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 { 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 { LDLM_ENQUEUE, "ldlm_enqueue" },
105 { LDLM_CONVERT, "ldlm_convert" },
106 { LDLM_CANCEL, "ldlm_cancel" },
107 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
108 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
109 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
110 { LDLM_SET_INFO, "ldlm_set_info" },
111 { MGS_CONNECT, "mgs_connect" },
112 { MGS_DISCONNECT, "mgs_disconnect" },
113 { MGS_EXCEPTION, "mgs_exception" },
114 { MGS_TARGET_REG, "mgs_target_reg" },
115 { MGS_TARGET_DEL, "mgs_target_del" },
116 { MGS_SET_INFO, "mgs_set_info" },
117 { MGS_CONFIG_READ, "mgs_config_read" },
118 { OBD_PING, "obd_ping" },
119 { OBD_LOG_CANCEL, "llog_origin_handle_cancel" },
120 { OBD_QC_CALLBACK, "obd_quota_callback" },
121 { OBD_IDX_READ, "dt_index_read" },
122 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
123 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
124 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
125 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
126 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
127 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
128 { LLOG_CATINFO, "llog_catinfo" },
129 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
130 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
131 { QUOTA_DQACQ, "quota_acquire" },
132 { QUOTA_DQREL, "quota_release" },
133 { SEQ_QUERY, "seq_query" },
134 { SEC_CTX_INIT, "sec_ctx_init" },
135 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
136 { SEC_CTX_FINI, "sec_ctx_fini" },
137 { FLD_QUERY, "fld_query" }
143 } ll_eopcode_table[EXTRA_LAST_OPC] = {
144 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
145 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
146 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
147 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
148 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
149 { MDS_REINT_SETATTR, "mds_reint_setattr" },
150 { MDS_REINT_CREATE, "mds_reint_create" },
151 { MDS_REINT_LINK, "mds_reint_link" },
152 { MDS_REINT_UNLINK, "mds_reint_unlink" },
153 { MDS_REINT_RENAME, "mds_reint_rename" },
154 { MDS_REINT_OPEN, "mds_reint_open" },
155 { MDS_REINT_SETXATTR, "mds_reint_setxattr" },
156 { BRW_READ_BYTES, "read_bytes" },
157 { BRW_WRITE_BYTES, "write_bytes" },
160 const char *ll_opcode2str(__u32 opcode)
162 /* When one of the assertions below fail, chances are that:
163 * 1) A new opcode was added in include/lustre/lustre_idl.h,
164 * but is missing from the table above.
165 * or 2) The opcode space was renumbered or rearranged,
166 * and the opcode_offset() function in
167 * ptlrpc_internal.h needs to be modified.
169 __u32 offset = opcode_offset(opcode);
170 LASSERTF(offset < LUSTRE_MAX_OPCODES,
171 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
172 offset, LUSTRE_MAX_OPCODES);
173 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
174 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
175 offset, ll_rpc_opcode_table[offset].opcode, opcode);
176 return ll_rpc_opcode_table[offset].opname;
179 const char* ll_eopcode2str(__u32 opcode)
181 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
182 return ll_eopcode_table[opcode].opname;
185 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
186 char *name, 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,0);
199 if (svc_stats == NULL)
203 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
204 if (IS_ERR(svc_procroot)) {
205 lprocfs_free_stats(&svc_stats);
212 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
213 svc_counter_config, "req_waittime", "usec");
214 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
215 svc_counter_config, "req_qdepth", "reqs");
216 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
217 svc_counter_config, "req_active", "reqs");
218 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
219 svc_counter_config, "req_timeout", "sec");
220 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
221 svc_counter_config, "reqbuf_avail", "bufs");
222 for (i = 0; i < EXTRA_LAST_OPC; i++) {
226 case BRW_WRITE_BYTES:
