1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 Cluster File Systems, Inc.
6 * This file is part of the Lustre file system, http://www.lustre.org
7 * Lustre is a trademark of Cluster File Systems, Inc.
9 * You may have signed or agreed to another license before downloading
10 * this software. If so, you are bound by the terms and conditions
11 * of that agreement, and the following does not apply to you. See the
12 * LICENSE file included with this distribution for more information.
14 * If you did not agree to a different license, then this copy of Lustre
15 * is open source software; you can redistribute it and/or modify it
16 * under the terms of version 2 of the GNU General Public License as
17 * published by the Free Software Foundation.
19 * In either case, Lustre is distributed in the hope that it will be
20 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * license text for more details.
25 #define DEBUG_SUBSYSTEM S_CLASS
28 # include <liblustre.h>
31 #include <obd_support.h>
33 #include <lprocfs_status.h>
34 #include <lustre/lustre_idl.h>
35 #include <lustre_net.h>
36 #include <obd_class.h>
37 #include "ptlrpc_internal.h"
40 struct ll_rpc_opcode {
43 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
44 { OST_REPLY, "ost_reply" },
45 { OST_GETATTR, "ost_getattr" },
46 { OST_SETATTR, "ost_setattr" },
47 { OST_READ, "ost_read" },
48 { OST_WRITE, "ost_write" },
49 { OST_CREATE , "ost_create" },
50 { OST_DESTROY, "ost_destroy" },
51 { OST_GET_INFO, "ost_get_info" },
52 { OST_CONNECT, "ost_connect" },
53 { OST_DISCONNECT, "ost_disconnect" },
54 { OST_PUNCH, "ost_punch" },
55 { OST_OPEN, "ost_open" },
56 { OST_CLOSE, "ost_close" },
57 { OST_STATFS, "ost_statfs" },
58 { 14, NULL }, /* formerly OST_SAN_READ */
59 { 15, NULL }, /* formerly OST_SAN_WRITE */
60 { OST_SYNC, "ost_sync" },
61 { OST_SET_INFO, "ost_set_info" },
62 { OST_QUOTACHECK, "ost_quotacheck" },
63 { OST_QUOTACTL, "ost_quotactl" },
64 { MDS_GETATTR, "mds_getattr" },
65 { MDS_GETATTR_NAME, "mds_getattr_lock" },
66 { MDS_CLOSE, "mds_close" },
67 { MDS_REINT, "mds_reint" },
68 { MDS_READPAGE, "mds_readpage" },
69 { MDS_CONNECT, "mds_connect" },
70 { MDS_DISCONNECT, "mds_disconnect" },
71 { MDS_GETSTATUS, "mds_getstatus" },
72 { MDS_STATFS, "mds_statfs" },
73 { MDS_PIN, "mds_pin" },
74 { MDS_UNPIN, "mds_unpin" },
75 { MDS_SYNC, "mds_sync" },
76 { MDS_DONE_WRITING, "mds_done_writing" },
77 { MDS_SET_INFO, "mds_set_info" },
78 { MDS_QUOTACHECK, "mds_quotacheck" },
79 { MDS_QUOTACTL, "mds_quotactl" },
80 { MDS_GETXATTR, "mds_getxattr" },
81 { MDS_SETXATTR, "mds_setxattr" },
82 { MDS_WRITEPAGE, "mds_writepage" },
83 { MDS_IS_SUBDIR, "mds_is_subdir" },
84 { LDLM_ENQUEUE, "ldlm_enqueue" },
85 { LDLM_CONVERT, "ldlm_convert" },
86 { LDLM_CANCEL, "ldlm_cancel" },
87 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
88 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
89 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
90 { MGS_CONNECT, "mgs_connect" },
91 { MGS_DISCONNECT, "mgs_disconnect" },
92 { MGS_EXCEPTION, "mgs_exception" },
93 { MGS_TARGET_REG, "mgs_target_reg" },
94 { MGS_TARGET_DEL, "mgs_target_del" },
95 { MGS_SET_INFO, "mgs_set_info" },
96 { OBD_PING, "obd_ping" },
97 { OBD_LOG_CANCEL, "llog_origin_handle_cancel" },
98 { OBD_QC_CALLBACK, "obd_qc_callback" },
99 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
100 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
101 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
102 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
103 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
104 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
105 { LLOG_CATINFO, "llog_catinfo" },
106 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
107 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
108 { FLD_QUERY, "fld_query" },
109 { SEQ_QUERY, "seq_query" },
110 { SEC_CTX_INIT, "sec_ctx_init" },
111 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
112 { SEC_CTX_FINI, "sec_ctx_fini" }
118 } ll_eopcode_table[EXTRA_LAST_OPC] = {
119 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
