1 /*
2  *  Device operations for the pnfs nfs4 file layout driver.
3  *
4  *  Copyright (c) 2002
5  *  The Regents of the University of Michigan
6  *  All Rights Reserved
7  *
8  *  Dean Hildebrand <dhildebz@umich.edu>
9  *  Garth Goodson   <Garth.Goodson@netapp.com>
10  *
11  *  Permission is granted to use, copy, create derivative works, and
12  *  redistribute this software and such derivative works for any purpose,
13  *  so long as the name of the University of Michigan is not used in
14  *  any advertising or publicity pertaining to the use or distribution
15  *  of this software without specific, written prior authorization. If
16  *  the above copyright notice or any other identification of the
17  *  University of Michigan is included in any copy of any portion of
18  *  this software, then the disclaimer below must also be included.
19  *
20  *  This software is provided as is, without representation or warranty
21  *  of any kind either express or implied, including without limitation
22  *  the implied warranties of merchantability, fitness for a particular
23  *  purpose, or noninfringement.  The Regents of the University of
24  *  Michigan shall not be liable for any damages, including special,
25  *  indirect, incidental, or consequential damages, with respect to any
26  *  claim arising out of or in connection with the use of the software,
27  *  even if it has been or is hereafter advised of the possibility of
28  *  such damages.
29  */
30 
31 #include <linux/nfs_fs.h>
32 #include <linux/vmalloc.h>
33 #include <linux/module.h>
34 #include <linux/sunrpc/addr.h>
35 
36 #include "../internal.h"
37 #include "../nfs4session.h"
38 #include "filelayout.h"
39 
40 #define NFSDBG_FACILITY		NFSDBG_PNFS_LD
41 
42 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
43 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
44 
45 /*
46  * Data server cache
47  *
48  * Data servers can be mapped to different device ids.
49  * nfs4_pnfs_ds reference counting
50  *   - set to 1 on allocation
51  *   - incremented when a device id maps a data server already in the cache.
52  *   - decremented when deviceid is removed from the cache.
53  */
54 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
55 static LIST_HEAD(nfs4_data_server_cache);
56 
57 /* Debug routines */
58 void
59 print_ds(struct nfs4_pnfs_ds *ds)
60 {
61 	if (ds == NULL) {
62 		printk("%s NULL device\n", __func__);
63 		return;
64 	}
65 	printk("        ds %s\n"
66 		"        ref count %d\n"
67 		"        client %p\n"
68 		"        cl_exchange_flags %x\n",
69 		ds->ds_remotestr,
70 		atomic_read(&ds->ds_count), ds->ds_clp,
71 		ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
72 }
73 
74 static bool
75 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
76 {
77 	struct sockaddr_in *a, *b;
78 	struct sockaddr_in6 *a6, *b6;
79 
80 	if (addr1->sa_family != addr2->sa_family)
81 		return false;
82 
83 	switch (addr1->sa_family) {
84 	case AF_INET:
85 		a = (struct sockaddr_in *)addr1;
86 		b = (struct sockaddr_in *)addr2;
87 
88 		if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
89 		    a->sin_port == b->sin_port)
90 			return true;
91 		break;
92 
93 	case AF_INET6:
94 		a6 = (struct sockaddr_in6 *)addr1;
95 		b6 = (struct sockaddr_in6 *)addr2;
96 
97 		/* LINKLOCAL addresses must have matching scope_id */
98 		if (ipv6_addr_src_scope(&a6->sin6_addr) ==
99 		    IPV6_ADDR_SCOPE_LINKLOCAL &&
100 		    a6->sin6_scope_id != b6->sin6_scope_id)
101 			return false;
102 
103 		if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
104 		    a6->sin6_port == b6->sin6_port)
105 			return true;
106 		break;
107 
108 	default:
109 		dprintk("%s: unhandled address family: %u\n",
