xref: /openbmc/linux/fs/nfs/pnfs_nfs.c (revision bb0eb050)
1 /*
2  * Common NFS I/O  operations for the pnfs file based
3  * layout drivers.
4  *
5  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
6  *
7  * Tom Haynes <loghyr@primarydata.com>
8  */
9 
10 #include <linux/nfs_fs.h>
11 #include <linux/nfs_page.h>
12 #include <linux/sunrpc/addr.h>
13 #include <linux/module.h>
14 
15 #include "nfs4session.h"
16 #include "internal.h"
17 #include "pnfs.h"
18 
19 #define NFSDBG_FACILITY		NFSDBG_PNFS
20 
21 void pnfs_generic_rw_release(void *data)
22 {
23 	struct nfs_pgio_header *hdr = data;
24 
25 	nfs_put_client(hdr->ds_clp);
26 	hdr->mds_ops->rpc_release(data);
27 }
28 EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
29 
30 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
31 void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data)
32 {
33 	struct nfs_page *first = nfs_list_entry(data->pages.next);
34 
35 	data->task.tk_status = 0;
36 	memcpy(&data->verf.verifier, &first->wb_verf,
37 	       sizeof(data->verf.verifier));
38 	data->verf.verifier.data[0]++; /* ensure verifier mismatch */
39 }
40 EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes);
41 
42 void pnfs_generic_write_commit_done(struct rpc_task *task, void *data)
43 {
44 	struct nfs_commit_data *wdata = data;
45 
46 	/* Note this may cause RPC to be resent */
47 	wdata->mds_ops->rpc_call_done(task, data);
48 }
49 EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done);
50 
51 void pnfs_generic_commit_release(void *calldata)
52 {
53 	struct nfs_commit_data *data = calldata;
54 
55 	data->completion_ops->completion(data);
56 	pnfs_put_lseg(data->lseg);
57 	nfs_put_client(data->ds_clp);
58 	nfs_commitdata_release(data);
59 }
60 EXPORT_SYMBOL_GPL(pnfs_generic_commit_release);
61 
62 /* The generic layer is about to remove the req from the commit list.
63  * If this will make the bucket empty, it will need to put the lseg reference.
64  * Note this must be called holding i_lock
65  */
66 void
67 pnfs_generic_clear_request_commit(struct nfs_page *req,
68 				  struct nfs_commit_info *cinfo)
69 {
70 	struct pnfs_layout_segment *freeme = NULL;
71 
72 	if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
73 		goto out;
74 	cinfo->ds->nwritten--;
75 	if (list_is_singular(&req->wb_list)) {
76 		struct pnfs_commit_bucket *bucket;
77 
78 		bucket = list_first_entry(&req->wb_list,
79 					  struct pnfs_commit_bucket,
80 					  written);
81 		freeme = bucket->wlseg;
82 		bucket->wlseg = NULL;
83 	}
84 out:
85 	nfs_request_remove_commit_list(req, cinfo);
86 	pnfs_put_lseg_locked(freeme);
87 }
88 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
89 
90 static int
91 pnfs_generic_transfer_commit_list(struct list_head *src, struct list_head *dst,
92 				  struct nfs_commit_info *cinfo, int max)
93 {
94 	struct nfs_page *req, *tmp;
95 	int ret = 0;
96 
97 	list_for_each_entry_safe(req, tmp, src, wb_list) {
98 		if (!nfs_lock_request(req))
99 			continue;
100 		kref_get(&req->wb_kref);
101 		if (cond_resched_lock(&cinfo->inode->i_lock))
102 			list_safe_reset_next(req, tmp, wb_list);
103 		nfs_request_remove_commit_list(req, cinfo);
104 		clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
105 		nfs_list_add_request(req, dst);
106 		ret++;
107 		if ((ret == max) && !cinfo->dreq)
108 			break;
109 	}
110 	return ret;
111 }
112 
113 static int
114 pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
115 				 struct nfs_commit_info *cinfo,
116 				 int max)
117 {
118 	struct list_head *src = &bucket->written;
119 	struct list_head *dst = &bucket->committing;
120 	int ret;
121 
122 	lockdep_assert_held(&cinfo->inode->i_lock);
123 	ret = pnfs_generic_transfer_commit_list(src, dst, cinfo, max);
124 	if (ret) {
125 		cinfo->ds->nwritten -= ret;
126 		cinfo->ds->ncommitting += ret;
127 		if (bucket->clseg == NULL)
128 			bucket->clseg = pnfs_get_lseg(bucket->wlseg);
129 		if (list_empty(src)) {
130 			pnfs_put_lseg_locked(bucket->wlseg);
131 			bucket->wlseg = NULL;
132 		}
133 	}
134 	return ret;
135 }
136 
137 /* Move reqs from written to committing lists, returning count
138  * of number moved.
