Kevin McCarthy [Fri, 1 Jan 2016 17:48:51 +0000 (09:48 -0800)]
Fix segfault when deleting and reusing attachment slots. (closes #3800)
When attachments are deleted, delete_attachment() slides the entries
down in the idx array, but forgets to NULL out the last vacated slot.
If more attachments are added later on via OP_COMPOSE_EDIT_HEADERS and
the Attach: pseudo-header, mutt_gen_attach_list() will attempt to
re-use the ATTACHPTR in that last slot because it wasn't set to NULL.
This will be pointing to freed memory and likely segfault (at best).
Kevin McCarthy [Tue, 8 Dec 2015 17:11:30 +0000 (09:11 -0800)]
Fix hash table key "use after free" in mh_check_mailbox(). (closes #3797)
The fnames hash uses the maildir->header->path as the key. As matches
are found, the headers are freed. This inadvertantly also freed the key
to the hashtable entry; the next hash_find() going to the same bucket
might end up comparing keys with a freed string.
This patch stores the path in the struct maildir canon_fname field (just
as maildir_check_mailbox() does) and uses that as the hash key instead.
This field isn't used outside of maildir_check_mailbox(), and is
automatically freed for us in the maildir_move_to_context() call at the
bottom of both functions.
Note there are other ways to fix this problem:
- Add a new mode to the hash table, causing it to strdup the keys and
free them itself.
- Delete the entries in the fnames hash, rather leaving them there.
The first seems the cleanest, but would end up touching much more code.
The second is also clean, but might have a negative performance impact.
Additionally, peeking back in history to changeset 1d45a50b6f9b, it
looks like the canon_fname used to be used by mh too, so perhaps
removing the strdup may have been a mistake during refactoring at some
point.
Kevin McCarthy [Thu, 3 Dec 2015 23:23:34 +0000 (15:23 -0800)]
Provide a better prompt and error for inline PGP with attachments. (closes #3738)
Change mutt_protect() to check for text/plain before trying to invoke
crypt_pgp_traditional_encryptsign(). This way, mutt can provide a bit
more specific prompt and error message.
Since pgp_mime_auto says it will prompt in the event of any failure,
keep the more generic prompt after the encryptsign call too.
Kevin McCarthy [Wed, 2 Dec 2015 02:20:27 +0000 (18:20 -0800)]
Loosen mutt_signed_handler() protocol value consistency check. (closes #3639)
Apparently, for S/MIME, some MUAs mismatch the protocol value of
the multipart/signed and the content-type of the signature: putting
"pkcs7-signature" in one and "x-pkcs7-signature" in the other.
Change mutt_signed_handler() to independently verify the values of the
protocol and the content-type. This still checks for correct values but
doesn't ensure they match between the two (for S/MIME).
Kevin McCarthy [Mon, 30 Nov 2015 23:52:30 +0000 (15:52 -0800)]
smime: allow signing message digest algorithm to be specified.
Currently, Mutt hardcodes micalg=sha1 for signed messages.
Unfortunately, the actual message digest algorithm used defaults to
the value in the "Signature Algorithm" field in the signing key's
certificate.
Add a new configuration option $smime_sign_digest_alg, defaulting
to sha256. Add a new printf format string, %d, to be used in the
signing command to specify the digest algorithm. Modify the sample
$smime_sign_command to include "-md %d".
Note: This solution requires using the modified $smime_sign_command,
or else the micalg parameter again may not match the algorithm used.
An alternative solution would be to query the certificate "Signature
Algorithm" field and try to change the micalg to match it, but this
method is easier to implement and provides better control for the user
to configure, in any case.
Kevin McCarthy [Thu, 26 Nov 2015 19:01:19 +0000 (11:01 -0800)]
Clean up address_uses_unicode() (closes #3794)
Pull the null check out of the loop. Use a bit comparison to detect if
the high bit is set: this avoids a warning for platforms where char is
implicitly signed (where comparing < 128 is always true).
