blob: 7b48e48b996834816a6a28f4d5392088d015a05b [file] [log] [blame]
#!/bin/sh
# A Poor (but Free) Man's dtrace
#
# Copyright (C) 2014-2021 Free Software Foundation, Inc.
#
# Contributed by Oracle, Inc.
#
# This file is part of GDB.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see
# <http://www.gnu.org/licenses/>.
# DISCLAIMER DISCLAIMER DISCLAIMER
# This script is a test tool. As such it is in no way intended to
# replace the "real" dtrace command for any practical purpose, apart
# from testing the DTrace USDT probes support in GDB.
# that said...
#
# pdtrace is a limited dtrace program, implementing a subset of its
# functionality:
#
# - The generation of an ELF file containing an embedded dtrace
# program. Equivalent to dtrace -G.
#
# - The generation of a header file with definitions for static
# probes. Equivalent to dtrace -h.
#
# This allows to generate DTrace static probes without having to use
# the user-level DTrace components. The generated objects are 100%
# compatible with DTrace and can be traced by the dtrace kernel module
# like if they were generated by dtrace.
#
# Some of the known limitations of this implementation are:
# - The input d-script must describe one provider, and only one.
# - The "probe " directives in the d-file must not include argument
# names, just the types. Thus something like `char *' is valid, but
# `char *name' is not.
# - The command line options must precede other arguments, since the
# script uses the (more) portable getopts.
# - Each probe header in the d-script must be contained in
# a single line.
# - strip -K removes the debugging information from the input object
# file.
# - The supported target platforms are i[3456]86 and x86_64.
#
# Please keep this code as portable as possible. Restrict yourself to
# POSIX sh.
# This script uses the following external programs, defined in
# variables. Some of them are substituted by autoconf.
TR=tr
NM=@NM_TRANSFORM_NAME@
EGREP=egrep
SED=sed
CUT=cut
READELF=@READELF_TRANSFORM_NAME@
SORT=sort
EXPR=expr
WC=wc
UNIQ=uniq
HEAD=head
SEQ=seq
AS=@GAS_TRANSFORM_NAME@
STRIP=@STRIP_TRANSFORM_NAME@
TRUE=true
# Sizes for several DOF structures, in bytes.
#
# See linux/dtrace/dof.h for the definition of the referred
# structures.
dof_hdrsize=64 # sizeof(dtrace_dof_hdr)
dof_secsize=32 # sizeof(dtrace_dof_sect)
dof_probesize=48 # sizeof(dtrace_dof_probe)
dof_providersize=44 # sizeof(dtrace_dof_provider)
# Types for the several DOF sections.
#
# See linux/dtrace/dof_defines.h for a complete list of section types
# along with their values.
dof_sect_type_strtab=8
dof_sect_type_provider=15
dof_sect_type_probes=16
dof_sect_type_prargs=17
dof_sect_type_proffs=18
dof_sect_type_prenoffs=26
### Functions
# Write a message to the standard error output and exit with an error
# status.
#
# Arguments:
# $1 error message.
f_panic()
{
echo "error: $1" 1>&2; exit 1
}
# Write a usage message to the standard output and exit with an error
# status.
f_usage()
{
printf "Usage: pdtrace [-32|-64] [-GhV] [-o output] [-s script] [ args ... ]\n\n"
printf "\t-32 generate 32-bit ELF files\n"
printf "\t-64 generate 64-bit ELF files\n\n"
printf "\t-G generate an ELF file containing embedded dtrace program\n"
printf "\t-h generate a header file with definitions for static probes\n"
printf "\t-o set output file\n"
printf "\t-s handle probes according to the specified D script\n"
printf "\t-V report the DTrace API version implemented by the tool\n"
exit 2
}
# Write a version message to the standard output and exit with a
# successful status.
f_version()
{
echo "pdtrace: Sun D 1.6.3"
exit
}
# Add a new record to a list and return it.
#
# Arguments:
# $1 is the list.
# $2 is the new record
f_add_record()
{
rec=$1
test -n "$rec" && \
{ rec=$(printf %s\\n "$rec"; echo x); rec=${rec%x}; }
printf %s "$rec$2"
