Commit | Line | Data |
---|---|---|
52e112b3 | 1 | /* Copyright (c) 2007 Coraid, Inc. See COPYING for GPL terms. */ |
1da177e4 LT |
2 | /* |
3 | * aoecmd.c | |
4 | * Filesystem request handling methods | |
5 | */ | |
6 | ||
04b3ab52 | 7 | #include <linux/ata.h> |
5a0e3ad6 | 8 | #include <linux/slab.h> |
1da177e4 LT |
9 | #include <linux/hdreg.h> |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/skbuff.h> | |
12 | #include <linux/netdevice.h> | |
3ae1c24e | 13 | #include <linux/genhd.h> |
68e0d42f | 14 | #include <linux/moduleparam.h> |
896831f5 EC |
15 | #include <linux/workqueue.h> |
16 | #include <linux/kthread.h> | |
881d966b | 17 | #include <net/net_namespace.h> |
475172fb | 18 | #include <asm/unaligned.h> |
896831f5 | 19 | #include <linux/uio.h> |
1da177e4 LT |
20 | #include "aoe.h" |
21 | ||
896831f5 EC |
22 | #define MAXIOC (8192) /* default meant to avoid most soft lockups */ |
23 | ||
24 | static void ktcomplete(struct frame *, struct sk_buff *); | |
25 | ||
69cf2d85 EC |
26 | static struct buf *nextbuf(struct aoedev *); |
27 | ||
b751e8b6 EC |
28 | static int aoe_deadsecs = 60 * 3; |
29 | module_param(aoe_deadsecs, int, 0644); | |
30 | MODULE_PARM_DESC(aoe_deadsecs, "After aoe_deadsecs seconds, give up and fail dev."); | |
1da177e4 | 31 | |
7df620d8 EC |
32 | static int aoe_maxout = 16; |
33 | module_param(aoe_maxout, int, 0644); | |
34 | MODULE_PARM_DESC(aoe_maxout, | |
35 | "Only aoe_maxout outstanding packets for every MAC on eX.Y."); | |
36 | ||
896831f5 EC |
37 | static wait_queue_head_t ktiowq; |
38 | static struct ktstate kts; | |
39 | ||
40 | /* io completion queue */ | |
41 | static struct { | |
42 | struct list_head head; | |
43 | spinlock_t lock; | |
44 | } iocq; | |
45 | ||
68e0d42f | 46 | static struct sk_buff * |
e407a7f6 | 47 | new_skb(ulong len) |
1da177e4 LT |
48 | { |
49 | struct sk_buff *skb; | |
50 | ||
51 | skb = alloc_skb(len, GFP_ATOMIC); | |
52 | if (skb) { | |
459a98ed | 53 | skb_reset_mac_header(skb); |
c1d2bbe1 | 54 | skb_reset_network_header(skb); |
1da177e4 | 55 | skb->protocol = __constant_htons(ETH_P_AOE); |
8babe8cc | 56 | skb_checksum_none_assert(skb); |
1da177e4 LT |
57 | } |
58 | return skb; | |
59 | } | |
60 | ||
1da177e4 | 61 | static struct frame * |
64a80f5a | 62 | getframe(struct aoedev *d, u32 tag) |
1da177e4 | 63 | { |
896831f5 EC |
64 | struct frame *f; |
65 | struct list_head *head, *pos, *nx; | |
66 | u32 n; | |
1da177e4 | 67 | |
896831f5 | 68 | n = tag % NFACTIVE; |
64a80f5a | 69 | head = &d->factive[n]; |
896831f5 EC |
70 | list_for_each_safe(pos, nx, head) { |
71 | f = list_entry(pos, struct frame, head); | |
72 | if (f->tag == tag) { | |
73 | list_del(pos); | |
1da177e4 | 74 | return f; |
896831f5 EC |
75 | } |
76 | } | |
1da177e4 LT |
77 | return NULL; |
78 | } | |
79 | ||
80 | /* | |
81 | * Leave the top bit clear so we have tagspace for userland. | |
82 | * The bottom 16 bits are the xmit tick for rexmit/rttavg processing. | |
83 | * This driver reserves tag -1 to mean "unused frame." | |
84 | */ | |
85 | static int | |
64a80f5a | 86 | newtag(struct aoedev *d) |
1da177e4 LT |
87 | { |
88 | register ulong n; | |
89 | ||
90 | n = jiffies & 0xffff; | |
64a80f5a | 91 | return n |= (++d->lasttag & 0x7fff) << 16; |
1da177e4 LT |
92 | } |
93 | ||
896831f5 | 94 | static u32 |
68e0d42f | 95 | aoehdr_atainit(struct aoedev *d, struct aoetgt *t, struct aoe_hdr *h) |
1da177e4 | 96 | { |
64a80f5a | 97 | u32 host_tag = newtag(d); |
1da177e4 | 98 | |
68e0d42f EC |
99 | memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src); |
100 | memcpy(h->dst, t->addr, sizeof h->dst); | |
63e9cc5d | 101 | h->type = __constant_cpu_to_be16(ETH_P_AOE); |
1da177e4 | 102 | h->verfl = AOE_HVER; |
63e9cc5d | 103 | h->major = cpu_to_be16(d->aoemajor); |
1da177e4 LT |
104 | h->minor = d->aoeminor; |
105 | h->cmd = AOECMD_ATA; | |
63e9cc5d | 106 | h->tag = cpu_to_be32(host_tag); |
1da177e4 LT |
107 | |
108 | return host_tag; | |
109 | } | |
110 | ||
19bf2635 EC |
111 | static inline void |
112 | put_lba(struct aoe_atahdr *ah, sector_t lba) | |
113 | { | |
114 | ah->lba0 = lba; | |
115 | ah->lba1 = lba >>= 8; | |
116 | ah->lba2 = lba >>= 8; | |
117 | ah->lba3 = lba >>= 8; | |
118 | ah->lba4 = lba >>= 8; | |
119 | ah->lba5 = lba >>= 8; | |
120 | } | |
121 | ||
3f0f0133 | 122 | static struct aoeif * |
68e0d42f EC |
123 | ifrotate(struct aoetgt *t) |
124 | { | |
3f0f0133 EC |
125 | struct aoeif *ifp; |
126 | ||
127 | ifp = t->ifp; | |
128 | ifp++; | |
129 | if (ifp >= &t->ifs[NAOEIFS] || ifp->nd == NULL) | |
130 | ifp = t->ifs; | |
131 | if (ifp->nd == NULL) | |
132 | return NULL; | |
133 | return t->ifp = ifp; | |
68e0d42f EC |
134 | } |
135 | ||
9bb237b6 EC |
136 | static void |
137 | skb_pool_put(struct aoedev *d, struct sk_buff *skb) | |
138 | { | |
e9bb8fb0 | 139 | __skb_queue_tail(&d->skbpool, skb); |
9bb237b6 EC |
140 | } |
141 | ||
142 | static struct sk_buff * | |
143 | skb_pool_get(struct aoedev *d) | |
144 | { | |
e9bb8fb0 | 145 | struct sk_buff *skb = skb_peek(&d->skbpool); |
9bb237b6 | 146 | |
9bb237b6 | 147 | if (skb && atomic_read(&skb_shinfo(skb)->dataref) == 1) { |
e9bb8fb0 | 148 | __skb_unlink(skb, &d->skbpool); |
9bb237b6 EC |
149 | return skb; |
150 | } | |
e9bb8fb0 DM |
151 | if (skb_queue_len(&d->skbpool) < NSKBPOOLMAX && |
152 | (skb = new_skb(ETH_ZLEN))) | |
9bb237b6 | 153 | return skb; |
e9bb8fb0 | 154 | |
9bb237b6 EC |
155 | return NULL; |
156 | } | |
157 | ||
896831f5 EC |
158 | void |
159 | aoe_freetframe(struct frame *f) | |
160 | { | |
161 | struct aoetgt *t; | |
162 | ||
163 | t = f->t; | |
164 | f->buf = NULL; | |
165 | f->bv = NULL; | |
166 | f->r_skb = NULL; | |
167 | list_add(&f->head, &t->ffree); | |
168 | } | |
169 | ||
68e0d42f | 170 | static struct frame * |
896831f5 | 171 | newtframe(struct aoedev *d, struct aoetgt *t) |
1da177e4 | 172 | { |
896831f5 | 173 | struct frame *f; |
9bb237b6 | 174 | struct sk_buff *skb; |
896831f5 EC |
175 | struct list_head *pos; |
176 | ||
177 | if (list_empty(&t->ffree)) { | |
178 | if (t->falloc >= NSKBPOOLMAX*2) | |
179 | return NULL; | |
