aoe: improve handling of misbehaving network paths

index bfd765c..b6d2b16 100644
@@ -91,6 +91,9 @@ enum {
 	RTTDSCALE = 3,
 	RTTAVG_INIT = USEC_PER_SEC / 4 << RTTSCALE,
 	RTTDEV_INIT = RTTAVG_INIT / 4,
+
+	HARD_SCORN_SECS = 10,	/* try another remote port after this */
+	MAX_TAINT = 1000,	/* cap on aoetgt taint */
 };
 
 struct buf {

@@ -103,6 +106,10 @@ struct buf {
 	struct request *rq;
 };
 
+enum frame_flags {
+	FFL_PROBE = 1,
+};
+
 struct frame {
 	struct list_head head;
 	u32 tag;

@@ -118,6 +125,7 @@ struct frame {
 	struct bio_vec *bv;
 	ulong bcnt;
 	ulong bv_off;
+	char flags;
 };
 
 struct aoeif {

@@ -138,8 +146,10 @@ struct aoetgt {
 	ushort next_cwnd;	/* incr maxout after decrementing to zero */
 	ushort ssthresh;	/* slow start threshold */
 	ulong falloc;		/* number of allocated frames */
+	int taint;		/* how much we want to avoid this aoetgt */
 	int minbcnt;
 	int wpkts, rpkts;
+	char nout_probes;
 };
 
 struct aoedev {

@@ -174,7 +184,6 @@ struct aoedev {
 	struct list_head rexmitq; /* deferred retransmissions */
 	struct aoetgt *targets[NTARGETS];
 	struct aoetgt **tgt;	/* target in use when working */
-	struct aoetgt *htgt;	/* target needing rexmit assistance */
 	ulong ntargets;
 	ulong kicked;
 	char ident[512];

602 lines not on this page · raw message

index aaaea66..f0c0c74 100644
@@ -223,7 +223,6 @@ aoedev_downdev(struct aoedev *d)
 
 	/* clean out the in-process request (if any) */
 	aoe_failip(d);
-	d->htgt = NULL;
 
 	/* fast fail all pending I/O */
 	if (d->blkq) {

index b6d2b16..d50e945 100644
@@ -74,8 +74,11 @@ enum {
 	DEVFL_TKILL = (1<<1),	/* flag for timer to know when to kill self */
 	DEVFL_EXT = (1<<2),	/* device accepts lba48 commands */
 	DEVFL_GDALLOC = (1<<3),	/* need to alloc gendisk */
-	DEVFL_KICKME = (1<<4),	/* slow polling network card catch */
-	DEVFL_NEWSIZE = (1<<5),	/* need to update dev size in block layer */
+	DEVFL_GD_NOW = (1<<4),	/* allocating gendisk */
+	DEVFL_KICKME = (1<<5),	/* slow polling network card catch */
+	DEVFL_NEWSIZE = (1<<6),	/* need to update dev size in block layer */
+	DEVFL_FREEING = (1<<7),	/* set when device is being cleaned up */
+	DEVFL_FREED = (1<<8),	/* device has been cleaned up */
 };
 
 enum {

index 57ac72c..6b5b787 100644
@@ -147,9 +147,18 @@ aoeblk_open(struct block_device *bdev, fmode_t mode)
 	struct aoedev *d = bdev->bd_disk->private_data;
 	ulong flags;
 
+	if (!virt_addr_valid(d)) {
+		pr_crit("aoe: invalid device pointer in %s\n",
+			__func__);
+		WARN_ON(1);
+		return -ENODEV;
+	}
+	if (!(d->flags & DEVFL_UP) || d->flags & DEVFL_TKILL)
+		return -ENODEV;
+
 	mutex_lock(&aoeblk_mutex);
 	spin_lock_irqsave(&d->lock, flags);
-	if (d->flags & DEVFL_UP) {
+	if (d->flags & DEVFL_UP && !(d->flags & DEVFL_TKILL)) {
 		d->nopen++;
 		spin_unlock_irqrestore(&d->lock, flags);
 		mutex_unlock(&aoeblk_mutex);

