[PATCH v4 07/10] PID namespace translation support

Ákos Uzonyi uzonyi.akos at gmail.com
Sat Jun 13 16:18:34 UTC 2020


* defs.h (perform_ns_resolution, pid_ns, pid_ns_inited): New variables.
(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, printpid, printpid_translation):
New function definitions.
* pidns.c: New file.
* trie.c: New file.
* trie.h: New file.
* Makefile.am (libstrace_a_SOURCES): Add trie.c, pidns.c.
* strace.c (perform_ns_resolution): New variable.
(init): Add -Y flag.

Co-Authored-by: Eugene Syromyatnikov <evgsyr at gmail.com>
---
 Makefile.am |   2 +
 defs.h      |  26 ++-
 pidns.c     | 634 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 strace.c    |   8 +-
 syscall.c   |  15 ++
 trie.c      | 344 ++++++++++++++++++++++++++++
 trie.h      |  90 ++++++++
 7 files changed, 1117 insertions(+), 2 deletions(-)
 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..aa9a8d78 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -68,6 +68,7 @@ libstrace_a_SOURCES =	\
 	bpf_fprog.h	\
 	bpf_seccomp_filter.c \
 	bpf_sock_filter.c \
+	trie.c \
 	btrfs.c		\
 	cacheflush.c	\
 	capability.c	\
@@ -233,6 +234,7 @@ libstrace_a_SOURCES =	\
 	personality.c	\
 	pidfd_getfd.c	\
 	pidfd_open.c	\
+	pidns.c		\
 	pkeys.c		\
 	poll.c		\
 	prctl.c		\
diff --git a/defs.h b/defs.h
index d8bd5135..cc8dae04 100644
--- a/defs.h
+++ b/defs.h
@@ -280,6 +280,9 @@ struct tcb {
 	struct timespec etime;	/* Syscall entry time (CLOCK_MONOTONIC) */
 	struct timespec delay_expiration_time; /* When does the delay end */
 
+	uint64_t pid_ns;
+	bool pid_ns_inited;
+
 	struct mmap_cache_t *mmap_cache;
 
 	/*
@@ -413,7 +416,11 @@ 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 +435,16 @@ 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 +486,7 @@ extern int Tflag_scale;
 extern int Tflag_width;
 extern bool iflag;
 extern bool count_wallclock;
+extern unsigned int perform_ns_resolution;
 /* are we filtering traces based on paths? */
 extern struct path_set {
 	const char **paths_selected;
@@ -984,6 +1002,11 @@ 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);
+
+extern int translate_pid(struct tcb *tcp, 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);
 
@@ -1059,6 +1082,7 @@ 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 *tcp, int pid, enum pid_type type);
 extern void print_sockaddr(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/pidns.c b/pidns.c
new file mode 100644
index 00000000..1c696034
--- /dev/null
+++ b/pidns.c
@@ -0,0 +1,634 @@
+#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 "trie.h"
+#include "nsfs.h"
+#include "xmalloc.h"
+
+struct trie *ns_pid_to_proc_pid[PT_COUNT];
+struct trie *proc_data_cache;
+
+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
+
+struct proc_data {
+	int proc_pid;
+	int ns_count;
+	uint64_t ns_hierarchy[MAX_NS_DEPTH]; /* from bottom to top of NS hierarchy */
+	int id_count[PT_COUNT];
+	int *id_hierarchy[PT_COUNT]; /* from top to bottom of NS hierarchy */
+};
+
+static uint8_t
+lg2(uint64_t n)
+{
+	uint8_t res = 0;
+	while (n) {
+		res++;
+		n >>= 1;
+	}
+	return res;
+}
+
+static int
+get_pid_max(void)
+{
+	static int pid_max = -1;
+
+	if (pid_max < 0) {
+		pid_max = INT_MAX;
