[deliverable/linux.git] / drivers / block / aoe / aoedev.c

/* Copyright (c) 2012 Coraid, Inc.  See COPYING for GPL terms. */
/*
 * aoedev.c
 * AoE device utility functions; maintains device list.
 */

#include <linux/hdreg.h>
#include <linux/blkdev.h>
#include <linux/netdevice.h>
#include <linux/delay.h>
#include <linux/slab.h>
#include <linux/bitmap.h>
#include <linux/kdev_t.h>
#include <linux/moduleparam.h>
#include "aoe.h"

static void dummy_timer(ulong);
static void freetgt(struct aoedev *d, struct aoetgt *t);
static void skbpoolfree(struct aoedev *d);

static int aoe_dyndevs = 1;
module_param(aoe_dyndevs, int, 0644);
MODULE_PARM_DESC(aoe_dyndevs, "Use dynamic minor numbers for devices.");

static struct aoedev *devlist;
static DEFINE_SPINLOCK(devlist_lock);

/* Because some systems will have one, many, or no
 *   - partitions,
 *   - slots per shelf,
 *   - or shelves,
 * we need some flexibility in the way the minor numbers
 * are allocated.  So they are dynamic.
 */
#define N_DEVS ((1U<<MINORBITS)/AOE_PARTITIONS)

static DEFINE_SPINLOCK(used_minors_lock);
static DECLARE_BITMAP(used_minors, N_DEVS);

static int
minor_get_dyn(ulong *sysminor)
{
	ulong flags;
	ulong n;
	int error = 0;

	spin_lock_irqsave(&used_minors_lock, flags);
	n = find_first_zero_bit(used_minors, N_DEVS);
	if (n < N_DEVS)
		set_bit(n, used_minors);
	else
		error = -1;
	spin_unlock_irqrestore(&used_minors_lock, flags);

	*sysminor = n * AOE_PARTITIONS;
	return error;
}

static int
minor_get_static(ulong *sysminor, ulong aoemaj, int aoemin)
{
	ulong flags;
	ulong n;
	int error = 0;
	enum {
		/* for backwards compatibility when !aoe_dyndevs,
		 * a static number of supported slots per shelf */
		NPERSHELF = 16,
	};

	if (aoemin >= NPERSHELF) {
		pr_err("aoe: %s %d slots per shelf\n",
			"static minor device numbers support only",
			NPERSHELF);
		error = -1;
		goto out;
	}

	n = aoemaj * NPERSHELF + aoemin;
	if (n >= N_DEVS) {
		pr_err("aoe: %s with e%ld.%d\n",
			"cannot use static minor device numbers",
			aoemaj, aoemin);
		error = -1;
		goto out;
	}

	spin_lock_irqsave(&used_minors_lock, flags);
	if (test_bit(n, used_minors)) {
		pr_err("aoe: %s %lu\n",
			"existing device already has static minor number",
			n);
		error = -1;
	} else
		set_bit(n, used_minors);
	spin_unlock_irqrestore(&used_minors_lock, flags);
	*sysminor = n * AOE_PARTITIONS;
out:
	return error;
}

static int
minor_get(ulong *sysminor, ulong aoemaj, int aoemin)
{
	if (aoe_dyndevs)
		return minor_get_dyn(sysminor);
	else
		return minor_get_static(sysminor, aoemaj, aoemin);
}

static void
minor_free(ulong minor)
{
	ulong flags;

	minor /= AOE_PARTITIONS;
	BUG_ON(minor >= N_DEVS);

	spin_lock_irqsave(&used_minors_lock, flags);
	BUG_ON(!test_bit(minor, used_minors));
	clear_bit(minor, used_minors);
	spin_unlock_irqrestore(&used_minors_lock, flags);
}

/*
 * Users who grab a pointer to the device with aoedev_by_aoeaddr
 * automatically get a reference count and must be responsible
 * for performing a aoedev_put.  With the addition of async
 * kthread processing I'm no longer confident that we can
 * guarantee consistency in the face of device flushes.
 *
 * For the time being, we only bother to add extra references for
 * frames sitting on the iocq.  When the kthreads finish processing
 * these frames, they will aoedev_put the device.
 */

void
aoedev_put(struct aoedev *d)
{
	ulong flags;

	spin_lock_irqsave(&devlist_lock, flags);
	d->ref--;
	spin_unlock_irqrestore(&devlist_lock, flags);
}

static void
dummy_timer(ulong vp)
{
	struct aoedev *d;

	d = (struct aoedev *)vp;
	if (d->flags & DEVFL_TKILL)
		return;
	d->timer.expires = jiffies + HZ;
	add_timer(&d->timer);
}