234 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
236 ll_eopcode2str(i), units);
238 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
239 __u32 opcode = ll_rpc_opcode_table[i].opcode;
240 lprocfs_counter_init(svc_stats,
241 EXTRA_MAX_OPCODES + i, svc_counter_config,
242 ll_opcode2str(opcode), "usec");
245 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
248 lprocfs_remove(&svc_procroot);
249 lprocfs_free_stats(&svc_stats);
252 *procroot_ret = svc_procroot;
253 *stats_ret = svc_stats;
258 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
259 int count, int *eof, void *data)
261 struct ptlrpc_service *svc = data;
262 struct ptlrpc_service_part *svcpt;
268 ptlrpc_service_for_each_part(svcpt, i, svc)
269 total += svcpt->scp_hist_nrqbds;
271 return snprintf(page, count, "%d\n", total);
275 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
276 int count, int *eof, void *data)
278 struct ptlrpc_service *svc = data;
279 struct ptlrpc_service_part *svcpt;
284 ptlrpc_service_for_each_part(svcpt, i, svc)
285 total += svc->srv_hist_nrqbds_cpt_max;
287 return snprintf(page, count, "%d\n", total);
291 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
292 unsigned long count, void *data)
294 struct ptlrpc_service *svc = data;
299 rc = lprocfs_write_helper(buffer, count, &val);
306 /* This sanity check is more of an insanity check; we can still
307 * hose a kernel by allowing the request history to grow too
309 bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
310 if (val > cfs_num_physpages/(2 * bufpages))
313 spin_lock(&svc->srv_lock);
316 svc->srv_hist_nrqbds_cpt_max = 0;
318 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
320 spin_unlock(&svc->srv_lock);
326 ptlrpc_lprocfs_rd_threads_min(char *page, char **start, off_t off,
327 int count, int *eof, void *data)
329 struct ptlrpc_service *svc = data;
331 return snprintf(page, count, "%d\n",
332 svc->srv_nthrs_cpt_init * svc->srv_ncpts);
336 ptlrpc_lprocfs_wr_threads_min(struct file *file, const char *buffer,
337 unsigned long count, void *data)
339 struct ptlrpc_service *svc = data;
341 int rc = lprocfs_write_helper(buffer, count, &val);
346 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
349 spin_lock(&svc->srv_lock);
350 if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
351 spin_unlock(&svc->srv_lock);
355 svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
357 spin_unlock(&svc->srv_lock);
363 ptlrpc_lprocfs_rd_threads_started(char *page, char **start, off_t off,
364 int count, int *eof, void *data)
366 struct ptlrpc_service *svc = data;
367 struct ptlrpc_service_part *svcpt;
371 LASSERT(svc->srv_parts != NULL);
372 ptlrpc_service_for_each_part(svcpt, i, svc)
373 total += svcpt->scp_nthrs_running;
375 return snprintf(page, count, "%d\n", total);
379 ptlrpc_lprocfs_rd_threads_max(char *page, char **start, off_t off,
380 int count, int *eof, void *data)
382 struct ptlrpc_service *svc = data;
384 return snprintf(page, count, "%d\n",
385 svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
389 ptlrpc_lprocfs_wr_threads_max(struct file *file, const char *buffer,
390 unsigned long count, void *data)
392 struct ptlrpc_service *svc = data;
394 int rc = lprocfs_write_helper(buffer, count, &val);
399 if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
402 spin_lock(&svc->srv_lock);
403 if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
404 spin_unlock(&svc->srv_lock);
408 svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
410 spin_unlock(&svc->srv_lock);
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
427 nrs_state2str(enum ptlrpc_nrs_pol_state state)
432 case NRS_POL_STATE_INVALID:
434 case NRS_POL_STATE_UNAVAIL:
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
456 nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
457 struct ptlrpc_nrs_pol_info *info)
459 LASSERT(policy != NULL);
460 LASSERT(info != NULL);
461 LASSERT(spin_is_locked(&policy->pol_nrs->nrs_lock));
463 memcpy(info->pi_name, policy->pol_name, NRS_POL_NAME_MAX);
465 info->pi_fallback = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
466 info->pi_state = policy->pol_state;
468 * XXX: These are accessed without holding
469 * ptlrpc_service_part::scp_req_lock.
