1 /* Copyright (c) 2013 Coraid, Inc. See COPYING for GPL terms. */
4 * AoE device utility functions; maintains device list.
7 #include <linux/hdreg.h>
8 #include <linux/blkdev.h>
9 #include <linux/netdevice.h>
10 #include <linux/delay.h>
11 #include <linux/slab.h>
12 #include <linux/bitmap.h>
13 #include <linux/kdev_t.h>
14 #include <linux/moduleparam.h>
15 #include <linux/string.h>
18 static void freetgt(struct aoedev *d, struct aoetgt *t);
19 static void skbpoolfree(struct aoedev *d);
21 static int aoe_dyndevs = 1;
22 module_param(aoe_dyndevs, int, 0644);
23 MODULE_PARM_DESC(aoe_dyndevs, "Use dynamic minor numbers for devices.");
25 static struct aoedev *devlist;
26 static DEFINE_SPINLOCK(devlist_lock);
28 /* Because some systems will have one, many, or no
32 * we need some flexibility in the way the minor numbers
33 * are allocated. So they are dynamic.
35 #define N_DEVS ((1U<<MINORBITS)/AOE_PARTITIONS)
37 static DEFINE_SPINLOCK(used_minors_lock);
38 static DECLARE_BITMAP(used_minors, N_DEVS);
41 minor_get_dyn(ulong *sysminor)
47 spin_lock_irqsave(&used_minors_lock, flags);
48 n = find_first_zero_bit(used_minors, N_DEVS);
50 set_bit(n, used_minors);
53 spin_unlock_irqrestore(&used_minors_lock, flags);
55 *sysminor = n * AOE_PARTITIONS;
60 minor_get_static(ulong *sysminor, ulong aoemaj, int aoemin)
66 /* for backwards compatibility when !aoe_dyndevs,
67 * a static number of supported slots per shelf */
71 if (aoemin >= NPERSHELF) {
72 pr_err("aoe: %s %d slots per shelf\n",
73 "static minor device numbers support only",
79 n = aoemaj * NPERSHELF + aoemin;
81 pr_err("aoe: %s with e%ld.%d\n",
82 "cannot use static minor device numbers",
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",
95 set_bit(n, used_minors);
96 spin_unlock_irqrestore(&used_minors_lock, flags);
97 *sysminor = n * AOE_PARTITIONS;
103 minor_get(ulong *sysminor, ulong aoemaj, int aoemin)
106 return minor_get_dyn(sysminor);
108 return minor_get_static(sysminor, aoemaj, aoemin);
112 minor_free(ulong minor)
116 minor /= AOE_PARTITIONS;
117 BUG_ON(minor >= N_DEVS);
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);
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.
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.
138 aoedev_put(struct aoedev *d)
142 spin_lock_irqsave(&devlist_lock, flags);
144 spin_unlock_irqrestore(&devlist_lock, flags);
148 dummy_timer(struct timer_list *t)
152 d = from_timer(d, t, timer);
153 if (d->flags & DEVFL_TKILL)
155 d->timer.expires = jiffies + HZ;
156 add_timer(&d->timer);
160 aoe_failip(struct aoedev *d)
166 aoe_failbuf(d, d->ip.buf);
171 while ((bio = d->ip.nxbio)) {
172 bio->bi_status = BLK_STS_IOERR;
173 d->ip.nxbio = bio->bi_next;
174 n = (unsigned long) rq->special;
175 rq->special = (void *) --n;
177 if ((unsigned long) rq->special == 0)
178 aoe_end_request(d, rq, 0);
182 downdev_frame(struct list_head *pos)
186 f = list_entry(pos, struct frame, head);
189 f->buf->nframesout--;
190 aoe_failbuf(f->t->d, f->buf);
196 aoedev_downdev(struct aoedev *d)
198 struct aoetgt *t, **tt, **te;
199 struct list_head *head, *pos, *nx;
203 d->flags &= ~DEVFL_UP;
205 /* clean out active and to-be-retransmitted buffers */
206 for (i = 0; i < NFACTIVE; i++) {
207 head = &d->factive[i];
208 list_for_each_safe(pos, nx, head)
212 list_for_each_safe(pos, nx, head)
215 /* reset window dressings */
217 te = tt + d->ntargets;
218 for (; tt < te && (t = *tt); tt++) {
223 /* clean out the in-process request (if any) */
226 /* fast fail all pending I/O */
228 while ((rq = blk_peek_request(d->blkq))) {
229 blk_start_request(rq);
230 aoe_end_request(d, rq, 1);
235 set_capacity(d->gd, 0);
238 /* return whether the user asked for this particular
239 * device to be flushed
242 user_req(char *s, size_t slen, struct aoedev *d)
249 p = kbasename(d->gd->disk_name);
250 lim = sizeof(d->gd->disk_name);
251 lim -= p - d->gd->disk_name;
255 return !strncmp(s, p, lim);
259 freedev(struct aoedev *d)
261 struct aoetgt **t, **e;
265 spin_lock_irqsave(&d->lock, flags);
