103 { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
104 { MDS_SWAP_LAYOUTS, "mds_swap_layouts" },
105 { LDLM_ENQUEUE, "ldlm_enqueue" },
106 { LDLM_CONVERT, "ldlm_convert" },
107 { LDLM_CANCEL, "ldlm_cancel" },
108 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
109 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
110 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
111 { LDLM_SET_INFO, "ldlm_set_info" },
112 { MGS_CONNECT, "mgs_connect" },
113 { MGS_DISCONNECT, "mgs_disconnect" },
114 { MGS_EXCEPTION, "mgs_exception" },
115 { MGS_TARGET_REG, "mgs_target_reg" },
116 { MGS_TARGET_DEL, "mgs_target_del" },
117 { MGS_SET_INFO, "mgs_set_info" },
118 { MGS_CONFIG_READ, "mgs_config_read" },
119 { OBD_PING, "obd_ping" },
120 { OBD_LOG_CANCEL, "llog_cancel" },
121 { OBD_QC_CALLBACK, "obd_quota_callback" },
122 { OBD_IDX_READ, "dt_index_read" },
123 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
124 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
125 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
126 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
127 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
128 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
129 { LLOG_CATINFO, "llog_catinfo" },
130 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
131 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
132 { QUOTA_DQACQ, "quota_acquire" },
133 { QUOTA_DQREL, "quota_release" },
134 { SEQ_QUERY, "seq_query" },
135 { SEC_CTX_INIT, "sec_ctx_init" },
136 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
137 { SEC_CTX_FINI, "sec_ctx_fini" },
138 { FLD_QUERY, "fld_query" },
139 { FLD_READ, "fld_read" },
140 { UPDATE_OBJ, "update_obj" },
146 } ll_eopcode_table[EXTRA_LAST_OPC] = {
147 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
148 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
149 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
150 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
151 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
152 { MDS_REINT_SETATTR, "mds_reint_setattr" },
153 { MDS_REINT_CREATE, "mds_reint_create" },
154 { MDS_REINT_LINK, "mds_reint_link" },
155 { MDS_REINT_UNLINK, "mds_reint_unlink" },
156 { MDS_REINT_RENAME, "mds_reint_rename" },
157 { MDS_REINT_OPEN, "mds_reint_open" },
158 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
159 { BRW_READ_BYTES, "read_bytes" },
160 { BRW_WRITE_BYTES, "write_bytes" },
163 const char *ll_opcode2str(__u32 opcode)
165 /* When one of the assertions below fail, chances are that:
166 * 1) A new opcode was added in include/lustre/lustre_idl.h,
167 * but is missing from the table above.
168 * or 2) The opcode space was renumbered or rearranged,
169 * and the opcode_offset() function in
170 * ptlrpc_internal.h needs to be modified.
172 __u32 offset = opcode_offset(opcode);
173 LASSERTF(offset < LUSTRE_MAX_OPCODES,
174 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
175 offset, LUSTRE_MAX_OPCODES);
176 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
177 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
178 offset, ll_rpc_opcode_table[offset].opcode, opcode);
179 return ll_rpc_opcode_table[offset].opname;
182 const char* ll_eopcode2str(__u32 opcode)
184 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
185 return ll_eopcode_table[opcode].opname;
188 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
189 char *name, struct proc_dir_entry **procroot_ret,
190 struct lprocfs_stats **stats_ret)
192 struct proc_dir_entry *svc_procroot;
193 struct lprocfs_stats *svc_stats;
195 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
198 LASSERT(*procroot_ret == NULL);
199 LASSERT(*stats_ret == NULL);
201 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
202 if (svc_stats == NULL)
206 svc_procroot = lprocfs_seq_register(dir, root, NULL, NULL);
207 if (IS_ERR(svc_procroot)) {
208 lprocfs_free_stats(&svc_stats);
215 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
216 svc_counter_config, "req_waittime", "usec");
217 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
218 svc_counter_config, "req_qdepth", "reqs");
219 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
220 svc_counter_config, "req_active", "reqs");
221 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
222 svc_counter_config, "req_timeout", "sec");
223 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
224 svc_counter_config, "reqbuf_avail", "bufs");
225 for (i = 0; i < EXTRA_LAST_OPC; i++) {
229 case BRW_WRITE_BYTES:
237 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
239 ll_eopcode2str(i), units);
241 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
242 __u32 opcode = ll_rpc_opcode_table[i].opcode;
243 lprocfs_counter_init(svc_stats,
244 EXTRA_MAX_OPCODES + i, svc_counter_config,
245 ll_opcode2str(opcode), "usec");
248 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
251 lprocfs_remove(&svc_procroot);
252 lprocfs_free_stats(&svc_stats);
255 *procroot_ret = svc_procroot;
256 *stats_ret = svc_stats;
