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 [sun.com URL with a
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" }
143 const char *ll_opcode2str(__u32 opcode)
145 /* When one of the assertions below fail, chances are that:
146 * 1) A new opcode was added in lustre_idl.h, but was
147 * is missing from the table above.
148 * or 2) The opcode space was renumbered or rearranged,
149 * and the opcode_offset() function in
150 * ptlrpc_internal.h needs to be modified.
152 __u32 offset = opcode_offset(opcode);
153 LASSERT(offset < LUSTRE_MAX_OPCODES);
154 LASSERT(ll_rpc_opcode_table[offset].opcode == opcode);
155 return ll_rpc_opcode_table[offset].opname;
158 const char* ll_eopcode2str(__u32 opcode)
160 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
161 return ll_eopcode_table[opcode].opname;
164 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
165 char *name, struct proc_dir_entry **procroot_ret,
166 struct lprocfs_stats **stats_ret)
168 struct proc_dir_entry *svc_procroot;
169 struct lprocfs_stats *svc_stats;
171 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
174 LASSERT(*procroot_ret == NULL);
175 LASSERT(*stats_ret == NULL);
177 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES, 0);
178 if (svc_stats == NULL)
182 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
183 if (IS_ERR(svc_procroot)) {
184 lprocfs_free_stats(&svc_stats);
191 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
192 svc_counter_config, "req_waittime", "usec");
193 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
194 svc_counter_config, "req_qdepth", "reqs");
195 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
196 svc_counter_config, "req_active", "reqs");
197 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
198 svc_counter_config, "req_timeout", "sec");
199 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
200 svc_counter_config, "reqbuf_avail", "bufs");
201 for (i = 0; i < EXTRA_LAST_OPC; i++) {
202 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
204 ll_eopcode2str(i), "reqs");
206 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
207 __u32 opcode = ll_rpc_opcode_table[i].opcode;
208 lprocfs_counter_init(svc_stats,
209 EXTRA_MAX_OPCODES + i, svc_counter_config,
210 ll_opcode2str(opcode), "usec");
213 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
216 lprocfs_remove(&svc_procroot);
217 lprocfs_free_stats(&svc_stats);
220 *procroot_ret = svc_procroot;
221 *stats_ret = svc_stats;
226 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
227 int count, int *eof, void *data)
229 struct ptlrpc_service *svc = data;
232 return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
236 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
237 int count, int *eof, void *data)
239 struct ptlrpc_service *svc = data;
242 return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
246 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
247 unsigned long count, void *data)
249 struct ptlrpc_service *svc = data;
252 int rc = lprocfs_write_helper(buffer, count, &val);
260 /* This sanity check is more of an insanity check; we can still
261 * hose a kernel by allowing the request history to grow too
263 bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
264 if (val > num_physpages/(2 * bufpages))
267 spin_lock(&svc->srv_lock);
268 svc->srv_max_history_rqbds = val;
269 spin_unlock(&svc->srv_lock);
274 struct ptlrpc_srh_iterator {
276 struct ptlrpc_request *srhi_req;
280 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
281 struct ptlrpc_srh_iterator *srhi,
285 struct ptlrpc_request *req;
287 if (srhi->srhi_req != NULL &&
288 srhi->srhi_seq > svc->srv_request_max_cull_seq &&
289 srhi->srhi_seq <= seq) {
290 /* If srhi_req was set previously, hasn't been culled and
291 * we're searching for a seq on or after it (i.e. more
292 * recent), search from it onwards.
