[mirror_ubuntu-zesty-kernel.git] / drivers / block / aoe / aoedev.c

/* Copyright (c) 2013 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 <linux/string.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)) {
		bio->bi_error = -EIO;
		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)
{
	const char *p;
	size_t lim;

	if (!d->gd)
		return 0;
	p = kbasename(d->gd->disk_name);
	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_debugfs(d);
		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
ca47bbd9	1	/* Copyright (c) 2013 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>
a88c1f0c	15	#include <linux/string.h>
1da177e4 LT	16	#include "aoe.h"
1da177e4 LT	17
262bf541	18	static void dummy_timer(ulong);
9bb237b6 EC	19	static void freetgt(struct aoedev d, struct aoetgt t);
9bb237b6 EC	20	static void skbpoolfree(struct aoedev *d);
262bf541	21
08b60623	22	static int aoe_dyndevs = 1;
4bcce1a3 EC	23	module_param(aoe_dyndevs, int, 0644);
	24	MODULE_PARM_DESC(aoe_dyndevs, "Use dynamic minor numbers for devices.");
	25
1da177e4	26	static struct aoedev *devlist;
476aed38	27	static DEFINE_SPINLOCK(devlist_lock);
1da177e4	28
0c966214 EC	29	/* Because some systems will have one, many, or no
	30	* - partitions,
	31	* - slots per shelf,
	32	* - or shelves,
	33	* we need some flexibility in the way the minor numbers
	34	* are allocated. So they are dynamic.
69cf2d85	35	*/
0c966214 EC	36	#define N_DEVS ((1U<<MINORBITS)/AOE_PARTITIONS)
	37
	38	static DEFINE_SPINLOCK(used_minors_lock);
	39	static DECLARE_BITMAP(used_minors, N_DEVS);
	40
	41	static int
4bcce1a3	42	minor_get_dyn(ulong *sysminor)
1da177e4	43	{
1da177e4	44	ulong flags;
0c966214 EC	45	ulong n;
	46	int error = 0;
	47
	48	spin_lock_irqsave(&used_minors_lock, flags);
	49	n = find_first_zero_bit(used_minors, N_DEVS);
	50	if (n < N_DEVS)
	51	set_bit(n, used_minors);
	52	else
	53	error = -1;
	54	spin_unlock_irqrestore(&used_minors_lock, flags);
	55
4bcce1a3	56	sysminor = n AOE_PARTITIONS;
0c966214 EC	57	return error;
0c966214 EC	58	}
1da177e4	59
4bcce1a3 EC	60	static int
	61	minor_get_static(ulong *sysminor, ulong aoemaj, int aoemin)
	62	{
	63	ulong flags;
	64	ulong n;
	65	int error = 0;
	66	enum {
	67	/* for backwards compatibility when !aoe_dyndevs,
	68	* a static number of supported slots per shelf */
	69	NPERSHELF = 16,
	70	};
	71
e0b2bbab EC	72	if (aoemin >= NPERSHELF) {
	73	pr_err("aoe: %s %d slots per shelf\n",
	74	"static minor device numbers support only",
	75	NPERSHELF);
	76	error = -1;
	77	goto out;
	78	}
	79
4bcce1a3	80	n = aoemaj * NPERSHELF + aoemin;
e0b2bbab	81	if (n >= N_DEVS) {
4bcce1a3 EC	82	pr_err("aoe: %s with e%ld.%d\n",
	83	"cannot use static minor device numbers",
	84	aoemaj, aoemin);
	85	error = -1;
e0b2bbab	86	goto out;
4bcce1a3 EC	87	}
4bcce1a3 EC	88
e0b2bbab EC	89	spin_lock_irqsave(&used_minors_lock, flags);
	90	if (test_bit(n, used_minors)) {
	91	pr_err("aoe: %s %lu\n",
	92	"existing device already has static minor number",
	93	n);
	94	error = -1;
	95	} else
	96	set_bit(n, used_minors);
	97	spin_unlock_irqrestore(&used_minors_lock, flags);
b91316f2	98	sysminor = n AOE_PARTITIONS;
e0b2bbab	99	out:
4bcce1a3 EC	100	return error;
	101	}
	102
	103	static int
	104	minor_get(ulong *sysminor, ulong aoemaj, int aoemin)
	105	{
	106	if (aoe_dyndevs)
	107	return minor_get_dyn(sysminor);
	108	else
	109	return minor_get_static(sysminor, aoemaj, aoemin);
	110	}
	111
0c966214 EC	112	static void
	113	minor_free(ulong minor)
	114	{
	115	ulong flags;
1da177e4	116
0c966214 EC	117	minor /= AOE_PARTITIONS;
0c966214 EC	118	BUG_ON(minor >= N_DEVS);
1da177e4	119
0c966214 EC	120	spin_lock_irqsave(&used_minors_lock, flags);
	121	BUG_ON(!test_bit(minor, used_minors));
	122	clear_bit(minor, used_minors);
	123	spin_unlock_irqrestore(&used_minors_lock, flags);
1da177e4 LT	124	}
1da177e4 LT	125
0c966214 EC	126	/*
	127	* Users who grab a pointer to the device with aoedev_by_aoeaddr
	128	* automatically get a reference count and must be responsible
	129	* for performing a aoedev_put. With the addition of async
	130	* kthread processing I'm no longer confident that we can
	131	* guarantee consistency in the face of device flushes.
