[PATCH v6 1/5] PID namespace translation support

Ákos Uzonyi uzonyi.akos at gmail.com
Thu Aug 6 19:02:07 UTC 2020


* defs.h (pidns_translation): New variable.
(tcb): Add pid_ns field.
(RVAL_MASK): Change value from 013 to 017.
(RVAL_TID, RVAL_SID, RVAL_TGID, RVAL_PGID): New definitions.
(pid_type): New enum.
(pidns_init, translate_pid, get_proc_pid, printpid, printpid_tgid_pgid):
New function declarations.
* largefile_wrappers.h (fstat_fd): New macro.
* pidns.c: New file.
* trie.c: New file.
* trie.h: New file.
* Makefile.am (libstrace_a_SOURCES): Add trie.c, trie.h, pidns.c.
* strace.c (pidns_translation): New variable.
(init): Add --pidns-translation option.
* syscall.c (syscall_exiting_trace): Handle RVAL_* return values.
* NEWS: Mention this.
* strace.1.in: Add description for new option.

Co-Authored-by: Eugene Syromyatnikov <evgsyr at gmail.com>
---
 Makefile.am          |   3 +
 NEWS                 |   1 +
 defs.h               |  53 +++-
 largefile_wrappers.h |   2 +
 pidns.c              | 613 +++++++++++++++++++++++++++++++++++++++++++
 strace.1.in          |   4 +
 strace.c             |   9 +
 syscall.c            |  15 ++
 trie.c               | 250 ++++++++++++++++++
 trie.h               |  89 +++++++
 10 files changed, 1038 insertions(+), 1 deletion(-)
 create mode 100644 pidns.c
 create mode 100644 trie.c
 create mode 100644 trie.h

diff --git a/Makefile.am b/Makefile.am
index f5447811..73c22035 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -233,6 +233,7 @@ libstrace_a_SOURCES =	\
 	personality.c	\
 	pidfd_getfd.c	\
 	pidfd_open.c	\
+	pidns.c		\
 	pkeys.c		\
 	poll.c		\
 	prctl.c		\
@@ -344,6 +345,8 @@ libstrace_a_SOURCES =	\
 	time.c		\
 	times.c		\
 	trace_event.h	\
+	trie.c 		\
+	trie.h 		\
 	truncate.c	\
 	ubi.c		\
 	ucopy.c		\
diff --git a/NEWS b/NEWS
index b83f9e80..9086fde9 100644
--- a/NEWS
+++ b/NEWS
@@ -14,6 +14,7 @@ Noteworthy changes in release ?.? (????-??-??)
     KEYCTL_*, KVM_*, LOOP_*, NDA_*, RTC_*, TCA_*, STATX_*, and *_MAGIC
     constants.
   * Updated lists of ioctl commands from Linux 5.8.
+  * Added --pidns-translation opition for PID namespace translation.
 
 * Bug fixes
   * Added statx syscall to %fstat trace class.
diff --git a/defs.h b/defs.h
index f7ac54f3..692fa0d3 100644
--- a/defs.h
+++ b/defs.h
@@ -280,6 +280,13 @@ struct tcb {
 	struct timespec etime;	/* Syscall entry time (CLOCK_MONOTONIC) */
 	struct timespec delay_expiration_time; /* When does the delay end */
 
+	/*
+	 * The ID of the PID namespace of this process
+	 * (inode number of /proc/<pid>/ns/pid)
+	 * (0: not initialized)
+	 */
+	unsigned int pid_ns;
+
 	struct mmap_cache_t *mmap_cache;
 
 	/*
@@ -413,7 +420,11 @@ extern const struct xlat whence_codes[];
 # define RVAL_HEX	001	/* hex format */
 # define RVAL_OCTAL	002	/* octal format */
 # define RVAL_FD		010	/* file descriptor */
-# define RVAL_MASK	013	/* mask for these values */
+# define RVAL_TID	011	/* task ID */
+# define RVAL_SID	012	/* session ID */
+# define RVAL_TGID	013	/* thread group ID */
+# define RVAL_PGID	014	/* process group ID */
+# define RVAL_MASK	017	/* mask for these values */
 
 # define RVAL_STR	020	/* Print `auxstr' field after return val */
 # define RVAL_NONE	040	/* Print nothing */
@@ -428,6 +439,16 @@ extern const struct xlat whence_codes[];
 
 # define indirect_ipccall(tcp) (tcp_sysent(tcp)->sys_flags & TRACE_INDIRECT_SUBCALL)
 
+enum pid_type {
+	PT_TID,
+	PT_TGID,
+	PT_PGID,
+	PT_SID,
+
+	PT_COUNT,
+	PT_NONE = -1
+};
+
 enum sock_proto {
 	SOCK_PROTO_UNKNOWN,
 	SOCK_PROTO_UNIX,
@@ -469,6 +490,7 @@ extern int Tflag_scale;
 extern int Tflag_width;
 extern bool iflag;
 extern bool count_wallclock;
+extern unsigned int pidns_translation;
 /* are we filtering traces based on paths? */
 extern struct path_set {
 	const char **paths_selected;
@@ -983,6 +1005,27 @@ print_local_array_ex(struct tcb *tcp,
 extern kernel_ulong_t *
 fetch_indirect_syscall_args(struct tcb *, kernel_ulong_t addr, unsigned int n_args);
 