180 | f = kcalloc(1, sizeof(*f), GFP_ATOMIC); | |
181 | if (f == NULL) | |
182 | return NULL; | |
183 | t->falloc++; | |
184 | f->t = t; | |
185 | } else { | |
186 | pos = t->ffree.next; | |
187 | list_del(pos); | |
188 | f = list_entry(pos, struct frame, head); | |
189 | } | |
190 | ||
191 | skb = f->skb; | |
192 | if (skb == NULL) { | |
193 | f->skb = skb = new_skb(ETH_ZLEN); | |
194 | if (!skb) { | |
195 | bail: aoe_freetframe(f); | |
196 | return NULL; | |
197 | } | |
198 | } | |
199 | ||
200 | if (atomic_read(&skb_shinfo(skb)->dataref) != 1) { | |
201 | skb = skb_pool_get(d); | |
202 | if (skb == NULL) | |
203 | goto bail; | |
204 | skb_pool_put(d, f->skb); | |
205 | f->skb = skb; | |
206 | } | |
207 | ||
208 | skb->truesize -= skb->data_len; | |
209 | skb_shinfo(skb)->nr_frags = skb->data_len = 0; | |
210 | skb_trim(skb, 0); | |
211 | return f; | |
212 | } | |
213 | ||
214 | static struct frame * | |
215 | newframe(struct aoedev *d) | |
216 | { | |
217 | struct frame *f; | |
218 | struct aoetgt *t, **tt; | |
219 | int totout = 0; | |
68e0d42f EC |
220 | |
221 | if (d->targets[0] == NULL) { /* shouldn't happen, but I'm paranoid */ | |
222 | printk(KERN_ERR "aoe: NULL TARGETS!\n"); | |
223 | return NULL; | |
224 | } | |
896831f5 | 225 | tt = d->tgt; /* last used target */ |
9bb237b6 | 226 | for (;;) { |
896831f5 EC |
227 | tt++; |
228 | if (tt >= &d->targets[NTARGETS] || !*tt) | |
229 | tt = d->targets; | |
230 | t = *tt; | |
231 | totout += t->nout; | |
232 | if (t->nout < t->maxout | |
9bb237b6 | 233 | && t != d->htgt |
896831f5 EC |
234 | && t->ifp->nd) { |
235 | f = newtframe(d, t); | |
236 | if (f) { | |
896831f5 | 237 | ifrotate(t); |
3f0f0133 | 238 | d->tgt = tt; |
68e0d42f EC |
239 | return f; |
240 | } | |
68e0d42f | 241 | } |
896831f5 | 242 | if (tt == d->tgt) /* we've looped and found nada */ |
9bb237b6 | 243 | break; |
896831f5 EC |
244 | } |
245 | if (totout == 0) { | |
246 | d->kicked++; | |
247 | d->flags |= DEVFL_KICKME; | |
9bb237b6 | 248 | } |
68e0d42f EC |
249 | return NULL; |
250 | } | |
251 | ||
3d5b0605 EC |
252 | static void |
253 | skb_fillup(struct sk_buff *skb, struct bio_vec *bv, ulong off, ulong cnt) | |
254 | { | |
255 | int frag = 0; | |
256 | ulong fcnt; | |
257 | loop: | |
258 | fcnt = bv->bv_len - (off - bv->bv_offset); | |
259 | if (fcnt > cnt) | |
260 | fcnt = cnt; | |
261 | skb_fill_page_desc(skb, frag++, bv->bv_page, off, fcnt); | |
262 | cnt -= fcnt; | |
263 | if (cnt <= 0) | |
264 | return; | |
265 | bv++; | |
266 | off = bv->bv_offset; | |
267 | goto loop; | |
268 | } | |
269 | ||
896831f5 EC |
270 | static void |
271 | fhash(struct frame *f) | |
272 | { | |
64a80f5a | 273 | struct aoedev *d = f->t->d; |
896831f5 EC |
274 | u32 n; |
275 | ||
276 | n = f->tag % NFACTIVE; | |
64a80f5a | 277 | list_add_tail(&f->head, &d->factive[n]); |
896831f5 EC |
278 | } |
279 | ||
68e0d42f EC |
280 | static int |
281 | aoecmd_ata_rw(struct aoedev *d) | |
282 | { | |
283 | struct frame *f; | |
1da177e4 LT |
284 | struct aoe_hdr *h; |
285 | struct aoe_atahdr *ah; | |
286 | struct buf *buf; | |
68e0d42f EC |
287 | struct bio_vec *bv; |
288 | struct aoetgt *t; | |
1da177e4 | 289 | struct sk_buff *skb; |
69cf2d85 | 290 | struct sk_buff_head queue; |
3d5b0605 | 291 | ulong bcnt, fbcnt; |
1da177e4 LT |
292 | char writebit, extbit; |
293 | ||
294 | writebit = 0x10; | |
295 | extbit = 0x4; | |
296 | ||
69cf2d85 EC |
297 | buf = nextbuf(d); |
298 | if (buf == NULL) | |
299 | return 0; | |
896831f5 | 300 | f = newframe(d); |
68e0d42f EC |
301 | if (f == NULL) |
302 | return 0; | |
303 | t = *d->tgt; | |
68e0d42f | 304 | bv = buf->bv; |
3f0f0133 | 305 | bcnt = d->maxbcnt; |
68e0d42f EC |
306 | if (bcnt == 0) |
307 | bcnt = DEFAULTBCNT; | |
3d5b0605 EC |
308 | if (bcnt > buf->resid) |
309 | bcnt = buf->resid; | |
310 | fbcnt = bcnt; | |
311 | f->bv = buf->bv; | |
312 | f->bv_off = f->bv->bv_offset + (f->bv->bv_len - buf->bv_resid); | |
313 | do { | |
314 | if (fbcnt < buf->bv_resid) { | |
315 | buf->bv_resid -= fbcnt; | |
316 | buf->resid -= fbcnt; | |
317 | break; | |
318 | } | |
319 | fbcnt -= buf->bv_resid; | |
320 | buf->resid -= buf->bv_resid; | |
321 | if (buf->resid == 0) { | |
69cf2d85 | 322 | d->ip.buf = NULL; |
3d5b0605 EC |
323 | break; |
324 | } | |
325 | buf->bv++; | |
326 | buf->bv_resid = buf->bv->bv_len; | |
327 | WARN_ON(buf->bv_resid == 0); | |
328 | } while (fbcnt); | |
329 | ||
1da177e4 | 330 | /* initialize the headers & frame */ |
e407a7f6 | 331 | skb = f->skb; |
abdbf94d | 332 | h = (struct aoe_hdr *) skb_mac_header(skb); |
1da177e4 | 333 | ah = (struct aoe_atahdr *) (h+1); |
19900cde EC |
334 | skb_put(skb, sizeof *h + sizeof *ah); |
335 | memset(h, 0, skb->len); | |
68e0d42f | 336 | f->tag = aoehdr_atainit(d, t, h); |
896831f5 | 337 | fhash(f); |
68e0d42f | 338 | t->nout++; |
1da177e4 LT |
339 | f->waited = 0; |
340 | f->buf = buf; | |
19bf2635 | 341 | f->bcnt = bcnt; |
68e0d42f | 342 | f->lba = buf->sector; |
1da177e4 LT |
343 | |
344 | /* set up ata header */ | |
345 | ah->scnt = bcnt >> 9; | |
68e0d42f | 346 | put_lba(ah, buf->sector); |
1da177e4 LT |
347 | if (d->flags & DEVFL_EXT) { |
348 | ah->aflags |= AOEAFL_EXT; | |
1da177e4 LT |
349 | } else { |
350 | extbit = 0; | |
351 | ah->lba3 &= 0x0f; | |
352 | ah->lba3 |= 0xe0; /* LBA bit + obsolete 0xa0 */ | |
353 | } | |
1da177e4 | 354 | if (bio_data_dir(buf->bio) == WRITE) { |
3d5b0605 | 355 | skb_fillup(skb, f->bv, f->bv_off, bcnt); |
1da177e4 | 356 | ah->aflags |= AOEAFL_WRITE; |
4f51dc5e EC |
357 | skb->len += bcnt; |
358 | skb->data_len = bcnt; | |
3d5b0605 | 359 | skb->truesize += bcnt; |
68e0d42f | 360 | t->wpkts++; |
1da177e4 | 361 | } else { |
68e0d42f | 362 | t->rpkts++; |
1da177e4 | 363 | writebit = 0; |
1da177e4 LT |
364 | } |
365 | ||
04b3ab52 | 366 | ah->cmdstat = ATA_CMD_PIO_READ | writebit | extbit; |
1da177e4 LT |
367 | |
368 | /* mark all tracking fields and load out */ | |
369 | buf->nframesout += 1; | |
1da177e4 | 370 | buf->sector += bcnt >> 9; |
1da177e4 | 371 | |
68e0d42f | 372 | skb->dev = t->ifp->nd; |
4f51dc5e | 373 | skb = skb_clone(skb, GFP_ATOMIC); |