static void
aoe_failip(struct aoedev *d)
{
	struct request *rq;
	struct bio *bio;
	unsigned long n;

	aoe_failbuf(d, d->ip.buf);

	rq = d->ip.rq;
	if (rq == NULL)
		return;
	while ((bio = d->ip.nxbio)) {
		clear_bit(BIO_UPTODATE, &bio->bi_flags);
		d->ip.nxbio = bio->bi_next;
		n = (unsigned long) rq->special;
		rq->special = (void *) --n;
	}
	if ((unsigned long) rq->special == 0)
		aoe_end_request(d, rq, 0);
}

static void
downdev_frame(struct list_head *pos)
{
	struct frame *f;

	f = list_entry(pos, struct frame, head);
	list_del(pos);
	if (f->buf) {
		f->buf->nframesout--;
		aoe_failbuf(f->t->d, f->buf);
	}
	aoe_freetframe(f);
}

void
aoedev_downdev(struct aoedev *d)
{
	struct aoetgt *t, **tt, **te;
	struct list_head *head, *pos, *nx;
	struct request *rq;
	int i;

	d->flags &= ~DEVFL_UP;

	/* clean out active and to-be-retransmitted buffers */
	for (i = 0; i < NFACTIVE; i++) {
		head = &d->factive[i];
		list_for_each_safe(pos, nx, head)
			downdev_frame(pos);
	}
	head = &d->rexmitq;
	list_for_each_safe(pos, nx, head)
		downdev_frame(pos);

	/* reset window dressings */
	tt = d->targets;
	te = tt + d->ntargets;
	for (; tt < te && (t = *tt); tt++) {
		aoecmd_wreset(t);
		t->nout = 0;
	}

	/* clean out the in-process request (if any) */
	aoe_failip(d);

	/* fast fail all pending I/O */
	if (d->blkq) {
		while ((rq = blk_peek_request(d->blkq))) {
			blk_start_request(rq);
			aoe_end_request(d, rq, 1);
		}
	}

	if (d->gd)
		set_capacity(d->gd, 0);
}

/* return whether the user asked for this particular
 * device to be flushed
 */
static int
user_req(char *s, size_t slen, struct aoedev *d)
{
	char *p;
	size_t lim;

	if (!d->gd)
		return 0;
	p = strrchr(d->gd->disk_name, '/');
	if (!p)
		p = d->gd->disk_name;
	else
		p += 1;
	lim = sizeof(d->gd->disk_name);
	lim -= p - d->gd->disk_name;
	if (slen < lim)
		lim = slen;

	return !strncmp(s, p, lim);
}

static void
freedev(struct aoedev *d)
{
	struct aoetgt **t, **e;
	int freeing = 0;
	unsigned long flags;

	spin_lock_irqsave(&d->lock, flags);
	if (d->flags & DEVFL_TKILL
	&& !(d->flags & DEVFL_FREEING)) {
		d->flags |= DEVFL_FREEING;
		freeing = 1;
	}
	spin_unlock_irqrestore(&d->lock, flags);
	if (!freeing)
		return;

	del_timer_sync(&d->timer);
	if (d->gd) {
		aoedisk_rm_sysfs(d);
		del_gendisk(d->gd);
		put_disk(d->gd);
		blk_cleanup_queue(d->blkq);
	}
	t = d->targets;
	e = t + d->ntargets;
	for (; t < e && *t; t++)
		freetgt(d, *t);
	if (d->bufpool)
		mempool_destroy(d->bufpool);
	skbpoolfree(d);
	minor_free(d->sysminor);

	spin_lock_irqsave(&d->lock, flags);
	d->flags |= DEVFL_FREED;
	spin_unlock_irqrestore(&d->lock, flags);
}

enum flush_parms {
	NOT_EXITING = 0,
	EXITING = 1,
};

static int
flush(const char __user *str, size_t cnt, int exiting)
{
	ulong flags;
	struct aoedev *d, **dd;
	char buf[16];
	int all = 0;
	int specified = 0;	/* flush a specific device */
	unsigned int skipflags;

	skipflags = DEVFL_GDALLOC | DEVFL_NEWSIZE | DEVFL_TKILL;

	if (!exiting && cnt >= 3) {
		if (cnt > sizeof buf)
			cnt = sizeof buf;
		if (copy_from_user(buf, str, cnt))
			return -EFAULT;
		all = !strncmp(buf, "all", 3);
		if (!all)
			specified = 1;
	}

	flush_scheduled_work();
	/* pass one: without sleeping, do aoedev_downdev */
	spin_lock_irqsave(&devlist_lock, flags);
	for (d = devlist; d; d = d->next) {
		spin_lock(&d->lock);
		if (exiting) {
			/* unconditionally take each device down */
		} else if (specified) {
			if (!user_req(buf, cnt, d))
				goto cont;
		} else if ((!all && (d->flags & DEVFL_UP))
		|| d->flags & skipflags
		|| d->nopen
		|| d->ref)
			goto cont;

		aoedev_downdev(d);
		d->flags |= DEVFL_TKILL;
cont:
		spin_unlock(&d->lock);
	}
	spin_unlock_irqrestore(&devlist_lock, flags);