261 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
263 struct ptlrpc_service *svc = m->private;
264 struct ptlrpc_service_part *svcpt;
268 ptlrpc_service_for_each_part(svcpt, i, svc)
269 total += svcpt->scp_hist_nrqbds;
271 return seq_printf(m, "%d\n", total);
273 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
276 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
278 struct ptlrpc_service *svc = m->private;
279 struct ptlrpc_service_part *svcpt;
283 ptlrpc_service_for_each_part(svcpt, i, svc)
284 total += svc->srv_hist_nrqbds_cpt_max;
286 return seq_printf(m, "%d\n", total);
290 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file, const char *buffer,
291 size_t count, loff_t *off)
293 struct seq_file *m = file->private_data;
294 struct ptlrpc_service *svc = m->private;
299 rc = lprocfs_write_helper(buffer, count, &val);
306 /* This sanity check is more of an insanity check; we can still
307 * hose a kernel by allowing the request history to grow too
309 bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
311 if (val > totalram_pages/(2 * bufpages))
314 spin_lock(&svc->srv_lock);
317 svc->srv_hist_nrqbds_cpt_max = 0;
319 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
321 spin_unlock(&svc->srv_lock);
325 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
328 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
330 struct ptlrpc_service *svc = m->private;
332 return seq_printf(m, "%d\n",
333 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
337 ptlrpc_lprocfs_threads_min_seq_write(struct file *file, const char *buffer,
338 size_t count, loff_t *off)
340 struct seq_file *m = file->private_data;
341 struct ptlrpc_service *svc = m->private;
343 int rc = lprocfs_write_helper(buffer, count, &val);
348 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
351 spin_lock(&svc->srv_lock);
352 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
353 spin_unlock(&svc->srv_lock);
357 svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
359 spin_unlock(&svc->srv_lock);
363 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
366 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
368 struct ptlrpc_service *svc = m->private;
369 struct ptlrpc_service_part *svcpt;
373 ptlrpc_service_for_each_part(svcpt, i, svc)
374 total += svcpt->scp_nthrs_running;
376 return seq_printf(m, "%d\n", total);
378 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
381 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
383 struct ptlrpc_service *svc = m->private;
385 return seq_printf(m, "%d\n",
386 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
390 ptlrpc_lprocfs_threads_max_seq_write(struct file *file, const char *buffer,
391 size_t count, loff_t *off)
393 struct seq_file *m = file->private_data;
394 struct ptlrpc_service *svc = m->private;
396 int rc = lprocfs_write_helper(buffer, count, &val);
401 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
404 spin_lock(&svc->srv_lock);
405 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
406 spin_unlock(&svc->srv_lock);
410 svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
412 spin_unlock(&svc->srv_lock);
416 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
422 extern struct nrs_core nrs_core;
425 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
427 * \param[in] state The policy state
429 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
434 case NRS_POL_STATE_INVALID:
436 case NRS_POL_STATE_STOPPED:
438 case NRS_POL_STATE_STOPPING:
440 case NRS_POL_STATE_STARTING:
442 case NRS_POL_STATE_STARTED:
448 * Obtains status information for \a policy.
450 * Information is copied in \a info.
452 * \param[in] policy The policy
453 * \param[out] info Holds returned status information
455 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
456 struct ptlrpc_nrs_pol_info *info)
458 LASSERT(policy != NULL);
459 LASSERT(info != NULL);
460 LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
462 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
464 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
465 info->pi_state = policy->pol_state;
467 * XXX: These are accessed without holding
468 * ptlrpc_service_part::scp_req_lock.
470 info->pi_req_queued = policy->pol_req_queued;
471 info->pi_req_started = policy->pol_req_started;
475 * Reads and prints policy status information for all policies of a PTLRPC
478 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
480 struct ptlrpc_service *svc = m->private;
481 struct ptlrpc_service_part *svcpt;
482 struct ptlrpc_nrs *nrs;
483 struct ptlrpc_nrs_policy *policy;
484 struct ptlrpc_nrs_pol_info *infos;
485 struct ptlrpc_nrs_pol_info tmp;
487 unsigned pol_idx = 0;
494 * Serialize NRS core lprocfs operations with policy registration/
497 mutex_lock(&nrs_core.nrs_mutex);
500 * Use the first service partition's regular NRS head in order to obtain
501 * the number of policies registered with NRS heads of this service. All
502 * service partitions will have the same number of policies.
