util-linux/misc-utils/lsfd.c
Karel Zak 96de80b659 Merge branch 'lsfd--xmode' of https://github.com/masatake/util-linux
* 'lsfd--xmode' of https://github.com/masatake/util-linux:
  lsfd: add 'D' flag for representing deleted files to XMODE column
  lsfd: introduce XMODE column, extensible variant of MODE
  test: (lsfd) add a subcase for testing NAME column for a deleted file
  test: (lsfd) add a case for testing DELETED column
  test: (mkfds) add "make-regular-file" factory
2023-07-03 12:41:04 +02:00

2100 lines
53 KiB
C

/*
* lsfd(1) - list file descriptors
*
* Copyright (C) 2021 Red Hat, Inc. All rights reserved.
* Written by Masatake YAMATO <yamato@redhat.com>
* Karel Zak <kzak@redhat.com>
*
* Very generally based on lsof(8) by Victor A. Abell <abe@purdue.edu>
* It supports multiple OSes. lsfd specializes to Linux.
*
* 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 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it would 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, write to the Free Software Foundation,
* Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include <stdio.h>
#include <sys/types.h>
#include <inttypes.h>
#include <sys/stat.h>
#include <unistd.h>
#include <getopt.h>
#include <ctype.h>
#include <search.h>
#include <linux/sched.h>
#include <sys/syscall.h>
#include <linux/kcmp.h>
static int kcmp(pid_t pid1, pid_t pid2, int type,
unsigned long idx1, unsigned long idx2)
{
return syscall(SYS_kcmp, pid1, pid2, type, idx1, idx2);
}
/* See proc(5).
* Defined in linux/include/linux/sched.h private header file. */
#define PF_KTHREAD 0x00200000 /* I am a kernel thread */
#include "c.h"
#include "nls.h"
#include "xalloc.h"
#include "list.h"
#include "closestream.h"
#include "strutils.h"
#include "procfs.h"
#include "fileutils.h"
#include "idcache.h"
#include "pathnames.h"
#include "libsmartcols.h"
#include "lsfd.h"
#include "lsfd-filter.h"
#include "lsfd-counter.h"
/*
* /proc/$pid/mountinfo entries
*/
struct nodev {
struct list_head nodevs;
unsigned long minor;
char *filesystem;
};
struct nodev_table {
#define NODEV_TABLE_SIZE 97
struct list_head tables[NODEV_TABLE_SIZE];
};
static struct nodev_table nodev_table;
struct name_manager {
struct idcache *cache;
unsigned long next_id;
};
/*
* /proc/devices entries
*/
struct devdrv {
struct list_head devdrvs;
unsigned long major;
char *name;
};
static struct list_head chrdrvs;
static struct list_head blkdrvs;
/*
* IPC table
*/
#define IPC_TABLE_SIZE 997
struct ipc_table {
struct list_head tables[IPC_TABLE_SIZE];
};
static struct ipc_table ipc_table;
/*
* Column related stuffs
*/
/* column names */
struct colinfo {
const char *name;
double whint;
int flags;
int json_type;
const char *help;
};
/* columns descriptions */
static const struct colinfo infos[] = {
[COL_AINODECLASS] = { "AINODECLASS",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("class of anonymous inode") },
[COL_ASSOC] = { "ASSOC",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("association between file and process") },
[COL_BLKDRV] = { "BLKDRV",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("block device driver name resolved by /proc/devices") },
[COL_CHRDRV] = { "CHRDRV",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("character device driver name resolved by /proc/devices") },
[COL_COMMAND] = { "COMMAND",
0.3, SCOLS_FL_TRUNC, SCOLS_JSON_STRING,
N_("command of the process opening the file") },
[COL_DELETED] = { "DELETED",
0, SCOLS_FL_RIGHT, SCOLS_JSON_BOOLEAN,
N_("reachability from the file system") },
[COL_DEV] = { "DEV",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("ID of device containing file") },
[COL_DEVTYPE] = { "DEVTYPE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("device type (blk, char, or nodev)") },
[COL_ENDPOINTS] = { "ENDPOINTS",
0, SCOLS_FL_WRAP, SCOLS_JSON_ARRAY_STRING,
N_("IPC endpoints information communicated with the fd") },
[COL_EVENTFD_ID] = {"EVENTFD.ID",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("eventfd ID") },
[COL_EVENTPOLL_TFDS] = {"EVENTPOLL.TFDS",
0, SCOLS_FL_WRAP, SCOLS_JSON_ARRAY_NUMBER,
N_("file descriptors targeted by the eventpoll file") },
[COL_FD] = { "FD",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("file descriptor for the file") },
[COL_FLAGS] = { "FLAGS",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("flags specified when opening the file") },
[COL_FUID] = { "FUID",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("user ID number of the file's owner") },
[COL_INET_LADDR] = { "INET.LADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("local IP address") },
[COL_INET_RADDR] = { "INET.RADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("remote IP address") },
[COL_INET6_LADDR] = { "INET6.LADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("local IPv6 address") },
[COL_INET6_RADDR] = { "INET6.RADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("remote IPv6 address") },
[COL_INODE] = { "INODE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("inode number") },
[COL_INOTIFY_INODES] = { "INOTIFY.INODES",
0, SCOLS_FL_WRAP, SCOLS_JSON_ARRAY_STRING,
N_("list of monitoring inodes (cooked)") },
[COL_INOTIFY_INODES_RAW]={ "INOTIFY.INODES.RAW",
0, SCOLS_FL_WRAP, SCOLS_JSON_ARRAY_STRING,
N_("list of monitoring inodes (raw, don't decode devices)") },
[COL_KNAME] = { "KNAME",
0.4, SCOLS_FL_TRUNC, SCOLS_JSON_STRING,
N_("name of the file (raw)") },
[COL_KTHREAD] = { "KTHREAD",
0, SCOLS_FL_RIGHT, SCOLS_JSON_BOOLEAN,
N_("opened by a kernel thread") },
[COL_MAJMIN] = { "MAJ:MIN",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("device ID for special, or ID of device containing file") },
[COL_MAPLEN] = { "MAPLEN",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("length of file mapping (in page)") },
[COL_MISCDEV] = { "MISCDEV",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("misc character device name resolved by /proc/misc") },
[COL_MNT_ID] = { "MNTID",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("mount id") },
[COL_MODE] = { "MODE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("access mode (rwx)") },
[COL_NAME] = { "NAME",
0.4, SCOLS_FL_TRUNC, SCOLS_JSON_STRING,
N_("name of the file (cooked)") },
[COL_NETLINK_GROUPS] = { "NETLINK.GROUPS",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("netlink multicast groups") },
[COL_NETLINK_LPORT] = { "NETLINK.LPORT",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("netlink local port id") },
[COL_NETLINK_PROTOCOL] = { "NETLINK.PROTOCOL",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("netlink protocol") },
[COL_NLINK] = { "NLINK",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("link count") },
