From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.129.124]) by sourceware.org (Postfix) with ESMTPS id 25D403871D1A for ; Wed, 14 Dec 2022 11:51:31 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 25D403871D1A Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=redhat.com Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=redhat.com DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1671018690; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: in-reply-to:in-reply-to:references:references; bh=zKc5raN/RowY2nb973LDOIyi3LVczp8EUceYMAZAsGk=; b=TysdYtBxXh3sNCtLSDI658WNwT5ngQkpvVmCQFvNTrsYNt7C428M0XsHuOvCZaPnIiorSu YJOXz5MBe5d7zGhPf7745xMQeUxbBqy3EkHcapEWDDVqJRSnFgALUNMFFYKO9irJwtwcjT bx2geRterjSnzMxMO76Dr/o/SDYzs4Y= Received: from mail-wm1-f69.google.com (mail-wm1-f69.google.com [209.85.128.69]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.3, cipher=TLS_AES_128_GCM_SHA256) id us-mta-551-gN73sCMuONmnk2wqpF2Vaw-1; Wed, 14 Dec 2022 06:51:29 -0500 X-MC-Unique: gN73sCMuONmnk2wqpF2Vaw-1 Received: by mail-wm1-f69.google.com with SMTP id ay19-20020a05600c1e1300b003cf758f1617so7113315wmb.5 for ; Wed, 14 Dec 2022 03:51:29 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=mime-version:message-id:date:references:in-reply-to:subject:cc:to :from:x-gm-message-state:from:to:cc:subject:date:message-id:reply-to; bh=zKc5raN/RowY2nb973LDOIyi3LVczp8EUceYMAZAsGk=; b=m4ei+eCBJtscOJ+2h1/jiPi2ZYIRXrrGQuBH8BYcqfZJL9/Lyb+OetE0dro+hJm2Yo n+Sr9dpdAv/ViIYKe1/XeL9p85noXexQOwwkZR3mCim8JGPe4HPUyjc23TTxFi4v06we /P/ltEPmai1U/PG4XCfU8E10T/mB1hEo77TGVYrMZ+3p/QMKcikrjCkB1bPuOq1SOo+a Xbk9fR1ylXcfjl8fJRj6WPJdG8HNJJvHSf3/pQKdw2x6e4+OSS1PRJ3XoOAYDh90dqaM Bu3cWJ0SB8r/H+XdJZyEYi3jby4lRnZZRs3yrygj/Dg4gtpDKkc5I/ZzlC+Zwb6Bg+lz ljdw== X-Gm-Message-State: ANoB5pnZWlw8Knbnd4mjmq8XogpKFKQhWMIZEQsw9XjccVZXMtFh2xIU DrWmEfhZXraIen5963o3WERYiK+UxYLl1LQUCXu9KwPbXUYcDZzSCv/oLt3rkI/Rxk7DO8UO0oM pFwGQjPH0PKrvbZTp/fS3u5JFaum7F24fJ8lLG2+A6wOkba7LLqZ9kY0Z/W878OdyUM1trYty0w == X-Received: by 2002:a05:600c:5112:b0:3cf:8ed7:712d with SMTP id o18-20020a05600c511200b003cf8ed7712dmr18268947wms.14.1671018687982; Wed, 14 Dec 2022 03:51:27 -0800 (PST) X-Google-Smtp-Source: AA0mqf6I88jbibs0uYSM7SEnE1VvIFftR5sNQ2OR8g1naCrEaDhRSsaNIPOAuSZUSscw4b8ynDWPdg== X-Received: by 2002:a05:600c:5112:b0:3cf:8ed7:712d with SMTP id o18-20020a05600c511200b003cf8ed7712dmr18268919wms.14.1671018687497; Wed, 14 Dec 2022 03:51:27 -0800 (PST) Received: from localhost ([31.111.84.238]) by smtp.gmail.com with ESMTPSA id d20-20020a1c7314000000b003cfd42821dasm2320771wmb.3.2022.12.14.03.51.26 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Wed, 14 Dec 2022 03:51:26 -0800 (PST) From: Andrew Burgess To: gdb-patches@sourceware.org Cc: Dr Lancelot SIX , Asaf Fisher Subject: Re: [PATCHv4] gdb: handle loading shared libraries from /proc/self/fd/ In-Reply-To: <96170aac143fdbcd13595e3285e125e62fe800c8.1669052725.git.aburgess@redhat.com> References: <87zgck7c8j.fsf@redhat.com> <96170aac143fdbcd13595e3285e125e62fe800c8.1669052725.git.aburgess@redhat.com> Date: Wed, 14 Dec 2022 11:51:26 +0000 Message-ID: <87k02u19vl.fsf@redhat.com> MIME-Version: 1.0 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Type: text/plain X-Spam-Status: No, score=-11.7 required=5.0 tests=BAYES_00,DKIMWL_WL_HIGH,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,GIT_PATCH_0,KAM_SHORT,RCVD_IN_DNSWL_NONE,RCVD_IN_MSPIKE_H2,SPF_HELO_NONE,SPF_NONE,TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: Ping! Anyone else have any thoughts on this change? Thanks, Andrew Andrew Burgess writes: > From: Asaf Fisher via Gdb-patches > > Lancelot, > > I dug into