+extern void pidns_init(void);
+
+/**
+ * Returns the pid of the tracee as present in /proc (can be different from
+ * tcp->pid if /proc and the tracee process are in different PID namespaces).
+ */
+extern int get_proc_pid(struct tcb *);
+
+/**
+ * Translates a pid from tracee's namespace to our namepace.
+ *
+ * @param tcp             The tcb of the tracee
+ *                        (NULL: from_id is in strace's namespace. Useful for
+ *                         getting the proc PID of from_id)
+ * @param from_id         The id to be translated
+ * @param type            The PID type of from_id
+ * @param proc_pid_ptr    If not NULL, writes the proc PID to this location
+ */
+extern int translate_pid(struct tcb *, int dest_id, enum pid_type type,
+		    int *proc_pid_ptr);
+
 extern void
 dumpiov_in_msghdr(struct tcb *, kernel_ulong_t addr, kernel_ulong_t data_size);
 
@@ -1058,6 +1101,14 @@ printfd(struct tcb *tcp, int fd)
  * of the tracee the descriptor tcp).  This is a stub.
  */
 extern void printfd_pid_tracee_ns(struct tcb *tcp, pid_t pid, int fd);
+
+extern void printpid(struct tcb *, int pid, enum pid_type type);
+
+/**
+ * Prints pid as a TGID if positive, and PGID if negative
+ * (like the first argument of kill).
+ */
+extern void printpid_tgid_pgid(struct tcb *, int pid);
 extern void print_sockaddr(struct tcb *, const void *sa, int len);
 extern bool
 print_inet_addr(int af, const void *addr, unsigned int len, const char *var_name);
diff --git a/largefile_wrappers.h b/largefile_wrappers.h
index 116e7048..9d8f5c92 100644
--- a/largefile_wrappers.h
+++ b/largefile_wrappers.h
@@ -29,6 +29,7 @@
 #  else
 #   define fcntl_fd fcntl
 #  endif
+#  define fstat_fd fstat64
 #  define strace_stat_t struct stat64
 #  define stat_file stat64
 #  define struct_dirent struct dirent64
@@ -39,6 +40,7 @@
 #  define open_file open
 #  define fopen_stream fopen
 #  define fcntl_fd fcntl
+#  define fstat_fd fstat
 #  define strace_stat_t struct stat
 #  define stat_file stat
 #  define struct_dirent struct dirent
diff --git a/pidns.c b/pidns.c
new file mode 100644
index 00000000..6a99fbb4
--- /dev/null
+++ b/pidns.c
@@ -0,0 +1,613 @@
+/*
+ * Copyright (c) 2020 Ákos Uzonyi <uzonyi.akos at gmail.com>
+ * All rights reserved.
+ *
+ * SPDX-License-Identifier: LGPL-2.1-or-later
+ */
+
+#include "defs.h"
+
+
+#include <dirent.h>
+#include <fcntl.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+
+#include <asm/unistd.h>
+
+#include <sys/ioctl.h>
+#include <sys/param.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+
+#include "largefile_wrappers.h"
+#include "trie.h"
+#include "nsfs.h"
+#include "xmalloc.h"
+#include "xstring.h"
+
+/**
+ * Key:   PID NS ID
+ * Value: a btree:
+ *           Key:   a process PID in NS
+ *           Value: the process's PID as present in /proc
+ */
+static struct trie *ns_pid_to_proc_pid[PT_COUNT];
+
+/**
+ * Key:   Proc PID
+ * Value: stuct proc_data
+ */
+static struct trie *proc_data_cache;
+
+static bool ns_get_parent_enotty = false;
+
+static const char tid_str[]  = "NSpid:\t";
+static const char tgid_str[] = "NStgid:\t";
+static const char pgid_str[] = "NSpgid:\t";
+static const char sid_str[]  = "NSsid:\t";
+
+static const struct {
+	const char *str;
+	size_t size;
+} id_strs[PT_COUNT] = {
+	[PT_TID] =  { tid_str,  sizeof(tid_str)  - 1 },
+	[PT_TGID] = { tgid_str, sizeof(tgid_str) - 1 },
+	[PT_PGID] = { pgid_str, sizeof(pgid_str) - 1 },
+	[PT_SID] =  { sid_str,  sizeof(sid_str)  - 1 },
+};
+
+
+/**
+ * Limit on PID NS hierarchy depth, imposed since Linux 3.7. NS traversal
+ * is not possible before Linux 4.9, so we consider this limit pretty universal.
+ */
+#define MAX_NS_DEPTH 32
+
+static const size_t ns_id_size = sizeof(unsigned int);
+static const uint8_t ptr_sz_lg = (sizeof(void *) == 8 ? 6 : 5);
+
+static int pid_max;
+static uint8_t pid_max_size, pid_max_size_lg;
+
+struct proc_data {
+	int proc_pid;
+	int ns_count;
+	unsigned int ns_hierarchy[MAX_NS_DEPTH];
+	int id_count[PT_COUNT];
+	int id_hierarchy[PT_COUNT][MAX_NS_DEPTH];
+};
+
+/**
+ * Helper function for creating a trie.
+ *
+ * For node_key_bits and data_block_key_bits 4 is used (so trie height is 32 / 4
+ * = 8, and node sizes are 8 byte * 2^4 = 128 bytes), which seems to be a good
