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, 2012, 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 "../include/obd_support.h"
40 #include "../include/obd.h"
41 #include "../include/lprocfs_status.h"
42 #include "../include/lustre/lustre_idl.h"
43 #include "../include/lustre_net.h"
44 #include "../include/obd_class.h"
45 #include "ptlrpc_internal.h"
48 struct ll_rpc_opcode {
51 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
52 { OST_REPLY, "ost_reply" },
53 { OST_GETATTR, "ost_getattr" },
54 { OST_SETATTR, "ost_setattr" },
55 { OST_READ, "ost_read" },
56 { OST_WRITE, "ost_write" },
57 { OST_CREATE , "ost_create" },
58 { OST_DESTROY, "ost_destroy" },
59 { OST_GET_INFO, "ost_get_info" },
60 { OST_CONNECT, "ost_connect" },
61 { OST_DISCONNECT, "ost_disconnect" },
62 { OST_PUNCH, "ost_punch" },
63 { OST_OPEN, "ost_open" },
64 { OST_CLOSE, "ost_close" },
65 { OST_STATFS, "ost_statfs" },
66 { 14, NULL }, /* formerly OST_SAN_READ */
67 { 15, NULL }, /* formerly OST_SAN_WRITE */
68 { OST_SYNC, "ost_sync" },
69 { OST_SET_INFO, "ost_set_info" },
70 { OST_QUOTACHECK, "ost_quotacheck" },
71 { OST_QUOTACTL, "ost_quotactl" },
72 { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
73 { MDS_GETATTR, "mds_getattr" },
74 { MDS_GETATTR_NAME, "mds_getattr_lock" },
75 { MDS_CLOSE, "mds_close" },
76 { MDS_REINT, "mds_reint" },
77 { MDS_READPAGE, "mds_readpage" },
78 { MDS_CONNECT, "mds_connect" },
79 { MDS_DISCONNECT, "mds_disconnect" },
80 { MDS_GETSTATUS, "mds_getstatus" },
81 { MDS_STATFS, "mds_statfs" },
82 { MDS_PIN, "mds_pin" },
83 { MDS_UNPIN, "mds_unpin" },
84 { MDS_SYNC, "mds_sync" },
85 { MDS_DONE_WRITING, "mds_done_writing" },
86 { MDS_SET_INFO, "mds_set_info" },
87 { MDS_QUOTACHECK, "mds_quotacheck" },
88 { MDS_QUOTACTL, "mds_quotactl" },
89 { MDS_GETXATTR, "mds_getxattr" },
90 { MDS_SETXATTR, "mds_setxattr" },
91 { MDS_WRITEPAGE, "mds_writepage" },
92 { MDS_IS_SUBDIR, "mds_is_subdir" },
93 { MDS_GET_INFO, "mds_get_info" },
94 { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
95 { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
96 { MDS_HSM_ACTION, "mds_hsm_action" },
97 { MDS_HSM_PROGRESS, "mds_hsm_progress" },
98 { MDS_HSM_REQUEST, "mds_hsm_request" },
99 { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
100 { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
101 { MDS_SWAP_LAYOUTS, "mds_swap_layouts" },
102 { LDLM_ENQUEUE, "ldlm_enqueue" },
103 { LDLM_CONVERT, "ldlm_convert" },
104 { LDLM_CANCEL, "ldlm_cancel" },
105 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
106 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
107 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
108 { LDLM_SET_INFO, "ldlm_set_info" },
109 { MGS_CONNECT, "mgs_connect" },
110 { MGS_DISCONNECT, "mgs_disconnect" },
111 { MGS_EXCEPTION, "mgs_exception" },
112 { MGS_TARGET_REG, "mgs_target_reg" },
113 { MGS_TARGET_DEL, "mgs_target_del" },
114 { MGS_SET_INFO, "mgs_set_info" },
115 { MGS_CONFIG_READ, "mgs_config_read" },
116 { OBD_PING, "obd_ping" },
117 { OBD_LOG_CANCEL, "llog_cancel" },
118 { OBD_QC_CALLBACK, "obd_quota_callback" },
119 { OBD_IDX_READ, "dt_index_read" },
120 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_open" },
121 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
122 { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
123 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
124 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
125 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
126 { LLOG_CATINFO, "llog_catinfo" },
127 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
128 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
129 { QUOTA_DQACQ, "quota_acquire" },
130 { QUOTA_DQREL, "quota_release" },
131 { SEQ_QUERY, "seq_query" },
132 { SEC_CTX_INIT, "sec_ctx_init" },
133 { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
134 { SEC_CTX_FINI, "sec_ctx_fini" },
135 { FLD_QUERY, "fld_query" },
136 { UPDATE_OBJ, "update_obj" },
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 char *ll_eopcode2str(__u32 opcode)
180 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
181 return ll_eopcode_table[opcode].opname;
183 #if defined (CONFIG_PROC_FS)
184 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
186 struct proc_dir_entry **procroot_ret,
187 struct lprocfs_stats **stats_ret)
189 struct proc_dir_entry *svc_procroot;
190 struct lprocfs_stats *svc_stats;
192 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
195 LASSERT(*procroot_ret == NULL);
196 LASSERT(*stats_ret == NULL);
198 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,
200 if (svc_stats == NULL)
204 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
205 if (IS_ERR(svc_procroot)) {
