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_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_read_req_history_len(char *page, char **start, off_t off,
262 int count, int *eof, void *data)
264 struct ptlrpc_service *svc = data;
265 struct ptlrpc_service_part *svcpt;
271 ptlrpc_service_for_each_part(svcpt, i, svc)
272 total += svcpt->scp_hist_nrqbds;
274 return snprintf(page, count, "%d\n", total);
278 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
279 int count, int *eof, void *data)
281 struct ptlrpc_service *svc = data;
282 struct ptlrpc_service_part *svcpt;
287 ptlrpc_service_for_each_part(svcpt, i, svc)
288 total += svc->srv_hist_nrqbds_cpt_max;
290 return snprintf(page, count, "%d\n", total);
294 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
295 unsigned long count, void *data)
297 struct ptlrpc_service *svc = data;
302 rc = lprocfs_write_helper(buffer, count, &val);
309 /* This sanity check is more of an insanity check; we can still
310 * hose a kernel by allowing the request history to grow too
312 bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >>
314 if (val > totalram_pages/(2 * bufpages))
317 spin_lock(&svc->srv_lock);
320 svc->srv_hist_nrqbds_cpt_max = 0;
322 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
324 spin_unlock(&svc->srv_lock);
330 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
331 int count, int *eof, void *data)
333 struct ptlrpc_service *svc = data;
335 return snprintf(page, count, "%d\n",
336 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
340 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
341 unsigned long count, void *data)
343 struct ptlrpc_service *svc = data;
345 int rc = lprocfs_write_helper(buffer, count, &val);
350 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
353 spin_lock(&svc->srv_lock);
354 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
355 spin_unlock(&svc->srv_lock);
359 svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
361 spin_unlock(&svc->srv_lock);
367 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
368 int count, int *eof, void *data)
370 struct ptlrpc_service *svc = data;
371 struct ptlrpc_service_part *svcpt;
375 ptlrpc_service_for_each_part(svcpt, i, svc)
376 total += svcpt->scp_nthrs_running;
378 return snprintf(page, count, "%d\n", total);
382 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
383 int count, int *eof, void *data)
385 struct ptlrpc_service *svc = data;
387 return snprintf(page, count, "%d\n",
388 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
392 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
393 unsigned long count, void *data)
395 struct ptlrpc_service *svc = data;
397 int rc = lprocfs_write_helper(buffer, count, &val);
402 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
405 spin_lock(&svc->srv_lock);
406 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
407 spin_unlock(&svc->srv_lock);
411 svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
413 spin_unlock(&svc->srv_lock);
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_rd_nrs(char *page, char **start, off_t off,
479 int count, int *eof, void *data)
481 struct ptlrpc_service *svc = data;
482 struct ptlrpc_service_part *svcpt;
483 struct ptlrpc_nrs *nrs;
484 struct ptlrpc_nrs_policy *policy;
485 struct ptlrpc_nrs_pol_info *infos;
486 struct ptlrpc_nrs_pol_info tmp;
488 unsigned pol_idx = 0;
496 * Serialize NRS core lprocfs operations with policy registration/
499 mutex_lock(&nrs_core.nrs_mutex);
502 * Use the first service partition's regular NRS head in order to obtain
503 * the number of policies registered with NRS heads of this service. All
504 * service partitions will have the same number of policies.
506 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
508 spin_lock(&nrs->nrs_lock);
509 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
510 spin_unlock(&nrs->nrs_lock);
512 OBD_ALLOC(infos, num_pols * sizeof(*infos));
514 GOTO(out, rc = -ENOMEM);
517 ptlrpc_service_for_each_part(svcpt, i, svc) {
518 nrs = nrs_svcpt2nrs(svcpt, hp);
519 spin_lock(&nrs->nrs_lock);
523 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
525 LASSERT(pol_idx < num_pols);
527 nrs_policy_get_info_locked(policy, &tmp);
529 * Copy values when handling the first service
533 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
535 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
536 sizeof(tmp.pi_state));
537 infos[pol_idx].pi_fallback = tmp.pi_fallback;
539 * For the rest of the service partitions
540 * sanity-check the values we get.
543 LASSERT(strncmp(infos[pol_idx].pi_name,
545 NRS_POL_NAME_MAX) == 0);
547 * Not asserting ptlrpc_nrs_pol_info::pi_state,
548 * because it may be different between
549 * instances of the same policy in different
550 * service partitions.
552 LASSERT(infos[pol_idx].pi_fallback ==
556 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
557 infos[pol_idx].pi_req_started += tmp.pi_req_started;
561 spin_unlock(&nrs->nrs_lock);
565 * Policy status information output is in YAML format.
