]> granicus.if.org Git - strace/blobdiff - strace.c
x32: fix decoding of flags argument of preadv2 and pwritev2 syscalls
[strace] / strace.c
index 9b93e7968dbb632294328a4abdf6acb867d4529a..7214e6efeba28b9def805f7d6c5d879a1fb78fdb 100644 (file)
--- a/strace.c
+++ b/strace.c
@@ -43,7 +43,9 @@
 #ifdef HAVE_PRCTL
 # include <sys/prctl.h>
 #endif
+#include <asm/unistd.h>
 
+#include "scno.h"
 #include "ptrace.h"
 #include "printsiginfo.h"
 
@@ -57,15 +59,7 @@ extern char *optarg;
 bool stack_trace_enabled = false;
 #endif
 
-#if defined __NR_tkill
-# define my_tkill(tid, sig) syscall(__NR_tkill, (tid), (sig))
-#else
-   /* kill() may choose arbitrarily the target task of the process group
-      while we later wait on a that specific TID.  PID process waits become
-      TID task specific waits for a process under ptrace(2).  */
-# warning "tkill(2) not available, risk of strace hangs!"
-# define my_tkill(tid, sig) kill((tid), (sig))
-#endif
+#define my_tkill(tid, sig) syscall(__NR_tkill, (tid), (sig))
 
 /* Glue for systems without a MMU that cannot provide fork() */
 #if !defined(HAVE_FORK)
@@ -80,7 +74,8 @@ const unsigned int syscall_trap_sig = SIGTRAP | 0x80;
 
 cflag_t cflag = CFLAG_NONE;
 unsigned int followfork = 0;
-unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC;
+unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC
+                                | PTRACE_O_TRACEEXIT;
 unsigned int xflag = 0;
 bool debug_flag = 0;
 bool Tflag = 0;
@@ -133,12 +128,8 @@ bool not_failing_only = 0;
 unsigned int show_fd_path = 0;
 
 static bool detach_on_execve = 0;
-/* Are we "strace PROG" and need to skip detach on first execve? */
-static bool skip_one_b_execve = 0;
-/* Are we "strace PROG" and need to hide everything until execve? */
-bool hide_log_until_execve = 0;
 
-static int exit_code = 0;
+static int exit_code;
 static int strace_child = 0;
 static int strace_tracer_pid = 0;
 