206 lprocfs_free_stats(&svc_stats);
213 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
214 svc_counter_config, "req_waittime", "usec");
215 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
216 svc_counter_config, "req_qdepth", "reqs");
217 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
218 svc_counter_config, "req_active", "reqs");
219 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
220 svc_counter_config, "req_timeout", "sec");
221 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
222 svc_counter_config, "reqbuf_avail", "bufs");
223 for (i = 0; i < EXTRA_LAST_OPC; i++) {
227 case BRW_WRITE_BYTES:
235 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
237 ll_eopcode2str(i), units);
239 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
240 __u32 opcode = ll_rpc_opcode_table[i].opcode;
241 lprocfs_counter_init(svc_stats,
242 EXTRA_MAX_OPCODES + i, svc_counter_config,
243 ll_opcode2str(opcode), "usec");
246 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
249 lprocfs_remove(&svc_procroot);
250 lprocfs_free_stats(&svc_stats);
253 *procroot_ret = svc_procroot;
254 *stats_ret = svc_stats;
259 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
261 struct ptlrpc_service *svc = m->private;
262 struct ptlrpc_service_part *svcpt;
266 ptlrpc_service_for_each_part(svcpt, i, svc)
267 total += svcpt->scp_hist_nrqbds;
269 return seq_printf(m, "%d\n", total);
271 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
274 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
276 struct ptlrpc_service *svc = m->private;
277 struct ptlrpc_service_part *svcpt;
281 ptlrpc_service_for_each_part(svcpt, i, svc)
282 total += svc->srv_hist_nrqbds_cpt_max;
284 return seq_printf(m, "%d\n", total);
288 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
289 const char __user *buffer,
290 size_t count, loff_t *off)
292 struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
297 rc = lprocfs_write_helper(buffer, count, &val);
304 /* This sanity check is more of an insanity check; we can still
305 * hose a kernel by allowing the request history to grow too
307 bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
308 if (val > totalram_pages / (2 * bufpages))
311 spin_lock(&svc->srv_lock);
314 svc->srv_hist_nrqbds_cpt_max = 0;
316 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
318 spin_unlock(&svc->srv_lock);
322 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
325 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
327 struct ptlrpc_service *svc = m->private;
329 return seq_printf(m, "%d\n",
330 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
334 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
335 const char __user *buffer,
336 size_t count, loff_t *off)
338 struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
340 int rc = lprocfs_write_helper(buffer, count, &val);
345 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
348 spin_lock(&svc->srv_lock);
349 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
350 spin_unlock(&svc->srv_lock);
354 svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
356 spin_unlock(&svc->srv_lock);
360 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
363 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
365 struct ptlrpc_service *svc = m->private;
366 struct ptlrpc_service_part *svcpt;
370 ptlrpc_service_for_each_part(svcpt, i, svc)
371 total += svcpt->scp_nthrs_running;
373 return seq_printf(m, "%d\n", total);
375 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
378 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
380 struct ptlrpc_service *svc = m->private;
382 return seq_printf(m, "%d\n",
383 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
387 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
388 const char __user *buffer,
389 size_t count, loff_t *off)
391 struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
393 int rc = lprocfs_write_helper(buffer, count, &val);
398 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
401 spin_lock(&svc->srv_lock);
402 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
403 spin_unlock(&svc->srv_lock);
407 svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
409 spin_unlock(&svc->srv_lock);
413 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
419 extern struct nrs_core nrs_core;
422 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
424 * \param[in] state The policy state
426 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
431 case NRS_POL_STATE_INVALID:
433 case NRS_POL_STATE_STOPPED:
435 case NRS_POL_STATE_STOPPING:
437 case NRS_POL_STATE_STARTING:
439 case NRS_POL_STATE_STARTED:
445 * Obtains status information for \a policy.