@@ -259,6 +268,18 @@ aoeblk_gdalloc(void *vp)
 	struct request_queue *q;
 	enum { KB = 1024, MB = KB * KB, READ_AHEAD = 2 * MB, };
 	ulong flags;
+	int late = 0;
+
+	spin_lock_irqsave(&d->lock, flags);
+	if (d->flags & DEVFL_GDALLOC
+	&& !(d->flags & DEVFL_TKILL)
+	&& !(d->flags & DEVFL_GD_NOW))
+		d->flags |= DEVFL_GD_NOW;
+	else
+		late = 1;
+	spin_unlock_irqrestore(&d->lock, flags);
+	if (late)
+		return;
 
 	gd = alloc_disk(AOE_PARTITIONS);
 	if (gd == NULL) {

@@ -282,6 +303,11 @@ aoeblk_gdalloc(void *vp)
 	}
 
 	spin_lock_irqsave(&d->lock, flags);
+	WARN_ON(!(d->flags & DEVFL_GD_NOW));
+	WARN_ON(!(d->flags & DEVFL_GDALLOC));
+	WARN_ON(d->flags & DEVFL_TKILL);
+	WARN_ON(d->gd);
+	WARN_ON(d->flags & DEVFL_UP);
 	blk_queue_max_hw_sectors(q, BLK_DEF_MAX_SECTORS);
 	q->backing_dev_info.name = "aoe";
 	q->backing_dev_info.ra_pages = READ_AHEAD / PAGE_CACHE_SIZE;

@@ -306,6 +332,11 @@ aoeblk_gdalloc(void *vp)
 
 	add_disk(gd);
 	aoedisk_add_sysfs(d);
+
+	spin_lock_irqsave(&d->lock, flags);
+	WARN_ON(!(d->flags & DEVFL_GD_NOW));
+	d->flags &= ~DEVFL_GD_NOW;
+	spin_unlock_irqrestore(&d->lock, flags);
 	return;
 
 err_mempool:

@@ -314,7 +345,8 @@ err_disk:
 	put_disk(gd);
 err:
 	spin_lock_irqsave(&d->lock, flags);
-	d->flags &= ~DEVFL_GDALLOC;
+	d->flags &= ~DEVFL_GD_NOW;
+	schedule_work(&d->work);
 	spin_unlock_irqrestore(&d->lock, flags);
 }
 

227 lines not on this page · raw message

index d50e945..0f478dc 100644
@@ -84,7 +84,7 @@ enum {
 enum {
 	DEFAULTBCNT = 2 * 512,	/* 2 sectors */
 	MIN_BUFS = 16,
-	NTARGETS = 8,
+	NTARGETS = 4,
 	NAOEIFS = 8,
 	NSKBPOOLMAX = 256,
 	NFACTIVE = 61,

@@ -185,9 +185,9 @@ struct aoedev {
 	ulong maxbcnt;
 	struct list_head factive[NFACTIVE];	/* hash of active frames */
 	struct list_head rexmitq; /* deferred retransmissions */
-	struct aoetgt *targets[NTARGETS];
+	struct aoetgt **targets;
+	ulong ntargets;		/* number of allocated aoetgt pointers */
 	struct aoetgt **tgt;	/* target in use when working */
-	ulong ntargets;
 	ulong kicked;
 	char ident[512];
 };

index 6b5b787..a129f8c 100644
@@ -67,7 +67,7 @@ static ssize_t aoedisk_show_netif(struct device *dev,
 	nd = nds;
 	ne = nd + ARRAY_SIZE(nds);
 	t = d->targets;
-	te = t + NTARGETS;
+	te = t + d->ntargets;
 	for (; t < te && *t; t++) {
 		ifp = (*t)->ifs;
 		e = ifp + NAOEIFS;

104 lines not on this page · raw message

index 3776715..e66ccbf 100644
@@ -214,7 +214,7 @@ aoedev_downdev(struct aoedev *d)
 
 	/* reset window dressings */
 	tt = d->targets;
-	te = tt + NTARGETS;
+	te = tt + d->ntargets;
 	for (; tt < te && (t = *tt); tt++) {
 		aoecmd_wreset(t);
 		t->nout = 0;

