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 <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 { LDLM_ENQUEUE, "ldlm_enqueue" },
98 { LDLM_CONVERT, "ldlm_convert" },
99 { LDLM_CANCEL, "ldlm_cancel" },
100 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
101 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
102 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
103 { LDLM_SET_INFO, "ldlm_set_info" },
104 { MGS_CONNECT, "mgs_connect" },
105 { MGS_DISCONNECT, "mgs_disconnect" },
106 { MGS_EXCEPTION, "mgs_exception" },
107 { MGS_TARGET_REG, "mgs_target_reg" },
108 { MGS_TARGET_DEL, "mgs_target_del" },
109 { MGS_SET_INFO, "mgs_set_info" },
110 { MGS_CONFIG_READ, "mgs_config_read" },
111 { OBD_PING, "obd_ping" },
112 { OBD_LOG_CANCEL, "llog_origin_handle_cancel" },
113 { OBD_QC_CALLBACK, "obd_quota_callback" },
114 { OBD_IDX_READ, "dt_index_read" },
115 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
116 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
117 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
118 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
119 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
120 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
121 { LLOG_CATINFO, "llog_catinfo" },
122 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
123 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
124 { QUOTA_DQACQ, "quota_acquire" },
125 { QUOTA_DQREL, "quota_release" },
126 { SEQ_QUERY, "seq_query" },
127 { SEC_CTX_INIT, "sec_ctx_init" },
128 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
129 { SEC_CTX_FINI, "sec_ctx_fini" },
130 { FLD_QUERY, "fld_query" }
136 } ll_eopcode_table[EXTRA_LAST_OPC] = {
137 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
138 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
139 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
140 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
141 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
142 { MDS_REINT_SETATTR, "mds_reint_setattr" },
143 { MDS_REINT_CREATE, "mds_reint_create" },
144 { MDS_REINT_LINK, "mds_reint_link" },
145 { MDS_REINT_UNLINK, "mds_reint_unlink" },
146 { MDS_REINT_RENAME, "mds_reint_rename" },
147 { MDS_REINT_OPEN, "mds_reint_open" },
148 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
149 { BRW_READ_BYTES, "read_bytes" },
150 { BRW_WRITE_BYTES, "write_bytes" },
153 const char *ll_opcode2str(__u32 opcode)
155 /* When one of the assertions below fail, chances are that:
156 * 1) A new opcode was added in include/lustre/lustre_idl.h,
157 * but is missing from the table above.
158 * or 2) The opcode space was renumbered or rearranged,
159 * and the opcode_offset() function in
160 * ptlrpc_internal.h needs to be modified.
162 __u32 offset = opcode_offset(opcode);
163 LASSERTF(offset < LUSTRE_MAX_OPCODES,
164 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
165 offset, LUSTRE_MAX_OPCODES);
166 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
167 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
168 offset, ll_rpc_opcode_table[offset].opcode, opcode);
169 return ll_rpc_opcode_table[offset].opname;
172 const char* ll_eopcode2str(__u32 opcode)
174 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
175 return ll_eopcode_table[opcode].opname;
178 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
179 char *name, struct proc_dir_entry **procroot_ret,
180 struct lprocfs_stats **stats_ret)
182 struct proc_dir_entry *svc_procroot;
183 struct lprocfs_stats *svc_stats;
185 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
188 LASSERT(*procroot_ret == NULL);
189 LASSERT(*stats_ret == NULL);
191 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
192 if (svc_stats == NULL)
196 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
197 if (IS_ERR(svc_procroot)) {
198 lprocfs_free_stats(&svc_stats);
205 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
206 svc_counter_config, "req_waittime", "usec");
207 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
208 svc_counter_config, "req_qdepth", "reqs");
209 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
210 svc_counter_config, "req_active", "reqs");
211 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
212 svc_counter_config, "req_timeout", "sec");
213 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
214 svc_counter_config, "reqbuf_avail", "bufs");
215 for (i = 0; i < EXTRA_LAST_OPC; i++) {
219 case BRW_WRITE_BYTES:
227 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
229 ll_eopcode2str(i), units);
231 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
232 __u32 opcode = ll_rpc_opcode_table[i].opcode;
233 lprocfs_counter_init(svc_stats,
234 EXTRA_MAX_OPCODES + i, svc_counter_config,
235 ll_opcode2str(opcode), "usec");
238 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
241 lprocfs_remove(&svc_procroot);
242 lprocfs_free_stats(&svc_stats);
245 *procroot_ret = svc_procroot;
246 *stats_ret = svc_stats;
251 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
252 int count, int *eof, void *data)
254 struct ptlrpc_service *svc = data;
255 struct ptlrpc_service_part *svcpt;
261 ptlrpc_service_for_each_part(svcpt, i, svc)
262 total += svcpt->scp_hist_nrqbds;
264 return snprintf(page, count, "%d\n", total);
268 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
269 int count, int *eof, void *data)
271 struct ptlrpc_service *svc = data;
272 struct ptlrpc_service_part *svcpt;
277 ptlrpc_service_for_each_part(svcpt, i, svc)
278 total += svc->srv_hist_nrqbds_cpt_max;
280 return snprintf(page, count, "%d\n", total);
284 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
285 unsigned long count, void *data)
287 struct ptlrpc_service *svc = data;
292 rc = lprocfs_write_helper(buffer, count, &val);
299 /* This sanity check is more of an insanity check; we can still
300 * hose a kernel by allowing the request history to grow too
302 bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
303 if (val > cfs_num_physpages/(2 * bufpages))
306 spin_lock(&svc->srv_lock);
309 svc->srv_hist_nrqbds_cpt_max = 0;
311 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
313 spin_unlock(&svc->srv_lock);
319 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
320 int count, int *eof, void *data)
322 struct ptlrpc_service *svc = data;
324 return snprintf(page, count, "%d\n",
325 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
329 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
330 unsigned long count, void *data)
332 struct ptlrpc_service *svc = data;
334 int rc = lprocfs_write_helper(buffer, count, &val);
339 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
342 spin_lock(&svc->srv_lock);
343 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
344 spin_unlock(&svc->srv_lock);
348 svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
350 spin_unlock(&svc->srv_lock);
356 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
357 int count, int *eof, void *data)
359 struct ptlrpc_service *svc = data;
360 struct ptlrpc_service_part *svcpt;
364 LASSERT(svc->srv_parts != NULL);
365 ptlrpc_service_for_each_part(svcpt, i, svc)
366 total += svcpt->scp_nthrs_running;
368 return snprintf(page, count, "%d\n", total);
372 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
373 int count, int *eof, void *data)
375 struct ptlrpc_service *svc = data;
377 return snprintf(page, count, "%d\n",
378 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
382 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
383 unsigned long count, void *data)
385 struct ptlrpc_service *svc = data;
387 int rc = lprocfs_write_helper(buffer, count, &val);
392 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
395 spin_lock(&svc->srv_lock);
396 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
397 spin_unlock(&svc->srv_lock);
401 svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
403 spin_unlock(&svc->srv_lock);
412 extern struct nrs_core nrs_core;
415 * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
417 * \param[in] state The policy state
420 nrs_state2str(enum ptlrpc_nrs_pol_state state)
425 case NRS_POL_STATE_INVALID:
427 case NRS_POL_STATE_UNAVAIL:
429 case NRS_POL_STATE_STOPPED:
431 case NRS_POL_STATE_STOPPING:
433 case NRS_POL_STATE_STARTING:
435 case NRS_POL_STATE_STARTED:
441 * Obtains status information for \a policy.
443 * Information is copied in \a info.
445 * \param[in] policy The policy
446 * \param[out] info Holds returned status information
449 nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
450 struct ptlrpc_nrs_pol_info *info)
452 LASSERT(policy != NULL);
453 LASSERT(info != NULL);
454 LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
456 memcpy(info->pi_name, policy->pol_name, NRS_POL_NAME_MAX);
458 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
459 info->pi_state = policy->pol_state;
461 * XXX: These are accessed without holding
462 * ptlrpc_service_part::scp_req_lock.
