Skip to contentSkip to conversation

blktrace: fix pdu_len when tracing packet command requests

PATCH In review Li Zefan 3 participants 7 messages Apr 2, 2009 – Apr 7, 2009

Subsystems (2)
lore
Share & tools
lore

Li Zefan

author
#0 lore raw

PATCH 1/1 blktrace: fix pdu_len when tracing packet command requests

This bug has been here since 2.6.26.

=======

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add large command support"), struct request->cmd has been changed from unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

---
 kernel/trace/blktrace.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
kernel/trace/blktrace.c +1 -1
index fd1ff49..bafb8d3 100644
@@ -642,7 +642,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 	if (blk_pc_request(rq)) {
 		what |= BLK_TC_ACT(BLK_TC_PC);
 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
-				sizeof(rq->cmd), rq->cmd);
+				max_t(int, rq->cmd_len, BLK_MAX_CDB), rq->cmd);
 	} else  {
 		what |= BLK_TC_ACT(BLK_TC_FS);
 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
Signature · 1 line
1.5.4.rc3
#1 lore raw
Quoted 19 lines · On Thu, 02 Apr 2009 10:17:26 +0800 Li Zefan <[email protected]> wrote:

This bug has been here since 2.6.26.

=======

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add large command support"), struct request->cmd has been changed from unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

---
 kernel/trace/blktrace.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
kernel/trace/blktrace.c +0 -0
index fd1ff49..bafb8d3 100644

kernel/trace/blktrace.c static void blk_add_trace_rq(struct request_queue *q, struct request *rq, · line 648

Quoted 8 lines
kernel/trace/blktrace.c +1 -1
@@ -642,7 +642,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 	if (blk_pc_request(rq)) {
 		what |= BLK_TC_ACT(BLK_TC_PC);
 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
-				sizeof(rq->cmd), rq->cmd);
+				max_t(int, rq->cmd_len, BLK_MAX_CDB), rq->cmd);
 	} else  {
 		what |= BLK_TC_ACT(BLK_TC_FS);
 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,

The patch works however it would be cleaner to simply use cmd_len?

block/blk-core.c +1 -0
index 996ed90..54765d7 100644
@@ -132,6 +132,7 @@ void blk_rq_init(struct request_queue *q, struct request *rq)
 	INIT_HLIST_NODE(&rq->hash);
 	RB_CLEAR_NODE(&rq->rb_node);
 	rq->cmd = rq->__cmd;
+	rq->cmd_len = BLK_MAX_CDB;
 	rq->tag = -1;
 	rq->ref_count = 1;
 }
block/blktrace.c +1 -1
index 028120a..90b9ca4 100644
@@ -583,7 +583,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 	if (blk_pc_request(rq)) {
 		what |= BLK_TC_ACT(BLK_TC_PC);
 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
-				sizeof(rq->cmd), rq->cmd);
+				rq->cmd_len, rq->cmd);
 	} else  {
 		what |= BLK_TC_ACT(BLK_TC_FS);
 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
author
#2 lore raw
Quoted 16 lines · FUJITA Tomonori
Quoted 13 lines · On Thu, 02 Apr 2009 10:17:26 +0800 Li Zefan <[email protected]> wrote:

This bug has been here since 2.6.26.

=======

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add large command support"), struct request->cmd has been changed from unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

---
 kernel/trace/blktrace.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
Quoted 17 lines
Quoted 6 lines
kernel/trace/blktrace.c +0 -0
index fd1ff49..bafb8d3 100644

kernel/trace/blktrace.c static void blk_add_trace_rq(struct request_queue *q, struct request *rq, · line 648

Quoted 8 lines
kernel/trace/blktrace.c +1 -1
@@ -642,7 +642,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 	if (blk_pc_request(rq)) {
 		what |= BLK_TC_ACT(BLK_TC_PC);
 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
-				sizeof(rq->cmd), rq->cmd);
+				max_t(int, rq->cmd_len, BLK_MAX_CDB), rq->cmd);
 	} else  {
 		what |= BLK_TC_ACT(BLK_TC_FS);
 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,

The patch works however it would be cleaner to simply use cmd_len?

Right! This looks better, thanks.

