Skip to contentSkip to conversation

ALSA: seq: resize buffer for overflow

PATCH In review Mark Salyzyn 2 participants 3 messages Oct 6, 2017 – Oct 9, 2017 Sound

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 ALSA: seq: resize buffer for overflow

Can not replicate, issue discovered in fuzzing. Stack trace below. No functional or performance testing done regarding the fix.

Trap at (reformatted):

snd_seq_oss_readq_puts(struct seq_oss_readq *q, int dev,
unsigned char *data, int len)

{
    union evrec rec;
    int result;

    memset(&rec, 0, sizeof(rec));
    rec.c[0] = SEQ_MIDIPUTC;
    rec.c[2] = dev;

    while (len-- > 0) {
        rec.c[1] = *data++; // data is RBX  HERE

'data' pointer just passed a page boundary, so the buffer supplied was short. Caller must have been handed an ev->type equal to SNDDRV_SEQ_EVENT_SYSEX, which resulted in handing off ev->data.ext.ptr[ev->data.ext.len] buffer. Intuited that the source of the event and buffer was referenced in snd_midi_event_encode_byte() passing a larger length than the allocated buffer.

BUG: unable to handle kernel paging request at ffffc900008ab000
IP: [<ffffffff82e2fc05>] snd_seq_oss_readq_puts+0xd5/0x170 sound/core/seq/oss/seq_oss_readq.c:112
PGD 1da091067 PUD 1da092067
Oops: 0000 [#1] PREEMPT SMP KASAN
Dumping ftrace buffer:
   (ftrace buffer empty)
Modules linked in:
CPU: 1 PID: 3264 Comm: XXXXXXXX
Hardware name: XXXXXXXXXX
task: ffff8801cdd9e000 task.stack: ffff8801ce648000
RIP: 0010:[<ffffffff82e2fc05>]  [<ffffffff82e2fc05>] snd_seq_oss_readq_puts+0xd5/0x170 sound/core/seq/oss/seq_oss_readq.c:112
RSP: 0018:ffff8801ce64f1c0  EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffffc900008ab000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffffffff858e5780
RBP: ffff8801ce64f260 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 1ffff10039cc9df2 R12: 000000003fffffa4
R13: dffffc0000000000 R14: ffff8801ce64f238 R15: ffffc900008ab001
FS:  00007fe3d3d9e700(0000) GS:ffff8801db300000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffc900008ab000 CR3: 00000001d19b7000 CR4: 00000000001406e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Stack:
 1ffff10039cc9e3b ffff8801ce64f1f8 ffff8801d0b9aa00 0000000041b58ab3
 ffffffff841daf3c ffffffff82e2fb30 0000000000000286 0000000000000005
 ffffffff838aa5d5 ffffffff861962c0 dffffc0000000000 ffff8801ce64f260
Call Trace:
 [<ffffffff82e2ebbe>] send_midi_event sound/core/seq/oss/seq_oss_midi.c:616 [inline]
 [<ffffffff82e2ebbe>] snd_seq_oss_midi_input+0x8ce/0xa70 sound/core/seq/oss/seq_oss_midi.c:535
 [<ffffffff82e2758d>] snd_seq_oss_event_input+0x15d/0x220 sound/core/seq/oss/seq_oss_event.c:439
 [<ffffffff82e0b650>] snd_seq_deliver_single_event.constprop.11+0x310/0x7c0 sound/core/seq/seq_clientmgr.c:621
 [<ffffffff82e0be16>] deliver_to_subscribers sound/core/seq/seq_clientmgr.c:676 [inline]
 [<ffffffff82e0be16>] snd_seq_deliver_event+0x316/0x740 sound/core/seq/seq_clientmgr.c:807