[COL_NS_NAME] = { "NS.NAME",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("name of the namespace (NS.TYPE:[INODE])") },
[COL_NS_TYPE] = { "NS.TYPE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("type of the namespace") },
[COL_OWNER] = { "OWNER",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("owner of the file") },
[COL_PACKET_IFACE] = { "PACKET.IFACE",
0, SCOLS_FL_RIGHT,SCOLS_JSON_STRING,
N_("net interface associated with the packet socket") },
[COL_PACKET_PROTOCOL] = { "PACKET.PROTOCOL",
0, SCOLS_FL_RIGHT,SCOLS_JSON_STRING,
N_("L3 protocol associated with the packet socket") },
[COL_PARTITION] = { "PARTITION",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("block device name resolved by /proc/partition") },
[COL_PID] = { "PID",
5, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("PID of the process opening the file") },
[COL_PIDFD_COMM] = { "PIDFD.COMM",
0.2, SCOLS_FL_TRUNC, SCOLS_JSON_STRING,
N_("command of the process targeted by the pidfd") },
[COL_PIDFD_NSPID] = { "PIDFD.NSPID",
0.2, SCOLS_FL_TRUNC, SCOLS_JSON_STRING,
N_("NSpid field in fdinfo of the pidfd") },
[COL_PIDFD_PID] = { "PIDFD.PID",
5, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("PID of the process targeted by the pidfd") },
[COL_PING_ID] = { "PING.ID",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("ICMP echo request ID") },
[COL_POS] = { "POS",
5, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("file position") },
[COL_RAW_PROTOCOL] = { "RAW.PROTOCOL",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("protocol number of the raw socket") },
[COL_RDEV] = { "RDEV",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("device ID (if special file)") },
[COL_SIGNALFD_MASK] = { "SIGNALFD.MASK",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("masked signals") },
[COL_SIZE] = { "SIZE",
4, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("file size"), },
[COL_SOCK_LISTENING] = { "SOCK.LISTENING",
0, SCOLS_FL_RIGHT, SCOLS_JSON_BOOLEAN,
N_("listening socket") },
[COL_SOCK_NETNS] = { "SOCK.NETNS",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("inode identifying network namespace where the socket belongs to") },
[COL_SOCK_PROTONAME] = { "SOCK.PROTONAME",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("protocol name") },
[COL_SOCK_STATE] = { "SOCK.STATE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("State of socket") },
[COL_SOCK_TYPE] = { "SOCK.TYPE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("Type of socket") },
[COL_SOURCE] = { "SOURCE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("file system, partition, or device containing file") },
[COL_STTYPE] = { "STTYPE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("file type (raw)") },
[COL_TCP_LADDR] = { "TCP.LADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("local TCP address (INET address:TCP port)") },
[COL_TCP_RADDR] = { "TCP.RADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("remote TCP address (INET address:TCP port)") },
[COL_TCP_LPORT] = { "TCP.LPORT",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("local TCP port") },
[COL_TCP_RPORT] = { "TCP.RPORT",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("remote TCP port") },
[COL_TID] = { "TID",
5, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("thread ID of the process opening the file") },
[COL_TIMERFD_CLOCKID] = { "TIMERFD.CLOCKID",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("clockid") },
[COL_TIMERFD_INTERVAL] = { "TIMERFD.INTERVAL",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("interval") },
[COL_TIMERFD_REMAINING]= { "TIMERFD.REMAINING",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("remaining time") },
[COL_TUN_IFACE] = { "TUN.IFACE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("network interface behind the tun device") },
[COL_TYPE] = { "TYPE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("file type (cooked)") },
[COL_UDP_LADDR] = { "UDP.LADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("local UDP address (INET address:UDP port)") },
[COL_UDP_RADDR] = { "UDP.RADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("remote UDP address (INET address:UDP port)") },
[COL_UDP_LPORT] = { "UDP.LPORT",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("local UDP port") },
[COL_UDP_RPORT] = { "UDP.RPORT",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("remote UDP port") },
[COL_UDPLITE_LADDR] = { "UDPLITE.LADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("local UDPLite address (INET address:UDPLite port)") },
[COL_UDPLITE_RADDR] = { "UDPLITE.RADDR",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("remote UDPLite address (INET address:UDPLite port)") },
[COL_UDPLITE_LPORT] = { "UDPLITE.LPORT",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("local UDPLite port") },
[COL_UDPLITE_RPORT] = { "UDPLITE.RPORT",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("remote UDPLite port") },
[COL_UID] = { "UID",
0, SCOLS_FL_RIGHT, SCOLS_JSON_NUMBER,
N_("user ID number of the process") },
[COL_UNIX_PATH] = { "UNIX.PATH",
0.4, SCOLS_FL_TRUNC, SCOLS_JSON_STRING,
N_("filesystem pathname for UNIX domain socket") },
[COL_USER] = { "USER",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("user of the process") },
[COL_XMODE] = { "XMODE",
0, SCOLS_FL_RIGHT, SCOLS_JSON_STRING,
N_("access mode and more (rwx)") },
};
static const int default_columns[] = {
COL_COMMAND,
COL_PID,
COL_USER,
COL_ASSOC,
COL_XMODE,
COL_TYPE,
COL_SOURCE,
COL_MNT_ID,
COL_INODE,
COL_NAME,
};
static const int default_threads_columns[] = {
COL_COMMAND,
COL_PID,
COL_TID,
COL_USER,
COL_ASSOC,
COL_XMODE,
COL_TYPE,
COL_SOURCE,
COL_MNT_ID,
COL_INODE,
COL_NAME,
};
static int columns[ARRAY_SIZE(infos) * 2] = {-1};
static size_t ncolumns;
static ino_t *mnt_namespaces;
static size_t nspaces;
struct counter_spec {
struct list_head specs;
const char *name;
const char *expr;
};
static const struct counter_spec default_counter_specs[] = {
{
.name = N_("processes"),
.expr = "ASSOC == 'cwd'",
},
{
.name = N_("root owned processes"),
.expr = "(ASSOC == 'cwd') && (UID == 0)",
},
{
.name = N_("kernel threads"),
.expr = "(ASSOC == 'cwd') && KTHREAD",
},
{
.name = N_("open files"),
.expr = "FD >= 0",
},
{
.name = N_("RO open files"),
.expr = "(FD >= 0) and (MODE == 'r--')",
},
{
.name = N_("WO open files"),
.expr = "(FD >= 0) and (MODE == '-w-')",
},
{
.name = N_("shared mappings"),
.expr = "ASSOC == 'shm'",
},
{
.name = N_("RO shared mappings"),
.expr = "(ASSOC == 'shm') and (MODE == 'r--')",
},
{
.name = N_("WO shared mappings"),
.expr = "(ASSOC == 'shm') and (MODE == '-w-')",
},
{
.name = N_("regular files"),
.expr = "(FD >= 0) && (STTYPE == 'REG')",
