Bluetooth: hidp: make sure input buffers are big enough
Subsystems (3)
PATCH 1/1 Bluetooth: hidp: make sure input buffers are big enough
HID core expects the input buffers to be at least of size 4096 (HID_MAX_BUFFER_SIZE). Other sizes will result in buffer-overflows if an input-report is smaller than advertised. We could, like i2c, compute the biggest report-size instead of using HID_MAX_BUFFER_SIZE, but this will blow up if report-descriptors are changed after ->start() has been called. So lets be safe and just use the biggest buffer we have.
Note that this adds an additional copy to the HIDP input path. If there is a way to make sure the skb-buf is big enough, we should use that instead.
The best way would be to make hid-core honor the @size argument, though, that sounds easier than it is. So lets just fix the buffer-overflows for now and afterwards look for a faster way for all transport drivers.
---
Hi
Any ideas how to improve this patch? I'd like to avoid the extra copy but I have no clue how the skb stuff works exactly.
I also haven't figured out a nice way to make HID-core honor the "size" parameter. hid-input depends on getting the whole input-report.
Comments welcome!
David
net/bluetooth/hidp/core.c | 16 ++++++++++++++-- net/bluetooth/hidp/hidp.h | 4 ++++ 2 files changed, 18 insertions(+), 2 deletions(-)
net/bluetooth/hidp/core.c +14 -2
@@ -430,6 +430,16 @@ static void hidp_del_timer(struct hidp_session *session)del_timer (&session->timer); } +static void hidp_process_report (struct hidp_session *session, +int type,const u8 *data,int len,int intr) +{ +if (len >HID_MAX_BUFFER_SIZE ) + len =HID_MAX_BUFFER_SIZE ; + +memcpy (session->input_buf, data, len); +hid_input_report (session->hid, type, session->input_buf, len, intr); +} +static void hidp_process_handshake (struct hidp_session *session,unsigned char param) {
@@ -502,7 +512,8 @@ static int hidp_process_data(struct hidp_session *session, struct sk_buff *skb,hidp_input_report (session, skb);if (session->hid) -hid_input_report (session->hid,HID_INPUT_REPORT , skb->data, skb->len,0 ); +hidp_process_report (session,HID_INPUT_REPORT , + skb->data, skb->len,0 );break ;case HIDP_DATA_RTYPE_OTHER :
@@ -584,7 +595,8 @@ static void hidp_recv_intr_frame(struct hidp_session *session,hidp_input_report (session, skb);if (session->hid) { -hid_input_report (session->hid,HID_INPUT_REPORT , skb->data, skb->len,1 ); +hidp_process_report (session,HID_INPUT_REPORT , + skb->data, skb->len,1 );BT_DBG ("report len %d" , skb->len); } }else {
net/bluetooth/hidp/hidp.h +4 -0
@@ -24,6 +24,7 @@#define __HIDP_H#include <linux/types.h> +#include <linux/hid.h> #include <linux/kref.h> #include <net/bluetooth/bluetooth.h> #include <net/bluetooth/l2cap.h>
@@ -179,6 +180,9 @@ struct hidp_session {/* Used in hidp_output_raw_report() */ int output_report_success;/* boolean */ + +/* temporary input buffer */ +u8 input_buf[HID_MAX_BUFFER_SIZE ]; };/* HIDP init defines */
Signature · 1 line
1.8.5.1
Marcel Holtmann
maintainerHi David,
Quoted 14 lines
HID core expects the input buffers to be at least of size 4096 (
HID_MAX_BUFFER_SIZE). Other sizes will result in buffer-overflows if an input-report is smaller than advertised. We could, like i2c, compute the biggest report-size instead of usingHID_MAX_BUFFER_SIZE, but this will blow up if report-descriptors are changed after ->start() has been called. So lets be safe and just use the biggest buffer we have.Note that this adds an additional copy to the HIDP input path. If there is a way to make sure the skb-buf is big enough, we should use that instead.
