NAME
procmap
—
display process memory map
SYNOPSIS
procmap |
[-AadlmPsv ] [-D
number] [-M
core] [-N
system] [-p
pid] [pid ...] |
DESCRIPTION
The procmap
utility lists the virtual
memory mappings underlying the given process. The start address of each
entry is always given and, depending on the options given, other information
such as the end address, the underlying file's device and inode numbers, and
various protection information will be displayed, along with the path to the
file, if such data is available.
procmap
requires the ability to open
/dev/kmem which may be restricted based upon the
value of the kern.allowkmem
sysctl(8).
By default, procmap
displays information
for its parent process, so that when run from a shell prompt, the shell's
memory information is displayed. If other PIDs are given as arguments on the
command line, information for those processes will be printed also. If the
special PID of 0 is given, then information for the kernel's memory map is
printed.
The options are as follows:
-A
- Print more detailed information on anonymous map usage.
-a
- Display “all” information from the process's memory map. This output mode is an amalgam of the contents of the Solaris, Linux, and OpenBSD style output modes.
-D
number- Enable various debug facilities. The number is a bit
mask of the values:
1
- dump the process's vmspace structure
2
- dump the process's vm_map structure
4
- dump the vm_map.header structure
8
- dump each vm_map_entry in its entirety
16
- dump the namei cache as it is traversed
-d
- Dumps the vm_map and vm_map_entry structures in a style similar to that of
ddb(4). When combined with the
-v
option, the device number, inode number, name, vnode addresses, or other identifying information from the vm_map_entry fields will be printed. -l
- Dumps information in a format like the contents of the maps pseudo-file
under the /proc file system which was, in turn,
modeled after the similarly named entry in the Linux
/proc file system. When combined with the
-v
option, identifiers for all entries are printed. -M
core- Extract values associated with the name list from the specified core instead of the default /dev/kmem.
-m
- Dumps information in the same format as the map pseudo-file of the
/proc file system. When the
-v
option is also given, device number, inode number, and filename or other identifying information is printed. -N
system- Extract the name list from the specified system instead of the running kernel.
-P
- Causes
procmap
to print information about itself. -p
pid- Tells
procmap
to print information about the given process. If-p
pid occurs last on the command line, the-p
is optional. -s
- The Solaris style output format, modeled after the Solaris command “pmap”. This is the default output style.
-v
- Verbose output. When used with
-d
,-l
, or-m
, more information is printed, possibly including device and inode numbers, file path names, or other identifying information. If specified more than once, a ‘*’ will be printed in between two entries that are not adjacent, making the visual identification of spaces in the process's map easier to see.
The -P
and -p
options override each other, so the last one to appear on the command line
takes effect. If you do wish to see information about
procmap
and another process as the same time, simply
omit the -p
and place the extra PID at the end of
the command line.
EXIT STATUS
The procmap
utility exits 0 on
success, and >0 if an error occurs.
EXAMPLES
While the meaning most of the output is self-evident, some pieces of it may appear to be a little inscrutable.
Here a portion of the default output from
procmap
being run at a
sh(1)
prompt shows the starting address of the map entry, the size of the map
entry, the current protection level of the map entry, and either the name of
the file backing the entry or some other descriptive text.
$ procmap 08048000 420K read/exec /bin/sh 080B1000 8K read/write /bin/sh 080B3000 28K read/write [ anon ] 080BA000 16K read/write/exec [ heap ] ...
When the ddb(4) output style is selected, the first thing printed is the contents of the vm_map structure, followed by the individual map entries.
$ procmap -d MAP 0xcf7cac84: [0x0->0xbfbfe000] #ent=8, sz=34041856, ref=1, version=20, flags=0x21 pmap=0xcf44cee0(resident=<unknown>) - 0xcfa3a358: 0x8048000->0x80b1000: obj=0xcf45a8e8/0x0, amap=0x0/0 submap=F, cow=T, nc=T, prot(max)=5/7, inh=1, wc=0, adv=0 ...
