media: venus: hfi_cmds: Replace fake flex-array with flexible-array member
Subsystems (4)
PATCH 1/1 media: venus: hfi_cmds: Replace fake flex-array with flexible-array member
One-element arrays are deprecated, and we are replacing them with flexible array members instead. So, replace one-element arrays with flexible-array members in struct hfi_sys_set_resource_pkt, and refactor the rest of the code, accordingly.
This helps with the ongoing efforts to tighten the FORTIFY_SOURCE routines on memcpy() and help us make progress towards globally enabling -fstrict-flex-arrays=3 [1].
The only binary differences seen before/after changes are the following:
17ba: mov %rbx,%rdi
17bd: call 17c2 <pkt_sys_set_resource+0x42>
17be: R_X86_64_PLT32 __tsan_write4-0x4
- 17c2: movl $0x14,(%rbx)
+ 17c2: movl $0x10,(%rbx)
17c8: lea 0x4(%rbx),%rdi
17cc: call 17d1 <pkt_sys_set_resource+0x51>
17cd: R_X86_64_PLT32 __tsan_write4-0x4which is expected once this accounts for the following line of code at drivers/media/platform/qcom/venus/hfi_cmds.c:73
73 pkt->hdr.size = sizeof(*pkt);
and as *pkt is of type struct hfi_sys_set_resource_pkt, sizeof(*pkt) is reduced by 4 bytes, due to the flex-array transformation.
--- drivers/media/platform/qcom/venus/hfi_cmds.c | 2 +- drivers/media/platform/qcom/venus/hfi_cmds.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/qcom/venus/hfi_cmds.c +1 -1
@@ -83,7 +83,7 @@ int pkt_sys_set_resource(struct hfi_sys_set_resource_pkt *pkt, u32 id, u32 size, res->size = size; res->mem = addr; pkt->resource_type =HFI_RESOURCE_OCMEM ; - pkt->hdr.size +=sizeof (*res) -sizeof (u32); + pkt->hdr.size +=sizeof (*res);break ; }case VIDC_RESOURCE_NONE :
drivers/media/platform/qcom/venus/hfi_cmds.h +1 -1
@@ -56,7 +56,7 @@ struct hfi_sys_set_resource_pkt {struct hfi_pkt_hdr hdr;u32 resource_handle;u32 resource_type; -u32 resource_data[1 ]; +u32 resource_data[]; };struct hfi_sys_release_resource_pkt {
Signature · 1 line
2.34.1
Quoted 56 lines · Gustavo A. R. Silva
One-element arrays are deprecated, and we are replacing them with flexible array members instead. So, replace one-element arrays with flexible-array members in struct
hfi_sys_set_resource_pkt, and refactor the rest of the code, accordingly.This helps with the ongoing efforts to tighten the
FORTIFY_SOURCEroutines onmemcpy()and help us make progress towards globally enabling -fstrict-flex-arrays=3 [1].The only binary differences seen before/after changes are the following:
17ba: mov %rbx,%rdi 17bd: call 17c2 <pkt_sys_set_resource+0x42> 17be: R_X86_64_PLT32 __tsan_write4-0x4(quoted diff) +1
-1- 17c2: movl $0x14,(%rbx) + 17c2: movl $0x10,(%rbx) 17c8: lea 0x4(%rbx),%rdi 17cc: call 17d1 <pkt_sys_set_resource+0x51> 17cd: R_X86_64_PLT32 __tsan_write4-0x4which is expected once this accounts for the following line of code at
drivers/media/platform/qcom/venus/hfi_cmds.c:7373
pkt->hdr.size=sizeof(*pkt);and as *pkt is of type struct
hfi_sys_set_resource_pkt,sizeof(*pkt)is reduced by 4 bytes, due to the flex-array transformation.Signed-off-by Gustavo A. R. Silva <[email protected]>--- drivers/media/platform/qcom/venus/hfi_cmds.c | 2 +- drivers/media/platform/qcom/venus/hfi_cmds.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)drivers/media/platform/qcom/venus/hfi_cmds.c +1
-1@@ -83,7 +83,7 @@ int pkt_sys_set_resource(struct hfi_sys_set_resource_pkt *pkt, u32 id, u32 size, res->size = size; res->mem = addr; pkt->resource_type =HFI_RESOURCE_OCMEM ; - pkt->hdr.size +=sizeof (*res) -sizeof (u32); + pkt->hdr.size +=sizeof (*res);break ; }case VIDC_RESOURCE_NONE :drivers/media/platform/qcom/venus/hfi_cmds.h +0
-0
drivers/media/platform/qcom/venus/hfi_cmds.h struct hfi_sys_set_resource_pkt { · line 59
Quoted 5 lines
drivers/media/platform/qcom/venus/hfi_cmds.h +1
-1@@ -56,7 +56,7 @@ struct hfi_sys_set_resource_pkt {struct hfi_pkt_hdr hdr;u32 resource_handle;u32 resource_type; -u32 resource_data[1 ]; +u32 resource_data[];
Would making this an u32* be a better resolution?