Kevin McCarthy [Wed, 25 Nov 2015 05:45:58 +0000 (21:45 -0800)]
Fix bad idn error on local mailboxes. (closes #3795)
Commit 831abf39d53a pulled the mbox_to_udomain() call inside the
conversion functions. Unfortunately, this causes local (user only)
mailboxes to be considered conversion errors instead of just skipping
them.
Revert mbox_to_udomain() back to using a static buffer and pull back
into the mutt_addrlist_to_local/intl() functions.
Pass the user and domain into the conversion functions instead of the address.
Kevin McCarthy [Tue, 24 Nov 2015 23:49:26 +0000 (15:49 -0800)]
Rename idna functions and bits for smtputf8 changes.
This is patch 1 of 4 implementing support for SMTPUTF8 (RFC 6531).
Change mutt_idna.c to be always compiled. Remove the stub functions in
mutt_idna.h. Instead, put #ifdefs around the idna function calls. The
conversion functions will be fixed up in the next patch.
Rename the conversion functions to mutt_addrlist_to_intl() and
mutt_env_to_intl(). Rename the ADDRESS idna status bits to "intl"
status bits.
Kevin McCarthy [Sat, 21 Nov 2015 23:28:57 +0000 (15:28 -0800)]
Remove redundant mbox delimiter check in imap_browse(). (closes #3646)
imap_fix_path() removes duplicate and trailing delimiters, so the check
below it was redundant. This also made it appear list.delim could be
used uninitialized.
Remove the check, but add a check to make sure the "fixed" path has
len>0, to prevent oob accesses of mbox[n-1] below.
Lastly, remove a redundant n=strlen(mbox) inside the initial LIST
processing loop. The mbox isn't changed from above, so there is no
need to rerun strlen.
Kevin McCarthy [Tue, 10 Nov 2015 02:53:05 +0000 (18:53 -0800)]
Improve error messages for eat_date() and eat_regexp().
After calling mutt_extract_token(), s->dptr will typically be at the end
of the string if an error occurred. Save a pointer to the beginning of
the expression, so it can be properly displayed in the error message.
Convert eat_date() and eat_regexp() to use the same iconv strings for
the error message.
Kevin McCarthy [Mon, 9 Nov 2015 23:40:42 +0000 (15:40 -0800)]
Fix possible unintentional '\0' strchr matches.
After fixing the ticket 3787 strchr issue, this patch cleans up other
potentially incorrect uses of strchr for the '\0' case.
In mutt_multi_choice(), mutt_getch() can technically return 0. Although
it seems the user would have to try quite hard to do this, it's
incorrect to return that index into letters. Change "ch.ch==0" to be
considered the same as an abort.
is_email_wsp() is used in a couple places where it wasn't obvious
whether '\0' was being accounted for, so add an explicit check to the
function.
Inside eat_date(), if mutt_extract_token() had no input and returned
"", the strchr ("<>=", buffer.data[0]) below would return a pointer.
In actuality, this is prevented by an empty parameter check inside
mutt_pattern_comp(), but it doesn't hurt to make it the same as
eat_regexp() and have the check explicitly done here too.
rfc2047_encode() was another borderline case for adding a check. The
convert_string() sets a length, so it seems highly unlikely that *t
could be 0, but doesn't hurt to add the check.
The find_encoded_word() fix looks necessary. If the passed in s was
something like "=?charset?" (followed by EOS, '\0'), the strchr("BbQq",
q[1]) would in fact return a pointer and the following q[2] would read
past the end of string. If q[2] happened to be '?', it might even
continue reading in the for loop below.
Lastly, in parse_mailboxdomain(), the potential overread was already
fixed in changeset:a6919571eb59, but although the nonspecial and special
strchr() line happens to "work" for the case of '\0', it's pretty
fragile to leave as is. It's better to be explicit and just return if
we hit EOS without calling next_token().
Kevin McCarthy [Sun, 18 Oct 2015 11:45:51 +0000 (19:45 +0800)]
Fix next_token() oob read. (closes #3787)
With specially crafted input to 'mutt -H', the line "Return-Path:<() "
is read and passed to mutt_parse_rfc822_line(). "<() " is then passed
through to rfc822_parse_adrlist().