The best way would be to make hid-core honor the @size argument, though, that sounds easier than it is. So lets just fix the buffer-overflows for now and afterwards look for a faster way for all transport drivers.
Quoted 6 lines
Signed-off-by David Herrmann <[email protected]>---
HiAny ideas how to improve this patch? I'd like to avoid the extra copy but I have no clue how the skb stuff works exactly.
the buffers are coming straight from L2CAP. They might come in fragments. We have to reassemble them and while there are large packets for sure, we rather not want to allocate 4096 for every reassembled packet to make HID happy.
I am not super familiar with the underlying memory management of SKBs, but I assume they use slices of some sort internally. So uses large SKBs for 20 byte HID report will cause an issue with all other protocols running on top of L2CAP>
Quoted 2 lines
I also haven't figured out a nice way to make HID-core honor the "size" parameter. hid-input depends on getting the whole input-report.
I think this needs clearly fixing.
Quoted 61 lines
Comments welcome!
David
net/bluetooth/hidp/core.c| 16 ++++++++++++++--net/bluetooth/hidp/hidp.h| 4 ++++2 files changed, 18 insertions(+), 2 deletions(-)net/bluetooth/hidp/core.c +14
-2@@ -430,6 +430,16 @@ static void hidp_del_timer(struct hidp_session *session)del_timer (&session->timer); } +static void hidp_process_report (struct hidp_session *session, +int type,const u8 *data,int len,int intr) +{ +if (len >HID_MAX_BUFFER_SIZE ) + len =HID_MAX_BUFFER_SIZE ; + +memcpy (session->input_buf, data, len); +hid_input_report (session->hid, type, session->input_buf, len, intr); +} +static void hidp_process_handshake (struct hidp_session *session,unsigned char param) {@@ -502,7 +512,8 @@ static int hidp_process_data(struct hidp_session *session, struct sk_buff *skb,hidp_input_report (session, skb);if (session->hid) -hid_input_report (session->hid,HID_INPUT_REPORT , skb->data, skb->len,0 ); +hidp_process_report (session,HID_INPUT_REPORT , + skb->data, skb->len,0 );break ;case HIDP_DATA_RTYPE_OTHER :@@ -584,7 +595,8 @@ static void hidp_recv_intr_frame(struct hidp_session *session,hidp_input_report (session, skb);if (session->hid) { -hid_input_report (session->hid,HID_INPUT_REPORT , skb->data, skb->len,1 ); +hidp_process_report (session,HID_INPUT_REPORT , + skb->data, skb->len,1 );BT_DBG ("report len %d" , skb->len); } }else {net/bluetooth/hidp/hidp.h +1
-0@@ -24,6 +24,7 @@#define __HIDP_H#include <linux/types.h> +#include <linux/hid.h> #include <linux/kref.h> #include <net/bluetooth/bluetooth.h> #include <net/bluetooth/l2cap.h>
net/bluetooth/hidp/hidp.h struct hidp_session { · line 186
Quoted 7 lines
net/bluetooth/hidp/hidp.h +3
-0@@ -179,6 +180,9 @@ struct hidp_session {/* Used in hidp_output_raw_report() */ int output_report_success;/* boolean */ + +/* temporary input buffer */ +u8 input_buf[HID_MAX_BUFFER_SIZE ]; };
The report does not need any specific alignment?
Regards
Marcel
Hi Marcel
Quoted 20 lines · Marcel Holtmann
Hi David,
Quoted 18 lines
HID core expects the input buffers to be at least of size 4096 (
HID_MAX_BUFFER_SIZE). Other sizes will result in buffer-overflows if an input-report is smaller than advertised. We could, like i2c, compute the biggest report-size instead of usingHID_MAX_BUFFER_SIZE, but this will blow up if report-descriptors are changed after ->start() has been called. So lets be safe and just use the biggest buffer we have.Note that this adds an additional copy to the HIDP input path. If there is a way to make sure the skb-buf is big enough, we should use that instead.