Konrad
Quoted 3 lines
drivers/media/platform/qcom/venus/hfi_cmds.h +0
-0};struct hfi_sys_release_resource_pkt {
Quoted 57 lines · Konrad Dybcio
Quoted 54 lines · Gustavo A. R. Silva
One-element arrays are deprecated, and we are replacing them with flexible array members instead. So, replace one-element arrays with flexible-array members in struct
hfi_sys_set_resource_pkt, and refactor the rest of the code, accordingly.This helps with the ongoing efforts to tighten the
FORTIFY_SOURCEroutines onmemcpy()and help us make progress towards globally enabling -fstrict-flex-arrays=3 [1].The only binary differences seen before/after changes are the following:
17ba: mov %rbx,%rdi 17bd: call 17c2 <pkt_sys_set_resource+0x42> 17be: R_X86_64_PLT32 __tsan_write4-0x4(quoted diff) +1
-1- 17c2: movl $0x14,(%rbx) + 17c2: movl $0x10,(%rbx) 17c8: lea 0x4(%rbx),%rdi 17cc: call 17d1 <pkt_sys_set_resource+0x51> 17cd: R_X86_64_PLT32 __tsan_write4-0x4which is expected once this accounts for the following line of code at
drivers/media/platform/qcom/venus/hfi_cmds.c:7373
pkt->hdr.size=sizeof(*pkt);and as *pkt is of type struct
hfi_sys_set_resource_pkt,sizeof(*pkt)is reduced by 4 bytes, due to the flex-array transformation.Signed-off-by Gustavo A. R. Silva <[email protected]>--- drivers/media/platform/qcom/venus/hfi_cmds.c | 2 +- drivers/media/platform/qcom/venus/hfi_cmds.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)drivers/media/platform/qcom/venus/hfi_cmds.c +1
-1@@ -83,7 +83,7 @@ int pkt_sys_set_resource(struct hfi_sys_set_resource_pkt *pkt, u32 id, u32 size, res->size = size; res->mem = addr; pkt->resource_type =HFI_RESOURCE_OCMEM ; - pkt->hdr.size +=sizeof (*res) -sizeof (u32); + pkt->hdr.size +=sizeof (*res);break ; }case VIDC_RESOURCE_NONE :drivers/media/platform/qcom/venus/hfi_cmds.h +0
-1--- a/drivers/media/platform/qcom/venus/hfi_cmds.h
Quoted 8 lines
drivers/media/platform/qcom/venus/hfi_cmds.h struct hfi_sys_set_resource_pkt { · line 59
Quoted 7 lines
+++
b/drivers/media/platform/qcom/venus/hfi_cmds.hdrivers/media/platform/qcom/venus/hfi_cmds.h +1
-1@@ -56,7 +56,7 @@ struct hfi_sys_set_resource_pkt {struct hfi_pkt_hdr hdr;u32 resource_handle;u32 resource_type; -u32 resource_data[1 ]; +u32 resource_data[];Would making this an u32* be a better resolution?
Nevermind, I overthought this by thinking in the terms of its size and not the data within the struct...
Maybe struct_size could be used instead of subtracting sizeof(u32) though?