110 			__func__, addr1->sa_family);
111 		return false;
112 	}
113 
114 	return false;
115 }
116 
117 static bool
118 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
119 			       const struct list_head *dsaddrs2)
120 {
121 	struct nfs4_pnfs_ds_addr *da1, *da2;
122 
123 	/* step through both lists, comparing as we go */
124 	for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
125 	     da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
126 	     da1 != NULL && da2 != NULL;
127 	     da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
128 	     da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
129 		if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
130 				   (struct sockaddr *)&da2->da_addr))
131 			return false;
132 	}
133 	if (da1 == NULL && da2 == NULL)
134 		return true;
135 
136 	return false;
137 }
138 
139 /*
140  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
141  */
142 static struct nfs4_pnfs_ds *
143 _data_server_lookup_locked(const struct list_head *dsaddrs)
144 {
145 	struct nfs4_pnfs_ds *ds;
146 
147 	list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
148 		if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
149 			return ds;
150 	return NULL;
151 }
152 
153 /*
154  * Create an rpc connection to the nfs4_pnfs_ds data server
155  * Currently only supports IPv4 and IPv6 addresses
156  */
157 static int
158 nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
159 {
160 	struct nfs_client *clp = ERR_PTR(-EIO);
161 	struct nfs4_pnfs_ds_addr *da;
162 	int status = 0;
163 
164 	dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
165 		mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor);
166 
167 	list_for_each_entry(da, &ds->ds_addrs, da_node) {
168 		dprintk("%s: DS %s: trying address %s\n",
169 			__func__, ds->ds_remotestr, da->da_remotestr);
170 
171 		clp = nfs4_set_ds_client(mds_srv->nfs_client,
172 					(struct sockaddr *)&da->da_addr,
173 					da->da_addrlen, IPPROTO_TCP,
174 					dataserver_timeo, dataserver_retrans);
175 		if (!IS_ERR(clp))
176 			break;
177 	}
178 
179 	if (IS_ERR(clp)) {
180 		status = PTR_ERR(clp);
181 		goto out;
182 	}
183 
184 	status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time);
185 	if (status)
186 		goto out_put;
187 
188 	smp_wmb();
189 	ds->ds_clp = clp;
190 	dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
191 out:
192 	return status;
193 out_put:
194 	nfs_put_client(clp);
195 	goto out;
196 }
197 
198 static void
199 destroy_ds(struct nfs4_pnfs_ds *ds)
200 {
201 	struct nfs4_pnfs_ds_addr *da;
202 
203 	dprintk("--> %s\n", __func__);
204 	ifdebug(FACILITY)
205 		print_ds(ds);
206 
207 	if (ds->ds_clp)
208 		nfs_put_client(ds->ds_clp);
209 
210 	while (!list_empty(&ds->ds_addrs)) {
211 		da = list_first_entry(&ds->ds_addrs,
212 				      struct nfs4_pnfs_ds_addr,
213 				      da_node);
214 		list_del_init(&da->da_node);
215 		kfree(da->da_remotestr);
216 		kfree(da);
217 	}
218 
219 	kfree(ds->ds_remotestr);
220 	kfree(ds);
221 }
222 
223 void
224 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
225 {
226 	struct nfs4_pnfs_ds *ds;
227 	int i;
228 
229 	nfs4_print_deviceid(&dsaddr->id_node.deviceid);
230 
231 	for (i = 0; i < dsaddr->ds_num; i++) {
232 		ds = dsaddr->ds_list[i];
233 		if (ds != NULL) {
234 			if (atomic_dec_and_lock(&ds->ds_count,
235 						&nfs4_ds_cache_lock)) {
236 				list_del_init(&ds->ds_node);
237 				spin_unlock(&nfs4_ds_cache_lock);
238 				destroy_ds(ds);
239 			}
240 		}
241 	}
242 	kfree(dsaddr->stripe_indices);
243 	kfree(dsaddr);
244 }
245 
246 /*
247  * Create a string with a human readable address and port to avoid
248  * complicated setup around many dprinks.