293 * Since the service history is LRU (i.e. culled reqs will
294 * be near the head), we shouldn't have to do long
296 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
297 LASSERT (!list_empty(&svc->srv_request_history));
298 e = &srhi->srhi_req->rq_history_list;
300 /* search from start */
301 e = svc->srv_request_history.next;
304 while (e != &svc->srv_request_history) {
305 req = list_entry(e, struct ptlrpc_request, rq_history_list);
307 if (req->rq_history_seq >= seq) {
308 srhi->srhi_seq = req->rq_history_seq;
309 srhi->srhi_req = req;
319 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
321 struct ptlrpc_service *svc = s->private;
322 struct ptlrpc_srh_iterator *srhi;
325 OBD_ALLOC(srhi, sizeof(*srhi));
330 srhi->srhi_req = NULL;
332 spin_lock(&svc->srv_lock);
333 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos);
334 spin_unlock(&svc->srv_lock);
337 *pos = srhi->srhi_seq;
341 OBD_FREE(srhi, sizeof(*srhi));
346 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
348 struct ptlrpc_srh_iterator *srhi = iter;
351 OBD_FREE(srhi, sizeof(*srhi));
355 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
356 void *iter, loff_t *pos)
358 struct ptlrpc_service *svc = s->private;
359 struct ptlrpc_srh_iterator *srhi = iter;
362 spin_lock(&svc->srv_lock);
363 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos + 1);
364 spin_unlock(&svc->srv_lock);
367 OBD_FREE(srhi, sizeof(*srhi));
371 *pos = srhi->srhi_seq;
375 /* common ost/mdt srv_request_history_print_fn */
376 void target_print_req(void *seq_file, struct ptlrpc_request *req)
378 /* Called holding srv_lock with irqs disabled.
379 * Print specific req contents and a newline.
380 * CAVEAT EMPTOR: check request message length before printing!!!
381 * You might have received any old crap so you must be just as
382 * careful here as the service's request parser!!! */
383 struct seq_file *sf = seq_file;
385 switch (req->rq_phase) {
387 /* still awaiting a service thread's attention, or rejected
388 * because the generic request message didn't unpack */
389 seq_printf(sf, "<not swabbed>\n");
391 case RQ_PHASE_INTERPRET:
392 /* being handled, so basic msg swabbed, and opc is valid
393 * but racing with mds_handle() */
394 case RQ_PHASE_COMPLETE:
395 /* been handled by mds_handle() reply state possibly still
397 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
400 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
403 EXPORT_SYMBOL(target_print_req);
405 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
407 struct ptlrpc_service *svc = s->private;
408 struct ptlrpc_srh_iterator *srhi = iter;
409 struct ptlrpc_request *req;
412 spin_lock(&svc->srv_lock);
414 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, srhi->srhi_seq);
417 req = srhi->srhi_req;
419 /* Print common req fields.
420 * CAVEAT EMPTOR: we're racing with the service handler
421 * here. The request could contain any old crap, so you
422 * must be just as careful as the service's request
423 * parser. Currently I only print stuff here I know is OK
424 * to look at coz it was set up in request_in_callback()!!! */
425 seq_printf(s, LPD64":%s:%s:x"LPD64":%d:%s:%ld:%lds(%+lds) ",
426 req->rq_history_seq, libcfs_nid2str(req->rq_self),
427 libcfs_id2str(req->rq_peer), req->rq_xid,
428 req->rq_reqlen, ptlrpc_rqphase2str(req),
429 req->rq_arrival_time.tv_sec,
430 req->rq_sent - req->rq_arrival_time.tv_sec,
431 req->rq_sent - req->rq_deadline);
432 if (svc->srv_request_history_print_fn == NULL)
435 svc->srv_request_history_print_fn(s, srhi->srhi_req);
438 spin_unlock(&svc->srv_lock);
444 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
446 static struct seq_operations sops = {
447 .start = ptlrpc_lprocfs_svc_req_history_start,
448 .stop = ptlrpc_lprocfs_svc_req_history_stop,
449 .next = ptlrpc_lprocfs_svc_req_history_next,
450 .show = ptlrpc_lprocfs_svc_req_history_show,
452 struct proc_dir_entry *dp = PDE(inode);
453 struct seq_file *seqf;
456 LPROCFS_ENTRY_AND_CHECK(dp);
457 rc = seq_open(file, &sops);
463 seqf = file->private_data;
464 seqf->private = dp->data;
468 /* See also lprocfs_rd_timeouts */
469 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
470 int count, int *eof, void *data)
472 struct ptlrpc_service *svc = data;
473 unsigned int cur, worst;