}
# Collect the providers and probes information from the input object
# file.
#
# This function sets the values of the following global variables.
# The values are structured in records, each record in a line. The
# fields of each record are separated in some cases by white
# characters and in other cases by colon (:) characters.
#
# The type codes in the line format descriptors are:
# S: string, D: decimal number
#
# probes
# Regular probes and is-enabled probes.
# TYPE(S) PROVIDER(S) NAME(S) OFFSET(D) BASE(D) BASE_SYM(S)
# base_probes
# Base probes, i.e. probes sharing provider, name and container.
# PROVIDER(S) NAME(S) BASE(D) BASE_SYM(S)
# providers
# List of providers.
# PROVIDER(S)
# All the offsets are expressed in bytes.
#
# Input globals:
# objfile
# Output globals:
# probes, base_probes, providers
probes=
base_probes=
providers=
probes_args=
f_collect_probes()
{
# Probe points are function calls to undefined functions featuring
# distinct names for both normal probes and is-enabled probes.
PROBE_REGEX="(__dtrace_([a-zA-Z_]+)___([a-zA-Z_]+))"
EPROBE_REGEX="(__dtraceenabled_([a-zA-Z_]+)___([a-zA-Z_]+))"
while read type symbol provider name; do
test -z "$type" && f_panic "No probe points found in $objfile"
provider=$(printf %s $provider | $TR -s _)
name=$(printf %s $name | $TR -s _)
# Search the object file for relocations defined for the
# probe symbols. Then calculate the base address of the
# probe (along with the symbol associated with that base
# address) and the offset of the probe point.
for offset in $($READELF -W -r $objfile | $EGREP $symbol | $CUT -d' ' -f1)
do
# Figure out the base address for the probe. This is
# done finding the function name in the text section of
# the object file located above the probed point. But
# note that the relocation is for the address operand of
# the call instruction, so we have to subtract 1 to find
# the real probed point.
offset=$((0x$offset - 1))
# The addresses of is-enabled probes must point to the
# first NOP instruction in their patched instructions
# sequences, so modify them (see f_patch_objfile for the
# instruction sequences).
if test "$type" = "e"; then
if test "$objbits" -eq "32"; then
offset=$((offset + 2))
else # 64 bits
offset=$((offset + 3))
fi
fi
# Determine the base address of the probe and its
# corresponding function name.
funcs=$($NM -td $objfile | $EGREP "^[0-9]+ T " \
| $CUT -d' ' -f1,3 | $SORT -n -r | $TR ' ' :)
for fun in $funcs; do
func_off=$(printf %s $fun | $CUT -d: -f1)
func_sym=$(printf %s $fun | $CUT -d: -f2)
# Note that `expr' is used to remove leading zeros
# to avoid FUNC_OFF to be interpreted as an octal
# number in arithmetic contexts.
test "$func_off" -le "$offset" && \
{ base=$($EXPR $func_off + 0); break; }
done
test -n "$base" || \
f_panic "could not find base address for probe at $objfile($o)"
# Emit the record for the probe.
probes=$(f_add_record "$probes" \
"$type $provider $name $(($offset - $base)) $base $func_sym")
done
done <<EOF
$($NM $objfile | $EGREP " U $PROBE_REGEX" \
| $SED -E -e "s/.*$PROBE_REGEX.*/p \1 \2 \3/";
$NM $objfile | $EGREP " U $EPROBE_REGEX" \
| $SED -E -e "s/.*$EPROBE_REGEX.*/e \1 \2 \3/")
EOF
# Build the list of providers and of base probes from the probes.
while read type provider name offset base base_sym; do
providers=$(f_add_record "$providers" "$provider")
base_probes=$(f_add_record "$base_probes" "$provider $name $base $base_sym")
done <<EOF
$probes
EOF
providers=$(printf %s\\n "$providers" | $SORT | $UNIQ)
base_probes=$(printf %s\\n "$base_probes" | $SORT | $UNIQ)
}
# Collect the argument counts and type strings for all the probes
# described in the `probes' global variable. This is done by
# inspecting the d-script file provided by the user.
#
# This function sets the values of the following global variables.
# The values are structured in records, each record in a line. The
# fields of each record are separated in some cases by white
# characters and in other cases by colon (:) characters.
#
# The type codes in the line format descriptors are:
# S: string, D: decimal number
#
# probes_args
# Probes arguments.
# PROVIDER(S):NAME(S):NARGS(D):ARG1(S):ARG2(S):...:ARGn(S)
#
# Input globals:
# probes
# Output globals:
# probes_args
# Arguments:
# $1 is the d-script file from which to extract the arguments
# information.
f_collect_probes_args()
{
dscript=$1
while read type provider name offset base base_sym; do
# Process normal probes only. Is-enabled probes are not
# described in the d-script file and they don't receive any
# argument.
test "$type" = "p" || continue
# Names are mangled in d-script files to make it possible to
# have underscore characters as part of the provider name and
# probe name.
m_provider=$(printf %s $provider | $SED -e 's/_/__/g')
m_name=$(printf %s $name | $SED -e 's/_/__/g')