@@ -196,53 +187,83 @@ strerror(int err_no)
 #endif /* HAVE_STERRROR */
 
 static void
-usage(FILE *ofp, int exitval)
+print_version(void)
+{
+       printf("%s -- version %s\n"
+              "Copyright (C) %s The strace developers <%s>.\n"
+              "This is free software; see the source for copying conditions.  There is NO\n"
+              "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
+              PACKAGE_NAME, PACKAGE_VERSION, "1991-2017", PACKAGE_URL);
+}
+
+static void
+usage(void)
 {
-       fprintf(ofp, "\
-usage: strace [-CdffhiqrtttTvVxxy] [-I n] [-e expr]...\n\
+       printf("\
+usage: strace [-CdffhiqrtttTvVwxxy] [-I n] [-e expr]...\n\
               [-a column] [-o file] [-s strsize] [-P path]...\n\
               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
-   or: strace -c[df] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
+   or: strace -c[dfw] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
--c -- count time, calls, and errors for each syscall and report summary\n\
--C -- like -c but also print regular output\n\
--w -- summarise syscall latency (default is system time)\n\
--d -- enable debug output to stderr\n\
--D -- run tracer process as a detached grandchild, not as parent\n\
--f -- follow forks, -ff -- with output into separate files\n\
--i -- print instruction pointer at time of syscall\n\
--q -- suppress messages about attaching, detaching, etc.\n\
--r -- print relative timestamp, -t -- absolute timestamp, -tt -- with usecs\n\
--T -- print time spent in each syscall\n\
--v -- verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
--x -- print non-ascii strings in hex, -xx -- print all strings in hex\n\
--y -- print paths associated with file descriptor arguments\n\
--yy -- print ip:port pairs associated with socket file descriptors\n\
--h -- print help message, -V -- print version\n\
--a column -- alignment COLUMN for printing syscall results (default %d)\n\
--b execve -- detach on this syscall\n\
--e expr -- a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
-   options: trace, abbrev, verbose, raw, signal, read, write\n\
--I interruptible --\n\
-   1: no signals are blocked\n\
-   2: fatal signals are blocked while decoding syscall (default)\n\
-   3: fatal signals are always blocked (default if '-o FILE PROG')\n\
-   4: fatal signals and SIGTSTP (^Z) are always blocked\n\
-      (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
--o file -- send trace output to FILE instead of stderr\n\
--O overhead -- set overhead for tracing syscalls to OVERHEAD usecs\n\
--p pid -- trace process with process id PID, may be repeated\n\
--s strsize -- limit length of print strings to STRSIZE chars (default %d)\n\
--S sortby -- sort syscall counts by: time, calls, name, nothing (default %s)\n\
--u username -- run command as username handling setuid and/or setgid\n\
--E var=val -- put var=val in the environment for command\n\
--E var -- remove var from the environment for command\n\
--P path -- trace accesses to path\n\
+\n\
+Output format:\n\
+  -a column      alignment COLUMN for printing syscall results (default %d)\n\
+  -i             print instruction pointer at time of syscall\n\
 "
 #ifdef USE_LIBUNWIND
-"-k obtain stack trace between each syscall (experimental)\n\
+"\
+  -k             obtain stack trace between each syscall (experimental)\n\
 "
 #endif
+"\
+  -o file        send trace output to FILE instead of stderr\n\
+  -q             suppress messages about attaching, detaching, etc.\n\
+  -r             print relative timestamp\n\
+  -s strsize     limit length of print strings to STRSIZE chars (default %d)\n\
+  -t             print absolute timestamp\n\
+  -tt            print absolute timestamp with usecs\n\
+  -T             print time spent in each syscall\n\
+  -x             print non-ascii strings in hex\n\
+  -xx            print all strings in hex\n\
+  -y             print paths associated with file descriptor arguments\n\
+  -yy            print protocol specific information associated with socket file descriptors\n\
+\n\
+Statistics:\n\
+  -c             count time, calls, and errors for each syscall and report summary\n\
+  -C             like -c but also print regular output\n\
+  -O overhead    set overhead for tracing syscalls to OVERHEAD usecs\n\
+  -S sortby      sort syscall counts by: time, calls, name, nothing (default %s)\n\
+  -w             summarise syscall latency (default is system time)\n\
+\n\
+Filtering:\n\
+  -e expr        a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
+     options:    trace, abbrev, verbose, raw, signal, read, write, fault\n\
+  -P path        trace accesses to path\n\
+\n\
+Tracing:\n\
+  -b execve      detach on execve syscall\n\
+  -D             run tracer process as a detached grandchild, not as parent\n\
+  -f             follow forks\n\
+  -ff            follow forks with output into separate files\n\
+  -I interruptible\n\
+     1:          no signals are blocked\n\
+     2:          fatal signals are blocked while decoding syscall (default)\n\
+     3:          fatal signals are always blocked (default if '-o FILE PROG')\n\
+     4:          fatal signals and SIGTSTP (^Z) are always blocked\n\
+                 (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
+\n\
+Startup:\n\
+  -E var         remove var from the environment for command\n\
+  -E var=val     put var=val in the environment for command\n\
+  -p pid         trace process with process id PID, may be repeated\n\
+  -u username    run command as username handling setuid and/or setgid\n\
+\n\
+Miscellaneous:\n\
+  -d             enable debug output to stderr\n\
+  -v             verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
+  -h             print help message\n\
+  -V             print version\n\
+"
 /* ancient, no one should use it
 -F -- attempt to follow vforks (deprecated, use -f)\n\
  */
@@ -250,7 +271,7 @@ usage: strace [-CdffhiqrtttTvVxxy] [-I n] [-e expr]...\n\
 -z -- print only succeeding syscalls\n\
  */
 , DEFAULT_ACOLUMN, DEFAULT_STRLEN, DEFAULT_SORTBY);
-       exit(exitval);
+       exit(0);
 }
 
 static void ATTRIBUTE_NORETURN
@@ -310,6 +331,17 @@ void error_msg_and_die(const char *fmt, ...)
        die();
 }
 
+void error_msg_and_help(const char *fmt, ...)
+{
+       if (fmt != NULL) {
+               va_list p;
+               va_start(p, fmt);
+               verror_msg(0, fmt, p);
+       }
+       fprintf(stderr, "Try '%s -h' for more information.\n", progname);
+       die();
+}
+
 void perror_msg(const char *fmt, ...)
 {
        va_list p;
@@ -329,25 +361,29 @@ void perror_msg_and_die(const char *fmt, ...)
 static void
 error_opt_arg(int opt, const char *arg)
 {
-       error_msg_and_die("Invalid -%c argument: '%s'", opt, arg);
+       error_msg_and_help("invalid -%c argument: '%s'", opt, arg);
 }
 
-#if USE_SEIZE
+static const char *ptrace_attach_cmd;
+
 static int
 ptrace_attach_or_seize(int pid)
 {
+#if USE_SEIZE
        int r;
        if (!use_seize)
-               return ptrace(PTRACE_ATTACH, pid, 0L, 0L);
+               return ptrace_attach_cmd = "PTRACE_ATTACH",
+                      ptrace(PTRACE_ATTACH, pid, 0L, 0L);
        r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long) ptrace_setoptions);
        if (r)
-               return r;
+               return ptrace_attach_cmd = "PTRACE_SEIZE", r;
        r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L);
-       return r;
-}
+       return ptrace_attach_cmd = "PTRACE_INTERRUPT", r;
 #else
-# define ptrace_attach_or_seize(pid) ptrace(PTRACE_ATTACH, (pid), 0, 0)
+               return ptrace_attach_cmd = "PTRACE_ATTACH",
+                      ptrace(PTRACE_ATTACH, pid, 0L, 0L);
 #endif
+}
 
 /*
  * Used when we want to unblock stopped traced process.
@@ -357,26 +393,31 @@ ptrace_attach_or_seize(int pid)
  * Otherwise prints error message and returns -1.
  */
 static int
-ptrace_restart(int op, struct tcb *tcp, int sig)
+ptrace_restart(const unsigned int op, struct tcb *const tcp, unsigned int sig)
 {
        int err;
        const char *msg;
 
        errno = 0;
-       ptrace(op, tcp->pid, (void *) 0, (long) sig);
+       ptrace(op, tcp->pid, 0L, (unsigned long) sig);
        err = errno;
        if (!err)
                return 0;
 