The best way would be to make hid-core honor the @size argument, though, that sounds easier than it is. So lets just fix the buffer-overflows for now and afterwards look for a faster way for all transport drivers.
Signed-off-by David Herrmann <[email protected]>---
Hi
Quoted 6 lines
Quoted 2 lines
Any ideas how to improve this patch? I'd like to avoid the extra copy but I have no clue how the skb stuff works exactly.
the buffers are coming straight from L2CAP. They might come in fragments. We have to reassemble them and while there are large packets for sure, we rather not want to allocate 4096 for every reassembled packet to make HID happy.
I am not super familiar with the underlying memory management of SKBs, but I assume they use slices of some sort internally. So uses large SKBs for 20 byte HID report will cause an issue with all other protocols running on top of L2CAP
I was more thinking of an skb call to increase the underlying buffer of a single package. We could call this together with skb_linearize() in HIDP to guarantee the skb-buf is big enough. Technically, this will probably result in the same behavior as my own patch so probably not the way to go.
Quoted 4 lines
Quoted 2 lines
I also haven't figured out a nice way to make HID-core honor the "size" parameter. hid-input depends on getting the whole input-report.
I think this needs clearly fixing.
And we have a volunteer, yippie! ;)
I think hid_input_report() in hid-core.c is the only place that relies on this. Maybe it really is easier to fix it. But I am not even slightly familiar with that code. So if no-one of the HID core developers steps up to fix it, we should take the transport-driver fixes instead. As nearly all transport-drivers are affected by this, maybe we should even move this buffer into hid_device and provide hid_input_truncated_report() which does this.
Anyhow, waiting for Jiri's comments on this.
Quoted 62 lines
Quoted 62 lines
Comments welcome!
David
net/bluetooth/hidp/core.c| 16 ++++++++++++++--net/bluetooth/hidp/hidp.h| 4 ++++2 files changed, 18 insertions(+), 2 deletions(-)net/bluetooth/hidp/core.c +14
-2@@ -430,6 +430,16 @@ static void hidp_del_timer(struct hidp_session *session)del_timer (&session->timer); } +static void hidp_process_report (struct hidp_session *session, +int type,const u8 *data,int len,int intr) +{ +if (len >HID_MAX_BUFFER_SIZE ) + len =HID_MAX_BUFFER_SIZE ; + +memcpy (session->input_buf, data, len); +hid_input_report (session->hid, type, session->input_buf, len, intr); +} +static void hidp_process_handshake (struct hidp_session *session,unsigned char param) {@@ -502,7 +512,8 @@ static int hidp_process_data(struct hidp_session *session, struct sk_buff *skb,hidp_input_report (session, skb);if (session->hid) -hid_input_report (session->hid,HID_INPUT_REPORT , skb->data, skb->len,0 ); +hidp_process_report (session,HID_INPUT_REPORT , + skb->data, skb->len,0 );break ;case HIDP_DATA_RTYPE_OTHER :@@ -584,7 +595,8 @@ static void hidp_recv_intr_frame(struct hidp_session *session,hidp_input_report (session, skb);if (session->hid) { -hid_input_report (session->hid,HID_INPUT_REPORT , skb->data, skb->len,1 ); +hidp_process_report (session,HID_INPUT_REPORT , + skb->data, skb->len,1 );BT_DBG ("report len %d" , skb->len); } }else {net/bluetooth/hidp/hidp.h +1
-0@@ -24,6 +24,7 @@#define __HIDP_H#include <linux/types.h> +#include <linux/hid.h> #include <linux/kref.h> #include <net/bluetooth/bluetooth.h> #include <net/bluetooth/l2cap.h> @@ -179,6 +180,9 @@ struct hidp_session {
Quoted 8 lines
net/bluetooth/hidp/hidp.h struct hidp_session { · line 186
Quoted 6 lines
net/bluetooth/hidp/hidp.h +3
-0/* Used in hidp_output_raw_report() */ int output_report_success;/* boolean */ + +/* temporary input buffer */ +u8 input_buf[HID_MAX_BUFFER_SIZE ]; };The report does not need any specific alignment?