# Ignore this probe if the d-script file does not describe its
# provider.
$EGREP -q "provider +$m_provider" $dscript || continue
# Look for the line containing the description of the probe.
# If we can't find it then ignore this probe.
line=$($EGREP "^ *probe +$m_name *\(.*\);" $dscript)
test -n "$line" || continue
# Ok, extract the argument types from the probe prototype.
# This is fragile as hell as it requires the prototype to be
# in a single line.
args=""; nargs=0; line=$(printf %s "$line" | $SED -e 's/.*(\(.*\)).*/\1/')
set -f; IFS=,
for arg in $line; do
args="$args:$arg"
nargs=$((nargs + 1))
done
set +f; unset IFS
# Emit the record for the probe arguments.
probes_args=$(f_add_record "$probes_args" "$provider:$name:$nargs$args")
done <<EOF
$probes
EOF
}
# Functions to manipulate the global BCOUNT.
BCOUNT=0
f_incr_bcount()
{
BCOUNT=$((BCOUNT + $1))
}
f_align_bcount()
{
test $((BCOUNT % $1)) -eq 0 || BCOUNT=$((BCOUNT + ($1 - (BCOUNT % $1))))
}
# Generate a line of assembly code and add it to the asmprogram global
# variable.
#
# Arguments:
# $1 string to generate in a line.
asmprogram=
f_gen_asm()
{
line=$(printf "\t$1")
asmprogram=$(f_add_record "$asmprogram" "$line")
}
# Helper function to generate the assembly code of a DOF section
# header.
#
# This function is used by `f_gen_dof_program'.
#
# Arguments:
# $1 is the name of the described section.
# $2 is the type of the described section.
# $3 is the alignment of the described section.
# $4 is the number of entities stored in the described section.
# $5 is the offset in the DOF program of the described section.
# $6 is the size of the described section, in bytes.
f_gen_dof_sect_header()
{
f_gen_asm ""
f_gen_asm "/* dtrace_dof_sect for the $1 section. */"
f_gen_asm ".balign 8"
f_gen_asm ".4byte $2\t/* uint32_t dofs_type */"
f_gen_asm ".4byte $3\t/* uint32_t dofs_align */"
# The DOF_SECF_LOAD flag is 1 => loadable section.
f_gen_asm ".4byte 1\t/* uint32_t dofs_flags */"
f_gen_asm ".4byte $4\t/* uint32_t dofs_entsize */"
f_gen_asm ".8byte $5\t/* uint64_t dofs_offset */"
f_gen_asm ".8byte $6\t/* uint64_t dofs_size */"
}
# Generate a DOF program and assembly it in the output file.
#
# The DOF program generated by this function has the following
# structure:
#
# HEADER
# STRTAB OFFTAB EOFFTAB [PROBES PROVIDER]...
# STRTAB_SECT OFFTAB_SECT EOFFTAB_SECT ARGTAB_SECT [PROBES_SECT PROVIDER_SECT]...
#
# Input globals:
# probes, base_probes, providers, probes_args, BCOUNT
f_gen_dof_program()