69cf2d85 EC |
374 | if (skb) { |
375 | __skb_queue_head_init(&queue); | |
376 | __skb_queue_tail(&queue, skb); | |
377 | aoenet_xmit(&queue); | |
378 | } | |
68e0d42f | 379 | return 1; |
1da177e4 LT |
380 | } |
381 | ||
3ae1c24e EC |
382 | /* some callers cannot sleep, and they can call this function, |
383 | * transmitting the packets later, when interrupts are on | |
384 | */ | |
e9bb8fb0 DM |
385 | static void |
386 | aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff_head *queue) | |
3ae1c24e EC |
387 | { |
388 | struct aoe_hdr *h; | |
389 | struct aoe_cfghdr *ch; | |
e9bb8fb0 | 390 | struct sk_buff *skb; |
3ae1c24e EC |
391 | struct net_device *ifp; |
392 | ||
840a185d ED |
393 | rcu_read_lock(); |
394 | for_each_netdev_rcu(&init_net, ifp) { | |
3ae1c24e EC |
395 | dev_hold(ifp); |
396 | if (!is_aoe_netif(ifp)) | |
7562f876 | 397 | goto cont; |
3ae1c24e | 398 | |
e407a7f6 | 399 | skb = new_skb(sizeof *h + sizeof *ch); |
3ae1c24e | 400 | if (skb == NULL) { |
a12c93f0 | 401 | printk(KERN_INFO "aoe: skb alloc failure\n"); |
7562f876 | 402 | goto cont; |
3ae1c24e | 403 | } |
19900cde | 404 | skb_put(skb, sizeof *h + sizeof *ch); |
e407a7f6 | 405 | skb->dev = ifp; |
e9bb8fb0 | 406 | __skb_queue_tail(queue, skb); |
abdbf94d | 407 | h = (struct aoe_hdr *) skb_mac_header(skb); |
3ae1c24e EC |
408 | memset(h, 0, sizeof *h + sizeof *ch); |
409 | ||
410 | memset(h->dst, 0xff, sizeof h->dst); | |
411 | memcpy(h->src, ifp->dev_addr, sizeof h->src); | |
412 | h->type = __constant_cpu_to_be16(ETH_P_AOE); | |
413 | h->verfl = AOE_HVER; | |
414 | h->major = cpu_to_be16(aoemajor); | |
415 | h->minor = aoeminor; | |
416 | h->cmd = AOECMD_CFG; | |
417 | ||
7562f876 PE |
418 | cont: |
419 | dev_put(ifp); | |
3ae1c24e | 420 | } |
840a185d | 421 | rcu_read_unlock(); |
3ae1c24e EC |
422 | } |
423 | ||
1da177e4 | 424 | static void |
896831f5 | 425 | resend(struct aoedev *d, struct frame *f) |
1da177e4 LT |
426 | { |
427 | struct sk_buff *skb; | |
69cf2d85 | 428 | struct sk_buff_head queue; |
1da177e4 | 429 | struct aoe_hdr *h; |
19bf2635 | 430 | struct aoe_atahdr *ah; |
896831f5 | 431 | struct aoetgt *t; |
1da177e4 LT |
432 | char buf[128]; |
433 | u32 n; | |
1da177e4 | 434 | |
896831f5 | 435 | t = f->t; |
64a80f5a | 436 | n = newtag(d); |
68e0d42f | 437 | skb = f->skb; |
3f0f0133 EC |
438 | if (ifrotate(t) == NULL) { |
439 | /* probably can't happen, but set it up to fail anyway */ | |
440 | pr_info("aoe: resend: no interfaces to rotate to.\n"); | |
441 | ktcomplete(f, NULL); | |
442 | return; | |
443 | } | |
68e0d42f EC |
444 | h = (struct aoe_hdr *) skb_mac_header(skb); |
445 | ah = (struct aoe_atahdr *) (h+1); | |
1da177e4 LT |
446 | |
447 | snprintf(buf, sizeof buf, | |
411c41ee | 448 | "%15s e%ld.%d oldtag=%08x@%08lx newtag=%08x s=%pm d=%pm nout=%d\n", |
68e0d42f | 449 | "retransmit", d->aoemajor, d->aoeminor, f->tag, jiffies, n, |
411c41ee | 450 | h->src, h->dst, t->nout); |
1da177e4 LT |
451 | aoechr_error(buf); |
452 | ||
1da177e4 | 453 | f->tag = n; |
896831f5 | 454 | fhash(f); |
63e9cc5d | 455 | h->tag = cpu_to_be32(n); |
68e0d42f EC |
456 | memcpy(h->dst, t->addr, sizeof h->dst); |
457 | memcpy(h->src, t->ifp->nd->dev_addr, sizeof h->src); | |
458 | ||
68e0d42f | 459 | skb->dev = t->ifp->nd; |
4f51dc5e EC |
460 | skb = skb_clone(skb, GFP_ATOMIC); |
461 | if (skb == NULL) | |
462 | return; | |
69cf2d85 EC |
463 | __skb_queue_head_init(&queue); |
464 | __skb_queue_tail(&queue, skb); | |
465 | aoenet_xmit(&queue); | |
1da177e4 LT |
466 | } |
467 | ||
468 | static int | |
896831f5 | 469 | tsince(u32 tag) |
1da177e4 LT |
470 | { |
471 | int n; | |
472 | ||
473 | n = jiffies & 0xffff; | |
474 | n -= tag & 0xffff; | |
475 | if (n < 0) | |
476 | n += 1<<16; | |
477 | return n; | |
478 | } | |
479 | ||
68e0d42f EC |
480 | static struct aoeif * |
481 | getif(struct aoetgt *t, struct net_device *nd) | |
482 | { | |
483 | struct aoeif *p, *e; | |
484 | ||
485 | p = t->ifs; | |
486 | e = p + NAOEIFS; | |
487 | for (; p < e; p++) | |
488 | if (p->nd == nd) | |
489 | return p; | |
490 | return NULL; | |
491 | } | |
492 | ||
68e0d42f EC |
493 | static void |
494 | ejectif(struct aoetgt *t, struct aoeif *ifp) | |
495 | { | |
496 | struct aoeif *e; | |
1b86fda9 | 497 | struct net_device *nd; |
68e0d42f EC |
498 | ulong n; |
499 | ||
1b86fda9 | 500 | nd = ifp->nd; |
68e0d42f EC |
501 | e = t->ifs + NAOEIFS - 1; |
502 | n = (e - ifp) * sizeof *ifp; | |
503 | memmove(ifp, ifp+1, n); | |
504 | e->nd = NULL; | |
1b86fda9 | 505 | dev_put(nd); |
68e0d42f EC |
506 | } |
507 | ||
508 | static int | |
509 | sthtith(struct aoedev *d) | |
510 | { | |
896831f5 EC |
511 | struct frame *f, *nf; |
512 | struct list_head *nx, *pos, *head; | |
68e0d42f | 513 | struct sk_buff *skb; |
896831f5 EC |
514 | struct aoetgt *ht = d->htgt; |
515 | int i; | |
516 | ||
517 | for (i = 0; i < NFACTIVE; i++) { | |
64a80f5a | 518 | head = &d->factive[i]; |
896831f5 EC |
519 | list_for_each_safe(pos, nx, head) { |
520 | f = list_entry(pos, struct frame, head); | |
64a80f5a EC |
521 | if (f->t != ht) |
522 | continue; | |
523 | ||
896831f5 EC |
524 | nf = newframe(d); |
525 | if (!nf) | |
526 | return 0; | |
527 | ||
528 | /* remove frame from active list */ | |
529 | list_del(pos); | |
530 | ||
531 | /* reassign all pertinent bits to new outbound frame */ | |
532 | skb = nf->skb; | |
533 | nf->skb = f->skb; | |
534 | nf->buf = f->buf; | |
535 | nf->bcnt = f->bcnt; | |
536 | nf->lba = f->lba; | |
537 | nf->bv = f->bv; | |
538 | nf->bv_off = f->bv_off; | |
539 | nf->waited = 0; | |
540 | f->skb = skb; | |
541 | aoe_freetframe(f); | |
542 | ht->nout--; | |
543 | nf->t->nout++; | |
544 | resend(d, nf); | |
545 | } | |
68e0d42f | 546 | } |
3f0f0133 EC |
547 | /* We've cleaned up the outstanding so take away his |
548 | * interfaces so he won't be used. We should remove him from | |
549 | * the target array here, but cleaning up a target is | |
550 | * involved. PUNT! | |
551 | */ | |
68e0d42f EC |
552 | memset(ht->ifs, 0, sizeof ht->ifs); |
553 | d->htgt = NULL; | |
554 | return 1; | |
555 | } | |