120 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
121 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
122 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
123 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
124 { MDS_REINT_CREATE, "mds_reint_create" },
125 { MDS_REINT_LINK, "mds_reint_link" },
126 { MDS_REINT_OPEN, "mds_reint_open" },
127 { MDS_REINT_SETATTR, "mds_reint_setattr" },
128 { MDS_REINT_RENAME, "mds_reint_rename" },
129 { MDS_REINT_UNLINK, "mds_reint_unlink" }
132 const char *ll_opcode2str(__u32 opcode)
134 /* When one of the assertions below fail, chances are that:
135 * 1) A new opcode was added in lustre_idl.h, but was
136 * is missing from the table above.
137 * or 2) The opcode space was renumbered or rearranged,
138 * and the opcode_offset() function in
139 * ptlrpc_internal.h needs to be modified.
141 __u32 offset = opcode_offset(opcode);
142 LASSERT(offset < LUSTRE_MAX_OPCODES);
143 LASSERT(ll_rpc_opcode_table[offset].opcode == opcode);
144 return ll_rpc_opcode_table[offset].opname;
147 const char* ll_eopcode2str(__u32 opcode)
149 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
150 return ll_eopcode_table[opcode].opname;
153 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
154 char *name, struct proc_dir_entry **procroot_ret,
155 struct lprocfs_stats **stats_ret)
157 struct proc_dir_entry *svc_procroot;
158 struct lprocfs_stats *svc_stats;
160 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
163 LASSERT(*procroot_ret == NULL);
164 LASSERT(*stats_ret == NULL);
166 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES, 0);
167 if (svc_stats == NULL)
171 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
172 if (IS_ERR(svc_procroot)) {
173 lprocfs_free_stats(&svc_stats);
180 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
181 svc_counter_config, "req_waittime", "usec");
182 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
183 svc_counter_config, "req_qdepth", "reqs");
184 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
185 svc_counter_config, "req_active", "reqs");
186 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
187 svc_counter_config, "reqbuf_avail", "bufs");
188 for (i = 0; i < EXTRA_LAST_OPC; i++) {
189 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
191 ll_eopcode2str(i), "reqs");
193 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
194 __u32 opcode = ll_rpc_opcode_table[i].opcode;
195 lprocfs_counter_init(svc_stats,
196 EXTRA_MAX_OPCODES + i, svc_counter_config,
197 ll_opcode2str(opcode), "usec");
200 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
203 lprocfs_remove(&svc_procroot);
204 lprocfs_free_stats(&svc_stats);
207 *procroot_ret = svc_procroot;
208 *stats_ret = svc_stats;
213 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
214 int count, int *eof, void *data)
216 struct ptlrpc_service *svc = data;
219 return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
223 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
224 int count, int *eof, void *data)
226 struct ptlrpc_service *svc = data;
229 return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
233 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
234 unsigned long count, void *data)
236 struct ptlrpc_service *svc = data;
239 int rc = lprocfs_write_helper(buffer, count, &val);
247 /* This sanity check is more of an insanity check; we can still
248 * hose a kernel by allowing the request history to grow too
250 bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
251 if (val > num_physpages/(2 * bufpages))
254 spin_lock(&svc->srv_lock);
255 svc->srv_max_history_rqbds = val;
256 spin_unlock(&svc->srv_lock);
261 struct ptlrpc_srh_iterator {
263 struct ptlrpc_request *srhi_req;
267 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
268 struct ptlrpc_srh_iterator *srhi,
272 struct ptlrpc_request *req;
274 if (srhi->srhi_req != NULL &&
275 srhi->srhi_seq > svc->srv_request_max_cull_seq &&
276 srhi->srhi_seq <= seq) {
277 /* If srhi_req was set previously, hasn't been culled and
278 * we're searching for a seq on or after it (i.e. more
279 * recent), search from it onwards.
