]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
ALSA: seq: Avoid concurrent access to queue flags
authorTakashi Iwai <tiwai@suse.de>
Fri, 14 Feb 2020 11:13:14 +0000 (12:13 +0100)
committerTakashi Iwai <tiwai@suse.de>
Fri, 14 Feb 2020 14:52:59 +0000 (15:52 +0100)
The queue flags are represented in bit fields and the concurrent
access may result in unexpected results.  Although the current code
should be mostly OK as it's only reading a field while writing other
fields as KCSAN reported, it's safer to cover both with a proper
spinlock protection.

This patch fixes the possible concurrent read by protecting with
q->owner_lock.  Also the queue owner field is protected as well since
it's the field to be protected by the lock itself.

Reported-by: syzbot+65c6c92d04304d0a8efc@syzkaller.appspotmail.com
Reported-by: syzbot+e60ddfa48717579799dd@syzkaller.appspotmail.com
Link: https://lore.kernel.org/r/20200214111316.26939-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
sound/core/seq/seq_queue.c

index caf68bf42f134be3e802c976b6b824f38d64c82d..20c552cf83987c8a0eeaa879e99403b5d28efc74 100644 (file)
@@ -392,6 +392,7 @@ int snd_seq_queue_check_access(int queueid, int client)
 int snd_seq_queue_set_owner(int queueid, int client, int locked)
 {
        struct snd_seq_queue *q = queueptr(queueid);
+       unsigned long flags;
 
        if (q == NULL)
                return -EINVAL;
@@ -401,8 +402,10 @@ int snd_seq_queue_set_owner(int queueid, int client, int locked)
                return -EPERM;
        }
 
+       spin_lock_irqsave(&q->owner_lock, flags);
        q->locked = locked ? 1 : 0;
        q->owner = client;
+       spin_unlock_irqrestore(&q->owner_lock, flags);
        queue_access_unlock(q);
        queuefree(q);
 
@@ -539,15 +542,17 @@ void snd_seq_queue_client_termination(int client)
        unsigned long flags;
        int i;
        struct snd_seq_queue *q;
+       bool matched;
 
        for (i = 0; i < SNDRV_SEQ_MAX_QUEUES; i++) {
                if ((q = queueptr(i)) == NULL)
                        continue;
                spin_lock_irqsave(&q->owner_lock, flags);
-               if (q->owner == client)
+               matched = (q->owner == client);
+               if (matched)
                        q->klocked = 1;
                spin_unlock_irqrestore(&q->owner_lock, flags);
-               if (q->owner == client) {
+               if (matched) {
                        if (q->timer->running)
                                snd_seq_timer_stop(q->timer);
                        snd_seq_timer_reset(q->timer);
@@ -739,6 +744,8 @@ void snd_seq_info_queues_read(struct snd_info_entry *entry,
        int i, bpm;
        struct snd_seq_queue *q;
        struct snd_seq_timer *tmr;
+       bool locked;
+       int owner;
 
        for (i = 0; i < SNDRV_SEQ_MAX_QUEUES; i++) {
                if ((q = queueptr(i)) == NULL)
@@ -750,9 +757,14 @@ void snd_seq_info_queues_read(struct snd_info_entry *entry,
                else
                        bpm = 0;
 
+               spin_lock_irq(&q->owner_lock);
+               locked = q->locked;
+               owner = q->owner;
+               spin_unlock_irq(&q->owner_lock);
+
                snd_iprintf(buffer, "queue %d: [%s]\n", q->queue, q->name);
-               snd_iprintf(buffer, "owned by client    : %d\n", q->owner);
-               snd_iprintf(buffer, "lock status        : %s\n", q->locked ? "Locked" : "Free");
+               snd_iprintf(buffer, "owned by client    : %d\n", owner);
+               snd_iprintf(buffer, "lock status        : %s\n", locked ? "Locked" : "Free");
                snd_iprintf(buffer, "queued time events : %d\n", snd_seq_prioq_avail(q->timeq));
                snd_iprintf(buffer, "queued tick events : %d\n", snd_seq_prioq_avail(q->tickq));
                snd_iprintf(buffer, "timer state        : %s\n", tmr->running ? "Running" : "Stopped");