266 if (d->flags & DEVFL_TKILL
267 && !(d->flags & DEVFL_FREEING)) {
268 d->flags |= DEVFL_FREEING;
271 spin_unlock_irqrestore(&d->lock, flags);
275 del_timer_sync(&d->timer);
277 aoedisk_rm_debugfs(d);
281 blk_cleanup_queue(d->blkq);
285 for (; t < e && *t; t++)
288 mempool_destroy(d->bufpool);
290 minor_free(d->sysminor);
292 spin_lock_irqsave(&d->lock, flags);
293 d->flags |= DEVFL_FREED;
294 spin_unlock_irqrestore(&d->lock, flags);
303 flush(const char __user *str, size_t cnt, int exiting)
306 struct aoedev *d, **dd;
309 int specified = 0; /* flush a specific device */
310 unsigned int skipflags;
312 skipflags = DEVFL_GDALLOC | DEVFL_NEWSIZE | DEVFL_TKILL;
314 if (!exiting && cnt >= 3) {
315 if (cnt > sizeof buf)
317 if (copy_from_user(buf, str, cnt))
319 all = !strncmp(buf, "all", 3);
324 flush_scheduled_work();
325 /* pass one: without sleeping, do aoedev_downdev */
326 spin_lock_irqsave(&devlist_lock, flags);
327 for (d = devlist; d; d = d->next) {
330 /* unconditionally take each device down */
331 } else if (specified) {
332 if (!user_req(buf, cnt, d))
334 } else if ((!all && (d->flags & DEVFL_UP))
335 || d->flags & skipflags
341 d->flags |= DEVFL_TKILL;
343 spin_unlock(&d->lock);
345 spin_unlock_irqrestore(&devlist_lock, flags);
347 /* pass two: call freedev, which might sleep,
348 * for aoedevs marked with DEVFL_TKILL
351 spin_lock_irqsave(&devlist_lock, flags);
352 for (d = devlist; d; d = d->next) {
354 if (d->flags & DEVFL_TKILL
355 && !(d->flags & DEVFL_FREEING)) {
356 spin_unlock(&d->lock);
357 spin_unlock_irqrestore(&devlist_lock, flags);
361 spin_unlock(&d->lock);
364 /* pass three: remove aoedevs marked with DEVFL_FREED */
365 for (dd = &devlist, d = *dd; d; d = *dd) {
366 struct aoedev *doomed = NULL;
369 if (d->flags & DEVFL_FREED) {
375 spin_unlock(&d->lock);
377 kfree(doomed->targets);
380 spin_unlock_irqrestore(&devlist_lock, flags);
386 aoedev_flush(const char __user *str, size_t cnt)
388 return flush(str, cnt, NOT_EXITING);
391 /* This has been confirmed to occur once with Tms=3*1000 due to the
392 * driver changing link and not processing its transmit ring. The
393 * problem is hard enough to solve by returning an error that I'm
394 * still punting on "solving" this.
397 skbfree(struct sk_buff *skb)
399 enum { Sms = 250, Tms = 30 * 1000};
404 while (atomic_read(&skb_shinfo(skb)->dataref) != 1 && i-- > 0)
408 "aoe: %s holds ref: %s\n",
409 skb->dev ? skb->dev->name : "netif",
410 "cannot free skb -- memory leaked.");
413 skb->truesize -= skb->data_len;
414 skb_shinfo(skb)->nr_frags = skb->data_len = 0;
420 skbpoolfree(struct aoedev *d)
422 struct sk_buff *skb, *tmp;
424 skb_queue_walk_safe(&d->skbpool, skb, tmp)
427 __skb_queue_head_init(&d->skbpool);
430 /* find it or allocate it */
432 aoedev_by_aoeaddr(ulong maj, int min, int do_alloc)
439 spin_lock_irqsave(&devlist_lock, flags);
441 for (d=devlist; d; d=d->next)
442 if (d->aoemajor == maj && d->aoeminor == min) {
444 if (d->flags & DEVFL_TKILL) {
445 spin_unlock(&d->lock);
450 spin_unlock(&d->lock);
453 if (d || !do_alloc || minor_get(&sysminor, maj, min) < 0)
455 d = kcalloc(1, sizeof *d, GFP_ATOMIC);
458 d->targets = kcalloc(NTARGETS, sizeof(*d->targets), GFP_ATOMIC);
464 d->ntargets = NTARGETS;
465 INIT_WORK(&d->work, aoecmd_sleepwork);
466 spin_lock_init(&d->lock);
467 skb_queue_head_init(&d->skbpool);
468 timer_setup(&d->timer, dummy_timer, 0);
469 d->timer.expires = jiffies + HZ;
470 add_timer(&d->timer);
471 d->bufpool = NULL; /* defer to aoeblk_gdalloc */
474 for (i = 0; i < NFACTIVE; i++)
475 INIT_LIST_HEAD(&d->factive[i]);
476 INIT_LIST_HEAD(&d->rexmitq);
477 d->sysminor = sysminor;
480 d->rttavg = RTTAVG_INIT;
481 d->rttdev = RTTDEV_INIT;
485 spin_unlock_irqrestore(&devlist_lock, flags);
490 freetgt(struct aoedev *d, struct aoetgt *t)
493 struct list_head *pos, *nx, *head;
496 for (ifp = t->ifs; ifp < &t->ifs[NAOEIFS]; ++ifp) {
503 list_for_each_safe(pos, nx, head) {
505 f = list_entry(pos, struct frame, head);
515 flush_scheduled_work();
516 flush(NULL, 0, EXITING);