	/* pass two: call freedev, which might sleep,
	 * for aoedevs marked with DEVFL_TKILL
	 */
restart:
	spin_lock_irqsave(&devlist_lock, flags);
	for (d = devlist; d; d = d->next) {
		spin_lock(&d->lock);
		if (d->flags & DEVFL_TKILL
		&& !(d->flags & DEVFL_FREEING)) {
			spin_unlock(&d->lock);
			spin_unlock_irqrestore(&devlist_lock, flags);
			freedev(d);
			goto restart;
		}
		spin_unlock(&d->lock);
	}

	/* pass three: remove aoedevs marked with DEVFL_FREED */
	for (dd = &devlist, d = *dd; d; d = *dd) {
		struct aoedev *doomed = NULL;

		spin_lock(&d->lock);
		if (d->flags & DEVFL_FREED) {
			*dd = d->next;
			doomed = d;
		} else {
			dd = &d->next;
		}
		spin_unlock(&d->lock);
		if (doomed)
			kfree(doomed->targets);
		kfree(doomed);
	}
	spin_unlock_irqrestore(&devlist_lock, flags);

	return 0;
}

int
aoedev_flush(const char __user *str, size_t cnt)
{
	return flush(str, cnt, NOT_EXITING);
}

/* This has been confirmed to occur once with Tms=3*1000 due to the
 * driver changing link and not processing its transmit ring.  The
 * problem is hard enough to solve by returning an error that I'm
 * still punting on "solving" this.
 */
static void
skbfree(struct sk_buff *skb)
{
	enum { Sms = 250, Tms = 30 * 1000};
	int i = Tms / Sms;

	if (skb == NULL)
		return;
	while (atomic_read(&skb_shinfo(skb)->dataref) != 1 && i-- > 0)
		msleep(Sms);
	if (i < 0) {
		printk(KERN_ERR
			"aoe: %s holds ref: %s\n",
			skb->dev ? skb->dev->name : "netif",
			"cannot free skb -- memory leaked.");
		return;
	}
	skb->truesize -= skb->data_len;
	skb_shinfo(skb)->nr_frags = skb->data_len = 0;
	skb_trim(skb, 0);
	dev_kfree_skb(skb);
}

static void
skbpoolfree(struct aoedev *d)
{
	struct sk_buff *skb, *tmp;

	skb_queue_walk_safe(&d->skbpool, skb, tmp)
		skbfree(skb);

	__skb_queue_head_init(&d->skbpool);
}

/* find it or allocate it */
struct aoedev *
aoedev_by_aoeaddr(ulong maj, int min, int do_alloc)
{
	struct aoedev *d;
	int i;
	ulong flags;
	ulong sysminor = 0;

	spin_lock_irqsave(&devlist_lock, flags);

	for (d=devlist; d; d=d->next)
		if (d->aoemajor == maj && d->aoeminor == min) {
			spin_lock(&d->lock);
			if (d->flags & DEVFL_TKILL) {
				spin_unlock(&d->lock);
				d = NULL;
				goto out;
			}
			d->ref++;
			spin_unlock(&d->lock);
			break;
		}
	if (d || !do_alloc || minor_get(&sysminor, maj, min) < 0)
		goto out;
	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);
		d = NULL;
		goto out;
	}
	d->ntargets = NTARGETS;
	INIT_WORK(&d->work, aoecmd_sleepwork);
	spin_lock_init(&d->lock);
	skb_queue_head_init(&d->skbpool);
	init_timer(&d->timer);
	d->timer.data = (ulong) d;
	d->timer.function = dummy_timer;
	d->timer.expires = jiffies + HZ;
	add_timer(&d->timer);
	d->bufpool = NULL;	/* defer to aoeblk_gdalloc */
	d->tgt = d->targets;
	d->ref = 1;
	for (i = 0; i < NFACTIVE; i++)
		INIT_LIST_HEAD(&d->factive[i]);
	INIT_LIST_HEAD(&d->rexmitq);
	d->sysminor = sysminor;
	d->aoemajor = maj;
	d->aoeminor = min;
	d->rttavg = RTTAVG_INIT;
	d->rttdev = RTTDEV_INIT;
	d->next = devlist;
	devlist = d;
 out:
	spin_unlock_irqrestore(&devlist_lock, flags);
	return d;
}