This is already aligned to native size, isn't it? Anyhow, HID core does not have any alignment-restrictions I am aware of. Jiri? I thought the extract()/implement() stuff was fixed recently.
Thanks
David
Quoted 16 lines · David Herrmann
Quoted 5 lines
Quoted 3 lines
I also haven't figured out a nice way to make HID-core honor the "size" parameter. hid-input depends on getting the whole input-report.
I think this needs clearly fixing.
And we have a volunteer, yippie! ;)
I think
hid_input_report()inhid-core.cis the only place that relies on this. Maybe it really is easier to fix it. But I am not even slightly familiar with that code. So if no-one of the HID core developers steps up to fix it, we should take the transport-driver fixes instead. As nearly all transport-drivers are affected by this, maybe we should even move this buffer intohid_deviceand providehid_input_truncated_report()which does this.
Quoted 1 line
Anyhow, waiting for Jiri's comments on this.
Hmm, this is really unfortunate situation.
I am now looking into making hid_input_report() honor 'size' properly, but no matter how it'll be done in the end, it'll absolutely not be a simple 'fix'. So definitely can be done for 3.15 or so, but not as a fix for current kernels.
So doing kzalloc(rsize, GFP_ATOMIC) in the HID-core for now, and copying the buffer around, seems like only viable solution for now, with the outlook of removing this ugliness once hid-core honors 'size' properly.
I will keep looking into this and maybe some easy way how to hack this together will materialize, but I don't currently see it.
Hmm ...
Signature · 2 lines
Jiri Kosina SUSE Labs
Hi Jiri
Quoted 22 lines · Jiri Kosina
Quoted 16 lines · David Herrmann
Quoted 5 lines
Quoted 3 lines
I also haven't figured out a nice way to make HID-core honor the "size" parameter. hid-input depends on getting the whole input-report.
I think this needs clearly fixing.
And we have a volunteer, yippie! ;)
I think
hid_input_report()inhid-core.cis the only place that relies on this. Maybe it really is easier to fix it. But I am not even slightly familiar with that code. So if no-one of the HID core developers steps up to fix it, we should take the transport-driver fixes instead. As nearly all transport-drivers are affected by this, maybe we should even move this buffer intohid_deviceand providehid_input_truncated_report()which does this.Quoted 1 line
Anyhow, waiting for Jiri's comments on this.
Hmm, this is really unfortunate situation.
Quoted 8 lines
I am now looking into making
hid_input_report()honor 'size' properly, but no matter how it'll be done in the end, it'll absolutely not be a simple 'fix'. So definitely can be done for 3.15 or so, but not as a fix for current kernels.So doing
kzalloc(rsize, GFP_ATOMIC)in the HID-core for now, and copying the buffer around, seems like only viable solution for now, with the outlook of removing this ugliness once hid-core honors 'size' properly.
Should I resend the patches and move it to hid_input_report() for now? Or are you getting something in yourself?
Given the amount of reports on the list and bugzilla, I think we should get this fix in asap. We can always fix it properly in -next.
Thanks
David
Quoted 8 lines
I will keep looking into this and maybe some easy way how to hack this together will materialize, but I don't currently see it.
Hmm ...
--
Jiri Kosina
SUSE Labs
Quoted 6 lines · David Herrmann
Quoted 3 lines
So doing
kzalloc(rsize, GFP_ATOMIC)in the HID-core for now, and copying the buffer around, seems like only viable solution for now, with the outlook of removing this ugliness once hid-core honors 'size' properly.Should I resend the patches and move it to
hid_input_report()for now? Or are you getting something in yourself?
Due to various reasons I will not have access to any testing HW for the upcoming 2-3 days, so if you can cook something up in that timeframe, it'd be appreciated.
Otherwise I'll be working on it by the end of this week.
Quoted 2 lines
Given the amount of reports on the list and bugzilla, I think we should get this fix in asap. We can always fix it properly in -next.