Quoted 27 lines
block/blk-core.c +1 -0
index 996ed90..54765d7 100644
@@ -132,6 +132,7 @@ void blk_rq_init(struct request_queue *q, struct request *rq)
 	INIT_HLIST_NODE(&rq->hash);
 	RB_CLEAR_NODE(&rq->rb_node);
 	rq->cmd = rq->__cmd;
+	rq->cmd_len = BLK_MAX_CDB;
 	rq->tag = -1;
 	rq->ref_count = 1;
 }
block/blktrace.c +1 -1
index 028120a..90b9ca4 100644
@@ -583,7 +583,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 	if (blk_pc_request(rq)) {
 		what |= BLK_TC_ACT(BLK_TC_PC);
 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
-				sizeof(rq->cmd), rq->cmd);
+				rq->cmd_len, rq->cmd);
 	} else  {
 		what |= BLK_TC_ACT(BLK_TC_FS);
 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
author
#3 lore raw

PATCH 1/1 blktrace: fix pdu_len when tracing packet command requests

In reply to Li Zefan

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add large command support"), struct request->cmd has been changed from unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

v1 -> v2:
- make sure rq->cmd_len is always intialized, and then we can use rq->cmd_len instead of BLK_MAX_CDB.

---
 block/blk-core.c        |    1 +
 kernel/trace/blktrace.c |    2 +-
 2 files changed, 2 insertions(+), 1 deletions(-)
block/blk-core.c +1 -0
index 29bcfac..859879d 100644
@@ -132,6 +132,7 @@ void blk_rq_init(struct request_queue *q, struct request *rq)
 	INIT_HLIST_NODE(&rq->hash);
 	RB_CLEAR_NODE(&rq->rb_node);
 	rq->cmd = rq->__cmd;
+	rq->cmd_len = BLK_MAX_CDB;
 	rq->tag = -1;
 	rq->ref_count = 1;
 }
kernel/trace/blktrace.c +1 -1
index 947c5b3..cc6a9c2 100644
@@ -642,7 +642,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 	if (blk_pc_request(rq)) {
 		what |= BLK_TC_ACT(BLK_TC_PC);
 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
-				sizeof(rq->cmd), rq->cmd);
+				rq->cmd_len, rq->cmd);
 	} else  {
 		what |= BLK_TC_ACT(BLK_TC_FS);
 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
Signature · 1 line
1.5.4.rc3
#4 lore raw
Quoted 7 lines · Li Zefan

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add large command support"), struct request->cmd has been changed from unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

v1 -> v2:
- make sure rq->cmd_len is always intialized, and then we can use rq->cmd_len instead of BLK_MAX_CDB.

Thanks. I've added a 'v2-by: FUJITA Tomonori' and the Ack from Fujita-san as well to document the precise lineage of the fix.

Note: there's an important robustness and security issue to check before we can apply this fully.

variable-size records are always tricky and need a full audit of the software stack.

rq->cmd_len comes from sg device ioctls, and the sg command header can have an arbitrary value for sg_io_v4::header_len. The only limit in the block layer at the moment is that it must fit into a single kmalloc() - and that - in theory - can be very large.

So:

1) the ftrace ring-buffer code has to be checked (does it work well with larger than 4K records). Steve .. how well will it work?

2) and the user-space blktrace+blkparse code has to be checked for overflows and static sizes as well. Jens, Alan?

I had a quick look at the user-space code. It seems mostly fine. There appears to be one minor bug in blkrawverify.c:

                        pdu_buf = malloc(bit->pdu_len);
                        n = fread(pdu_buf, bit->pdu_len, 1, ifp);
                        if (n == 0) {
                                INC_BAD("bad pdu");

malloc() can return NULL under memory pressure - shouldnt we check it for NULL instead of passing it to fread()?

Oh, there does seem to be a buffer-overflow problem in blkparse_fmt.c:

   static char *dump_pdu(unsigned char *pdu_buf, int pdu_len)
   {
        static char p[4096];
        int i, len;
        if (!pdu_buf || !pdu_len)
                return NULL;

        for (len = 0, i = 0; i < pdu_len; i++) {
                if (i)
                        len += sprintf(p + len, " ");

                len += sprintf(p + len, "%02x", pdu_buf[i]);

[...]

that p[4096] is a buffer-overflow if the pdu_len goes over 4096. This is a small potential security issue if we apply this patch. Should be changed to malloc(pdu_len) instead.

( Relatively small because SG_IO ioctls are not normally allowed
to unprivileged users so generating intentionally large packets to exploit a sysadmin running blkparse seems like a stretch of a threat model. )

Anyway, this needs to be fixed and fully audited, as we can literally get a 128K packet traced here - even though the hardware itself wont be able to do much with it - most packet commands are in the few bytes range up to 16 bytes typically - but the blktrace layer will forward it.

Please double-check that blkparse is not surprised by the (now-again-) variable length packet command output either.

Quoted 1 line

From v2.6.26 on we only emitted the first 4/8 bytes depending on

bitness.