static void
freetgt(struct aoedev *d, struct aoetgt *t)
{
	struct frame *f;
	struct list_head *pos, *nx, *head;
	struct aoeif *ifp;

	for (ifp = t->ifs; ifp < &t->ifs[NAOEIFS]; ++ifp) {
		if (!ifp->nd)
			break;
		dev_put(ifp->nd);
	}

	head = &t->ffree;
	list_for_each_safe(pos, nx, head) {
		list_del(pos);
		f = list_entry(pos, struct frame, head);
		skbfree(f->skb);
		kfree(f);
	}
	kfree(t);
}

void
aoedev_exit(void)
{
	flush_scheduled_work();
	flush(NULL, 0, EXITING);
}

int __init
aoedev_init(void)
{
	return 0;
}
Commit	Line	Data
fea05a26	1	/* Copyright (c) 2012 Coraid, Inc. See COPYING for GPL terms. */
1da177e4 LT	2	/*
	3	* aoedev.c
	4	* AoE device utility functions; maintains device list.
	5	*/
	6
	7	#include <linux/hdreg.h>
	8	#include <linux/blkdev.h>
	9	#include <linux/netdevice.h>
9bb237b6	10	#include <linux/delay.h>
5a0e3ad6	11	#include <linux/slab.h>
0c966214 EC	12	#include <linux/bitmap.h>
0c966214 EC	13	#include <linux/kdev_t.h>
4bcce1a3	14	#include <linux/moduleparam.h>
1da177e4 LT	15	#include "aoe.h"
1da177e4 LT	16
262bf541	17	static void dummy_timer(ulong);
9bb237b6 EC	18	static void freetgt(struct aoedev d, struct aoetgt t);
9bb237b6 EC	19	static void skbpoolfree(struct aoedev *d);
262bf541	20
08b60623	21	static int aoe_dyndevs = 1;
4bcce1a3 EC	22	module_param(aoe_dyndevs, int, 0644);
	23	MODULE_PARM_DESC(aoe_dyndevs, "Use dynamic minor numbers for devices.");
	24
1da177e4	25	static struct aoedev *devlist;
476aed38	26	static DEFINE_SPINLOCK(devlist_lock);
1da177e4	27
0c966214 EC	28	/* Because some systems will have one, many, or no
	29	* - partitions,
	30	* - slots per shelf,
	31	* - or shelves,
	32	* we need some flexibility in the way the minor numbers
	33	* are allocated. So they are dynamic.
69cf2d85	34	*/
0c966214 EC	35	#define N_DEVS ((1U<<MINORBITS)/AOE_PARTITIONS)
	36
	37	static DEFINE_SPINLOCK(used_minors_lock);
	38	static DECLARE_BITMAP(used_minors, N_DEVS);
	39
	40	static int
4bcce1a3	41	minor_get_dyn(ulong *sysminor)
1da177e4	42	{
1da177e4	43	ulong flags;
0c966214 EC	44	ulong n;
	45	int error = 0;
	46
	47	spin_lock_irqsave(&used_minors_lock, flags);
	48	n = find_first_zero_bit(used_minors, N_DEVS);
	49	if (n < N_DEVS)
	50	set_bit(n, used_minors);
	51	else
	52	error = -1;
	53	spin_unlock_irqrestore(&used_minors_lock, flags);
	54
4bcce1a3	55	sysminor = n AOE_PARTITIONS;
0c966214 EC	56	return error;
0c966214 EC	57	}
1da177e4	58
4bcce1a3 EC	59	static int
	60	minor_get_static(ulong *sysminor, ulong aoemaj, int aoemin)
	61	{
	62	ulong flags;
	63	ulong n;
	64	int error = 0;
	65	enum {
	66	/* for backwards compatibility when !aoe_dyndevs,
	67	* a static number of supported slots per shelf */
	68	NPERSHELF = 16,
	69	};
	70
e0b2bbab EC	71	if (aoemin >= NPERSHELF) {
	72	pr_err("aoe: %s %d slots per shelf\n",
	73	"static minor device numbers support only",
	74	NPERSHELF);
	75	error = -1;
	76	goto out;
	77	}
	78
4bcce1a3	79	n = aoemaj * NPERSHELF + aoemin;
e0b2bbab	80	if (n >= N_DEVS) {
4bcce1a3 EC	81	pr_err("aoe: %s with e%ld.%d\n",
	82	"cannot use static minor device numbers",
	83	aoemaj, aoemin);
	84	error = -1;
e0b2bbab	85	goto out;