+ * tradeoff between between memory usage and lookup time. It should not be too
+ * large, since there can be large holes between PIDs, and it would be just a
+ * waste of memory having large nodes with lot of NULL pointers in them.
+ */
+static struct trie *
+create_trie_4(uint8_t key_size, uint8_t item_size_lg, uint64_t empty_value)
+{
+	struct trie *t = trie_create(key_size, item_size_lg, 4, 4, empty_value);
+	if (!t)
+		error_msg_and_die("creating trie failed");
+
+	return t;
+}
+
+void
+pidns_init(void)
+{
+	if (proc_data_cache)
+		return;
+
+	pid_max = INT_MAX;
+	if (read_int_from_file("/proc/sys/kernel/pid_max", &pid_max) < 0)
+		debug_func_perror_msg("reading /proc/sys/kernel/pid_max");
+	pid_max_size = ilog2_32(pid_max - 1) + 1;
+	pid_max_size_lg = ilog2_32(pid_max_size - 1) + 1;
+
+	for (int i = 0; i < PT_COUNT; i++)
+		ns_pid_to_proc_pid[i] = create_trie_4(ns_id_size, ptr_sz_lg, 0);
+
+	proc_data_cache = create_trie_4(pid_max_size, ptr_sz_lg, 0);
+}
+
+static void
+put_proc_pid(unsigned int ns, int ns_pid, enum pid_type type, int proc_pid)
+{
+	struct trie *b = (struct trie *) (uintptr_t) trie_get(ns_pid_to_proc_pid[type], ns);
+	if (!b) {
+		b = create_trie_4(pid_max_size, pid_max_size_lg, 0);
+		trie_set(ns_pid_to_proc_pid[type], ns, (uint64_t) (uintptr_t) b);
+	}
+	trie_set(b, ns_pid, proc_pid);
+}
+
+static int
+get_cached_proc_pid(unsigned int ns, int ns_pid, enum pid_type type)
+{
+	struct trie *b = (struct trie *) (uintptr_t)
+		trie_get(ns_pid_to_proc_pid[type], ns);
+	if (!b)
+		return 0;
+
+	return trie_get(b, ns_pid);
+}
+
+/**
+ * Helper function, converts pid to string, or to "self" for pid == 0.
+ * Uses static buffer for operation.
+ */
+static const char *
+pid_to_str(pid_t pid)
+{
+	if (!pid)
+		return "self";
+
+	static char buf[sizeof("-2147483648")];
+	xsprintf(buf, "%d", pid);
+	return buf;
+}
+
+/**
+ * Returns a list of PID NS IDs for the specified PID.
+ *
+ * @param proc_pid PID (as present in /proc) to get information for.
+ * @param ns_buf   Pointer to buffer that is able to contain at least
+ *                 ns_buf_size items.
+ * @return         Amount of NS in list. 0 indicates error.
+ */
+static size_t
+get_ns_hierarchy(int proc_pid, unsigned int *ns_buf, size_t ns_buf_size)
+{
+	char path[PATH_MAX + 1];
+	xsprintf(path, "/proc/%s/ns/pid", pid_to_str(proc_pid));
+
+	int fd = open_file(path, O_RDONLY);
+	if (fd < 0)
+		return 0;
+
+	size_t n = 0;
+	while (n < ns_buf_size) {
+		strace_stat_t st;
+		if (fstat_fd(fd, &st))
+			break;
+
+		ns_buf[n++] = st.st_ino;
+		if (n >= ns_buf_size)
+			break;
+
+		if (ns_get_parent_enotty)
+			break;
+
+		int parent_fd = ioctl(fd, NS_GET_PARENT);
+		if (parent_fd < 0) {
+			switch (errno) {
+			case EPERM:
+				break;
+
+			case ENOTTY:
+				ns_get_parent_enotty = true;
+				error_msg("NS_* ioctl commands are not "
+					  "supported by the kernel");
+				break;
+
+			default:
+				perror_func_msg("ioctl(NS_GET_PARENT)");
+				break;
+			}
+
+			break;
+		}
+
+		close(fd);
+		fd = parent_fd;
+	}
+
+	close(fd);
+
+	return n;
+}
+
+/**
+ * Get list of IDs present in NS* proc status record. IDs are placed as they are
+ * stored in /proc (from top to bottom of NS hierarchy).
+ *
+ * @param proc_pid    PID (as present in /proc) to get information for.
+ * @param id_buf      Pointer to buffer that is able to contain at least
+ *                    MAX_NS_DEPTH items. Can be NULL.
+ * @param type        Type of ID requested.
+ * @return            Number of items stored in id_list. 0 indicates error.
+ */
+static size_t
+get_id_list(int proc_pid, int *id_buf, enum pid_type type)
+{
+	const char *ns_str = id_strs[type].str;
+	size_t ns_str_size = id_strs[type].size;
+
+	size_t n = 0;
+
+	char status_path[PATH_MAX + 1];
+	xsprintf(status_path, "/proc/%s/status", pid_to_str(proc_pid));
+	FILE *f = fopen_stream(status_path, "r");
+	if (!f)
+		return 0;
+
+	char *line = NULL;
+	size_t linesize = 0;
+	char *p = NULL;
+
+	while (getline(&line, &linesize, f) > 0) {
+		if (strncmp(line, ns_str, ns_str_size) == 0) {
+			p = line + ns_str_size;
+			break;
+		}
+	}
+
+	while (p) {
+		errno = 0;
+		long id = strtol(p, NULL, 10);
+
+		if (errno || (id < 1) || (id > INT_MAX)) {
+			perror_func_msg("converting pid to int");