Agreed.
Thanks,
Signature · 2 lines
Jiri Kosina SUSE Labs
Quoted 4 lines · Jiri Kosina
Quoted 4 lines
Quoted 3 lines
So doing
kzalloc(rsize, GFP_ATOMIC)in the HID-core for now, and copying the buffer around, seems like only viable solution for now, with the outlook of removing this ugliness once hid-core honors 'size' properly.
Quoted 8 lines
Quoted 2 lines
Should I resend the patches and move it to
hid_input_report()for now? Or are you getting something in yourself?Due to various reasons I will not have access to any testing HW for the upcoming 2-3 days, so if you can cook something up in that timeframe, it'd be appreciated.
Otherwise I'll be working on it by the end of this week.
David,
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Quoted 4 lines
Quoted 2 lines
Given the amount of reports on the list and bugzilla, I think we should get this fix in asap. We can always fix it properly in -next.
Agreed.
Signature · 2 lines
Jiri Kosina SUSE Labs
Hi
Quoted 13 lines · Jiri Kosina
Quoted 11 lines · Jiri Kosina
Quoted 6 lines
Quoted 3 lines
So doing
kzalloc(rsize, GFP_ATOMIC)in the HID-core for now, and copying the buffer around, seems like only viable solution for now, with the outlook of removing this ugliness once hid-core honors 'size' properly.Should I resend the patches and move it to
hid_input_report()for now? Or are you getting something in yourself?Due to various reasons I will not have access to any testing HW for the upcoming 2-3 days, so if you can cook something up in that timeframe, it'd be appreciated.
Quoted 6 lines
Quoted 1 line
Otherwise I'll be working on it by the end of this week.
David,
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Thanks
David
Quoted 6 lines · David Herrmann
Quoted 6 lines
Quoted 5 lines
Due to various reasons I will not have access to any testing HW for the upcoming 2-3 days, so if you can cook something up in that timeframe, it'd be appreciated.
Otherwise I'll be working on it by the end of this week.
Quoted 7 lines
Quoted 4 lines
David,
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Okay, thanks. I then guess we should proceed with this bandaid (double allocation) for 3.14
Signature · 2 lines
Jiri Kosina SUSE Labs
Hi Jiri
Quoted 10 lines · Jiri Kosina
Quoted 8 lines · David Herrmann
Quoted 8 lines
Quoted 5 lines
Due to various reasons I will not have access to any testing HW for the upcoming 2-3 days, so if you can cook something up in that timeframe, it'd be appreciated.
Otherwise I'll be working on it by the end of this week.
David,
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Okay, thanks. I then guess we should proceed with this bandaid (double allocation) for 3.14
It would be really nice if we could get the HIDP patch into 3.14-rc2 and backported to stable. There have been quite a bunch of reports now and I dislike adding a GFP_ATOMIC allocation in HID core. I am back home now and will try to make HID core honor the length, but that's quite invasive and shouldn't be used for stable.
Sorry for the delay,
David
Quoted 14 lines · David Herrmann
Quoted 14 lines
Quoted 13 lines
Quoted 10 lines
Quoted 5 lines
Due to various reasons I will not have access to any testing HW for the upcoming 2-3 days, so if you can cook something up in that timeframe, it'd be appreciated.
Otherwise I'll be working on it by the end of this week.
David,
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Quoted 6 lines
Quoted 2 lines
Okay, thanks. I then guess we should proceed with this bandaid (double allocation) for 3.14
It would be really nice if we could get the HIDP patch into 3.14-rc2 and backported to stable. There have been quite a bunch of reports now and I dislike adding a
GFP_ATOMICallocation in HID core.
I agree with David; Gustavo, what is your take on this, please?
Thanks,
Signature · 2 lines
Jiri Kosina SUSE Labs
Hi Jiri,
Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
Quoted 13 lines
Quoted 10 lines
Quoted 5 lines
Due to various reasons I will not have access to any testing HW for the upcoming 2-3 days, so if you can cook something up in that timeframe, it'd be appreciated.