+
+		FILE *f = fopen("/proc/sys/kernel/pid_max", "r");
+		if (!f)
+			perror_msg("get_pid_max: opening /proc/sys/kernel/pid_max");
+		else
+			fscanf(f, "%d", &pid_max);
+	}
+
+	return pid_max;
+}
+
+void
+pidns_init(void)
+{
+	static bool inited = false;
+	if (inited)
+		return;
+
+	for (int i = 0; i < PT_COUNT; i++)
+		ns_pid_to_proc_pid[i] = trie_create(6, 16, 16, 64, 0);
+
+	proc_data_cache = trie_create(6, 16, 16, lg2(get_pid_max() - 1), 0);
+
+	inited = true;
+}
+
+static void
+put_proc_pid(uint64_t ns, int ns_pid, enum pid_type type, int proc_pid)
+{
+	struct trie *b = (struct trie *) trie_get(ns_pid_to_proc_pid[type], ns);
+	if (!b) {
+		int pid_max = get_pid_max();
+		uint8_t pid_max_size = lg2(pid_max - 1);
+		uint8_t pid_max_size_lg = lg2(pid_max_size - 1);
+		b = trie_create(pid_max_size_lg, 16, 16, pid_max_size, 0);
+
+		trie_set(ns_pid_to_proc_pid[type], ns, (uint64_t) b);
+	}
+	trie_set(b, ns_pid, proc_pid);
+}
+
+static int
+get_cached_proc_pid(uint64_t ns, int ns_pid, enum pid_type type)
+{
+	struct trie *b = (struct trie *) 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)
+{
+	static char buf[sizeof("-2147483648")];
+	ssize_t ret;
+
+	if (!pid)
+		return "self";
+
+	ret = snprintf(buf, sizeof(buf), "%d", pid);
+
+	if ((ret < 0) || ((size_t) ret >= sizeof(buf)))
+		perror_msg_and_die("pid_to_str: snprintf");
+
+	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
+ *                 MAX_NS_DEPTH items.
+ * @param last     ID of NS on which ascencion can be interrupted.
+ *                 0 for no interruption.
+ * @return         Amount of NS in list. 0 indicates error, MAX_NS_DEPTH + 1
+ *                 indicates that ascension limit hasn't been reached (only
+ *                 MAX_NS_DEPTH values have been written to the array, however).
+ */
+static size_t
+get_ns_hierarchy(int proc_pid, uint64_t *ns_buf, size_t ns_buf_size)
+{
+	char path[PATH_MAX + 1];
+	struct stat st;
+	ssize_t ret;
+	size_t n = 0;
+	int fd;
+	int parent_fd;
+
+	ret = snprintf(path, sizeof(path), "/proc/%s/ns/pid",
+		       pid_to_str(proc_pid));
+
+	if ((ret < 0) || ((size_t) ret >= sizeof(path)))
+		return 0;
+
+	fd = open(path, O_RDONLY | O_NONBLOCK);
+	if (fd < 0)
+		return 0;
+
+	while (1) {
+		ret = fstat(fd, &st);
+		if (ret)
+			break;
+
+		/* 32 is the hierarchy depth on modern Linux */
+		if ((n >= MAX_NS_DEPTH) || (n >= ns_buf_size)) {
+			n++;
+			break;
+		}
+
+		ns_buf[n] = st.st_ino;
+		if (debug_flag)
+			error_msg("Got NS: %" PRIu64, ns_buf[n]);
+
+		n++;
+
+		parent_fd = ioctl(fd, NS_GET_PARENT);
+		if (parent_fd == -1) {
+			switch (errno) {
+			case EPERM:
+				if (debug_flag)
+					error_msg("Terminating NS ascending "
+						  "after %zu levels on NS %"
+						  PRIu64, n, ns_buf[n - 1]);
+				break;
+
+			case ENOTTY:
+				error_msg("NS_* ioctl commands are not "
+					  "supported by the kernel");
+				break;
+			default:
+				perror_msg("get_ns_hierarchy: "
+					   "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,
+ *                    MAX_NS_DEPTH + 1 indicates that status record contains
+ *                    more that MAX_NS_DEPTH records and the id_buf provided
+ *                    is unusable.