556 | ||
557 | static inline unsigned char | |
558 | ata_scnt(unsigned char *packet) { | |
559 | struct aoe_hdr *h; | |
560 | struct aoe_atahdr *ah; | |
561 | ||
562 | h = (struct aoe_hdr *) packet; | |
563 | ah = (struct aoe_atahdr *) (h+1); | |
564 | return ah->scnt; | |
565 | } | |
566 | ||
1da177e4 LT |
567 | static void |
568 | rexmit_timer(ulong vp) | |
569 | { | |
570 | struct aoedev *d; | |
68e0d42f EC |
571 | struct aoetgt *t, **tt, **te; |
572 | struct aoeif *ifp; | |
896831f5 EC |
573 | struct frame *f; |
574 | struct list_head *head, *pos, *nx; | |
575 | LIST_HEAD(flist); | |
1da177e4 LT |
576 | register long timeout; |
577 | ulong flags, n; | |
896831f5 | 578 | int i; |
1da177e4 LT |
579 | |
580 | d = (struct aoedev *) vp; | |
1da177e4 LT |
581 | |
582 | /* timeout is always ~150% of the moving average */ | |
583 | timeout = d->rttavg; | |
584 | timeout += timeout >> 1; | |
585 | ||
586 | spin_lock_irqsave(&d->lock, flags); | |
587 | ||
588 | if (d->flags & DEVFL_TKILL) { | |
1c6f3fca | 589 | spin_unlock_irqrestore(&d->lock, flags); |
1da177e4 LT |
590 | return; |
591 | } | |
896831f5 EC |
592 | |
593 | /* collect all frames to rexmit into flist */ | |
64a80f5a EC |
594 | for (i = 0; i < NFACTIVE; i++) { |
595 | head = &d->factive[i]; | |
596 | list_for_each_safe(pos, nx, head) { | |
597 | f = list_entry(pos, struct frame, head); | |
598 | if (tsince(f->tag) < timeout) | |
599 | break; /* end of expired frames */ | |
600 | /* move to flist for later processing */ | |
601 | list_move_tail(pos, &flist); | |
68e0d42f | 602 | } |
64a80f5a EC |
603 | } |
604 | /* window check */ | |
605 | tt = d->targets; | |
606 | te = tt + d->ntargets; | |
607 | for (; tt < te && (t = *tt); tt++) { | |
68e0d42f EC |
608 | if (t->nout == t->maxout |
609 | && t->maxout < t->nframes | |
610 | && (jiffies - t->lastwadj)/HZ > 10) { | |
611 | t->maxout++; | |
612 | t->lastwadj = jiffies; | |
1da177e4 LT |
613 | } |
614 | } | |
68e0d42f | 615 | |
69cf2d85 EC |
616 | if (!list_empty(&flist)) { /* retransmissions necessary */ |
617 | n = d->rttavg <<= 1; | |
618 | if (n > MAXTIMER) | |
619 | d->rttavg = MAXTIMER; | |
620 | } | |
621 | ||
896831f5 EC |
622 | /* process expired frames */ |
623 | while (!list_empty(&flist)) { | |
624 | pos = flist.next; | |
625 | f = list_entry(pos, struct frame, head); | |
626 | n = f->waited += timeout; | |
627 | n /= HZ; | |
628 | if (n > aoe_deadsecs) { | |
629 | /* Waited too long. Device failure. | |
630 | * Hang all frames on first hash bucket for downdev | |
631 | * to clean up. | |
632 | */ | |
64a80f5a | 633 | list_splice(&flist, &d->factive[0]); |
896831f5 EC |
634 | aoedev_downdev(d); |
635 | break; | |
636 | } | |
637 | list_del(pos); | |
638 | ||
639 | t = f->t; | |
d54d35ac EC |
640 | if (n > aoe_deadsecs/2) |
641 | d->htgt = t; /* see if another target can help */ | |
642 | ||
896831f5 EC |
643 | if (t->nout == t->maxout) { |
644 | if (t->maxout > 1) | |
645 | t->maxout--; | |
646 | t->lastwadj = jiffies; | |
647 | } | |
648 | ||
649 | ifp = getif(t, f->skb->dev); | |
650 | if (ifp && ++ifp->lost > (t->nframes << 1) | |
651 | && (ifp != t->ifs || t->ifs[1].nd)) { | |
652 | ejectif(t, ifp); | |
653 | ifp = NULL; | |
654 | } | |
655 | resend(d, f); | |
656 | } | |
657 | ||
69cf2d85 | 658 | if ((d->flags & DEVFL_KICKME || d->htgt) && d->blkq) { |
4f51dc5e | 659 | d->flags &= ~DEVFL_KICKME; |
69cf2d85 | 660 | d->blkq->request_fn(d->blkq); |
4f51dc5e | 661 | } |
1da177e4 | 662 | |
1da177e4 LT |
663 | d->timer.expires = jiffies + TIMERTICK; |
664 | add_timer(&d->timer); | |
665 | ||
666 | spin_unlock_irqrestore(&d->lock, flags); | |
69cf2d85 | 667 | } |
1da177e4 | 668 | |
69cf2d85 EC |
669 | static unsigned long |
670 | rqbiocnt(struct request *r) | |
671 | { | |
672 | struct bio *bio; | |
673 | unsigned long n = 0; | |
674 | ||
675 | __rq_for_each_bio(bio, r) | |
676 | n++; | |
677 | return n; | |
678 | } | |
679 | ||
680 | /* This can be removed if we are certain that no users of the block | |
681 | * layer will ever use zero-count pages in bios. Otherwise we have to | |
682 | * protect against the put_page sometimes done by the network layer. | |
683 | * | |
684 | * See http://oss.sgi.com/archives/xfs/2007-01/msg00594.html for | |
685 | * discussion. | |
686 | * | |
687 | * We cannot use get_page in the workaround, because it insists on a | |
688 | * positive page count as a precondition. So we use _count directly. | |
689 | */ | |
690 | static void | |
691 | bio_pageinc(struct bio *bio) | |
692 | { | |
693 | struct bio_vec *bv; | |
694 | struct page *page; | |
695 | int i; | |
696 | ||
697 | bio_for_each_segment(bv, bio, i) { | |
698 | page = bv->bv_page; | |
699 | /* Non-zero page count for non-head members of | |
700 | * compound pages is no longer allowed by the kernel, | |
701 | * but this has never been seen here. | |
702 | */ | |
703 | if (unlikely(PageCompound(page))) | |
704 | if (compound_trans_head(page) != page) { | |
705 | pr_crit("page tail used for block I/O\n"); | |
706 | BUG(); | |
707 | } | |
708 | atomic_inc(&page->_count); | |
709 | } | |
710 | } | |
711 | ||
712 | static void | |
713 | bio_pagedec(struct bio *bio) | |
714 | { | |
715 | struct bio_vec *bv; | |
716 | int i; | |
717 | ||
718 | bio_for_each_segment(bv, bio, i) | |
719 | atomic_dec(&bv->bv_page->_count); | |
720 | } | |
721 | ||
722 | static void | |
723 | bufinit(struct buf *buf, struct request *rq, struct bio *bio) | |
724 | { | |
725 | struct bio_vec *bv; | |
726 | ||
727 | memset(buf, 0, sizeof(*buf)); | |
728 | buf->rq = rq; | |
729 | buf->bio = bio; | |
730 | buf->resid = bio->bi_size; | |
731 | buf->sector = bio->bi_sector; | |
732 | bio_pageinc(bio); | |
733 | buf->bv = bv = &bio->bi_io_vec[bio->bi_idx]; | |
734 | buf->bv_resid = bv->bv_len; | |
735 | WARN_ON(buf->bv_resid == 0); | |
736 | } | |
737 | ||
738 | static struct buf * | |
739 | nextbuf(struct aoedev *d) | |
740 | { | |
741 | struct request *rq; | |
742 | struct request_queue *q; | |
743 | struct buf *buf; | |
744 | struct bio *bio; | |
745 | ||
746 | q = d->blkq; | |