 [<ffffffff82e0cf9e>] snd_seq_kernel_client_dispatch+0x11e/0x150 sound/core/seq/seq_clientmgr.c:2314
 [<ffffffff82e31775>] dummy_input+0x235/0x320 sound/core/seq/seq_dummy.c:104
 [<ffffffff82e0b650>] snd_seq_deliver_single_event.constprop.11+0x310/0x7c0 sound/core/seq/seq_clientmgr.c:621
 [<ffffffff82e0bc2d>] snd_seq_deliver_event+0x12d/0x740 sound/core/seq/seq_clientmgr.c:818
 [<ffffffff82e0d0ed>] snd_seq_dispatch_event+0x11d/0x520 sound/core/seq/seq_clientmgr.c:892
 [<ffffffff82e1117e>] snd_seq_check_queue.part.3+0x38e/0x510 sound/core/seq/seq_queue.c:285
 [<ffffffff82e11d8d>] snd_seq_check_queue sound/core/seq/seq_queue.c:357 [inline]
 [<ffffffff82e11d8d>] snd_seq_enqueue_event+0x32d/0x3d0 sound/core/seq/seq_queue.c:363
 [<ffffffff82e0c444>] snd_seq_client_enqueue_event+0x204/0x3e0 sound/core/seq/seq_clientmgr.c:951
 [<ffffffff82e0cc55>] kernel_client_enqueue.part.10+0xb5/0xd0 sound/core/seq/seq_clientmgr.c:2251
 [<ffffffff82e0cd3f>] kernel_client_enqueue sound/core/seq/seq_clientmgr.c:2241 [inline]
 [<ffffffff82e0cd3f>] snd_seq_kernel_client_enqueue_blocking+0xcf/0x110 sound/core/seq/seq_clientmgr.c:2279
 [<ffffffff82e27f68>] insert_queue sound/core/seq/oss/seq_oss_rw.c:189 [inline]
 [<ffffffff82e27f68>] snd_seq_oss_write+0x538/0x850 sound/core/seq/oss/seq_oss_rw.c:148
 [<ffffffff82e1fab4>] odev_write+0x64/0x90 sound/core/seq/oss/seq_oss.c:177
 [<ffffffff8156ab43>] __vfs_write+0x103/0x680 fs/read_write.c:510
 [<ffffffff8156ec70>] vfs_write+0x170/0x4e0 fs/read_write.c:560
 [<ffffffff81572669>] SYSC_write fs/read_write.c:607 [inline]
 [<ffffffff81572669>] SyS_write+0xd9/0x1b0 fs/read_write.c:599
 [<ffffffff838aad85>] entry_SYSCALL_64_fastpath+0x23/0xc6
Code: 4d 9a eb 4e e8 2d aa 53 fe 4c 8d 7b 01 48 89 d8 48 89 d9 48 c1 e8 03 83 e1 07 42 0f b6 04 28 38 c8 7f 08 84 c0 0f 85 80 00 00 00 <41> 0f b6 47 ff 41 83 ec 01 48 8b b5 68 ff ff ff 48 8b bd 70 ff
RIP  [<ffffffff82e2fc05>] snd_seq_oss_readq_puts+0xd5/0x170 sound/core/seq/oss/seq_oss_readq.c:112
 RSP <ffff8801ce64f1c0>

CR2: ffffc900008ab000

v2: removed nested locks in snd_midi_event_resize_buffer

---
 sound/core/seq/seq_midi_event.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
sound/core/seq/seq_midi_event.c +3 -5
index 90bbbdbeba03..ed3206ef0cd4 100644
@@ -192,8 +192,7 @@ EXPORT_SYMBOL(snd_midi_event_no_status);
 /*
  * resize buffer
  */
-#if 0
-int snd_midi_event_resize_buffer(struct snd_midi_event *dev, int bufsize)
+static int snd_midi_event_resize_buffer(struct snd_midi_event *dev, int bufsize)
 {
 	unsigned char *new_buf, *old_buf;
 	unsigned long flags;
@@ -203,16 +202,13 @@ int snd_midi_event_resize_buffer(struct snd_midi_event *dev, int bufsize)
 	new_buf = kmalloc(bufsize, GFP_KERNEL);
 	if (new_buf == NULL)
 		return -ENOMEM;
-	spin_lock_irqsave(&dev->lock, flags);
 	old_buf = dev->buf;
 	dev->buf = new_buf;
 	dev->bufsize = bufsize;
 	reset_encode(dev);
-	spin_unlock_irqrestore(&dev->lock, flags);
 	kfree(old_buf);
 	return 0;
 }
-#endif  /*  0  */
 
