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 { MDS_GETATTR, "mds_getattr" },
76 { MDS_GETATTR_NAME, "mds_getattr_lock" },
77 { MDS_CLOSE, "mds_close" },
78 { MDS_REINT, "mds_reint" },
79 { MDS_READPAGE, "mds_readpage" },
80 { MDS_CONNECT, "mds_connect" },
81 { MDS_DISCONNECT, "mds_disconnect" },
82 { MDS_GETSTATUS, "mds_getstatus" },
83 { MDS_STATFS, "mds_statfs" },
84 { MDS_PIN, "mds_pin" },
85 { MDS_UNPIN, "mds_unpin" },
86 { MDS_SYNC, "mds_sync" },
87 { MDS_DONE_WRITING, "mds_done_writing" },
88 { MDS_SET_INFO, "mds_set_info" },
89 { MDS_QUOTACHECK, "mds_quotacheck" },
90 { MDS_QUOTACTL, "mds_quotactl" },
91 { MDS_GETXATTR, "mds_getxattr" },
92 { MDS_SETXATTR, "mds_setxattr" },
93 { MDS_WRITEPAGE, "mds_writepage" },
94 { MDS_IS_SUBDIR, "mds_is_subdir" },
95 { LDLM_ENQUEUE, "ldlm_enqueue" },
96 { LDLM_CONVERT, "ldlm_convert" },
97 { LDLM_CANCEL, "ldlm_cancel" },
98 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
99 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
100 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
101 { MGS_CONNECT, "mgs_connect" },
102 { MGS_DISCONNECT, "mgs_disconnect" },
103 { MGS_EXCEPTION, "mgs_exception" },
104 { MGS_TARGET_REG, "mgs_target_reg" },
105 { MGS_TARGET_DEL, "mgs_target_del" },
106 { MGS_SET_INFO, "mgs_set_info" },
107 { OBD_PING, "obd_ping" },
108 { OBD_LOG_CANCEL, "llog_origin_handle_cancel" },
109 { OBD_QC_CALLBACK, "obd_qc_callback" },
110 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
111 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
112 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
113 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
114 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
115 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
116 { LLOG_CATINFO, "llog_catinfo" },
117 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
118 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
119 { FLD_QUERY, "fld_query" },
120 { SEQ_QUERY, "seq_query" },
121 { SEC_CTX_INIT, "sec_ctx_init" },
122 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
123 { SEC_CTX_FINI, "sec_ctx_fini" }
129 } ll_eopcode_table[EXTRA_LAST_OPC] = {
130 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
131 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
132 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
133 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
134 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
135 { MDS_REINT_CREATE, "mds_reint_create" },
136 { MDS_REINT_LINK, "mds_reint_link" },
137 { MDS_REINT_OPEN, "mds_reint_open" },
138 { MDS_REINT_SETATTR, "mds_reint_setattr" },
139 { MDS_REINT_RENAME, "mds_reint_rename" },
140 { MDS_REINT_UNLINK, "mds_reint_unlink" },
141 { BRW_READ_BYTES, "read_bytes" },
142 { BRW_WRITE_BYTES, "write_bytes" },
145 const char *ll_opcode2str(__u32 opcode)
147 /* When one of the assertions below fail, chances are that:
148 * 1) A new opcode was added in lustre_idl.h, but was
149 * is missing from the table above.
150 * or 2) The opcode space was renumbered or rearranged,
151 * and the opcode_offset() function in
152 * ptlrpc_internal.h needs to be modified.