747 | if (q == NULL) | |
748 | return NULL; /* initializing */ | |
749 | if (d->ip.buf) | |
750 | return d->ip.buf; | |
751 | rq = d->ip.rq; | |
752 | if (rq == NULL) { | |
753 | rq = blk_peek_request(q); | |
754 | if (rq == NULL) | |
755 | return NULL; | |
756 | blk_start_request(rq); | |
757 | d->ip.rq = rq; | |
758 | d->ip.nxbio = rq->bio; | |
759 | rq->special = (void *) rqbiocnt(rq); | |
760 | } | |
761 | buf = mempool_alloc(d->bufpool, GFP_ATOMIC); | |
762 | if (buf == NULL) { | |
763 | pr_err("aoe: nextbuf: unable to mempool_alloc!\n"); | |
764 | return NULL; | |
765 | } | |
766 | bio = d->ip.nxbio; | |
767 | bufinit(buf, rq, bio); | |
768 | bio = bio->bi_next; | |
769 | d->ip.nxbio = bio; | |
770 | if (bio == NULL) | |
771 | d->ip.rq = NULL; | |
772 | return d->ip.buf = buf; | |
1da177e4 LT |
773 | } |
774 | ||
68e0d42f EC |
775 | /* enters with d->lock held */ |
776 | void | |
777 | aoecmd_work(struct aoedev *d) | |
778 | { | |
68e0d42f EC |
779 | if (d->htgt && !sthtith(d)) |
780 | return; | |
69cf2d85 EC |
781 | while (aoecmd_ata_rw(d)) |
782 | ; | |
68e0d42f EC |
783 | } |
784 | ||
3ae1c24e EC |
785 | /* this function performs work that has been deferred until sleeping is OK |
786 | */ | |
787 | void | |
c4028958 | 788 | aoecmd_sleepwork(struct work_struct *work) |
3ae1c24e | 789 | { |
c4028958 | 790 | struct aoedev *d = container_of(work, struct aoedev, work); |
3ae1c24e EC |
791 | |
792 | if (d->flags & DEVFL_GDALLOC) | |
793 | aoeblk_gdalloc(d); | |
794 | ||
795 | if (d->flags & DEVFL_NEWSIZE) { | |
796 | struct block_device *bd; | |
797 | unsigned long flags; | |
798 | u64 ssize; | |
799 | ||
80795aef | 800 | ssize = get_capacity(d->gd); |
3ae1c24e EC |
801 | bd = bdget_disk(d->gd, 0); |
802 | ||
803 | if (bd) { | |
804 | mutex_lock(&bd->bd_inode->i_mutex); | |
805 | i_size_write(bd->bd_inode, (loff_t)ssize<<9); | |
806 | mutex_unlock(&bd->bd_inode->i_mutex); | |
807 | bdput(bd); | |
808 | } | |
809 | spin_lock_irqsave(&d->lock, flags); | |
810 | d->flags |= DEVFL_UP; | |
811 | d->flags &= ~DEVFL_NEWSIZE; | |
812 | spin_unlock_irqrestore(&d->lock, flags); | |
813 | } | |
814 | } | |
815 | ||
1da177e4 | 816 | static void |
68e0d42f | 817 | ataid_complete(struct aoedev *d, struct aoetgt *t, unsigned char *id) |
1da177e4 LT |
818 | { |
819 | u64 ssize; | |
820 | u16 n; | |
821 | ||
822 | /* word 83: command set supported */ | |
f885f8d1 | 823 | n = get_unaligned_le16(&id[83 << 1]); |
1da177e4 LT |
824 | |
825 | /* word 86: command set/feature enabled */ | |
f885f8d1 | 826 | n |= get_unaligned_le16(&id[86 << 1]); |
1da177e4 LT |
827 | |
828 | if (n & (1<<10)) { /* bit 10: LBA 48 */ | |
829 | d->flags |= DEVFL_EXT; | |
830 | ||
831 | /* word 100: number lba48 sectors */ | |
f885f8d1 | 832 | ssize = get_unaligned_le64(&id[100 << 1]); |
1da177e4 LT |
833 | |
834 | /* set as in ide-disk.c:init_idedisk_capacity */ | |
835 | d->geo.cylinders = ssize; | |
836 | d->geo.cylinders /= (255 * 63); | |
837 | d->geo.heads = 255; | |
838 | d->geo.sectors = 63; | |
839 | } else { | |
840 | d->flags &= ~DEVFL_EXT; | |
841 | ||
842 | /* number lba28 sectors */ | |
f885f8d1 | 843 | ssize = get_unaligned_le32(&id[60 << 1]); |
1da177e4 LT |
844 | |
845 | /* NOTE: obsolete in ATA 6 */ | |
f885f8d1 HH |
846 | d->geo.cylinders = get_unaligned_le16(&id[54 << 1]); |
847 | d->geo.heads = get_unaligned_le16(&id[55 << 1]); | |
848 | d->geo.sectors = get_unaligned_le16(&id[56 << 1]); | |
1da177e4 | 849 | } |
3ae1c24e EC |
850 | |
851 | if (d->ssize != ssize) | |
1d75981a | 852 | printk(KERN_INFO |
411c41ee HH |
853 | "aoe: %pm e%ld.%d v%04x has %llu sectors\n", |
854 | t->addr, | |
3ae1c24e EC |
855 | d->aoemajor, d->aoeminor, |
856 | d->fw_ver, (long long)ssize); | |
1da177e4 LT |
857 | d->ssize = ssize; |
858 | d->geo.start = 0; | |
6b9699bb EC |
859 | if (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE)) |
860 | return; | |
1da177e4 | 861 | if (d->gd != NULL) { |
80795aef | 862 | set_capacity(d->gd, ssize); |
3ae1c24e | 863 | d->flags |= DEVFL_NEWSIZE; |
68e0d42f | 864 | } else |
3ae1c24e | 865 | d->flags |= DEVFL_GDALLOC; |
1da177e4 | 866 | schedule_work(&d->work); |
1da177e4 LT |
867 | } |
868 | ||
869 | static void | |
870 | calc_rttavg(struct aoedev *d, int rtt) | |
871 | { | |
872 | register long n; | |
873 | ||
874 | n = rtt; | |
dced3a05 EC |
875 | if (n < 0) { |
876 | n = -rtt; | |
877 | if (n < MINTIMER) | |
878 | n = MINTIMER; | |
879 | else if (n > MAXTIMER) | |
880 | n = MAXTIMER; | |
881 | d->mintimer += (n - d->mintimer) >> 1; | |
882 | } else if (n < d->mintimer) | |
883 | n = d->mintimer; | |
1da177e4 LT |
884 | else if (n > MAXTIMER) |
885 | n = MAXTIMER; | |
886 | ||
887 | /* g == .25; cf. Congestion Avoidance and Control, Jacobson & Karels; 1988 */ | |
888 | n -= d->rttavg; | |
889 | d->rttavg += n >> 2; | |
890 | } | |
891 | ||
68e0d42f EC |
892 | static struct aoetgt * |
893 | gettgt(struct aoedev *d, char *addr) | |
894 | { | |
895 | struct aoetgt **t, **e; | |
896 | ||
897 | t = d->targets; | |
898 | e = t + NTARGETS; | |
899 | for (; t < e && *t; t++) | |
900 | if (memcmp((*t)->addr, addr, sizeof((*t)->addr)) == 0) | |
901 | return *t; | |
902 | return NULL; | |
903 | } | |
904 | ||
3d5b0605 | 905 | static void |
896831f5 | 906 | bvcpy(struct bio_vec *bv, ulong off, struct sk_buff *skb, long cnt) |
3d5b0605 EC |
907 | { |
908 | ulong fcnt; | |
909 | char *p; | |
910 | int soff = 0; | |
911 | loop: | |
912 | fcnt = bv->bv_len - (off - bv->bv_offset); | |
913 | if (fcnt > cnt) | |
914 | fcnt = cnt; | |
915 | p = page_address(bv->bv_page) + off; | |
916 | skb_copy_bits(skb, soff, p, fcnt); | |
917 | soff += fcnt; | |
918 | cnt -= fcnt; | |
919 | if (cnt <= 0) | |
920 | return; | |
921 | bv++; | |
922 | off = bv->bv_offset; | |
923 | goto loop; | |
924 | } | |
925 | ||
69cf2d85 EC |
926 | void |
927 | aoe_end_request(struct aoedev *d, struct request *rq, int fastfail) | |
928 | { | |
929 | struct bio *bio; | |
930 | int bok; | |
931 | struct request_queue *q; | |
932 | ||
933 | q = d->blkq; | |
934 | if (rq == d->ip.rq) | |
935 | d->ip.rq = NULL; | |