 /*
  *  read bytes and encode to sequencer event if finished
@@ -297,6 +293,8 @@ int snd_midi_event_encode_byte(struct snd_midi_event *dev, int c,
 	} else 	if (dev->type == ST_SYSEX) {
 		if (c == MIDI_CMD_COMMON_SYSEX_END ||
 		    dev->read >= dev->bufsize) {
+			if (dev->read > dev->bufsize)
+				snd_midi_event_resize_buffer(dev, dev->read);
 			ev->flags &= ~SNDRV_SEQ_EVENT_LENGTH_MASK;
 			ev->flags |= SNDRV_SEQ_EVENT_LENGTH_VARIABLE;
 			ev->type = SNDRV_SEQ_EVENT_SYSEX;
Signature · 1 line
2.14.2.920.gcf0c67979c-goog
#1 lore raw
Quoted 19 lines · Mark Salyzyn

Can not replicate, issue discovered in fuzzing. Stack trace below. No functional or performance testing done regarding the fix.

Trap at (reformatted):

snd_seq_oss_readq_puts(struct seq_oss_readq *q, int dev,
unsigned char *data, int len)

{
    union evrec rec;
    int result;

    memset(&rec, 0, sizeof(rec));
    rec.c[0] = SEQ_MIDIPUTC;
    rec.c[2] = dev;

    while (len-- > 0) {
        rec.c[1] = *data++; // data is RBX  HERE
Quoted 7 lines

'data' pointer just passed a page boundary, so the buffer supplied was short. Caller must have been handed an ev->type equal to SNDDRV_SEQ_EVENT_SYSEX, which resulted in handing off ev->data.ext.ptr[ev->data.ext.len] buffer. Intuited that the source of the event and buffer was referenced in snd_midi_event_encode_byte() passing a larger length than the allocated buffer.

I doubt it came from snd_midi_event_encode_byte(). Judging from the call trace below, the event originated from the OSS sequencer write, i.e. it received an OSS event packet, and it was delivered again to another OSS sequencer port back via dummy client.

If so, it should have received some EV_SYSEX packet, and it was processed via snd_seq_oss_synth_sysex(), and the encoded event was delivered.

Now the question is how it triggers this Oops. I couldn't find any obvious cause, but one thing I noticed is a possible race when writing to OSS sequencer concurrently. Something wrong might happen.

BTW, about your patch is buggy regarding the call kmalloc() with GFP_KERNEL inside spinlock.