Eventually, inside next_token(), is_special(*s) is called when s
points to the end of the string ('\0'). This macro calls strchr,
which will actually match and return a pointer to the trailing '\0' in
RFC822Specials! This causes "s + 1" to be returned, skipping past the
end of string inside parse_mailboxdomain().
This patch adds a check to make sure *s is non-null before calling
is_special(*s).
Kevin McCarthy [Sat, 17 Oct 2015 03:15:01 +0000 (11:15 +0800)]
Fix error message for attach-message. (closes #3785)
Currently if mx_open_mailbox() fails when trying to attach a message,
mutt_perror() is called. Change this to call mutt_error() instead,
since errno isn't set for all failure cases.
Kevin McCarthy [Sun, 4 Oct 2015 02:08:49 +0000 (10:08 +0800)]
Create a separate macro/push/exec event buffer. (closes #3779)
Currently, the SSL and TLS certficate prompts turn on
OPTUNBUFFEREDINPUT, (to prevent macros and such from running right
through the dialog). Unfortunately, the menu dialog processing in
menu_dialog_dokey() is using mutt_ungetch() to forward non-dialog keys
on to standard menu processing. With OPTUNBUFFEREDINPUT set, those keys
never make it to the menu and are buffered until after the menu dialog.
This patch creates a new event buffer, separate from the standard
"unget" buffer, for use by macros, exec, and push events. These events
can be temporarily ignored by setting OPTIGNOREMACROEVENTS (renamed
from OPTUNBUFFEREDINPUT), while continuing to allow unget events to be
processed.
Since the "push" and "unget" functions now go to different buffers,
function names were slightly renamed, to make it less easy to
unintentionally use the wrong function at the wrong time.
Kevin McCarthy [Sun, 4 Oct 2015 02:08:45 +0000 (10:08 +0800)]
Fix menu type in certificate prompt. (see #3779)
The menu type is used in several places as a direct index into
Keymaps[], so passing in -1 to mutt_new_menu() was leading to illegal
memory accesses later on.
Add a range check in mutt_new_menu(), defaulting to MENU_GENERIC, to
prevent this problem in the future.
Kevin McCarthy [Wed, 30 Sep 2015 03:25:28 +0000 (11:25 +0800)]
Fix chomp in smime_handle_cert_email.
During a review of the previous patch, Oswald Buddenhagen noticed two
of the fixed oob reads had another problem: they were "chomping" (the
newline) without verifying there actually was a newline at the end of
the string.
Kevin McCarthy [Wed, 30 Sep 2015 03:21:06 +0000 (11:21 +0800)]
Fix oob reads when fgets returns "\0". (closes #3776)
The ticket reported an out of bounds read in mutt_read_rfc822_line()
when a '\0' was embedded on its own line in the headers. The function
assumed if fgets() didn't return NULL, then the string would have at
least one character.
I scanned the rest of the code and found three other places making the
same assumption for fgets.
Thanks to hanno for finding this with the "american fuzzy lop" tool.
Kevin McCarthy [Sun, 6 Sep 2015 23:31:57 +0000 (16:31 -0700)]
Fix double-decode during IMAP browse.
cmd_parse_list() already calls imap_unmunge_mbox_name() on the mailbox
names returned from the server. However, browse_add_list_result() was
taking those mailbox names and passing them to imap_add_folder(), which
was calling imap_unmunge_mbox_name() yet again.
The reason is that imap_browse() was directly calling imap_add_folder()
too, passing in a previously encoded "mbox" name. After looking
carefully at the code, I could find no reason that mbox needed to
be encoded outside of the LIST commands.. Therefore I changed
imap_browse() to call imap_munge_mbox_name() on mbox for the
two LIST commands generated from it instead, and removed the
imap_unmunge_mbox_name() call inside imap_add_folder().
Kevin McCarthy [Sun, 6 Sep 2015 14:40:06 +0000 (07:40 -0700)]
Fix use after free of ctx->last_tag. (closes #3775)
When using imap to access gmail, tagging and saving messages to "all
mail" and pressing <sync-mailbox> can result in the call path:
mx_check_mailbox()
imap_check_mailbox()
imap_cmd_finish()
imap_expunge_mailbox()
mx_update_tables()
followed by:
mx_sync_mailbox()
The HEADER pointed to by ctx->last_tag will be removed and FREE'ed in
mx_update_tables(), but will subsequently be accessed in mx_sync_mailbox().