},
{
.name = N_("directories"),
.expr = "(FD >= 0) && (STTYPE == 'DIR')",
},
{
.name = N_("sockets"),
.expr = "(FD >= 0) && (STTYPE == 'SOCK')",
},
{
.name = N_("fifos/pipes"),
.expr = "(FD >= 0) && (STTYPE == 'FIFO')",
},
{
.name = N_("character devices"),
.expr = "(FD >= 0) && (STTYPE == 'CHR')",
},
{
.name = N_("block devices"),
.expr = "(FD >= 0) && (STTYPE == 'BLK')",
},
{
.name = N_("unknown types"),
.expr = "(FD >= 0) && (STTYPE == 'UNKN')",
}
};
struct lsfd_control {
struct libscols_table *tb; /* output */
struct list_head procs; /* list of all processes */
unsigned int noheadings : 1,
raw : 1,
json : 1,
notrunc : 1,
threads : 1,
show_main : 1, /* print main table */
show_summary : 1, /* print summary/counters */
sockets_only : 1; /* display only SOCKETS */
struct lsfd_filter *filter;
struct lsfd_counter **counters; /* NULL terminated array. */
};
static void *proc_tree; /* for tsearch/tfind */
static int proc_tree_compare(const void *a, const void *b)
{
return ((struct proc *)a)->pid - ((struct proc *)b)->pid;
}
struct proc *get_proc(pid_t pid)
{
struct proc **node = tfind(&pid, &proc_tree, proc_tree_compare);
if (node)
return *node;
return NULL;
}
static int column_name_to_id(const char *name, size_t namesz)
{
size_t i;
for (i = 0; i < ARRAY_SIZE(infos); i++) {
const char *cn = infos[i].name;
if (!strncasecmp(name, cn, namesz) && !*(cn + namesz))
return i;
}
warnx(_("unknown column: %s"), name);
return LSFD_FILTER_UNKNOWN_COL_ID;
}
static int column_name_to_id_cb(const char *name, void *data __attribute__((__unused__)))
{
return column_name_to_id(name, strlen(name));
}
static int get_column_id(int num)
{
assert(num >= 0);
assert((size_t) num < ncolumns);
assert(columns[num] < (int) ARRAY_SIZE(infos));
return columns[num];
}
static const struct colinfo *get_column_info(int num)
{
return &infos[ get_column_id(num) ];
}
static struct libscols_column *add_column(struct libscols_table *tb, const struct colinfo *col)
{
struct libscols_column *cl;
int flags = col->flags;
cl = scols_table_new_column(tb, col->name, col->whint, flags);
if (cl) {
scols_column_set_json_type(cl, col->json_type);
if (col->flags & SCOLS_FL_WRAP) {
scols_column_set_wrapfunc(cl,
scols_wrapnl_chunksize,
scols_wrapnl_nextchunk,
NULL);
scols_column_set_safechars(cl, "\n");
}
}
return cl;
}
static struct libscols_column *add_column_by_id_cb(struct libscols_table *tb, int colid, void *data)
{
struct libscols_column *cl;
if (ncolumns >= ARRAY_SIZE(columns))
errx(EXIT_FAILURE, _("too many columns are added via filter expression"));
assert(colid < LSFD_N_COLS);
cl = add_column(tb, infos + colid);
if (!cl)
err(EXIT_FAILURE, _("failed to allocate output column"));
columns[ncolumns++] = colid;
if (colid == COL_TID) {
struct lsfd_control *ctl = data;
ctl->threads = 1;
}
return cl;
}
static int has_mnt_ns(ino_t id)
{
size_t i;
for (i = 0; i < nspaces; i++) {
if (mnt_namespaces[i] == id)
return 1;
}
return 0;
}
static void add_mnt_ns(ino_t id)
{
size_t nmax = 0;
if (nspaces)
nmax = (nspaces + 16) / 16 * 16;
if (nmax <= nspaces + 1) {
nmax += 16;
mnt_namespaces = xrealloc(mnt_namespaces,
sizeof(ino_t) * nmax);
}
mnt_namespaces[nspaces++] = id;
}
static const struct file_class *stat2class(struct stat *sb)
{
dev_t dev;
assert(sb);
switch (sb->st_mode & S_IFMT) {
case S_IFCHR:
return &cdev_class;
case S_IFBLK:
return &bdev_class;
case S_IFSOCK:
return &sock_class;
case S_IFIFO:
return &fifo_class;
case S_IFLNK:
case S_IFDIR:
return &file_class;
case S_IFREG:
dev = sb->st_dev;
if (major(dev) != 0)
return &file_class;
if (is_nsfs_dev(dev))
return &nsfs_file_class;
if (is_mqueue_dev(dev))
return &mqueue_file_class;
return &file_class;
default:
break;
}
return &unkn_class;
}
static struct file *new_file(struct proc *proc, const struct file_class *class)
{
struct file *file;
assert(class);
file = xcalloc(1, class->size);
file->class = class;
file->proc = proc;
INIT_LIST_HEAD(&file->files);
list_add_tail(&file->files, &proc->files);
return file;
}
static struct file *copy_file(struct file *old)
{
struct file *file = xcalloc(1, old->class->size);
INIT_LIST_HEAD(&file->files);
file->proc = old->proc;
list_add_tail(&file->files, &old->proc->files);
file->class = old->class;
file->association = old->association;
file->name = xstrdup(old->name);
file->stat = old->stat;
return file;
}
static void file_set_path(struct file *file, struct stat *sb, const char *name, int association)
{
file->association = association;
file->name = xstrdup(name);
file->stat = *sb;
}
static void file_init_content(struct file *file)
{
if (file->class && file->class->initialize_content)
file->class->initialize_content(file);
}
static void free_file(struct file *file)
{
const struct file_class *class = file->class;
while (class) {
if (class->free_content)
class->free_content(file);
class = class->super;
}
free(file);
}
static struct proc *new_process(pid_t pid, struct proc *leader)
{
struct proc *proc = xcalloc(1, sizeof(*proc));
proc->pid = pid;
proc->leader = leader? leader: proc;
proc->command = NULL;
INIT_LIST_HEAD(&proc->files);
INIT_LIST_HEAD(&proc->procs);
proc->kthread = 0;
return proc;
}
static void free_proc(struct proc *proc)
{
list_free(&proc->files, struct file, files, free_file);
free(proc->command);
free(proc);
}
static void read_fdinfo(struct file *file, FILE *fdinfo)
{
char buf[1024];
while (fgets(buf, sizeof(buf), fdinfo)) {
const struct file_class *class;
char *val = strchr(buf, ':');
if (!val)
continue;
*val++ = '\0'; /* terminate key */
val = (char *) skip_space(val);
rtrim_whitespace((unsigned char *) val);
class = file->class;
while (class) {
if (class->handle_fdinfo
&& class->handle_fdinfo(file, buf, val))
break;
class = class->super;
}
}
}
static struct file *collect_file_symlink(struct path_cxt *pc,
struct proc *proc,
const char *name,
int assoc,
bool sockets_only)
{
char sym[PATH_MAX] = { '\0' };
struct stat sb;
struct file *f, *prev;
if (ul_path_readlink(pc, sym, sizeof(sym), name) < 0)
return NULL;
/* The /proc/#/{fd,ns} often contains the same file (e.g. /dev/tty)
* more than once. Let's try to reuse the previous file if the real
* path is the same to save stat() call.
*/
prev = list_last_entry(&proc->files, struct file, files);
if (prev && prev->name && strcmp(prev->name, sym) == 0) {
f = copy_file(prev);
f->association = assoc;
} else {
const struct file_class *class;
if (ul_path_stat(pc, &sb, 0, name) < 0)
return NULL;
class = stat2class(&sb);
if (sockets_only
/* A nsfs file is not a socket but the nsfs file can
* be used as a entry point to collect information from
* other network namespaces. Besed on the information,
* various columns of sockets can be filled.