+ */
+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;
+	char *buf;
+	size_t bufsize = 0;
+	char *p;
+	char *endp;
+	FILE *f;
+	int idx = 0;
+	ssize_t ret;
+
+	ret = asprintf(&buf, "/proc/%s/status", pid_to_str(proc_pid));
+	if (ret < 0)
+		return 0;
+
+	f = fopen(buf, "r");
+	if (!f)
+		return 0;
+
+	free(buf);
+	buf = NULL;
+
+	while (getline(&buf, &bufsize, f) > 0) {
+		if (strncmp(buf, ns_str, ns_str_size))
+			continue;
+
+		p = buf + ns_str_size;
+
+		for (idx = 0; idx < MAX_NS_DEPTH; idx++) {
+			if (!p)
+				break;
+
+			errno = 0;
+			int id = strtol(p, &endp, 10);
+
+			if (errno && (p[0] != '\t')) {
+				perror_msg("get_id_list: converting pid to int");
+				idx = 0;
+				goto get_id_list_exit;
+			}
+
+			if (debug_flag)
+				error_msg("PID %d: %s[%d]: %d",
+					  proc_pid, ns_str, idx, id);
+
+			if (id_buf)
+				id_buf[idx] = id;
+
+			strsep(&p, "\t");
+		}
+
+		if (p)
+			idx++;
+
+		break;
+	}
+
+get_id_list_exit:
+	if (f)
+		fclose(f);
+	if (buf)
+		free(buf);
+
+	return idx;
+}
+
+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;
+}
+
+static uint64_t
+get_ns(struct tcb *tcp)
+{
+	if (!tcp->pid_ns_inited) {
+		int pid = tcp->pid;
+
+		if (!is_proc_ours())
+			if (translate_pid(NULL, tcp->pid, PT_TID, &pid) < 1)
+				pid = -1;
+
+		if ((pid == -1) || !get_ns_hierarchy(pid, &tcp->pid_ns, 1))
+			tcp->pid_ns = 0;
+
+		tcp->pid_ns_inited = true;
+	}
+
+	return tcp->pid_ns;
+}
+
+static uint64_t
+get_our_ns(void)
+{
+	static uint64_t 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 *) 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) 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;
+
+	if (!pd->id_hierarchy[type])
+		pd->id_hierarchy[type] = calloc(MAX_NS_DEPTH,
+			sizeof(pd->id_hierarchy[type][0]));
+	if (!pd->id_hierarchy[type])
+		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:
+	if (pd)
+		free(pd);
+
+	trie_set(proc_data_cache, pd->proc_pid, (uint64_t) NULL);
+	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 */
+	uint64_t 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) {
+		perror_msg("translate_id_dir: opening dir: %s", path);
+		return;
+	}
+
+	while (!tip->result_id) {
+		errno = 0;
+		struct dirent *entry = readdir(dir);
+		if (!entry) {
+			if (errno)
+				perror_msg("translate_id_dir: 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];
+			snprintf(task_dir_path, sizeof(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 *) 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; /* According to cache, this is not what we are looking for, continue */
+
+	/* Now translate it from actual data in /proc, to check cache validity */
+	translate_id_proc_pid(tip, pd->proc_pid);
+}
+
+/**
+ * Translates an ID from tcp's namespace to our namepace
+ *
+ * @param tcp             The tcb whose namepace from_id is in (NULL means strace's namespace)
+ * @param from_id         The id to be translated
+ * @param type            The type of ID
+ * @param proc_pid_ptr    If not NULL, writes the proc PID to this location
+ */
+int
+translate_pid(struct tcb *tcp, int from_id, enum pid_type type, int *proc_pid_ptr)
+{
+	const uint64_t our_ns = get_our_ns();
+
+	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.type >= PT_COUNT) || (tip.type < 0))
+		goto translate_pid_exit;
+
+	/* 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 translate_pid_exit;
+	}
+
+	/* 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 translate_pid_exit;
+	}
+
+	/* Iterate through the cache, find potential proc_data */
+	trie_iterate_keys(proc_data_cache, 0, get_pid_max(), 0, proc_data_cache_iterator_fn, &tip);