{
###### Variables used to cache information needed later.
# Number of section headers in the generated DOF program.
dof_secnum=0
# Offset of section headers in the generated DOF program, in bytes.
dof_secoff=0
# Sizes of the STRTAB, OFFTAB and EOFFTAB sections, in bytes.
strtab_size=0
offtab_size=0
eofftab_size=0
# Offsets of the STRTAB, OFFTAB EOFFTAB and PROBES sections in the
# generated DOF program. In bytes.
strtab_offset=0
offtab_offset=0
eofftab_offset=0
argtab_offset=0
probes_offset=0
# Indexes of the section headers of the STRTAB, OFFTAB, EOFFTAB and
# PROBES sections in the sections array.
strtab_sect_index=0
offtab_sect_index=0
eofftab_sect_index=0
argtab_sect_index=0
probes_sect_index=0
# First offsets and eoffsets of the base-probes.
# Lines: PROVIDER(S) NAME(S) BASE(D) (DOF_OFFSET(D)|DOF_EOFFSET(D))
probes_dof_offsets=
probes_dof_eoffsets=
# Offsets in the STRTAB section for the first type of base probes.
# Record per line: PROVIDER(S) NAME(S) BASE(D) OFFSET(D)
probes_dof_types=
# Offsets of the provider names in the provider's STRTAB section.
# Lines: PROVIDER(S) OFFSET(D)
providers_dof_names=
# Offsets of the base-probe names in the provider's STRTAB section.
# Lines: PROVIDER(S) NAME(S) BASE(D) OFFSET(D)
probes_dof_names=
# Offsets of the provider sections in the DOF program.
# Lines: PROVIDER(S) OFFSET(D)
providers_offsets=
###### Generation phase.
# The header of the DOF program contains a `struct
# dtrace_dof_hdr'. Record its size, but it is written at the end
# of the function.
f_incr_bcount $dof_hdrsize; f_align_bcount 8
# The STRTAB section immediately follows the header. It contains
# the following set of packed null-terminated strings:
#
# [PROVIDER [BASE_PROBE_NAME [BASE_PROBE_ARG_TYPE...]]...]...
strtab_offset=$BCOUNT
strtab_sect_index=$dof_secnum
dof_secnum=$((dof_secnum + 1))
f_gen_asm ""
f_gen_asm "/* The STRTAB section. */"
f_gen_asm ".balign 8"
# Add the provider names.
off=0
while read provider; do
strtab_size=$(($strtab_size + ${#prov} + 1))
# Note the funny mangling...
f_gen_asm ".asciz \"$(printf %s $provider | $TR _ -)\""
providers_dof_names=$(f_add_record "$providers_dof_names" \
"$provider $off")
off=$(($off + ${#provider} + 1))
# Add the base-probe names.
while read p_provider name base base_sym; do
test "$p_provider" = "$provider" || continue
# And yes, more funny mangling...
f_gen_asm ".asciz \"$(printf %s $name | $TR _ -)\""
probes_dof_names=$(f_add_record "$probes_dof_names" \
"$p_provider $name $base $off")
off=$(($off + ${#name} + 1))
while read args; do
a_provider=$(printf %s "$args" | $CUT -d: -f1)
a_name=$(printf %s "$args" | $CUT -d: -f2)
test "$a_provider" = "$p_provider" \
&& test "$a_name" = "$name" \
|| continue
probes_dof_types=$(f_add_record "$probes_dof_types" \
"$a_provider $name $base $off")
nargs=$(printf %s "$args" | $CUT -d: -f3)
for n in $($SEQ $nargs); do
arg=$(printf %s "$args" | $CUT -d: -f$(($n + 3)))
f_gen_asm ".asciz \"${arg}\""
off=$(($off + ${#arg} + 1))
done
done <<EOF
$probes_args
EOF
done <<EOF
$base_probes
EOF
done <<EOF
$providers
EOF
strtab_size=$off
f_incr_bcount $strtab_size; f_align_bcount 8
# The OFFTAB section contains a set of 32bit words, one per
# defined regular probe.
offtab_offset=$BCOUNT
offtab_sect_index=$dof_secnum
dof_secnum=$((dof_secnum + 1))
f_gen_asm ""
f_gen_asm "/* The OFFTAB section. */"
f_gen_asm ".balign 8"
off=0
while read type provider name offset base base_sym; do
test "$type" = "p" || continue
f_gen_asm ".4byte $offset\t/* probe ${provider}:${name} */"
probes_dof_offsets=$(f_add_record "$probes_dof_offsets" \
"$provider $name $base $off")
off=$(($off + 4))
done <<EOF
$probes
EOF
offtab_size=$off
f_incr_bcount $offtab_size; f_align_bcount 8
# The EOFFTAB section contains a set of 32bit words, one per
# defined is-enabled probe.
eofftab_offset=$BCOUNT
eofftab_sect_index=$dof_secnum
dof_secnum=$((dof_secnum + 1))
f_gen_asm ""
f_gen_asm "/* The EOFFTAB section. */"
f_gen_asm ".balign 8"
off=0
while read type provider name offset base base_sym; do
test "$type" = "e" || continue
f_gen_asm ".4byte $offset\t/* is-enabled probe ${provider}:${name} */"
probes_dof_eoffsets=$(f_add_record "$probes_dof_eoffsets" \
"$provider $name $base $off")
off=$(($off + 4))
done <<EOF
$probes
EOF
eofftab_size=$off
f_incr_bcount $eofftab_size; f_align_bcount 8
# The ARGTAB section is empty, but nonetheless has a section
# header, so record its section index here.
argtab_offset=0
argtab_sect_index=$dof_secnum
dof_secnum=$((dof_secnum + 1))