447 * Information is copied in \a info.
449 * \param[in] policy The policy
450 * \param[out] info Holds returned status information
452 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
453 struct ptlrpc_nrs_pol_info *info)
455 LASSERT(policy != NULL);
456 LASSERT(info != NULL);
457 assert_spin_locked(&policy->pol_nrs->nrs_lock);
459 memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
461 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
462 info->pi_state = policy->pol_state;
464 * XXX: These are accessed without holding
465 * ptlrpc_service_part::scp_req_lock.
467 info->pi_req_queued = policy->pol_req_queued;
468 info->pi_req_started = policy->pol_req_started;
472 * Reads and prints policy status information for all policies of a PTLRPC
475 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
477 struct ptlrpc_service *svc = m->private;
478 struct ptlrpc_service_part *svcpt;
479 struct ptlrpc_nrs *nrs;
480 struct ptlrpc_nrs_policy *policy;
481 struct ptlrpc_nrs_pol_info *infos;
482 struct ptlrpc_nrs_pol_info tmp;
484 unsigned pol_idx = 0;
490 * Serialize NRS core lprocfs operations with policy registration/
493 mutex_lock(&nrs_core.nrs_mutex);
496 * Use the first service partition's regular NRS head in order to obtain
497 * the number of policies registered with NRS heads of this service. All
498 * service partitions will have the same number of policies.
500 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
502 spin_lock(&nrs->nrs_lock);
503 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
504 spin_unlock(&nrs->nrs_lock);
506 OBD_ALLOC(infos, num_pols * sizeof(*infos));
513 ptlrpc_service_for_each_part(svcpt, i, svc) {
514 nrs = nrs_svcpt2nrs(svcpt, hp);
515 spin_lock(&nrs->nrs_lock);
519 list_for_each_entry(policy, &nrs->nrs_policy_list,
521 LASSERT(pol_idx < num_pols);
523 nrs_policy_get_info_locked(policy, &tmp);
525 * Copy values when handling the first service
529 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
531 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
532 sizeof(tmp.pi_state));
533 infos[pol_idx].pi_fallback = tmp.pi_fallback;
535 * For the rest of the service partitions
536 * sanity-check the values we get.
539 LASSERT(strncmp(infos[pol_idx].pi_name,
541 NRS_POL_NAME_MAX) == 0);
543 * Not asserting ptlrpc_nrs_pol_info::pi_state,
544 * because it may be different between
545 * instances of the same policy in different
546 * service partitions.
548 LASSERT(infos[pol_idx].pi_fallback ==
552 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
553 infos[pol_idx].pi_req_started += tmp.pi_req_started;
557 spin_unlock(&nrs->nrs_lock);
561 * Policy status information output is in YAML format.
577 * high_priority_requests:
590 seq_printf(m, "%s\n",
591 !hp ? "\nregular_requests:" : "high_priority_requests:");
593 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
594 seq_printf(m, " - name: %s\n"
598 " active: %-20d\n\n",
599 infos[pol_idx].pi_name,
600 nrs_state2str(infos[pol_idx].pi_state),
601 infos[pol_idx].pi_fallback ? "yes" : "no",
602 (int)infos[pol_idx].pi_req_queued,
603 (int)infos[pol_idx].pi_req_started);
606 if (!hp && nrs_svc_has_hp(svc)) {
607 memset(infos, 0, num_pols * sizeof(*infos));
610 * Redo the processing for the service's HP NRS heads' policies.
618 OBD_FREE(infos, num_pols * sizeof(*infos));
620 mutex_unlock(&nrs_core.nrs_mutex);
626 * The longest valid command string is the maximum policy name size, plus the
627 * length of the " reg" substring
629 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
632 * Starts and stops a given policy on a PTLRPC service.
