https://github.com/torvalds/linux
Revision 646017a612e72f19bd9f991fe25287a149c5f627 authored by Keith Busch on 24 February 2016, 16:15:54 UTC, committed by Jens Axboe on 03 March 2016, 21:42:49 UTC
This patch makes nvme namespace removal lockless. It is up to the caller
to ensure no active namespace scanning is occuring. To ensure no scan
work occurs, the nvme pci driver adds a removing state to the controller
device to avoid queueing scan work during removal. The work is flushed
after setting the state, so no new scan work can be queued.

The lockless removal allows the driver to cleanup a namespace
request_queue if the controller fails during removal. Previously this
could deadlock trying to acquire the namespace mutex in order to handle
such events.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
1 parent 075790e
Raw File
Tip revision: 646017a612e72f19bd9f991fe25287a149c5f627 authored by Keith Busch on 24 February 2016, 16:15:54 UTC
NVMe: Fix namespace removal deadlock
Tip revision: 646017a
cfserl.c
/*
 * Copyright (C) ST-Ericsson AB 2010
 * Author:	Sjur Brendeland
 * License terms: GNU General Public License (GPL) version 2
 */

#define pr_fmt(fmt) KBUILD_MODNAME ":%s(): " fmt, __func__

#include <linux/stddef.h>
#include <linux/spinlock.h>
#include <linux/slab.h>
#include <net/caif/caif_layer.h>
#include <net/caif/cfpkt.h>
#include <net/caif/cfserl.h>

#define container_obj(layr) ((struct cfserl *) layr)

#define CFSERL_STX 0x02
#define SERIAL_MINIUM_PACKET_SIZE 4
#define SERIAL_MAX_FRAMESIZE 4096
struct cfserl {
	struct cflayer layer;
	struct cfpkt *incomplete_frm;
	/* Protects parallel processing of incoming packets */
	spinlock_t sync;
	bool usestx;
};

static int cfserl_receive(struct cflayer *layr, struct cfpkt *pkt);
static int cfserl_transmit(struct cflayer *layr, struct cfpkt *pkt);
static void cfserl_ctrlcmd(struct cflayer *layr, enum caif_ctrlcmd ctrl,
			   int phyid);

struct cflayer *cfserl_create(int instance, bool use_stx)
{
	struct cfserl *this = kzalloc(sizeof(struct cfserl), GFP_ATOMIC);
	if (!this)
		return NULL;
	caif_assert(offsetof(struct cfserl, layer) == 0);
	this->layer.receive = cfserl_receive;
	this->layer.transmit = cfserl_transmit;
	this->layer.ctrlcmd = cfserl_ctrlcmd;
	this->usestx = use_stx;
	spin_lock_init(&this->sync);
	snprintf(this->layer.name, CAIF_LAYER_NAME_SZ, "ser1");
	return &this->layer;
}

static int cfserl_receive(struct cflayer *l, struct cfpkt *newpkt)
{
	struct cfserl *layr = container_obj(l);
	u16 pkt_len;
	struct cfpkt *pkt = NULL;
	struct cfpkt *tail_pkt = NULL;
	u8 tmp8;
	u16 tmp;
	u8 stx = CFSERL_STX;
	int ret;
	u16 expectlen = 0;

	caif_assert(newpkt != NULL);
	spin_lock(&layr->sync);

	if (layr->incomplete_frm != NULL) {
		layr->incomplete_frm =
		    cfpkt_append(layr->incomplete_frm, newpkt, expectlen);
		pkt = layr->incomplete_frm;
		if (pkt == NULL) {
			spin_unlock(&layr->sync);
			return -ENOMEM;
		}
	} else {
		pkt = newpkt;
	}
	layr->incomplete_frm = NULL;

	do {
		/* Search for STX at start of pkt if STX is used */
		if (layr->usestx) {
			cfpkt_extr_head(pkt, &tmp8, 1);
			if (tmp8 != CFSERL_STX) {
				while (cfpkt_more(pkt)
				       && tmp8 != CFSERL_STX) {
					cfpkt_extr_head(pkt, &tmp8, 1);
				}
				if (!cfpkt_more(pkt)) {
					cfpkt_destroy(pkt);
					layr->incomplete_frm = NULL;
					spin_unlock(&layr->sync);
					return -EPROTO;
				}
			}
		}

		pkt_len = cfpkt_getlen(pkt);

		/*
		 *  pkt_len is the accumulated length of the packet data
		 *  we have received so far.
		 *  Exit if frame doesn't hold length.
		 */

		if (pkt_len < 2) {
			if (layr->usestx)
				cfpkt_add_head(pkt, &stx, 1);
			layr->incomplete_frm = pkt;
			spin_unlock(&layr->sync);
			return 0;
		}

		/*
		 *  Find length of frame.
		 *  expectlen is the length we need for a full frame.
		 */
		cfpkt_peek_head(pkt, &tmp, 2);
		expectlen = le16_to_cpu(tmp) + 2;
		/*
		 * Frame error handling
		 */
		if (expectlen < SERIAL_MINIUM_PACKET_SIZE
		    || expectlen > SERIAL_MAX_FRAMESIZE) {
			if (!layr->usestx) {
				if (pkt != NULL)
					cfpkt_destroy(pkt);
				layr->incomplete_frm = NULL;
				expectlen = 0;
				spin_unlock(&layr->sync);
				return -EPROTO;
			}
			continue;
		}

		if (pkt_len < expectlen) {
			/* Too little received data */
			if (layr->usestx)
				cfpkt_add_head(pkt, &stx, 1);
			layr->incomplete_frm = pkt;
			spin_unlock(&layr->sync);
			return 0;
		}

		/*
		 * Enough data for at least one frame.
		 * Split the frame, if too long
		 */
		if (pkt_len > expectlen)
			tail_pkt = cfpkt_split(pkt, expectlen);
		else
			tail_pkt = NULL;

		/* Send the first part of packet upwards.*/
		spin_unlock(&layr->sync);
		ret = layr->layer.up->receive(layr->layer.up, pkt);
		spin_lock(&layr->sync);
		if (ret == -EILSEQ) {
			if (layr->usestx) {
				if (tail_pkt != NULL)
					pkt = cfpkt_append(pkt, tail_pkt, 0);
				/* Start search for next STX if frame failed */
				continue;
			} else {
				cfpkt_destroy(pkt);
				pkt = NULL;
			}
		}

		pkt = tail_pkt;

	} while (pkt != NULL);

	spin_unlock(&layr->sync);
	return 0;
}

static int cfserl_transmit(struct cflayer *layer, struct cfpkt *newpkt)
{
	struct cfserl *layr = container_obj(layer);
	u8 tmp8 = CFSERL_STX;
	if (layr->usestx)
		cfpkt_add_head(newpkt, &tmp8, 1);
	return layer->dn->transmit(layer->dn, newpkt);
}

static void cfserl_ctrlcmd(struct cflayer *layr, enum caif_ctrlcmd ctrl,
			   int phyid)
{
	layr->up->ctrlcmd(layr->up, ctrl, phyid);
}
back to top