464 info->pi_req_queued = policy->pol_req_queued;
465 info->pi_req_started = policy->pol_req_started;
469 * Reads and prints policy status information for all policies of a PTLRPC
473 ptlrpc_lprocfs_rd_nrs(char *page, char **start, off_t off,
474 int count, int *eof, void *data)
476 struct ptlrpc_service *svc = data;
477 struct ptlrpc_service_part *svcpt;
478 struct ptlrpc_nrs *nrs;
479 struct ptlrpc_nrs_policy *policy;
480 struct ptlrpc_nrs_pol_info *infos;
481 struct ptlrpc_nrs_pol_info tmp;
483 unsigned pol_idx = 0;
491 * Serialize NRS core lprocfs operations with policy registration/
494 mutex_lock(&nrs_core.nrs_mutex);
497 * Use the first service partition's regular NRS head in order to obtain
498 * the number of policies registered with NRS heads of this service. All
499 * service partitions will have the same number of policies.
501 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
503 spin_lock(&nrs->nrs_lock);
504 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
505 spin_unlock(&nrs->nrs_lock);
507 OBD_ALLOC(infos, num_pols * sizeof(*infos));
509 GOTO(out, rc = -ENOMEM);
512 ptlrpc_service_for_each_part(svcpt, i, svc) {
513 nrs = nrs_svcpt2nrs(svcpt, hp);
514 spin_lock(&nrs->nrs_lock);
518 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
520 LASSERT(pol_idx < num_pols);
522 nrs_policy_get_info_locked(policy, &tmp);
524 * Copy values when handling the first service
528 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
530 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
531 sizeof(tmp.pi_state));
532 infos[pol_idx].pi_fallback = tmp.pi_fallback;
534 * For the rest of the service partitions
535 * sanity-check the values we get.
538 LASSERT(strncmp(infos[pol_idx].pi_name,
540 NRS_POL_NAME_MAX) == 0);
542 * Not asserting ptlrpc_nrs_pol_info::pi_state,
543 * because it may be different between
544 * instances of the same policy in different
545 * service partitions.
547 LASSERT(infos[pol_idx].pi_fallback ==
551 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
552 infos[pol_idx].pi_req_started += tmp.pi_req_started;
556 spin_unlock(&nrs->nrs_lock);
560 * Policy status information output is in YAML format.
576 * high_priority_requests:
589 rc2 = snprintf(page + rc, count - rc,
591 "\nregular_requests:" :
592 "high_priority_requests:");
594 if (rc2 >= count - rc) {
595 /** Output was truncated */
596 GOTO(out, rc = -EFBIG);
601 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
602 rc2 = snprintf(page + rc, count - rc,
607 " active: %-20d\n\n",
608 infos[pol_idx].pi_name,
609 nrs_state2str(infos[pol_idx].pi_state),
610 infos[pol_idx].pi_fallback ? "yes" : "no",
611 (int)infos[pol_idx].pi_req_queued,
612 (int)infos[pol_idx].pi_req_started);
615 if (rc2 >= count - rc) {
616 /** Output was truncated */
617 GOTO(out, rc = -EFBIG);
623 if (!hp && nrs_svc_has_hp(svc)) {
624 memset(infos, 0, num_pols * sizeof(*infos));
627 * Redo the processing for the service's HP NRS heads' policies.
637 OBD_FREE(infos, num_pols * sizeof(*infos));
639 mutex_unlock(&nrs_core.nrs_mutex);
645 * The longest valid command string is the maxium policy name size, plus the
646 * length of the " reg" substring
648 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
651 * Starts and stops a given policy on a PTLRPC service.
653 * Commands consist of the policy name, followed by an optional [reg|hp] token;
654 * if the optional token is omitted, the operation is performed on both the
655 * regular and high-priority (if the service has one) NRS head.
