1#!/usr/sbin/dtrace -s
2/*
3 * fspaging.d - file system read/write and paging tracing.
4 *              Written using DTrace (Solaris 10 3/05)
5 *
6 * This traces file related activity: system call reads and writes,
7 * vnode logical read and writes (fop), vnode putpage and getpage activity,
8 * and disk I/O. It can be used to examine the behaviour of each I/O
9 * layer, from the syscall interface to what the disk is doing. Behaviour
10 * such as read-ahead, and max I/O size breakup can be observed.
11 *
12 * This is a verbose version of fsrw.d, as this also traces paging activity.
13 *
14 * $Id: fspaging.d 3 2007-08-01 10:50:08Z brendan $
15 *
16 * USAGE:	fspaging.d
17 *
18 * FIELDS:
19 *		Event		Traced event (see EVENTS below)
20 *		Device		Device, for disk I/O
21 *		RW		Either Read or Write
22 *		Size		Size of I/O in bytes, if known
23 *		Offset		Offset of I/O in kilobytes, if known
24 *		Path		Path to file on disk
25 *
26 * EVENTS:
27 *		sc-read		System call read
28 *		sc-write	System call write
29 *		fop_read	Logical read
30 *		fop_write	Logical write
31 *		fop_getpage	Logical get page
32 *		fop_putpage	Logical put page
33 *		disk_io		Physical disk I/O
34 *		disk_ra		Physical disk I/O, read ahead
35 *
36 * The events are drawn with a level of indentation, which can sometimes
37 * help identify related events.
38 *
39 * SEE ALSO: fsrw.d
40 *
41 * IDEA: Richard McDougall, Solaris Internals 2nd Ed, FS Chapter.
42 *
43 * COPYRIGHT: Copyright (c) 2006 Brendan Gregg.
44 *
45 * CDDL HEADER START
46 *
47 *  The contents of this file are subject to the terms of the
48 *  Common Development and Distribution License, Version 1.0 only
49 *  (the "License").  You may not use this file except in compliance
50 *  with the License.
51 *
52 *  You can obtain a copy of the license at Docs/cddl1.txt
53 *  or http://www.opensolaris.org/os/licensing.
54 *  See the License for the specific language governing permissions
55 *  and limitations under the License.
56 *
57 * CDDL HEADER END
58 *
59 * ToDo: readv()
60 *
61 * 20-Mar-2006  Brendan Gregg   Created this.
62 * 23-Apr-2006	   "      "	Last update.
63 */
64
65#pragma D option quiet
66#pragma D option switchrate=10hz
67
68dtrace:::BEGIN
69{
70	printf("%-13s %10s %2s %8s %6s %s\n",
71	    "Event", "Device", "RW", "Size", "Offset", "Path");
72}
73
74syscall::*read:entry,
75syscall::*write*:entry
76{
77	/*
78	 * starting with a file descriptior, dig out useful info
79	 * from the corresponding file_t and vnode_t.
80	 */
81	this->filistp = curthread->t_procp->p_user.u_finfo.fi_list;
82	this->ufentryp = (uf_entry_t *)((uint64_t)this->filistp +
83	    (uint64_t)arg0 * (uint64_t)sizeof (uf_entry_t));
84	this->filep = this->ufentryp->uf_file;
85	self->offset = this->filep->f_offset;
86	this->vnodep = this->filep != 0 ? this->filep->f_vnode : 0;
87	self->vpath = this->vnodep ? (this->vnodep->v_path != 0 ?
88	    cleanpath(this->vnodep->v_path) : "<unknown>") : "<unknown>";
89	self->sc_trace = this->vnodep ? this->vnodep->v_type == 1 ||
90	    this->vnodep->v_type == 2 ? 1 : 0 : 0;
91}
92
93syscall::*read:entry
94/self->sc_trace/
95{
96	printf("sc-%-10s %10s %2s %8d %6d %s\n", probefunc, ".", "R",
97	    (int)arg2, self->offset / 1024, self->vpath);
98}
99
100syscall::*write*:entry
101/self->sc_trace/
102{
103	printf("sc-%-10s %10s %2s %8d %6d %s\n", probefunc, ".", "W",
104	    (int)arg2, self->offset / 1024, self->vpath);
105}
106
107syscall::*read:return,
108syscall::*write*:return
109{
110	self->vpath = 0;
111	self->offset = 0;
112	self->sc_trace = 0;
113}
114
115fbt::fop_putpage:entry,
116fbt::fop_getpage:entry
117/self->sc_trace && args[0]->v_path/
118{
119	printf("  %-11s %10s %2s %8d %6d %s\n", probefunc, ".",
120	    probefunc == "fop_getpage" ? "R" : "W", (uint64_t)arg2,
121	    args[1] / 1024, cleanpath(args[0]->v_path));
122}
123
124
125fbt::fop_read:entry,
126fbt::fop_write:entry
127/self->sc_trace && args[0]->v_path/
128{
129	printf("  %-11s %10s %2s %8d %6d %s\n", probefunc, ".",
130	    probefunc == "fop_read" ? "R" : "W", args[1]->uio_resid,
131	    args[1]->_uio_offset._f / 1024, cleanpath(args[0]->v_path));
132}
133
134fbt:ufs:ufs_getpage_ra:entry
135{
136	/* fetch the real offset (file_t is unaware of this) */
137	self->offset = ((inode_t *)args[0]->v_data)->i_nextrio;
138	self->read_ahead = 1;
139}
140
141fbt:ufs:ufs_getpage_ra:return
142{
143	self->read_ahead = 0;
144	self->offset = 0;
145}
146
147io::bdev_strategy:start
148{
149	this->offset = self->read_ahead ? self->offset : args[2]->fi_offset;
150	printf("    %-9s %10s %2s %8d %6d %s\n",
151	    self->read_ahead ? "disk_ra" : "disk_io", args[1]->dev_statname,
152	    args[0]->b_flags & B_READ ? "R" : "W", args[0]->b_bcount,
153	    this->offset / 1024, args[2]->fi_pathname);
154}
155