154 __u32 offset = opcode_offset(opcode);
155 LASSERT(offset < LUSTRE_MAX_OPCODES);
156 LASSERT(ll_rpc_opcode_table[offset].opcode == opcode);
157 return ll_rpc_opcode_table[offset].opname;
160 const char* ll_eopcode2str(__u32 opcode)
162 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
163 return ll_eopcode_table[opcode].opname;
166 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
167 char *name, struct proc_dir_entry **procroot_ret,
168 struct lprocfs_stats **stats_ret)
170 struct proc_dir_entry *svc_procroot;
171 struct lprocfs_stats *svc_stats;
173 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
176 LASSERT(*procroot_ret == NULL);
177 LASSERT(*stats_ret == NULL);
179 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES, 0);
180 if (svc_stats == NULL)
184 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
185 if (IS_ERR(svc_procroot)) {
186 lprocfs_free_stats(&svc_stats);
193 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
194 svc_counter_config, "req_waittime", "usec");
195 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
196 svc_counter_config, "req_qdepth", "reqs");
197 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
198 svc_counter_config, "req_active", "reqs");
199 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
200 svc_counter_config, "req_timeout", "sec");
201 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
202 svc_counter_config, "reqbuf_avail", "bufs");
203 for (i = 0; i < EXTRA_LAST_OPC; i++) {
207 case BRW_WRITE_BYTES:
215 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
217 ll_eopcode2str(i), units);
219 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
220 __u32 opcode = ll_rpc_opcode_table[i].opcode;
221 lprocfs_counter_init(svc_stats,
222 EXTRA_MAX_OPCODES + i, svc_counter_config,
223 ll_opcode2str(opcode), "usec");
226 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
229 lprocfs_remove(&svc_procroot);
230 lprocfs_free_stats(&svc_stats);
233 *procroot_ret = svc_procroot;
234 *stats_ret = svc_stats;
239 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
240 int count, int *eof, void *data)
242 struct ptlrpc_service *svc = data;
245 return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
249 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
250 int count, int *eof, void *data)
252 struct ptlrpc_service *svc = data;
255 return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
259 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
260 unsigned long count, void *data)
262 struct ptlrpc_service *svc = data;
265 int rc = lprocfs_write_helper(buffer, count, &val);
273 /* This sanity check is more of an insanity check; we can still
274 * hose a kernel by allowing the request history to grow too
276 bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
277 if (val > num_physpages/(2 * bufpages))
280 spin_lock(&svc->srv_lock);
281 svc->srv_max_history_rqbds = val;
282 spin_unlock(&svc->srv_lock);
287 struct ptlrpc_srh_iterator {
289 struct ptlrpc_request *srhi_req;
293 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
294 struct ptlrpc_srh_iterator *srhi,
298 struct ptlrpc_request *req;
300 if (srhi->srhi_req != NULL &&
301 srhi->srhi_seq > svc->srv_request_max_cull_seq &&
302 srhi->srhi_seq <= seq) {
303 /* If srhi_req was set previously, hasn't been culled and
304 * we're searching for a seq on or after it (i.e. more
305 * recent), search from it onwards.
306 * Since the service history is LRU (i.e. culled reqs will
307 * be near the head), we shouldn't have to do long
309 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
310 LASSERT (!list_empty(&svc->srv_request_history));
311 e = &srhi->srhi_req->rq_history_list;
313 /* search from start */
314 e = svc->srv_request_history.next;
317 while (e != &svc->srv_request_history) {
318 req = list_entry(e, struct ptlrpc_request, rq_history_list);
320 if (req->rq_history_seq >= seq) {
321 srhi->srhi_seq = req->rq_history_seq;
322 srhi->srhi_req = req;
332 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
334 struct ptlrpc_service *svc = s->private;
335 struct ptlrpc_srh_iterator *srhi;
338 OBD_ALLOC(srhi, sizeof(*srhi));
343 srhi->srhi_req = NULL;
345 spin_lock(&svc->srv_lock);
346 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos);
347 spin_unlock(&svc->srv_lock);
350 *pos = srhi->srhi_seq;
354 OBD_FREE(srhi, sizeof(*srhi));
359 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
361 struct ptlrpc_srh_iterator *srhi = iter;
364 OBD_FREE(srhi, sizeof(*srhi));
368 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
369 void *iter, loff_t *pos)
371 struct ptlrpc_service *svc = s->private;
372 struct ptlrpc_srh_iterator *srhi = iter;
375 spin_lock(&svc->srv_lock);
376 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos + 1);
377 spin_unlock(&svc->srv_lock);
380 OBD_FREE(srhi, sizeof(*srhi));
384 *pos = srhi->srhi_seq;
388 /* common ost/mdt srv_request_history_print_fn */
389 void target_print_req(void *seq_file, struct ptlrpc_request *req)
391 /* Called holding srv_lock with irqs disabled.