581 * high_priority_requests:
594 rc2 = snprintf(page + rc, count - rc,
596 "\nregular_requests:" :
597 "high_priority_requests:");
599 if (rc2 >= count - rc) {
600 /** Output was truncated */
601 GOTO(out, rc = -EFBIG);
606 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
607 rc2 = snprintf(page + rc, count - rc,
612 " active: %-20d\n\n",
613 infos[pol_idx].pi_name,
614 nrs_state2str(infos[pol_idx].pi_state),
615 infos[pol_idx].pi_fallback ? "yes" : "no",
616 (int)infos[pol_idx].pi_req_queued,
617 (int)infos[pol_idx].pi_req_started);
620 if (rc2 >= count - rc) {
621 /** Output was truncated */
622 GOTO(out, rc = -EFBIG);
628 if (!hp && nrs_svc_has_hp(svc)) {
629 memset(infos, 0, num_pols * sizeof(*infos));
632 * Redo the processing for the service's HP NRS heads' policies.
642 OBD_FREE(infos, num_pols * sizeof(*infos));
644 mutex_unlock(&nrs_core.nrs_mutex);
650 * The longest valid command string is the maxium policy name size, plus the
651 * length of the " reg" substring
653 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
656 * Starts and stops a given policy on a PTLRPC service.
658 * Commands consist of the policy name, followed by an optional [reg|hp] token;
659 * if the optional token is omitted, the operation is performed on both the
660 * regular and high-priority (if the service has one) NRS head.
662 static int ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
663 unsigned long count, void *data)
665 struct ptlrpc_service *svc = data;
666 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
668 char *cmd_copy = NULL;
673 if (count >= LPROCFS_NRS_WR_MAX_CMD)
674 GOTO(out, rc = -EINVAL);
676 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
678 GOTO(out, rc = -ENOMEM);
680 * strsep() modifies its argument, so keep a copy
684 if (copy_from_user(cmd, buffer, count))
685 GOTO(out, rc = -EFAULT);
689 token = strsep(&cmd, " ");
691 if (strlen(token) > NRS_POL_NAME_MAX - 1)
692 GOTO(out, rc = -EINVAL);
695 * No [reg|hp] token has been specified
701 * The second token is either NULL, or an optional [reg|hp] string
703 if (strcmp(cmd, "reg") == 0)
704 queue = PTLRPC_NRS_QUEUE_REG;
705 else if (strcmp(cmd, "hp") == 0)
706 queue = PTLRPC_NRS_QUEUE_HP;
708 GOTO(out, rc = -EINVAL);
712 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
713 GOTO(out, rc = -ENODEV);
714 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
715 queue = PTLRPC_NRS_QUEUE_REG;
718 * Serialize NRS core lprocfs operations with policy registration/
721 mutex_lock(&nrs_core.nrs_mutex);
723 rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
726 mutex_unlock(&nrs_core.nrs_mutex);
729 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
731 RETURN(rc < 0 ? rc : count);
736 struct ptlrpc_srh_iterator {
739 struct ptlrpc_request *srhi_req;
743 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
744 struct ptlrpc_srh_iterator *srhi,
748 struct ptlrpc_request *req;
750 if (srhi->srhi_req != NULL &&
751 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
752 srhi->srhi_seq <= seq) {
753 /* If srhi_req was set previously, hasn't been culled and
754 * we're searching for a seq on or after it (i.e. more
755 * recent), search from it onwards.
756 * Since the service history is LRU (i.e. culled reqs will
757 * be near the head), we shouldn't have to do long
759 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
760 "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
761 svcpt->scp_service->srv_name, svcpt->scp_cpt,
762 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
763 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
764 "%s:%d: seek offset "LPU64", request seq "LPU64", "
765 "last culled "LPU64"\n",
766 svcpt->scp_service->srv_name, svcpt->scp_cpt,
767 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
768 e = &srhi->srhi_req->rq_history_list;
770 /* search from start */
771 e = svcpt->scp_hist_reqs.next;
774 while (e != &svcpt->scp_hist_reqs) {
775 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
777 if (req->rq_history_seq >= seq) {
778 srhi->srhi_seq = req->rq_history_seq;
779 srhi->srhi_req = req;
789 * ptlrpc history sequence is used as "position" of seq_file, in some case,
790 * seq_read() will increase "position" to indicate reading the next
791 * element, however, low bits of history sequence are reserved for CPT id
792 * (check the details from comments before ptlrpc_req_add_history), which
793 * means seq_read() might change CPT id of history sequence and never
794 * finish reading of requests on a CPT. To make it work, we have to shift
795 * CPT id to high bits and timestamp to low bits, so seq_read() will only
796 * increase timestamp which can correctly indicate the next position.