471 info->pi_req_queued = policy->pol_req_queued;
472 info->pi_req_started = policy->pol_req_started;
476 * Reads and prints policy status information for all policies of a PTLRPC
480 ptlrpc_lprocfs_rd_nrs(char *page, char **start, off_t off,
481 int count, int *eof, void *data)
483 struct ptlrpc_service *svc = data;
484 struct ptlrpc_service_part *svcpt;
485 struct ptlrpc_nrs *nrs;
486 struct ptlrpc_nrs_policy *policy;
487 struct ptlrpc_nrs_pol_info *infos;
488 struct ptlrpc_nrs_pol_info tmp;
490 unsigned pol_idx = 0;
498 * Serialize NRS core lprocfs operations with policy registration/
501 mutex_lock(&nrs_core.nrs_mutex);
504 * Use the first service partition's regular NRS head in order to obtain
505 * the number of policies registered with NRS heads of this service. All
506 * service partitions will have the same number of policies.
508 nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
510 spin_lock(&nrs->nrs_lock);
511 num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
512 spin_unlock(&nrs->nrs_lock);
514 OBD_ALLOC(infos, num_pols * sizeof(*infos));
516 GOTO(out, rc = -ENOMEM);
519 ptlrpc_service_for_each_part(svcpt, i, svc) {
520 nrs = nrs_svcpt2nrs(svcpt, hp);
521 spin_lock(&nrs->nrs_lock);
525 cfs_list_for_each_entry(policy, &nrs->nrs_policy_list,
527 LASSERT(pol_idx < num_pols);
529 nrs_policy_get_info_locked(policy, &tmp);
531 * Copy values when handling the first service
535 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
537 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
538 sizeof(tmp.pi_state));
539 infos[pol_idx].pi_fallback = tmp.pi_fallback;
541 * For the rest of the service partitions
542 * sanity-check the values we get.
545 LASSERT(strncmp(infos[pol_idx].pi_name,
547 NRS_POL_NAME_MAX) == 0);
549 * Not asserting ptlrpc_nrs_pol_info::pi_state,
550 * because it may be different between
551 * instances of the same policy in different
552 * service partitions.
554 LASSERT(infos[pol_idx].pi_fallback ==
558 infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
559 infos[pol_idx].pi_req_started += tmp.pi_req_started;
563 spin_unlock(&nrs->nrs_lock);
567 * Policy status information output is in YAML format.
583 * high_priority_requests:
596 rc2 = snprintf(page + rc, count - rc,
598 "\nregular_requests:" :
599 "high_priority_requests:");
601 if (rc2 >= count - rc) {
602 /** Output was truncated */
603 GOTO(out, rc = -EFBIG);
608 for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
609 rc2 = snprintf(page + rc, count - rc,
614 " active: %-20d\n\n",
615 infos[pol_idx].pi_name,
616 nrs_state2str(infos[pol_idx].pi_state),
617 infos[pol_idx].pi_fallback ? "yes" : "no",
618 (int)infos[pol_idx].pi_req_queued,
619 (int)infos[pol_idx].pi_req_started);
622 if (rc2 >= count - rc) {
623 /** Output was truncated */
624 GOTO(out, rc = -EFBIG);
630 if (!hp && nrs_svc_has_hp(svc)) {
631 memset(infos, 0, num_pols * sizeof(*infos));
634 * Redo the processing for the service's HP NRS heads' policies.
644 OBD_FREE(infos, num_pols * sizeof(*infos));
646 mutex_unlock(&nrs_core.nrs_mutex);
652 * The longest valid command string is the maxium policy name size, plus the
653 * length of the " reg" substring
655 #define LPROCFS_NRS_WR_MAX_CMD (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
658 * Starts and stops a given policy on a PTLRPC service.