936 | do { | |
937 | bio = rq->bio; | |
938 | bok = !fastfail && test_bit(BIO_UPTODATE, &bio->bi_flags); | |
939 | } while (__blk_end_request(rq, bok ? 0 : -EIO, bio->bi_size)); | |
940 | ||
941 | /* cf. http://lkml.org/lkml/2006/10/31/28 */ | |
942 | if (!fastfail) | |
943 | q->request_fn(q); | |
944 | } | |
945 | ||
946 | static void | |
947 | aoe_end_buf(struct aoedev *d, struct buf *buf) | |
948 | { | |
949 | struct request *rq; | |
950 | unsigned long n; | |
951 | ||
952 | if (buf == d->ip.buf) | |
953 | d->ip.buf = NULL; | |
954 | rq = buf->rq; | |
955 | bio_pagedec(buf->bio); | |
956 | mempool_free(buf, d->bufpool); | |
957 | n = (unsigned long) rq->special; | |
958 | rq->special = (void *) --n; | |
959 | if (n == 0) | |
960 | aoe_end_request(d, rq, 0); | |
961 | } | |
962 | ||
3d5b0605 | 963 | static void |
896831f5 | 964 | ktiocomplete(struct frame *f) |
3d5b0605 | 965 | { |
896831f5 EC |
966 | struct aoe_hdr *hin, *hout; |
967 | struct aoe_atahdr *ahin, *ahout; | |
968 | struct buf *buf; | |
969 | struct sk_buff *skb; | |
970 | struct aoetgt *t; | |
971 | struct aoeif *ifp; | |
972 | struct aoedev *d; | |
973 | long n; | |
3d5b0605 | 974 | |
896831f5 | 975 | if (f == NULL) |
3d5b0605 | 976 | return; |
896831f5 EC |
977 | |
978 | t = f->t; | |
979 | d = t->d; | |
980 | ||
981 | hout = (struct aoe_hdr *) skb_mac_header(f->skb); | |
982 | ahout = (struct aoe_atahdr *) (hout+1); | |
983 | buf = f->buf; | |
984 | skb = f->r_skb; | |
985 | if (skb == NULL) | |
986 | goto noskb; /* just fail the buf. */ | |
987 | ||
988 | hin = (struct aoe_hdr *) skb->data; | |
989 | skb_pull(skb, sizeof(*hin)); | |
990 | ahin = (struct aoe_atahdr *) skb->data; | |
991 | skb_pull(skb, sizeof(*ahin)); | |
992 | if (ahin->cmdstat & 0xa9) { /* these bits cleared on success */ | |
993 | pr_err("aoe: ata error cmd=%2.2Xh stat=%2.2Xh from e%ld.%d\n", | |
994 | ahout->cmdstat, ahin->cmdstat, | |
995 | d->aoemajor, d->aoeminor); | |
996 | noskb: if (buf) | |
69cf2d85 | 997 | clear_bit(BIO_UPTODATE, &buf->bio->bi_flags); |
896831f5 | 998 | goto badrsp; |
3d5b0605 | 999 | } |
896831f5 EC |
1000 | |
1001 | n = ahout->scnt << 9; | |
1002 | switch (ahout->cmdstat) { | |
1003 | case ATA_CMD_PIO_READ: | |
1004 | case ATA_CMD_PIO_READ_EXT: | |
1005 | if (skb->len < n) { | |
1006 | pr_err("aoe: runt data size in read. skb->len=%d need=%ld\n", | |
1007 | skb->len, n); | |
69cf2d85 | 1008 | clear_bit(BIO_UPTODATE, &buf->bio->bi_flags); |
896831f5 EC |
1009 | break; |
1010 | } | |
1011 | bvcpy(f->bv, f->bv_off, skb, n); | |
1012 | case ATA_CMD_PIO_WRITE: | |
1013 | case ATA_CMD_PIO_WRITE_EXT: | |
1014 | spin_lock_irq(&d->lock); | |
1015 | ifp = getif(t, skb->dev); | |
3f0f0133 | 1016 | if (ifp) |
896831f5 | 1017 | ifp->lost = 0; |
896831f5 EC |
1018 | if (d->htgt == t) /* I'll help myself, thank you. */ |
1019 | d->htgt = NULL; | |
1020 | spin_unlock_irq(&d->lock); | |
1021 | break; | |
1022 | case ATA_CMD_ID_ATA: | |
1023 | if (skb->len < 512) { | |
1024 | pr_info("aoe: runt data size in ataid. skb->len=%d\n", | |
1025 | skb->len); | |
1026 | break; | |
1027 | } | |
1028 | if (skb_linearize(skb)) | |
1029 | break; | |
1030 | spin_lock_irq(&d->lock); | |
1031 | ataid_complete(d, t, skb->data); | |
1032 | spin_unlock_irq(&d->lock); | |
1033 | break; | |
1034 | default: | |
1035 | pr_info("aoe: unrecognized ata command %2.2Xh for %d.%d\n", | |
1036 | ahout->cmdstat, | |
1037 | be16_to_cpu(get_unaligned(&hin->major)), | |
1038 | hin->minor); | |
1039 | } | |
1040 | badrsp: | |
1041 | spin_lock_irq(&d->lock); | |
1042 | ||
1043 | aoe_freetframe(f); | |
1044 | ||
69cf2d85 EC |
1045 | if (buf && --buf->nframesout == 0 && buf->resid == 0) |
1046 | aoe_end_buf(d, buf); | |
896831f5 | 1047 | |
69cf2d85 EC |
1048 | aoecmd_work(d); |
1049 | ||
1050 | spin_unlock_irq(&d->lock); | |
1051 | aoedev_put(d); | |
896831f5 | 1052 | dev_kfree_skb(skb); |
3d5b0605 EC |
1053 | } |
1054 | ||
896831f5 EC |
1055 | /* Enters with iocq.lock held. |
1056 | * Returns true iff responses needing processing remain. | |
1057 | */ | |
1058 | static int | |
1059 | ktio(void) | |
1060 | { | |
1061 | struct frame *f; | |
1062 | struct list_head *pos; | |
1063 | int i; | |
1064 | ||
1065 | for (i = 0; ; ++i) { | |
1066 | if (i == MAXIOC) | |
1067 | return 1; | |
1068 | if (list_empty(&iocq.head)) | |
1069 | return 0; | |
1070 | pos = iocq.head.next; | |
1071 | list_del(pos); | |
1072 | spin_unlock_irq(&iocq.lock); | |
1073 | f = list_entry(pos, struct frame, head); | |
1074 | ktiocomplete(f); | |
1075 | spin_lock_irq(&iocq.lock); | |
1076 | } | |
1077 | } | |
1078 | ||
1079 | static int | |
1080 | kthread(void *vp) | |
1081 | { | |
1082 | struct ktstate *k; | |
1083 | DECLARE_WAITQUEUE(wait, current); | |
1084 | int more; | |
1085 | ||
1086 | k = vp; | |
1087 | current->flags |= PF_NOFREEZE; | |
1088 | set_user_nice(current, -10); | |
1089 | complete(&k->rendez); /* tell spawner we're running */ | |
1090 | do { | |
1091 | spin_lock_irq(k->lock); | |
1092 | more = k->fn(); | |
1093 | if (!more) { | |
1094 | add_wait_queue(k->waitq, &wait); | |
1095 | __set_current_state(TASK_INTERRUPTIBLE); | |
1096 | } | |
1097 | spin_unlock_irq(k->lock); | |
1098 | if (!more) { | |
1099 | schedule(); | |
1100 | remove_wait_queue(k->waitq, &wait); | |
1101 | } else | |
1102 | cond_resched(); | |
1103 | } while (!kthread_should_stop()); | |
1104 | complete(&k->rendez); /* tell spawner we're stopping */ | |
1105 | return 0; | |
1106 | } | |
1107 | ||
eb086ec5 | 1108 | void |
896831f5 EC |
1109 | aoe_ktstop(struct ktstate *k) |
1110 | { | |
1111 | kthread_stop(k->task); | |
1112 | wait_for_completion(&k->rendez); | |
1113 | } | |
1114 | ||
eb086ec5 | 1115 | int |
896831f5 EC |
1116 | aoe_ktstart(struct ktstate *k) |
1117 | { | |
1118 | struct task_struct *task; | |
1119 | ||
1120 | init_completion(&k->rendez); | |
1121 | task = kthread_run(kthread, k, k->name); | |
1122 | if (task == NULL || IS_ERR(task)) | |
1123 | return -ENOMEM; | |
1124 | k->task = task; | |
1125 | wait_for_completion(&k->rendez); /* allow kthread to start */ | |