634 * Commands consist of the policy name, followed by an optional [reg|hp] token;
635 * if the optional token is omitted, the operation is performed on both the
636 * regular and high-priority (if the service has one) NRS head.
638 static ssize_t ptlrpc_lprocfs_nrs_seq_write(struct file *file,
639 const char __user *buffer,
640 size_t count, loff_t *off)
642 struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
643 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
645 char *cmd_copy = NULL;
649 if (count >= LPROCFS_NRS_WR_MAX_CMD) {
654 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
660 * strsep() modifies its argument, so keep a copy
664 if (copy_from_user(cmd, buffer, count)) {
671 token = strsep(&cmd, " ");
673 if (strlen(token) > NRS_POL_NAME_MAX - 1) {
679 * No [reg|hp] token has been specified
685 * The second token is either NULL, or an optional [reg|hp] string
687 if (strcmp(cmd, "reg") == 0)
688 queue = PTLRPC_NRS_QUEUE_REG;
689 else if (strcmp(cmd, "hp") == 0)
690 queue = PTLRPC_NRS_QUEUE_HP;
698 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc)) {
701 } else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
702 queue = PTLRPC_NRS_QUEUE_REG;
705 * Serialize NRS core lprocfs operations with policy registration/
708 mutex_lock(&nrs_core.nrs_mutex);
710 rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
713 mutex_unlock(&nrs_core.nrs_mutex);
716 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
718 return rc < 0 ? rc : count;
720 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
724 struct ptlrpc_srh_iterator {
727 struct ptlrpc_request *srhi_req;
731 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
732 struct ptlrpc_srh_iterator *srhi,
736 struct ptlrpc_request *req;
738 if (srhi->srhi_req != NULL &&
739 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
740 srhi->srhi_seq <= seq) {
741 /* If srhi_req was set previously, hasn't been culled and
742 * we're searching for a seq on or after it (i.e. more
743 * recent), search from it onwards.
744 * Since the service history is LRU (i.e. culled reqs will
745 * be near the head), we shouldn't have to do long
747 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
748 "%s:%d: seek seq %llu, request seq %llu\n",
749 svcpt->scp_service->srv_name, svcpt->scp_cpt,
750 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
751 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
752 "%s:%d: seek offset %llu, request seq %llu, last culled %llu\n",
753 svcpt->scp_service->srv_name, svcpt->scp_cpt,
754 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
755 e = &srhi->srhi_req->rq_history_list;
757 /* search from start */
758 e = svcpt->scp_hist_reqs.next;
761 while (e != &svcpt->scp_hist_reqs) {
762 req = list_entry(e, struct ptlrpc_request, rq_history_list);
764 if (req->rq_history_seq >= seq) {
765 srhi->srhi_seq = req->rq_history_seq;
766 srhi->srhi_req = req;
776 * ptlrpc history sequence is used as "position" of seq_file, in some case,
777 * seq_read() will increase "position" to indicate reading the next
778 * element, however, low bits of history sequence are reserved for CPT id
779 * (check the details from comments before ptlrpc_req_add_history), which
780 * means seq_read() might change CPT id of history sequence and never
781 * finish reading of requests on a CPT. To make it work, we have to shift
782 * CPT id to high bits and timestamp to low bits, so seq_read() will only
783 * increase timestamp which can correctly indicate the next position.