+	/* (proc_data_cache_iterator_fn takes care about updating proc_data) */
+	if (tip.result_id)
+		goto translate_pid_exit;
+
+	/* No cache helped, read all entries in /proc */
+	translate_id_dir(&tip, "/proc", true);
+
+translate_pid_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)
+{
+	if (!is_proc_ours()) {
+		int ret;
+
+		if (translate_pid(NULL, tcp->pid, PT_TID, &ret) < 0)
+			return -1;
+
+		return ret;
+	}
+
+	return tcp->pid;
+}
+
+void
+printpid_translation(struct tcb *tcp, int pid, enum pid_type type)
+{
+	int strace_pid;
+
+	if (perform_ns_resolution) {
+		strace_pid = translate_pid(tcp, pid, type, NULL);
+
+		if ((strace_pid > 0) && (pid != strace_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);
+}
diff --git a/strace.c b/strace.c
index 311e4d62..bd8562bf 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 perform_ns_resolution;
+
 static bool detach_on_execve;
 
 static int exit_code;
@@ -2013,7 +2015,7 @@ init(int argc, char *argv[])
 	qualify_signals("all");
 
 	static const char optstring[] =
-		"+a:Ab:cCdDe:E:fFhiI:ko:O:p:P:qrs:S:tTu:U:vVwxX:yzZ";
+		"+a:Ab:cCdDe:E:fFhiI:ko:O:p:P:qrs:S:tTu:U:vVwxX:yYzZ";
 
 	enum {
 		GETOPT_SECCOMP = 0x100,
@@ -2285,6 +2287,10 @@ init(int argc, char *argv[])
 		case 'y':
 			yflag_short++;
 			break;
+		case 'Y':
+			perform_ns_resolution++;
+			pidns_init();
+			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 022cab5b..e0433e56 100644
--- a/syscall.c
+++ b/syscall.c
@@ -930,6 +930,21 @@ 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..80a4e03c
--- /dev/null
+++ b/trie.c
@@ -0,0 +1,344 @@
+/* Simple trie implementation for key-value mapping storage */
+
+#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);
+
+bool
+trie_check(uint8_t item_size_lg, uint8_t ptr_block_size_lg,
+	    uint8_t data_block_size_lg, uint8_t key_size)
+{
+	if (item_size_lg > 6)
+		return false;
+	if (key_size < 1 || key_size > 64)
+		return false;
+	if (ptr_block_size_lg < ptr_sz_lg || ptr_block_size_lg > PTR_BLOCK_SIZE_LG_MAX)
+		return false;
+	if (data_block_size_lg > DATA_BLOCK_SIZE_LG_MAX ||
+	    data_block_size_lg < 6 ||
+	    item_size_lg > data_block_size_lg)
+		return false;
+
+	return true;
+}
+
+void
+trie_init(struct trie *t, uint8_t item_size_lg, uint8_t ptr_block_size_lg,
+	   uint8_t data_block_size_lg, uint8_t key_size, uint64_t set_value)
+{
+	assert(trie_check(item_size_lg, ptr_block_size_lg, data_block_size_lg,
+			   key_size));
+
+	t->set_value = set_value;
+	t->data = TRIE_UNSET;
+	t->item_size_lg = item_size_lg;
+	t->ptr_block_size_lg = ptr_block_size_lg;
+	t->data_block_size_lg = data_block_size_lg;
+	t->key_size = key_size;
+}
+
+static uint8_t
+trie_get_depth(struct trie *t)
+{
+	return (t->key_size - (t->data_block_size_lg - t->item_size_lg) +
+		t->ptr_block_size_lg - ptr_sz_lg - 1) / (t->ptr_block_size_lg - ptr_sz_lg);
+}
+
+/**
+ * Returns lg2 of block 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_block_size(struct trie *t, uint8_t depth, int max_depth)
+{
+	if (max_depth < 0)
+		max_depth = trie_get_depth(t);
+
+	/* Last level contains data and we allow it having a different size */
+	if (depth == max_depth)
+		return t->data_block_size_lg;
+	/* Last level of the tree can be smaller */
+	if (depth == max_depth - 1)
+		return (t->key_size -
+			(t->data_block_size_lg - t->item_size_lg) - 1) %
+			(t->ptr_block_size_lg - ptr_sz_lg) + 1 + ptr_sz_lg;
+
+	return t->ptr_block_size_lg;
+}
+
+#define round_down(a, b) (((a) / (b)) * (b))
+
+/**
+ * Provides starting offset of bits in key corresponding to the block index
+ * at the specific level.