799 /* convert seq_file pos to cpt */
800 #define PTLRPC_REQ_POS2CPT(svc, pos) \
801 ((svc)->srv_cpt_bits == 0 ? 0 : \
802 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
804 /* make up seq_file pos from cpt */
805 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
806 ((svc)->srv_cpt_bits == 0 ? 0 : \
807 (cpt) << (64 - (svc)->srv_cpt_bits))
809 /* convert sequence to position */
810 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
811 ((svc)->srv_cpt_bits == 0 ? (seq) : \
812 ((seq) >> (svc)->srv_cpt_bits) | \
813 ((seq) << (64 - (svc)->srv_cpt_bits)))
815 /* convert position to sequence */
816 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
817 ((svc)->srv_cpt_bits == 0 ? (pos) : \
818 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
819 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
822 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
824 struct ptlrpc_service *svc = s->private;
825 struct ptlrpc_service_part *svcpt;
826 struct ptlrpc_srh_iterator *srhi;
831 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
832 CWARN("Failed to read request history because size of loff_t "
833 "%d can't match size of u64\n", (int)sizeof(loff_t));
837 OBD_ALLOC(srhi, sizeof(*srhi));
842 srhi->srhi_req = NULL;
844 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
846 ptlrpc_service_for_each_part(svcpt, i, svc) {
847 if (i < cpt) /* skip */
849 if (i > cpt) /* make up the lowest position for this CPT */
850 *pos = PTLRPC_REQ_CPT2POS(svc, i);
852 spin_lock(&svcpt->scp_lock);
853 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
854 PTLRPC_REQ_POS2SEQ(svc, *pos));
855 spin_unlock(&svcpt->scp_lock);
857 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
863 OBD_FREE(srhi, sizeof(*srhi));
868 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
870 struct ptlrpc_srh_iterator *srhi = iter;
873 OBD_FREE(srhi, sizeof(*srhi));
877 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
878 void *iter, loff_t *pos)
880 struct ptlrpc_service *svc = s->private;
881 struct ptlrpc_srh_iterator *srhi = iter;
882 struct ptlrpc_service_part *svcpt;
887 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
888 svcpt = svc->srv_parts[i];
890 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
891 srhi->srhi_req = NULL;
892 seq = srhi->srhi_seq = 0;
893 } else { /* the next sequence */
894 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
897 spin_lock(&svcpt->scp_lock);
898 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
899 spin_unlock(&svcpt->scp_lock);
901 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
907 OBD_FREE(srhi, sizeof(*srhi));
911 /* common ost/mdt so_req_printer */
912 void target_print_req(void *seq_file, struct ptlrpc_request *req)
914 /* Called holding srv_lock with irqs disabled.
915 * Print specific req contents and a newline.
916 * CAVEAT EMPTOR: check request message length before printing!!!
917 * You might have received any old crap so you must be just as
918 * careful here as the service's request parser!!! */
919 struct seq_file *sf = seq_file;
921 switch (req->rq_phase) {
923 /* still awaiting a service thread's attention, or rejected
924 * because the generic request message didn't unpack */
925 seq_printf(sf, "<not swabbed>\n");
927 case RQ_PHASE_INTERPRET:
928 /* being handled, so basic msg swabbed, and opc is valid
929 * but racing with mds_handle() */
930 case RQ_PHASE_COMPLETE:
931 /* been handled by mds_handle() reply state possibly still
933 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
936 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
939 EXPORT_SYMBOL(target_print_req);
941 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
943 struct ptlrpc_service *svc = s->private;
944 struct ptlrpc_srh_iterator *srhi = iter;
945 struct ptlrpc_service_part *svcpt;
946 struct ptlrpc_request *req;
949 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
951 svcpt = svc->srv_parts[srhi->srhi_idx];
953 spin_lock(&svcpt->scp_lock);
955 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
958 req = srhi->srhi_req;
960 /* Print common req fields.