Konrad
Quoted 5 lines
Konrad
Quoted 3 lines
drivers/media/platform/qcom/venus/hfi_cmds.h +0
-0};struct hfi_sys_release_resource_pkt {
Quoted 21 lines · Gustavo A. R. Silva
One-element arrays are deprecated, and we are replacing them with flexible array members instead. So, replace one-element arrays with flexible-array members in struct
hfi_sys_set_resource_pkt, and refactor the rest of the code, accordingly.This helps with the ongoing efforts to tighten the
FORTIFY_SOURCEroutines onmemcpy()and help us make progress towards globally enabling -fstrict-flex-arrays=3 [1].The only binary differences seen before/after changes are the following:
17ba: mov %rbx,%rdi 17bd: call 17c2 <pkt_sys_set_resource+0x42> 17be: R_X86_64_PLT32 __tsan_write4-0x4(quoted diff) +1
-1- 17c2: movl $0x14,(%rbx) + 17c2: movl $0x10,(%rbx) 17c8: lea 0x4(%rbx),%rdi 17cc: call 17d1 <pkt_sys_set_resource+0x51> 17cd: R_X86_64_PLT32 __tsan_write4-0x4
Quoted 7 lines
which is expected once this accounts for the following line of code at
drivers/media/platform/qcom/venus/hfi_cmds.c:7373
pkt->hdr.size=sizeof(*pkt);and as *pkt is of type struct
hfi_sys_set_resource_pkt,sizeof(*pkt)is reduced by 4 bytes, due to the flex-array transformation.
Based on the other place that was subtracting the 1 element, this looks like hfi_cmds.c:73 is an existing sizing bug that is now fixed with this patch, yes?
-Kees
Quoted 39 lines
Signed-off-by Gustavo A. R. Silva <[email protected]>--- drivers/media/platform/qcom/venus/hfi_cmds.c | 2 +- drivers/media/platform/qcom/venus/hfi_cmds.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)drivers/media/platform/qcom/venus/hfi_cmds.c +1
-1@@ -83,7 +83,7 @@ int pkt_sys_set_resource(struct hfi_sys_set_resource_pkt *pkt, u32 id, u32 size, res->size = size; res->mem = addr; pkt->resource_type =HFI_RESOURCE_OCMEM ; - pkt->hdr.size +=sizeof (*res) -sizeof (u32); + pkt->hdr.size +=sizeof (*res);break ; }case VIDC_RESOURCE_NONE :drivers/media/platform/qcom/venus/hfi_cmds.h +1
-1@@ -56,7 +56,7 @@ struct hfi_sys_set_resource_pkt {struct hfi_pkt_hdr hdr;u32 resource_handle;u32 resource_type; -u32 resource_data[1 ]; +u32 resource_data[]; };struct hfi_sys_release_resource_pkt {Signature · 1 line
2.34.1
Signature · 1 line
Kees Cook
Quoted 25 lines · Kees Cook
Quoted 24 lines · Gustavo A. R. Silva
One-element arrays are deprecated, and we are replacing them with flexible array members instead. So, replace one-element arrays with flexible-array members in struct
hfi_sys_set_resource_pkt, and refactor the rest of the code, accordingly.This helps with the ongoing efforts to tighten the
FORTIFY_SOURCEroutines onmemcpy()and help us make progress towards globally enabling -fstrict-flex-arrays=3 [1].The only binary differences seen before/after changes are the following:
17ba: mov %rbx,%rdi 17bd: call 17c2 <pkt_sys_set_resource+0x42> 17be: R_X86_64_PLT32 __tsan_write4-0x4(quoted diff) +1
-1- 17c2: movl $0x14,(%rbx) + 17c2: movl $0x10,(%rbx) 17c8: lea 0x4(%rbx),%rdi 17cc: call 17d1 <pkt_sys_set_resource+0x51> 17cd: R_X86_64_PLT32 __tsan_write4-0x4which is expected once this accounts for the following line of code at
drivers/media/platform/qcom/venus/hfi_cmds.c:73
Quoted 6 lines
Quoted 4 lines
73
pkt->hdr.size=sizeof(*pkt);and as *pkt is of type struct
hfi_sys_set_resource_pkt,sizeof(*pkt)is reduced by 4 bytes, due to the flex-array transformation.Based on the other place that was subtracting the 1 element, this looks
Do you mean the one you commented on yesterday?
https://lore.kernel.org/linux-hardening/ZGPk3PpvYzjD1+0%2F@work/ this?
--
Gustavo
Quoted 49 lines
like hfi_cmds.c:73 is an existing sizing bug that is now fixed with this patch, yes?