249  */
250 static char *
251 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
252 {
253 	struct nfs4_pnfs_ds_addr *da;
254 	char *remotestr;
255 	size_t len;
256 	char *p;
257 
258 	len = 3;        /* '{', '}' and eol */
259 	list_for_each_entry(da, dsaddrs, da_node) {
260 		len += strlen(da->da_remotestr) + 1;    /* string plus comma */
261 	}
262 
263 	remotestr = kzalloc(len, gfp_flags);
264 	if (!remotestr)
265 		return NULL;
266 
267 	p = remotestr;
268 	*(p++) = '{';
269 	len--;
270 	list_for_each_entry(da, dsaddrs, da_node) {
271 		size_t ll = strlen(da->da_remotestr);
272 
273 		if (ll > len)
274 			goto out_err;
275 
276 		memcpy(p, da->da_remotestr, ll);
277 		p += ll;
278 		len -= ll;
279 
280 		if (len < 1)
281 			goto out_err;
282 		(*p++) = ',';
283 		len--;
284 	}
285 	if (len < 2)
286 		goto out_err;
287 	*(p++) = '}';
288 	*p = '\0';
289 	return remotestr;
290 out_err:
291 	kfree(remotestr);
292 	return NULL;
293 }
294 
295 static struct nfs4_pnfs_ds *
296 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
297 {
298 	struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
299 	char *remotestr;
300 
301 	if (list_empty(dsaddrs)) {
302 		dprintk("%s: no addresses defined\n", __func__);
303 		goto out;
304 	}
305 
306 	ds = kzalloc(sizeof(*ds), gfp_flags);
307 	if (!ds)
308 		goto out;
309 
310 	/* this is only used for debugging, so it's ok if its NULL */
311 	remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
312 
313 	spin_lock(&nfs4_ds_cache_lock);
314 	tmp_ds = _data_server_lookup_locked(dsaddrs);
315 	if (tmp_ds == NULL) {
316 		INIT_LIST_HEAD(&ds->ds_addrs);
317 		list_splice_init(dsaddrs, &ds->ds_addrs);
318 		ds->ds_remotestr = remotestr;
319 		atomic_set(&ds->ds_count, 1);
320 		INIT_LIST_HEAD(&ds->ds_node);
321 		ds->ds_clp = NULL;
322 		list_add(&ds->ds_node, &nfs4_data_server_cache);
323 		dprintk("%s add new data server %s\n", __func__,
324 			ds->ds_remotestr);
325 	} else {
326 		kfree(remotestr);
327 		kfree(ds);
328 		atomic_inc(&tmp_ds->ds_count);
329 		dprintk("%s data server %s found, inc'ed ds_count to %d\n",
330 			__func__, tmp_ds->ds_remotestr,
331 			atomic_read(&tmp_ds->ds_count));
332 		ds = tmp_ds;
333 	}
334 	spin_unlock(&nfs4_ds_cache_lock);
335 out:
336 	return ds;
337 }
338 
339 /*
340  * Currently only supports ipv4, ipv6 and one multi-path address.