660 * Commands consist of the policy name, followed by an optional [reg|hp] token;
661 * if the optional token is omitted, the operation is performed on both the
662 * regular and high-priority (if the service has one) NRS head.
665 ptlrpc_lprocfs_wr_nrs(struct file *file, const char *buffer,
666 unsigned long count, void *data)
668 struct ptlrpc_service *svc = data;
669 enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
671 char *cmd_copy = NULL;
676 if (count >= LPROCFS_NRS_WR_MAX_CMD)
677 GOTO(out, rc = -EINVAL);
679 OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
681 GOTO(out, rc = -ENOMEM);
683 * strsep() modifies its argument, so keep a copy
687 if (cfs_copy_from_user(cmd, buffer, count))
688 GOTO(out, rc = -EFAULT);
692 token = strsep(&cmd, " ");
694 if (strlen(token) > NRS_POL_NAME_MAX - 1)
695 GOTO(out, rc = -EINVAL);
698 * No [reg|hp] token has been specified
704 * The second token is either NULL, or an optional [reg|hp] string
706 if (strcmp(cmd, "reg") == 0)
707 queue = PTLRPC_NRS_QUEUE_REG;
708 else if (strcmp(cmd, "hp") == 0)
709 queue = PTLRPC_NRS_QUEUE_HP;
711 GOTO(out, rc = -EINVAL);
715 if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
716 GOTO(out, rc = -ENODEV);
717 else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
718 queue = PTLRPC_NRS_QUEUE_REG;
721 * Serialize NRS core lprocfs operations with policy registration/
724 mutex_lock(&nrs_core.nrs_mutex);
726 rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
729 mutex_unlock(&nrs_core.nrs_mutex);
732 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
734 RETURN(rc < 0 ? rc : count);
739 struct ptlrpc_srh_iterator {
742 struct ptlrpc_request *srhi_req;
746 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
747 struct ptlrpc_srh_iterator *srhi,
751 struct ptlrpc_request *req;
753 if (srhi->srhi_req != NULL &&
754 srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
755 srhi->srhi_seq <= seq) {
756 /* If srhi_req was set previously, hasn't been culled and
757 * we're searching for a seq on or after it (i.e. more
758 * recent), search from it onwards.
759 * Since the service history is LRU (i.e. culled reqs will
760 * be near the head), we shouldn't have to do long
762 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
763 "%s:%d: seek seq "LPU64", request seq "LPU64"\n",
764 svcpt->scp_service->srv_name, svcpt->scp_cpt,
765 srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
766 LASSERTF(!cfs_list_empty(&svcpt->scp_hist_reqs),
767 "%s:%d: seek offset "LPU64", request seq "LPU64", "
768 "last culled "LPU64"\n",
769 svcpt->scp_service->srv_name, svcpt->scp_cpt,
770 seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
771 e = &srhi->srhi_req->rq_history_list;
773 /* search from start */
774 e = svcpt->scp_hist_reqs.next;
777 while (e != &svcpt->scp_hist_reqs) {
778 req = cfs_list_entry(e, struct ptlrpc_request, rq_history_list);
780 if (req->rq_history_seq >= seq) {
781 srhi->srhi_seq = req->rq_history_seq;
782 srhi->srhi_req = req;
792 * ptlrpc history sequence is used as "position" of seq_file, in some case,
793 * seq_read() will increase "position" to indicate reading the next
794 * element, however, low bits of history sequence are reserved for CPT id
795 * (check the details from comments before ptlrpc_req_add_history), which
796 * means seq_read() might change CPT id of history sequence and never
797 * finish reading of requests on a CPT. To make it work, we have to shift
798 * CPT id to high bits and timestamp to low bits, so seq_read() will only
799 * increase timestamp which can correctly indicate the next position.