4bcce1a3 EC	86	}
4bcce1a3 EC	87
e0b2bbab EC	88	spin_lock_irqsave(&used_minors_lock, flags);
	89	if (test_bit(n, used_minors)) {
	90	pr_err("aoe: %s %lu\n",
	91	"existing device already has static minor number",
	92	n);
	93	error = -1;
	94	} else
	95	set_bit(n, used_minors);
	96	spin_unlock_irqrestore(&used_minors_lock, flags);
b91316f2	97	sysminor = n AOE_PARTITIONS;
e0b2bbab	98	out:
4bcce1a3 EC	99	return error;
	100	}
	101
	102	static int
	103	minor_get(ulong *sysminor, ulong aoemaj, int aoemin)
	104	{
	105	if (aoe_dyndevs)
	106	return minor_get_dyn(sysminor);
	107	else
	108	return minor_get_static(sysminor, aoemaj, aoemin);
	109	}
	110
0c966214 EC	111	static void
	112	minor_free(ulong minor)
	113	{
	114	ulong flags;
1da177e4	115
0c966214 EC	116	minor /= AOE_PARTITIONS;
0c966214 EC	117	BUG_ON(minor >= N_DEVS);
1da177e4	118
0c966214 EC	119	spin_lock_irqsave(&used_minors_lock, flags);
	120	BUG_ON(!test_bit(minor, used_minors));
	121	clear_bit(minor, used_minors);
	122	spin_unlock_irqrestore(&used_minors_lock, flags);
1da177e4 LT	123	}
1da177e4 LT	124
0c966214 EC	125	/*
	126	* Users who grab a pointer to the device with aoedev_by_aoeaddr
	127	* automatically get a reference count and must be responsible
	128	* for performing a aoedev_put. With the addition of async
	129	* kthread processing I'm no longer confident that we can
	130	* guarantee consistency in the face of device flushes.
	131	*
	132	* For the time being, we only bother to add extra references for
	133	* frames sitting on the iocq. When the kthreads finish processing
	134	* these frames, they will aoedev_put the device.
	135	*/
	136
69cf2d85 EC	137	void
	138	aoedev_put(struct aoedev *d)
	139	{
	140	ulong flags;
	141
	142	spin_lock_irqsave(&devlist_lock, flags);
	143	d->ref--;
	144	spin_unlock_irqrestore(&devlist_lock, flags);
	145	}
	146
3ae1c24e EC	147	static void
	148	dummy_timer(ulong vp)
	149	{
	150	struct aoedev *d;
	151
	152	d = (struct aoedev *)vp;
	153	if (d->flags & DEVFL_TKILL)
	154	return;
	155	d->timer.expires = jiffies + HZ;
	156	add_timer(&d->timer);
	157	}
	158
69cf2d85 EC	159	static void
69cf2d85 EC	160	aoe_failip(struct aoedev *d)
1da177e4	161	{
69cf2d85	162	struct request *rq;
1da177e4	163	struct bio *bio;
69cf2d85 EC	164	unsigned long n;
	165
	166	aoe_failbuf(d, d->ip.buf);
1da177e4	167
69cf2d85 EC	168	rq = d->ip.rq;
69cf2d85 EC	169	if (rq == NULL)
896831f5	170	return;
69cf2d85 EC	171	while ((bio = d->ip.nxbio)) {
	172	clear_bit(BIO_UPTODATE, &bio->bi_flags);
	173	d->ip.nxbio = bio->bi_next;
	174	n = (unsigned long) rq->special;
	175	rq->special = (void *) --n;
1da177e4	176	}
69cf2d85 EC	177	if ((unsigned long) rq->special == 0)
69cf2d85 EC	178	aoe_end_request(d, rq, 0);
896831f5 EC	179	}
896831f5 EC	180
3fc9b032 EC	181	static void
	182	downdev_frame(struct list_head *pos)
	183	{
	184	struct frame *f;
	185
	186	f = list_entry(pos, struct frame, head);
	187	list_del(pos);
	188	if (f->buf) {
	189	f->buf->nframesout--;
	190	aoe_failbuf(f->t->d, f->buf);
	191	}
	192	aoe_freetframe(f);
	193	}
	194
896831f5 EC	195	void
	196	aoedev_downdev(struct aoedev *d)
	197	{
	198	struct aoetgt t, tt, *te;