*/
&& (class != &sock_class) && (class != &nsfs_file_class))
return NULL;
f = new_file(proc, class);
file_set_path(f, &sb, sym, assoc);
}
file_init_content(f);
if (is_association(f, NS_MNT))
proc->ns_mnt = f->stat.st_ino;
else if (is_association(f, NS_NET))
load_sock_xinfo(pc, name, f->stat.st_ino);
else if (assoc >= 0) {
/* file-descriptor based association */
FILE *fdinfo;
if (ul_path_stat(pc, &sb, AT_SYMLINK_NOFOLLOW, name) == 0)
f->mode = sb.st_mode;
if (is_nsfs_dev(f->stat.st_dev))
load_sock_xinfo(pc, name, f->stat.st_ino);
fdinfo = ul_path_fopenf(pc, "r", "fdinfo/%d", assoc);
if (fdinfo) {
read_fdinfo(f, fdinfo);
fclose(fdinfo);
}
}
return f;
}
/* read symlinks from /proc/#/fd
*/
static void collect_fd_files(struct path_cxt *pc, struct proc *proc,
bool sockets_only)
{
DIR *sub = NULL;
struct dirent *d = NULL;
char path[sizeof("fd/") + sizeof(stringify_value(UINT64_MAX))];
while (ul_path_next_dirent(pc, &sub, "fd", &d) == 0) {
uint64_t num;
if (ul_strtou64(d->d_name, &num, 10) != 0) /* only numbers */
continue;
snprintf(path, sizeof(path), "fd/%ju", (uintmax_t) num);
collect_file_symlink(pc, proc, path, num, sockets_only);
}
}
static void parse_maps_line(struct path_cxt *pc, char *buf, struct proc *proc)
{
uint64_t start, end, offset, ino;
unsigned long major, minor;
enum association assoc = ASSOC_MEM;
struct stat sb;
struct file *f, *prev;
char *path, modestr[5];
dev_t devno;
/* read rest of the map */
if (sscanf(buf, "%"SCNx64 /* start */
"-%"SCNx64 /* end */
" %4[^ ]" /* mode */
" %"SCNx64 /* offset */
" %lx:%lx" /* maj:min */
" %"SCNu64, /* inode */
&start, &end, modestr, &offset,
&major, &minor, &ino) != 7)
return;
/* Skip private anonymous mappings. */
if (major == 0 && minor == 0 && ino == 0)
return;
devno = makedev(major, minor);
if (modestr[3] == 's')
assoc = ASSOC_SHM;
/* The map usually contains the same file more than once, try to reuse
* the previous file (if devno and ino are the same) to save stat() call.
*/
prev = list_last_entry(&proc->files, struct file, files);
if (prev && prev->stat.st_dev == devno && prev->stat.st_ino == ino) {
f = copy_file(prev);
f->association = -assoc;
} else if ((path = strchr(buf, '/'))) {
rtrim_whitespace((unsigned char *) path);
if (stat(path, &sb) < 0)
/* If a file is mapped but deleted from the file system,
* "stat by the file name" may not work. In that case,
*/
goto try_map_files;
f = new_file(proc, stat2class(&sb));
file_set_path(f, &sb, path, -assoc);
} else {
/* As used in tcpdump, AF_PACKET socket can be mmap'ed. */
char map_file[sizeof("map_files/0000000000000000-ffffffffffffffff")];
char sym[PATH_MAX] = { '\0' };
try_map_files:
snprintf(map_file, sizeof(map_file), "map_files/%"PRIx64"-%"PRIx64, start, end);
if (ul_path_stat(pc, &sb, 0, map_file) < 0)
return;
if (ul_path_readlink(pc, sym, sizeof(sym), map_file) < 0)
return;
f = new_file(proc, stat2class(&sb));
file_set_path(f, &sb, sym, -assoc);
}
if (modestr[0] == 'r')
f->mode |= S_IRUSR;
if (modestr[1] == 'w')
f->mode |= S_IWUSR;
if (modestr[2] == 'x')
f->mode |= S_IXUSR;
f->map_start = start;
f->map_end = end;
f->pos = offset;
file_init_content(f);
}
static void collect_mem_files(struct path_cxt *pc, struct proc *proc)
{
FILE *fp;
char buf[BUFSIZ];
fp = ul_path_fopen(pc, "r", "maps");
if (!fp)
return;
while (fgets(buf, sizeof(buf), fp))
parse_maps_line(pc, buf, proc);
fclose(fp);
}
static void collect_outofbox_files(struct path_cxt *pc,
struct proc *proc,
enum association assocs[],
const char *names[],
size_t count,
bool sockets_only)
{
size_t i;
for (i = 0; i < count; i++)
collect_file_symlink(pc, proc, names[assocs[i]], assocs[i] * -1,
sockets_only);
}
static void collect_execve_file(struct path_cxt *pc, struct proc *proc,
bool sockets_only)
{
enum association assocs[] = { ASSOC_EXE };
const char *names[] = {
[ASSOC_EXE] = "exe",
};
collect_outofbox_files(pc, proc, assocs, names, ARRAY_SIZE(assocs),
sockets_only);
}
static void collect_fs_files(struct path_cxt *pc, struct proc *proc,
bool sockets_only)
{
enum association assocs[] = { ASSOC_EXE, ASSOC_CWD, ASSOC_ROOT };
const char *names[] = {
[ASSOC_CWD] = "cwd",
[ASSOC_ROOT] = "root",
};
collect_outofbox_files(pc, proc, assocs, names, ARRAY_SIZE(assocs),
sockets_only);
}
static void collect_namespace_files(struct path_cxt *pc, struct proc *proc)
{
enum association assocs[] = {
ASSOC_NS_CGROUP,
ASSOC_NS_IPC,
ASSOC_NS_MNT,
ASSOC_NS_NET,
ASSOC_NS_PID,
ASSOC_NS_PID4C,
ASSOC_NS_TIME,
ASSOC_NS_TIME4C,
ASSOC_NS_USER,
ASSOC_NS_UTS,
};
const char *names[] = {
[ASSOC_NS_CGROUP] = "ns/cgroup",
[ASSOC_NS_IPC] = "ns/ipc",
[ASSOC_NS_MNT] = "ns/mnt",
[ASSOC_NS_NET] = "ns/net",
[ASSOC_NS_PID] = "ns/pid",
[ASSOC_NS_PID4C] = "ns/pid_for_children",
[ASSOC_NS_TIME] = "ns/time",
[ASSOC_NS_TIME4C] = "ns/time_for_children",
[ASSOC_NS_USER] = "ns/user",
[ASSOC_NS_UTS] = "ns/uts",
};
collect_outofbox_files(pc, proc, assocs, names, ARRAY_SIZE(assocs),
/* Namespace information is alwasys needed. */
false);
}
static struct nodev *new_nodev(unsigned long minor, const char *filesystem)
{
struct nodev *nodev = xcalloc(1, sizeof(*nodev));
INIT_LIST_HEAD(&nodev->nodevs);
nodev->minor = minor;
nodev->filesystem = xstrdup(filesystem);
return nodev;
}
static void free_nodev(struct nodev *nodev)
{
free(nodev->filesystem);
free(nodev);
}
void add_nodev(unsigned long minor, const char *filesystem)
{
struct nodev *nodev = new_nodev(minor, filesystem);
unsigned long slot = nodev->minor % NODEV_TABLE_SIZE;
list_add_tail(&nodev->nodevs, &nodev_table.tables[slot]);
}
static void initialize_nodevs(void)
{
int i;
for (i = 0; i < NODEV_TABLE_SIZE; i++)
INIT_LIST_HEAD(&nodev_table.tables[i]);
}
static void finalize_nodevs(void)
{
int i;
for (i = 0; i < NODEV_TABLE_SIZE; i++)
list_free(&nodev_table.tables[i], struct nodev, nodevs, free_nodev);
free(mnt_namespaces);
}
const char *get_nodev_filesystem(unsigned long minor)
{
struct list_head *n;
int slot = minor % NODEV_TABLE_SIZE;
list_for_each (n, &nodev_table.tables[slot]) {
struct nodev *nodev = list_entry(n, struct nodev, nodevs);
if (nodev->minor == minor)
return nodev->filesystem;
}
return NULL;
}
static void add_nodevs(FILE *mnt)
{
/* This can be very long. A line in mountinfo can have more than 3
* paths. */
char line[PATH_MAX * 3 + 256];
while (fgets(line, sizeof(line), mnt)) {
unsigned long major, minor;
char filesystem[256];
/* 23 61 0:22 / /sys rw,nosuid,nodev,noexec,relatime shared:2 - sysfs sysfs rw,seclabel */
if(sscanf(line, "%*d %*d %lu:%lu %*s %*s %*s %*[^-] - %s %*[^\n]",
&major, &minor, filesystem) != 3)
/* 1600 1458 0:55 / / rw,nodev,relatime - overlay overlay rw,context="s... */
if (sscanf(line, "%*d %*d %lu:%lu %*s %*s %*s - %s %*[^\n]",
&major, &minor, filesystem) != 3)
continue;
if (major != 0)
continue;
if (get_nodev_filesystem(minor))
continue;
add_nodev(minor, filesystem);
}
}
static void initialize_ipc_table(void)
{
for (int i = 0; i < IPC_TABLE_SIZE; i++)
INIT_LIST_HEAD(ipc_table.tables + i);
}
static void free_ipc(struct ipc *ipc)
{
if (ipc->class->free)
ipc->class->free(ipc);