This patch simply sets ctx->last_tag=NULL if it is freed inside mx_update_tables().
Thanks to Peter Lekensteyn for the bug report and ASAN report.
Kevin McCarthy [Thu, 3 Sep 2015 01:11:28 +0000 (18:11 -0700)]
Add time_t conversion to fix gpgme segfault on OpenBSD.
time_t isn't the same size as gpgme_subkey_t->timestamp on OpenBSD.
Passing &subkey->timestamp to localtime was therefore passing an address
to the wrong size int and was causing a segfault.
Thanks to Hannes Wenzel for reporting the bug and providing a patch.
Kevin McCarthy [Mon, 17 Aug 2015 21:51:07 +0000 (14:51 -0700)]
Add idn to MUTTLIBS instead of LIBS (see #3638)
Currently, -lidn is included while linking all the mutt binaries. Add
it to MUTTLIBS instead so it's only used for the mutt linking.
Also, add $LIBICONV to $LIBS before checking for some of the idna
functions. On some sytems, there are apparently issues when using
static linking. (Thanks to grarpamp for reporting).
Kevin McCarthy [Mon, 17 Aug 2015 19:31:40 +0000 (12:31 -0700)]
Change mutt_adv_mktemp to call mutt_mktemp instead of mktemp. (see #3638).
mutt_mktemp is currently called in 95% of the cases in mutt, and
is already our "own rolled" version of mktemp. The "insecure mktemp
warning" discussion keeps coming up, so instead add prefix and suffix
functionality to mutt_mktemp() and call that.
All other uses of Tempdir in the mutt source did not call
mutt_expand_path() first, so remove that from mutt_adv_mktemp().
Kevin McCarthy [Wed, 29 Jul 2015 16:07:51 +0000 (09:07 -0700)]
Add error handling for ^ and other empty mailbox shortcuts.
(closes #2402) (closes #3735)
Explicitly mention the ^ example in the documentation added in 6d733cab6b45.
Add an error message for ^ when CurrentFolder is not set. Add checks
for other mailbox shortcuts that expand to the empty string. This
could happen if the @alias shortcut was accidentally used, or the value
referenced by a shortcut isn't set yet.
Kevin McCarthy [Mon, 27 Jul 2015 21:09:55 +0000 (14:09 -0700)]
Add "Mailbox Matching in Hooks" section to manual.
Folder-hook and mbox-hook perform mailbox shortcut expansion on the
regexp parameter. Add a section to the manual to give examples and make
the behavior clearer.
In certain circumstances, Exchange corrupts a multipart/encrypted block
into:
<multipart/mixed>
<text/plain>
<application/pgp-encrypted> [BASE64-encoded]
<application/octet-stream> [BASE64-encoded]
This patch pulls the full detection of valid/invalid multiparts
into mutt_body_handler(). It extracts a run_decode_and_handler()
function, which is reused by new intermediate handlers to decode
the application/octet-stream part before passing it directly to
crypt_pgp_encrypted_handler. These intermediate handlers then check
and set any GOODSIG flags back into the parent part.
This change may result in less error messages for invalid
multipart/encrypted parts. Instead, mutt will default to the
multipart_handler if it isn't fully "correct".
Viewing attachments uses crypt_pgp_decrypt_mime() which bypasses the
handler mechanism. Add decoding to the decrypt_mime() functions for pgp
and gpgme.
Thanks to Vincent Brillault for his analysis and initial patch.
Kevin McCarthy [Tue, 21 Jul 2015 19:34:49 +0000 (12:34 -0700)]
Imap: Fix flag caching after sync.
The flags cached in IMAP_HEADER_DATA were not updated to match the
HEADER flags after a sync. This means if a flag were toggled and synced
twice, the second sync was not sending the flag update.