+			break;
+		}
+
+		if (id_buf)
+			id_buf[n] = (int) id;
+
+		n++;
+		strsep(&p, "\t");
+	}
+
+	free(line);
+	fclose(f);
+
+	return n;
+}
+
+/**
+ * Returns whether the /proc filesystem's PID namespace is same as strace's.
+ */
+static bool
+is_proc_ours(void)
+{
+	static int cached_val = -1;
+
+	if (cached_val < 0)
+		cached_val = get_id_list(0, NULL, PT_TID) == 1;
+
+	return cached_val;
+}
+
+/**
+ * Returns the PID namespace of the tracee
+ */
+static unsigned int
+get_ns(struct tcb *tcp)
+{
+	if (!tcp->pid_ns) {
+		int proc_pid = 0;
+		translate_pid(NULL, tcp->pid, PT_TID, &proc_pid);
+
+		if (proc_pid)
+			get_ns_hierarchy(proc_pid, &tcp->pid_ns, 1);
+	}
+
+	return tcp->pid_ns;
+}
+
+/**
+ * Returns the PID namespace of strace
+ */
+static unsigned int
+get_our_ns(void)
+{
+	static unsigned int our_ns = 0;
+	static bool our_ns_initialised = false;
+
+	if (!our_ns_initialised) {
+		get_ns_hierarchy(0, &our_ns, 1);
+		our_ns_initialised = true;
+	}
+
+	return our_ns;
+}
+
+/**
+ * Returns the cached proc_data struct associated with proc_pid.
+ * If none found, allocates a new proc_data.
+ */
+static struct proc_data *
+get_or_create_proc_data(int proc_pid)
+{
+	struct proc_data *pd = (struct proc_data *) (uintptr_t)
+		trie_get(proc_data_cache, proc_pid);
+
+	if (!pd) {
+		pd = calloc(1, sizeof(*pd));
+		if (!pd)
+			return NULL;
+
+		pd->proc_pid = proc_pid;
+		trie_set(proc_data_cache, proc_pid, (uint64_t) (uintptr_t) pd);
+	}
+
+	return pd;
+}
+
+/**
+ * Updates the proc_data from /proc
+ * If the process does not exists, returns false, and frees the proc_data
+ */
+static bool
+update_proc_data(struct proc_data *pd, enum pid_type type)
+{
+	pd->ns_count = get_ns_hierarchy(pd->proc_pid,
+		pd->ns_hierarchy, MAX_NS_DEPTH);
+	if (!pd->ns_count)
+		goto fail;
+
+	pd->id_count[type] = get_id_list(pd->proc_pid,
+		pd->id_hierarchy[type], type);
+	if (!pd->id_count[type])
+		goto fail;
+
+	return true;
+
+fail:
+	trie_set(proc_data_cache, pd->proc_pid, (uint64_t) (uintptr_t) NULL);
+	free(pd);
+	return false;
+}
+
+/**
+ * Paramters for id translation
+ */
+struct translate_id_params {
+	/* The result (output) */
+	int result_id;
+	/* The proc data of the process (output) */
+	struct proc_data *pd;
+
+	/* The namespace to be translated from */
+	unsigned int from_ns;
+	/* The id to be translated */
+	int from_id;
+	/* The type of the id */
+	enum pid_type type;
+};
+
+/**
+ * Translates an id to our namespace, given the proc_pid of the process,
+ * by reading files in /proc.
+ *
+ * @param tip      The parameters
+ * @param proc_pid The proc pid of the process.
+ *                 If 0, use the cached values in tip->pd.
+ */
+static void
+translate_id_proc_pid(struct translate_id_params *tip, int proc_pid)
+{
+	struct proc_data *pd = proc_pid ?
+		get_or_create_proc_data(proc_pid) :
+		tip->pd;
+
+	tip->result_id = 0;
+	tip->pd = NULL;
+
+	if (!pd)
+		return;
+
+	if (proc_pid && !update_proc_data(pd, tip->type))
+		return;
+
+	if (!pd->ns_count || (pd->id_count[tip->type] < pd->ns_count))
+		return;
+
+	int our_ns_id_idx = pd->id_count[tip->type] - pd->ns_count;
+
+	for (int i = 0; i < pd->ns_count; i++) {
+		if (pd->ns_hierarchy[i] != tip->from_ns)
+			continue;
+
+		int id_idx = pd->id_count[tip->type] - i - 1;
+		if (pd->id_hierarchy[tip->type][id_idx] != tip->from_id)
+			return;
+
+		tip->result_id = pd->id_hierarchy[tip->type][our_ns_id_idx];
+		tip->pd = pd;
+		return;
+	}
+}
+
+/**
+ * Translates an id to our namespace, by reading all proc entries in dir.
+ *
+ * @param tip            The parameters
+ * @param path           The path of the dir to be read.
+ * @param read_task_dir  Whether recurse to "task" subdirectory.
+ */
+static void
+translate_id_dir(struct translate_id_params *tip, const char *path,
+                 bool read_task_dir)
+{
+	DIR *dir = opendir(path);
+	if (!dir) {
+		debug_func_perror_msg("opening dir: %s", path);
+		return;
+	}
+
+	while (!tip->result_id) {
+		errno = 0;
+		struct_dirent *entry = read_dir(dir);
+		if (!entry) {
+			if (errno)
+				perror_func_msg("readdir");
+
+			break;
+		}
+
+		if (entry->d_type != DT_DIR)
+			continue;
+
+		errno = 0;
+		long proc_pid = strtol(entry->d_name, NULL, 10);
+		if (errno)
+			continue;