# Generate a pair of sections PROBES and PROVIDER for each
# provider.
while read prov; do
# The PROBES section contains an array of `struct
# dtrace_dof_probe'.
#
# A `dtrace_dof_probe' entry characterizes the collection of
# probes and is-enabled probes sharing the same provider, name and
# base address.
probes_sect_index=$dof_secnum
dof_secnum=$((dof_secnum + 1))
probes_offset=$BCOUNT
num_base_probes=$(printf %s\\n "$base_probes" | $WC -l)
while read provider name base base_sym; do
name_offset=$(printf %s\\n "$probes_dof_names" \
| $EGREP "^$provider $name " | $CUT -d' ' -f4)
num_offsets=$(printf %s\\n "$probes_dof_offsets" \
| $EGREP "^$provider $name [0-9]+ " | $WC -l)
first_offset=0
test "$num_offsets" -gt 0 && \
first_offset=$(printf %s\\n "$probes_dof_offsets" \
| $EGREP "^$provider $name " | $CUT -d' ' -f4 | $HEAD -1)
num_eoffsets=$(printf %s\\n "$probes_dof_eoffsets" \
| $EGREP "^$provider $name [0-9]+ " | $WC -l)
first_eoffset=0
test "$num_eoffsets" -gt 0 && \
first_eoffset=$(printf %s "$probes_dof_eoffsets" \
| $EGREP "^$provider $name " | $CUT -d' ' -f4 | $HEAD -1)
num_args=$(printf %s "$probes_args" \
| $EGREP "^$provider:$name:" | $CUT -d: -f3 | $HEAD -1)
first_type=$(printf %s "$probes_dof_types" \
| $EGREP "^$provider $name $base " | $CUT -d' ' -f4 | $HEAD -1)
reloctype=R_X86_64_GLOB_DAT
test "$objbits" = "32" && reloctype=R_386_32
f_gen_asm ""
f_gen_asm "/* dtrace_dof_probe for ${provider}:${name} at ${base_sym} */"
f_gen_asm ".balign 8"
f_gen_asm ".reloc ., $reloctype, $base_sym + 0"
f_gen_asm ".8byte ${base}\t/* uint64_t dofpr_addr */"
f_gen_asm ".4byte 0\t/* uint32_t dofpr_func */"
f_gen_asm ".4byte $name_offset\t/* uint32_t dofpr_name */"
f_gen_asm ".4byte $first_type\t/* uint32_t dofpr_nargv */"
f_gen_asm ".4byte 0\t/* uint32_t dofpr_xargv */"
f_gen_asm ".4byte 0\t/* uint32_t dofpr_argidx */"
f_gen_asm ".4byte $(($first_offset/4))\t/* uint32_t dofpr_offidx */"
f_gen_asm ".byte $num_args\t/* uint8_t dofpr_nargc */"
f_gen_asm ".byte 0\t/* uint8_t dofpr_xargc */"
f_gen_asm ".2byte $num_offsets\t/* uint16_t dofpr_noffs */"
f_gen_asm ".4byte $(($first_eoffset/4))\t/* uint32_t dofpr_enoffidx */"
f_gen_asm ".2byte $num_eoffsets\t/* uint16_t dofpr_nenoffs */"
f_gen_asm ".2byte 0\t/* uint16_t dofpr_pad1 */"
f_gen_asm ".4byte 0\t/* uint16_t dofpr_pad2 */"
f_incr_bcount "$dof_probesize"
done <<EOF
$base_probes
EOF
# The PROVIDER section contains a `struct dtrace_dof_provider'
# instance describing the provider for the probes above.
dof_secnum=$((dof_secnum + 1))
providers_offsets=$(f_add_record "$providers_offsets" \
"$prov $BCOUNT")
# The dtrace_dof_provider.
provider_name_offset=$(printf %s "$providers_dof_names" \
| $EGREP "^$prov " | $CUT -d' ' -f2)
f_gen_asm ""
f_gen_asm "/* dtrace_dof_provider for $prov */"
f_gen_asm ".balign 8"
# Links to several DOF sections.
f_gen_asm ".4byte $strtab_sect_index\t/* uint32_t dofpv_strtab */"
f_gen_asm ".4byte $probes_sect_index\t/* uint32_t dofpv_probes */"
f_gen_asm ".4byte $argtab_sect_index\t/* uint32_t dofpv_prargs */"
f_gen_asm ".4byte $offtab_sect_index\t/* uint32_t dofpv_proffs */"
# Offset of the provider name into the STRTAB section.
f_gen_asm ".4byte $provider_name_offset\t/* uint32_t dofpv_name */"
# The rest of fields can be 0 for our modest purposes :)
f_gen_asm ".4byte 0\t/* uint32_t dofpv_provattr */"
f_gen_asm ".4byte 0\t/* uint32_t dofpv_modattr */"
f_gen_asm ".4byte 0\t/* uint32_t dofpv_funcattr */"
f_gen_asm ".4byte 0\t/* uint32_t dofpv_nameattr */"
f_gen_asm ".4byte 0\t/* uint32_t dofpv_argsattr */"
# But not this one, of course...
f_gen_asm ".4byte $eofftab_sect_index\t/* uint32_t dofpv_prenoffs */"