Kevin McCarthy [Thu, 23 Jul 2015 21:57:04 +0000 (14:57 -0700)]
Fix compiler type warnings. (closes #3765)
The output of mutt_local_tz() was being passed to abs(). Technically
the return type is time_t, but it represents a small value: the timezone
offset in seconds. Add a safe explicit cast to int.
Change the txt parameter of mutt_make_help() to type const char *.
Typically all calls run the txt parameter through _(), which
accepts const char * and returns a char *. However, if NLS is not
enabled, _() is a noop, simply returning the parameter itself. In
mutt_compile_help(), items[i].name is const char *, so it will generate
a warning when passed as the txt parameter of mutt_make_help().
On some systems, e.g. OS X, snprintf is defined as a macro. One call
in hcache.c was embedding directives inside the snprintf call. This is
apparently undefined behavior, so duplicate the call instead.
Kevin McCarthy [Thu, 9 Jul 2015 21:50:05 +0000 (14:50 -0700)]
Use $(VAR) instead of @VAR@ in Makefile.am files. (closes #3664)
The @VAR@ form is not overridable, such as the case for #3664 where the
reporter wanted to override DOTLOCK_GROUP.
It's doubtful targets need to be overriden, but it makes sense to be
consistent in the usage of automake substituted variables unless there
is a particular reason/bug to work around.
Kevin McCarthy [Thu, 2 Jul 2015 18:21:19 +0000 (11:21 -0700)]
Convert from using mkinstalldirs to $(MKDIR_P).
The automake mkinstalldirs script is now deprecated. "install-sh -d"
can be used in place. Configure.ac already includes AC_PROG_INSTALL,
which will ensure install-sh is bundled. Add AC_PROG_MKDIR_P, which
will set $(MKDIR_P) to either a thread-safe "mkdir -p" or will fall back
to using install-sh.
Kevin McCarthy [Mon, 29 Jun 2015 21:35:28 +0000 (14:35 -0700)]
Add note to manual clarifying push/exec and folder-hook order.
Folder-hooks are executed in the order declared. However, push/exec
commands will simply prepend to the keyboard buffer. If there are
multiple push/exec commands, each will prepend to the earlier one. The
result is that they will end up being "run" in reverse order once keyboard
buffer processing starts.
Kevin McCarthy [Fri, 26 Jun 2015 19:23:26 +0000 (12:23 -0700)]
Fix IMAP segfault due to NULL capstr.
After a failed login, the connection is left open but capstr is freed.
If a second login attempt is made, imap_auth_sasl was trying to strstr
using the NULL capstr.
Add a NONULL around the capstr parameter to strstr. Change
imap_conn_find() to keep the capstr around until a successful
authentication occurs.
Model the smime.c searching off of classic pgp searching. Create
smime_get_key_by_hash()/addr()/str() functions that use a single
smime_get_candidates() function to search the index.
Use the new smime_keys generated purpose flag in the index to filter by
KEYFLAG_CANENCRYPT and KEYFLAG_CANSIGN. Old style indices fall back to
all-inclusive behavior.
Kevin McCarthy [Sun, 7 Jun 2015 20:59:19 +0000 (13:59 -0700)]
smime_keys: Add purpose flag to index.
A subsequent patch will change smime.c to use the purpose to filter
certs and keys that can't be used for signing or encryption.
Issuer, trust, and purpose flags were also added to the keys index file,
to keep parsing simpler in smime_keys as well as smime.c. The trust and
purpose are derived from the cert. Issuer is set to '?' as this should
never be needed for keys.
Kevin McCarthy [Mon, 25 May 2015 17:59:50 +0000 (10:59 -0700)]
smime_keys: Handle certificate chains in add_cert. (closes #3339) (closes #3559)
Find all chains in the certificate provided. For each chain create a
separate leaf and intermediate certificate file. Because Mutt controls
the label prompt, use a single label for all chains.
Also, loosen up cert file parsing to allow attributes even if they
aren't delimited by a "Bag Attributes" header.
Thanks to David J. Weller-Fahy for his testing and feedback!