896831f5	199	struct list_head head, pos, *nx;
69cf2d85	200	struct request *rq;
896831f5 EC	201	int i;
896831f5 EC	202
69cf2d85 EC	203	d->flags &= ~DEVFL_UP;
69cf2d85 EC	204
3fc9b032	205	/* clean out active and to-be-retransmitted buffers */
64a80f5a EC	206	for (i = 0; i < NFACTIVE; i++) {
64a80f5a EC	207	head = &d->factive[i];
3fc9b032 EC	208	list_for_each_safe(pos, nx, head)
3fc9b032 EC	209	downdev_frame(pos);
64a80f5a	210	}
3fc9b032 EC	211	head = &d->rexmitq;
	212	list_for_each_safe(pos, nx, head)
	213	downdev_frame(pos);
	214
64a80f5a	215	/* reset window dressings */
896831f5	216	tt = d->targets;
71114ec4	217	te = tt + d->ntargets;
896831f5	218	for (; tt < te && (t = *tt); tt++) {
3a0c40d2	219	aoecmd_wreset(t);
896831f5 EC	220	t->nout = 0;
	221	}
	222
69cf2d85 EC	223	/* clean out the in-process request (if any) */
69cf2d85 EC	224	aoe_failip(d);
1da177e4	225
69cf2d85 EC	226	/* fast fail all pending I/O */
	227	if (d->blkq) {
	228	while ((rq = blk_peek_request(d->blkq))) {
	229	blk_start_request(rq);
	230	aoe_end_request(d, rq, 1);
	231	}
1da177e4 LT	232	}
1da177e4 LT	233
1da177e4	234	if (d->gd)
80795aef	235	set_capacity(d->gd, 0);
1da177e4 LT	236	}
1da177e4 LT	237
4ba9aa7f EC	238	/* return whether the user asked for this particular
	239	* device to be flushed
	240	*/
	241	static int
	242	user_req(char s, size_t slen, struct aoedev d)
	243	{
	244	char *p;
	245	size_t lim;
	246
	247	if (!d->gd)
	248	return 0;
	249	p = strrchr(d->gd->disk_name, '/');
	250	if (!p)
	251	p = d->gd->disk_name;
	252	else
	253	p += 1;
	254	lim = sizeof(d->gd->disk_name);
	255	lim -= p - d->gd->disk_name;
	256	if (slen < lim)
	257	lim = slen;
	258
	259	return !strncmp(s, p, lim);
	260	}
	261
e52a2932 EC	262	static void
	263	freedev(struct aoedev *d)
	264	{
	265	struct aoetgt t, e;
	266	int freeing = 0;
	267	unsigned long flags;
	268
	269	spin_lock_irqsave(&d->lock, flags);
	270	if (d->flags & DEVFL_TKILL
	271	&& !(d->flags & DEVFL_FREEING)) {
	272	d->flags \|= DEVFL_FREEING;
	273	freeing = 1;
	274	}
	275	spin_unlock_irqrestore(&d->lock, flags);
	276	if (!freeing)
	277	return;
	278
	279	del_timer_sync(&d->timer);
	280	if (d->gd) {
	281	aoedisk_rm_sysfs(d);
	282	del_gendisk(d->gd);
	283	put_disk(d->gd);
	284	blk_cleanup_queue(d->blkq);
	285	}
	286	t = d->targets;
71114ec4	287	e = t + d->ntargets;
e52a2932 EC	288	for (; t < e && *t; t++)
	289	freetgt(d, *t);
	290	if (d->bufpool)
	291	mempool_destroy(d->bufpool);
	292	skbpoolfree(d);
	293	minor_free(d->sysminor);
	294
	295	spin_lock_irqsave(&d->lock, flags);
	296	d->flags \|= DEVFL_FREED;
	297	spin_unlock_irqrestore(&d->lock, flags);
	298	}
	299
	300	enum flush_parms {
	301	NOT_EXITING = 0,
	302	EXITING = 1,
	303	};
	304
	305	static int
	306	flush(const char __user *str, size_t cnt, int exiting)
262bf541 EC	307	{
	308	ulong flags;
	309	struct aoedev d, *dd;
262bf541 EC	310	char buf[16];
262bf541 EC	311	int all = 0;
4ba9aa7f	312	int specified = 0; /* flush a specific device */
e52a2932 EC	313	unsigned int skipflags;
	314
	315	skipflags = DEVFL_GDALLOC \| DEVFL_NEWSIZE \| DEVFL_TKILL;
262bf541	316
e52a2932	317	if (!exiting && cnt >= 3) {