free(ipc);
}
static void finalize_ipc_table(void)
{
for (int i = 0; i < IPC_TABLE_SIZE; i++)
list_free(&ipc_table.tables[i], struct ipc, ipcs, free_ipc);
}
struct ipc *new_ipc(const struct ipc_class *class)
{
struct ipc *ipc = xcalloc(1, class->size);
ipc->class = class;
INIT_LIST_HEAD(&ipc->endpoints);
INIT_LIST_HEAD(&ipc->ipcs);
return ipc;
}
struct ipc *get_ipc(struct file *file)
{
int slot;
struct list_head *e;
const struct ipc_class *ipc_class;
if (!file->class->get_ipc_class)
return NULL;
ipc_class = file->class->get_ipc_class(file);
if (!ipc_class)
return NULL;
slot = ipc_class->get_hash(file) % IPC_TABLE_SIZE;
list_for_each (e, &ipc_table.tables[slot]) {
struct ipc *ipc = list_entry(e, struct ipc, ipcs);
if (ipc->class != ipc_class)
continue;
if (ipc_class->is_suitable_ipc(ipc, file))
return ipc;
}
return NULL;
}
void add_ipc(struct ipc *ipc, unsigned int hash)
{
int slot = hash % IPC_TABLE_SIZE;
list_add(&ipc->ipcs, &ipc_table.tables[slot]);
}
void init_endpoint(struct ipc_endpoint *endpoint)
{
INIT_LIST_HEAD(&endpoint->endpoints);
}
void add_endpoint(struct ipc_endpoint *endpoint, struct ipc *ipc)
{
endpoint->ipc = ipc;
list_add(&endpoint->endpoints, &ipc->endpoints);
}
static void fill_column(struct proc *proc,
struct file *file,
struct libscols_line *ln,
int column_id,
size_t column_index)
{
const struct file_class *class = file->class;
while (class) {
if (class->fill_column
&& class->fill_column(proc, file, ln,
column_id, column_index))
break;
class = class->super;
}
}
static void convert_file(struct proc *proc,
struct file *file,
struct libscols_line *ln)
{
size_t i;
for (i = 0; i < ncolumns; i++)
fill_column(proc, file, ln, get_column_id(i), i);
}
static void convert(struct list_head *procs, struct lsfd_control *ctl)
{
struct list_head *p;
list_for_each (p, procs) {
struct proc *proc = list_entry(p, struct proc, procs);
struct list_head *f;
list_for_each (f, &proc->files) {
struct file *file = list_entry(f, struct file, files);
struct libscols_line *ln = scols_table_new_line(ctl->tb, NULL);
struct lsfd_counter **counter = NULL;
if (!ln)
err(EXIT_FAILURE, _("failed to allocate output line"));
convert_file(proc, file, ln);
if (!lsfd_filter_apply(ctl->filter, ln)) {
scols_table_remove_line(ctl->tb, ln);
continue;
}
if (!ctl->counters)
continue;
for (counter = ctl->counters; *counter; counter++)
lsfd_counter_accumulate(*counter, ln);
}
}
}
static void delete(struct list_head *procs, struct lsfd_control *ctl)
{
struct list_head *p;
list_for_each (p, procs) {
struct proc *proc = list_entry(p, struct proc, procs);
tdelete(proc, &proc_tree, proc_tree_compare);
}
list_free(procs, struct proc, procs, free_proc);
scols_unref_table(ctl->tb);
lsfd_filter_free(ctl->filter);
if (ctl->counters) {
struct lsfd_counter **counter;
for (counter = ctl->counters; *counter; counter++)
lsfd_counter_free(*counter);
free(ctl->counters);
}
}
static void emit(struct lsfd_control *ctl)
{
scols_print_table(ctl->tb);
}
static void initialize_class(const struct file_class *class)
{
if (class->initialize_class)
class->initialize_class();
}
static void initialize_classes(void)
{
initialize_class(&file_class);
initialize_class(&cdev_class);
initialize_class(&bdev_class);
initialize_class(&sock_class);
initialize_class(&unkn_class);
}
static void finalize_class(const struct file_class *class)
{
if (class->finalize_class)
class->finalize_class();
}
static void finalize_classes(void)
{
finalize_class(&file_class);
finalize_class(&cdev_class);
finalize_class(&bdev_class);
finalize_class(&sock_class);
finalize_class(&unkn_class);
}
static struct devdrv *new_devdrv(unsigned long major, const char *name)
{
struct devdrv *devdrv = xcalloc(1, sizeof(*devdrv));
INIT_LIST_HEAD(&devdrv->devdrvs);
devdrv->major = major;
devdrv->name = xstrdup(name);
return devdrv;
}
static void free_devdrv(struct devdrv *devdrv)
{
free(devdrv->name);
free(devdrv);
}
#define READ_DEVICES_LINE_LEN 256
static struct devdrv *read_devdrv(const char *line)
{
unsigned long major;
char name[READ_DEVICES_LINE_LEN];
if (sscanf(line, "%lu %s", &major, name) != 2)
return NULL;
return new_devdrv(major, name);
}
static void read_devices(struct list_head *chrdrvs_list,
struct list_head *blkdrvs_list, FILE *devices_fp)
{
char line[READ_DEVICES_LINE_LEN];
/* Skip to the line "Character devices:". */
while (fgets(line, sizeof(line), devices_fp)) {
if (line[0] == 'C')
break;
continue;
}
while (fgets(line, sizeof(line), devices_fp)) {
/* Find the blank line before "Block devices:" line. */
if (line[0] == '\n')
break;
/* Read the character device drivers */
struct devdrv *devdrv = read_devdrv(line);
if (devdrv)
list_add_tail(&devdrv->devdrvs, chrdrvs_list);
}
/* Skip to the line "Block devices:". */
while (fgets(line, sizeof(line), devices_fp)) {
if (line[0] == 'B')
break;
continue;
}
/* Read the block device drivers */
while (fgets(line, sizeof(line), devices_fp)) {
struct devdrv *devdrv = read_devdrv(line);
if (devdrv)
list_add_tail(&devdrv->devdrvs, blkdrvs_list);
}
}
static void initialize_devdrvs(void)
{
FILE *devices_fp;
INIT_LIST_HEAD(&chrdrvs);
INIT_LIST_HEAD(&blkdrvs);
devices_fp = fopen("/proc/devices", "r");
if (devices_fp) {
read_devices(&chrdrvs, &blkdrvs, devices_fp);
fclose(devices_fp);
}
}
static void finalize_devdrvs(void)
{
list_free(&blkdrvs, struct devdrv, devdrvs, free_devdrv);
list_free(&chrdrvs, struct devdrv, devdrvs, free_devdrv);
}
static const char *get_devdrv(struct list_head *devdrvs_list, unsigned long major)
{
struct list_head *c;
list_for_each(c, devdrvs_list) {
struct devdrv *devdrv = list_entry(c, struct devdrv, devdrvs);
if (devdrv->major == major)
return devdrv->name;
}
return NULL;
}
const char *get_chrdrv(unsigned long major)
{
return get_devdrv(&chrdrvs, major);
}
const char *get_blkdrv(unsigned long major)
{
return get_devdrv(&blkdrvs, major);
}
struct name_manager *new_name_manager(void)
{
struct name_manager *nm = xcalloc(1, sizeof(struct name_manager));
nm->cache = new_idcache();
if (!nm->cache)
err(EXIT_FAILURE, _("failed to allocate an idcache"));
nm->next_id = 1; /* 0 is never issued as id. */
return nm;
}
void free_name_manager(struct name_manager *nm)
{
free_idcache(nm->cache);
free(nm);
}
const char *get_name(struct name_manager *nm, unsigned long id)
{
struct identry *e;
e = get_id(nm->cache, id);
return e? e->name: NULL;
}
unsigned long add_name(struct name_manager *nm, const char *name)
{
struct identry *e = NULL, *tmp;
for (tmp = nm->cache->ent; tmp; tmp = tmp->next) {
if (strcmp(tmp->name, name) == 0) {
e = tmp;
break;
}
}
if (e)
return e->id;
e = xmalloc(sizeof(struct identry));
e->name = xstrdup(name);
e->id = nm->next_id++;
e->next = nm->cache->ent;
nm->cache->ent = e;
return e->id;
}
static void read_process(struct lsfd_control *ctl, struct path_cxt *pc,
pid_t pid, struct proc *leader)
{
char buf[BUFSIZ];
struct proc *proc;
if (procfs_process_init_path(pc, pid) != 0)
return;
proc = new_process(pid, leader);
proc->command = procfs_process_get_cmdname(pc, buf, sizeof(buf)) > 0 ?