658 ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
659 unsigned long count, void *data)
661 struct ptlrpc_service *svc = data;
662 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
664 char *cmd_copy = NULL;
669 if (count >= LPROCFS_NRS_WR_MAX_CMD)
670 GOTO(out, rc = -EINVAL);
672 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
674 GOTO(out, rc = -ENOMEM);
676 * strsep() modifies its argument, so keep a copy
680 if (cfs_copy_from_user(cmd, buffer, count))
681 GOTO(out, rc = -EFAULT);
685 token = strsep(&cmd, " ");
687 if (strlen(token) > NRS_POL_NAME_MAX - 1)
688 GOTO(out, rc = -EINVAL);
691 * No [reg|hp] token has been specified
697 * The second token is either NULL, or an optional [reg|hp] string
699 if (strcmp(cmd, "reg") == 0)
700 queue = PTLRPC_NRS_QUEUE_REG;
701 else if (strcmp(cmd, "hp") == 0)
702 queue = PTLRPC_NRS_QUEUE_HP;
704 GOTO(out, rc = -EINVAL);
708 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
709 GOTO(out, rc = -ENODEV);
710 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
711 queue = PTLRPC_NRS_QUEUE_REG;
714 * Serialize NRS core lprocfs operations with policy registration/
717 mutex_lock(&nrs_core.nrs_mutex);
719 rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
722 mutex_unlock(&nrs_core.nrs_mutex);
725 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
727 RETURN(rc < 0 ? rc : count);
732 struct ptlrpc_srh_iterator {
735 struct ptlrpc_request *srhi_req;
739 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
740 struct ptlrpc_srh_iterator *srhi,
744 struct ptlrpc_request *req;
746 if (srhi->srhi_req != NULL &&
747 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
748 srhi->srhi_seq <= seq) {
749 /* If srhi_req was set previously, hasn't been culled and
750 * we're searching for a seq on or after it (i.e. more
751 * recent), search from it onwards.
752 * Since the service history is LRU (i.e. culled reqs will
753 * be near the head), we shouldn't have to do long
755 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
756 "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
757 svcpt->scp_service->srv_name, svcpt->scp_cpt,
758 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
759 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
760 "%s:%d: seek offset "LPU64", request seq "LPU64", "
761 "last culled "LPU64"\n",
762 svcpt->scp_service->srv_name, svcpt->scp_cpt,
763 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
764 e = &srhi->srhi_req->rq_history_list;
766 /* search from start */
767 e = svcpt->scp_hist_reqs.next;
770 while (e != &svcpt->scp_hist_reqs) {
771 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
773 if (req->rq_history_seq >= seq) {
774 srhi->srhi_seq = req->rq_history_seq;
775 srhi->srhi_req = req;
785 * ptlrpc history sequence is used as "position" of seq_file, in some case,
786 * seq_read() will increase "position" to indicate reading the next
787 * element, however, low bits of history sequence are reserved for CPT id
788 * (check the details from comments before ptlrpc_req_add_history), which
789 * means seq_read() might change CPT id of history sequence and never
790 * finish reading of requests on a CPT. To make it work, we have to shift
791 * CPT id to high bits and timestamp to low bits, so seq_read() will only
792 * increase timestamp which can correctly indicate the next position.
795 /* convert seq_file pos to cpt */
796 #define PTLRPC_REQ_POS2CPT(svc, pos) \
797 ((svc)->srv_cpt_bits == 0 ? 0 : \
798 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
800 /* make up seq_file pos from cpt */
801 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
802 ((svc)->srv_cpt_bits == 0 ? 0 : \
803 (cpt) << (64 - (svc)->srv_cpt_bits))
805 /* convert sequence to position */
806 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
807 ((svc)->srv_cpt_bits == 0 ? (seq) : \
808 ((seq) >> (svc)->srv_cpt_bits) | \
809 ((seq) << (64 - (svc)->srv_cpt_bits)))
811 /* convert position to sequence */
812 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
813 ((svc)->srv_cpt_bits == 0 ? (pos) : \
814 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
815 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
818 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
820 struct ptlrpc_service *svc = s->private;
821 struct ptlrpc_service_part *svcpt;
822 struct ptlrpc_srh_iterator *srhi;
827 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
828 CWARN("Failed to read request history because size of loff_t "
829 "%d can't match size of u64\n", (int)sizeof(loff_t));
833 OBD_ALLOC(srhi, sizeof(*srhi));
838 srhi->srhi_req = NULL;
840 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
842 ptlrpc_service_for_each_part(svcpt, i, svc) {
843 if (i < cpt) /* skip */
845 if (i > cpt) /* make up the lowest position for this CPT */
846 *pos = PTLRPC_REQ_CPT2POS(svc, i);
848 spin_lock(&svcpt->scp_lock);
849 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
850 PTLRPC_REQ_POS2SEQ(svc, *pos));
851 spin_unlock(&svcpt->scp_lock);
853 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
859 OBD_FREE(srhi, sizeof(*srhi));
864 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
866 struct ptlrpc_srh_iterator *srhi = iter;
869 OBD_FREE(srhi, sizeof(*srhi));
873 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
874 void *iter, loff_t *pos)
876 struct ptlrpc_service *svc = s->private;
877 struct ptlrpc_srh_iterator *srhi = iter;
878 struct ptlrpc_service_part *svcpt;
883 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
884 svcpt = svc->srv_parts[i];
886 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
887 srhi->srhi_req = NULL;
888 seq = srhi->srhi_seq = 0;
889 } else { /* the next sequence */
890 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
893 spin_lock(&svcpt->scp_lock);
894 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
895 spin_unlock(&svcpt->scp_lock);
897 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
903 OBD_FREE(srhi, sizeof(*srhi));
907 /* common ost/mdt so_req_printer */
908 void target_print_req(void *seq_file, struct ptlrpc_request *req)
910 /* Called holding srv_lock with irqs disabled.
