1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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 { LDLM_ENQUEUE, "ldlm_enqueue" },
97 { LDLM_CONVERT, "ldlm_convert" },
98 { LDLM_CANCEL, "ldlm_cancel" },
99 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
100 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
101 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
102 { MGS_CONNECT, "mgs_connect" },
103 { MGS_DISCONNECT, "mgs_disconnect" },
104 { MGS_EXCEPTION, "mgs_exception" },
105 { MGS_TARGET_REG, "mgs_target_reg" },
106 { MGS_TARGET_DEL, "mgs_target_del" },
107 { MGS_SET_INFO, "mgs_set_info" },
108 { OBD_PING, "obd_ping" },
109 { OBD_LOG_CANCEL, "llog_origin_handle_cancel" },
110 { OBD_QC_CALLBACK, "obd_quota_callback" },
111 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
112 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block"},
113 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
114 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
115 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
116 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
117 { LLOG_CATINFO, "llog_catinfo" },
118 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
119 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
120 { QUOTA_DQACQ, "quota_acquire" },
121 { QUOTA_DQREL, "quota_release" },
122 { SEQ_QUERY, "seq_query" },
128 } ll_eopcode_table[EXTRA_LAST_OPC] = {
129 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
130 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
131 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
132 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
133 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
134 { MDS_REINT_SETATTR, "mds_reint_setattr" },
135 { MDS_REINT_CREATE, "mds_reint_create" },
136 { MDS_REINT_LINK, "mds_reint_link" },
137 { MDS_REINT_UNLINK, "mds_reint_unlink" },
138 { MDS_REINT_RENAME, "mds_reint_rename" },
139 { MDS_REINT_OPEN, "mds_reint_open" },
140 { BRW_READ_BYTES, "read_bytes" },
141 { BRW_WRITE_BYTES, "write_bytes" },
144 const char *ll_opcode2str(__u32 opcode)
146 /* When one of the assertions below fail, chances are that:
147 * 1) A new opcode was added in include/lustre/lustre_idl.h,
148 * but is missing from the table above.
149 * or 2) The opcode space was renumbered or rearranged,
150 * and the opcode_offset() function in
151 * ptlrpc_internal.h needs to be modified.
153 __u32 offset = opcode_offset(opcode);
154 LASSERTF(offset < LUSTRE_MAX_OPCODES,
155 "offset %u >= LUSTRE_MAX_OPCODES %u\n",
156 offset, LUSTRE_MAX_OPCODES);
157 LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
158 "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
159 offset, ll_rpc_opcode_table[offset].opcode, opcode);
160 return ll_rpc_opcode_table[offset].opname;
163 const char* ll_eopcode2str(__u32 opcode)
165 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
166 return ll_eopcode_table[opcode].opname;
170 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
171 char *name, struct proc_dir_entry **procroot_ret,
172 struct lprocfs_stats **stats_ret)
174 struct proc_dir_entry *svc_procroot;
175 struct lprocfs_stats *svc_stats;
177 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
180 LASSERT(*procroot_ret == NULL);
181 LASSERT(*stats_ret == NULL);
183 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES, 0);
184 if (svc_stats == NULL)
188 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
189 if (IS_ERR(svc_procroot)) {
190 lprocfs_free_stats(&svc_stats);
197 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
198 svc_counter_config, "req_waittime", "usec");
199 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
200 svc_counter_config, "req_qdepth", "reqs");
201 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
202 svc_counter_config, "req_active", "reqs");
203 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
204 svc_counter_config, "req_timeout", "sec");
205 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
206 svc_counter_config, "reqbuf_avail", "bufs");
207 for (i = 0; i < EXTRA_LAST_OPC; i++) {
211 case BRW_WRITE_BYTES:
219 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
221 ll_eopcode2str(i), units);
223 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
224 __u32 opcode = ll_rpc_opcode_table[i].opcode;
225 lprocfs_counter_init(svc_stats,
226 EXTRA_MAX_OPCODES + i, svc_counter_config,
227 ll_opcode2str(opcode), "usec");
230 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
233 lprocfs_remove(&svc_procroot);
234 lprocfs_free_stats(&svc_stats);
237 *procroot_ret = svc_procroot;
238 *stats_ret = svc_stats;
243 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
244 int count, int *eof, void *data)
246 struct ptlrpc_service *svc = data;
249 return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
253 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
254 int count, int *eof, void *data)
256 struct ptlrpc_service *svc = data;
259 return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
263 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
264 unsigned long count, void *data)
266 struct ptlrpc_service *svc = data;
269 int rc = lprocfs_write_helper(buffer, count, &val);
277 /* This sanity check is more of an insanity check; we can still
278 * hose a kernel by allowing the request history to grow too
280 bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
281 if (val > num_physpages/(2*bufpages))
284 spin_lock(&svc->srv_lock);
285 svc->srv_max_history_rqbds = val;
286 spin_unlock(&svc->srv_lock);
291 struct ptlrpc_srh_iterator {
293 struct ptlrpc_request *srhi_req;
297 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
298 struct ptlrpc_srh_iterator *srhi,
302 struct ptlrpc_request *req;
304 if (srhi->srhi_req != NULL &&
305 srhi->srhi_seq > svc->srv_request_max_cull_seq &&
306 srhi->srhi_seq <= seq) {
307 /* If srhi_req was set previously, hasn't been culled and
308 * we're searching for a seq on or after it (i.e. more
309 * recent), search from it onwards.