f_incr_bcount $dof_providersize
done<<EOF
$providers
EOF
f_align_bcount 8
# The section headers follow, one per section defined above.
dof_secoff=$BCOUNT
f_gen_dof_sect_header STRTAB \
$dof_sect_type_strtab \
1 1 $strtab_offset $strtab_size
f_incr_bcount $dof_secsize; f_align_bcount 8
f_gen_dof_sect_header OFFTAB \
$dof_sect_type_proffs \
4 4 $offtab_offset $offtab_size
f_incr_bcount $dof_secsize; f_align_bcount 8
f_gen_dof_sect_header EOFFTAB \
$dof_sect_type_prenoffs \
4 4 $eofftab_offset $eofftab_size
f_incr_bcount $dof_secsize; f_align_bcount 8
f_gen_dof_sect_header ARGTAB \
$dof_sect_type_prargs \
4 1 $argtab_offset 0
f_incr_bcount $dof_secsize; f_align_bcount 8
while read provider; do
provider_offset=$(printf %s "$providers_offsets" \
| $EGREP "^$provider " | $CUT -d' ' -f2)
num_base_probes=$(printf %s\\n "$base_probes" | $WC -l)
f_gen_dof_sect_header "$provider probes" \
$dof_sect_type_probes \
8 $dof_probesize $probes_offset \
$((num_base_probes * dof_probesize))
f_incr_bcount $dof_secsize; f_align_bcount 8
f_gen_dof_sect_header "$provider provider" \
$dof_sect_type_provider \
8 1 $provider_offset $dof_providersize
f_incr_bcount $dof_secsize; f_align_bcount 8
done <<EOF
$providers
EOF
# Finally, cook the header.
asmbody="$asmprogram"
asmprogram=""
f_gen_asm "/* File generated by pdtrace. */"
f_gen_asm ""
f_gen_asm ".section .SUNW_dof,\"a\",\"progbits\""
f_gen_asm ".globl __SUNW_dof"
f_gen_asm ".hidden __SUNW_dof"
f_gen_asm ".size __SUNW_dof, ${BCOUNT}"
f_gen_asm ".type __SUNW_dof, @object"
f_gen_asm "__SUNW_dof:"
f_gen_asm ""
f_gen_asm "/* dtrace_dof_hdr */"
f_gen_asm ".balign 8"
f_gen_asm ".byte 0x7f, 'D, 'O, 'F\t/* dofh_ident[0..3] */"
f_gen_asm ".byte 2\t\t/* model: 1=ILP32, 2=LP64 */"
f_gen_asm ".byte 1\t\t/* encoding: 1: little-endian, 2: big-endian */"
f_gen_asm ".byte 2\t\t/* DOF version: 1 or 2. Latest is 2 */"
f_gen_asm ".byte 2\t\t/* DIF version: 1 or 2. Latest is 2 */"
f_gen_asm ".byte 8\t\t/* number of DIF integer registers */"
f_gen_asm ".byte 8\t\t/* number of DIF tuple registers */"
f_gen_asm ".byte 0, 0\t\t/* dofh_ident[10..11] */"
f_gen_asm ".4byte 0\t\t/* dofh_ident[12..15] */"
f_gen_asm ".4byte 0\t/* uint32_t dofh_flags */" # See Limitations above.
f_gen_asm ".4byte ${dof_hdrsize}\t/* uint32_t dofh_hdrsize */"
f_gen_asm ".4byte ${dof_secsize}\t/* uint32_t dofh_secsize */"
f_gen_asm ".4byte ${dof_secnum}\t/* uint32_t dofh_secnum */"
f_gen_asm ".8byte ${dof_secoff}\t/* uint64_t dofh_secoff */"
f_gen_asm ".8byte ${BCOUNT}\t/* uint64_t dofh_loadsz */"
f_gen_asm ".8byte ${BCOUNT}\t/* uint64_t dofh_filesz */"
f_gen_asm ".8byte 0\t/* uint64_t dofh_pad */"
f_gen_asm ""
# Ok, now assembly the program in OFILE
echo "$asmprogram$asmbody" | $AS -$objbits -o $ofile
# Next step is to change the sh_type of the ".SUNW_dof" section
# headers to 0x6ffffff4 (SHT_SUNW_dof).
#
# Note that this code relies in the fact that readelf will list
# the sections ordered in the same order than the section headers
# in the section header table of the file.
elfinfo=$($READELF -a $ofile)
