| Message ID | 20240530150057.444585-1-hadess@hadess.net |
|---|---|
| Headers | show |
| Series | Fix a number of static analysis issues #3 | expand |
Hi, to, 2024-05-30 kello 16:58 +0200, Bastien Nocera kirjoitti: > Don't manipulate the "req" structs as if they were flat arrays, static > analysis and humans are both equally confused by this kind of usage. struct start_req { union { struct seid required[1]; struct seid seids[0]; }; } __attribute__ ((packed)); and access only via req->seids? > > Error: ARRAY_VS_SINGLETON (CWE-119): [#def26] [important] > bluez-5.76/profiles/audio/avdtp.c:1675:2: address_of: Taking address with "&start->first_seid" yields a singleton pointer. > bluez-5.76/profiles/audio/avdtp.c:1675:2: assign: Assigning: "seid" = "&start->first_seid". > bluez-5.76/profiles/audio/avdtp.c:1679:25: ptr_arith: Using "seid" as an array. This might corrupt or misinterpret adjacent memory locations. > 1677| int i; > 1678| > 1679|-> for (i = 0; i < count; i++, seid++) { > 1680| if (seid->seid == id) { > 1681| req->collided = TRUE; > > Error: ARRAY_VS_SINGLETON (CWE-119): [#def27] [important] > bluez-5.76/profiles/audio/avdtp.c:1690:2: address_of: Taking address with "&suspend->first_seid" yields a singleton pointer. > bluez-5.76/profiles/audio/avdtp.c:1690:2: assign: Assigning: "seid" = "&suspend->first_seid". > bluez-5.76/profiles/audio/avdtp.c:1694:25: ptr_arith: Using "seid" as an array. This might corrupt or misinterpret adjacent memory locations. > 1692| int i; > 1693| > 1694|-> for (i = 0; i < count; i++, seid++) { > 1695| if (seid->seid == id) { > 1696| req->collided = TRUE; > > Error: ARRAY_VS_SINGLETON (CWE-119): [#def28] [important] > bluez-5.76/profiles/audio/avdtp.c:1799:2: address_of: Taking address with "&req->first_seid" yields a singleton pointer. > bluez-5.76/profiles/audio/avdtp.c:1799:2: assign: Assigning: "seid" = "&req->first_seid". > bluez-5.76/profiles/audio/avdtp.c:1801:30: ptr_arith: Using "seid" as an array. This might corrupt or misinterpret adjacent memory locations. > 1799| seid = &req->first_seid; > 1800| > 1801|-> for (i = 0; i < seid_count; i++, seid++) { > 1802| failed_seid = seid->seid; > 1803| > > Error: ARRAY_VS_SINGLETON (CWE-119): [#def29] [important] > bluez-5.76/profiles/audio/avdtp.c:1912:2: address_of: Taking address with "&req->first_seid" yields a singleton pointer. > bluez-5.76/profiles/audio/avdtp.c:1912:2: assign: Assigning: "seid" = "&req->first_seid". > bluez-5.76/profiles/audio/avdtp.c:1914:30: ptr_arith: Using "seid" as an array. This might corrupt or misinterpret adjacent memory locations. > 1912| seid = &req->first_seid; > 1913| > 1914|-> for (i = 0; i < seid_count; i++, seid++) { > 1915| failed_seid = seid->seid; > 1916| > --- > profiles/audio/avdtp.c | 33 ++++++++++++++++++++++----------- > 1 file changed, 22 insertions(+), 11 deletions(-) > > diff --git a/profiles/audio/avdtp.c b/profiles/audio/avdtp.c > index 3667e08400dd..38c1870e619d 100644 > --- a/profiles/audio/avdtp.c > +++ b/profiles/audio/avdtp.c > @@ -429,6 +429,20 @@ static void avdtp_sep_set_state(struct avdtp *session, > struct avdtp_local_sep *sep, > avdtp_state_t state); > > +#define REQ_GET_NTH_SEID(x) \ > + static struct seid * \ > + x##_req_get_nth_seid(struct x##_req *req, int count, int i) \ > + { \ > + if (count == 0 || i >= count) \ > + return NULL; \ > + if (i == 1) \ > + return &req->first_seid; \ > + return &req->other_seids[i]; \ (i == 0) and [i - 