The value of the flags field (in hexadecimal) is taken from the
include file
<uvm/uvm_map.h>
:
VM_MAP_PAGEABLE 0x01 ro: entries are pageable VM_MAP_INTRSAFE 0x02 ro: interrupt safe map VM_MAP_WIREFUTURE 0x04 rw: wire future mappings VM_MAP_BUSY 0x08 rw: map is busy VM_MAP_WANTLOCK 0x10 rw: want to write-lock
The “submap”, “cow”, and
“nc” fields are true or false, and indicate whether the map is
a submap, whether it is marked for copy on write, and whether it needs a
copy. The “prot” (or protection) field, along with
“max” (maximum protection allowed) are made up of the
following flags from
<uvm/uvm_extern.h>
:
PROT_READ 0x01 read allowed PROT_WRITE 0x02 write allowed PROT_EXEC 0x04 execute allowed
The “obj” and “amap” fields are pointers to, and offsets into, the underlying uvm_object or vm_amap object. The value for resident is always unknown because digging such information out of the kernel is beyond the scope of this application.
The two output styles that mirror the contents of the /proc file system appear as follows:
$ procmap -m 0x8048000 0x80b1000 r-x rwx COW NC 1 0 0 0x80b1000 0x80b3000 rw- rwx COW NC 1 0 0 0x80b3000 0x80ba000 rw- rwx COW NNC 1 0 0 0x80ba000 0x80be000 rwx rwx COW NNC 1 0 0 ... $ procmap -l 08048000-080b1000 r-xp 00000000 00:00 70173 /bin/sh 080b1000-080b3000 rw-p 00068000 00:00 70173 /bin/sh 080b3000-080ba000 rw-p 00000000 00:00 0 080ba000-080be000 rwxp 00000000 00:00 0 ...
Here the protection and maximum protection values are indicated with ‘r’, ‘w’, and ‘x’ characters, indicating read permission, write permission, and execute permission, respectively. The “COW”, “NC”, and “NNC” values that follow indicate, again, that the map is marked for copy on write and either needs or does not need a copy. It is also possible to see the value “NCOW” here, which indicates that an entry will not be copied. The three following numbers indicate the inheritance type of the map, the wired count of the map, and any advice value assigned via madvise(2).
In the second form, the permissions indicated are followed by a ‘p’ or ‘s’ character indicating whether the map entry is private or shared (copy on write or not), and the numbers are the offset into the underlying object, the device and numbers of the object if it is a file, and the path to the file (if available).
As noted above (see section DESCRIPTION), the “all” output format is an amalgam of the previous output formats.
$ procmap -a Start End Size Offset rwxpc RWX I/W/A ... 08048000-080b0fff 420k 00000000 r-xp+ (rwx) 1/0/0 ... ...
In this format, the column labeled “rwxpc” contains the permissions for the mapping along with the shared/private flag, and a character indicating whether the mapping needs to be copied on write (‘+’) or has already been copied (‘-’) and is followed by a column that indicates the maximum permissions for the map entry. The column labeled “I/W/A” indicates the inheritance, wired, and advice values for the map entry, as previously described.
SEE ALSO
ls(1), madvise(2), mmap(2), kvm(3), ddb(4), namei(9), vnode(9)
HISTORY
The procmap
utility first appeared in
OpenBSD 3.5. It was derived from the
NetBSD utility known as “pmap”.
AUTHORS
The procmap
utility and documentation was
written by Andrew Brown
<atatat@netbsd.org>.
BUGS
Very little will work unless procmap
is
reading from the correct kernel in order to retrieve the proper symbol
information.
Since processes can change state while
procmap
is running, some of the information printed
may be inaccurate. This is especially important to consider when examining
the kernel's map, since merely executing procmap
will cause some of the information to change.
The pathnames to files backing certain vnodes (such as the text and data sections of programs and shared libraries) are extracted from the kernel's namei cache which is considerably volatile. If a path is not found there in its entirety, as much information as was available will be printed. In most cases, simply running ls(1) with the expected path to the file will cause the information to be reentered into the cache.
The Solaris version (“pmap”) has some interesting
command line flags that would be nice to emulate here. In particular, the
-r
option that lists a process's reserved addresses,
and the -x
option that prints
resident/shared/private mapping details for each entry.
Some of the output modes can be or are wider than the standard 80 columns of a terminal. Some sort of formatting might be nice.