Kevin McCarthy [Mon, 18 May 2015 20:27:12 +0000 (13:27 -0700)]
smime_keys: Convert openssl execution to use open("-|",...). (see #3575) (see #2456)
This does a fork/exec, bypassing the shell, and so handles
spaces, quotes, and other shell-characters problems better than the
simple fix in changeset:c66a6bd5d0d5
This also fixes the "verify with crl" bug in #2456: the grep is now done
in perl.
Thank you Vincent Lefevre for your review and feedback!
Kevin McCarthy [Fri, 15 May 2015 17:47:38 +0000 (10:47 -0700)]
Start cleaning up and fixing smime_keys.pl (closes #3324) (see #2456)
* Convert to using File::Temp (#3324). This was also suggested at
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=775199
* Use File::Temp for add_p12 temp file. (#2456)
* Make the query_label() method a bit more robust with empty strings,
ctrl-d, and leading spaces.
* Clean up openssl_do_verify() logic. Mark cert as invalid
rather that die'ing if an openssl verify command fails.
* General cleanup:
- Clearly separate op handler, certificate management, and helper
functions by section and using prefixes.
- Create openssl helper functions to reduce copy/paste invocations
and make the code clearer.
- Make indentation consistent at 2 spaces.
- Change handle_add_pem() to re-use handle_add_chain() once the
correct files are identified.
- Change openssl_parse_pem() to return a single array of data
structures representing the parsed certs/keys.
Kevin McCarthy [Sun, 3 May 2015 23:25:45 +0000 (16:25 -0700)]
Provide SSL cipher selection option. (closes #3167)
Creates a $ssl_ciphers option that allows direct selection of the
ciphers for OpenSSL (via SSL_CTX_set_cipher_list) and GnuTLS (via
gnutls_priority_set_direct).
Fix performance regression for ~b/~B searching. (closes #3743)
In mutt_is_autoview(), changeset b58cdfacfb89 introduced a call to
rfc1524_mailcap_lookup() before checking if the MIME type should
be autoviewed based on the user's preferences. This caused a major
performance regression for ~b/~B searching.
Rearrange mutt_is_autoview() to check the user preferences first, then
search for a mailcap entry only if the MIME type should be autoviewed.
In order to preserve correct mime_lookup behavior, re-add a call to
mutt_check_lookup_list() before scanning the AutoViewList.
Kevin McCarthy [Sun, 19 Apr 2015 20:15:50 +0000 (13:15 -0700)]
Add $crypt_confirmhook option. (see #3727)
Allow the confirmation prompt for crypt-hooks to be disabled. This is
useful for multiple crypt-hook users (e.g. encrypted mailing lists), or
just for people who are certain of their crypt-hooks and don't want to
be prompted every time.
Kevin McCarthy [Sun, 19 Apr 2015 20:15:48 +0000 (13:15 -0700)]
Allow multiple crypt-hooks with the same regexp. (closes #3727).
Changes the crypt-hook to accumulate a LIST of hooks with
the same regexp, as opposed to replacing the hook data.
This is useful for the case of encrypted mailing lists.
Update pgp classic and gpgme to process a LIST of crypt-hook
values instead of just one.
This version of the patch creates a new _mutt_list_hook() function that
(in theory) other hooks could use if they were changed to return a list.
It also changes the behavior when a crypt-hook is declined: previously
it would immediately use the original recipient for key selection. Now
it will only do that if all crypt-hooks for a recipient are declined.
This allows all, a subset, or none of the hooks to be used.
Thanks to Rejo Zenger, Remco Rijnders, and Dale Woolridge for their work
on various versions of this patch.
Kevin McCarthy [Thu, 16 Apr 2015 01:57:17 +0000 (18:57 -0700)]
Use fingerprints instead of keyIDs internally. (see #3695)
Add a helper function, pgp_fpr_or_lkeyid(), that returns the fingerprint
if available, otherwise falls back to the long key id. Convert Mutt to
use that value for pgp command invocation.
Change gpgme to use an equivalent crypt_fpr_or_lkeyid() function in a
couple places too (for keylist generation and sign-as key selection).
Update documentation to mention fingerprints and the --with-fingerprint
option for gpg invocation.
Change pgp_long_ids to default: yes, but add a note mentioning
it's only used for the display of key IDs in a few places.