504 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
506 spin_lock(&nrs->nrs_lock);
507 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
508 spin_unlock(&nrs->nrs_lock);
510 OBD_ALLOC(infos, num_pols * sizeof(*infos));
512 GOTO(out, rc = -ENOMEM);
515 ptlrpc_service_for_each_part(svcpt, i, svc) {
516 nrs = nrs_svcpt2nrs(svcpt, hp);
517 spin_lock(&nrs->nrs_lock);
521 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
523 LASSERT(pol_idx < num_pols);
525 nrs_policy_get_info_locked(policy, &tmp);
527 * Copy values when handling the first service
531 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
533 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
534 sizeof(tmp.pi_state));
535 infos[pol_idx].pi_fallback = tmp.pi_fallback;
537 * For the rest of the service partitions
538 * sanity-check the values we get.
541 LASSERT(strncmp(infos[pol_idx].pi_name,
543 NRS_POL_NAME_MAX) == 0);
545 * Not asserting ptlrpc_nrs_pol_info::pi_state,
546 * because it may be different between
547 * instances of the same policy in different
548 * service partitions.
550 LASSERT(infos[pol_idx].pi_fallback ==
554 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
555 infos[pol_idx].pi_req_started += tmp.pi_req_started;
559 spin_unlock(&nrs->nrs_lock);
563 * Policy status information output is in YAML format.
579 * high_priority_requests:
592 seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
593 "high_priority_requests:");
595 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
596 seq_printf(m, " - name: %s\n"
600 " active: %-20d\n\n",
601 infos[pol_idx].pi_name,
602 nrs_state2str(infos[pol_idx].pi_state),
603 infos[pol_idx].pi_fallback ? "yes" : "no",
604 (int)infos[pol_idx].pi_req_queued,
605 (int)infos[pol_idx].pi_req_started);
608 if (!hp && nrs_svc_has_hp(svc)) {
609 memset(infos, 0, num_pols * sizeof(*infos));
612 * Redo the processing for the service's HP NRS heads' policies.
620 OBD_FREE(infos, num_pols * sizeof(*infos));
622 mutex_unlock(&nrs_core.nrs_mutex);
628 #define LPROCFS_NRS_WR_MAX_ARG (1024)
630 * The longest valid command string is the maxium policy name size, plus the
631 * length of the " reg" substring, plus the lenght of argument
633 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
634 + LPROCFS_NRS_WR_MAX_ARG)
637 * Starts and stops a given policy on a PTLRPC service.
639 * Commands consist of the policy name, followed by an optional [reg|hp] token;
640 * if the optional token is omitted, the operation is performed on both the
641 * regular and high-priority (if the service has one) NRS head.
644 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char *buffer,
645 size_t count, loff_t *off)
647 struct seq_file *m = file->private_data;
648 struct ptlrpc_service *svc = m->private;
649 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
651 char *cmd_copy = NULL;
657 if (count >= LPROCFS_NRS_WR_MAX_CMD)
658 GOTO(out, rc = -EINVAL);
660 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
662 GOTO(out, rc = -ENOMEM);
664 * strsep() modifies its argument, so keep a copy
668 if (copy_from_user(cmd, buffer, count))
669 GOTO(out, rc = -EFAULT);
673 policy_name = strsep(&cmd, " ");
675 if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
676 GOTO(out, rc = -EINVAL);
679 * No [reg|hp] token has been specified
684 queue_name = strsep(&cmd, " ");
686 * The second token is either an optional [reg|hp] string,
689 if (strcmp(queue_name, "reg") == 0)
690 queue = PTLRPC_NRS_QUEUE_REG;
691 else if (strcmp(queue_name, "hp") == 0)
692 queue = PTLRPC_NRS_QUEUE_HP;
701 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
702 GOTO(out, rc = -ENODEV);
703 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
704 queue = PTLRPC_NRS_QUEUE_REG;
707 * Serialize NRS core lprocfs operations with policy registration/
710 mutex_lock(&nrs_core.nrs_mutex);
712 rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
713 PTLRPC_NRS_CTL_START,
716 mutex_unlock(&nrs_core.nrs_mutex);
719 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
721 RETURN(rc < 0 ? rc : count);
723 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
727 struct ptlrpc_srh_iterator {
730 struct ptlrpc_request *srhi_req;
734 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
735 struct ptlrpc_srh_iterator *srhi,
739 struct ptlrpc_request *req;
741 if (srhi->srhi_req != NULL &&
742 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
743 srhi->srhi_seq <= seq) {
744 /* If srhi_req was set previously, hasn't been culled and
745 * we're searching for a seq on or after it (i.e. more
746 * recent), search from it onwards.