911 * Print specific req contents and a newline.
912 * CAVEAT EMPTOR: check request message length before printing!!!
913 * You might have received any old crap so you must be just as
914 * careful here as the service's request parser!!! */
915 struct seq_file *sf = seq_file;
917 switch (req->rq_phase) {
919 /* still awaiting a service thread's attention, or rejected
920 * because the generic request message didn't unpack */
921 seq_printf(sf, "<not swabbed>\n");
923 case RQ_PHASE_INTERPRET:
924 /* being handled, so basic msg swabbed, and opc is valid
925 * but racing with mds_handle() */
926 case RQ_PHASE_COMPLETE:
927 /* been handled by mds_handle() reply state possibly still
929 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
932 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
935 EXPORT_SYMBOL(target_print_req);
937 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
939 struct ptlrpc_service *svc = s->private;
940 struct ptlrpc_srh_iterator *srhi = iter;
941 struct ptlrpc_service_part *svcpt;
942 struct ptlrpc_request *req;
945 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
947 svcpt = svc->srv_parts[srhi->srhi_idx];
949 spin_lock(&svcpt->scp_lock);
951 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
954 req = srhi->srhi_req;
956 /* Print common req fields.
957 * CAVEAT EMPTOR: we're racing with the service handler
958 * here. The request could contain any old crap, so you
959 * must be just as careful as the service's request
960 * parser. Currently I only print stuff here I know is OK
961 * to look at coz it was set up in request_in_callback()!!! */
962 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
963 req->rq_history_seq, libcfs_nid2str(req->rq_self),
964 libcfs_id2str(req->rq_peer), req->rq_xid,
965 req->rq_reqlen, ptlrpc_rqphase2str(req),
966 req->rq_arrival_time.tv_sec,
967 req->rq_sent - req->rq_arrival_time.tv_sec,
968 req->rq_sent - req->rq_deadline);
969 if (svc->srv_ops.so_req_printer == NULL)
972 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
975 spin_unlock(&svcpt->scp_lock);
980 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
982 static struct seq_operations sops = {
983 .start = ptlrpc_lprocfs_svc_req_history_start,
984 .stop = ptlrpc_lprocfs_svc_req_history_stop,
985 .next = ptlrpc_lprocfs_svc_req_history_next,
986 .show = ptlrpc_lprocfs_svc_req_history_show,
988 struct proc_dir_entry *dp = PDE(inode);
989 struct seq_file *seqf;
992 LPROCFS_ENTRY_AND_CHECK(dp);
993 rc = seq_open(file, &sops);
999 seqf = file->private_data;
1000 seqf->private = dp->data;
1004 /* See also lprocfs_rd_timeouts */
1005 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1006 int count, int *eof, void *data)
1008 struct ptlrpc_service *svc = data;
1009 struct ptlrpc_service_part *svcpt;
1018 LASSERT(svc->srv_parts != NULL);
1021 rc += snprintf(page + rc, count - rc,
1022 "adaptive timeouts off, using obd_timeout %u\n",
1027 ptlrpc_service_for_each_part(svcpt, i, svc) {
1028 cur = at_get(&svcpt->scp_at_estimate);
1029 worst = svcpt->scp_at_estimate.at_worst_ever;
1030 worstt = svcpt->scp_at_estimate.at_worst_time;
1031 s2dhms(&ts, cfs_time_current_sec() - worstt);
1033 nob = snprintf(page, count,
1034 "%10s : cur %3u worst %3u (at %ld, "
1035 DHMS_FMT" ago) ", "service",
1036 cur, worst, worstt, DHMS_VARS(&ts));
1038 nob = lprocfs_at_hist_helper(page, count, nob,
1039 &svcpt->scp_at_estimate);
1045 * NB: for lustre proc read, the read count must be less
1046 * than PAGE_SIZE, please see details in lprocfs_fops_read.