139  */
140 int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo,
141 				   int max)
142 {
143 	int i, rv = 0, cnt;
144 
145 	lockdep_assert_held(&cinfo->inode->i_lock);
146 	for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
147 		cnt = pnfs_generic_scan_ds_commit_list(&cinfo->ds->buckets[i],
148 						       cinfo, max);
149 		max -= cnt;
150 		rv += cnt;
151 	}
152 	return rv;
153 }
154 EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
155 
156 /* Pull everything off the committing lists and dump into @dst.  */
157 void pnfs_generic_recover_commit_reqs(struct list_head *dst,
158 				      struct nfs_commit_info *cinfo)
159 {
160 	struct pnfs_commit_bucket *b;
161 	struct pnfs_layout_segment *freeme;
162 	int i;
163 
164 	lockdep_assert_held(&cinfo->inode->i_lock);
165 restart:
166 	for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
167 		if (pnfs_generic_transfer_commit_list(&b->written, dst,
168 						      cinfo, 0)) {
169 			freeme = b->wlseg;
170 			b->wlseg = NULL;
171 			spin_unlock(&cinfo->inode->i_lock);
172 			pnfs_put_lseg(freeme);
173 			spin_lock(&cinfo->inode->i_lock);
174 			goto restart;
175 		}
176 	}
177 	cinfo->ds->nwritten = 0;
178 }
179 EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
180 
181 static void pnfs_generic_retry_commit(struct nfs_commit_info *cinfo, int idx)
182 {
183 	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
184 	struct pnfs_commit_bucket *bucket;
185 	struct pnfs_layout_segment *freeme;
186 	LIST_HEAD(pages);
187 	int i;
188 
189 	spin_lock(&cinfo->inode->i_lock);
190 	for (i = idx; i < fl_cinfo->nbuckets; i++) {
191 		bucket = &fl_cinfo->buckets[i];
192 		if (list_empty(&bucket->committing))
193 			continue;
194 		freeme = bucket->clseg;
195 		bucket->clseg = NULL;
196 		list_splice_init(&bucket->committing, &pages);
197 		spin_unlock(&cinfo->inode->i_lock);
198 		nfs_retry_commit(&pages, freeme, cinfo, i);
199 		pnfs_put_lseg(freeme);
200 		spin_lock(&cinfo->inode->i_lock);
201 	}
202 	spin_unlock(&cinfo->inode->i_lock);
203 }
204 
205 static unsigned int
206 pnfs_generic_alloc_ds_commits(struct nfs_commit_info *cinfo,
207 			      struct list_head *list)
208 {
209 	struct pnfs_ds_commit_info *fl_cinfo;
210 	struct pnfs_commit_bucket *bucket;
211 	struct nfs_commit_data *data;
212 	int i;
213 	unsigned int nreq = 0;
214 
215 	fl_cinfo = cinfo->ds;
216 	bucket = fl_cinfo->buckets;
217 	for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
218 		if (list_empty(&bucket->committing))
219 			continue;
220 		/*
221 		 * If the layout segment is invalid, then let
222 		 * pnfs_generic_retry_commit() clean up the bucket.
223 		 */
224 		if (bucket->clseg && !pnfs_is_valid_lseg(bucket->clseg) &&
225 		    !test_bit(NFS_LSEG_LAYOUTRETURN, &bucket->clseg->pls_flags))
226 			break;
227 		data = nfs_commitdata_alloc(false);
228 		if (!data)
229 			break;
230 		data->ds_commit_index = i;
231 		list_add(&data->pages, list);
232 		nreq++;
233 	}
234 
235 	/* Clean up on error */
236 	pnfs_generic_retry_commit(cinfo, i);
237 	return nreq;
238 }
239 
240 static inline
241 void pnfs_fetch_commit_bucket_list(struct list_head *pages,
242 		struct nfs_commit_data *data,
243 		struct nfs_commit_info *cinfo)
244 {
245 	struct pnfs_commit_bucket *bucket;
246 
247 	bucket = &cinfo->ds->buckets[data->ds_commit_index];
248 	spin_lock(&cinfo->inode->i_lock);
249 	list_splice_init(&bucket->committing, pages);
250 	data->lseg = bucket->clseg;
251 	bucket->clseg = NULL;
252 	spin_unlock(&cinfo->inode->i_lock);
253 
254 }
255 
256 /* Helper function for pnfs_generic_commit_pagelist to catch an empty
257  * page list. This can happen when two commits race.