Otherwise I'll be working on it by the end of this week.
David,
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
Okay, thanks. I then guess we should proceed with this bandaid (double allocation) for 3.14
It would be really nice if we could get the HIDP patch into 3.14-rc2 and backported to stable. There have been quite a bunch of reports now and I dislike adding a
GFP_ATOMICallocation in HID core.I agree with David; Gustavo, what is your take on this, please?
I leave this up to Gustavo to get this into wireless tree for 3.14-rc2.
Regards
Marcel
Quoted 13 lines · Marcel Holtmann
Quoted 13 lines
Quoted 12 lines
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Okay, thanks. I then guess we should proceed with this bandaid (double allocation) for 3.14
It would be really nice if we could get the HIDP patch into 3.14-rc2 and backported to stable. There have been quite a bunch of reports now and I dislike adding a
GFP_ATOMICallocation in HID core.
Quoted 5 lines
Quoted 1 line
I agree with David; Gustavo, what is your take on this, please?
I leave this up to Gustavo to get this into wireless tree for 3.14-rc2.
Acked-by Marcel Holtmann <[email protected]>
Thanks a lot.
Gustavo, what is your take on this please? I can take it through hid.git in case you don't have anything else queued for -rc2.
Signature · 2 lines
Jiri Kosina SUSE Labs
Quoted 15 lines · Jiri Kosina
Quoted 15 lines
Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 2 lines
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Quoted 6 lines
Quoted 2 lines
Okay, thanks. I then guess we should proceed with this bandaid (double allocation) for 3.14
It would be really nice if we could get the HIDP patch into 3.14-rc2 and backported to stable. There have been quite a bunch of reports now and I dislike adding a
GFP_ATOMICallocation in HID core.I agree with David; Gustavo, what is your take on this, please?
Quoted 8 lines
Quoted 3 lines
I leave this up to Gustavo to get this into wireless tree for 3.14-rc2.
Acked-by Marcel Holtmann <[email protected]>Thanks a lot.
Gustavo, what is your take on this please? I can take it through hid.git in case you don't have anything else queued for -rc2.
... ping?
In case this doesn't get reacted upon by the end of the week, I am inclined to take it through hid.git.
Thanks,
Signature · 2 lines
Jiri Kosina SUSE Labs
Marcel Holtmann
maintainerHi Jiri,
Quoted 21 lines
Quoted 21 lines
Quoted 13 lines
Quoted 13 lines
Quoted 12 lines
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
just got back to this, finally ... did you have time to work on this at all, or should I just start from scratch?
Sorry, no. Fosdem is this weekend and I needed to get my code ready for that. But I'll finally have time again next week.
Okay, thanks. I then guess we should proceed with this bandaid (double allocation) for 3.14
It would be really nice if we could get the HIDP patch into 3.14-rc2 and backported to stable. There have been quite a bunch of reports now and I dislike adding a
GFP_ATOMICallocation in HID core.Quoted 5 lines
Quoted 1 line
I agree with David; Gustavo, what is your take on this, please?
I leave this up to Gustavo to get this into wireless tree for 3.14-rc2.
Acked-by Marcel Holtmann <[email protected]>Thanks a lot.
Quoted 7 lines
Quoted 2 lines
Gustavo, what is your take on this please? I can take it through hid.git in case you don't have anything else queued for -rc2.
... ping?
In case this doesn't get reacted upon by the end of the week, I am inclined to take it through hid.git.
take it through hid.git please.
Regards
Marcel
Quoted 9 lines · Marcel Holtmann
Quoted 7 lines
Quoted 2 lines
Gustavo, what is your take on this please? I can take it through hid.git in case you don't have anything else queued for -rc2.
... ping?
In case this doesn't get reacted upon by the end of the week, I am inclined to take it through hid.git.
take it through hid.git please.
Now applied, thanks.
Signature · 2 lines
Jiri Kosina SUSE Labs