1047 * It's unlikely that we exceed PAGE_SIZE at here because
1048 * it means the service has more than 50 partitions.
1051 CWARN("Can't fit AT information of %s in one page, "
1052 "please contact with developer to fix this.\n",
1061 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
1062 int count, int *eof, void *data)
1064 struct ptlrpc_service *svc = data;
1065 int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
1069 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
1070 unsigned long count, void *data)
1072 struct ptlrpc_service *svc = data;
1076 rc = lprocfs_write_helper(buffer, count, &val);
1083 spin_lock(&svc->srv_lock);
1084 svc->srv_hpreq_ratio = val;
1085 spin_unlock(&svc->srv_lock);
1090 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1091 struct ptlrpc_service *svc)
1093 struct lprocfs_vars lproc_vars[] = {
1094 {.name = "high_priority_ratio",
1095 .read_fptr = ptlrpc_lprocfs_rd_hp_ratio,
1096 .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
1098 {.name = "req_buffer_history_len",
1099 .read_fptr = ptlrpc_lprocfs_read_req_history_len,
1101 {.name = "req_buffer_history_max",
1102 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
1103 .read_fptr = ptlrpc_lprocfs_read_req_history_max,
1105 {.name = "threads_min",
1106 .read_fptr = ptlrpc_lprocfs_rd_threads_min,
1107 .write_fptr = ptlrpc_lprocfs_wr_threads_min,
1109 {.name = "threads_max",
1110 .read_fptr = ptlrpc_lprocfs_rd_threads_max,
1111 .write_fptr = ptlrpc_lprocfs_wr_threads_max,
1113 {.name = "threads_started",
1114 .read_fptr = ptlrpc_lprocfs_rd_threads_started,
1116 {.name = "timeouts",
1117 .read_fptr = ptlrpc_lprocfs_rd_timeouts,
1119 {.name = "nrs_policies",
1120 .read_fptr = ptlrpc_lprocfs_rd_nrs,
1121 .write_fptr = ptlrpc_lprocfs_wr_nrs,
1125 static struct file_operations req_history_fops = {
1126 .owner = THIS_MODULE,
1127 .open = ptlrpc_lprocfs_svc_req_history_open,
1129 .llseek = seq_lseek,
1130 .release = lprocfs_seq_release,
1135 ptlrpc_lprocfs_register(entry, svc->srv_name,
1136 "stats", &svc->srv_procroot,
1139 if (svc->srv_procroot == NULL)
1142 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1144 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1145 0400, &req_history_fops, svc);
1147 CWARN("Error adding the req_history file\n");
1150 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1152 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1153 &obddev->obd_svc_procroot,
1154 &obddev->obd_svc_stats);
1156 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1158 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1160 struct lprocfs_stats *svc_stats;
1161 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1162 int opc = opcode_offset(op);
1164 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1165 if (svc_stats == NULL || opc <= 0)
1167 LASSERT(opc < LUSTRE_MAX_OPCODES);
1168 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1169 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1172 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1174 struct lprocfs_stats *svc_stats;
1177 if (!req->rq_import)
1179 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1182 idx = lustre_msg_get_opc(req->rq_reqmsg);
1185 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1188 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1191 LASSERTF(0, "unsupported opcode %u\n", idx);
1195 lprocfs_counter_add(svc_stats, idx, bytes);
1198 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1200 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1202 if (svc->srv_procroot != NULL)
1203 lprocfs_remove(&svc->srv_procroot);
1206 lprocfs_free_stats(&svc->srv_stats);
1209 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1211 if (obd->obd_svc_procroot)
1212 lprocfs_remove(&obd->obd_svc_procroot);
1214 if (obd->obd_svc_stats)
1215 lprocfs_free_stats(&obd->obd_svc_stats);
1217 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1220 #define BUFLEN (UUID_MAX + 5)
1222 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1223 unsigned long count, void *data)
1225 struct obd_device *obd = data;
1229 OBD_ALLOC(kbuf, BUFLEN);
1234 * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1235 * bytes into kbuf, to ensure that the string is NUL-terminated.