802 /* convert seq_file pos to cpt */
803 #define PTLRPC_REQ_POS2CPT(svc, pos) \
804 ((svc)->srv_cpt_bits == 0 ? 0 : \
805 (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
807 /* make up seq_file pos from cpt */
808 #define PTLRPC_REQ_CPT2POS(svc, cpt) \
809 ((svc)->srv_cpt_bits == 0 ? 0 : \
810 (cpt) << (64 - (svc)->srv_cpt_bits))
812 /* convert sequence to position */
813 #define PTLRPC_REQ_SEQ2POS(svc, seq) \
814 ((svc)->srv_cpt_bits == 0 ? (seq) : \
815 ((seq) >> (svc)->srv_cpt_bits) | \
816 ((seq) << (64 - (svc)->srv_cpt_bits)))
818 /* convert position to sequence */
819 #define PTLRPC_REQ_POS2SEQ(svc, pos) \
820 ((svc)->srv_cpt_bits == 0 ? (pos) : \
821 ((__u64)(pos) << (svc)->srv_cpt_bits) | \
822 ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
825 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
827 struct ptlrpc_service *svc = s->private;
828 struct ptlrpc_service_part *svcpt;
829 struct ptlrpc_srh_iterator *srhi;
834 if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
835 CWARN("Failed to read request history because size of loff_t "
836 "%d can't match size of u64\n", (int)sizeof(loff_t));
840 OBD_ALLOC(srhi, sizeof(*srhi));
845 srhi->srhi_req = NULL;
847 cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
849 ptlrpc_service_for_each_part(svcpt, i, svc) {
850 if (i < cpt) /* skip */
852 if (i > cpt) /* make up the lowest position for this CPT */
853 *pos = PTLRPC_REQ_CPT2POS(svc, i);
855 spin_lock(&svcpt->scp_lock);
856 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
857 PTLRPC_REQ_POS2SEQ(svc, *pos));
858 spin_unlock(&svcpt->scp_lock);
860 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
866 OBD_FREE(srhi, sizeof(*srhi));
871 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
873 struct ptlrpc_srh_iterator *srhi = iter;
876 OBD_FREE(srhi, sizeof(*srhi));
880 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
881 void *iter, loff_t *pos)
883 struct ptlrpc_service *svc = s->private;
884 struct ptlrpc_srh_iterator *srhi = iter;
885 struct ptlrpc_service_part *svcpt;
890 for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
891 svcpt = svc->srv_parts[i];
893 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
894 srhi->srhi_req = NULL;
895 seq = srhi->srhi_seq = 0;
896 } else { /* the next sequence */
897 seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
900 spin_lock(&svcpt->scp_lock);
901 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
902 spin_unlock(&svcpt->scp_lock);
904 *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
910 OBD_FREE(srhi, sizeof(*srhi));
914 /* common ost/mdt so_req_printer */
915 void target_print_req(void *seq_file, struct ptlrpc_request *req)
917 /* Called holding srv_lock with irqs disabled.
918 * Print specific req contents and a newline.
919 * CAVEAT EMPTOR: check request message length before printing!!!
920 * You might have received any old crap so you must be just as
921 * careful here as the service's request parser!!! */
922 struct seq_file *sf = seq_file;
924 switch (req->rq_phase) {
926 /* still awaiting a service thread's attention, or rejected
927 * because the generic request message didn't unpack */
928 seq_printf(sf, "<not swabbed>\n");
930 case RQ_PHASE_INTERPRET:
931 /* being handled, so basic msg swabbed, and opc is valid
932 * but racing with mds_handle() */
933 case RQ_PHASE_COMPLETE:
934 /* been handled by mds_handle() reply state possibly still
936 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
939 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
942 EXPORT_SYMBOL(target_print_req);
944 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
946 struct ptlrpc_service *svc = s->private;
947 struct ptlrpc_srh_iterator *srhi = iter;
948 struct ptlrpc_service_part *svcpt;
949 struct ptlrpc_request *req;
952 LASSERT(srhi->srhi_idx < svc->srv_ncpts);
954 svcpt = svc->srv_parts[srhi->srhi_idx];
956 spin_lock(&svcpt->scp_lock);
958 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
961 req = srhi->srhi_req;
963 /* Print common req fields.