why the interesting shared library event is sometime the > first stop, and sometimes the second stop. I think it depends on > whether the stap probes based mechanism is used for handling shared > library events or not. > > When the probes are not used then the library is reported at the first > event (though if you continue again there is another stop, this time > with no new libraries reported). If you are using stap probes then > the library is reported at the second stop. > > I was able to reproduce the behaviour when I use a different Fedora > version to my default. What I haven't looked into yet is why one > install uses probes, but the other does not. I'll take a look at that > tomorrow. > > However, I don't think that matters for this patch, probes or > non-probes are both valid, so I've updated the test such that it > should handle both cases fine. > > Changes since v3: > > - Fixed NULL/nullptr issues Lancelot pointed out, > > - Updated the test case to take account of non-probes based shared > library event handling. > > Thanks, > Andrew > > --- > > Bug PR gdb/29586 describes a situations where a shared library is > created in memory, then written to a memory mapped file. The memory > mapped file will show up as a file descriptor within /proc/self/fd/, > and this path is then used with dlopen in order to call functions > within the in-memory shared library. > > When attempting to debug this GDB hangs. The problem is that, GDB > stops at the shared-library event, reads the shared library path from > the inferior, which is "/proc/self/fd/", and then GDB attempts to > open this file. > > Unfortunately, this means GDB tries to open a file within GDB's > /proc/self/fd/ directory, not within the inferior's directory. In the > case of our hang it turns out that the file descriptor that is opened > is a pipe, and GDB hangs trying to read from the pipe. > > However, the behaviour is really just undefined, depending on which > file descriptor the inferior tries to open, GDB will open, or fail to > open, random files within its /proc/self/fd directory. > > The solution proposed in this commit is to hook into solib_find_1, and > spot when GDB is looking for any file in /proc/self/, if this is the > case, then the filename is rewritten as /proc/, where is > the process-id of the current inferior. > > The test passes for the unix, native-gdbserver, and > native-extended-gdbserver targets. When testing with either of the > gdbserver targets, the test is run using the default empty sysroot, > and also using the 'target:' sysroot. > > Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29586 > > Co-authored-by: Andrew Burgess > --- > gdb/solib.c | 38 ++++- > gdb/testsuite/gdb.base/solib-proc-self.cc | 73 +++++++++ > gdb/testsuite/gdb.base/solib-proc-self.exp | 175 +++++++++++++++++++++ > 3 files changed, 282 insertions(+), 4 deletions(-) > create mode 100644 gdb/testsuite/gdb.base/solib-proc-self.cc > create mode 100644 gdb/testsuite/gdb.base/solib-proc-self.exp > > diff --git a/gdb/solib.c b/gdb/solib.c > index 7cfdd81114c..6aabcf0d68d 100644 > --- a/gdb/solib.c > +++ b/gdb/solib.c > @@ -50,6 +50,7 @@ > #include "gdb_bfd.h" > #include "gdbsupport/filestuff.h" > #include "gdbsupport/scoped_fd.h" > +#include "gdbsupport/pathstuff.h" > #include "debuginfod-support.h" > #include "source.h" > #include "cli/cli-style.h" > @@ -83,6 +84,34 @@ show_solib_search_path (struct ui_file *file, int from_tty, > # define DOS_BASED_FILE_SYSTEM 0 > #endif > > +/* Fix references to files in /proc/self/fd/ when opening a shared library. > + > + SO_NAME is the name of the shared library