+ */
+static uint8_t
+trie_get_block_bit_offs(struct trie *t, uint8_t depth, int max_depth)
+{
+	uint8_t offs;
+
+	if (max_depth < 0)
+		max_depth = trie_get_depth(t);
+
+	if (depth == max_depth)
+		return 0;
+
+	offs = t->data_block_size_lg - t->item_size_lg;
+
+	if (depth == max_depth - 1)
+		return offs;
+
+	/* data_block_size + remainder */
+	offs += trie_get_block_size(t, max_depth - 1, max_depth) - ptr_sz_lg;
+	offs += (max_depth - depth - 2) * (t->ptr_block_size_lg - ptr_sz_lg);
+
+	return offs;
+}
+
+struct trie *
+trie_create(uint8_t item_size_lg, uint8_t ptr_block_size_lg,
+	     uint8_t data_block_size_lg, uint8_t key_size, uint64_t set_value)
+{
+	struct trie *t;
+
+	if (!trie_check(item_size_lg, ptr_block_size_lg, data_block_size_lg,
+	    key_size))
+		return NULL;
+
+	t = malloc(sizeof(*t));
+	if (!t)
+		return NULL;
+
+	trie_init(t, item_size_lg, ptr_block_size_lg, data_block_size_lg,
+		   key_size, set_value);
+
+	return t;
+}
+
+static uint64_t
+trie_filler(uint64_t val, uint8_t item_size)
+{
+	val &= (1 << (1 << item_size)) - 1;
+
+	for (; item_size < 6; item_size++)
+		val |= val << (1 << item_size);
+
+	return val;
+}
+
+static uint64_t *
+trie_get_block(struct trie *t, uint64_t key, bool auto_create)
+{
+	void **cur_block = &(t->data);
+	unsigned i;
+	uint8_t cur_depth;
+	uint8_t max_depth;
+	uint8_t sz;
+
+	if (t->key_size < 64 && key > (uint64_t) 1 << t->key_size)
+		return NULL;
+
+	max_depth = trie_get_depth(t);
+
+	for (cur_depth = 0; cur_depth <= max_depth; cur_depth++) {
+		sz = trie_get_block_size(t, cur_depth, max_depth);
+
+		if (*cur_block == TRIE_SET || *cur_block == TRIE_UNSET) {
+			void *old_val = *cur_block;
+
+			if (!auto_create)
+				return (uint64_t *) (*cur_block);
+
+			*cur_block = xcalloc(1 << sz, 8);
+
+			if (old_val == TRIE_SET) {
+				uint64_t fill_value = cur_depth == max_depth ? t->set_value : (uintptr_t) TRIE_SET;
+				uint8_t fill_size = cur_depth == max_depth ? t->item_size_lg : ptr_sz_lg;
+
+				for (i = 0; i < ((unsigned int)1 << (sz - 3)); i++)
+					((uint64_t *) *cur_block)[i] = trie_filler(fill_value, fill_size);
+			}
+		}
+
+		if (cur_depth < max_depth) {
+			size_t pos = (key >> trie_get_block_bit_offs(t,
+				cur_depth, max_depth)) & ((1 << (sz - ptr_sz_lg)) - 1);
+
+			cur_block = (((void **) (*cur_block)) + pos);
+		}
+	}
+
+	return (uint64_t *) (*cur_block);
+}
+
+bool
+trie_set(struct trie *t, uint64_t key, uint64_t val)
+{
+	uint64_t *data = trie_get_block(t, key, true);
+	size_t mask = (1 << (t->data_block_size_lg - t->item_size_lg)) - 1;
+	size_t pos = (key & mask) >> (6 - t->item_size_lg);
+
+	if (!data)
+		return false;
+
+	if (t->item_size_lg == 6) {
+		data[pos] = val;
+	} else {
+		size_t offs = (key & ((1 << (6 - t->item_size_lg)) - 1)) * (1 << t->item_size_lg);
+		uint64_t mask = (((uint64_t) 1 << (1 << t->item_size_lg)) - 1) << offs;
+
+		data[pos] &= ~mask;
+		data[pos] |= (val << offs) & mask;
+	}
+
+	return true;
+}
+
+#if 0
+int
+trie_mask_set(struct trie *t, uint64_t key, uint8_t mask_bits)
+{
+}
+
+/**
+ * Sets to 0 all keys with 0-ed bits of mask equivalent to corresponding bits in
+ * key.