964 * CAVEAT EMPTOR: we're racing with the service handler
965 * here. The request could contain any old crap, so you
966 * must be just as careful as the service's request
967 * parser. Currently I only print stuff here I know is OK
968 * to look at coz it was set up in request_in_callback()!!! */
969 seq_printf(s, LPD64":%s:%s:x"LPU64":%d:%s:%ld:%lds(%+lds) ",
970 req->rq_history_seq, libcfs_nid2str(req->rq_self),
971 libcfs_id2str(req->rq_peer), req->rq_xid,
972 req->rq_reqlen, ptlrpc_rqphase2str(req),
973 req->rq_arrival_time.tv_sec,
974 req->rq_sent - req->rq_arrival_time.tv_sec,
975 req->rq_sent - req->rq_deadline);
976 if (svc->srv_ops.so_req_printer == NULL)
979 svc->srv_ops.so_req_printer(s, srhi->srhi_req);
982 spin_unlock(&svcpt->scp_lock);
987 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
989 static struct seq_operations sops = {
990 .start = ptlrpc_lprocfs_svc_req_history_start,
991 .stop = ptlrpc_lprocfs_svc_req_history_stop,
992 .next = ptlrpc_lprocfs_svc_req_history_next,
993 .show = ptlrpc_lprocfs_svc_req_history_show,
995 struct proc_dir_entry *dp = PDE(inode);
996 struct seq_file *seqf;
999 LPROCFS_ENTRY_AND_CHECK(dp);
1000 rc = seq_open(file, &sops);
1006 seqf = file->private_data;
1007 seqf->private = dp->data;
1011 /* See also lprocfs_rd_timeouts */
1012 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
1013 int count, int *eof, void *data)
1015 struct ptlrpc_service *svc = data;
1016 struct ptlrpc_service_part *svcpt;
1025 LASSERT(svc->srv_parts != NULL);
1028 rc += snprintf(page + rc, count - rc,
1029 "adaptive timeouts off, using obd_timeout %u\n",
1034 ptlrpc_service_for_each_part(svcpt, i, svc) {
1035 cur = at_get(&svcpt->scp_at_estimate);
1036 worst = svcpt->scp_at_estimate.at_worst_ever;
1037 worstt = svcpt->scp_at_estimate.at_worst_time;
1038 s2dhms(&ts, cfs_time_current_sec() - worstt);
1040 nob = snprintf(page, count,
1041 "%10s : cur %3u worst %3u (at %ld, "
1042 DHMS_FMT" ago) ", "service",
1043 cur, worst, worstt, DHMS_VARS(&ts));
1045 nob = lprocfs_at_hist_helper(page, count, nob,
1046 &svcpt->scp_at_estimate);
1052 * NB: for lustre proc read, the read count must be less
1053 * than PAGE_SIZE, please see details in lprocfs_fops_read.
1054 * It's unlikely that we exceed PAGE_SIZE at here because
1055 * it means the service has more than 50 partitions.