	132	*
	133	* For the time being, we only bother to add extra references for
	134	* frames sitting on the iocq. When the kthreads finish processing
	135	* these frames, they will aoedev_put the device.
	136	*/
	137
69cf2d85 EC	138	void
	139	aoedev_put(struct aoedev *d)
	140	{
	141	ulong flags;
	142
	143	spin_lock_irqsave(&devlist_lock, flags);
	144	d->ref--;
	145	spin_unlock_irqrestore(&devlist_lock, flags);
	146	}
	147
3ae1c24e EC	148	static void
	149	dummy_timer(ulong vp)
	150	{
	151	struct aoedev *d;
	152
	153	d = (struct aoedev *)vp;
	154	if (d->flags & DEVFL_TKILL)
	155	return;
	156	d->timer.expires = jiffies + HZ;
	157	add_timer(&d->timer);
	158	}
	159
69cf2d85 EC	160	static void
69cf2d85 EC	161	aoe_failip(struct aoedev *d)
1da177e4	162	{
69cf2d85	163	struct request *rq;
1da177e4	164	struct bio *bio;
69cf2d85 EC	165	unsigned long n;
	166
	167	aoe_failbuf(d, d->ip.buf);
1da177e4	168
69cf2d85 EC	169	rq = d->ip.rq;
69cf2d85 EC	170	if (rq == NULL)
896831f5	171	return;
69cf2d85	172	while ((bio = d->ip.nxbio)) {
4246a0b6	173	bio->bi_error = -EIO;
69cf2d85 EC	174	d->ip.nxbio = bio->bi_next;
	175	n = (unsigned long) rq->special;
	176	rq->special = (void *) --n;
1da177e4	177	}
69cf2d85 EC	178	if ((unsigned long) rq->special == 0)
69cf2d85 EC	179	aoe_end_request(d, rq, 0);
896831f5 EC	180	}
896831f5 EC	181
3fc9b032 EC	182	static void
	183	downdev_frame(struct list_head *pos)
	184	{
	185	struct frame *f;
	186
	187	f = list_entry(pos, struct frame, head);
	188	list_del(pos);
	189	if (f->buf) {
	190	f->buf->nframesout--;
	191	aoe_failbuf(f->t->d, f->buf);
	192	}
	193	aoe_freetframe(f);
	194	}
	195
896831f5 EC	196	void
	197	aoedev_downdev(struct aoedev *d)
	198	{
	199	struct aoetgt t, tt, *te;
896831f5	200	struct list_head head, pos, *nx;
69cf2d85	201	struct request *rq;
896831f5 EC	202	int i;
896831f5 EC	203
69cf2d85 EC	204	d->flags &= ~DEVFL_UP;
69cf2d85 EC	205
3fc9b032	206	/* clean out active and to-be-retransmitted buffers */
64a80f5a EC	207	for (i = 0; i < NFACTIVE; i++) {
64a80f5a EC	208	head = &d->factive[i];
3fc9b032 EC	209	list_for_each_safe(pos, nx, head)
3fc9b032 EC	210	downdev_frame(pos);
64a80f5a	211	}
3fc9b032 EC	212	head = &d->rexmitq;
	213	list_for_each_safe(pos, nx, head)
	214	downdev_frame(pos);
	215
64a80f5a	216	/* reset window dressings */
896831f5	217	tt = d->targets;
71114ec4	218	te = tt + d->ntargets;
896831f5	219	for (; tt < te && (t = *tt); tt++) {
3a0c40d2	220	aoecmd_wreset(t);
896831f5 EC	221	t->nout = 0;
	222	}
	223
69cf2d85 EC	224	/* clean out the in-process request (if any) */
69cf2d85 EC	225	aoe_failip(d);
1da177e4	226
69cf2d85 EC	227	/* fast fail all pending I/O */
	228	if (d->blkq) {
	229	while ((rq = blk_peek_request(d->blkq))) {
	230	blk_start_request(rq);
	231	aoe_end_request(d, rq, 1);