+ */
+int
+trie_mask_unset(struct trie *t, uint64_t key, uint8_t mask_bits)
+{
+}
+
+int
+trie_interval_set(struct trie *t, uint64_t begin, uint64_t end, uint64_t val)
+{
+}
+
+uint64_t
+trie_get_next_set_key(struct trie *t, uint64_t key)
+{
+}
+#endif
+
+static uint64_t
+trie_data_block_get(struct trie *t, uint64_t *data, uint64_t key)
+{
+	size_t mask;
+	size_t pos;
+	size_t offs;
+
+	if (!data)
+		return 0;
+	if ((void *) data == (void *) TRIE_SET)
+		return t->set_value;
+
+	mask = (1 << (t->data_block_size_lg - t->item_size_lg)) - 1;
+	pos = (key & mask) >> (6 - t->item_size_lg);
+
+	if (t->item_size_lg == 6)
+		return data[pos];
+
+	offs = (key & ((1 << (6 - t->item_size_lg)) - 1)) * (1 << t->item_size_lg);
+
+	return (data[pos] >> offs) & (((uint64_t)1 << (1 << t->item_size_lg)) - 1);
+}
+
+uint64_t
+trie_get(struct trie *b, uint64_t key)
+{
+	return trie_data_block_get(b, trie_get_block(b, key, false), key);
+}
+
+static uint64_t
+trie_iterate_keys_block(struct trie *t, enum trie_iterate_flags flags,
+				trie_iterate_fn fn, void *fn_data,
+				void *block, uint64_t start, uint64_t end,
+				uint8_t depth, uint8_t max_depth)
+{
+	if (start > end)
+		return 0;
+
+	if ((block == TRIE_SET && !(flags & TRIE_ITERATE_KEYS_SET)) ||
+		(block == TRIE_UNSET && !(flags & TRIE_ITERATE_KEYS_UNSET)))
+		return 0;
+
+	if (block == TRIE_SET || block == TRIE_UNSET || depth == max_depth) {
+		for (uint64_t i = start; i <= end; i++)
+			fn(fn_data, i, trie_data_block_get(t, (uint64_t *) block, i));
+
+		return end - start + 1; //TODO: overflow
+	}
+
+	uint8_t parent_block_bit_off = depth == 0 ? t->key_size : trie_get_block_bit_offs(t, depth - 1, max_depth);
+	uint64_t first_key_in_block = start & (uint64_t) -1 << parent_block_bit_off;
+
+	uint8_t block_bit_off = trie_get_block_bit_offs(t, depth, max_depth);
+	uint8_t block_key_bits = parent_block_bit_off - block_bit_off;
+	uint64_t mask = ((uint64_t) 1 << (block_key_bits)) - 1;
+	uint64_t start_index = (start >> block_bit_off) & mask;
+	uint64_t end_index = (end >> block_bit_off) & mask;
+	uint64_t child_key_count = (uint64_t) 1 << block_bit_off;
+
+	uint64_t count = 0;
+
+	for (uint64_t i = start_index; i <= end_index; i++) {
+		uint64_t child_start = first_key_in_block + i * child_key_count;
+		uint64_t child_end = first_key_in_block + (i + 1) * child_key_count - 1;
+
+		if (child_start < start)
+			child_start = start;
+		if (child_end > end)
+			child_end = end;
+
+		count += trie_iterate_keys_block(t, flags, fn, fn_data,
+			((void **) block)[i], child_start, child_end,
+			depth + 1, max_depth);
+	}
+
+	return count;
+}
+
+uint64_t trie_iterate_keys(struct trie *t, uint64_t start, uint64_t end,
+			    enum trie_iterate_flags flags, trie_iterate_fn fn,
+			    void *fn_data)
+{
+	return trie_iterate_keys_block(t, flags, fn, fn_data, t->data,
+		start, end, 0, trie_get_depth(t));
+}
+
+void
+trie_free_block(struct trie *t, uint64_t **block, uint8_t depth,
+		 int max_depth)
+{
+	size_t sz;
+	size_t i;
+
+	if (block == TRIE_SET || block == TRIE_UNSET)
+		return;
+	if (max_depth < 0)
+		max_depth = trie_get_depth(t);
+	if (depth >= max_depth)
+		goto free_block;
+
+	sz = 1 << (trie_get_block_size(t, depth, max_depth) - ptr_sz_lg);
+
+	for (i = 0; i < sz; i++)