786 /* convert seq_file pos to cpt */
787 #define PTLRPC_REQ_POS2CPT(svc, pos) \
788 ((svc)->srv_cpt_bits == 0 ? 0 : \
789 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
791 /* make up seq_file pos from cpt */
792 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
793 ((svc)->srv_cpt_bits == 0 ? 0 : \
794 (cpt) << (64 - (svc)->srv_cpt_bits))
796 /* convert sequence to position */
797 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
798 ((svc)->srv_cpt_bits == 0 ? (seq) : \
799 ((seq) >> (svc)->srv_cpt_bits) | \
800 ((seq) << (64 - (svc)->srv_cpt_bits)))
802 /* convert position to sequence */
803 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
804 ((svc)->srv_cpt_bits == 0 ? (pos) : \
805 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
806 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
809 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
811 struct ptlrpc_service *svc = s->private;
812 struct ptlrpc_service_part *svcpt;
813 struct ptlrpc_srh_iterator *srhi;
818 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
819 CWARN("Failed to read request history because size of loff_t %d can't match size of u64\n",
820 (int)sizeof(loff_t));
824 OBD_ALLOC(srhi, sizeof(*srhi));
829 srhi->srhi_req = NULL;
831 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
833 ptlrpc_service_for_each_part(svcpt, i, svc) {
834 if (i < cpt) /* skip */
836 if (i > cpt) /* make up the lowest position for this CPT */
837 *pos = PTLRPC_REQ_CPT2POS(svc, i);
839 spin_lock(&svcpt->scp_lock);
840 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
841 PTLRPC_REQ_POS2SEQ(svc, *pos));
842 spin_unlock(&svcpt->scp_lock);
844 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
850 OBD_FREE(srhi, sizeof(*srhi));
855 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
857 struct ptlrpc_srh_iterator *srhi = iter;
860 OBD_FREE(srhi, sizeof(*srhi));
864 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
865 void *iter, loff_t *pos)
867 struct ptlrpc_service *svc = s->private;
868 struct ptlrpc_srh_iterator *srhi = iter;
869 struct ptlrpc_service_part *svcpt;
874 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
875 svcpt = svc->srv_parts[i];
877 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
878 srhi->srhi_req = NULL;
879 seq = srhi->srhi_seq = 0;
880 } else { /* the next sequence */
881 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
884 spin_lock(&svcpt->scp_lock);
885 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
886 spin_unlock(&svcpt->scp_lock);
888 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
894 OBD_FREE(srhi, sizeof(*srhi));
898 /* common ost/mdt so_req_printer */
899 void target_print_req(void *seq_file, struct ptlrpc_request *req)
901 /* Called holding srv_lock with irqs disabled.
902 * Print specific req contents and a newline.
903 * CAVEAT EMPTOR: check request message length before printing!!!
904 * You might have received any old crap so you must be just as
905 * careful here as the service's request parser!!! */
906 struct seq_file *sf = seq_file;
908 switch (req->rq_phase) {
910 /* still awaiting a service thread's attention, or rejected
911 * because the generic request message didn't unpack */
912 seq_printf(sf, "<not swabbed>\n");
914 case RQ_PHASE_INTERPRET:
915 /* being handled, so basic msg swabbed, and opc is valid
916 * but racing with mds_handle() */
917 case RQ_PHASE_COMPLETE:
918 /* been handled by mds_handle() reply state possibly still
920 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
923 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
926 EXPORT_SYMBOL(target_print_req);
928 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
930 struct ptlrpc_service *svc = s->private;
931 struct ptlrpc_srh_iterator *srhi = iter;
932 struct ptlrpc_service_part *svcpt;
933 struct ptlrpc_request *req;
936 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
938 svcpt = svc->srv_parts[srhi->srhi_idx];
940 spin_lock(&svcpt->scp_lock);
942 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
945 req = srhi->srhi_req;
947 /* Print common req fields.