479 cur = at_get(&svc->srv_at_estimate);
480 worst = svc->srv_at_estimate.at_worst_ever;
481 worstt = svc->srv_at_estimate.at_worst_time;
482 s2dhms(&ts, cfs_time_current_sec() - worstt);
484 rc += snprintf(page + rc, count - rc,
485 "adaptive timeouts off, using obd_timeout %u\n",
487 rc += snprintf(page + rc, count - rc,
488 "%10s : cur %3u worst %3u (at %ld, "DHMS_FMT" ago) ",
489 "service", cur, worst, worstt,
491 rc = lprocfs_at_hist_helper(page, count, rc,
492 &svc->srv_at_estimate);
496 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
497 struct ptlrpc_service *svc)
499 struct lprocfs_vars lproc_vars[] = {
500 {.name = "req_buffer_history_len",
502 .read_fptr = ptlrpc_lprocfs_read_req_history_len,
504 {.name = "req_buffer_history_max",
505 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
506 .read_fptr = ptlrpc_lprocfs_read_req_history_max,
509 .read_fptr = ptlrpc_lprocfs_rd_timeouts,
513 static struct file_operations req_history_fops = {
514 .owner = THIS_MODULE,
515 .open = ptlrpc_lprocfs_svc_req_history_open,
518 .release = lprocfs_seq_release,
523 ptlrpc_lprocfs_register(entry, svc->srv_name,
524 "stats", &svc->srv_procroot,
527 if (svc->srv_procroot == NULL)
530 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
532 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
533 0400, &req_history_fops, svc);
535 CWARN("Error adding the req_history file\n");
538 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
540 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
541 &obddev->obd_svc_procroot,
542 &obddev->obd_svc_stats);
544 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
546 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req)
548 struct lprocfs_stats *svc_stats;
549 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
550 int opc = opcode_offset(op);
552 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
553 if (svc_stats == NULL || opc <= 0)
555 LASSERT(opc < LUSTRE_MAX_OPCODES);
556 /* These two use the ptlrpc_lprocfs_brw below */
557 if (!(opc == OST_WRITE || opc == OST_READ || op == LDLM_ENQUEUE
559 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, 0);
562 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes)
564 struct lprocfs_stats *svc_stats;
565 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
568 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, bytes);
570 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
572 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
574 if (svc->srv_procroot != NULL)
575 lprocfs_remove(&svc->srv_procroot);
578 lprocfs_free_stats(&svc->srv_stats);
581 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
583 if (obd->obd_svc_procroot)
584 lprocfs_remove(&obd->obd_svc_procroot);
586 if (obd->obd_svc_stats)
587 lprocfs_free_stats(&obd->obd_svc_stats);
589 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
592 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
593 unsigned long count, void *data)
595 struct obd_device *obd = data;
596 char tmpbuf[sizeof(struct obd_uuid)];
598 /* Kludge code(deadlock situation): the lprocfs lock has been held
599 * since the client is evicted by writting client's
600 * uuid/nid to procfs "evict_client" entry. However,
601 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
602 * the proc entries under the being destroyed export{}, so I have
603 * to drop the lock at first here.
604 * - jay, jxiong@clusterfs.com */
608 sscanf(buffer, "%40s", tmpbuf);
609 if (strncmp(tmpbuf, "nid:", 4) == 0)
610 obd_export_evict_by_nid(obd, tmpbuf + 4);
611 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
612 obd_export_evict_by_uuid(obd, tmpbuf + 5);
614 obd_export_evict_by_uuid(obd, tmpbuf);
621 EXPORT_SYMBOL(lprocfs_wr_evict_client);
623 int lprocfs_wr_ping(struct file *file, const char *buffer,
624 unsigned long count, void *data)
626 struct obd_device *obd = data;
627 struct ptlrpc_request *req;
631 LPROCFS_CLIMP_CHECK(obd);
632 req = ptlrpc_request_alloc_pack(obd->u.cli.cl_import, &RQF_OBD_PING,
633 LUSTRE_OBD_VERSION, OBD_PING);
635 LPROCFS_CLIMP_EXIT(obd);
639 ptlrpc_request_set_replen(req);
640 req->rq_send_state = LUSTRE_IMP_FULL;
641 req->rq_no_resend = 1;
642 req->rq_no_delay = 1;
644 rc = ptlrpc_queue_wait(req);
646 ptlrpc_req_finished(req);
651 EXPORT_SYMBOL(lprocfs_wr_ping);