280 * Since the service history is LRU (i.e. culled reqs will
281 * be near the head), we shouldn't have to do long
283 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
284 LASSERT (!list_empty(&svc->srv_request_history));
285 e = &srhi->srhi_req->rq_history_list;
287 /* search from start */
288 e = svc->srv_request_history.next;
291 while (e != &svc->srv_request_history) {
292 req = list_entry(e, struct ptlrpc_request, rq_history_list);
294 if (req->rq_history_seq >= seq) {
295 srhi->srhi_seq = req->rq_history_seq;
296 srhi->srhi_req = req;
306 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
308 struct ptlrpc_service *svc = s->private;
309 struct ptlrpc_srh_iterator *srhi;
312 OBD_ALLOC(srhi, sizeof(*srhi));
317 srhi->srhi_req = NULL;
319 spin_lock(&svc->srv_lock);
320 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos);
321 spin_unlock(&svc->srv_lock);
324 *pos = srhi->srhi_seq;
328 OBD_FREE(srhi, sizeof(*srhi));
333 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
335 struct ptlrpc_srh_iterator *srhi = iter;
338 OBD_FREE(srhi, sizeof(*srhi));
342 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
343 void *iter, loff_t *pos)
345 struct ptlrpc_service *svc = s->private;
346 struct ptlrpc_srh_iterator *srhi = iter;
349 spin_lock(&svc->srv_lock);
350 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos + 1);
351 spin_unlock(&svc->srv_lock);
354 OBD_FREE(srhi, sizeof(*srhi));
358 *pos = srhi->srhi_seq;
362 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
364 struct ptlrpc_service *svc = s->private;
365 struct ptlrpc_srh_iterator *srhi = iter;
366 struct ptlrpc_request *req;
369 spin_lock(&svc->srv_lock);
371 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, srhi->srhi_seq);
374 req = srhi->srhi_req;
376 /* Print common req fields.
377 * CAVEAT EMPTOR: we're racing with the service handler
378 * here. The request could contain any old crap, so you
379 * must be just as careful as the service's request
380 * parser. Currently I only print stuff here I know is OK
381 * to look at coz it was set up in request_in_callback()!!! */
382 seq_printf(s, LPD64":%s:%s:"LPD64":%d:%s ",
383 req->rq_history_seq, libcfs_nid2str(req->rq_self),
384 libcfs_id2str(req->rq_peer), req->rq_xid,
385 req->rq_reqlen,ptlrpc_rqphase2str(req));
387 if (svc->srv_request_history_print_fn == NULL)
390 svc->srv_request_history_print_fn(s, srhi->srhi_req);
393 spin_unlock(&svc->srv_lock);
399 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
401 static struct seq_operations sops = {
402 .start = ptlrpc_lprocfs_svc_req_history_start,
403 .stop = ptlrpc_lprocfs_svc_req_history_stop,
404 .next = ptlrpc_lprocfs_svc_req_history_next,
405 .show = ptlrpc_lprocfs_svc_req_history_show,
407 struct proc_dir_entry *dp = PDE(inode);
408 struct seq_file *seqf;
411 LPROCFS_ENTRY_AND_CHECK(dp);
412 rc = seq_open(file, &sops);
418 seqf = file->private_data;
419 seqf->private = dp->data;
423 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
424 struct ptlrpc_service *svc)
426 struct lprocfs_vars lproc_vars[] = {
427 {.name = "req_buffer_history_len",
429 .read_fptr = ptlrpc_lprocfs_read_req_history_len,
431 {.name = "req_buffer_history_max",
432 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
433 .read_fptr = ptlrpc_lprocfs_read_req_history_max,
437 static struct file_operations req_history_fops = {