-       msg = "SYSCALL";
-       if (op == PTRACE_CONT)
-               msg = "CONT";
-       if (op == PTRACE_DETACH)
-               msg = "DETACH";
-#ifdef PTRACE_LISTEN
-       if (op == PTRACE_LISTEN)
-               msg = "LISTEN";
-#endif
+       switch (op) {
+               case PTRACE_CONT:
+                       msg = "CONT";
+                       break;
+               case PTRACE_DETACH:
+                       msg = "DETACH";
+                       break;
+               case PTRACE_LISTEN:
+                       msg = "LISTEN";
+                       break;
+               default:
+                       msg = "SYSCALL";
+       }
+
        /*
         * Why curcol != 0? Otherwise sometimes we get this:
         *
@@ -393,7 +434,7 @@ ptrace_restart(int op, struct tcb *tcp, int sig)
        if (err == ESRCH)
                return 0;
        errno = err;
-       perror_msg("ptrace(PTRACE_%s,pid:%d,sig:%d)", msg, tcp->pid, sig);
+       perror_msg("ptrace(PTRACE_%s,pid:%d,sig:%u)", msg, tcp->pid, sig);
        return -1;
 }
 
@@ -419,7 +460,8 @@ set_cloexec_flag(int fd)
        fcntl(fd, F_SETFD, newflags); /* never fails */
 }
 
-static void kill_save_errno(pid_t pid, int sig)
+static void
+kill_save_errno(pid_t pid, int sig)
 {
        int saved_errno = errno;
 
@@ -534,7 +576,7 @@ tprintf(const char *fmt, ...)
 
        va_start(args, fmt);
        if (current_tcp) {
-               int n = strace_vfprintf(current_tcp->outf, fmt, args);
+               int n = vfprintf(current_tcp->outf, fmt, args);
                if (n < 0) {
                        if (current_tcp->outf != stderr)
                                perror_msg("%s", outfname);
@@ -663,19 +705,25 @@ newoutf(struct tcb *tcp)
 static void
 expand_tcbtab(void)
 {
-       /* Allocate some more TCBs and expand the table.
+       /* Allocate some (more) TCBs (and expand the table).
           We don't want to relocate the TCBs because our
           callers have pointers and it would be a pain.
           So tcbtab is a table of pointers.  Since we never
           free the TCBs, we allocate a single chunk of many.  */
-       unsigned int i = tcbtabsize;
-       struct tcb *newtcbs = xcalloc(tcbtabsize, sizeof(newtcbs[0]));
-       struct tcb **newtab = xreallocarray(tcbtab, tcbtabsize * 2,
-                                           sizeof(tcbtab[0]));
-       tcbtabsize *= 2;
-       tcbtab = newtab;
-       while (i < tcbtabsize)
-               tcbtab[i++] = newtcbs++;
+       unsigned int new_tcbtabsize, alloc_tcbtabsize;
+       struct tcb *newtcbs;
+
+       if (tcbtabsize) {
+               alloc_tcbtabsize = tcbtabsize;
+               new_tcbtabsize = tcbtabsize * 2;
+       } else {
+               new_tcbtabsize = alloc_tcbtabsize = 1;
+       }
+
+       newtcbs = xcalloc(alloc_tcbtabsize, sizeof(newtcbs[0]));
+       tcbtab = xreallocarray(tcbtab, new_tcbtabsize, sizeof(tcbtab[0]));
+       while (tcbtabsize < new_tcbtabsize)
+               tcbtab[tcbtabsize++] = newtcbs++;
 }
 
 static struct tcb *
@@ -711,12 +759,49 @@ alloctcb(int pid)
        error_msg_and_die("bug in alloctcb");
 }
 
+void *
+get_tcb_priv_data(const struct tcb *tcp)
+{
+       return tcp->_priv_data;
+}
+
+int
+set_tcb_priv_data(struct tcb *tcp, void *const priv_data,
+                 void (*const free_priv_data)(void *))
+{
+       if (tcp->_priv_data)
+               return -1;
+
+       tcp->_free_priv_data = free_priv_data;
+       tcp->_priv_data = priv_data;
+
+       return 0;
+}
+
+void
+free_tcb_priv_data(struct tcb *tcp)
+{
+       if (tcp->_priv_data) {
+               if (tcp->_free_priv_data) {
+                       tcp->_free_priv_data(tcp->_priv_data);
+                       tcp->_free_priv_data = NULL;
+               }
+               tcp->_priv_data = NULL;
+       }
+}
+
 static void
 droptcb(struct tcb *tcp)
 {
        if (tcp->pid == 0)
                return;
 