258  *
259  * This must be called instead of nfs_init_commit - call one or the other, but
260  * not both!
261  */
262 static bool
263 pnfs_generic_commit_cancel_empty_pagelist(struct list_head *pages,
264 					  struct nfs_commit_data *data,
265 					  struct nfs_commit_info *cinfo)
266 {
267 	if (list_empty(pages)) {
268 		if (atomic_dec_and_test(&cinfo->mds->rpcs_out))
269 			wake_up_atomic_t(&cinfo->mds->rpcs_out);
270 		/* don't call nfs_commitdata_release - it tries to put
271 		 * the open_context which is not acquired until nfs_init_commit
272 		 * which has not been called on @data */
273 		WARN_ON_ONCE(data->context);
274 		nfs_commit_free(data);
275 		return true;
276 	}
277 
278 	return false;
279 }
280 
281 /* This follows nfs_commit_list pretty closely */
282 int
283 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
284 			     int how, struct nfs_commit_info *cinfo,
285 			     int (*initiate_commit)(struct nfs_commit_data *data,
286 						    int how))
287 {
288 	struct nfs_commit_data *data, *tmp;
289 	LIST_HEAD(list);
290 	unsigned int nreq = 0;
291 
292 	if (!list_empty(mds_pages)) {
293 		data = nfs_commitdata_alloc(true);
294 		data->ds_commit_index = -1;
295 		list_add(&data->pages, &list);
296 		nreq++;
297 	}
298 
299 	nreq += pnfs_generic_alloc_ds_commits(cinfo, &list);
300 
301 	if (nreq == 0)
302 		goto out;
303 
304 	atomic_add(nreq, &cinfo->mds->rpcs_out);
305 
306 	list_for_each_entry_safe(data, tmp, &list, pages) {
307 		list_del_init(&data->pages);
308 		if (data->ds_commit_index < 0) {
309 			/* another commit raced with us */
310 			if (pnfs_generic_commit_cancel_empty_pagelist(mds_pages,
311 				data, cinfo))
312 				continue;
313 
314 			nfs_init_commit(data, mds_pages, NULL, cinfo);
315 			nfs_initiate_commit(NFS_CLIENT(inode), data,
316 					    NFS_PROTO(data->inode),
317 					    data->mds_ops, how, 0);
318 		} else {
319 			LIST_HEAD(pages);
320 
321 			pnfs_fetch_commit_bucket_list(&pages, data, cinfo);
322 
323 			/* another commit raced with us */
324 			if (pnfs_generic_commit_cancel_empty_pagelist(&pages,
325 				data, cinfo))
326 				continue;
327 
328 			nfs_init_commit(data, &pages, data->lseg, cinfo);
329 			initiate_commit(data, how);
330 		}
331 	}
332 out:
333 	cinfo->ds->ncommitting = 0;
334 	return PNFS_ATTEMPTED;
335 }
336 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
337 
338 /*
339  * Data server cache
340  *
341  * Data servers can be mapped to different device ids.
342  * nfs4_pnfs_ds reference counting
343  *   - set to 1 on allocation
344  *   - incremented when a device id maps a data server already in the cache.
345  *   - decremented when deviceid is removed from the cache.