being loaded. This function > + returns a possibly modified name which should be used as the path to the > + shared library. > + > + If SO_NAME starts with /proc/self, then the returned name will be > + modified to start with /proc/PID where 'PID' is the pid of the current > + inferior. */ > + > +static gdb::unique_xmalloc_ptr > +filter_proc_self_filenames (gdb::unique_xmalloc_ptr so_name) > +{ > + static const char *proc_self_prefix = "/proc/self"; > + > + /* Is the path really a /proc/self? */ > + if (!startswith (so_name.get (), proc_self_prefix)) > + return so_name; > + > + /* Get the part of the path after /proc/self. For example given > + '/proc/self/fd' we find the '/fd' part. */ > + const char *tail = so_name.get () + strlen (proc_self_prefix); > + > + /* Build a replacement path. */ > + int inferior_pid = inferior_ptid.pid (); > + return xstrprintf ("/proc/%d%s", inferior_pid, tail); > +} > + > /* Return the full pathname of a binary file (the main executable or a > shared library file), or NULL if not found. If FD is non-NULL, *FD > is set to either -1 or an open file handle for the binary file. > @@ -172,6 +201,9 @@ solib_find_1 (const char *in_pathname, int *fd, bool is_solib) > } > } > > + temp_pathname.reset (xstrdup (in_pathname)); > + temp_pathname = filter_proc_self_filenames (std::move (temp_pathname)); > + > /* Note, we're interested in IS_TARGET_ABSOLUTE_PATH, not > IS_ABSOLUTE_PATH. The latter is for host paths only, while > IN_PATHNAME is a target path. For example, if we're supposed to > @@ -184,9 +216,7 @@ solib_find_1 (const char *in_pathname, int *fd, bool is_solib) > 3rd attempt, c:/foo/bar.dll ==> /sysroot/foo/bar.dll > */ > > - if (!IS_TARGET_ABSOLUTE_PATH (fskind, in_pathname) || sysroot == NULL) > - temp_pathname.reset (xstrdup (in_pathname)); > - else > + if (IS_TARGET_ABSOLUTE_PATH (fskind, in_pathname) && sysroot != nullptr) > { > bool need_dir_separator; > > @@ -213,7 +243,7 @@ solib_find_1 (const char *in_pathname, int *fd, bool is_solib) > /* Cat the prefixed pathname together. */ > temp_pathname.reset (concat (sysroot, > need_dir_separator ? SLASH_STRING : "", > - in_pathname, (char *) NULL)); > + temp_pathname.get (), nullptr)); > } > > /* Handle files to be accessed via the target. */ > diff --git a/gdb/testsuite/gdb.base/solib-proc-self.cc b/gdb/testsuite/gdb.base/solib-proc-self.cc > new file mode 100644 > index 00000000000..f2439d738a3 > --- /dev/null > +++ b/gdb/testsuite/gdb.base/solib-proc-self.cc > @@ -0,0 +1,73 @@ > +/* This testcase is part of GDB, the GNU debugger. > + > + Copyright 2022 Free Software Foundation, Inc. > + > + This program is free software; you can redistribute it and/or modify > + it under the terms of the GNU General Public License as published by > + the Free Software Foundation; either version 3 of the License, or > + (at your option) any later version. > + > + This program is distributed in the hope that it will be useful, > + but WITHOUT ANY WARRANTY; without even the implied warranty of > + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + GNU General Public License for more details. > + > + You should have received a copy of the GNU General Public License > + along with this program. If not, see . */ > + > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > + > +#ifdef __WIN32__ > +#include > +#define dlopen(name, mode) LoadLibrary (name) > +#define dlclose(handle) FreeLibrary (handle) > +#define dlerror() "an error occurred" > +#else > +#include > +#endif > + > +int > +main () > +{ > + /* Read the shared libraries contents into a buffer. */ > + std::ifstream read_so_file = std::ifstream (SHLIB_NAME); > + read_so_file.seekg (0, std::ios::end); > + std::streamsize size = read_so_file.tellg (); > + read_so_file.seekg (0, std::ios::beg); > + std::vector buffer (size); > + if (!read_so_file.read (buffer.data (), size)) > + { > + fprintf (stderr, "Failed to load solib\n"); > + exit (1); > + } > + > + /* Create a memory mapped file, then write the shared library to that > + new memory mapped file. */ > + int mem_fd = memfd_create ("test", 0); > + write (mem_fd, buffer.data (), buffer.size ()); > + > + /* Generate the /proc/self/fd/[num] path for the memory mapped file. */ > + std::string proc_self_fd_path; /* break-here */ > + std::stringstream proc_self_fd_path_stream > + = std::stringstream (proc_self_fd_path); > + proc_self_fd_path_stream << "/proc/self/fd/" << mem_fd; > + > + /* Call dlopen on it. */ > + void *handle = dlopen (proc_self_fd_path_stream.str ().c_str (), RTLD_LAZY); > + if (!handle) > + { > + fprintf (stderr, "%s\n", dlerror ()); > + exit (1); > + } > + /* It worked. */ > + dlclose (handle); > + > + return 0; > +} > diff --git a/gdb/testsuite/gdb.base/solib-proc-self.exp b/gdb/testsuite/gdb.base/solib-proc-self.exp > new file mode 100644 > index 00000000000..794a03556ff > --- /dev/null > +++ b/gdb/testsuite/gdb.base/solib-proc-self.exp > @@ -0,0 +1,175 @@ > +# Copyright 2022 Free Software Foundation, Inc. > + > +# This program is free software; you can redistribute it and/or modify > +# it under the terms of the GNU General Public License as published by > +# the Free Software Foundation; either version 3 of the License, or > +# (at your option) any later version. > +# > +# This program is distributed in the hope that it will be useful, > +# but WITHOUT ANY WARRANTY; without even the implied warranty of > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > +# GNU General Public License for more details. > +# > +# You should have received a copy of the GNU General Public License > +# along with this program. If not, see . */ > + > +# Test connecting and disconnecting at shared library events. > + > +if {[skip_shlib_tests]} { > + untested "could not run to main" > + return 0 > +} > + > +standard_testfile .cc > + > +# Reuse an existing library, we don't care about the library contents > +# for this test. > +set libfile so-disc-shr > +set libsrc "${srcdir}/${subdir}/${libfile}.c" > +set libname "${libfile}.so" > +set libobj [standard_output_file ${libname}] > + > +# Compile the shared library. > +if { [gdb_compile_shlib $libsrc $libobj {debug}] != ""} { > + return -1 > +} > + > +# Compile the test executable. > +if [ build_executable "failed to prepare" $testfile $srcfile \ > + [list shlib_load debug c++ additional_flags=-DSHLIB_NAME="${libobj}"]] { > + return -1 > +} > + > +# Start GDB and run to the point where the test program tries to dlopen a file > +# from within /proc/self/fd/. Catch the shared library event and check that > +# we actually try to load a file from /proc//fd/. > +# > +# If SYSROOT is not the empty string, then this is set as the value of GDB's > +# sysroot immediately after starting GDB. The only value that is (currently) > +# supported, other than the empty string, is 'target:'. > +proc do_test { {sysroot ""} } { > + clean_restart $::binfile > + > + if {$sysroot != ""} { > + gdb_test_no_output "set sysroot ${sysroot}" > + } > + > + gdb_load_shlib $::libobj > + > + if ![runto_main] then { > + return 0 > + } > + > + # Get inferior's PID for later. > + set inferior_pid [get_inferior_pid] > + > + # Turn on the solib-events so we can see that gdb resolves everything > + # correctly. > + gdb_test_no_output "set stop-on-solib-events 1" > + > + # Run to the 'break-here' marker. > + gdb_breakpoint [gdb_get_line_number "break-here"] > + gdb_continue_to_breakpoint "break-here" ".