# Mind the endianness.
if printf %s "$elfinfo" | $EGREP -q "little endian"; then
sht_sunw_dof=$(printf %s%s%s%s \\364 \\377 \\377 \\157)
else
sht_sunw_dof=$(printf %s%s%s%s \\157 \\377 \\377 \\364)
fi
shdr_start=$(printf %s "$elfinfo" \
| $EGREP "^[ \t]*Start of section headers:" \
| $SED -E -e 's/.*headers:[ \t]*([0-9]+).*/\1/')
test -n "$shdr_start" \
|| f_panic "could not extract the start of shdr from $ofile"
shdr_num_entries=$(printf %s "$elfinfo" \
| $EGREP "^[ \t]*Size of section headers:" \
| $SED -E -e 's/.*headers:[ \t]*([0-9]+).*/\1/')
test -n "$shdr_num_entries" \
|| f_panic "could not extract the number of shdr entries from $ofile"
shdr_entry_size=$(printf %s "$elfinfo" \
| $EGREP "^[ \t]*Size of section headers:" \
| $SED -E -e 's/.*headers:[ \t]*([0-9]+).*/\1/')
test -n "$shdr_entry_size" \
|| f_panic "could not fetch the size of section headers from $ofile"
while read line; do
data=$(printf %s "$line" \
| $SED -E -e 's/.*\[(.*)\][ \t]+([a-zA-Z_.]+).*/\1:\2/')
num=$(printf %s "$data" | $CUT -d: -f1)
name=$(printf %s "$data" | $CUT -d: -f2)
if test "$name" = ".SUNW_dof"; then
# Patch the new sh_type in the proper entry of the section
# header table.
printf "$sht_sunw_dof" \
| dd of=$ofile conv=notrunc count=4 ibs=1 bs=1 \
seek=$((shdr_start + (shdr_entry_size * num) + 4)) \
2> /dev/null
break
fi
done <<EOF
$(printf %s "$elfinfo" | $EGREP "^[ \t]*\[[0-9 ]+\].*[A-Z]+.*PROGBITS")
EOF
}
# Patch the probed points in the given object file, replacing the
# function calls with NOPs.
#
# The probed points in the input object files are function calls.
# This function replaces these function calls by some other
# instruction sequences. Which replacement to use depends on several
# factors, as documented below.
#
# Arguments:
# $1 is the object file to patch.
f_patch_objfile()
{
objfile=$1
# Several x86_64 instruction opcodes, in octal.
x86_op_nop=$(printf \\220)
x86_op_ret=$(printf \\303)
x86_op_call=$(printf \\350)
x86_op_jmp32=$(printf \\351)
x86_op_rex_rax=$(printf \\110)
x86_op_xor_eax_0=$(printf \\063)
x86_op_xor_eax_1=$(printf \\300)
# Figure out the file offset of the text section in the object
# file.
text_off=0x$(objdump -j .text -h $objfile \
| grep \.text | $TR -s ' ' | $CUT -d' ' -f 7)
while read type provider name offset base base_sym; do
# Calculate the offset of the probed point in the object file.
# Note that the `offset' of is-enabled probes is tweaked in
# `f_collect_probes" to point ahead the patching point.
probe_off=$((text_off + base + offset))
if test "$type" = "e"; then
if test "$objbits" -eq "32"; then
probe_off=$((probe_off - 2))
else # 64 bits
probe_off=$((probe_off - 3))
fi
fi
# The probed point can be either a CALL instruction or a JMP
# instruction (a tail call). This has an impact on the
# patching sequence. Fetch the first byte at the probed point
# and do the right thing.
nopret="$x86_op_nop"
byte=$(dd if=$objfile count=1 ibs=1 bs=1 skip=$probe_off 2> /dev/null)
test "$byte" = "$x86_op_jmp32" && nopret="$x86_op_ret"
# Determine the patching sequence. It depends on the type of
# probe at hand (regular or is-enabled) and also if
# manipulating a 32bit or 64bit binary.
patchseq=
case $type in
p) patchseq=$(printf %s%s%s%s%s \
"$nopret" \
"$x86_op_nop" \
"$x86_op_nop" \
"$x86_op_nop" \
"$x86_op_nop")
;;
e) test "$objbits" -eq 64 && \
patchseq=$(printf %s%s%s%s%s \
"$x86_op_rex_rax" \
"$x86_op_xor_eax_0" \
"$x86_op_xor_eax_1" \
"$nopret" \
"$x86_op_nop")
test "$objbits" -eq 32 && \
patchseq=$(printf %s%s%s%s%s \
"$x86_op_xor_eax_0" \
"$x86_op_xor_eax_1" \
"$nopret" \
"$x86_op_nop" \
"$x86_op_nop")
;;
*) f_panic "internal error: wrong probe type $type";;