+       int p;
+       for (p = 0; p < SUPPORTED_PERSONALITIES; ++p)
+               free(tcp->inject_vec[p]);
+
+       free_tcb_priv_data(tcp);
+
 #ifdef USE_LIBUNWIND
        if (stack_trace_enabled) {
                unwind_tcb_fin(tcp);
@@ -764,10 +849,6 @@ detach(struct tcb *tcp)
         * before detaching.  Arghh.  We go through hoops
         * to make a clean break of things.
         */
-#if defined(SPARC)
-# undef PTRACE_DETACH
-# define PTRACE_DETACH PTRACE_SUNDETACH
-#endif
 
        if (!(tcp->flags & TCB_ATTACHED))
                goto drop;
@@ -942,9 +1023,73 @@ process_opt_p_list(char *opt)
        }
 }
 
+static void
+attach_tcb(struct tcb *const tcp)
+{
+       if (ptrace_attach_or_seize(tcp->pid) < 0) {
+               perror_msg("attach: ptrace(%s, %d)",
+                          ptrace_attach_cmd, tcp->pid);
+               droptcb(tcp);
+               return;
+       }
+
+       tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
+       newoutf(tcp);
+       if (debug_flag)
+               error_msg("attach to pid %d (main) succeeded", tcp->pid);
+
+       char procdir[sizeof("/proc/%d/task") + sizeof(int) * 3];
+       DIR *dir;
+       unsigned int ntid = 0, nerr = 0;
+
+       if (followfork && tcp->pid != strace_child &&
+           sprintf(procdir, "/proc/%d/task", tcp->pid) > 0 &&
+           (dir = opendir(procdir)) != NULL) {
+               struct_dirent *de;
+
+               while ((de = read_dir(dir)) != NULL) {
+                       if (de->d_fileno == 0)
+                               continue;
+
+                       int tid = string_to_uint(de->d_name);
+                       if (tid <= 0 || tid == tcp->pid)
+                               continue;
+
+                       ++ntid;
+                       if (ptrace_attach_or_seize(tid) < 0) {
+                               ++nerr;
+                               if (debug_flag)
+                                       perror_msg("attach: ptrace(%s, %d)",
+                                                  ptrace_attach_cmd, tid);
+                               continue;
+                       }
+                       if (debug_flag)
+                               error_msg("attach to pid %d succeeded", tid);
+
+                       struct tcb *tid_tcp = alloctcb(tid);
+                       tid_tcp->flags |= TCB_ATTACHED | TCB_STARTUP |
+                                         post_attach_sigstop;
+                       newoutf(tid_tcp);
+               }
+
+               closedir(dir);
+       }
+
+       if (!qflag) {
+               if (ntid > nerr)
+                       error_msg("Process %u attached"
+                                 " with %u threads",
+                                 tcp->pid, ntid - nerr + 1);
+               else
+                       error_msg("Process %u attached",
+                                 tcp->pid);
+       }
+}
+
 static void
 startup_attach(void)
 {
+       pid_t parent_pid = strace_tracer_pid;
        unsigned int tcbi;
        struct tcb *tcp;
 
@@ -987,92 +1132,32 @@ startup_attach(void)
                if (tcp->flags & TCB_ATTACHED)
                        continue; /* no, we already attached it */
 
-               if (followfork && !daemonized_tracer) {
-                       char procdir[sizeof("/proc/%d/task") + sizeof(int) * 3];
-                       DIR *dir;
-
-                       sprintf(procdir, "/proc/%d/task", tcp->pid);
-                       dir = opendir(procdir);
-                       if (dir != NULL) {
-                               unsigned int ntid = 0, nerr = 0;
-                               struct_dirent *de;
-
-                               while ((de = read_dir(dir)) != NULL) {
-                                       struct tcb *cur_tcp;
-                                       int tid;
-
-                                       if (de->d_fileno == 0)
-                                               continue;
-                                       /* we trust /proc filesystem */
-                                       tid = atoi(de->d_name);
-                                       if (tid <= 0)
-                                               continue;
-                                       ++ntid;
-                                       if (ptrace_attach_or_seize(tid) < 0) {
-                                               ++nerr;
-                                               if (debug_flag)
-                                                       error_msg("attach to pid %d failed", tid);
-                                               continue;
-                                       }
-                                       if (debug_flag)
-                                               error_msg("attach to pid %d succeeded", tid);
-                                       cur_tcp = tcp;
-                                       if (tid != tcp->pid)
-                                               cur_tcp = alloctcb(tid);
-                                       cur_tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
-                                       newoutf(cur_tcp);
-                               }
-                               closedir(dir);
-                               if (interactive) {
-                                       sigprocmask(SIG_SETMASK, &empty_set, NULL);
-                                       if (interrupted)
-                                               goto ret;
-                                       sigprocmask(SIG_BLOCK, &blocked_set, NULL);
-                               }
-                               ntid -= nerr;
-                               if (ntid == 0) {
-                                       perror_msg("attach: ptrace(PTRACE_ATTACH, ...)");
-                                       droptcb(tcp);
-                                       continue;
-                               }
-                               if (!qflag) {
-                                       error_msg(ntid > 1
-? "Process %u attached with %u threads"
-: "Process %u attached",
-                                               tcp->pid, ntid);
-                               }
-                               if (!(tcp->flags & TCB_ATTACHED)) {
-                                       /* -p PID, we failed to attach to PID itself
-                                        * but did attach to some of its sibling threads.
-                                        * Drop PID's tcp.
-                                        */
-                                       droptcb(tcp);
-                               }
-                               continue;
-                       } /* if (opendir worked) */
-               } /* if (-f) */
-               if (ptrace_attach_or_seize(tcp->pid) < 0) {
-                       perror_msg("attach: ptrace(PTRACE_ATTACH, ...)");
+               if (tcp->pid == parent_pid || tcp->pid == strace_tracer_pid) {
+                       errno = EPERM;
+                       perror_msg("attach: pid %d", tcp->pid);
                        droptcb(tcp);
                        continue;
                }
-               tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
-               newoutf(tcp);
-               if (debug_flag)
-                       error_msg("attach to pid %d (main) succeeded", tcp->pid);
 