1058 CWARN("Can't fit AT information of %s in one page, "
1059 "please contact with developer to fix this.\n",
1068 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
1069 int count, int *eof, void *data)
1071 struct ptlrpc_service *svc = data;
1072 int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
1076 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
1077 unsigned long count, void *data)
1079 struct ptlrpc_service *svc = data;
1083 rc = lprocfs_write_helper(buffer, count, &val);
1090 spin_lock(&svc->srv_lock);
1091 svc->srv_hpreq_ratio = val;
1092 spin_unlock(&svc->srv_lock);
1097 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1098 struct ptlrpc_service *svc)
1100 struct lprocfs_vars lproc_vars[] = {
1101 {.name = "high_priority_ratio",
1102 .read_fptr = ptlrpc_lprocfs_rd_hp_ratio,
1103 .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
1105 {.name = "req_buffer_history_len",
1106 .read_fptr = ptlrpc_lprocfs_read_req_history_len,
1108 {.name = "req_buffer_history_max",
1109 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
1110 .read_fptr = ptlrpc_lprocfs_read_req_history_max,
1112 {.name = "threads_min",
1113 .read_fptr = ptlrpc_lprocfs_rd_threads_min,
1114 .write_fptr = ptlrpc_lprocfs_wr_threads_min,
1116 {.name = "threads_max",
1117 .read_fptr = ptlrpc_lprocfs_rd_threads_max,
1118 .write_fptr = ptlrpc_lprocfs_wr_threads_max,
1120 {.name = "threads_started",
1121 .read_fptr = ptlrpc_lprocfs_rd_threads_started,
1123 {.name = "timeouts",
1124 .read_fptr = ptlrpc_lprocfs_rd_timeouts,
1126 {.name = "nrs_policies",
1127 .read_fptr = ptlrpc_lprocfs_rd_nrs,
1128 .write_fptr = ptlrpc_lprocfs_wr_nrs,
1132 static struct file_operations req_history_fops = {
1133 .owner = THIS_MODULE,
1134 .open = ptlrpc_lprocfs_svc_req_history_open,
1136 .llseek = seq_lseek,
1137 .release = lprocfs_seq_release,
1142 ptlrpc_lprocfs_register(entry, svc->srv_name,
1143 "stats", &svc->srv_procroot,
1146 if (svc->srv_procroot == NULL)
1149 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1151 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1152 0400, &req_history_fops, svc);
1154 CWARN("Error adding the req_history file\n");
1157 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1159 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1160 &obddev->obd_svc_procroot,
1161 &obddev->obd_svc_stats);
1163 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1165 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1167 struct lprocfs_stats *svc_stats;
1168 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1169 int opc = opcode_offset(op);
1171 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1172 if (svc_stats == NULL || opc <= 0)
1174 LASSERT(opc < LUSTRE_MAX_OPCODES);
1175 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1176 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1179 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1181 struct lprocfs_stats *svc_stats;
1184 if (!req->rq_import)
1186 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1189 idx = lustre_msg_get_opc(req->rq_reqmsg);
1192 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1195 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1198 LASSERTF(0, "unsupported opcode %u\n", idx);
1202 lprocfs_counter_add(svc_stats, idx, bytes);
1205 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1207 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1209 if (svc->srv_procroot != NULL)
1210 lprocfs_remove(&svc->srv_procroot);
1213 lprocfs_free_stats(&svc->srv_stats);
1216 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1218 if (obd->obd_svc_procroot)
1219 lprocfs_remove(&obd->obd_svc_procroot);
1221 if (obd->obd_svc_stats)
1222 lprocfs_free_stats(&obd->obd_svc_stats);
1224 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1227 #define BUFLEN (UUID_MAX + 5)
1229 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
1230 unsigned long count, void *data)
1232 struct obd_device *obd = data;
1236 OBD_ALLOC(kbuf, BUFLEN);
1241 * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
1242 * bytes into kbuf, to ensure that the string is NUL-terminated.
1243 * UUID_MAX should include a trailing NUL already.
1245 if (cfs_copy_from_user(kbuf, buffer,
1246 min_t(unsigned long, BUFLEN - 1, count))) {
1250 tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
1251 /* Kludge code(deadlock situation): the lprocfs lock has been held
1252 * since the client is evicted by writting client's
1253 * uuid/nid to procfs "evict_client" entry. However,
1254 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
1255 * the proc entries under the being destroyed export{}, so I have
1256 * to drop the lock at first here.