+		if ((proc_pid < 1) || (proc_pid > INT_MAX))
+			continue;
+
+		if (read_task_dir) {
+			char task_dir_path[PATH_MAX + 1];
+			xsprintf(task_dir_path, "/proc/%ld/task", proc_pid);
+			translate_id_dir(tip, task_dir_path, false);
+		}
+
+		if (tip->result_id)
+			break;
+
+		translate_id_proc_pid(tip, proc_pid);
+	}
+
+	closedir(dir);
+}
+
+/**
+ * Iterator function of the proc_data_cache for id translation.
+ * If the cache contains the id we are looking for, reads the corresponding
+ * directory in /proc, and if cache is valid, saves the result.
+ */
+static void
+proc_data_cache_iterator_fn(void* fn_data, uint64_t key, uint64_t val)
+{
+	struct translate_id_params *tip = (struct translate_id_params *)fn_data;
+	struct proc_data *pd = (struct proc_data *) (uintptr_t) val;
+
+	if (!pd)
+		return;
+
+	/* Result already found in an earlier iteration */
+	if (tip->result_id)
+		return;
+
+	/* Translate from cache */
+	tip->pd = pd;
+	translate_id_proc_pid(tip, 0);
+	if (!tip->result_id)
+		return;
+
+	/* Now translate from actual data in /proc, to check cache validity */
+	translate_id_proc_pid(tip, pd->proc_pid);
+}
+
+int
+translate_pid(struct tcb *tcp, int from_id, enum pid_type type,
+              int *proc_pid_ptr)
+{
+	if ((from_id <= 0) || (type < 0) || (type >= PT_COUNT))
+		return 0;
+
+	const unsigned int our_ns = get_our_ns();
+	if (!our_ns)
+		return 0;
+
+	struct translate_id_params tip = {
+		.result_id = 0,
+		.pd = NULL,
+		.from_ns = tcp ? get_ns(tcp) : our_ns,
+		.from_id = from_id,
+		.type = type,
+	};
+
+	if (!tip.from_ns)
+		return 0;
+
+	/* If translation is trivial */
+	if (tip.from_ns == our_ns && (is_proc_ours() || !proc_pid_ptr)) {
+		if (proc_pid_ptr)
+			*proc_pid_ptr = from_id;
+
+		tip.result_id = tip.from_id;
+		goto exit;
+	}
+
+	if (ns_get_parent_enotty)
+		return 0;
+
+	/* Look for a cached proc_pid for this (from_ns, from_id) pair */
+	int cached_proc_pid = get_cached_proc_pid(tip.from_ns, tip.from_id,
+		tip.type);
+	if (cached_proc_pid) {
+		translate_id_proc_pid(&tip, cached_proc_pid);
+		if (tip.result_id)
+			goto exit;
+	}
+
+	/* Iterate through the cache, find potential proc_data */
+	trie_iterate_keys(proc_data_cache, 0, pid_max,
+		proc_data_cache_iterator_fn, &tip);
+	/* (proc_data_cache_iterator_fn takes care about updating proc_data) */
+	if (tip.result_id)
+		goto exit;
+
+	/* No cache helped, read all entries in /proc */
+	translate_id_dir(&tip, "/proc", true);
+
+exit:
+	if (tip.pd) {
+		if (tip.pd->proc_pid)
+			put_proc_pid(tip.from_ns, tip.from_id, tip.type,
+				tip.pd->proc_pid);
+
+		if (proc_pid_ptr)
+			*proc_pid_ptr = tip.pd->proc_pid;
+	}
+
+	return tip.result_id;
+}
+
+int
+get_proc_pid(struct tcb *tcp)
+{
+	int ret = 0;
+
+	if (is_proc_ours())
+		ret = tcp->pid;
+	else
+		translate_pid(NULL, tcp->pid, PT_TID, &ret);
+
+	return ret;
+}
+
+static void
+printpid_translation(struct tcb *tcp, int pid, enum pid_type type)
+{
+	if (!pidns_translation)
+		return;
+
+	int strace_pid = translate_pid(tcp, pid, type, NULL);
+	if (strace_pid && strace_pid != pid)
+		tprintf_comment("%d in strace's PID NS", strace_pid);
+}
+
+void
+printpid(struct tcb *tcp, int pid, enum pid_type type)
+{
+	tprintf("%d", pid);
+	printpid_translation(tcp, pid, type);
+}
+
+void
+printpid_tgid_pgid(struct tcb *tcp, int pid)
+{
+	tprintf("%d", pid);
+	if (pid > 0)
+		printpid_translation(tcp,  pid, PT_TGID);
+	else if (pid < -1)
+		printpid_translation(tcp, -pid, PT_PGID);
+}
diff --git a/strace.1.in b/strace.1.in
index 3b21caec..adca194c 100644
--- a/strace.1.in
+++ b/strace.1.in
@@ -1075,6 +1075,10 @@ Print all available information associated with file descritors:
 protocol-specific information associated with socket file descriptors,
 block/character device number associated with device file descriptors,
 and PIDs asociated with pidfd file descriptors.
+.TP
+.B \-\-pidns\-translation
+If strace and tracee are in different PID namespaces, print PIDs in
+strace's namespace also.
 .SS Statistics
 .TP 12
 .B \-c
diff --git a/strace.c b/strace.c
index 4c96a98b..249533ea 100644
--- a/strace.c
+++ b/strace.c
@@ -133,6 +133,8 @@ static unsigned int daemonized_tracer;
 static int post_attach_sigstop = TCB_IGNORE_ONE_SIGSTOP;
 #define use_seize (post_attach_sigstop == 0)
 