* break-here .*" > + > + set memfd "" > + gdb_test_multiple "p mem_fd" "Get file descriptor" { > + -re -wrap "\\\$$::decimal = (\[^\r\n\]*)" { > + set memfd $expect_out(1,string) > + pass $gdb_test_name > + } > + } > + > + # Now continue forward until the solib event is detected, and > + # check that the loaded library is found through the /proc/PID/fd > + # rather than /proc/self/fd. > + # > + # We need to handle the possibility of the interesting event > + # showing the first, or second time we stop, as depending on which > + # mechanism GDB is using to handle the shared library events (the > + # newer probes based interface, or the old non-probes interface), > + # the library will be reported at the first or second stop. > + set saw_expected_event false > + set saw_no_event_stop false > + gdb_test_multiple "continue" "continue to solib evnt" { > + -re "^continue\r\n" { > + exp_continue > + } > + > + -re "^Continuing\\.\r\n" { > + exp_continue > + } > + > + -early -re "Stopped due to shared library event \\(no libraries added or removed\\)\r\n" { > + # This non-interesting event shows up first when using the > + # probes based mechanism for dealing with shared library > + # events. > + # > + # We set a flag here, and, once the prompt has appeared, > + # we send another continue, the next event will contain > + # the information we want. > + set saw_no_event_stop true > + exp_continue > + } > + > + -re "Stopped due to shared library event:\r\n Inferior loaded (?:target:)?/proc/${inferior_pid}/fd/$memfd\r\n" { > + # This event, which includes the information we are > + # looking for, occurs first when using the non-probes > + # based mechanism for handling shared library events, and > + # occurs second when using the probes mechanism. > + # > + # Either way, record here that we say the output we expect. > + set saw_expected_event true > + exp_continue > + } > + > + -re "$::gdb_prompt $" { > + if {$saw_no_event_stop} { > + set saw_no_event_stop false > + send_gdb "continue\n" > + exp_continue > + } else { > + gdb_assert {$saw_expected_event} $gdb_test_name > + } > + } > + } > +} > + > +# First run of the test. > +do_test > + > +# Possible second run of the test. If we are using a remote target then we > +# should consider setting the sysroot to 'target:' and re-running the test. > +if {[target_info exists gdb_protocol] > + && ([target_info gdb_protocol] == "remote" > + || [target_info gdb_protocol] == "extended-remote")} { > + # GDB will already be running after the first call to do_test, so we can > + # take a peek at the current sysroot setting, and decide if we should > + # repeat the test with a different setting. > + > + set new_sysroot "" > + gdb_test_multiple "show sysroot" "" { > + -wrap -re "The current system root is \"\"\\." { > + pass $gdb_test_name > + > + # Repeat the test with 'target:' sysroot. > + set new_sysroot "target:" > + } > + -wrap -re "The current system root is \"target:\"\\." { > + pass $gdb_test_name > + > + # Nothing else to do, we already tested with target: sysroot. > + } > + -re "$gdb_prompt $" { > + pass $gdb_test_name > + > + # If already testing with any other sysroot, we probably should > + # not try to adjust things, so don't do any further testing. > + } > + } > + > + with_test_prefix "sysroot $new_sysroot" { > + if { $new_sysroot != "" } { > + do_test $new_sysroot > + } > + } > +} > > base-commit: 84f9fbe90e5429adb9dee68f04f44c92fa9e2345 > -- > 2.25.4