-               if (daemonized_tracer) {
-                       /*
-                        * Make parent go away.
-                        * Also makes grandparent's wait() unblock.
-                        */
-                       kill(getppid(), SIGKILL);
-               }
+               attach_tcb(tcp);
 
-               if (!qflag)
-                       error_msg("Process %u attached", tcp->pid);
+               if (interactive) {
+                       sigprocmask(SIG_SETMASK, &empty_set, NULL);
+                       if (interrupted)
+                               goto ret;
+                       sigprocmask(SIG_BLOCK, &blocked_set, NULL);
+               }
        } /* for each tcbtab[] */
 
+       if (daemonized_tracer) {
+               /*
+                * Make parent go away.
+                * Also makes grandparent's wait() unblock.
+                */
+               kill(parent_pid, SIGKILL);
+               strace_child = 0;
+       }
+
  ret:
        if (interactive)
                sigprocmask(SIG_SETMASK, &empty_set, NULL);
@@ -1145,6 +1230,70 @@ exec_or_die(void)
        perror_msg_and_die("exec");
 }
 
+/*
+ * Open a dummy descriptor for use as a placeholder.
+ * The descriptor is O_RDONLY with FD_CLOEXEC flag set.
+ * A read attempt from such descriptor ends with EOF,
+ * a write attempt is rejected with EBADF.
+ */
+static int
+open_dummy_desc(void)
+{
+       int fds[2];
+
+       if (pipe(fds))
+               perror_msg_and_die("pipe");
+       close(fds[1]);
+       set_cloexec_flag(fds[0]);
+       return fds[0];
+}
+
+/* placeholder fds status for stdin and stdout */
+static bool fd_is_placeholder[2];
+
+/*
+ * Ensure that all standard file descriptors are open by opening placeholder
+ * file descriptors for those standard file descriptors that are not open.
+ *
+ * The information which descriptors have been made open is saved
+ * in fd_is_placeholder for later use.
+ */
+static void
+ensure_standard_fds_opened(void)
+{
+       int fd;
+
+       while ((fd = open_dummy_desc()) <= 2) {
+               if (fd == 2)
+                       break;
+               fd_is_placeholder[fd] = true;
+       }
+
+       if (fd > 2)
+               close(fd);
+}
+
+/*
+ * Redirect stdin and stdout unless they have been opened earlier
+ * by ensure_standard_fds_opened as placeholders.
+ */
+static void
+redirect_standard_fds(void)
+{
+       int i;
+
+       /*
+        * It might be a good idea to redirect stderr as well,
+        * but we sometimes need to print error messages.
+        */
+       for (i = 0; i <= 1; ++i) {
+               if (!fd_is_placeholder[i]) {
+                       close(i);
+                       open_dummy_desc();
+               }
+       }
+}
+
 static void
 startup_child(char **argv)
 {
@@ -1263,7 +1412,8 @@ startup_child(char **argv)
                                }
                                if (!WIFSTOPPED(status) || WSTOPSIG(status) != SIGSTOP) {
                                        kill_save_errno(pid, SIGKILL);
-                                       perror_msg_and_die("Unexpected wait status %x", status);
+                                       perror_msg_and_die("Unexpected wait status %#x",
+                                                          status);
                                }
                        }
                        /* Else: NOMMU case, we have no way to sync.
@@ -1273,24 +1423,24 @@ startup_child(char **argv)
 
                        if (ptrace_attach_or_seize(pid)) {
                                kill_save_errno(pid, SIGKILL);
-                               perror_msg_and_die("Can't attach to %d", pid);
+                               perror_msg_and_die("attach: ptrace(%s, %d)",
+                                                  ptrace_attach_cmd, pid);
                        }
                        if (!NOMMU_SYSTEM)
                                kill(pid, SIGCONT);
                }
                tcp = alloctcb(pid);
-               if (!NOMMU_SYSTEM)
-                       tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
-               else
-                       tcp->flags |= TCB_ATTACHED | TCB_STARTUP;
+               tcp->flags |= TCB_ATTACHED | TCB_STARTUP
+                           | TCB_SKIP_DETACH_ON_FIRST_EXEC
+                           | (NOMMU_SYSTEM ? 0 : (TCB_HIDE_LOG | post_attach_sigstop));
                newoutf(tcp);
        }
        else {
-               /* With -D, we are *child* here, IOW: different pid. Fetch it: */
+               /* With -D, we are *child* here, the tracee is our parent. */
+               strace_child = strace_tracer_pid;
                strace_tracer_pid = getpid();
-               /* The tracee is our parent: */
-               pid = getppid();
-               alloctcb(pid);
+               tcp = alloctcb(strace_child);
+               tcp->flags |= TCB_SKIP_DETACH_ON_FIRST_EXEC | TCB_HIDE_LOG;
                /* attaching will be done later, by startup_attach */
                /* note: we don't do newoutf(tcp) here either! */
 