262bf541 EC	318	if (cnt > sizeof buf)
	319	cnt = sizeof buf;
	320	if (copy_from_user(buf, str, cnt))
	321	return -EFAULT;
	322	all = !strncmp(buf, "all", 3);
4ba9aa7f EC	323	if (!all)
4ba9aa7f EC	324	specified = 1;
262bf541 EC	325	}
262bf541 EC	326
e52a2932 EC	327	flush_scheduled_work();
e52a2932 EC	328	/* pass one: without sleeping, do aoedev_downdev */
262bf541	329	spin_lock_irqsave(&devlist_lock, flags);
e52a2932	330	for (d = devlist; d; d = d->next) {
262bf541	331	spin_lock(&d->lock);
e52a2932 EC	332	if (exiting) {
	333	/* unconditionally take each device down */
	334	} else if (specified) {
4ba9aa7f	335	if (!user_req(buf, cnt, d))
e52a2932	336	goto cont;
4ba9aa7f	337	} else if ((!all && (d->flags & DEVFL_UP))
e52a2932	338	\|\| d->flags & skipflags
69cf2d85	339	\|\| d->nopen
4ba9aa7f	340	\|\| d->ref)
e52a2932	341	goto cont;
4ba9aa7f	342
262bf541 EC	343	aoedev_downdev(d);
262bf541 EC	344	d->flags \|= DEVFL_TKILL;
e52a2932	345	cont:
262bf541	346	spin_unlock(&d->lock);
262bf541 EC	347	}
262bf541 EC	348	spin_unlock_irqrestore(&devlist_lock, flags);
e52a2932 EC	349
	350	/* pass two: call freedev, which might sleep,
	351	* for aoedevs marked with DEVFL_TKILL
	352	*/
	353	restart:
	354	spin_lock_irqsave(&devlist_lock, flags);
	355	for (d = devlist; d; d = d->next) {
	356	spin_lock(&d->lock);
	357	if (d->flags & DEVFL_TKILL
	358	&& !(d->flags & DEVFL_FREEING)) {
	359	spin_unlock(&d->lock);
	360	spin_unlock_irqrestore(&devlist_lock, flags);
	361	freedev(d);
	362	goto restart;
	363	}
	364	spin_unlock(&d->lock);
262bf541	365	}
e52a2932 EC	366
	367	/* pass three: remove aoedevs marked with DEVFL_FREED */
	368	for (dd = &devlist, d = dd; d; d = dd) {
	369	struct aoedev *doomed = NULL;
	370
	371	spin_lock(&d->lock);
	372	if (d->flags & DEVFL_FREED) {
	373	*dd = d->next;
	374	doomed = d;
	375	} else {
	376	dd = &d->next;
	377	}
	378	spin_unlock(&d->lock);
71114ec4 EC	379	if (doomed)
71114ec4 EC	380	kfree(doomed->targets);
e52a2932 EC	381	kfree(doomed);
	382	}
	383	spin_unlock_irqrestore(&devlist_lock, flags);
	384
262bf541 EC	385	return 0;
	386	}
	387
e52a2932 EC	388	int
	389	aoedev_flush(const char __user *str, size_t cnt)
	390	{
	391	return flush(str, cnt, NOT_EXITING);
	392	}
	393
69cf2d85 EC	394	/* This has been confirmed to occur once with Tms=3*1000 due to the
	395	* driver changing link and not processing its transmit ring. The
	396	* problem is hard enough to solve by returning an error that I'm
	397	* still punting on "solving" this.
	398	*/
9bb237b6 EC	399	static void
	400	skbfree(struct sk_buff *skb)
	401	{
69cf2d85	402	enum { Sms = 250, Tms = 30 * 1000};
9bb237b6 EC	403	int i = Tms / Sms;
	404
	405	if (skb == NULL)
	406	return;
	407	while (atomic_read(&skb_shinfo(skb)->dataref) != 1 && i-- > 0)
	408	msleep(Sms);
94873111	409	if (i < 0) {
9bb237b6 EC	410	printk(KERN_ERR
	411	"aoe: %s holds ref: %s\n",
	412	skb->dev ? skb->dev->name : "netif",
	413	"cannot free skb -- memory leaked.");
	414	return;
	415	}
3d5b0605	416	skb->truesize -= skb->data_len;
9bb237b6 EC	417	skb_shinfo(skb)->nr_frags = skb->data_len = 0;
	418	skb_trim(skb, 0);