xstrdup(buf) : xstrdup(_("(unknown)"));
procfs_process_get_uid(pc, &proc->uid);
if (procfs_process_get_stat(pc, buf, sizeof(buf)) > 0) {
char *p;
unsigned int flags;
char *pat = NULL;
/* See proc(5) about the column in the line. */
xstrappend(&pat, "%*d (");
for (p = proc->command; *p != '\0'; p++) {
if (*p == '%')
xstrappend(&pat, "%%");
else
xstrputc(&pat, *p);
}
xstrappend(&pat, ") %*c %*d %*d %*d %*d %*d %u %*[^\n]");
if (sscanf(buf, pat, &flags) == 1)
proc->kthread = !!(flags & PF_KTHREAD);
free(pat);
}
collect_execve_file(pc, proc, ctl->sockets_only);
if (proc->pid == proc->leader->pid
|| kcmp(proc->leader->pid, proc->pid, KCMP_FS, 0, 0) != 0)
collect_fs_files(pc, proc, ctl->sockets_only);
if (proc->ns_mnt == 0 || !has_mnt_ns(proc->ns_mnt)) {
FILE *mnt = ul_path_fopen(pc, "r", "mountinfo");
if (mnt) {
add_nodevs(mnt);
if (proc->ns_mnt)
add_mnt_ns(proc->ns_mnt);
fclose(mnt);
}
}
collect_namespace_files(pc, proc);
/* If kcmp is not available,
* there is no way to no whether threads share resources.
* In such cases, we must pay the costs: call collect_mem_files()
* and collect_fd_files().
*/
if ((!ctl->sockets_only)
&& (proc->pid == proc->leader->pid
|| kcmp(proc->leader->pid, proc->pid, KCMP_VM, 0, 0) != 0))
collect_mem_files(pc, proc);
if (proc->pid == proc->leader->pid
|| kcmp(proc->leader->pid, proc->pid, KCMP_FILES, 0, 0) != 0)
collect_fd_files(pc, proc, ctl->sockets_only);
list_add_tail(&proc->procs, &ctl->procs);
if (tsearch(proc, &proc_tree, proc_tree_compare) == NULL)
errx(EXIT_FAILURE, _("failed to allocate memory"));
/* The tasks collecting overwrites @pc by /proc/<task-pid>/. Keep it as
* the last path based operation in read_process()
*/
if (ctl->threads && leader == NULL) {
DIR *sub = NULL;
pid_t tid = 0;
while (procfs_process_next_tid(pc, &sub, &tid) == 0) {
if (tid == pid)
continue;
read_process(ctl, pc, tid, proc);
}
}
/* Let's be careful with number of open files */
ul_path_close_dirfd(pc);
}
static void parse_pids(const char *str, pid_t **pids, int *count)
{
long v;
char *next = NULL;
if (*str == '\0')
return;
errno = 0;
v = strtol(str, &next, 10);
if (errno)
err(EXIT_FAILURE, _("unexpected value for pid specification: %s"), str);
if (next == str)
errx(EXIT_FAILURE, _("garbage at the end of pid specification: %s"), str);
if (v < 0)
errx(EXIT_FAILURE, _("out of range value for pid specification: %ld"), v);
(*count)++;
*pids = xrealloc(*pids, (*count) * sizeof(**pids));
(*pids)[*count - 1] = (pid_t)v;
while (next && *next != '\0'
&& (isspace((unsigned char)*next) || *next == ','))
next++;
if (*next != '\0')
parse_pids(next, pids, count);
}
static int pidcmp(const void *a, const void *b)
{
pid_t pa = *(pid_t *)a;
pid_t pb = *(pid_t *)b;
if (pa < pb)
return -1;
else if (pa == pb)
return 0;
else
return 1;
}
static void sort_pids(pid_t pids[], const int count)
{
qsort(pids, count, sizeof(pid_t), pidcmp);
}
static bool member_pids(const pid_t pid, const pid_t pids[], const int count)
{
return bsearch(&pid, pids, count, sizeof(pid_t), pidcmp)? true: false;
}
static void collect_processes(struct lsfd_control *ctl, const pid_t pids[], int n_pids)
{
DIR *dir;
struct dirent *d;
struct path_cxt *pc = NULL;
pc = ul_new_path(NULL);
if (!pc)
err(EXIT_FAILURE, _("failed to alloc procfs handler"));
dir = opendir(_PATH_PROC);
if (!dir)
err(EXIT_FAILURE, _("failed to open /proc"));
while ((d = readdir(dir))) {
pid_t pid;
if (procfs_dirent_get_pid(d, &pid) != 0)
continue;
if (n_pids == 0 || member_pids(pid, pids, n_pids))
read_process(ctl, pc, pid, 0);
}
closedir(dir);
ul_unref_path(pc);
}
static void __attribute__((__noreturn__)) usage(void)
{
FILE *out = stdout;
size_t i;
fputs(USAGE_HEADER, out);
fprintf(out, _(" %s [options]\n"), program_invocation_short_name);
fputs(USAGE_OPTIONS, out);
fputs(_(" -l, --threads list in threads level\n"), out);
fputs(_(" -J, --json use JSON output format\n"), out);
fputs(_(" -n, --noheadings don't print headings\n"), out);
fputs(_(" -o, --output <list> output columns\n"), out);
fputs(_(" -r, --raw use raw output format\n"), out);
fputs(_(" -u, --notruncate don't truncate text in columns\n"), out);
fputs(_(" -p, --pid <pid(s)> collect information only specified processes\n"), out);
fputs(_(" -i[4|6], --inet[=4|6] list only IPv4 and/or IPv6 sockets\n"), out);
fputs(_(" -Q, --filter <expr> apply display filter\n"), out);
fputs(_(" --debug-filter dump the internal data structure of filter and exit\n"), out);
fputs(_(" -C, --counter <name>:<expr>\n"
" define custom counter for --summary output\n"), out);
fputs(_(" --dump-counters dump counter definitions\n"), out);
fputs(_(" --summary[=<when>] print summary information (only, append, or never)\n"), out);
fputs(USAGE_SEPARATOR, out);
printf(USAGE_HELP_OPTIONS(30));
fprintf(out, USAGE_COLUMNS);
for (i = 0; i < ARRAY_SIZE(infos); i++)
fprintf(out, " %16s %-10s%s\n", infos[i].name,
infos[i].json_type == SCOLS_JSON_STRING? "<string>":
infos[i].json_type == SCOLS_JSON_ARRAY_STRING? "<string>":
infos[i].json_type == SCOLS_JSON_ARRAY_NUMBER? "<string>":
infos[i].json_type == SCOLS_JSON_NUMBER? "<number>":
"<boolean>",
_(infos[i].help));
printf(USAGE_MAN_TAIL("lsfd(1)"));
exit(EXIT_SUCCESS);
}
static void append_filter_expr(char **a, const char *b, bool and)
{
if (*a == NULL) {