+unsigned int pidns_translation;
+
 static bool detach_on_execve;
 
 static int exit_code;
@@ -1998,6 +2000,8 @@ init(int argc, char *argv[])
 
 	os_release = get_os_release();
 
+	pidns_init();
+
 	shared_log = stderr;
 	set_sortby(DEFAULT_SORTBY);
 	set_personality(DEFAULT_PERSONALITY);
@@ -2022,6 +2026,7 @@ init(int argc, char *argv[])
 		GETOPT_FOLLOWFORKS,
 		GETOPT_OUTPUT_SEPARATELY,
 		GETOPT_TS,
+		GETOPT_PIDNS_TRANSLATION,
 
 		GETOPT_QUAL_TRACE,
 		GETOPT_QUAL_ABBREV,
@@ -2072,6 +2077,7 @@ init(int argc, char *argv[])
 		{ "summary-wall-clock", no_argument,	   0, 'w' },
 		{ "strings-in-hex",	optional_argument, 0, GETOPT_HEX_STR },
 		{ "const-print-style",	required_argument, 0, 'X' },
+		{ "pidns-translation",	no_argument      , 0, GETOPT_PIDNS_TRANSLATION },
 		{ "successful-only",	no_argument,	   0, 'z' },
 		{ "failed-only",	no_argument,	   0, 'Z' },
 		{ "failing-only",	no_argument,	   0, 'Z' },
@@ -2285,6 +2291,9 @@ init(int argc, char *argv[])
 		case 'y':
 			yflag_short++;
 			break;
+		case GETOPT_PIDNS_TRANSLATION:
+			pidns_translation++;
+			break;
 		case 'z':
 			clear_number_set_array(status_set, 1);
 			add_number_to_set(STATUS_SUCCESSFUL, status_set);
diff --git a/syscall.c b/syscall.c
index bcc87025..0f4bab6d 100644
--- a/syscall.c
+++ b/syscall.c
@@ -937,6 +937,21 @@ syscall_exiting_trace(struct tcb *tcp, struct timespec *ts, int res)
 					tprintf("= %" PRI_kld, tcp->u_rval);
 				}
 				break;
+			case RVAL_TID:
+			case RVAL_SID:
+			case RVAL_TGID:
+			case RVAL_PGID: {
+				#define _(_t) [RVAL_##_t - RVAL_TID] = PT_##_t
+				static const enum pid_type types[] = {
+					_(TID), _(SID), _(TGID), _(PGID),
+				};
+				#undef _
+
+				tprints("= ");
+				printpid(tcp, tcp->u_rval,
+					 types[(sys_res & RVAL_MASK) - RVAL_TID]);
+				break;
+			}
 			default:
 				error_msg("invalid rval format");
 				break;
diff --git a/trie.c b/trie.c
new file mode 100644
index 00000000..76cecc25
--- /dev/null
+++ b/trie.c
@@ -0,0 +1,250 @@
+/*
+ * Simple trie implementation for key-value mapping storage
+ *
+ * Copyright (c) 2020 Ákos Uzonyi <uzonyi.akos at gmail.com>
+ * All rights reserved.
+ *
+ * SPDX-License-Identifier: LGPL-2.1-or-later
+ */
+
+#include "defs.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+#include "trie.h"
+
+static const uint8_t ptr_sz_lg = (sizeof(uint64_t *) == 8 ? 6 : 5);
+
+/**
+ * Returns lg2 of node size for the specific level of the trie. If max_depth
+ * provided is less than zero, it is calculated via trie_get_depth call.
+ */
+static uint8_t
+trie_get_node_size(struct trie *t, uint8_t depth)
+{
+	/* Last level contains data and we allow it having a different size */
+	if (depth == t->max_depth)
+		return t->data_block_key_bits + t->item_size_lg;
+	/* Last level of the tree can be smaller */
+	if (depth == t->max_depth - 1)
+		return (t->key_size - t->data_block_key_bits - 1) %
+		t->node_key_bits + 1 + ptr_sz_lg;
+
+	return t->node_key_bits + ptr_sz_lg;
+}
+
+/**
+ * Provides starting offset of bits in key corresponding to the node index
+ * at the specific level.
+ */
+static uint8_t
+trie_get_node_bit_offs(struct trie *t, uint8_t depth)
+{
+	uint8_t offs;
+
+	if (depth == t->max_depth)
+		return 0;
+
+	offs = t->data_block_key_bits;
+
+	if (depth == t->max_depth - 1)
+		return offs;
+
+	/* data_block_size + remainder */
+	offs += trie_get_node_size(t, t->max_depth - 1) - ptr_sz_lg;
+	offs += (t->max_depth - depth - 2) * t->node_key_bits;
+
+	return offs;
+}
+
+struct trie *
+trie_create(uint8_t key_size, uint8_t item_size_lg, uint8_t node_key_bits,
+            uint8_t data_block_key_bits, uint64_t empty_value)
+{
+	if (item_size_lg > 6)
+		return NULL;
+	if (key_size > 64)
+		return NULL;