1126 | init_completion(&k->rendez); /* for waiting for exit later */ | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | /* pass it off to kthreads for processing */ | |
1131 | static void | |
1132 | ktcomplete(struct frame *f, struct sk_buff *skb) | |
1133 | { | |
1134 | ulong flags; | |
1135 | ||
1136 | f->r_skb = skb; | |
1137 | spin_lock_irqsave(&iocq.lock, flags); | |
1138 | list_add_tail(&f->head, &iocq.head); | |
1139 | spin_unlock_irqrestore(&iocq.lock, flags); | |
1140 | wake_up(&ktiowq); | |
1141 | } | |
1142 | ||
1143 | struct sk_buff * | |
1da177e4 LT |
1144 | aoecmd_ata_rsp(struct sk_buff *skb) |
1145 | { | |
1146 | struct aoedev *d; | |
896831f5 | 1147 | struct aoe_hdr *h; |
1da177e4 | 1148 | struct frame *f; |
68e0d42f | 1149 | struct aoetgt *t; |
896831f5 | 1150 | u32 n; |
1da177e4 LT |
1151 | ulong flags; |
1152 | char ebuf[128]; | |
32465c65 | 1153 | u16 aoemajor; |
1154 | ||
896831f5 EC |
1155 | h = (struct aoe_hdr *) skb->data; |
1156 | aoemajor = be16_to_cpu(get_unaligned(&h->major)); | |
1157 | d = aoedev_by_aoeaddr(aoemajor, h->minor); | |
1da177e4 LT |
1158 | if (d == NULL) { |
1159 | snprintf(ebuf, sizeof ebuf, "aoecmd_ata_rsp: ata response " | |
1160 | "for unknown device %d.%d\n", | |
896831f5 | 1161 | aoemajor, h->minor); |
1da177e4 | 1162 | aoechr_error(ebuf); |
896831f5 | 1163 | return skb; |
1da177e4 LT |
1164 | } |
1165 | ||
1166 | spin_lock_irqsave(&d->lock, flags); | |
1167 | ||
896831f5 | 1168 | n = be32_to_cpu(get_unaligned(&h->tag)); |
64a80f5a | 1169 | f = getframe(d, n); |
1da177e4 | 1170 | if (f == NULL) { |
dced3a05 | 1171 | calc_rttavg(d, -tsince(n)); |
1da177e4 | 1172 | spin_unlock_irqrestore(&d->lock, flags); |
69cf2d85 | 1173 | aoedev_put(d); |
1da177e4 LT |
1174 | snprintf(ebuf, sizeof ebuf, |
1175 | "%15s e%d.%d tag=%08x@%08lx\n", | |
1176 | "unexpected rsp", | |
896831f5 EC |
1177 | get_unaligned_be16(&h->major), |
1178 | h->minor, | |
1179 | get_unaligned_be32(&h->tag), | |
1da177e4 LT |
1180 | jiffies); |
1181 | aoechr_error(ebuf); | |
896831f5 | 1182 | return skb; |
1da177e4 | 1183 | } |
64a80f5a | 1184 | t = f->t; |
1da177e4 | 1185 | calc_rttavg(d, tsince(f->tag)); |
68e0d42f | 1186 | t->nout--; |
1da177e4 | 1187 | aoecmd_work(d); |
1da177e4 LT |
1188 | |
1189 | spin_unlock_irqrestore(&d->lock, flags); | |
896831f5 EC |
1190 | |
1191 | ktcomplete(f, skb); | |
1192 | ||
1193 | /* | |
1194 | * Note here that we do not perform an aoedev_put, as we are | |
1195 | * leaving this reference for the ktio to release. | |
1196 | */ | |
1197 | return NULL; | |
1da177e4 LT |
1198 | } |
1199 | ||
1200 | void | |
1201 | aoecmd_cfg(ushort aoemajor, unsigned char aoeminor) | |
1202 | { | |
e9bb8fb0 | 1203 | struct sk_buff_head queue; |
1da177e4 | 1204 | |
e9bb8fb0 DM |
1205 | __skb_queue_head_init(&queue); |
1206 | aoecmd_cfg_pkts(aoemajor, aoeminor, &queue); | |
1207 | aoenet_xmit(&queue); | |
1da177e4 LT |
1208 | } |
1209 | ||
68e0d42f | 1210 | struct sk_buff * |
1da177e4 LT |
1211 | aoecmd_ata_id(struct aoedev *d) |
1212 | { | |
1213 | struct aoe_hdr *h; | |
1214 | struct aoe_atahdr *ah; | |
1215 | struct frame *f; | |
1216 | struct sk_buff *skb; | |
68e0d42f | 1217 | struct aoetgt *t; |
1da177e4 | 1218 | |
896831f5 | 1219 | f = newframe(d); |
68e0d42f | 1220 | if (f == NULL) |
1da177e4 | 1221 | return NULL; |
68e0d42f EC |
1222 | |
1223 | t = *d->tgt; | |
1da177e4 LT |
1224 | |
1225 | /* initialize the headers & frame */ | |
e407a7f6 | 1226 | skb = f->skb; |
abdbf94d | 1227 | h = (struct aoe_hdr *) skb_mac_header(skb); |
1da177e4 | 1228 | ah = (struct aoe_atahdr *) (h+1); |
19900cde EC |
1229 | skb_put(skb, sizeof *h + sizeof *ah); |
1230 | memset(h, 0, skb->len); | |
68e0d42f | 1231 | f->tag = aoehdr_atainit(d, t, h); |
896831f5 | 1232 | fhash(f); |
68e0d42f | 1233 | t->nout++; |
1da177e4 | 1234 | f->waited = 0; |
1da177e4 | 1235 | |
1da177e4 LT |
1236 | /* set up ata header */ |
1237 | ah->scnt = 1; | |
04b3ab52 | 1238 | ah->cmdstat = ATA_CMD_ID_ATA; |
1da177e4 LT |
1239 | ah->lba3 = 0xa0; |
1240 | ||
68e0d42f | 1241 | skb->dev = t->ifp->nd; |
1da177e4 | 1242 | |
3ae1c24e | 1243 | d->rttavg = MAXTIMER; |
1da177e4 | 1244 | d->timer.function = rexmit_timer; |
1da177e4 | 1245 | |
4f51dc5e | 1246 | return skb_clone(skb, GFP_ATOMIC); |
1da177e4 LT |
1247 | } |
1248 | ||
68e0d42f EC |
1249 | static struct aoetgt * |
1250 | addtgt(struct aoedev *d, char *addr, ulong nframes) | |
1251 | { | |
1252 | struct aoetgt *t, **tt, **te; | |
68e0d42f EC |
1253 | |
1254 | tt = d->targets; | |
1255 | te = tt + NTARGETS; | |
1256 | for (; tt < te && *tt; tt++) | |
1257 | ; | |
1258 | ||
578c4aa0 EC |
1259 | if (tt == te) { |
1260 | printk(KERN_INFO | |
1261 | "aoe: device addtgt failure; too many targets\n"); | |
68e0d42f | 1262 | return NULL; |
578c4aa0 | 1263 | } |
896831f5 EC |
1264 | t = kzalloc(sizeof(*t), GFP_ATOMIC); |
1265 | if (!t) { | |
578c4aa0 | 1266 | printk(KERN_INFO "aoe: cannot allocate memory to add target\n"); |
9bb237b6 EC |
1267 | return NULL; |
1268 | } | |
1269 | ||
896831f5 | 1270 | d->ntargets++; |
68e0d42f | 1271 | t->nframes = nframes; |
896831f5 | 1272 | t->d = d; |
68e0d42f EC |
1273 | memcpy(t->addr, addr, sizeof t->addr); |
1274 | t->ifp = t->ifs; | |
1275 | t->maxout = t->nframes; | |
896831f5 | 1276 | INIT_LIST_HEAD(&t->ffree); |
68e0d42f | 1277 | return *tt = t; |
68e0d42f EC |
1278 | } |
1279 | ||
3f0f0133 EC |
1280 | static void |
1281 | setdbcnt(struct aoedev *d) | |
1282 | { | |
1283 | struct aoetgt **t, **e; | |
1284 | int bcnt = 0; | |
1285 | ||
1286 | t = d->targets; | |
1287 | e = t + NTARGETS; | |
1288 | for (; t < e && *t; t++) | |
1289 | if (bcnt == 0 || bcnt > (*t)->minbcnt) | |
1290 | bcnt = (*t)->minbcnt; | |
1291 | if (bcnt != d->maxbcnt) { | |
1292 | d->maxbcnt = bcnt; | |
1293 | pr_info("aoe: e%ld.%d: setting %d byte data frames\n", | |
1294 | d->aoemajor, d->aoeminor, bcnt); | |
1295 | } | |
1296 | } | |
1297 | ||
1298 | static void | |
1299 | setifbcnt(struct aoetgt *t, struct net_device *nd, int bcnt) | |
1300 | { | |
1301 | struct aoedev *d; | |
1302 | struct aoeif *p, *e; | |