1]? > + } > + > +REQ_GET_NTH_SEID(start) > +REQ_GET_NTH_SEID(suspend) > + > static const char *avdtp_statestr(avdtp_state_t state) > { > switch (state) { > @@ -1672,11 +1686,11 @@ static void check_seid_collision(struct pending_req *req, uint8_t id) > static void check_start_collision(struct pending_req *req, uint8_t id) > { > struct start_req *start = req->data; > - struct seid *seid = &start->first_seid; > int count = 1 + req->data_size - sizeof(struct start_req); > int i; > > - for (i = 0; i < count; i++, seid++) { > + for (i = 0; i < count; i++) { > + struct seid *seid = start_req_get_nth_seid(start, count, i); > if (seid->seid == id) { > req->collided = TRUE; > return; > @@ -1687,11 +1701,11 @@ static void check_start_collision(struct pending_req *req, uint8_t id) > static void check_suspend_collision(struct pending_req *req, uint8_t id) > { > struct suspend_req *suspend = req->data; > - struct seid *seid = &suspend->first_seid; > int count = 1 + req->data_size - sizeof(struct suspend_req); > int i; > > - for (i = 0; i < count; i++, seid++) { > + for (i = 0; i < count; i++) { > + struct seid *seid = suspend_req_get_nth_seid(suspend, count, i); > if (seid->seid == id) { > req->collided = TRUE; > return; > @@ -1785,7 +1799,6 @@ static gboolean avdtp_start_cmd(struct avdtp *session, uint8_t transaction, > struct avdtp_local_sep *sep; > struct avdtp_stream *stream; > struct stream_rej rej; > - struct seid *seid; > uint8_t err, failed_seid; > int seid_count, i; > > @@ -1796,9 +1809,9 @@ static gboolean avdtp_start_cmd(struct avdtp *session, uint8_t transaction, > > seid_count = 1 + size - sizeof(struct start_req); > > - seid = &req->first_seid; > + for (i = 0; i < seid_count; i++) { > + struct seid *seid = start_req_get_nth_seid(req, seid_count, i); > > - for (i = 0; i < seid_count; i++, seid++) { > failed_seid = seid->seid; > > sep = find_local_sep_by_seid(session, seid->seid); > @@ -1898,7 +1911,6 @@ static gboolean avdtp_suspend_cmd(struct avdtp *session, uint8_t transaction, > struct avdtp_local_sep *sep; > struct avdtp_stream *stream; > struct stream_rej rej; > - struct seid *seid; > uint8_t err, failed_seid; > int seid_count, i; > > @@ -1909,9 +1921,8 @@ static gboolean avdtp_suspend_cmd(struct avdtp *session, uint8_t transaction, > > seid_count = 1 + size - sizeof(struct suspend_req); > > - seid = &req->first_seid; > - > - for (i = 0; i < seid_count; i++, seid++) { > + for (i = 0; i < seid_count; i++) { > + struct seid *seid = suspend_req_get_nth_seid(req, seid_count, i); > failed_seid = seid->seid; > > sep = find_local_sep_by_seid(session, seid->seid);
On Thu, 2024-05-30 at 19:57 +0300, Pauli Virtanen wrote: > Hi, > > to, 2024-05-30 kello 16:58 +0200, Bastien Nocera kirjoitti: > > Don't manipulate the "req" structs as if they were flat arrays, > > static > > analysis and humans are both equally confused by this kind of > > usage. > > struct start_req { > union { > struct seid required[1]; > struct seid seids[0]; > }; > } __attribute__ ((packed)); > > and access only via req->seids? That's a good idea, I'll give it a try. > <snip> > > +#define > > REQ_GET_NTH_SEID(x) \ > > + static struct seid > > * \ > > + x##_req_get_nth_seid(struct x##_req *req, int count, int > > i) \ > > + { > > \ > > + if (count == 0 || i >= > > count) \ > > + return > > NULL; \ > > + if (i == > > 1) \ > > + return &req- > > >first_seid; \ > > + return &req- > > >other_seids[i]; \ > > (i == 0) and [i - 1]? *facepalm* Yes, this will need a v2, thanks for spotting that.