@@ -1315,6 +1465,18 @@ startup_child(char **argv)
                 * to create a genuine separate stack and execute on it.
                 */
        }
+       /*
+        * A case where straced process is part of a pipe:
+        * { sleep 1; yes | head -n99999; } | strace -o/dev/null sh -c 'exec <&-; sleep 9'
+        * If strace won't close its fd#0, closing it in tracee is not enough:
+        * the pipe is still open, it has a reader. Thus, "head" will not get its
+        * SIGPIPE at once, on the first write.
+        *
+        * Preventing it by redirecting strace's stdin/out.
+        * (Don't leave fds 0 and 1 closed, this is bad practice: future opens
+        * will reuse them, unexpectedly making a newly opened object "stdin").
+        */
+       redirect_standard_fds();
 }
 
 #if USE_SEIZE
@@ -1364,8 +1526,8 @@ test_ptrace_seize(void)
                if (WIFSIGNALED(status)) {
                        return;
                }
-               error_msg_and_die("%s: unexpected wait status %x",
-                               __func__, status);
+               error_msg_and_die("%s: unexpected wait status %#x",
+                                 __func__, status);
        }
 }
 #else /* !USE_SEIZE */
@@ -1416,10 +1578,8 @@ get_os_release(void)
 static void ATTRIBUTE_NOINLINE
 init(int argc, char *argv[])
 {
-       struct tcb *tcp;
        int c, i;
        int optF = 0;
-       unsigned int tcbi;
        struct sigaction sa;
 
        progname = argv[0] ? argv[0] : "strace";
@@ -1435,13 +1595,6 @@ init(int argc, char *argv[])
 
        os_release = get_os_release();
 
-       /* Allocate the initial tcbtab.  */
-       tcbtabsize = argc;      /* Surely enough for all -p args.  */
-       tcbtab = xcalloc(tcbtabsize, sizeof(tcbtab[0]));
-       tcp = xcalloc(tcbtabsize, sizeof(*tcp));
-       for (tcbi = 0; tcbi < tcbtabsize; tcbi++)
-               tcbtab[tcbi] = tcp++;
-
        shared_log = stderr;
        set_sortby(DEFAULT_SORTBY);
        set_personality(DEFAULT_PERSONALITY);
@@ -1468,13 +1621,13 @@ init(int argc, char *argv[])
                        break;
                case 'c':
                        if (cflag == CFLAG_BOTH) {
-                               error_msg_and_die("-c and -C are mutually exclusive");
+                               error_msg_and_help("-c and -C are mutually exclusive");
                        }
                        cflag = CFLAG_ONLY_STATS;
                        break;
                case 'C':
                        if (cflag == CFLAG_ONLY_STATS) {
-                               error_msg_and_die("-c and -C are mutually exclusive");
+                               error_msg_and_help("-c and -C are mutually exclusive");
                        }
                        cflag = CFLAG_BOTH;
                        break;
@@ -1491,7 +1644,7 @@ init(int argc, char *argv[])
                        followfork++;
                        break;
                case 'h':
-                       usage(stdout, 0);
+                       usage();
                        break;
                case 'i':
                        iflag = 1;
@@ -1501,7 +1654,7 @@ init(int argc, char *argv[])
                        break;
                case 'r':
                        rflag = 1;
-                       /* fall through to tflag++ */
+                       break;
                case 't':
                        tflag++;
                        break;
@@ -1521,7 +1674,7 @@ init(int argc, char *argv[])
                        qualify("abbrev=none");
                        break;
                case 'V':
-                       printf("%s -- version %s\n", PACKAGE_NAME, VERSION);
+                       print_version();
                        exit(0);
                        break;
                case 'z':
@@ -1572,12 +1725,12 @@ init(int argc, char *argv[])
                                die_out_of_memory();
                        break;
                case 'I':
-                       opt_intr = string_to_uint(optarg);
-                       if (opt_intr <= 0 || opt_intr >= NUM_INTR_OPTS)
+                       opt_intr = string_to_uint_upto(optarg, NUM_INTR_OPTS - 1);
+                       if (opt_intr <= 0)
                                error_opt_arg(c, optarg);
                        break;
                default:
-                       usage(stderr, 1);
+                       error_msg_and_help(NULL);
                        break;
                }
        }
@@ -1588,23 +1741,23 @@ init(int argc, char *argv[])
        memset(acolumn_spaces, ' ', acolumn);
        acolumn_spaces[acolumn] = '\0';
 
-       /* Must have PROG [ARGS], or -p PID. Not both. */
-       if (!argv[0] == !nprocs)
-               usage(stderr, 1);
+       if (!argv[0] && !nprocs) {
+               error_msg_and_help("must have PROG [ARGS] or -p PID");
+       }
 
