public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: Florian Weimer <fweimer@redhat.com>
Cc: Adhemerval Zanella via Libc-alpha <libc-alpha@sourceware.org>,
	jma14 <jma14@rice.edu>, Carlos O'Donell <carlos@redhat.com>,
	John Mellor-Crummey <johnmc@rice.edu>
Subject: Re: [PATCH v10 2/4] elf: Fix initial-exec TLS access on audit modules (BZ #28096)
Date: Tue, 18 Jan 2022 10:26:26 -0300	[thread overview]
Message-ID: <45c1d187-3ce8-975d-7d78-644be2f505d5@linaro.org> (raw)
In-Reply-To: <875yqhjiy4.fsf@oldenburg.str.redhat.com>



On 18/01/2022 10:02, Florian Weimer wrote:
> * Adhemerval Zanella:
> 
>> On 18/01/2022 08:33, Florian Weimer wrote:
>>> * Adhemerval Zanella via Libc-alpha:
>>>
>>>> diff --git a/elf/dl-tls.c b/elf/dl-tls.c
>>>> index 8ba70c9a9d..8ed91ff599 100644
>>>> --- a/elf/dl-tls.c
>>>> +++ b/elf/dl-tls.c
>>>> @@ -520,7 +520,7 @@ _dl_resize_dtv (dtv_t *dtv, size_t max_modid)
>>>>  
>>>>  
>>>>  void *
>>>> -_dl_allocate_tls_init (void *result)
>>>> +_dl_allocate_tls_init (void *result, bool init_tls)
>>>>  {
>>>>    if (result == NULL)
>>>>      /* The memory allocation failed.  */
>>>> @@ -593,7 +593,14 @@ _dl_allocate_tls_init (void *result)
>>>>  	     some platforms use in static programs requires it.  */
>>>>  	  dtv[map->l_tls_modid].pointer.val = dest;
>>>>  
>>>> -	  /* Copy the initialization image and clear the BSS part.  */
>>>> +	  /* Copy the initialization image and clear the BSS part.  For
>>>> +	     audit modules or depedencies with initial-exec TLS, we can not
>>>> +	     set the initial TLS image on default loader initialization
>>>> +	     because it would already be set by the audit setup.  However,
>>>> +	     subsequent thread creation would need to follow the default
>>>> +	     behaviour.   */
>>>> +	  if (__glibc_unlikely (map->l_auditing && !init_tls))
>>>> +	    continue;
>>>>  	  memset (__mempcpy (dest, map->l_tls_initimage,
>>>>  			     map->l_tls_initimage_size), '\0',
>>>>  		  map->l_tls_blocksize - map->l_tls_initimage_size);
>>>
>>> I don't understand why the map->l_auditing condition is correct.
>>> Shouldn't it be
>>>
>>>   map->l_ns == LM_ID_BASE && !init_tls
> 
> Not sure if this is what I meant, the “continue” inverts the condition.
> 
> The initialization should happen for pthread_create or l_ns == LM_ID_BASE.
> 
>>> ?  Everything else has been loaded via dlopen or dlmopen, so TLS
>>> initialization for the main thread has already happened.
>>
>> This is not suffice (I tried before in fact) because we need to skip
>> solely for the audit modules and dependencies at the startup.  For
>> instance with tst-audit21 we have with some _dl_debug_printf to 
>> show what is happening
>>
>>    1533769:     [rtld.c:966] load_audit_module
>>    1533769:     [rtld.c:973] load_audit_module
>>    1533769:     [rtld.c:2465] dl_main
>>    1533769:     [../elf/dl-tls.c:603] map->l_name= map->l_ns=0 map->l_auditing=0 init_tls=0 [INIT]
>>    1533769:     [../elf/dl-tls.c:603] map->l_name=[...]/elf/tst-auditmod21a.so map->l_ns=1 map->l_auditing=1 init_tls=0
>>    1533769:     [../elf/dl-tls.c:603] map->l_name=[...]/libc.so.6 map->l_ns=1 map->l_auditing=1 init_tls=0
>>    1533769:     [../elf/dl-tls.c:603] map->l_name=[...]/libc.so.6 map->l_ns=0 map->l_auditing=0 init_tls=0 [INIT]
>>    1533769:     [../elf/dl-tls.c:621]
>>    1533769:     [rtld.c:2467] dl_main
>> [allocatestack.c:435] allocate_stack
>>    1533769:     [../elf/dl-tls.c:633] _dl_allocate_tls
>>    1533769:     [../elf/dl-tls.c:603] map->l_name= map->l_ns=0 map->l_auditing=0 init_tls=1 [INIT]
>>    1533769:     [../elf/dl-tls.c:603] [...]/elf/tst-auditmod21a.so map->l_ns=1 map->l_auditing=1 init_tls=1 [INIT]
>>    1533769:     [../elf/dl-tls.c:603] [...]/libc.so.6 map->l_ns=1 map->l_auditing=1 init_tls=1 [INIT]
>>    1533769:     [../elf/dl-tls.c:603] [...]/libc.so.6 map->l_ns=0 map->l_auditing=0 init_tls=1 [INIT]
>>    1533769:     [../elf/dl-tls.c:621]
>>    1533769:     [../elf/dl-tls.c:642] _dl_allocate_tls
>> [allocatestack.c:446] allocate_stack
>>
>>
>> So we need to avoid initialization only for the libc.so loading by
>> audit libraries, the default one should still be initialized (marked
>> as [INIT]).  Later when new threads are created we need to initialized
>> everything.
> 
> That's what I tried to express with the LM_ID_BASE check.