1257 * - jay, jxiong@clusterfs.com */
1259 class_incref(obd, __FUNCTION__, cfs_current());
1261 if (strncmp(tmpbuf, "nid:", 4) == 0)
1262 obd_export_evict_by_nid(obd, tmpbuf + 4);
1263 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
1264 obd_export_evict_by_uuid(obd, tmpbuf + 5);
1266 obd_export_evict_by_uuid(obd, tmpbuf);
1268 class_decref(obd, __FUNCTION__, cfs_current());
1272 OBD_FREE(kbuf, BUFLEN);
1275 EXPORT_SYMBOL(lprocfs_wr_evict_client);
1279 int lprocfs_wr_ping(struct file *file, const char *buffer,
1280 unsigned long count, void *data)
1282 struct obd_device *obd = data;
1283 struct ptlrpc_request *req;
1287 LPROCFS_CLIMP_CHECK(obd);
1288 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1289 LPROCFS_CLIMP_EXIT(obd);
1293 req->rq_send_state = LUSTRE_IMP_FULL;
1295 rc = ptlrpc_queue_wait(req);
1297 ptlrpc_req_finished(req);
1302 EXPORT_SYMBOL(lprocfs_wr_ping);
1304 /* Write the connection UUID to this file to attempt to connect to that node.
1305 * The connection UUID is a node's primary NID. For example,
1306 * "echo connection=192.168.0.1@tcp0::instance > .../import".
1308 int lprocfs_wr_import(struct file *file, const char *buffer,
1309 unsigned long count, void *data)
1311 struct obd_device *obd = data;
1312 struct obd_import *imp = obd->u.cli.cl_import;
1317 const char prefix[] = "connection=";
1318 const int prefix_len = sizeof(prefix) - 1;
1320 if (count > CFS_PAGE_SIZE - 1 || count <= prefix_len)
1323 OBD_ALLOC(kbuf, count + 1);
1327 if (cfs_copy_from_user(kbuf, buffer, count))
1328 GOTO(out, count = -EFAULT);
1332 /* only support connection=uuid::instance now */
1333 if (strncmp(prefix, kbuf, prefix_len) != 0)
1334 GOTO(out, count = -EINVAL);
1336 uuid = kbuf + prefix_len;
1337 ptr = strstr(uuid, "::");
1344 ptr += strlen("::");
1345 inst = simple_strtol(ptr, &endptr, 10);
1347 CERROR("config: wrong instance # %s\n", ptr);
1348 } else if (inst != imp->imp_connect_data.ocd_instance) {
1349 CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1350 "target(%u/%u), reconnecting...\n",
1351 imp->imp_obd->obd_name,
1352 imp->imp_connect_data.ocd_instance, inst);
1355 CDEBUG(D_INFO, "IR: %s has already been connecting to "
1357 imp->imp_obd->obd_name, inst);
1362 ptlrpc_recover_import(imp, uuid, 1);
1365 OBD_FREE(kbuf, count + 1);
1368 EXPORT_SYMBOL(lprocfs_wr_import);
1370 int lprocfs_rd_pinger_recov(char *page, char **start, off_t off,
1371 int count, int *eof, void *data)
1373 struct obd_device *obd = data;
1374 struct obd_import *imp = obd->u.cli.cl_import;
1377 LPROCFS_CLIMP_CHECK(obd);
1378 rc = snprintf(page, count, "%d\n", !imp->imp_no_pinger_recover);
1379 LPROCFS_CLIMP_EXIT(obd);
1383 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1385 int lprocfs_wr_pinger_recov(struct file *file, const char *buffer,
1386 unsigned long count, void *data)
1388 struct obd_device *obd = data;
1389 struct client_obd *cli = &obd->u.cli;
1390 struct obd_import *imp = cli->cl_import;
1393 rc = lprocfs_write_helper(buffer, count, &val);
1397 if (val != 0 && val != 1)
1400 LPROCFS_CLIMP_CHECK(obd);
1401 spin_lock(&imp->imp_lock);
1402 imp->imp_no_pinger_recover = !val;
1403 spin_unlock(&imp->imp_lock);
1404 LPROCFS_CLIMP_EXIT(obd);
1409 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);
1411 #endif /* LPROCFS */