392 * Print specific req contents and a newline.
393 * CAVEAT EMPTOR: check request message length before printing!!!
394 * You might have received any old crap so you must be just as
395 * careful here as the service's request parser!!! */
396 struct seq_file *sf = seq_file;
398 switch (req->rq_phase) {
400 /* still awaiting a service thread's attention, or rejected
401 * because the generic request message didn't unpack */
402 seq_printf(sf, "<not swabbed>\n");
404 case RQ_PHASE_INTERPRET:
405 /* being handled, so basic msg swabbed, and opc is valid
406 * but racing with mds_handle() */
407 case RQ_PHASE_COMPLETE:
408 /* been handled by mds_handle() reply state possibly still
410 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
413 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
416 EXPORT_SYMBOL(target_print_req);
418 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
420 struct ptlrpc_service *svc = s->private;
421 struct ptlrpc_srh_iterator *srhi = iter;
422 struct ptlrpc_request *req;
425 spin_lock(&svc->srv_lock);
427 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, srhi->srhi_seq);
430 req = srhi->srhi_req;
432 /* Print common req fields.
433 * CAVEAT EMPTOR: we're racing with the service handler
434 * here. The request could contain any old crap, so you
435 * must be just as careful as the service's request
436 * parser. Currently I only print stuff here I know is OK
437 * to look at coz it was set up in request_in_callback()!!! */
438 seq_printf(s, LPD64":%s:%s:x"LPD64":%d:%s:%ld:%lds(%+lds) ",
439 req->rq_history_seq, libcfs_nid2str(req->rq_self),
440 libcfs_id2str(req->rq_peer), req->rq_xid,
441 req->rq_reqlen, ptlrpc_rqphase2str(req),
442 req->rq_arrival_time.tv_sec,
443 req->rq_sent - req->rq_arrival_time.tv_sec,
444 req->rq_sent - req->rq_deadline);
445 if (svc->srv_request_history_print_fn == NULL)
448 svc->srv_request_history_print_fn(s, srhi->srhi_req);
451 spin_unlock(&svc->srv_lock);
457 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
459 static struct seq_operations sops = {
460 .start = ptlrpc_lprocfs_svc_req_history_start,
461 .stop = ptlrpc_lprocfs_svc_req_history_stop,
462 .next = ptlrpc_lprocfs_svc_req_history_next,
463 .show = ptlrpc_lprocfs_svc_req_history_show,
465 struct proc_dir_entry *dp = PDE(inode);
466 struct seq_file *seqf;
469 LPROCFS_ENTRY_AND_CHECK(dp);
470 rc = seq_open(file, &sops);
476 seqf = file->private_data;
477 seqf->private = dp->data;
481 /* See also lprocfs_rd_timeouts */
482 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
483 int count, int *eof, void *data)
485 struct ptlrpc_service *svc = data;
486 unsigned int cur, worst;
492 cur = at_get(&svc->srv_at_estimate);
493 worst = svc->srv_at_estimate.at_worst_ever;
494 worstt = svc->srv_at_estimate.at_worst_time;
495 s2dhms(&ts, cfs_time_current_sec() - worstt);
497 rc += snprintf(page + rc, count - rc,
498 "adaptive timeouts off, using obd_timeout %u\n",
500 rc += snprintf(page + rc, count - rc,
501 "%10s : cur %3u worst %3u (at %ld, "DHMS_FMT" ago) ",
502 "service", cur, worst, worstt,
504 rc = lprocfs_at_hist_helper(page, count, rc,
505 &svc->srv_at_estimate);
509 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
510 struct ptlrpc_service *svc)
512 struct lprocfs_vars lproc_vars[] = {
513 {.name = "req_buffer_history_len",
515 .read_fptr = ptlrpc_lprocfs_read_req_history_len,
517 {.name = "req_buffer_history_max",
518 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
519 .read_fptr = ptlrpc_lprocfs_read_req_history_max,