@@ -284,7 +284,7 @@ freedev(struct aoedev *d)
 		blk_cleanup_queue(d->blkq);
 	}
 	t = d->targets;
-	e = t + NTARGETS;
+	e = t + d->ntargets;
 	for (; t < e && *t; t++)
 		freetgt(d, *t);
 	if (d->bufpool)

@@ -376,6 +376,8 @@ restart:
 			dd = &d->next;
 		}
 		spin_unlock(&d->lock);
+		if (doomed)
+			kfree(doomed->targets);
 		kfree(doomed);
 	}
 	spin_unlock_irqrestore(&devlist_lock, flags);

@@ -456,6 +458,12 @@ aoedev_by_aoeaddr(ulong maj, int min, int do_alloc)
 	d = kcalloc(1, sizeof *d, GFP_ATOMIC);
 	if (!d)
 		goto out;
+	d->targets = kcalloc(NTARGETS, sizeof(*d->targets), GFP_ATOMIC);
+	if (!d->targets) {
+		kfree(d);
+		goto out;
+	}
+	d->ntargets = NTARGETS;
 	INIT_WORK(&d->work, aoecmd_sleepwork);
 	spin_lock_init(&d->lock);
 	skb_queue_head_init(&d->skbpool);

index da360f9..abf4ad2 100644
@@ -812,7 +812,9 @@ rexmit_timer(ulong vp)
 		since = tsince_hr(f);
 		n = f->waited_total + since;
 		n /= USEC_PER_SEC;
-		if (n > aoe_deadsecs && !(f->flags & FFL_PROBE)) {
+		if (aoe_deadsecs
+		&& n > aoe_deadsecs
+		&& !(f->flags & FFL_PROBE)) {
 			/* Waited too long.  Device failure.
 			 * Hang all frames on first hash bucket for downdev
 			 * to clean up.

index 0f478dc..f8d6c7e 100644
@@ -11,7 +11,7 @@
 #define AOE_PARTITIONS (16)
 #endif
 
-#define WHITESPACE " \t\v\f\n"
+#define WHITESPACE " \t\v\f\n,"
 
 enum {
 	AOECMD_ATA,

index 2e47404..71d3ea8 100644
@@ -31,7 +31,7 @@ enum {
 
 static char aoe_iflist[IFLISTSZ];
 module_param_string(aoe_iflist, aoe_iflist, IFLISTSZ, 0600);
-MODULE_PARM_DESC(aoe_iflist, "aoe_iflist=\"dev1 [dev2 ...]\"");
+MODULE_PARM_DESC(aoe_iflist, "aoe_iflist=dev1[,dev2...]");
 
 static wait_queue_head_t txwq;
 static struct ktstate kts;

index abf4ad2..25ef5c0 100644
@@ -1217,8 +1217,10 @@ noskb:		if (buf)
 	case ATA_CMD_PIO_READ:
 	case ATA_CMD_PIO_READ_EXT:
 		if (skb->len < n) {
-			pr_err("aoe: runt data size in read.  skb->len=%d need=%ld\n",
-				skb->len, n);
+			pr_err("%s e%ld.%d.  skb->len=%d need=%ld\n",
+				"aoe: runt data size in read from",
+				(long) d->aoemajor, d->aoeminor,
+			       skb->len, n);
 			clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
 			break;
 		}

@@ -1233,7 +1235,9 @@ noskb:		if (buf)
 		break;
 	case ATA_CMD_ID_ATA:
 		if (skb->len < 512) {
-			pr_info("aoe: runt data size in ataid.  skb->len=%d\n",
+			pr_info("%s e%ld.%d.  skb->len=%d need=512\n",
+				"aoe: runt data size in ataid from",
+				(long) d->aoemajor, d->aoeminor,
 				skb->len);
 			break;
 		}

index f8d6c7e..1756494 100644
@@ -1,6 +1,5 @@
-
 /* Copyright (c) 2012 Coraid, Inc.  See COPYING for GPL terms. */
-#define VERSION "64+"
+#define VERSION "81"
 #define AOE_MAJOR 152
 #define DEVICE_NAME "aoe"