948 * CAVEAT EMPTOR: we're racing with the service handler
949 * here. The request could contain any old crap, so you
950 * must be just as careful as the service's request
951 * parser. Currently I only print stuff here I know is OK
952 * to look at coz it was set up in request_in_callback()!!! */
953 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%ld:%lds(%+lds) ",
954 req->rq_history_seq, libcfs_nid2str(req->rq_self),
955 libcfs_id2str(req->rq_peer), req->rq_xid,
956 req->rq_reqlen, ptlrpc_rqphase2str(req),
957 req->rq_arrival_time.tv_sec,
958 req->rq_sent - req->rq_arrival_time.tv_sec,
959 req->rq_sent - req->rq_deadline);
960 if (svc->srv_ops.so_req_printer == NULL)
963 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
966 spin_unlock(&svcpt->scp_lock);
971 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
973 static struct seq_operations sops = {
974 .start = ptlrpc_lprocfs_svc_req_history_start,
975 .stop = ptlrpc_lprocfs_svc_req_history_stop,
976 .next = ptlrpc_lprocfs_svc_req_history_next,
977 .show = ptlrpc_lprocfs_svc_req_history_show,
979 struct seq_file *seqf;
982 rc = seq_open(file, &sops);
986 seqf = file->private_data;
987 seqf->private = PDE_DATA(inode);
991 /* See also lprocfs_rd_timeouts */
992 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
994 struct ptlrpc_service *svc = m->private;
995 struct ptlrpc_service_part *svcpt;
1003 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1008 ptlrpc_service_for_each_part(svcpt, i, svc) {
1009 cur = at_get(&svcpt->scp_at_estimate);
1010 worst = svcpt->scp_at_estimate.at_worst_ever;
1011 worstt = svcpt->scp_at_estimate.at_worst_time;
1012 s2dhms(&ts, get_seconds() - worstt);
1014 seq_printf(m, "%10s : cur %3u worst %3u (at %ld, "
1015 DHMS_FMT" ago) ", "service",
1016 cur, worst, worstt, DHMS_VARS(&ts));
1018 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1023 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1025 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1027 struct ptlrpc_service *svc = m->private;
1028 return seq_printf(m, "%d", svc->srv_hpreq_ratio);
1031 static ssize_t ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file,
1032 const char __user *buffer,
1036 struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
1040 rc = lprocfs_write_helper(buffer, count, &val);
1047 spin_lock(&svc->srv_lock);
1048 svc->srv_hpreq_ratio = val;
1049 spin_unlock(&svc->srv_lock);
1053 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1055 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1056 struct ptlrpc_service *svc)
1058 struct lprocfs_vars lproc_vars[] = {
1059 {.name = "high_priority_ratio",
1060 .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1062 {.name = "req_buffer_history_len",
1063 .fops = &ptlrpc_lprocfs_req_history_len_fops,
1065 {.name = "req_buffer_history_max",
1066 .fops = &ptlrpc_lprocfs_req_history_max_fops,
1068 {.name = "threads_min",
1069 .fops = &ptlrpc_lprocfs_threads_min_fops,
1071 {.name = "threads_max",
1072 .fops = &ptlrpc_lprocfs_threads_max_fops,
1074 {.name = "threads_started",
1075 .fops = &ptlrpc_lprocfs_threads_started_fops,
1077 {.name = "timeouts",
1078 .fops = &ptlrpc_lprocfs_timeouts_fops,
1080 {.name = "nrs_policies",
1081 .fops = &ptlrpc_lprocfs_nrs_fops,
1085 static struct file_operations req_history_fops = {
1086 .owner = THIS_MODULE,
1087 .open = ptlrpc_lprocfs_svc_req_history_open,
1089 .llseek = seq_lseek,
1090 .release = lprocfs_seq_release,
1095 ptlrpc_lprocfs_register(entry, svc->srv_name,
1096 "stats", &svc->srv_procroot,
1099 if (svc->srv_procroot == NULL)
1102 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1104 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1105 0400, &req_history_fops, svc);
1107 CWARN("Error adding the req_history file\n");
1110 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1112 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1113 &obddev->obd_svc_procroot,
1114 &obddev->obd_svc_stats);
1116 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1118 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1120 struct lprocfs_stats *svc_stats;
1121 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1122 int opc = opcode_offset(op);
1124 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1125 if (svc_stats == NULL || opc <= 0)
1127 LASSERT(opc < LUSTRE_MAX_OPCODES);
1128 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1129 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1132 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1134 struct lprocfs_stats *svc_stats;
1137 if (!req->rq_import)
1139 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1142 idx = lustre_msg_get_opc(req->rq_reqmsg);
1145 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1148 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1151 LASSERTF(0, "unsupported opcode %u\n", idx);
1155 lprocfs_counter_add(svc_stats, idx, bytes);
1158 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1160 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1162 if (svc->srv_procroot != NULL)
1163 lprocfs_remove(&svc->srv_procroot);
1166 lprocfs_free_stats(&svc->srv_stats);
1169 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1171 if (obd->obd_svc_procroot)
1172 lprocfs_remove(&obd->obd_svc_procroot);
1174 if (obd->obd_svc_stats)
1175 lprocfs_free_stats(&obd->obd_svc_stats);
1177 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1180 #define BUFLEN (UUID_MAX + 5)
1182 int lprocfs_wr_evict_client(struct file *file, const char __user *buffer,
1183 size_t count, loff_t *off)
1185 struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1189 OBD_ALLOC(kbuf, BUFLEN);
1194 * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1195 * bytes into kbuf, to ensure that the string is NUL-terminated.