1303 | int minbcnt; | |
1304 | ||
1305 | d = t->d; | |
1306 | minbcnt = bcnt; | |
1307 | p = t->ifs; | |
1308 | e = p + NAOEIFS; | |
1309 | for (; p < e; p++) { | |
1310 | if (p->nd == NULL) | |
1311 | break; /* end of the valid interfaces */ | |
1312 | if (p->nd == nd) { | |
1313 | p->bcnt = bcnt; /* we're updating */ | |
1314 | nd = NULL; | |
1315 | } else if (minbcnt > p->bcnt) | |
1316 | minbcnt = p->bcnt; /* find the min interface */ | |
1317 | } | |
1318 | if (nd) { | |
1319 | if (p == e) { | |
1320 | pr_err("aoe: device setifbcnt failure; too many interfaces.\n"); | |
1321 | return; | |
1322 | } | |
1b86fda9 | 1323 | dev_hold(nd); |
3f0f0133 EC |
1324 | p->nd = nd; |
1325 | p->bcnt = bcnt; | |
1326 | } | |
1327 | t->minbcnt = minbcnt; | |
1328 | setdbcnt(d); | |
1329 | } | |
1330 | ||
1da177e4 LT |
1331 | void |
1332 | aoecmd_cfg_rsp(struct sk_buff *skb) | |
1333 | { | |
1334 | struct aoedev *d; | |
1335 | struct aoe_hdr *h; | |
1336 | struct aoe_cfghdr *ch; | |
68e0d42f | 1337 | struct aoetgt *t; |
63e9cc5d | 1338 | ulong flags, sysminor, aoemajor; |
1da177e4 | 1339 | struct sk_buff *sl; |
69cf2d85 | 1340 | struct sk_buff_head queue; |
19bf2635 | 1341 | u16 n; |
1da177e4 | 1342 | |
69cf2d85 | 1343 | sl = NULL; |
abdbf94d | 1344 | h = (struct aoe_hdr *) skb_mac_header(skb); |
1da177e4 LT |
1345 | ch = (struct aoe_cfghdr *) (h+1); |
1346 | ||
1347 | /* | |
1348 | * Enough people have their dip switches set backwards to | |
1349 | * warrant a loud message for this special case. | |
1350 | */ | |
823ed72e | 1351 | aoemajor = get_unaligned_be16(&h->major); |
1da177e4 | 1352 | if (aoemajor == 0xfff) { |
a12c93f0 | 1353 | printk(KERN_ERR "aoe: Warning: shelf address is all ones. " |
6bb6285f | 1354 | "Check shelf dip switches.\n"); |
1da177e4 LT |
1355 | return; |
1356 | } | |
6583303c EC |
1357 | if (h->minor >= NPERSHELF) { |
1358 | pr_err("aoe: e%ld.%d %s, %d\n", | |
1359 | aoemajor, h->minor, | |
1360 | "slot number larger than the maximum", | |
1361 | NPERSHELF-1); | |
1362 | return; | |
1363 | } | |
1da177e4 LT |
1364 | |
1365 | sysminor = SYSMINOR(aoemajor, h->minor); | |
fc458dcd | 1366 | if (sysminor * AOE_PARTITIONS + AOE_PARTITIONS > MINORMASK) { |
a12c93f0 | 1367 | printk(KERN_INFO "aoe: e%ld.%d: minor number too large\n", |
fc458dcd | 1368 | aoemajor, (int) h->minor); |
1da177e4 LT |
1369 | return; |
1370 | } | |
1371 | ||
19bf2635 | 1372 | n = be16_to_cpu(ch->bufcnt); |
7df620d8 EC |
1373 | if (n > aoe_maxout) /* keep it reasonable */ |
1374 | n = aoe_maxout; | |
1da177e4 | 1375 | |
68e0d42f | 1376 | d = aoedev_by_sysminor_m(sysminor); |
1da177e4 | 1377 | if (d == NULL) { |
a12c93f0 | 1378 | printk(KERN_INFO "aoe: device sysminor_m failure\n"); |
1da177e4 LT |
1379 | return; |
1380 | } | |
1381 | ||
1382 | spin_lock_irqsave(&d->lock, flags); | |
1383 | ||
68e0d42f EC |
1384 | t = gettgt(d, h->src); |
1385 | if (!t) { | |
1386 | t = addtgt(d, h->src, n); | |
69cf2d85 EC |
1387 | if (!t) |
1388 | goto bail; | |
68e0d42f | 1389 | } |
3f0f0133 EC |
1390 | n = skb->dev->mtu; |
1391 | n -= sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr); | |
1392 | n /= 512; | |
1393 | if (n > ch->scnt) | |
1394 | n = ch->scnt; | |
1395 | n = n ? n * 512 : DEFAULTBCNT; | |
1396 | setifbcnt(t, skb->dev, n); | |
3ae1c24e EC |
1397 | |
1398 | /* don't change users' perspective */ | |
69cf2d85 EC |
1399 | if (d->nopen == 0) { |
1400 | d->fw_ver = be16_to_cpu(ch->fwver); | |
1401 | sl = aoecmd_ata_id(d); | |
1da177e4 | 1402 | } |
69cf2d85 | 1403 | bail: |
1da177e4 | 1404 | spin_unlock_irqrestore(&d->lock, flags); |
69cf2d85 | 1405 | aoedev_put(d); |
e9bb8fb0 | 1406 | if (sl) { |
e9bb8fb0 DM |
1407 | __skb_queue_head_init(&queue); |
1408 | __skb_queue_tail(&queue, sl); | |
1409 | aoenet_xmit(&queue); | |
1410 | } | |
1da177e4 LT |
1411 | } |
1412 | ||
68e0d42f EC |
1413 | void |
1414 | aoecmd_cleanslate(struct aoedev *d) | |
1415 | { | |
1416 | struct aoetgt **t, **te; | |
68e0d42f EC |
1417 | |
1418 | d->mintimer = MINTIMER; | |
3f0f0133 | 1419 | d->maxbcnt = 0; |
68e0d42f EC |
1420 | |
1421 | t = d->targets; | |
1422 | te = t + NTARGETS; | |
3f0f0133 | 1423 | for (; t < te && *t; t++) |
68e0d42f | 1424 | (*t)->maxout = (*t)->nframes; |
68e0d42f | 1425 | } |
896831f5 | 1426 | |
69cf2d85 EC |
1427 | void |
1428 | aoe_failbuf(struct aoedev *d, struct buf *buf) | |
1429 | { | |
1430 | if (buf == NULL) | |
1431 | return; | |
1432 | buf->resid = 0; | |
1433 | clear_bit(BIO_UPTODATE, &buf->bio->bi_flags); | |
1434 | if (buf->nframesout == 0) | |
1435 | aoe_end_buf(d, buf); | |
1436 | } | |
1437 | ||
1438 | void | |
1439 | aoe_flush_iocq(void) | |
896831f5 EC |
1440 | { |
1441 | struct frame *f; | |
1442 | struct aoedev *d; | |
1443 | LIST_HEAD(flist); | |
1444 | struct list_head *pos; | |
1445 | struct sk_buff *skb; | |
1446 | ulong flags; | |
1447 | ||
1448 | spin_lock_irqsave(&iocq.lock, flags); | |
1449 | list_splice_init(&iocq.head, &flist); | |
1450 | spin_unlock_irqrestore(&iocq.lock, flags); | |
1451 | while (!list_empty(&flist)) { | |
1452 | pos = flist.next; | |
1453 | list_del(pos); | |
1454 | f = list_entry(pos, struct frame, head); | |
1455 | d = f->t->d; | |
1456 | skb = f->r_skb; | |
1457 | spin_lock_irqsave(&d->lock, flags); | |
1458 | if (f->buf) { | |
1459 | f->buf->nframesout--; | |
1460 | aoe_failbuf(d, f->buf); | |
1461 | } | |
1462 | aoe_freetframe(f); | |
1463 | spin_unlock_irqrestore(&d->lock, flags); | |
1464 | dev_kfree_skb(skb); | |
69cf2d85 | 1465 | aoedev_put(d); |
896831f5 EC |
1466 | } |
1467 | } | |
1468 | ||
1469 | int __init | |
1470 | aoecmd_init(void) | |
1471 | { | |
1472 | INIT_LIST_HEAD(&iocq.head); | |
1473 | spin_lock_init(&iocq.lock); | |
1474 | init_waitqueue_head(&ktiowq); | |
1475 | kts.name = "aoe_ktio"; | |
1476 | kts.fn = ktio; | |
1477 | kts.waitq = &ktiowq; | |
1478 | kts.lock = &iocq.lock; | |
1479 | return aoe_ktstart(&kts); | |
1480 | } | |
1481 | ||
1482 | void | |
1483 | aoecmd_exit(void) | |
1484 | { | |
1485 | aoe_ktstop(&kts); | |
69cf2d85 | 1486 | aoe_flush_iocq(); |
896831f5 | 1487 | } |