438 .owner = THIS_MODULE,
439 .open = ptlrpc_lprocfs_svc_req_history_open,
442 .release = lprocfs_seq_release,
444 struct proc_dir_entry *req_history;
446 ptlrpc_lprocfs_register(entry, svc->srv_name,
447 "stats", &svc->srv_procroot,
450 if (svc->srv_procroot == NULL)
453 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
455 req_history = create_proc_entry("req_history", 0400,
457 if (req_history != NULL) {
458 req_history->data = svc;
459 req_history->proc_fops = &req_history_fops;
463 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
465 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
466 &obddev->obd_svc_procroot,
467 &obddev->obd_svc_stats);
469 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
471 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req)
473 struct lprocfs_stats *svc_stats;
474 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
475 int opc = opcode_offset(op);
477 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
478 if (svc_stats == NULL || opc <= 0)
480 LASSERT(opc < LUSTRE_MAX_OPCODES);
481 /* These two use the ptlrpc_lprocfs_brw below */
482 if (!(opc == OST_WRITE || opc == OST_READ || op == LDLM_ENQUEUE
484 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, 0);
487 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes)
489 struct lprocfs_stats *svc_stats;
490 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
493 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, bytes);
495 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
497 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
499 if (svc->srv_procroot != NULL)
500 lprocfs_remove(&svc->srv_procroot);
503 lprocfs_free_stats(&svc->srv_stats);
506 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
508 if (obd->obd_svc_procroot)
509 lprocfs_remove(&obd->obd_svc_procroot);
511 if (obd->obd_svc_stats)
512 lprocfs_free_stats(&obd->obd_svc_stats);
514 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
517 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
518 unsigned long count, void *data)
520 struct obd_device *obd = data;
521 char tmpbuf[sizeof(struct obd_uuid)];
523 /* Kludge code(deadlock situation): the lprocfs lock has been held
524 * since the client is evicted by writting client's
525 * uuid/nid to procfs "evict_client" entry. However,
526 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
527 * the proc entries under the being destroyed export{}, so I have
528 * to drop the lock at first here.
529 * - jay, jxiong@clusterfs.com */
533 sscanf(buffer, "%40s", tmpbuf);
534 if (strncmp(tmpbuf, "nid:", 4) == 0)
535 obd_export_evict_by_nid(obd, tmpbuf + 4);
536 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
537 obd_export_evict_by_uuid(obd, tmpbuf + 5);
539 obd_export_evict_by_uuid(obd, tmpbuf);
546 EXPORT_SYMBOL(lprocfs_wr_evict_client);
548 int lprocfs_wr_ping(struct file *file, const char *buffer,
549 unsigned long count, void *data)
551 struct obd_device *obd = data;
552 struct ptlrpc_request *req;
556 LPROCFS_CLIMP_CHECK(obd);
557 req = ptlrpc_request_alloc_pack(obd->u.cli.cl_import, &RQF_OBD_PING,
558 LUSTRE_OBD_VERSION, OBD_PING);
560 LPROCFS_CLIMP_EXIT(obd);
564 ptlrpc_request_set_replen(req);
565 req->rq_send_state = LUSTRE_IMP_FULL;
566 req->rq_no_resend = 1;
568 rc = ptlrpc_queue_wait(req);
570 ptlrpc_req_finished(req);
575 EXPORT_SYMBOL(lprocfs_wr_ping);