747 * Since the service history is LRU (i.e. culled reqs will
748 * be near the head), we shouldn't have to do long
750 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
751 "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
752 svcpt->scp_service->srv_name, svcpt->scp_cpt,
753 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
754 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
755 "%s:%d: seek offset "LPU64", request seq "LPU64", "
756 "last culled "LPU64"\n",
757 svcpt->scp_service->srv_name, svcpt->scp_cpt,
758 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
759 e = &srhi->srhi_req->rq_history_list;
761 /* search from start */
762 e = svcpt->scp_hist_reqs.next;
765 while (e != &svcpt->scp_hist_reqs) {
766 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
768 if (req->rq_history_seq >= seq) {
769 srhi->srhi_seq = req->rq_history_seq;
770 srhi->srhi_req = req;
780 * ptlrpc history sequence is used as "position" of seq_file, in some case,
781 * seq_read() will increase "position" to indicate reading the next
782 * element, however, low bits of history sequence are reserved for CPT id
783 * (check the details from comments before ptlrpc_req_add_history), which
784 * means seq_read() might change CPT id of history sequence and never
785 * finish reading of requests on a CPT. To make it work, we have to shift
786 * CPT id to high bits and timestamp to low bits, so seq_read() will only
787 * increase timestamp which can correctly indicate the next position.
790 /* convert seq_file pos to cpt */
791 #define PTLRPC_REQ_POS2CPT(svc, pos) \
792 ((svc)->srv_cpt_bits == 0 ? 0 : \
793 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
795 /* make up seq_file pos from cpt */
796 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
797 ((svc)->srv_cpt_bits == 0 ? 0 : \
798 (cpt) << (64 - (svc)->srv_cpt_bits))
800 /* convert sequence to position */
801 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
802 ((svc)->srv_cpt_bits == 0 ? (seq) : \
803 ((seq) >> (svc)->srv_cpt_bits) | \
804 ((seq) << (64 - (svc)->srv_cpt_bits)))
806 /* convert position to sequence */
807 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
808 ((svc)->srv_cpt_bits == 0 ? (pos) : \
809 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
810 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
813 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
815 struct ptlrpc_service *svc = s->private;
816 struct ptlrpc_service_part *svcpt;
817 struct ptlrpc_srh_iterator *srhi;
822 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
823 CWARN("Failed to read request history because size of loff_t "
824 "%d can't match size of u64\n", (int)sizeof(loff_t));
828 OBD_ALLOC(srhi, sizeof(*srhi));
833 srhi->srhi_req = NULL;
835 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
837 ptlrpc_service_for_each_part(svcpt, i, svc) {
838 if (i < cpt) /* skip */
840 if (i > cpt) /* make up the lowest position for this CPT */
841 *pos = PTLRPC_REQ_CPT2POS(svc, i);
843 spin_lock(&svcpt->scp_lock);
844 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
845 PTLRPC_REQ_POS2SEQ(svc, *pos));
846 spin_unlock(&svcpt->scp_lock);
848 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
854 OBD_FREE(srhi, sizeof(*srhi));
859 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
861 struct ptlrpc_srh_iterator *srhi = iter;
864 OBD_FREE(srhi, sizeof(*srhi));
868 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
869 void *iter, loff_t *pos)
871 struct ptlrpc_service *svc = s->private;
872 struct ptlrpc_srh_iterator *srhi = iter;
873 struct ptlrpc_service_part *svcpt;
878 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
879 svcpt = svc->srv_parts[i];
881 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
882 srhi->srhi_req = NULL;
883 seq = srhi->srhi_seq = 0;
884 } else { /* the next sequence */
885 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
888 spin_lock(&svcpt->scp_lock);
889 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
890 spin_unlock(&svcpt->scp_lock);
892 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
898 OBD_FREE(srhi, sizeof(*srhi));
902 /* common ost/mdt so_req_printer */
903 void target_print_req(void *seq_file, struct ptlrpc_request *req)
905 /* Called holding srv_lock with irqs disabled.