961 * CAVEAT EMPTOR: we're racing with the service handler
962 * here. The request could contain any old crap, so you
963 * must be just as careful as the service's request
964 * parser. Currently I only print stuff here I know is OK
965 * to look at coz it was set up in request_in_callback()!!! */
966 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
967 req->rq_history_seq, libcfs_nid2str(req->rq_self),
968 libcfs_id2str(req->rq_peer), req->rq_xid,
969 req->rq_reqlen, ptlrpc_rqphase2str(req),
970 req->rq_arrival_time.tv_sec,
971 req->rq_sent - req->rq_arrival_time.tv_sec,
972 req->rq_sent - req->rq_deadline);
973 if (svc->srv_ops.so_req_printer == NULL)
976 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
979 spin_unlock(&svcpt->scp_lock);
984 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
986 static struct seq_operations sops = {
987 .start = ptlrpc_lprocfs_svc_req_history_start,
988 .stop = ptlrpc_lprocfs_svc_req_history_stop,
989 .next = ptlrpc_lprocfs_svc_req_history_next,
990 .show = ptlrpc_lprocfs_svc_req_history_show,
992 struct proc_dir_entry *dp = PDE(inode);
993 struct seq_file *seqf;
996 LPROCFS_ENTRY_CHECK(dp);
997 rc = seq_open(file, &sops);
1001 seqf = file->private_data;
1002 seqf->private = dp->data;
1006 /* See also lprocfs_rd_timeouts */
1007 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1008 int count, int *eof, void *data)
1010 struct ptlrpc_service *svc = data;
1011 struct ptlrpc_service_part *svcpt;
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 (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 class_incref(obd, __FUNCTION__, current);
1246 if (strncmp(tmpbuf, "nid:", 4) == 0)
1247 obd_export_evict_by_nid(obd, tmpbuf + 4);
1248 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1249 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1251 obd_export_evict_by_uuid(obd, tmpbuf);
1253 class_decref(obd, __FUNCTION__, current);
1256 OBD_FREE(kbuf, BUFLEN);
1259 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1263 int lprocfs_wr_ping(struct file *file, const char *buffer,
1264 unsigned long count, void *data)
1266 struct obd_device *obd = data;
1267 struct ptlrpc_request *req;
1271 LPROCFS_CLIMP_CHECK(obd);
1272 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1273 LPROCFS_CLIMP_EXIT(obd);
1277 req->rq_send_state = LUSTRE_IMP_FULL;
1279 rc = ptlrpc_queue_wait(req);
1281 ptlrpc_req_finished(req);
1286 EXPORT_SYMBOL(lprocfs_wr_ping);
1288 /* Write the connection UUID to this file to attempt to connect to that node.
1289 * The connection UUID is a node's primary NID. For example,
1290 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1292 int lprocfs_wr_import(struct file *file, const char *buffer,
1293 unsigned long count, void *data)
1295 struct obd_device *obd = data;
1296 struct obd_import *imp = obd->u.cli.cl_import;
1301 const char prefix[] = "connection=";
1302 const int prefix_len = sizeof(prefix) - 1;
1304 if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1307 OBD_ALLOC(kbuf, count + 1);
1311 if (copy_from_user(kbuf, buffer, count))
1312 GOTO(out, count = -EFAULT);
1316 /* only support connection=uuid::instance now */
1317 if (strncmp(prefix, kbuf, prefix_len) != 0)
1318 GOTO(out, count = -EINVAL);
1320 uuid = kbuf + prefix_len;
1321 ptr = strstr(uuid, "::");
1328 ptr += strlen("::");
1329 inst = simple_strtol(ptr, &endptr, 10);
1331 CERROR("config: wrong instance # %s\n", ptr);
1332 } else if (inst != imp->imp_connect_data.ocd_instance) {
1333 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1334 "target(%u/%u), reconnecting...\n",
1335 imp->imp_obd->obd_name,
1336 imp->imp_connect_data.ocd_instance, inst);
1339 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1341 imp->imp_obd->obd_name, inst);
1346 ptlrpc_recover_import(imp, uuid, 1);
1349 OBD_FREE(kbuf, count + 1);
1352 EXPORT_SYMBOL(lprocfs_wr_import);
1354 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1355 int count, int *eof, void *data)
1357 struct obd_device *obd = data;
1358 struct obd_import *imp = obd->u.cli.cl_import;
1361 LPROCFS_CLIMP_CHECK(obd);
1362 rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1363 LPROCFS_CLIMP_EXIT(obd);
1367 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1369 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1370 unsigned long count, void *data)
1372 struct obd_device *obd = data;
1373 struct client_obd *cli = &obd->u.cli;
1374 struct obd_import *imp = cli->cl_import;
1377 rc = lprocfs_write_helper(buffer, count, &val);
1381 if (val != 0 && val != 1)
1384 LPROCFS_CLIMP_CHECK(obd);
1385 spin_lock(&imp->imp_lock);
1386 imp->imp_no_pinger_recover = !val;
1387 spin_unlock(&imp->imp_lock);
1388 LPROCFS_CLIMP_EXIT(obd);
1393 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1395 #endif /* LPROCFS */