+		trie_free_block(t, (uint64_t **) (block[i]), depth + 1, max_depth);
+
+free_block:
+	free(block);
+}
+
+void
+trie_free(struct trie *t)
+{
+	trie_free_block(t, t->data, 0, -1);
+	free(t);
+}
diff --git a/trie.h b/trie.h
new file mode 100644
index 00000000..32637e87
--- /dev/null
+++ b/trie.h
@@ -0,0 +1,90 @@
+#ifndef STRACE_TRIE_H
+#define STRACE_TRIE_H
+
+/* Simple trie interface */
+
+#define TRIE_SET   ((void *) ~(intptr_t) 0)
+#define TRIE_UNSET ((void *) NULL)
+
+#define PTR_BLOCK_SIZE_LG_MAX   24
+#define DATA_BLOCK_SIZE_LG_MAX  23
+
+enum trie_iterate_flags {
+	TRIE_ITERATE_KEYS_SET   = 1 << 0,
+	TRIE_ITERATE_KEYS_UNSET = 1 << 1,
+};
+
+/**
+ * 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 pointer block size and data block size.
+ *  * It can be used for mask storage - supports storing the flag that all keys
+ *    are set/unset in the middle tree layers. See also trie_mask_set() and
+ *    trie_mask_unset().
+ *
+ * How bits of key are used for different block levels:
+ *
+ *     highest bits                                         lowest bits
+ *     | ptr_block_size_lg | ... | < remainder > | data_block_size_lg |
+ *     \______________________________________________________________/
+ *                                 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 {
+	uint64_t set_value;         /**< Default set value */
+	void *data;
+	uint8_t item_size_lg;       /**< Item size log2, in bits, 0..6. */
+	/** Pointer block size log2, in bits. 14-20, usually. */
+	uint8_t ptr_block_size_lg;
+	/** Data block size log2, in bits. 11-17, usually. */
+	uint8_t data_block_size_lg;
+	uint8_t key_size;           /**< Key size, in bits, 1..64. */
+};
+
+typedef void (*trie_iterate_fn)(void *data, uint64_t key, uint64_t val);
+
+bool trie_check(uint8_t item_size_lg, uint8_t ptr_block_size_lg,
+		 uint8_t data_block_size_lg, uint8_t key_size);
+void trie_init(struct trie *t, uint8_t item_size_lg,
+		uint8_t ptr_block_size_lg, uint8_t data_block_size_lg,
+		uint8_t key_size, uint64_t set_value);
+struct trie * trie_create(uint8_t item_size_lg, uint8_t ptr_block_size_lg,
+			    uint8_t data_block_size_lg, uint8_t key_size,
+			    uint64_t set_value);
+
+bool trie_set(struct trie *t, uint64_t key, uint64_t val);
+#if 0
+/**
+ * Sets to the value b->set_value all keys with 0-ed bits of mask equivalent to
+ * corresponding bits in key.
+ */
+int trie_mask_set(struct trie *t, uint64_t key, uint8_t mask_bits);
+/**
+ * Sets to 0 all keys with 0-ed bits of mask equivalent to corresponding bits in
+ * key.
+ */
+int trie_mask_unset(struct trie *t, uint64_t key, uint8_t mask_bits);
+int trie_interval_set(struct trie *t, uint64_t begin, uint64_t end,
+		       uint64_t val);
+
+uint64_t trie_get_next_set_key(struct trie *t, uint64_t key);
+#endif
+
+uint64_t trie_iterate_keys(struct trie *t, uint64_t start, uint64_t end,
+			    enum trie_iterate_flags flags, trie_iterate_fn fn,
+			    void *fn_data);
+
+uint64_t trie_get(struct trie *t, uint64_t key);
+
+void trie_free_block(struct trie *t, uint64_t **block, uint8_t depth,
+		      int max_depth);
+void trie_free(struct trie *t);
+
+#endif /* !STRACE_TRIE_H */
-- 
2.27.0



More information about the Strace-devel mailing list