906 * Print specific req contents and a newline.
907 * CAVEAT EMPTOR: check request message length before printing!!!
908 * You might have received any old crap so you must be just as
909 * careful here as the service's request parser!!! */
910 struct seq_file *sf = seq_file;
912 switch (req->rq_phase) {
914 /* still awaiting a service thread's attention, or rejected
915 * because the generic request message didn't unpack */
916 seq_printf(sf, "<not swabbed>\n");
918 case RQ_PHASE_INTERPRET:
919 /* being handled, so basic msg swabbed, and opc is valid
920 * but racing with mds_handle() */
921 case RQ_PHASE_COMPLETE:
922 /* been handled by mds_handle() reply state possibly still
924 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
927 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
930 EXPORT_SYMBOL(target_print_req);
932 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
934 struct ptlrpc_service *svc = s->private;
935 struct ptlrpc_srh_iterator *srhi = iter;
936 struct ptlrpc_service_part *svcpt;
937 struct ptlrpc_request *req;
940 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
942 svcpt = svc->srv_parts[srhi->srhi_idx];
944 spin_lock(&svcpt->scp_lock);
946 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
949 req = srhi->srhi_req;
951 /* Print common req fields.
952 * CAVEAT EMPTOR: we're racing with the service handler
953 * here. The request could contain any old crap, so you
954 * must be just as careful as the service's request
955 * parser. Currently I only print stuff here I know is OK
956 * to look at coz it was set up in request_in_callback()!!! */
957 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
958 req->rq_history_seq, libcfs_nid2str(req->rq_self),
959 libcfs_id2str(req->rq_peer), req->rq_xid,
960 req->rq_reqlen, ptlrpc_rqphase2str(req),
961 req->rq_arrival_time.tv_sec,
962 req->rq_sent - req->rq_arrival_time.tv_sec,
963 req->rq_sent - req->rq_deadline);
964 if (svc->srv_ops.so_req_printer == NULL)
967 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
970 spin_unlock(&svcpt->scp_lock);
975 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
977 static struct seq_operations sops = {
978 .start = ptlrpc_lprocfs_svc_req_history_start,
979 .stop = ptlrpc_lprocfs_svc_req_history_stop,
980 .next = ptlrpc_lprocfs_svc_req_history_next,
981 .show = ptlrpc_lprocfs_svc_req_history_show,
983 struct seq_file *seqf;
986 LPROCFS_ENTRY_CHECK(PDE(inode));
987 rc = seq_open(file, &sops);
991 seqf = file->private_data;
992 seqf->private = PDE_DATA(inode);
996 /* See also lprocfs_rd_timeouts */
997 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
999 struct ptlrpc_service *svc = m->private;
1000 struct ptlrpc_service_part *svcpt;
1008 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1013 ptlrpc_service_for_each_part(svcpt, i, svc) {
1014 cur = at_get(&svcpt->scp_at_estimate);
1015 worst = svcpt->scp_at_estimate.at_worst_ever;
1016 worstt = svcpt->scp_at_estimate.at_worst_time;
1017 s2dhms(&ts, cfs_time_current_sec() - worstt);
1019 seq_printf(m, "%10s : cur %3u worst %3u (at %ld, "
1020 DHMS_FMT" ago) ", "service",
1021 cur, worst, worstt, DHMS_VARS(&ts));
1023 lprocfs_seq_at_hist_helper(m, &svcpt->scp_at_estimate);
1028 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1030 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1032 struct ptlrpc_service *svc = m->private;
1033 return seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1037 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char *buffer,
1038 size_t count, loff_t *off)
1040 struct seq_file *m = file->private_data;
1041 struct ptlrpc_service *svc = m->private;
1045 rc = lprocfs_write_helper(buffer, count, &val);
1052 spin_lock(&svc->srv_lock);
1053 svc->srv_hpreq_ratio = val;
1054 spin_unlock(&svc->srv_lock);
1058 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1060 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1061 struct ptlrpc_service *svc)
1063 struct lprocfs_seq_vars lproc_vars[] = {
1064 { .name = "high_priority_ratio",
1065 .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1067 { .name = "req_buffer_history_len",