*a = xstrdup(b);
return;
}
char *tmp = *a;
*a = NULL;
xstrappend(a, "(");
xstrappend(a, tmp);
xstrappend(a, ")");
if (and)
xstrappend(a, "and(");
else
xstrappend(a, "or(");
xstrappend(a, b);
xstrappend(a, ")");
}
static struct lsfd_filter *new_filter(const char *expr, bool debug, const char *err_prefix, struct lsfd_control *ctl)
{
struct lsfd_filter *filter;
const char *errmsg;
filter = lsfd_filter_new(expr, ctl->tb,
LSFD_N_COLS,
column_name_to_id_cb,
add_column_by_id_cb, ctl);
errmsg = lsfd_filter_get_errmsg(filter);
if (errmsg)
errx(EXIT_FAILURE, "%s%s", err_prefix, errmsg);
if (debug) {
lsfd_filter_dump(filter, stdout);
exit(EXIT_SUCCESS);
}
return filter;
}
static struct counter_spec *new_counter_spec(const char *spec_str)
{
char *sep;
struct counter_spec *spec;
if (spec_str[0] == '\0')
errx(EXIT_FAILURE,
_("too short counter specification: -C/--counter %s"),
spec_str);
if (spec_str[0] == ':')
errx(EXIT_FAILURE,
_("no name for counter: -C/--counter %s"),
spec_str);
sep = strchr(spec_str, ':');
if (sep == NULL)
errx(EXIT_FAILURE,
_("no name for counter: -C/--counter %s"),
spec_str);
if (sep[1] == '\0')
errx(EXIT_FAILURE,
_("empty counter expression given: -C/--counter %s"),
spec_str);
/* Split the spec_str in to name and expr. */
*sep = '\0';
if (strchr(spec_str, '{'))
errx(EXIT_FAILURE,
_("don't use `{' in the name of a counter: %s"),
spec_str);
spec = xmalloc(sizeof(struct counter_spec));
INIT_LIST_HEAD(&spec->specs);
spec->name = spec_str;
spec->expr = sep + 1;
return spec;
}
static void free_counter_spec(struct counter_spec *counter_spec)
{
free(counter_spec);
}
static struct lsfd_counter *new_counter(const struct counter_spec *spec, struct lsfd_control *ctl)
{
struct lsfd_filter *filter;
filter = new_filter(spec->expr, false,
_("failed in making filter for a counter: "),
ctl);
return lsfd_counter_new(spec->name, filter);
}
static struct lsfd_counter **new_counters(struct list_head *specs, struct lsfd_control *ctl)
{
struct lsfd_counter **counters;
size_t len = list_count_entries(specs);
size_t i = 0;
struct list_head *s;
counters = xcalloc(len + 1, sizeof(struct lsfd_counter *));
list_for_each(s, specs) {
struct counter_spec *spec = list_entry(s, struct counter_spec, specs);
counters[i++] = new_counter(spec, ctl);
}
assert(counters[len] == NULL);
return counters;
}
static struct lsfd_counter **new_default_counters(struct lsfd_control *ctl)
{
struct lsfd_counter **counters;
size_t len = ARRAY_SIZE(default_counter_specs);
size_t i;
counters = xcalloc(len + 1, sizeof(struct lsfd_counter *));
for (i = 0; i < len; i++) {
const struct counter_spec *spec = default_counter_specs + i;
counters[i] = new_counter(spec, ctl);
}
assert(counters[len] == NULL);
return counters;
}
static void dump_default_counter_specs(void)
{
size_t len = ARRAY_SIZE(default_counter_specs);
size_t i;
puts("default counter specs:");
for (i = 0; i < len; i++) {
const struct counter_spec *spec = default_counter_specs + i;
printf("\t%s:%s\n", spec->name, spec->expr);
}
}
static void dump_counter_specs(struct list_head *specs)
{
struct list_head *s;
puts("custom counter specs:");
list_for_each(s, specs) {
struct counter_spec *spec = list_entry(s, struct counter_spec, specs);
printf("\t%s:%s\n", spec->name, spec->expr);
}
}
static struct libscols_table *new_summary_table(struct lsfd_control *ctl)
{
struct libscols_table *tb = scols_new_table();
struct libscols_column *name_cl, *value_cl;
if (!tb)
err(EXIT_FAILURE, _("failed to allocate summary table"));
scols_table_enable_noheadings(tb, ctl->noheadings);
scols_table_enable_raw(tb, ctl->raw);
scols_table_enable_json(tb, ctl->json);
if(ctl->json)
scols_table_set_name(tb, "lsfd-summary");
value_cl = scols_table_new_column(tb, _("VALUE"), 0, SCOLS_FL_RIGHT);
if (!value_cl)
err(EXIT_FAILURE, _("failed to allocate summary column"));
if (ctl->json)
scols_column_set_json_type(value_cl, SCOLS_JSON_NUMBER);
name_cl = scols_table_new_column(tb, _("COUNTER"), 0, 0);
if (!name_cl)
err(EXIT_FAILURE, _("failed to allocate summary column"));
if (ctl->json)
scols_column_set_json_type(name_cl, SCOLS_JSON_STRING);
return tb;
}
static void fill_summary_line(struct libscols_line *ln, struct lsfd_counter *counter)
{
char *str = NULL;
xasprintf(&str, "%llu", (unsigned long long)lsfd_counter_value(counter));
if (!str)
err(EXIT_FAILURE, _("failed to add summary data"));
if (scols_line_refer_data(ln, 0, str))
err(EXIT_FAILURE, _("failed to add summary data"));
if (scols_line_set_data(ln, 1, lsfd_counter_name(counter)))
err(EXIT_FAILURE, _("failed to add summary data"));
}
static void emit_summary(struct lsfd_control *ctl, struct lsfd_counter **counter)
{
struct libscols_table *tb = new_summary_table(ctl);
for (; *counter; counter++) {
struct libscols_line *ln = scols_table_new_line(tb, NULL);
fill_summary_line(ln, *counter);
}
scols_print_table(tb);
scols_unref_table(tb);
}
static void attach_xinfos(struct list_head *procs)
{
struct list_head *p;
list_for_each (p, procs) {
struct proc *proc = list_entry(p, struct proc, procs);
struct list_head *f;
list_for_each (f, &proc->files) {
struct file *file = list_entry(f, struct file, files);
if (file->class->attach_xinfo)
file->class->attach_xinfo(file);
}
}
}
/* Filter expressions for implementing -i option.