310 * Since the service history is LRU (i.e. culled reqs will
311 * be near the head), we shouldn't have to do long
313 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
314 LASSERT (!list_empty(&svc->srv_request_history));
315 e = &srhi->srhi_req->rq_history_list;
317 /* search from start */
318 e = svc->srv_request_history.next;
321 while (e != &svc->srv_request_history) {
322 req = list_entry(e, struct ptlrpc_request, rq_history_list);
324 if (req->rq_history_seq >= seq) {
325 srhi->srhi_seq = req->rq_history_seq;
326 srhi->srhi_req = req;
336 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
338 struct ptlrpc_service *svc = s->private;
339 struct ptlrpc_srh_iterator *srhi;
342 OBD_ALLOC(srhi, sizeof(*srhi));
347 srhi->srhi_req = NULL;
349 spin_lock(&svc->srv_lock);
350 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos);
351 spin_unlock(&svc->srv_lock);
354 *pos = srhi->srhi_seq;
358 OBD_FREE(srhi, sizeof(*srhi));
363 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
365 struct ptlrpc_srh_iterator *srhi = iter;
368 OBD_FREE(srhi, sizeof(*srhi));
372 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
373 void *iter, loff_t *pos)
375 struct ptlrpc_service *svc = s->private;
376 struct ptlrpc_srh_iterator *srhi = iter;
379 spin_lock(&svc->srv_lock);
380 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos + 1);
381 spin_unlock(&svc->srv_lock);
384 OBD_FREE(srhi, sizeof(*srhi));
388 *pos = srhi->srhi_seq;
392 /* common ost/mdt srv_request_history_print_fn */
393 void target_print_req(void *seq_file, struct ptlrpc_request *req)
395 /* Called holding srv_lock with irqs disabled.
396 * Print specific req contents and a newline.
397 * CAVEAT EMPTOR: check request message length before printing!!!
398 * You might have received any old crap so you must be just as
399 * careful here as the service's request parser!!! */
400 struct seq_file *sf = seq_file;
402 switch (req->rq_phase) {
404 /* still awaiting a service thread's attention, or rejected
405 * because the generic request message didn't unpack */
406 seq_printf(sf, "<not swabbed>\n");
408 case RQ_PHASE_INTERPRET:
409 /* being handled, so basic msg swabbed, and opc is valid
410 * but racing with mds_handle() */
411 case RQ_PHASE_COMPLETE:
412 /* been handled by mds_handle() reply state possibly still
414 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
417 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
420 EXPORT_SYMBOL(target_print_req);
422 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
424 struct ptlrpc_service *svc = s->private;
425 struct ptlrpc_srh_iterator *srhi = iter;
426 struct ptlrpc_request *req;
429 spin_lock(&svc->srv_lock);
431 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, srhi->srhi_seq);
434 req = srhi->srhi_req;
436 /* Print common req fields.
437 * CAVEAT EMPTOR: we're racing with the service handler
438 * here. The request could contain any old crap, so you
439 * must be just as careful as the service's request
440 * parser. Currently I only print stuff here I know is OK
441 * to look at coz it was set up in request_in_callback()!!! */
442 seq_printf(s, LPD64":%s:%s:x"LPD64":%d:%s:%ld:%lds(%+lds) ",
443 req->rq_history_seq, libcfs_nid2str(req->rq_self),
444 libcfs_id2str(req->rq_peer), req->rq_xid,
445 req->rq_reqlen, ptlrpc_rqphase2str(req),
446 req->rq_arrival_time.tv_sec,
447 req->rq_sent - req->rq_arrival_time.tv_sec,
448 req->rq_sent - req->rq_deadline);
449 if (svc->srv_request_history_print_fn == NULL)
452 svc->srv_request_history_print_fn(s, srhi->srhi_req);
455 spin_unlock(&svc->srv_lock);
461 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
463 static struct seq_operations sops = {
464 .start = ptlrpc_lprocfs_svc_req_history_start,
465 .stop = ptlrpc_lprocfs_svc_req_history_stop,
466 .next = ptlrpc_lprocfs_svc_req_history_next,
467 .show = ptlrpc_lprocfs_svc_req_history_show,
469 struct proc_dir_entry *dp = PDE(inode);
470 struct seq_file *seqf;
473 LPROCFS_ENTRY_AND_CHECK(dp);
474 rc = seq_open(file, &sops);
480 seqf = file->private_data;
481 seqf->private = dp->data;
485 /* See also lprocfs_rd_timeouts */
486 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
487 int count, int *eof, void *data)
489 struct ptlrpc_service *svc = data;
490 unsigned int cur, worst;
496 cur = at_get(&svc->srv_at_estimate);
497 worst = svc->srv_at_estimate.at_worst_ever;
498 worstt = svc->srv_at_estimate.at_worst_time;
499 s2dhms(&ts, cfs_time_current_sec() - worstt);
501 rc += snprintf(page + rc, count - rc,
502 "adaptive timeouts off, using obd_timeout %u\n",
504 rc += snprintf(page + rc, count - rc,
505 "%10s : cur %3u worst %3u (at %ld, "DHMS_FMT" ago) ",