341  */
342 static struct nfs4_pnfs_ds_addr *
343 decode_ds_addr(struct net *net, struct xdr_stream *streamp, gfp_t gfp_flags)
344 {
345 	struct nfs4_pnfs_ds_addr *da = NULL;
346 	char *buf, *portstr;
347 	__be16 port;
348 	int nlen, rlen;
349 	int tmp[2];
350 	__be32 *p;
351 	char *netid, *match_netid;
352 	size_t len, match_netid_len;
353 	char *startsep = "";
354 	char *endsep = "";
355 
356 
357 	/* r_netid */
358 	p = xdr_inline_decode(streamp, 4);
359 	if (unlikely(!p))
360 		goto out_err;
361 	nlen = be32_to_cpup(p++);
362 
363 	p = xdr_inline_decode(streamp, nlen);
364 	if (unlikely(!p))
365 		goto out_err;
366 
367 	netid = kmalloc(nlen+1, gfp_flags);
368 	if (unlikely(!netid))
369 		goto out_err;
370 
371 	netid[nlen] = '\0';
372 	memcpy(netid, p, nlen);
373 
374 	/* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
375 	p = xdr_inline_decode(streamp, 4);
376 	if (unlikely(!p))
377 		goto out_free_netid;
378 	rlen = be32_to_cpup(p);
379 
380 	p = xdr_inline_decode(streamp, rlen);
381 	if (unlikely(!p))
382 		goto out_free_netid;
383 
384 	/* port is ".ABC.DEF", 8 chars max */
385 	if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
386 		dprintk("%s: Invalid address, length %d\n", __func__,
387 			rlen);
388 		goto out_free_netid;
389 	}
390 	buf = kmalloc(rlen + 1, gfp_flags);
391 	if (!buf) {
392 		dprintk("%s: Not enough memory\n", __func__);
393 		goto out_free_netid;
394 	}
395 	buf[rlen] = '\0';
396 	memcpy(buf, p, rlen);
397 
398 	/* replace port '.' with '-' */
399 	portstr = strrchr(buf, '.');
400 	if (!portstr) {
401 		dprintk("%s: Failed finding expected dot in port\n",
402 			__func__);
403 		goto out_free_buf;
404 	}
405 	*portstr = '-';
406 
407 	/* find '.' between address and port */
408 	portstr = strrchr(buf, '.');
409 	if (!portstr) {
410 		dprintk("%s: Failed finding expected dot between address and "
411 			"port\n", __func__);
412 		goto out_free_buf;
413 	}
414 	*portstr = '\0';
415 
416 	da = kzalloc(sizeof(*da), gfp_flags);
417 	if (unlikely(!da))
418 		goto out_free_buf;
419 
420 	INIT_LIST_HEAD(&da->da_node);
421 
422 	if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
423 		      sizeof(da->da_addr))) {
424 		dprintk("%s: error parsing address %s\n", __func__, buf);
425 		goto out_free_da;
426 	}
427 
428 	portstr++;
429 	sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
430 	port = htons((tmp[0] << 8) | (tmp[1]));
431 
432 	switch (da->da_addr.ss_family) {
433 	case AF_INET:
434 		((struct sockaddr_in *)&da->da_addr)->sin_port = port;
435 		da->da_addrlen = sizeof(struct sockaddr_in);
436 		match_netid = "tcp";
437 		match_netid_len = 3;
438 		break;
439 
440 	case AF_INET6:
441 		((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
442 		da->da_addrlen = sizeof(struct sockaddr_in6);
443 		match_netid = "tcp6";
444 		match_netid_len = 4;
445 		startsep = "[";
446 		endsep = "]";
447 		break;
448 
449 	default:
450 		dprintk("%s: unsupported address family: %u\n",
451 			__func__, da->da_addr.ss_family);
452 		goto out_free_da;
453 	}
454 
455 	if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
456 		dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
457 			__func__, netid, match_netid);
458 		goto out_free_da;
459 	}
460 
461 	/* save human readable address */
462 	len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
463 	da->da_remotestr = kzalloc(len, gfp_flags);
464 
465 	/* NULL is ok, only used for dprintk */
466 	if (da->da_remotestr)
467 		snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
468 			 buf, endsep, ntohs(port));
469 
470 	dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
471 	kfree(buf);
472 	kfree(netid);
473 	return da;
474 
475 out_free_da:
476 	kfree(da);
477 out_free_buf:
478 	dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
479 	kfree(buf);
480 out_free_netid:
481 	kfree(netid);
482 out_err:
483 	return NULL;
484 }
485 
486 /* Decode opaque device data and return the result */
487 struct nfs4_file_layout_dsaddr *
488 nfs4_fl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
489 		gfp_t gfp_flags)
490 {
491 	int i;
492 	u32 cnt, num;
493 	u8 *indexp;
494 	__be32 *p;
495 	u8 *stripe_indices;
496 	u8 max_stripe_index;
497 	struct nfs4_file_layout_dsaddr *dsaddr = NULL;
498 	struct xdr_stream stream;
499 	struct xdr_buf buf;
500 	struct page *scratch;
501 	struct list_head dsaddrs;
502 	struct nfs4_pnfs_ds_addr *da;
503 
504 	/* set up xdr stream */
505 	scratch = alloc_page(gfp_flags);
506 	if (!scratch)
507 		goto out_err;
508 
509 	xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
510 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
511 
512 	/* Get the stripe count (number of stripe index) */
513 	p = xdr_inline_decode(&stream, 4);
514 	if (unlikely(!p))
515 		goto out_err_free_scratch;
516 
517 	cnt = be32_to_cpup(p);
518 	dprintk("%s stripe count  %d\n", __func__, cnt);
519 	if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
520 		printk(KERN_WARNING "NFS: %s: stripe count %d greater than "
521 		       "supported maximum %d\n", __func__,
522 			cnt, NFS4_PNFS_MAX_STRIPE_CNT);
523 		goto out_err_free_scratch;
524 	}
525 
526 	/* read stripe indices */
527 	stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
528 	if (!stripe_indices)
529 		goto out_err_free_scratch;
530 
531 	p = xdr_inline_decode(&stream, cnt << 2);
532 	if (unlikely(!p))
533 		goto out_err_free_stripe_indices;
534 
535 	indexp = &stripe_indices[0];
536 	max_stripe_index = 0;
537 	for (i = 0; i < cnt; i++) {
538 		*indexp = be32_to_cpup(p++);
539 		max_stripe_index = max(max_stripe_index, *indexp);
540 		indexp++;
541 	}
542 
543 	/* Check the multipath list count */
544 	p = xdr_inline_decode(&stream, 4);
545 	if (unlikely(!p))
546 		goto out_err_free_stripe_indices;
547 
548 	num = be32_to_cpup(p);
549 	dprintk("%s ds_num %u\n", __func__, num);
550 	if (num > NFS4_PNFS_MAX_MULTI_CNT) {
551 		printk(KERN_WARNING "NFS: %s: multipath count %d greater than "
552 			"supported maximum %d\n", __func__,
553 			num, NFS4_PNFS_MAX_MULTI_CNT);
554 		goto out_err_free_stripe_indices;
555 	}
556 
557 	/* validate stripe indices are all < num */
558 	if (max_stripe_index >= num) {
559 		printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n",
560 			__func__, max_stripe_index, num);
561 		goto out_err_free_stripe_indices;
562 	}
563 
564 	dsaddr = kzalloc(sizeof(*dsaddr) +
565 			(sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
566 			gfp_flags);
567 	if (!dsaddr)
568 		goto out_err_free_stripe_indices;
569 
570 	dsaddr->stripe_count = cnt;
571 	dsaddr->stripe_indices = stripe_indices;
572 	stripe_indices = NULL;
573 	dsaddr->ds_num = num;
574 	nfs4_init_deviceid_node(&dsaddr->id_node, server, &pdev->dev_id);
575 
576 	INIT_LIST_HEAD(&dsaddrs);
577 
578 	for (i = 0; i < dsaddr->ds_num; i++) {
579 		int j;
580 		u32 mp_count;
581 
582 		p = xdr_inline_decode(&stream, 4);
583 		if (unlikely(!p))
584 			goto out_err_free_deviceid;
585 
586 		mp_count = be32_to_cpup(p); /* multipath count */
587 		for (j = 0; j < mp_count; j++) {
588 			da = decode_ds_addr(server->nfs_client->cl_net,
589 					    &stream, gfp_flags);
590 			if (da)
591 				list_add_tail(&da->da_node, &dsaddrs);
592 		}
593 		if (list_empty(&dsaddrs)) {
594 			dprintk("%s: no suitable DS addresses found\n",
595 				__func__);
596 			goto out_err_free_deviceid;
597 		}
598 
599 		dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
600 		if (!dsaddr->ds_list[i])
601 			goto out_err_drain_dsaddrs;
602 
603 		/* If DS was already in cache, free ds addrs */
604 		while (!list_empty(&dsaddrs)) {
605 			da = list_first_entry(&dsaddrs,
606 					      struct nfs4_pnfs_ds_addr,
607 					      da_node);
608 			list_del_init(&da->da_node);