thanks,

Takashi

Quoted 109 lines
BUG: unable to handle kernel paging request at ffffc900008ab000
IP: [<ffffffff82e2fc05>] snd_seq_oss_readq_puts+0xd5/0x170 sound/core/seq/oss/seq_oss_readq.c:112
PGD 1da091067 PUD 1da092067
Oops: 0000 [#1] PREEMPT SMP KASAN
Dumping ftrace buffer:
   (ftrace buffer empty)
Modules linked in:
CPU: 1 PID: 3264 Comm: XXXXXXXX
Hardware name: XXXXXXXXXX
task: ffff8801cdd9e000 task.stack: ffff8801ce648000
RIP: 0010:[<ffffffff82e2fc05>]  [<ffffffff82e2fc05>] snd_seq_oss_readq_puts+0xd5/0x170 sound/core/seq/oss/seq_oss_readq.c:112
RSP: 0018:ffff8801ce64f1c0  EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffffc900008ab000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffffffff858e5780
RBP: ffff8801ce64f260 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 1ffff10039cc9df2 R12: 000000003fffffa4
R13: dffffc0000000000 R14: ffff8801ce64f238 R15: ffffc900008ab001
FS:  00007fe3d3d9e700(0000) GS:ffff8801db300000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffc900008ab000 CR3: 00000001d19b7000 CR4: 00000000001406e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Stack:
 1ffff10039cc9e3b ffff8801ce64f1f8 ffff8801d0b9aa00 0000000041b58ab3
 ffffffff841daf3c ffffffff82e2fb30 0000000000000286 0000000000000005
 ffffffff838aa5d5 ffffffff861962c0 dffffc0000000000 ffff8801ce64f260
Call Trace:
 [<ffffffff82e2ebbe>] send_midi_event sound/core/seq/oss/seq_oss_midi.c:616 [inline]
 [<ffffffff82e2ebbe>] snd_seq_oss_midi_input+0x8ce/0xa70 sound/core/seq/oss/seq_oss_midi.c:535
 [<ffffffff82e2758d>] snd_seq_oss_event_input+0x15d/0x220 sound/core/seq/oss/seq_oss_event.c:439
 [<ffffffff82e0b650>] snd_seq_deliver_single_event.constprop.11+0x310/0x7c0 sound/core/seq/seq_clientmgr.c:621
 [<ffffffff82e0be16>] deliver_to_subscribers sound/core/seq/seq_clientmgr.c:676 [inline]
 [<ffffffff82e0be16>] snd_seq_deliver_event+0x316/0x740 sound/core/seq/seq_clientmgr.c:807
 [<ffffffff82e0cf9e>] snd_seq_kernel_client_dispatch+0x11e/0x150 sound/core/seq/seq_clientmgr.c:2314
 [<ffffffff82e31775>] dummy_input+0x235/0x320 sound/core/seq/seq_dummy.c:104
 [<ffffffff82e0b650>] snd_seq_deliver_single_event.constprop.11+0x310/0x7c0 sound/core/seq/seq_clientmgr.c:621
 [<ffffffff82e0bc2d>] snd_seq_deliver_event+0x12d/0x740 sound/core/seq/seq_clientmgr.c:818
 [<ffffffff82e0d0ed>] snd_seq_dispatch_event+0x11d/0x520 sound/core/seq/seq_clientmgr.c:892
 [<ffffffff82e1117e>] snd_seq_check_queue.part.3+0x38e/0x510 sound/core/seq/seq_queue.c:285
 [<ffffffff82e11d8d>] snd_seq_check_queue sound/core/seq/seq_queue.c:357 [inline]
 [<ffffffff82e11d8d>] snd_seq_enqueue_event+0x32d/0x3d0 sound/core/seq/seq_queue.c:363
 [<ffffffff82e0c444>] snd_seq_client_enqueue_event+0x204/0x3e0 sound/core/seq/seq_clientmgr.c:951
 [<ffffffff82e0cc55>] kernel_client_enqueue.part.10+0xb5/0xd0 sound/core/seq/seq_clientmgr.c:2251
 [<ffffffff82e0cd3f>] kernel_client_enqueue sound/core/seq/seq_clientmgr.c:2241 [inline]
 [<ffffffff82e0cd3f>] snd_seq_kernel_client_enqueue_blocking+0xcf/0x110 sound/core/seq/seq_clientmgr.c:2279
 [<ffffffff82e27f68>] insert_queue sound/core/seq/oss/seq_oss_rw.c:189 [inline]
 [<ffffffff82e27f68>] snd_seq_oss_write+0x538/0x850 sound/core/seq/oss/seq_oss_rw.c:148
 [<ffffffff82e1fab4>] odev_write+0x64/0x90 sound/core/seq/oss/seq_oss.c:177
 [<ffffffff8156ab43>] __vfs_write+0x103/0x680 fs/read_write.c:510
 [<ffffffff8156ec70>] vfs_write+0x170/0x4e0 fs/read_write.c:560
 [<ffffffff81572669>] SYSC_write fs/read_write.c:607 [inline]
 [<ffffffff81572669>] SyS_write+0xd9/0x1b0 fs/read_write.c:599
 [<ffffffff838aad85>] entry_SYSCALL_64_fastpath+0x23/0xc6
Code: 4d 9a eb 4e e8 2d aa 53 fe 4c 8d 7b 01 48 89 d8 48 89 d9 48 c1 e8 03 83 e1 07 42 0f b6 04 28 38 c8 7f 08 84 c0 0f 85 80 00 00 00 <41> 0f b6 47 ff 41 83 ec 01 48 8b b5 68 ff ff ff 48 8b bd 70 ff
RIP  [<ffffffff82e2fc05>] snd_seq_oss_readq_puts+0xd5/0x170 sound/core/seq/oss/seq_oss_readq.c:112
 RSP <ffff8801ce64f1c0>