+	if (node_key_bits < 1)
+		return NULL;
+	if (data_block_key_bits < 1 || data_block_key_bits > key_size)
+		return NULL;
+
+	struct trie *t = malloc(sizeof(*t));
+	if (!t)
+		return NULL;
+
+	t->empty_value = empty_value;
+	t->data = NULL;
+	t->item_size_lg = item_size_lg;
+	t->node_key_bits = node_key_bits;
+	t->data_block_key_bits = data_block_key_bits;
+	t->key_size = key_size;
+	t->max_depth = (key_size - data_block_key_bits + node_key_bits - 1)
+		/ t->node_key_bits;
+
+	return t;
+}
+
+static uint64_t *
+trie_get_node(struct trie *t, uint64_t key, bool auto_create)
+{
+	void **cur_node = &(t->data);
+
+	if (t->key_size < 64 && key > (uint64_t) 1 << t->key_size)
+		return NULL;
+
+	for (uint8_t cur_depth = 0; cur_depth <= t->max_depth; cur_depth++) {
+		uint8_t offs = trie_get_node_bit_offs(t, cur_depth);
+		uint8_t sz = trie_get_node_size(t, cur_depth);
+
+		if (!*cur_node) {
+			if (!auto_create)
+				return NULL;
+
+			*cur_node = xcalloc(1 << sz, 1);
+		}
+
+		if (cur_depth >= t->max_depth)
+			break;
+
+		size_t pos = (key >> offs) & ((1 << (sz - ptr_sz_lg)) - 1);
+		cur_node = (((void **) (*cur_node)) + pos);
+	}
+
+	return (uint64_t *) (*cur_node);
+}
+
+static void
+trie_data_block_calc_pos(struct trie *t, uint64_t key,
+                         uint64_t *pos, uint64_t *mask, uint64_t *offs)
+{
+	uint64_t key_mask;
+
+	key_mask = (1 << t->data_block_key_bits) - 1;
+	*pos = (key & key_mask) >> (6 - t->item_size_lg);
+
+	key_mask = (1 << (6 - t->item_size_lg)) - 1;
+	*offs = (key & key_mask) * (1 << t->item_size_lg);
+
+	*mask = (((uint64_t) 1 << (1 << t->item_size_lg)) - 1) << *offs;
+}
+
+bool
+trie_set(struct trie *t, uint64_t key, uint64_t val)
+{
+	uint64_t *data = trie_get_node(t, key, true);
+	if (!data)
+		return false;
+
+	uint64_t pos, mask, offs;
+	trie_data_block_calc_pos(t, key, &pos, &mask, &offs);
+
+	data[pos] &= ~mask;
+	data[pos] |= (val << offs) & mask;
+
+	return true;
+}
+
+static uint64_t
+trie_data_block_get(struct trie *t, uint64_t *data, uint64_t key)
+{
+	if (!data)
+		return t->empty_value;
+
+	uint64_t pos, mask, offs;
+	trie_data_block_calc_pos(t, key, &pos, &mask, &offs);
+
+	return (data[pos] & mask) >> offs;
+}
+
+uint64_t
+trie_get(struct trie *b, uint64_t key)
+{
+	return trie_data_block_get(b, trie_get_node(b, key, false), key);
+}
+
+static uint64_t
+trie_iterate_keys_node(struct trie *t,
+                       trie_iterate_fn fn, void *fn_data,
+                       void *node, uint64_t start, uint64_t end,
+                       uint8_t depth)
+{
+	if (start > end || !node)
+		return 0;
+
+	if (depth == t->max_depth) {
+		for (uint64_t i = start; i <= end; i++)
+			fn(fn_data, i, trie_data_block_get(t,
+				(uint64_t *) node, i));
+
+		return end - start + 1; //TODO: overflow
+	}
+
+	uint8_t parent_node_bit_off = depth == 0 ?
+		t->key_size :
+		trie_get_node_bit_offs(t, depth - 1);
+
+	uint64_t first_key_in_node = start &
+		(uint64_t) -1 << parent_node_bit_off;
+
+	uint8_t node_bit_off = trie_get_node_bit_offs(t, depth);
+	uint8_t node_key_bits = parent_node_bit_off - node_bit_off;
+	uint64_t mask = ((uint64_t) 1 << (node_key_bits)) - 1;
+	uint64_t start_index = (start >> node_bit_off) & mask;
+	uint64_t end_index = (end >> node_bit_off) & mask;
+	uint64_t child_key_count = (uint64_t) 1 << node_bit_off;
+
+	uint64_t count = 0;
+
+	for (uint64_t i = start_index; i <= end_index; i++) {
+		uint64_t child_start = first_key_in_node + i * child_key_count;
+		uint64_t child_end = first_key_in_node +
+			(i + 1) * child_key_count - 1;
+
+		if (child_start < start)
+			child_start = start;
+		if (child_end > end)
+			child_end = end;
+
+		count += trie_iterate_keys_node(t, fn, fn_data,