Thanks,

Ingo

author
#6 lore raw
In reply to Li Zefan
Commit-ID:  e2494e1b42ebac402324105d57646489d19e2b01
Gitweb:     http://git.kernel.org/tip/e2494e1b42ebac402324105d57646489d19e2b01
Author:     Li Zefan <[email protected]>
AuthorDate: Thu, 2 Apr 2009 13:43:26 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Fri, 3 Apr 2009 15:29:26 +0200

blktrace: fix pdu_len when tracing packet command requests

Impact: output all of packet commands - not just the first 4 / 8 bytes

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add large command support"), struct request->cmd has been changed from unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

v1 -> v2: by: FUJITA Tomonori <[email protected]>

- make sure rq->cmd_len is always intialized, and then we can use rq->cmd_len instead of BLK_MAX_CDB.

Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Jens Axboe <[email protected]>

LKML-Reference: <[email protected]>

---
 block/blk-core.c        |    1 +
 kernel/trace/blktrace.c |    2 +-
 2 files changed, 2 insertions(+), 1 deletions(-)
block/blk-core.c +1 -0
index 29bcfac..859879d 100644
@@ -132,6 +132,7 @@ void blk_rq_init(struct request_queue *q, struct request *rq)
 	INIT_HLIST_NODE(&rq->hash);
 	RB_CLEAR_NODE(&rq->rb_node);
 	rq->cmd = rq->__cmd;
+	rq->cmd_len = BLK_MAX_CDB;
 	rq->tag = -1;
 	rq->ref_count = 1;
 }
kernel/trace/blktrace.c +1 -1
index 73d7860..b32ff44 100644
@@ -643,7 +643,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 	if (blk_pc_request(rq)) {
 		what |= BLK_TC_ACT(BLK_TC_PC);
 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
-				sizeof(rq->cmd), rq->cmd);
+				rq->cmd_len, rq->cmd);
 	} else  {
 		what |= BLK_TC_ACT(BLK_TC_FS);
 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
author
#5 lore raw
In reply to Ingo Molnar
Quoted 24 lines · Ingo Molnar
Quoted 7 lines · Li Zefan

Since commit d7e3c3249ef23b4617393c69fe464765b4ff1645 ("block: add large command support"), struct request->cmd has been changed from unsinged char cmd[BLK_MAX_CDB] to unsigned char *cmd.

v1 -> v2:
- make sure rq->cmd_len is always intialized, and then we can use rq->cmd_len instead of BLK_MAX_CDB.

Thanks. I've added a 'v2-by: FUJITA Tomonori' and the Ack from Fujita-san as well to document the precise lineage of the fix.

Note: there's an important robustness and security issue to check before we can apply this fully.

variable-size records are always tricky and need a full audit of the software stack.

rq->cmd_len comes from sg device ioctls, and the sg command header can have an arbitrary value for sg_io_v4::header_len. The only limit in the block layer at the moment is that it must fit into a single kmalloc() - and that - in theory - can be very large.

Quoted 5 lines

So:

1) the ftrace ring-buffer code has to be checked (does it work well with larger than 4K records). Steve .. how well will it work?

There is a check:

ring_buffer_lock_reserve(length)

{
	...
	length = rb_calculate_event_length(length);
	if (length > BUF_PAGE_SIZE)
		goto out;
	...
}

so if the record is around PAGE_SIZE, the event will not be recorded.

Quoted 11 lines

2) and the user-space blktrace+blkparse code has to be checked for overflows and static sizes as well. Jens, Alan?

I had a quick look at the user-space code. It seems mostly fine. There appears to be one minor bug in blkrawverify.c:

                        pdu_buf = malloc(bit->pdu_len);
                        n = fread(pdu_buf, bit->pdu_len, 1, ifp);
                        if (n == 0) {
                                INC_BAD("bad pdu");
Quoted 6 lines

malloc() can return NULL under memory pressure - shouldnt we check it for NULL instead of passing it to fread()?

Oh, there does seem to be a buffer-overflow problem in blkparse_fmt.c:

and this can be fixed easily.

Quoted 24 lines
   static char *dump_pdu(unsigned char *pdu_buf, int pdu_len)
   {
        static char p[4096];
        int i, len;
        if (!pdu_buf || !pdu_len)
                return NULL;

        for (len = 0, i = 0; i < pdu_len; i++) {
                if (i)
                        len += sprintf(p + len, " ");

                len += sprintf(p + len, "%02x", pdu_buf[i]);

[...]

that p[4096] is a buffer-overflow if the pdu_len goes over 4096. This is a small potential security issue if we apply this patch. Should be changed to malloc(pdu_len) instead.

( Relatively small because SG_IO ioctls are not normally allowed
to unprivileged users so generating intentionally large packets to exploit a sysadmin running blkparse seems like a stretch of a threat model. )