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 * 23-Apr-2006, ver 0.50
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 */
63
64#pragma D option quiet
65#pragma D option switchrate=10hz
66
67dtrace:::BEGIN
68{
69	printf("%-13s %10s %2s %8s %6s %s\n",
70	    "Event", "Device", "RW", "Size", "Offset", "Path");
71}
72
73syscall::*read:entry,
74syscall::*write*:entry
75{
76	/*
77	 * starting with a file descriptior, dig out useful info
78	 * from the corresponding file_t and vnode_t.
79	 */
80	this->filistp = curthread->t_procp->p_user.u_finfo.fi_list;
81	this->ufentryp = (uf_entry_t *)((uint64_t)this->filistp +
82	    (uint64_t)arg0 * (uint64_t)sizeof (uf_entry_t));
83	this->filep = this->ufentryp->uf_file;
84	self->offset = this->filep->f_offset;
85	this->vnodep = this->filep != 0 ? this->filep->f_vnode : 0;
86	self->vpath = this->vnodep ? (this->vnodep->v_path != 0 ?
87	    cleanpath(this->vnodep->v_path) : "<unknown>") : "<unknown>";
88	self->sc_trace = this->vnodep ? this->vnodep->v_type == 1 ||
89	    this->vnodep->v_type == 2 ? 1 : 0 : 0;
90}
91
92syscall::*read:entry
93/self->sc_trace/
94{
95	printf("sc-%-10s %10s %2s %8d %6d %s\n", probefunc, ".", "R",
96	    (int)arg2, self->offset / 1024, self->vpath);
97}
98
99syscall::*write*:entry
100/self->sc_trace/
101{
102	printf("sc-%-10s %10s %2s %8d %6d %s\n", probefunc, ".", "W",
103	    (int)arg2, self->offset / 1024, self->vpath);
104}
105
106syscall::*read:return,
107syscall::*write*:return
108{
109	self->vpath = 0;
110	self->offset = 0;
111	self->sc_trace = 0;
112}
113
114fbt::fop_putpage:entry,
115fbt::fop_getpage:entry
116/self->sc_trace && args[0]->v_path/
117{
118	printf("  %-11s %10s %2s %8d %6d %s\n", probefunc, ".",
119	    probefunc == "fop_getpage" ? "R" : "W", (uint64_t)arg2,
120	    args[1] / 1024, cleanpath(args[0]->v_path));
121}
122
123
124fbt::fop_read:entry,
125fbt::fop_write:entry
126/self->sc_trace && args[0]->v_path/
127{
128	printf("  %-11s %10s %2s %8d %6d %s\n", probefunc, ".",
129	    probefunc == "fop_read" ? "R" : "W", args[1]->uio_resid,
130	    args[1]->_uio_offset._f / 1024, cleanpath(args[0]->v_path));
131}
132
133fbt:ufs:ufs_getpage_ra:entry
134{
135	/* fetch the real offset (file_t is unaware of this) */
136	self->offset = ((inode_t *)args[0]->v_data)->i_nextrio;
137	self->read_ahead = 1;
138}
139
140fbt:ufs:ufs_getpage_ra:return
141{
142	self->read_ahead = 0;
143	self->offset = 0;
144}
145
146io::bdev_strategy:start
147{
148	this->offset = self->read_ahead ? self->offset : args[2]->fi_offset;
149	printf("    %-9s %10s %2s %8d %6d %s\n",
150	    self->read_ahead ? "disk_ra" : "disk_io", args[1]->dev_statname,
151	    args[0]->b_flags & B_READ ? "R" : "W", args[0]->b_bcount,
152	    this->offset / 1024, args[2]->fi_pathname);
153}
154