	419	dev_kfree_skb(skb);
	420	}
	421
	422	static void
	423	skbpoolfree(struct aoedev *d)
	424	{
e9bb8fb0	425	struct sk_buff skb, tmp;
9bb237b6	426
e9bb8fb0	427	skb_queue_walk_safe(&d->skbpool, skb, tmp)
9bb237b6	428	skbfree(skb);
e9bb8fb0 DM	429
e9bb8fb0 DM	430	__skb_queue_head_init(&d->skbpool);
9bb237b6 EC	431	}
9bb237b6 EC	432
0c966214	433	/* find it or allocate it */
1da177e4	434	struct aoedev *
0c966214	435	aoedev_by_aoeaddr(ulong maj, int min, int do_alloc)
1da177e4 LT	436	{
1da177e4 LT	437	struct aoedev *d;
64a80f5a	438	int i;
1da177e4	439	ulong flags;
10935d05	440	ulong sysminor = 0;
1da177e4 LT	441
	442	spin_lock_irqsave(&devlist_lock, flags);
	443
	444	for (d=devlist; d; d=d->next)
0c966214	445	if (d->aoemajor == maj && d->aoeminor == min) {
e52a2932 EC	446	spin_lock(&d->lock);
	447	if (d->flags & DEVFL_TKILL) {
	448	spin_unlock(&d->lock);
	449	d = NULL;
	450	goto out;
	451	}
69cf2d85	452	d->ref++;
e52a2932	453	spin_unlock(&d->lock);
1da177e4	454	break;
69cf2d85	455	}
4bcce1a3	456	if (d \|\| !do_alloc \|\| minor_get(&sysminor, maj, min) < 0)
68e0d42f EC	457	goto out;
	458	d = kcalloc(1, sizeof *d, GFP_ATOMIC);
	459	if (!d)
	460	goto out;
71114ec4 EC	461	d->targets = kcalloc(NTARGETS, sizeof(*d->targets), GFP_ATOMIC);
	462	if (!d->targets) {
	463	kfree(d);
31279b14	464	d = NULL;
71114ec4 EC	465	goto out;
	466	}
	467	d->ntargets = NTARGETS;
68e0d42f EC	468	INIT_WORK(&d->work, aoecmd_sleepwork);
68e0d42f EC	469	spin_lock_init(&d->lock);
e9bb8fb0	470	skb_queue_head_init(&d->skbpool);
68e0d42f EC	471	init_timer(&d->timer);
	472	d->timer.data = (ulong) d;
	473	d->timer.function = dummy_timer;
	474	d->timer.expires = jiffies + HZ;
	475	add_timer(&d->timer);
	476	d->bufpool = NULL; /* defer to aoeblk_gdalloc */
	477	d->tgt = d->targets;
69cf2d85	478	d->ref = 1;
64a80f5a EC	479	for (i = 0; i < NFACTIVE; i++)
64a80f5a EC	480	INIT_LIST_HEAD(&d->factive[i]);
3a0c40d2	481	INIT_LIST_HEAD(&d->rexmitq);
68e0d42f	482	d->sysminor = sysminor;
0c966214 EC	483	d->aoemajor = maj;
0c966214 EC	484	d->aoeminor = min;
3a0c40d2 EC	485	d->rttavg = RTTAVG_INIT;
3a0c40d2 EC	486	d->rttdev = RTTDEV_INIT;
68e0d42f EC	487	d->next = devlist;
	488	devlist = d;
	489	out:
3ae1c24e	490	spin_unlock_irqrestore(&devlist_lock, flags);
1da177e4 LT	491	return d;
	492	}
	493
	494	static void
9bb237b6	495	freetgt(struct aoedev d, struct aoetgt t)
1da177e4	496	{
896831f5 EC	497	struct frame *f;
896831f5 EC	498	struct list_head pos, nx, *head;
1b86fda9 EC	499	struct aoeif *ifp;
	500
	501	for (ifp = t->ifs; ifp < &t->ifs[NAOEIFS]; ++ifp) {
	502	if (!ifp->nd)
	503	break;
	504	dev_put(ifp->nd);
	505	}
e407a7f6	506
896831f5 EC	507	head = &t->ffree;
	508	list_for_each_safe(pos, nx, head) {
	509	list_del(pos);
	510	f = list_entry(pos, struct frame, head);
9bb237b6	511	skbfree(f->skb);
896831f5 EC	512	kfree(f);
896831f5 EC	513	}
68e0d42f EC	514	kfree(t);
	515	}
	516
1da177e4 LT	517	void
	518	aoedev_exit(void)
	519	{
e52a2932	520	flush_scheduled_work();
e52a2932	521	flush(NULL, 0, EXITING);
1da177e4 LT	522	}
	523
	524	int __init
	525	aoedev_init(void)
	526	{
1da177e4 LT	527	return 0;
1da177e4 LT	528	}