esac
# Patch!
printf %s "$patchseq" \
| dd of=$objfile conv=notrunc count=5 ibs=1 bs=1 seek=$probe_off 2> /dev/null
done <<EOF
$probes
EOF
# Finally, we have to remove the __dtrace_* and __dtraceenabled_*
# symbols from the object file, along with their respective
# relocations.
#
# Note that the most obvious call:
# strip -v -N whatever -w foo.o
# will not work:
# strip: not stripping symbol `whatever' because it is named in a relocation
#
# Fortunately using `-K !whatever' instead tricks strip to do the
# right thing, but this is black magic and may eventually stop
# working...
$STRIP -K '!__dtrace_*' -w $objfile
$STRIP -K '!__dtraceenabled_*' -w $objfile
}
# Read the input .d file and print a header file with macros to
# invoke the probes defined in it.
f_gen_header_file()
{
guard=$(basename $ofile | $TR - _ | $CUT -d. -f1 | $TR a-z A-Z)
printf "/*\n * Generated by pdtrace.\n */\n\n"
printf "#ifndef _${guard}_H\n"
printf "#define _${guard}_H\n\n"
printf "#include <unistd.h>\n"
printf "#include <inttypes.h>\n"
printf \\n\\n
printf "#ifdef __cplusplus\nextern \"C\" {\n#endif\n"
printf "#define _DTRACE_VERSION 1\n\n"
provider=$(cat $dfile | $EGREP "^ *provider +([a-zA-Z_]+)" \
| $SED -E -e 's/^ *provider +([a-zA-Z]+).*/\1/')
test -z "$provider" \
&& f_panic "unable to parse the provider name from $dfile."
u_provider=$(printf %s "$provider" | $TR a-z A-Z | $TR -s _)
cat $dfile | $EGREP "^ *probe +[a-zA-Z_]+ *\(.*\);" | \
while read line; do
# Extract the probe name.
name=$(printf %s "$line" \
| $SED -E -e 's/^ *probe +([a-zA-Z_]+).*/\1/')
u_name=$(printf %s "$name" | $TR a-z A-Z | $TR -s _)
# Generate an arg1,arg2,...,argN line for the probe.
args=""; nargs=0; aline=$(printf %s "$line" | $SED -e 's/.*(\(.*\)).*/\1/')
set -f; IFS=,
for arg in $aline; do
args="${args}arg${nargs},"
nargs=$((nargs + 1))
done
set +f; unset IFS
args=${args%,}
echo "#if _DTRACE_VERSION"
echo ""
# Emit the macros for the probe.
echo "#define ${u_provider}_${u_name}($args) \\"
echo " __dtrace_${provider}___${name}($args)"
echo "#define ${u_provider}_${u_name}_ENABLED() \\"
echo " __dtraceenabled_${provider}___${name}()"
# Emit the extern definitions for the probe dummy
# functions.
echo ""
printf %s\\n "$line" \
| $SED -E -e "s/^ *probe +/extern void __dtrace_${provider}___/"
echo "extern int __dtraceenabled_${provider}___${name}(void);"
printf "\n#else\n"
# Emit empty macros for the probe
echo "#define ${u_provider}_${u_name}($args)"
echo "#define ${u_provider}_${u_name}_ENABLED() (0)"
printf "\n#endif /* _DTRACE_VERSION */\n"
done
printf "#ifdef __cplusplus\n}\n#endif\n\n"
printf "#endif /* _${guard}_H */\n"
}
### Main program.
# Process command line arguments.
test "$#" -eq "0" && f_usage
genelf=0
genheader=0
objbits=64
ofile=
dfile=
while getopts VG3264hs:o: name; do
case $name in
V) f_version;;
s) dfile="$OPTARG";
test -f "$dfile" || f_panic "cannot read $dfile";;
o) ofile="$OPTARG";;
G) genelf=1;;
h) genheader=1;;
# Note the trick to support -32
3) objbits=666;;
2) test "$objbits" -eq 666 || f_usage; objbits=32;;
# Likewise for -64
6) objbits=777;;
4) test "$objbits" -eq 777 || f_usage; objbits=64;;
?) f_usage;;
esac
done
shift $(($OPTIND - 1))
test "$objbits" -eq "32" || test "$objbits" -eq "64" \
|| f_usage
test $((genelf + genheader)) -gt 1 && \
{ echo "Please use either -G or -h."; f_usage; }
test -n "$dfile" || { echo "Please specify a .d file with -s."; exit 2; }
if test "$genelf" -gt 0; then
# In this mode there must be a remaining argument: the name of the
# object file to inspect for probed points.
test "$#" -ne "1" && f_usage
test -f "$1" || f_panic "cannot read $1"
objfile=$1
# Collect probe information from the input object file and the
# d-script.
f_collect_probes $objfile
f_collect_probes_args $dfile
# Generate the assembly code and assemble the DOF program in
# OFILE. Then patch OBJFILE to remove the dummy probe calls.
f_gen_dof_program
f_patch_objfile $objfile
fi
if test "$genheader" -gt 0; then
test -n "$ofile" || { echo "Please specify an output file with -o."; exit 2; }
# In this mode no extra arguments shall be present.
test "$#" -ne "0" && f_usage
f_gen_header_file > $ofile
fi
# pdtrace ends here.