	232	}
1da177e4 LT	233	}
1da177e4 LT	234
1da177e4	235	if (d->gd)
80795aef	236	set_capacity(d->gd, 0);
1da177e4 LT	237	}
1da177e4 LT	238
4ba9aa7f EC	239	/* return whether the user asked for this particular
	240	* device to be flushed
	241	*/
	242	static int
	243	user_req(char s, size_t slen, struct aoedev d)
	244	{
a88c1f0c	245	const char *p;
4ba9aa7f EC	246	size_t lim;
	247
	248	if (!d->gd)
	249	return 0;
a88c1f0c	250	p = kbasename(d->gd->disk_name);
4ba9aa7f EC	251	lim = sizeof(d->gd->disk_name);
	252	lim -= p - d->gd->disk_name;
	253	if (slen < lim)
	254	lim = slen;
	255
	256	return !strncmp(s, p, lim);
	257	}
	258
e52a2932 EC	259	static void
	260	freedev(struct aoedev *d)
	261	{
	262	struct aoetgt t, e;
	263	int freeing = 0;
	264	unsigned long flags;
	265
	266	spin_lock_irqsave(&d->lock, flags);
	267	if (d->flags & DEVFL_TKILL
	268	&& !(d->flags & DEVFL_FREEING)) {
	269	d->flags \|= DEVFL_FREEING;
	270	freeing = 1;
	271	}
	272	spin_unlock_irqrestore(&d->lock, flags);
	273	if (!freeing)
	274	return;
	275
	276	del_timer_sync(&d->timer);
	277	if (d->gd) {
e8866cf2	278	aoedisk_rm_debugfs(d);
e52a2932 EC	279	aoedisk_rm_sysfs(d);
	280	del_gendisk(d->gd);
	281	put_disk(d->gd);
	282	blk_cleanup_queue(d->blkq);
	283	}
	284	t = d->targets;
71114ec4	285	e = t + d->ntargets;
e52a2932 EC	286	for (; t < e && *t; t++)
	287	freetgt(d, *t);
	288	if (d->bufpool)
	289	mempool_destroy(d->bufpool);
	290	skbpoolfree(d);
	291	minor_free(d->sysminor);
	292
	293	spin_lock_irqsave(&d->lock, flags);
	294	d->flags \|= DEVFL_FREED;
	295	spin_unlock_irqrestore(&d->lock, flags);
	296	}
	297
	298	enum flush_parms {
	299	NOT_EXITING = 0,
	300	EXITING = 1,
	301	};
	302
	303	static int
	304	flush(const char __user *str, size_t cnt, int exiting)
262bf541 EC	305	{
	306	ulong flags;
	307	struct aoedev d, *dd;
262bf541 EC	308	char buf[16];
262bf541 EC	309	int all = 0;
4ba9aa7f	310	int specified = 0; /* flush a specific device */
e52a2932 EC	311	unsigned int skipflags;
	312
	313	skipflags = DEVFL_GDALLOC \| DEVFL_NEWSIZE \| DEVFL_TKILL;
262bf541	314
e52a2932	315	if (!exiting && cnt >= 3) {
262bf541 EC	316	if (cnt > sizeof buf)
	317	cnt = sizeof buf;
	318	if (copy_from_user(buf, str, cnt))
	319	return -EFAULT;
	320	all = !strncmp(buf, "all", 3);
4ba9aa7f EC	321	if (!all)
4ba9aa7f EC	322	specified = 1;
262bf541 EC	323	}
262bf541 EC	324
e52a2932 EC	325	flush_scheduled_work();
e52a2932 EC	326	/* pass one: without sleeping, do aoedev_downdev */
262bf541	327	spin_lock_irqsave(&devlist_lock, flags);
e52a2932	328	for (d = devlist; d; d = d->next) {
262bf541	329	spin_lock(&d->lock);
e52a2932 EC	330	if (exiting) {
	331	/* unconditionally take each device down */
	332	} else if (specified) {
4ba9aa7f	333	if (!user_req(buf, cnt, d))
e52a2932	334	goto cont;