1236 * UUID_MAX should include a trailing NUL already.
1238 if (cfs_copy_from_user(kbuf, buffer,
1239 min_t(unsigned long, BUFLEN - 1, count))) {
1243 tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1244 /* Kludge code(deadlock situation): the lprocfs lock has been held
1245 * since the client is evicted by writting client's
1246 * uuid/nid to procfs "evict_client" entry. However,
1247 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1248 * the proc entries under the being destroyed export{}, so I have
1249 * to drop the lock at first here.
1250 * - jay, jxiong@clusterfs.com */
1252 class_incref(obd, __FUNCTION__, cfs_current());
1254 if (strncmp(tmpbuf, "nid:", 4) == 0)
1255 obd_export_evict_by_nid(obd, tmpbuf + 4);
1256 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1257 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1259 obd_export_evict_by_uuid(obd, tmpbuf);
1261 class_decref(obd, __FUNCTION__, cfs_current());
1265 OBD_FREE(kbuf, BUFLEN);
1268 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1272 int lprocfs_wr_ping(struct file *file, const char *buffer,
1273 unsigned long count, void *data)
1275 struct obd_device *obd = data;
1276 struct ptlrpc_request *req;
1280 LPROCFS_CLIMP_CHECK(obd);
1281 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1282 LPROCFS_CLIMP_EXIT(obd);
1286 req->rq_send_state = LUSTRE_IMP_FULL;
1288 rc = ptlrpc_queue_wait(req);
1290 ptlrpc_req_finished(req);
1295 EXPORT_SYMBOL(lprocfs_wr_ping);
1297 /* Write the connection UUID to this file to attempt to connect to that node.
1298 * The connection UUID is a node's primary NID. For example,
1299 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1301 int lprocfs_wr_import(struct file *file, const char *buffer,
1302 unsigned long count, void *data)
1304 struct obd_device *obd = data;
1305 struct obd_import *imp = obd->u.cli.cl_import;
1310 const char prefix[] = "connection=";
1311 const int prefix_len = sizeof(prefix) - 1;
1313 if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
1316 OBD_ALLOC(kbuf, count + 1);
1320 if (cfs_copy_from_user(kbuf, buffer, count))
1321 GOTO(out, count = -EFAULT);
1325 /* only support connection=uuid::instance now */
1326 if (strncmp(prefix, kbuf, prefix_len) != 0)
1327 GOTO(out, count = -EINVAL);
1329 uuid = kbuf + prefix_len;
1330 ptr = strstr(uuid, "::");
1337 ptr += strlen("::");
1338 inst = simple_strtol(ptr, &endptr, 10);
1340 CERROR("config: wrong instance # %s\n", ptr);
1341 } else if (inst != imp->imp_connect_data.ocd_instance) {
1342 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1343 "target(%u/%u), reconnecting...\n",
1344 imp->imp_obd->obd_name,
1345 imp->imp_connect_data.ocd_instance, inst);
1348 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1350 imp->imp_obd->obd_name, inst);
1355 ptlrpc_recover_import(imp, uuid, 1);
1358 OBD_FREE(kbuf, count + 1);
1361 EXPORT_SYMBOL(lprocfs_wr_import);
1363 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1364 int count, int *eof, void *data)
1366 struct obd_device *obd = data;
1367 struct obd_import *imp = obd->u.cli.cl_import;
1370 LPROCFS_CLIMP_CHECK(obd);
1371 rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1372 LPROCFS_CLIMP_EXIT(obd);
1376 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1378 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1379 unsigned long count, void *data)
1381 struct obd_device *obd = data;
1382 struct client_obd *cli = &obd->u.cli;
1383 struct obd_import *imp = cli->cl_import;
1386 rc = lprocfs_write_helper(buffer, count, &val);
1390 if (val != 0 && val != 1)
1393 LPROCFS_CLIMP_CHECK(obd);
1394 spin_lock(&imp->imp_lock);
1395 imp->imp_no_pinger_recover = !val;
1396 spin_unlock(&imp->imp_lock);
1397 LPROCFS_CLIMP_EXIT(obd);
1402 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1404 #endif /* LPROCFS */