Reviewed-by Kees Cook <[email protected]>-Kees
Quoted 39 lines
Signed-off-by Gustavo A. R. Silva <[email protected]>--- drivers/media/platform/qcom/venus/hfi_cmds.c | 2 +- drivers/media/platform/qcom/venus/hfi_cmds.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)drivers/media/platform/qcom/venus/hfi_cmds.c +1
-1@@ -83,7 +83,7 @@ int pkt_sys_set_resource(struct hfi_sys_set_resource_pkt *pkt, u32 id, u32 size, res->size = size; res->mem = addr; pkt->resource_type =HFI_RESOURCE_OCMEM ; - pkt->hdr.size +=sizeof (*res) -sizeof (u32); + pkt->hdr.size +=sizeof (*res);break ; }case VIDC_RESOURCE_NONE :drivers/media/platform/qcom/venus/hfi_cmds.h +1
-1@@ -56,7 +56,7 @@ struct hfi_sys_set_resource_pkt {struct hfi_pkt_hdr hdr;u32 resource_handle;u32 resource_type; -u32 resource_data[1 ]; +u32 resource_data[]; };struct hfi_sys_release_resource_pkt {Signature · 1 line
2.34.1Signature · 1 line
Kees Cook
Quoted 25 lines · Kees Cook
Quoted 24 lines · Gustavo A. R. Silva
One-element arrays are deprecated, and we are replacing them with flexible array members instead. So, replace one-element arrays with flexible-array members in struct
hfi_sys_set_resource_pkt, and refactor the rest of the code, accordingly.This helps with the ongoing efforts to tighten the
FORTIFY_SOURCEroutines onmemcpy()and help us make progress towards globally enabling -fstrict-flex-arrays=3 [1].The only binary differences seen before/after changes are the following:
17ba: mov %rbx,%rdi 17bd: call 17c2 <pkt_sys_set_resource+0x42> 17be: R_X86_64_PLT32 __tsan_write4-0x4(quoted diff) +1
-1- 17c2: movl $0x14,(%rbx) + 17c2: movl $0x10,(%rbx) 17c8: lea 0x4(%rbx),%rdi 17cc: call 17d1 <pkt_sys_set_resource+0x51> 17cd: R_X86_64_PLT32 __tsan_write4-0x4which is expected once this accounts for the following line of code at
drivers/media/platform/qcom/venus/hfi_cmds.c:73
Quoted 8 lines
Quoted 4 lines
73
pkt->hdr.size=sizeof(*pkt);and as *pkt is of type struct
hfi_sys_set_resource_pkt,sizeof(*pkt)is reduced by 4 bytes, due to the flex-array transformation.Based on the other place that was subtracting the 1 element, this looks like hfi_cmds.c:73 is an existing sizing bug that is now fixed with this patch, yes?
It is not a bug. The way it is structured is like [packet header + payload]. The packet header size should be (size of header + payload). Line 73 calculates the size of packet header, while line 86 calculates size of payload. Since payload starts from the last u32, the total size needs to be reduced by size of u32.
Quoted 9 lines
Reviewed-by Kees Cook <[email protected]>-Kees
Quoted 5 lines
The patch looks good. As stated in previous patch, lets combine this into a single series.
Quoted 35 lines
Quoted 34 lines
--- drivers/media/platform/qcom/venus/hfi_cmds.c | 2 +- drivers/media/platform/qcom/venus/hfi_cmds.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)drivers/media/platform/qcom/venus/hfi_cmds.c +1
-1@@ -83,7 +83,7 @@ int pkt_sys_set_resource(struct hfi_sys_set_resource_pkt *pkt, u32 id, u32 size, res->size = size; res->mem = addr; pkt->resource_type =HFI_RESOURCE_OCMEM ; - pkt->hdr.size +=sizeof (*res) -sizeof (u32); + pkt->hdr.size +=sizeof (*res);break ; }case VIDC_RESOURCE_NONE :drivers/media/platform/qcom/venus/hfi_cmds.h +1
-1@@ -56,7 +56,7 @@ struct hfi_sys_set_resource_pkt {struct hfi_pkt_hdr hdr;u32 resource_handle;u32 resource_type; -u32 resource_data[1 ]; +u32 resource_data[]; };struct hfi_sys_release_resource_pkt {Signature · 1 line
2.34.1