*
* To list up the protocol names, use the following command line
*
* cd linux/net;
* find . -type f -exec grep -A 1 --color=auto -nH --null -e 'struct proto .*{' \{\} +
*
*/
#define INET_SUBEXP_BEGIN "(SOCK.PROTONAME =~ \"^("
#define INET4_REG "TCP|UDP|RAW|PING|UDP-Lite|SCTP|DCCP|L2TP/IP|SMC"
#define INET6_REG "TCPv6|UDPv6|RAWv6|PINGv6|UDPLITEv6|SCTPv6|DCCPv6|L2TP/IPv6|SMC6"
#define INET_SUBEXP_END ")$\")"
static const char *inet4_subexpr = INET_SUBEXP_BEGIN
INET4_REG
INET_SUBEXP_END;
static const char *inet6_subexpr = INET_SUBEXP_BEGIN
INET6_REG
INET_SUBEXP_END;
static const char *inet46_subexpr = INET_SUBEXP_BEGIN
INET4_REG "|" INET6_REG
INET_SUBEXP_END;
int main(int argc, char *argv[])
{
int c;
size_t i;
char *outarg = NULL;
char *filter_expr = NULL;
bool debug_filter = false;
bool dump_counters = false;
pid_t *pids = NULL;
int n_pids = 0;
struct list_head counter_specs;
struct lsfd_control ctl = {
.show_main = 1
};
INIT_LIST_HEAD(&counter_specs);
enum {
OPT_DEBUG_FILTER = CHAR_MAX + 1,
OPT_SUMMARY,
OPT_DUMP_COUNTERS,
};
static const struct option longopts[] = {
{ "noheadings", no_argument, NULL, 'n' },
{ "output", required_argument, NULL, 'o' },
{ "version", no_argument, NULL, 'V' },
{ "help", no_argument, NULL, 'h' },
{ "json", no_argument, NULL, 'J' },
{ "raw", no_argument, NULL, 'r' },
{ "threads", no_argument, NULL, 'l' },
{ "notruncate", no_argument, NULL, 'u' },
{ "pid", required_argument, NULL, 'p' },
{ "inet", optional_argument, NULL, 'i' },
{ "filter", required_argument, NULL, 'Q' },
{ "debug-filter",no_argument, NULL, OPT_DEBUG_FILTER },
{ "summary", optional_argument, NULL, OPT_SUMMARY },
{ "counter", required_argument, NULL, 'C' },
{ "dump-counters",no_argument, NULL, OPT_DUMP_COUNTERS },
{ NULL, 0, NULL, 0 },
};
setlocale(LC_ALL, "");
bindtextdomain(PACKAGE, LOCALEDIR);
textdomain(PACKAGE);
close_stdout_atexit();
while ((c = getopt_long(argc, argv, "no:JrVhluQ:p:i::C:s", longopts, NULL)) != -1) {
switch (c) {
case 'n':
ctl.noheadings = 1;
break;
case 'o':
outarg = optarg;
break;
case 'J':
ctl.json = 1;
break;
case 'r':
ctl.raw = 1;
break;
case 'l':
ctl.threads = 1;
break;
case 'u':
ctl.notrunc = 1;
break;
case 'p':
parse_pids(optarg, &pids, &n_pids);
break;
case 'i': {
const char *subexpr = NULL;
ctl.sockets_only = 1;
if (optarg == NULL)
subexpr = inet46_subexpr;
else if (strcmp(optarg, "4") == 0)
subexpr = inet4_subexpr;
else if (strcmp(optarg, "6") == 0)
subexpr = inet6_subexpr;
else
errx(EXIT_FAILURE,
_("unknown -i/--inet argument: %s"),
optarg);
append_filter_expr(&filter_expr, subexpr, true);
break;
}
case 'Q':
append_filter_expr(&filter_expr, optarg, true);
break;
case 'C': {
struct counter_spec *c = new_counter_spec(optarg);
list_add_tail(&c->specs, &counter_specs);
break;
}
case OPT_DEBUG_FILTER:
debug_filter = true;
break;
case OPT_SUMMARY:
if (optarg) {
if (strcmp(optarg, "never") == 0)
ctl.show_summary = 0, ctl.show_main = 1;
else if (strcmp(optarg, "only") == 0)
ctl.show_summary = 1, ctl.show_main = 0;
else if (strcmp(optarg, "append") == 0)
ctl.show_summary = 1, ctl.show_main = 1;
else
errx(EXIT_FAILURE, _("unsupported --summary argument"));
} else
ctl.show_summary = 1, ctl.show_main = 0;
break;
case OPT_DUMP_COUNTERS:
dump_counters = true;
break;
case 'V':
print_version(EXIT_SUCCESS);
case 'h':
usage();
default:
errtryhelp(EXIT_FAILURE);
}
}
if (argv[optind])
errtryhelp(EXIT_FAILURE);
#define INITIALIZE_COLUMNS(COLUMN_SPEC) \
for (i = 0; i < ARRAY_SIZE(COLUMN_SPEC); i++) \
columns[ncolumns++] = COLUMN_SPEC[i]
if (!ncolumns) {
if (ctl.threads)
INITIALIZE_COLUMNS(default_threads_columns);
else
INITIALIZE_COLUMNS(default_columns);
}
if (outarg && string_add_to_idarray(outarg, columns, ARRAY_SIZE(columns),
&ncolumns, column_name_to_id) < 0)
return EXIT_FAILURE;
scols_init_debug(0);
INIT_LIST_HEAD(&ctl.procs);
/* inilialize scols table */
ctl.tb = scols_new_table();
if (!ctl.tb)
err(EXIT_FAILURE, _("failed to allocate output table"));
scols_table_enable_noheadings(ctl.tb, ctl.noheadings);
scols_table_enable_raw(ctl.tb, ctl.raw);
scols_table_enable_json(ctl.tb, ctl.json);
if (ctl.json)
scols_table_set_name(ctl.tb, "lsfd");
/* create output columns */
for (i = 0; i < ncolumns; i++) {
const struct colinfo *col = get_column_info(i);
struct libscols_column *cl = add_column(ctl.tb, col);
if (!cl)
err(EXIT_FAILURE, _("failed to allocate output column"));
if (ctl.notrunc) {
int flags = scols_column_get_flags(cl);
flags &= ~SCOLS_FL_TRUNC;
scols_column_set_flags(cl, flags);
}
}
/* make fitler */
if (filter_expr) {
ctl.filter = new_filter(filter_expr, debug_filter, "", &ctl);
free(filter_expr);
}
if (dump_counters) {
if (list_empty(&counter_specs))
dump_default_counter_specs();
else
dump_counter_specs(&counter_specs);
return 0;
}
/* make counters */
if (ctl.show_summary) {
if (list_empty(&counter_specs))
ctl.counters = new_default_counters(&ctl);
else {
ctl.counters = new_counters(&counter_specs, &ctl);
list_free(&counter_specs, struct counter_spec, specs,
free_counter_spec);
}
}
if (n_pids > 0)
sort_pids(pids, n_pids);
/* collect data */
initialize_nodevs();
initialize_classes();
initialize_devdrvs();
initialize_ipc_table();
collect_processes(&ctl, pids, n_pids);
free(pids);
attach_xinfos(&ctl.procs);
convert(&ctl.procs, &ctl);
/* print */
if (ctl.show_main)
emit(&ctl);
if (ctl.show_summary && ctl.counters)
emit_summary(&ctl, ctl.counters);
/* cleanup */
delete(&ctl.procs, &ctl);
finalize_ipc_table();
finalize_devdrvs();
finalize_classes();
finalize_nodevs();
return 0;
}