-       if (nprocs != 0 && daemonized_tracer) {
-               error_msg_and_die("-D and -p are mutually exclusive");
+       if (!argv[0] && daemonized_tracer) {
+               error_msg_and_help("PROG [ARGS] must be specified with -D");
        }
 
        if (!followfork)
                followfork = optF;
 
        if (followfork >= 2 && cflag) {
-               error_msg_and_die("(-c or -C) and -ff are mutually exclusive");
+               error_msg_and_help("(-c or -C) and -ff are mutually exclusive");
        }
 
        if (count_wallclock && !cflag) {
-               error_msg_and_die("-w must be given with (-c or -C)");
+               error_msg_and_help("-w must be given with (-c or -C)");
        }
 
        if (cflag == CFLAG_ONLY_STATS) {
@@ -1624,9 +1777,21 @@ init(int argc, char *argv[])
                        error_msg("-%c has no effect with -c", 'y');
        }
 
+       if (rflag) {
+               if (tflag > 1)
+                       error_msg("-tt has no effect with -r");
+               tflag = 1;
+       }
+
 #ifdef USE_LIBUNWIND
-       if (stack_trace_enabled)
+       if (stack_trace_enabled) {
+               unsigned int tcbi;
+
                unwind_init();
+               for (tcbi = 0; tcbi < tcbtabsize; ++tcbi) {
+                       unwind_tcb_init(tcbtab[tcbi]);
+               }
+       }
 #endif
 
        /* See if they want to run as another user. */
@@ -1656,6 +1821,19 @@ init(int argc, char *argv[])
                error_msg("ptrace_setoptions = %#x", ptrace_setoptions);
        test_ptrace_seize();
 
+       /*
+        * Is something weird with our stdin and/or stdout -
+        * for example, may they be not open? In this case,
+        * ensure that none of the future opens uses them.
+        *
+        * This was seen in the wild when /proc/sys/kernel/core_pattern
+        * was set to "|/bin/strace -o/tmp/LOG PROG":
+        * kernel runs coredump helper with fd#0 open but fd#1 closed (!),
+        * therefore LOG gets opened to fd#1, and fd#1 is closed by
+        * "don't hold up stdin/out open" code soon after.
+        */
+       ensure_standard_fds_opened();
+
        /* Check if they want to redirect the output. */
        if (outfname) {
                /* See if they want to pipe the output. */
@@ -1665,7 +1843,7 @@ init(int argc, char *argv[])
                         * when using popen, so prohibit it.
                         */
                        if (followfork >= 2)
-                               error_msg_and_die("Piping the output and -ff are mutually exclusive");
+                               error_msg_and_help("piping the output and -ff are mutually exclusive");
                        shared_log = strace_popen(outfname + 1);
                }
                else if (followfork < 2)
@@ -1677,8 +1855,7 @@ init(int argc, char *argv[])
        }
 
        if (!outfname || outfname[0] == '|' || outfname[0] == '!') {
-               char *buf = xmalloc(BUFSIZ);
-               setvbuf(shared_log, buf, _IOLBF, BUFSIZ);
+               setvbuf(shared_log, NULL, _IOLBF, 0);
        }
        if (outfname && argv[0]) {
                if (!opt_intr)
@@ -1690,9 +1867,9 @@ init(int argc, char *argv[])
                opt_intr = INTR_WHILE_WAIT;
 
        /* argv[0]      -pPID   -oFILE  Default interactive setting
-        * yes          0       0       INTR_WHILE_WAIT
+        * yes          *       0       INTR_WHILE_WAIT
         * no           1       0       INTR_WHILE_WAIT
-        * yes          0       1       INTR_NEVER
+        * yes          *       1       INTR_NEVER
         * no           1       1       INTR_WHILE_WAIT
         */
 
@@ -1705,9 +1882,6 @@ init(int argc, char *argv[])
         * in the startup_child() mode we kill the spawned process anyway.
         */
        if (argv[0]) {
-               if (!NOMMU_SYSTEM || daemonized_tracer)
-                       hide_log_until_execve = 1;
-               skip_one_b_execve = 1;
                startup_child(argv);
        }
 
@@ -1877,7 +2051,7 @@ maybe_allocate_tcb(const int pid, int status)
                /* This can happen if a clone call used
                 * CLONE_PTRACE itself.
                 */
-               ptrace(PTRACE_CONT, pid, (char *) 0, 0);
+               ptrace(PTRACE_CONT, pid, NULL, 0);
                error_msg("Stop of unknown pid %u seen, PTRACE_CONTed it", pid);
                return NULL;
        }
@@ -1890,7 +2064,7 @@ maybe_switch_tcbs(struct tcb *tcp, const int pid)
        struct tcb *execve_thread;
        long old_pid = 0;
 
-       if (ptrace(PTRACE_GETEVENTMSG, pid, NULL, (long) &old_pid) < 0)
+       if (ptrace(PTRACE_GETEVENTMSG, pid, NULL, &old_pid) < 0)
                return tcp;
        /* Avoid truncation in pid2tcb() param passing */
        if (old_pid <= 0 || old_pid == pid)
@@ -1941,8 +2115,7 @@ print_signalled(struct tcb *tcp, const int pid, int status)
        }
 
        if (cflag != CFLAG_ONLY_STATS
-        && (qual_flags[WTERMSIG(status)] & QUAL_SIGNAL)
-       ) {
+           && is_number_in_set(WTERMSIG(status), &signal_set)) {
                printleader(tcp);
 #ifdef WCOREDUMP
                tprintf("+++ killed by %s %s+++\n",
@@ -1976,13 +2149,12 @@ static void
 print_stopped(struct tcb *tcp, const siginfo_t *si, const unsigned int sig)
 {
        if (cflag != CFLAG_ONLY_STATS
-           && !hide_log_until_execve
-           && (qual_flags[sig] & QUAL_SIGNAL)
-          ) {
+           && !hide_log(tcp)
+           && is_number_in_set(sig, &signal_set)) {
                printleader(tcp);
                if (si) {
                        tprintf("--- %s ", signame(sig));
-                       printsiginfo(si, verbose(tcp));
+                       printsiginfo(si);
                        tprints(" ---\n");
                } else
                        tprintf("--- stopped by %s ---\n", signame(sig));
@@ -2011,6 +2183,43 @@ startup_tcb(struct tcb *tcp)
        }
 }
 