CR2: ffffc900008ab000

v2: removed nested locks in snd_midi_event_resize_buffer

---
 sound/core/seq/seq_midi_event.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
sound/core/seq/seq_midi_event.c +3 -5
index 90bbbdbeba03..ed3206ef0cd4 100644
@@ -192,8 +192,7 @@ EXPORT_SYMBOL(snd_midi_event_no_status);
 /*
  * resize buffer
  */
-#if 0
-int snd_midi_event_resize_buffer(struct snd_midi_event *dev, int bufsize)
+static int snd_midi_event_resize_buffer(struct snd_midi_event *dev, int bufsize)
 {
 	unsigned char *new_buf, *old_buf;
 	unsigned long flags;
@@ -203,16 +202,13 @@ int snd_midi_event_resize_buffer(struct snd_midi_event *dev, int bufsize)
 	new_buf = kmalloc(bufsize, GFP_KERNEL);
 	if (new_buf == NULL)
 		return -ENOMEM;
-	spin_lock_irqsave(&dev->lock, flags);
 	old_buf = dev->buf;
 	dev->buf = new_buf;
 	dev->bufsize = bufsize;
 	reset_encode(dev);
-	spin_unlock_irqrestore(&dev->lock, flags);
 	kfree(old_buf);
 	return 0;
 }
-#endif  /*  0  */
 
 /*
  *  read bytes and encode to sequencer event if finished
@@ -297,6 +293,8 @@ int snd_midi_event_encode_byte(struct snd_midi_event *dev, int c,
 	} else 	if (dev->type == ST_SYSEX) {
 		if (c == MIDI_CMD_COMMON_SYSEX_END ||
 		    dev->read >= dev->bufsize) {
+			if (dev->read > dev->bufsize)
+				snd_midi_event_resize_buffer(dev, dev->read);
 			ev->flags &= ~SNDRV_SEQ_EVENT_LENGTH_MASK;
 			ev->flags |= SNDRV_SEQ_EVENT_LENGTH_VARIABLE;
 			ev->type = SNDRV_SEQ_EVENT_SYSEX;
Signature · 1 line
2.14.2.920.gcf0c67979c-goog
#2 lore raw
Quoted 5 lines · Takashi Iwai

I doubt it came from snd_midi_event_encode_byte(). Judging from the call trace below, the event originated from the OSS sequencer write, i.e. it received an OSS event packet, and it was delivered again to another OSS sequencer port back via dummy client.

Quoted 7 lines

If so, it should have received some EV_SYSEX packet, and it was processed via snd_seq_oss_synth_sysex(), and the encoded event was delivered.

Now the question is how it triggers this Oops. I couldn't find any obvious cause, but one thing I noticed is a possible race when writing to OSS sequencer concurrently. Something wrong might happen.

Concurrent writing, thanks, I will switch gears and see if that represents the replication path!

Quoted 2 lines

BTW, about your patch is buggy regarding the call kmalloc() with GFP_KERNEL inside spinlock.

<urrrrk> yup, withdraw this patch, and please erase it from my permanent record ;->

Thanks for the review, it was immensely helpful!

-- Mark