The problem is 'map->l_ns == LM_ID_BASE && !init_tls' will not initilize
the required libc.so for main application during audit module loading.
With tst-audit21:

   1593586:     [rtld.c:965] load_audit_module
   1593586:     [rtld.c:972] load_audit_module
   1593586:     [rtld.c:2464] dl_main
   1593586:     [../elf/dl-tls.c:605] map->l_name= map->l_ns=0 map->l_auditing=0 init_tls=0
   1593586:     [../elf/dl-tls.c:610] map->l_name=[...]/elf/tst-auditmod21a.so map->l_ns=1 map->l_auditing=1 init_tls=0 [INIT]
   1593586:     [../elf/dl-tls.c:610] map->l_name=[...]/libc.so.6 map->l_ns=1 map->l_auditing=1 init_tls=0 [INIT]
   1593586:     [../elf/dl-tls.c:605] map->l_name=[...]/libc.so.6 map->l_ns=0 map->l_auditing=0 init_tls=0
   1593586:     [../elf/dl-tls.c:627]
   1593586:     [rtld.c:2466] dl_main
Segmentation fault (core dumped)


  reply	other threads:[~2022-01-18 13:26 UTC|newest]

Thread overview: 21+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-01-17 21:31 [PATCH v10 0/4] Multiple rtld-audit fixes Adhemerval Zanella
2022-01-17 21:31 ` [PATCH v10 1/4] elf: Add la_activity during application exit Adhemerval Zanella
2022-01-18 11:29   ` Florian Weimer
2022-01-18 13:10     ` Adhemerval Zanella
2022-01-21 10:50       ` Florian Weimer
2022-01-21 10:50   ` Florian Weimer
2022-01-24 12:53     ` Adhemerval Zanella
2022-01-17 21:31 ` [PATCH v10 2/4] elf: Fix initial-exec TLS access on audit modules (BZ #28096) Adhemerval Zanella
2022-01-18 11:33   ` Florian Weimer
2022-01-18 12:58     ` Adhemerval Zanella
2022-01-18 13:02       ` Florian Weimer
2022-01-18 13:26         ` Adhemerval Zanella [this message]
2022-01-18 13:29           ` Florian Weimer
2022-01-18 13:33             ` Adhemerval Zanella
2022-01-18 13:46               ` Adhemerval Zanella
2022-01-18 13:47                 ` Adhemerval Zanella
2022-01-17 21:31 ` [PATCH v10 3/4] elf: Issue la_symbind for bind-now (BZ #23734) Adhemerval Zanella
2022-01-21 11:18   ` Florian Weimer
2022-01-17 21:31 ` [PATCH v10 4/4] elf: Fix runtime linker auditing on aarch64 (BZ #26643) Adhemerval Zanella
2022-01-18 13:36   ` Szabolcs Nagy
2022-01-21 11:19   ` Florian Weimer

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=45c1d187-3ce8-975d-7d78-644be2f505d5@linaro.org \
    --to=adhemerval.zanella@linaro.org \
    --cc=carlos@redhat.com \
    --cc=fweimer@redhat.com \
    --cc=jma14@rice.edu \
    --cc=johnmc@rice.edu \
    --cc=libc-alpha@sourceware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).