+static void
+print_event_exit(struct tcb *tcp)
+{
+       if (entering(tcp) || filtered(tcp) || hide_log(tcp)
+           || cflag == CFLAG_ONLY_STATS) {
+               return;
+       }
+
+       if (followfork < 2 && printing_tcp && printing_tcp != tcp
+           && printing_tcp->curcol != 0) {
+               current_tcp = printing_tcp;
+               tprints(" <unfinished ...>\n");
+               fflush(printing_tcp->outf);
+               printing_tcp->curcol = 0;
+               current_tcp = tcp;
+       }
+
+       if ((followfork < 2 && printing_tcp != tcp)
+           || (tcp->flags & TCB_REPRINT)) {
+               tcp->flags &= ~TCB_REPRINT;
+               printleader(tcp);
+               tprintf("<... %s resumed>", tcp->s_ent->sys_name);
+       }
+
+       if (!(tcp->sys_func_rval & RVAL_DECODED)) {
+               /*
+                * The decoder has probably decided to print something
+                * on exiting syscall which is not going to happen.
+                */
+               tprints(" <unfinished ...>");
+       }
+       tprints(") ");
+       tabto();
+       tprints("= ?\n");
+       line_ended();
+}
+
 /* Returns true iff the main trace loop has to continue. */
 static bool
 trace(void)
@@ -2112,11 +2321,14 @@ trace(void)
                if (os_release >= KERNEL_VERSION(3,0,0))
                        tcp = maybe_switch_tcbs(tcp, pid);
 
-               if (detach_on_execve && !skip_one_b_execve) {
-                       detach(tcp); /* do "-b execve" thingy */
-                       return true;
+               if (detach_on_execve) {
+                       if (tcp->flags & TCB_SKIP_DETACH_ON_FIRST_EXEC) {
+                               tcp->flags &= ~TCB_SKIP_DETACH_ON_FIRST_EXEC;
+                       } else {
+                               detach(tcp); /* do "-b execve" thingy */
+                               return true;
+                       }
                }
-               skip_one_b_execve = 0;
        }
 
        /* Set current output file */
@@ -2158,10 +2370,14 @@ trace(void)
 
        sig = WSTOPSIG(status);
 
-       if (event != 0) {
-               /* Ptrace event */
+       switch (event) {
+               case 0:
+                       break;
+               case PTRACE_EVENT_EXIT:
+                       print_event_exit(tcp);
+                       goto restart_tracee_with_sig_0;
 #if USE_SEIZE
-               if (event == PTRACE_EVENT_STOP) {
+               case PTRACE_EVENT_STOP:
                        /*
                         * PTRACE_INTERRUPT-stop or group-stop.
                         * PTRACE_INTERRUPT-stop has sig == SIGTRAP here.
@@ -2174,9 +2390,10 @@ trace(void)
                                        stopped = true;
                                        goto show_stopsig;
                        }
-               }
+                       /* fall through */
 #endif
-               goto restart_tracee_with_sig_0;
+               default:
+                       goto restart_tracee_with_sig_0;
        }
 
        /*
@@ -2194,7 +2411,7 @@ trace(void)
        }
 
        if (sig != syscall_trap_sig) {
-               siginfo_t si;
+               siginfo_t si = {};
 
                /*
                 * True if tracee is stopped by signal
@@ -2202,7 +2419,7 @@ trace(void)
                 * TODO: shouldn't we check for errno == EINVAL too?
                 * We can get ESRCH instead, you know...
                 */
-               stopped = ptrace(PTRACE_GETSIGINFO, pid, 0, (long) &si) < 0;
+               stopped = ptrace(PTRACE_GETSIGINFO, pid, 0, &si) < 0;
 #if USE_SEIZE
 show_stopsig:
 #endif
@@ -2238,7 +2455,8 @@ show_stopsig:
         * This should be syscall entry or exit.
         * Handle it.
         */
-       if (trace_syscall(tcp) < 0) {
+       sig = 0;
+       if (trace_syscall(tcp, &sig) < 0) {
                /*
                 * ptrace() failed in trace_syscall().
                 * Likely a result of process disappearing mid-flight.
@@ -2252,6 +2470,7 @@ show_stopsig:
                 */
                return true;
        }
+       goto restart_tracee;
 
 restart_tracee_with_sig_0:
        sig = 0;
@@ -2271,6 +2490,8 @@ main(int argc, char *argv[])
 {
        init(argc, argv);
 
+       exit_code = !nprocs;
+
        while (trace())
                ;