Hi Bastien, On Fri, May 31, 2024 at 11:51 AM Bastien Nocera <hadess@hadess.net> wrote: > > On Thu, 2024-05-30 at 19:57 +0300, Pauli Virtanen wrote: > > Hi, > > > > to, 2024-05-30 kello 16:58 +0200, Bastien Nocera kirjoitti: > > > Don't manipulate the "req" structs as if they were flat arrays, > > > static > > > analysis and humans are both equally confused by this kind of > > > usage. > > > > struct start_req { > > union { > > struct seid required[1]; > > struct seid seids[0]; > > }; > > } __attribute__ ((packed)); > > > > and access only via req->seids? > > That's a good idea, I'll give it a try. > > > <snip> > > > +#define > > > REQ_GET_NTH_SEID(x) \ > > > + static struct seid > > > * \ > > > + x##_req_get_nth_seid(struct x##_req *req, int count, int > > > i) \ > > > + { > > > \ > > > + if (count == 0 || i >= > > > count) \ > > > + return > > > NULL; \ > > > + if (i == > > > 1) \ > > > + return &req- > > > >first_seid; \ > > > + return &req- > > > >other_seids[i]; \ > > > > (i == 0) and [i - 1]? > > *facepalm* > > Yes, this will need a v2, thanks for spotting that. Ive applied the set except for this one, please resend once you are done with the suggested changes.
Hello: This series was applied to bluetooth/bluez.git (master) by Luiz Augusto von Dentz <luiz.von.dentz@intel.com>: On Thu, 30 May 2024 16:57:54 +0200 you wrote: > 14 defects fixed, and 1 error check added. > > Let me know whether there's any problems with the implementation, I'm > thinking in particular of the avdtp changes which are pretty invasive. > > Cheers > > [...] Here is the summary with links: - [BlueZ,1/9] rctest: Fix possible overrun https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=24cf04939502 - [BlueZ,2/9] mgmt-tester: Fix buffer overrun https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=aa54087f13d5 - [BlueZ,3/9] l2test: Add missing error checking https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=ccec5e8ef171 - [BlueZ,4/9] rfkill: Avoid using a signed int for an unsigned variable https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=c9fcea121f9a - [BlueZ,5/9] shared/mainloop: Fix integer overflow https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=6cf9117bfd3f - [BlueZ,6/9] sdp: Fix ineffective error guard https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=bd954700e631 - [BlueZ,7/9] obexd: Fix buffer overrun https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=1764cea5c7fd - [BlueZ,8/9] bap: Fix more memory leaks on error https://git.kernel.org/pub/scm/bluetooth/bluez.git/?id=cc8e6ef63509 - [BlueZ,9/9] avdtp: Fix manipulating struct as an array (no matching commit) You are awesome, thank you!
On Mon, 2024-06-03 at 15:19 -0400, Luiz Augusto von Dentz wrote: > Hi Bastien, > > On Fri, May 31, 2024 at 11:51 AM Bastien Nocera <hadess@hadess.net> > wrote: > > > > On Thu, 2024-05-30 at 19:57 +0300, Pauli Virtanen wrote: > > > Hi, > > > > > > to, 2024-05-30 kello 16:58 +0200, Bastien Nocera kirjoitti: > > > > Don't manipulate the "req" structs as if they were flat arrays, > > > > static > > > > analysis and humans are both equally confused by this kind of > > > > usage. > > > > > > struct start_req { > > > union { > > > struct seid required[1]; > > > struct seid seids[0]; > > > }; > > > } __attribute__ ((packed)); > > > > > > and access only via req->seids? > > > > That's a good idea, I'll give it a try. > > > > > <snip> > > > > +#define > > > > REQ_GET_NTH_SEID(x) \ > > > > + static struct seid > > > > * \ > > > > + x##_req_get_nth_seid(struct x##_req *req, int count, int > > > > i) \ > > > > + { > > > > \ > > > > + if (count == 0 || i >= > > > > count) \ > > > > + return > > > > NULL; \ > > > > + if (i == > > > > 1) \ > > > > + return &req- > > > > > first_seid; \ > > > > + return &req- > > > > > other_seids[i]; \ > > > > > > (i == 0) and [i - 1]? > > > > *facepalm* > > > > Yes, this will need a v2, thanks for spotting that. > > Ive applied the set except for this one, please resend once you are > done with the suggested changes. I've redone this patch, it's now in: https://patchwork.kernel.org/project/bluetooth/list/?series=867448 https://lore.kernel.org/linux-bluetooth/20240702084900.773620-3-hadess@hadess.net/T/#u Thanks to Pauli for the help. >