346  */
347 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
348 static LIST_HEAD(nfs4_data_server_cache);
349 
350 /* Debug routines */
351 static void
352 print_ds(struct nfs4_pnfs_ds *ds)
353 {
354 	if (ds == NULL) {
355 		printk(KERN_WARNING "%s NULL device\n", __func__);
356 		return;
357 	}
358 	printk(KERN_WARNING "        ds %s\n"
359 		"        ref count %d\n"
360 		"        client %p\n"
361 		"        cl_exchange_flags %x\n",
362 		ds->ds_remotestr,
363 		atomic_read(&ds->ds_count), ds->ds_clp,
364 		ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
365 }
366 
367 static bool
368 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
369 {
370 	struct sockaddr_in *a, *b;
371 	struct sockaddr_in6 *a6, *b6;
372 
373 	if (addr1->sa_family != addr2->sa_family)
374 		return false;
375 
376 	switch (addr1->sa_family) {
377 	case AF_INET:
378 		a = (struct sockaddr_in *)addr1;
379 		b = (struct sockaddr_in *)addr2;
380 
381 		if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
382 		    a->sin_port == b->sin_port)
383 			return true;
384 		break;
385 
386 	case AF_INET6:
387 		a6 = (struct sockaddr_in6 *)addr1;
388 		b6 = (struct sockaddr_in6 *)addr2;
389 
390 		/* LINKLOCAL addresses must have matching scope_id */
391 		if (ipv6_addr_src_scope(&a6->sin6_addr) ==
392 		    IPV6_ADDR_SCOPE_LINKLOCAL &&
393 		    a6->sin6_scope_id != b6->sin6_scope_id)
394 			return false;
395 
396 		if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
397 		    a6->sin6_port == b6->sin6_port)
398 			return true;
399 		break;
400 
401 	default:
402 		dprintk("%s: unhandled address family: %u\n",
403 			__func__, addr1->sa_family);
404 		return false;
405 	}
406 
407 	return false;
408 }
409 
410 /*
411  * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
412  * declare a match.
413  */
414 static bool
415 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
416 			       const struct list_head *dsaddrs2)
417 {
418 	struct nfs4_pnfs_ds_addr *da1, *da2;
419 	struct sockaddr *sa1, *sa2;
420 	bool match = false;
421 
422 	list_for_each_entry(da1, dsaddrs1, da_node) {
423 		sa1 = (struct sockaddr *)&da1->da_addr;
424 		match = false;
425 		list_for_each_entry(da2, dsaddrs2, da_node) {
426 			sa2 = (struct sockaddr *)&da2->da_addr;
427 			match = same_sockaddr(sa1, sa2);
428 			if (match)
429 				break;
430 		}
431 		if (!match)
432 			break;
433 	}
434 	return match;
435 }
436 
437 /*
438  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
439  */
440 static struct nfs4_pnfs_ds *
441 _data_server_lookup_locked(const struct list_head *dsaddrs)
442 {
443 	struct nfs4_pnfs_ds *ds;
444 
445 	list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
446 		if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
447 			return ds;
448 	return NULL;
449 }
450 
451 static void destroy_ds(struct nfs4_pnfs_ds *ds)
452 {
453 	struct nfs4_pnfs_ds_addr *da;
454 
455 	dprintk("--> %s\n", __func__);
456 	ifdebug(FACILITY)
457 		print_ds(ds);
458 
459 	nfs_put_client(ds->ds_clp);
460 
461 	while (!list_empty(&ds->ds_addrs)) {
462 		da = list_first_entry(&ds->ds_addrs,
463 				      struct nfs4_pnfs_ds_addr,
464 				      da_node);
465 		list_del_init(&da->da_node);
466 		kfree(da->da_remotestr);
467 		kfree(da);
468 	}
469 
470 	kfree(ds->ds_remotestr);
471 	kfree(ds);
472 }
473 
474 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
475 {
476 	if (atomic_dec_and_lock(&ds->ds_count,
477 				&nfs4_ds_cache_lock)) {
478 		list_del_init(&ds->ds_node);
479 		spin_unlock(&nfs4_ds_cache_lock);
480 		destroy_ds(ds);
481 	}
482 }
483 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
484 
485 /*
486  * Create a string with a human readable address and port to avoid
487  * complicated setup around many dprinks.
488  */
489 static char *
490 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
491 {
492 	struct nfs4_pnfs_ds_addr *da;
493 	char *remotestr;
494 	size_t len;
495 	char *p;
496 
497 	len = 3;        /* '{', '}' and eol */
498 	list_for_each_entry(da, dsaddrs, da_node) {
499 		len += strlen(da->da_remotestr) + 1;    /* string plus comma */
500 	}
501 
502 	remotestr = kzalloc(len, gfp_flags);
503 	if (!remotestr)
504 		return NULL;
505 
506 	p = remotestr;
507 	*(p++) = '{';
508 	len--;
509 	list_for_each_entry(da, dsaddrs, da_node) {
510 		size_t ll = strlen(da->da_remotestr);
511 
512 		if (ll > len)
513 			goto out_err;
514 
515 		memcpy(p, da->da_remotestr, ll);
516 		p += ll;
517 		len -= ll;
518 
519 		if (len < 1)
520 			goto out_err;
521 		(*p++) = ',';
522 		len--;
523 	}
524 	if (len < 2)
525 		goto out_err;
526 	*(p++) = '}';
527 	*p = '\0';
528 	return remotestr;
529 out_err:
530 	kfree(remotestr);
531 	return NULL;
532 }
533 
534 /*
535  * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
536  * uncached and return cached struct nfs4_pnfs_ds.