1068 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1070 { .name = "req_buffer_history_max",
1071 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1073 { .name = "threads_min",
1074 .fops = &ptlrpc_lprocfs_threads_min_fops,
1076 { .name = "threads_max",
1077 .fops = &ptlrpc_lprocfs_threads_max_fops,
1079 { .name = "threads_started",
1080 .fops = &ptlrpc_lprocfs_threads_started_fops,
1082 { .name = "timeouts",
1083 .fops = &ptlrpc_lprocfs_timeouts_fops,
1085 { .name = "nrs_policies",
1086 .fops = &ptlrpc_lprocfs_nrs_fops,
1090 static struct file_operations req_history_fops = {
1091 .owner = THIS_MODULE,
1092 .open = ptlrpc_lprocfs_svc_req_history_open,
1094 .llseek = seq_lseek,
1095 .release = lprocfs_seq_release,
1100 ptlrpc_lprocfs_register(entry, svc->srv_name,
1101 "stats", &svc->srv_procroot,
1103 if (svc->srv_procroot == NULL)
1106 lprocfs_seq_add_vars(svc->srv_procroot, lproc_vars, NULL);
1108 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1109 0400, &req_history_fops, svc);
1111 CWARN("Error adding the req_history file\n");
1114 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1116 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1117 &obddev->obd_svc_procroot,
1118 &obddev->obd_svc_stats);
1120 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1122 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1124 struct lprocfs_stats *svc_stats;
1125 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1126 int opc = opcode_offset(op);
1128 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1129 if (svc_stats == NULL || opc <= 0)
1131 LASSERT(opc < LUSTRE_MAX_OPCODES);
1132 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1133 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1136 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1138 struct lprocfs_stats *svc_stats;
1141 if (!req->rq_import)
1143 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1146 idx = lustre_msg_get_opc(req->rq_reqmsg);
1149 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1152 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1155 LASSERTF(0, "unsupported opcode %u\n", idx);
1159 lprocfs_counter_add(svc_stats, idx, bytes);
1162 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1164 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1166 if (svc->srv_procroot != NULL)
1167 lprocfs_remove(&svc->srv_procroot);
1170 lprocfs_free_stats(&svc->srv_stats);
1173 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1175 if (obd->obd_svc_procroot)
1176 lprocfs_remove(&obd->obd_svc_procroot);
1178 if (obd->obd_svc_stats)
1179 lprocfs_free_stats(&obd->obd_svc_stats);
1181 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1183 #ifndef HAVE_ONLY_PROCFS_SEQ
1184 int lprocfs_wr_ping(struct file *file, const char *buffer,
1185 unsigned long count, void *data)
1187 struct obd_device *obd = data;
1188 struct ptlrpc_request *req;
1192 LPROCFS_CLIMP_CHECK(obd);
1193 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1194 LPROCFS_CLIMP_EXIT(obd);
1198 req->rq_send_state = LUSTRE_IMP_FULL;
1200 rc = ptlrpc_queue_wait(req);
1202 ptlrpc_req_finished(req);
1207 EXPORT_SYMBOL(lprocfs_wr_ping);
1209 int lprocfs_wr_import(struct file *file, const char *buffer,
1210 unsigned long count, void *data)
1212 struct obd_device *obd = data;
1213 struct obd_import *imp = obd->u.cli.cl_import;
1218 const char prefix[] = "connection=";
1219 const int prefix_len = sizeof(prefix) - 1;
1221 if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1224 OBD_ALLOC(kbuf, count + 1);
1228 if (copy_from_user(kbuf, buffer, count))
1229 GOTO(out, count = -EFAULT);
1233 /* only support connection=uuid::instance now */
1234 if (strncmp(prefix, kbuf, prefix_len) != 0)
1235 GOTO(out, count = -EINVAL);
1237 uuid = kbuf + prefix_len;
1238 ptr = strstr(uuid, "::");
1245 ptr += strlen("::");
1246 inst = simple_strtol(ptr, &endptr, 10);
1248 CERROR("config: wrong instance # %s\n", ptr);
1249 } else if (inst != imp->imp_connect_data.ocd_instance) {
1250 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1251 "target(%u/%u), reconnecting...\n",
1252 imp->imp_obd->obd_name,