522 .read_fptr = ptlrpc_lprocfs_rd_timeouts,
526 static struct file_operations req_history_fops = {
527 .owner = THIS_MODULE,
528 .open = ptlrpc_lprocfs_svc_req_history_open,
531 .release = lprocfs_seq_release,
536 ptlrpc_lprocfs_register(entry, svc->srv_name,
537 "stats", &svc->srv_procroot,
540 if (svc->srv_procroot == NULL)
543 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
545 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
546 0400, &req_history_fops, svc);
548 CWARN("Error adding the req_history file\n");
551 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
553 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
554 &obddev->obd_svc_procroot,
555 &obddev->obd_svc_stats);
557 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
559 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req)
561 struct lprocfs_stats *svc_stats;
562 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
563 int opc = opcode_offset(op);
565 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
566 if (svc_stats == NULL || opc <= 0)
568 LASSERT(opc < LUSTRE_MAX_OPCODES);
569 if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
570 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, 0);
573 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
575 struct lprocfs_stats *svc_stats;
580 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
583 idx = lustre_msg_get_opc(req->rq_reqmsg);
586 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
589 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
592 LASSERTF(0, "unsupported opcode %u\n", idx);
596 lprocfs_counter_add(svc_stats, idx, bytes);
599 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
601 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
603 if (svc->srv_procroot != NULL)
604 lprocfs_remove(&svc->srv_procroot);
607 lprocfs_free_stats(&svc->srv_stats);
610 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
612 if (obd->obd_svc_procroot)
613 lprocfs_remove(&obd->obd_svc_procroot);
615 if (obd->obd_svc_stats)
616 lprocfs_free_stats(&obd->obd_svc_stats);
618 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
621 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
622 unsigned long count, void *data)
624 struct obd_device *obd = data;
625 char tmpbuf[sizeof(struct obd_uuid)];
627 /* Kludge code(deadlock situation): the lprocfs lock has been held
628 * since the client is evicted by writting client's
629 * uuid/nid to procfs "evict_client" entry. However,
630 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
631 * the proc entries under the being destroyed export{}, so I have
632 * to drop the lock at first here.
633 * - jay, jxiong@clusterfs.com */
637 sscanf(buffer, "%40s", tmpbuf);
638 if (strncmp(tmpbuf, "nid:", 4) == 0)
639 obd_export_evict_by_nid(obd, tmpbuf + 4);
640 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
641 obd_export_evict_by_uuid(obd, tmpbuf + 5);
643 obd_export_evict_by_uuid(obd, tmpbuf);
650 EXPORT_SYMBOL(lprocfs_wr_evict_client);
652 int lprocfs_wr_ping(struct file *file, const char *buffer,
653 unsigned long count, void *data)
655 struct obd_device *obd = data;
656 struct ptlrpc_request *req;
660 LPROCFS_CLIMP_CHECK(obd);
661 req = ptlrpc_request_alloc_pack(obd->u.cli.cl_import, &RQF_OBD_PING,
662 LUSTRE_OBD_VERSION, OBD_PING);
664 LPROCFS_CLIMP_EXIT(obd);
668 ptlrpc_request_set_replen(req);
669 req->rq_send_state = LUSTRE_IMP_FULL;
670 req->rq_no_resend = 1;
671 req->rq_no_delay = 1;
673 rc = ptlrpc_queue_wait(req);
675 ptlrpc_req_finished(req);
680 EXPORT_SYMBOL(lprocfs_wr_ping);