4ba9aa7f	335	} else if ((!all && (d->flags & DEVFL_UP))
e52a2932	336	\|\| d->flags & skipflags
69cf2d85	337	\|\| d->nopen
4ba9aa7f	338	\|\| d->ref)
e52a2932	339	goto cont;
4ba9aa7f	340
262bf541 EC	341	aoedev_downdev(d);
262bf541 EC	342	d->flags \|= DEVFL_TKILL;
e52a2932	343	cont:
262bf541	344	spin_unlock(&d->lock);
262bf541 EC	345	}
262bf541 EC	346	spin_unlock_irqrestore(&devlist_lock, flags);
e52a2932 EC	347
	348	/* pass two: call freedev, which might sleep,
	349	* for aoedevs marked with DEVFL_TKILL
	350	*/
	351	restart:
	352	spin_lock_irqsave(&devlist_lock, flags);
	353	for (d = devlist; d; d = d->next) {
	354	spin_lock(&d->lock);
	355	if (d->flags & DEVFL_TKILL
	356	&& !(d->flags & DEVFL_FREEING)) {
	357	spin_unlock(&d->lock);
	358	spin_unlock_irqrestore(&devlist_lock, flags);
	359	freedev(d);
	360	goto restart;
	361	}
	362	spin_unlock(&d->lock);
262bf541	363	}
e52a2932 EC	364
	365	/* pass three: remove aoedevs marked with DEVFL_FREED */
	366	for (dd = &devlist, d = dd; d; d = dd) {
	367	struct aoedev *doomed = NULL;
	368
	369	spin_lock(&d->lock);
	370	if (d->flags & DEVFL_FREED) {
	371	*dd = d->next;
	372	doomed = d;
	373	} else {
	374	dd = &d->next;
	375	}
	376	spin_unlock(&d->lock);
71114ec4 EC	377	if (doomed)
71114ec4 EC	378	kfree(doomed->targets);
e52a2932 EC	379	kfree(doomed);
	380	}
	381	spin_unlock_irqrestore(&devlist_lock, flags);
	382
262bf541 EC	383	return 0;
	384	}
	385
e52a2932 EC	386	int
	387	aoedev_flush(const char __user *str, size_t cnt)
	388	{
	389	return flush(str, cnt, NOT_EXITING);
	390	}
	391
69cf2d85 EC	392	/* This has been confirmed to occur once with Tms=3*1000 due to the
	393	* driver changing link and not processing its transmit ring. The
	394	* problem is hard enough to solve by returning an error that I'm
	395	* still punting on "solving" this.
	396	*/
9bb237b6 EC	397	static void
	398	skbfree(struct sk_buff *skb)
	399	{
69cf2d85	400	enum { Sms = 250, Tms = 30 * 1000};
9bb237b6 EC	401	int i = Tms / Sms;
	402
	403	if (skb == NULL)
	404	return;
	405	while (atomic_read(&skb_shinfo(skb)->dataref) != 1 && i-- > 0)
	406	msleep(Sms);
94873111	407	if (i < 0) {
9bb237b6 EC	408	printk(KERN_ERR
	409	"aoe: %s holds ref: %s\n",
	410	skb->dev ? skb->dev->name : "netif",
	411	"cannot free skb -- memory leaked.");
	412	return;
	413	}
3d5b0605	414	skb->truesize -= skb->data_len;
9bb237b6 EC	415	skb_shinfo(skb)->nr_frags = skb->data_len = 0;
	416	skb_trim(skb, 0);
	417	dev_kfree_skb(skb);
	418	}
	419
	420	static void
	421	skbpoolfree(struct aoedev *d)
	422	{
e9bb8fb0	423	struct sk_buff skb, tmp;
9bb237b6	424
e9bb8fb0	425	skb_queue_walk_safe(&d->skbpool, skb, tmp)
9bb237b6	426	skbfree(skb);
e9bb8fb0 DM	427
e9bb8fb0 DM	428	__skb_queue_head_init(&d->skbpool);
9bb237b6 EC	429	}
9bb237b6 EC	430