1196 * UUID_MAX should include a trailing NUL already.
1198 if (copy_from_user(kbuf, buffer,
1199 min_t(unsigned long, BUFLEN - 1, count))) {
1203 tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1204 /* Kludge code(deadlock situation): the lprocfs lock has been held
1205 * since the client is evicted by writing client's
1206 * uuid/nid to procfs "evict_client" entry. However,
1207 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1208 * the proc entries under the being destroyed export{}, so I have
1209 * to drop the lock at first here.
1210 * - jay, jxiong@clusterfs.com */
1211 class_incref(obd, __func__, current);
1213 if (strncmp(tmpbuf, "nid:", 4) == 0)
1214 obd_export_evict_by_nid(obd, tmpbuf + 4);
1215 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1216 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1218 obd_export_evict_by_uuid(obd, tmpbuf);
1220 class_decref(obd, __func__, current);
1223 OBD_FREE(kbuf, BUFLEN);
1226 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1230 int lprocfs_wr_ping(struct file *file, const char __user *buffer,
1231 size_t count, loff_t *off)
1233 struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1234 struct ptlrpc_request *req;
1237 LPROCFS_CLIMP_CHECK(obd);
1238 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1239 LPROCFS_CLIMP_EXIT(obd);
1243 req->rq_send_state = LUSTRE_IMP_FULL;
1245 rc = ptlrpc_queue_wait(req);
1247 ptlrpc_req_finished(req);
1252 EXPORT_SYMBOL(lprocfs_wr_ping);
1254 /* Write the connection UUID to this file to attempt to connect to that node.
1255 * The connection UUID is a node's primary NID. For example,
1256 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1258 int lprocfs_wr_import(struct file *file, const char __user *buffer,
1259 size_t count, loff_t *off)
1261 struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1262 struct obd_import *imp = obd->u.cli.cl_import;
1267 const char prefix[] = "connection=";
1268 const int prefix_len = sizeof(prefix) - 1;
1270 if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1273 OBD_ALLOC(kbuf, count + 1);
1277 if (copy_from_user(kbuf, buffer, count)) {
1284 /* only support connection=uuid::instance now */
1285 if (strncmp(prefix, kbuf, prefix_len) != 0) {
1290 uuid = kbuf + prefix_len;
1291 ptr = strstr(uuid, "::");
1298 ptr += strlen("::");
1299 inst = simple_strtol(ptr, &endptr, 10);
1301 CERROR("config: wrong instance # %s\n", ptr);
1302 } else if (inst != imp->imp_connect_data.ocd_instance) {
1303 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1304 imp->imp_obd->obd_name,
1305 imp->imp_connect_data.ocd_instance, inst);
1308 CDEBUG(D_INFO, "IR: %s has already been connecting to new target(%u)\n",
1309 imp->imp_obd->obd_name, inst);
1314 ptlrpc_recover_import(imp, uuid, 1);
1317 OBD_FREE(kbuf, count + 1);
1320 EXPORT_SYMBOL(lprocfs_wr_import);
1322 int lprocfs_rd_pinger_recov(struct seq_file *m, void *n)
1324 struct obd_device *obd = m->private;
1325 struct obd_import *imp = obd->u.cli.cl_import;
1328 LPROCFS_CLIMP_CHECK(obd);
1329 rc = seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1330 LPROCFS_CLIMP_EXIT(obd);
1334 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1336 int lprocfs_wr_pinger_recov(struct file *file, const char __user *buffer,
1337 size_t count, loff_t *off)
1339 struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1340 struct client_obd *cli = &obd->u.cli;
1341 struct obd_import *imp = cli->cl_import;
1344 rc = lprocfs_write_helper(buffer, count, &val);
1348 if (val != 0 && val != 1)
1351 LPROCFS_CLIMP_CHECK(obd);
1352 spin_lock(&imp->imp_lock);
1353 imp->imp_no_pinger_recover = !val;
1354 spin_unlock(&imp->imp_lock);
1355 LPROCFS_CLIMP_EXIT(obd);
1360 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1362 #endif /* CONFIG_PROC_FS */