506 "service", cur, worst, worstt,
508 rc = lprocfs_at_hist_helper(page, count, rc,
509 &svc->srv_at_estimate);
513 static int ptlrpc_lprocfs_rd_hp_ratio(char *page, char **start, off_t off,
514 int count, int *eof, void *data)
516 struct ptlrpc_service *svc = data;
517 int rc = snprintf(page, count, "%d", svc->srv_hpreq_ratio);
521 static int ptlrpc_lprocfs_wr_hp_ratio(struct file *file, const char *buffer,
522 unsigned long count, void *data)
524 struct ptlrpc_service *svc = data;
527 rc = lprocfs_write_helper(buffer, count, &val);
533 spin_lock(&svc->srv_lock);
534 svc->srv_hpreq_ratio = val;
535 spin_unlock(&svc->srv_lock);
539 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
540 struct ptlrpc_service *svc)
542 struct lprocfs_vars lproc_vars[] = {
543 {.name = "req_buffer_history_len",
545 .read_fptr = ptlrpc_lprocfs_read_req_history_len,
547 {.name = "req_buffer_history_max",
548 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
549 .read_fptr = ptlrpc_lprocfs_read_req_history_max,
552 .read_fptr = ptlrpc_lprocfs_rd_timeouts,
554 {.name = "high_priority_ratio",
555 .read_fptr = ptlrpc_lprocfs_rd_hp_ratio,
556 .write_fptr = ptlrpc_lprocfs_wr_hp_ratio,
560 static struct file_operations req_history_fops = {
561 .owner = THIS_MODULE,
562 .open = ptlrpc_lprocfs_svc_req_history_open,
565 .release = lprocfs_seq_release,
567 struct proc_dir_entry *req_history;
569 ptlrpc_lprocfs_register(entry, svc->srv_name,
570 "stats", &svc->srv_procroot,
573 if (svc->srv_procroot == NULL)
576 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
578 req_history = create_proc_entry("req_history", 0400,
580 if (req_history != NULL) {
581 req_history->data = svc;
582 req_history->proc_fops = &req_history_fops;
586 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
588 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
589 &obddev->obd_svc_procroot,
590 &obddev->obd_svc_stats);
592 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
594 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req)
596 struct lprocfs_stats *svc_stats;
597 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
598 int opc = opcode_offset(op);
600 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
601 if (svc_stats == NULL || opc <= 0)
603 LASSERT(opc < LUSTRE_MAX_OPCODES);
604 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
605 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, 0);
608 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
610 struct lprocfs_stats *svc_stats;
615 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
618 idx = lustre_msg_get_opc(req->rq_reqmsg);
621 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
624 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
627 LASSERTF(0, "unsupported opcode %u\n", idx);
631 lprocfs_counter_add(svc_stats, idx, bytes);
634 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
636 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
638 if (svc->srv_procroot != NULL)
639 lprocfs_remove(&svc->srv_procroot);
641 lprocfs_free_stats(&svc->srv_stats);
644 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
646 if (obd->obd_svc_procroot)
647 lprocfs_remove(&obd->obd_svc_procroot);
648 if (obd->obd_svc_stats)
649 lprocfs_free_stats(&obd->obd_svc_stats);
651 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
654 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
655 unsigned long count, void *data)
657 struct obd_device *obd = data;
658 char tmpbuf[sizeof(struct obd_uuid)];
660 /* Kludge code(deadlock situation): the lprocfs lock has been held
661 * since the client is evicted by writting client's
662 * uuid/nid to procfs "evict_client" entry. However,
663 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
664 * the proc entries under the being destroyed export{}, so I have
665 * to drop the lock at first here.
666 * - jay, jxiong@clusterfs.com */
670 sscanf(buffer, "%40s", tmpbuf);
671 if (strncmp(tmpbuf, "nid:", 4) == 0)
672 obd_export_evict_by_nid(obd, tmpbuf + 4);
673 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
674 obd_export_evict_by_uuid(obd, tmpbuf + 5);
676 obd_export_evict_by_uuid(obd, tmpbuf);
683 EXPORT_SYMBOL(lprocfs_wr_evict_client);
685 int lprocfs_wr_ping(struct file *file, const char *buffer,
686 unsigned long count, void *data)
688 struct obd_device *obd = data;
689 struct ptlrpc_request *req;
693 LPROCFS_CLIMP_CHECK(obd);
694 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
695 LPROCFS_CLIMP_EXIT(obd);
699 req->rq_send_state = LUSTRE_IMP_FULL;
701 rc = ptlrpc_queue_wait(req);
703 ptlrpc_req_finished(req);
708 EXPORT_SYMBOL(lprocfs_wr_ping);