1253 imp->imp_connect_data.ocd_instance, inst);
1256 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1258 imp->imp_obd->obd_name, inst);
1263 ptlrpc_recover_import(imp, uuid, 1);
1266 OBD_FREE(kbuf, count + 1);
1269 EXPORT_SYMBOL(lprocfs_wr_import);
1271 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1272 int count, int *eof, void *data)
1274 struct obd_device *obd = data;
1275 struct obd_import *imp = obd->u.cli.cl_import;
1278 LPROCFS_CLIMP_CHECK(obd);
1279 rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1280 LPROCFS_CLIMP_EXIT(obd);
1284 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1286 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1287 unsigned long count, void *data)
1289 struct obd_device *obd = data;
1290 struct client_obd *cli = &obd->u.cli;
1291 struct obd_import *imp = cli->cl_import;
1294 rc = lprocfs_write_helper(buffer, count, &val);
1298 if (val != 0 && val != 1)
1301 LPROCFS_CLIMP_CHECK(obd);
1302 spin_lock(&imp->imp_lock);
1303 imp->imp_no_pinger_recover = !val;
1304 spin_unlock(&imp->imp_lock);
1305 LPROCFS_CLIMP_EXIT(obd);
1310 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1312 #endif /* HAVE_ONLY_PROCFS_SEQ */
1315 lprocfs_ping_seq_write(struct file *file, const char *buffer,
1316 size_t count, loff_t *off)
1318 struct seq_file *m = file->private_data;
1319 struct obd_device *obd = m->private;
1320 struct ptlrpc_request *req;
1324 LPROCFS_CLIMP_CHECK(obd);
1325 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1326 LPROCFS_CLIMP_EXIT(obd);
1330 req->rq_send_state = LUSTRE_IMP_FULL;
1332 rc = ptlrpc_queue_wait(req);
1334 ptlrpc_req_finished(req);
1339 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1341 /* Write the connection UUID to this file to attempt to connect to that node.
1342 * The connection UUID is a node's primary NID. For example,
1343 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1346 lprocfs_import_seq_write(struct file *file, const char *buffer, size_t count,
1349 struct seq_file *m = file->private_data;
1350 struct obd_device *obd = m->private;
1351 struct obd_import *imp = obd->u.cli.cl_import;
1356 const char prefix[] = "connection=";
1357 const int prefix_len = sizeof(prefix) - 1;
1359 if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1362 OBD_ALLOC(kbuf, count + 1);
1366 if (copy_from_user(kbuf, buffer, count))
1367 GOTO(out, count = -EFAULT);
1371 /* only support connection=uuid::instance now */
1372 if (strncmp(prefix, kbuf, prefix_len) != 0)
1373 GOTO(out, count = -EINVAL);
1375 uuid = kbuf + prefix_len;
1376 ptr = strstr(uuid, "::");
1383 ptr += strlen("::");
1384 inst = simple_strtol(ptr, &endptr, 10);
1386 CERROR("config: wrong instance # %s\n", ptr);
1387 } else if (inst != imp->imp_connect_data.ocd_instance) {
1388 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1389 "target(%u/%u), reconnecting...\n",
1390 imp->imp_obd->obd_name,
1391 imp->imp_connect_data.ocd_instance, inst);
1394 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1396 imp->imp_obd->obd_name, inst);
1401 ptlrpc_recover_import(imp, uuid, 1);
1404 OBD_FREE(kbuf, count + 1);
1407 EXPORT_SYMBOL(lprocfs_import_seq_write);
1409 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1411 struct obd_device *obd = m->private;
1412 struct obd_import *imp = obd->u.cli.cl_import;
1415 LPROCFS_CLIMP_CHECK(obd);
1416 rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1417 LPROCFS_CLIMP_EXIT(obd);
1420 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1423 lprocfs_pinger_recov_seq_write(struct file *file, const char *buffer,
1424 size_t count, loff_t *off)
1426 struct seq_file *m = file->private_data;
1427 struct obd_device *obd = m->private;
1428 struct client_obd *cli = &obd->u.cli;
1429 struct obd_import *imp = cli->cl_import;
1432 rc = lprocfs_write_helper(buffer, count, &val);
1436 if (val != 0 && val != 1)
1439 LPROCFS_CLIMP_CHECK(obd);
1440 spin_lock(&imp->imp_lock);
1441 imp->imp_no_pinger_recover = !val;
1442 spin_unlock(&imp->imp_lock);
1443 LPROCFS_CLIMP_EXIT(obd);
1446 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1448 #endif /* LPROCFS */