0c966214	431	/* find it or allocate it */
1da177e4	432	struct aoedev *
0c966214	433	aoedev_by_aoeaddr(ulong maj, int min, int do_alloc)
1da177e4 LT	434	{
1da177e4 LT	435	struct aoedev *d;
64a80f5a	436	int i;
1da177e4	437	ulong flags;
10935d05	438	ulong sysminor = 0;
1da177e4 LT	439
	440	spin_lock_irqsave(&devlist_lock, flags);
	441
	442	for (d=devlist; d; d=d->next)
0c966214	443	if (d->aoemajor == maj && d->aoeminor == min) {
e52a2932 EC	444	spin_lock(&d->lock);
	445	if (d->flags & DEVFL_TKILL) {
	446	spin_unlock(&d->lock);
	447	d = NULL;
	448	goto out;
	449	}
69cf2d85	450	d->ref++;
e52a2932	451	spin_unlock(&d->lock);
1da177e4	452	break;
69cf2d85	453	}
4bcce1a3	454	if (d \|\| !do_alloc \|\| minor_get(&sysminor, maj, min) < 0)
68e0d42f EC	455	goto out;
	456	d = kcalloc(1, sizeof *d, GFP_ATOMIC);
	457	if (!d)
	458	goto out;
71114ec4 EC	459	d->targets = kcalloc(NTARGETS, sizeof(*d->targets), GFP_ATOMIC);
	460	if (!d->targets) {
	461	kfree(d);
31279b14	462	d = NULL;
71114ec4 EC	463	goto out;
	464	}
	465	d->ntargets = NTARGETS;
68e0d42f EC	466	INIT_WORK(&d->work, aoecmd_sleepwork);
68e0d42f EC	467	spin_lock_init(&d->lock);
e9bb8fb0	468	skb_queue_head_init(&d->skbpool);
68e0d42f EC	469	init_timer(&d->timer);
	470	d->timer.data = (ulong) d;
	471	d->timer.function = dummy_timer;
	472	d->timer.expires = jiffies + HZ;
	473	add_timer(&d->timer);
	474	d->bufpool = NULL; /* defer to aoeblk_gdalloc */
	475	d->tgt = d->targets;
69cf2d85	476	d->ref = 1;
64a80f5a EC	477	for (i = 0; i < NFACTIVE; i++)
64a80f5a EC	478	INIT_LIST_HEAD(&d->factive[i]);
3a0c40d2	479	INIT_LIST_HEAD(&d->rexmitq);
68e0d42f	480	d->sysminor = sysminor;
0c966214 EC	481	d->aoemajor = maj;
0c966214 EC	482	d->aoeminor = min;
3a0c40d2 EC	483	d->rttavg = RTTAVG_INIT;
3a0c40d2 EC	484	d->rttdev = RTTDEV_INIT;
68e0d42f EC	485	d->next = devlist;
	486	devlist = d;
	487	out:
3ae1c24e	488	spin_unlock_irqrestore(&devlist_lock, flags);
1da177e4 LT	489	return d;
	490	}
	491
	492	static void
9bb237b6	493	freetgt(struct aoedev d, struct aoetgt t)
1da177e4	494	{
896831f5 EC	495	struct frame *f;
896831f5 EC	496	struct list_head pos, nx, *head;
1b86fda9 EC	497	struct aoeif *ifp;
	498
	499	for (ifp = t->ifs; ifp < &t->ifs[NAOEIFS]; ++ifp) {
	500	if (!ifp->nd)
	501	break;
	502	dev_put(ifp->nd);
	503	}
e407a7f6	504
896831f5 EC	505	head = &t->ffree;
	506	list_for_each_safe(pos, nx, head) {
	507	list_del(pos);
	508	f = list_entry(pos, struct frame, head);
9bb237b6	509	skbfree(f->skb);
896831f5 EC	510	kfree(f);
896831f5 EC	511	}
68e0d42f EC	512	kfree(t);
	513	}
	514
1da177e4 LT	515	void
	516	aoedev_exit(void)
	517	{
e52a2932	518	flush_scheduled_work();
e52a2932	519	flush(NULL, 0, EXITING);
1da177e4 LT	520	}
	521
	522	int __init
	523	aoedev_init(void)
	524	{
1da177e4 LT	525	return 0;
1da177e4 LT	526	}