609 			kfree(da->da_remotestr);
610 			kfree(da);
611 		}
612 	}
613 
614 	__free_page(scratch);
615 	return dsaddr;
616 
617 out_err_drain_dsaddrs:
618 	while (!list_empty(&dsaddrs)) {
619 		da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
620 				      da_node);
621 		list_del_init(&da->da_node);
622 		kfree(da->da_remotestr);
623 		kfree(da);
624 	}
625 out_err_free_deviceid:
626 	nfs4_fl_free_deviceid(dsaddr);
627 	/* stripe_indicies was part of dsaddr */
628 	goto out_err_free_scratch;
629 out_err_free_stripe_indices:
630 	kfree(stripe_indices);
631 out_err_free_scratch:
632 	__free_page(scratch);
633 out_err:
634 	dprintk("%s ERROR: returning NULL\n", __func__);
635 	return NULL;
636 }
637 
638 void
639 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
640 {
641 	nfs4_put_deviceid_node(&dsaddr->id_node);
642 }
643 
644 /*
645  * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
646  * Then: ((res + fsi) % dsaddr->stripe_count)
647  */
648 u32
649 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
650 {
651 	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
652 	u64 tmp;
653 
654 	tmp = offset - flseg->pattern_offset;
655 	do_div(tmp, flseg->stripe_unit);
656 	tmp += flseg->first_stripe_index;
657 	return do_div(tmp, flseg->dsaddr->stripe_count);
658 }
659 
660 u32
661 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
662 {
663 	return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
664 }
665 
666 struct nfs_fh *
667 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
668 {
669 	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
670 	u32 i;
671 
672 	if (flseg->stripe_type == STRIPE_SPARSE) {
673 		if (flseg->num_fh == 1)
674 			i = 0;
675 		else if (flseg->num_fh == 0)
676 			/* Use the MDS OPEN fh set in nfs_read_rpcsetup */
677 			return NULL;
678 		else
679 			i = nfs4_fl_calc_ds_index(lseg, j);
680 	} else
681 		i = j;
682 	return flseg->fh_array[i];
683 }
684 
685 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
686 {
687 	might_sleep();
688 	wait_on_bit_action(&ds->ds_state, NFS4DS_CONNECTING,
689 			   nfs_wait_bit_killable, TASK_KILLABLE);
690 }
691 
692 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
693 {
694 	smp_mb__before_atomic();
695 	clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
696 	smp_mb__after_atomic();
697 	wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
698 }
699 
700 
701 struct nfs4_pnfs_ds *
702 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
703 {
704 	struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
705 	struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
706 	struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
707 	struct nfs4_pnfs_ds *ret = ds;
708 
709 	if (ds == NULL) {
710 		printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
711 			__func__, ds_idx);
712 		filelayout_mark_devid_invalid(devid);
713 		goto out;
714 	}
715 	smp_rmb();
716 	if (ds->ds_clp)
717 		goto out_test_devid;
718 
719 	if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
720 		struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
721 		int err;
722 
723 		err = nfs4_ds_connect(s, ds);
724 		if (err)
725 			nfs4_mark_deviceid_unavailable(devid);
726 		nfs4_clear_ds_conn_bit(ds);
727 	} else {
728 		/* Either ds is connected, or ds is NULL */
729 		nfs4_wait_ds_connect(ds);
730 	}
731 out_test_devid:
732 	if (filelayout_test_devid_unavailable(devid))
733 		ret = NULL;
734 out:
735 	return ret;
736 }
737 
738 module_param(dataserver_retrans, uint, 0644);
739 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
740 			"retries a request before it attempts further "
741 			" recovery  action.");
742 module_param(dataserver_timeo, uint, 0644);
743 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
744 			"NFSv4.1  client  waits for a response from a "
745 			" data server before it retries an NFS request.");
746