+			((void **) node)[i], child_start, child_end,
+			depth + 1);
+	}
+
+	return count;
+}
+
+uint64_t trie_iterate_keys(struct trie *t, uint64_t start, uint64_t end,
+                           trie_iterate_fn fn, void *fn_data)
+{
+	return trie_iterate_keys_node(t, fn, fn_data, t->data,
+		start, end, 0);
+}
+
+static void
+trie_free_node(struct trie *t, void *node, uint8_t depth)
+{
+	if (!node)
+		return;
+
+	if (depth >= t->max_depth)
+		goto free_node;
+
+	size_t sz = 1 << (trie_get_node_size(t, depth) - ptr_sz_lg);
+	for (size_t i = 0; i < sz; i++)
+		trie_free_node(t, ((void **) node)[i], depth + 1);
+
+free_node:
+	free(node);
+}
+
+void
+trie_free(struct trie *t)
+{
+	trie_free_node(t, t->data, 0);
+	free(t);
+}
diff --git a/trie.h b/trie.h
new file mode 100644
index 00000000..f7058c7e
--- /dev/null
+++ b/trie.h
@@ -0,0 +1,89 @@
+/*
+ * Simple trie interface
+ *
+ * Copyright (c) 2020 Ákos Uzonyi <uzonyi.akos at gmail.com>
+ * All rights reserved.
+ *
+ * SPDX-License-Identifier: LGPL-2.1-or-later
+ */
+
+#ifndef STRACE_TRIE_H
+#define STRACE_TRIE_H
+
+/**
+ * Trie control structure.
+ * Trie implemented here has the following properties:
+ *  * It allows storing values of the same size, the size can vary from 1 bit to
+ *    64 bit values (only power of 2 sizes are allowed).
+ *  * The key can be up to 64 bits in size.
+ *  * It has separate configuration for node size and data block size.
+ *
+ * How bits of key are used for different node levels:
+ *
+ *   highest bits                                                  lowest bits
+ *  | node_key_bits | node_key_bits | ... | <remainder> | data_block_key_bits |
+ *  \_________________________________________________________________________/
+ *                                 key_size
+ *
+ * So, the remainder is used on the lowest non-data node level.
+ *
+ * As of now, it doesn't implement any mechanisms for resizing/changing key
+ * size.  De-fragmentation is also unsupported currently.
+ */
+struct trie {
+	/** Return value of trie_get if key is not found */
+	uint64_t empty_value;
+
+	/** Pointer to root node */
+	void *data;
+
+	/** Key size in bits (0..64). */
+	uint8_t key_size;
+
+	/**
+	 * Size of the stored values in log2 bits (0..6).
+	 * (6: 64 bit values, 5: 32 bit values, ...)
+	 */
+	uint8_t item_size_lg;
+
+	/**
+	 * Number of bits in key that makes a symbol for a node.
+	 * (equals to log2 of the child count of the node)
+	 */
+	uint8_t node_key_bits;
+
+	/**
+	 * Number of bits in key that make a symbol for the data block (leaf).
+	 * (equals to log2 of the value count stored in a data block)
+	 */
+	uint8_t data_block_key_bits;
+
+	/** The depth of the data block. Caluclated from the values above */
+	uint8_t max_depth;
+};
+
+struct trie* trie_create(uint8_t key_size, uint8_t item_size_lg,
+			uint8_t node_key_bits, uint8_t data_block_key_bits,
+			uint64_t empty_value);
+
+bool trie_set(struct trie *t, uint64_t key, uint64_t val);
+uint64_t trie_get(struct trie *t, uint64_t key);
+
+typedef void (*trie_iterate_fn)(void *data, uint64_t key, uint64_t val);
+
+/**
+ * Calls trie_iterate_fn for each key-value pair where
+ * key is inside the [start, end] interval (inclusive).
+ *
+ * @param t        The trie.
+ * @param start    The start of the key interval (inclusive).
+ * @param end      The end of the key interval (inclusive).
+ * @param fn       The function to be called.
+ * @param fn_data  The value to be passed to fn.
+ */
+uint64_t trie_iterate_keys(struct trie *t, uint64_t start, uint64_t end,
+			    trie_iterate_fn fn, void *fn_data);
+
+void trie_free(struct trie *t);
+
+#endif /* !STRACE_TRIE_H */
-- 
2.27.0



More information about the Strace-devel mailing list