537  */
538 struct nfs4_pnfs_ds *
539 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
540 {
541 	struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
542 	char *remotestr;
543 
544 	if (list_empty(dsaddrs)) {
545 		dprintk("%s: no addresses defined\n", __func__);
546 		goto out;
547 	}
548 
549 	ds = kzalloc(sizeof(*ds), gfp_flags);
550 	if (!ds)
551 		goto out;
552 
553 	/* this is only used for debugging, so it's ok if its NULL */
554 	remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
555 
556 	spin_lock(&nfs4_ds_cache_lock);
557 	tmp_ds = _data_server_lookup_locked(dsaddrs);
558 	if (tmp_ds == NULL) {
559 		INIT_LIST_HEAD(&ds->ds_addrs);
560 		list_splice_init(dsaddrs, &ds->ds_addrs);
561 		ds->ds_remotestr = remotestr;
562 		atomic_set(&ds->ds_count, 1);
563 		INIT_LIST_HEAD(&ds->ds_node);
564 		ds->ds_clp = NULL;
565 		list_add(&ds->ds_node, &nfs4_data_server_cache);
566 		dprintk("%s add new data server %s\n", __func__,
567 			ds->ds_remotestr);
568 	} else {
569 		kfree(remotestr);
570 		kfree(ds);
571 		atomic_inc(&tmp_ds->ds_count);
572 		dprintk("%s data server %s found, inc'ed ds_count to %d\n",
573 			__func__, tmp_ds->ds_remotestr,
574 			atomic_read(&tmp_ds->ds_count));
575 		ds = tmp_ds;
576 	}
577 	spin_unlock(&nfs4_ds_cache_lock);
578 out:
579 	return ds;
580 }
581 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
582 
583 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
584 {
585 	might_sleep();
586 	wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
587 			TASK_KILLABLE);
588 }
589 
590 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
591 {
592 	smp_mb__before_atomic();
593 	clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
594 	smp_mb__after_atomic();
595 	wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
596 }
597 
598 static struct nfs_client *(*get_v3_ds_connect)(
599 			struct nfs_server *mds_srv,
600 			const struct sockaddr *ds_addr,
601 			int ds_addrlen,
602 			int ds_proto,
603 			unsigned int ds_timeo,
604 			unsigned int ds_retrans);
605 
606 static bool load_v3_ds_connect(void)
607 {
608 	if (!get_v3_ds_connect) {
609 		get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
610 		WARN_ON_ONCE(!get_v3_ds_connect);
611 	}
612 
613 	return(get_v3_ds_connect != NULL);
614 }
615 
616 void nfs4_pnfs_v3_ds_connect_unload(void)
617 {
618 	if (get_v3_ds_connect) {
619 		symbol_put(nfs3_set_ds_client);
620 		get_v3_ds_connect = NULL;
621 	}
622 }
623 
624 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
625 				 struct nfs4_pnfs_ds *ds,
626 				 unsigned int timeo,
627 				 unsigned int retrans)
628 {
629 	struct nfs_client *clp = ERR_PTR(-EIO);
630 	struct nfs4_pnfs_ds_addr *da;
631 	int status = 0;
632 
633 	dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
634 
635 	if (!load_v3_ds_connect())
636 		goto out;
637 
638 	list_for_each_entry(da, &ds->ds_addrs, da_node) {
639 		dprintk("%s: DS %s: trying address %s\n",
640 			__func__, ds->ds_remotestr, da->da_remotestr);
641 
642 		if (!IS_ERR(clp)) {
643 			struct xprt_create xprt_args = {
644 				.ident = XPRT_TRANSPORT_TCP,
645 				.net = clp->cl_net,
646 				.dstaddr = (struct sockaddr *)&da->da_addr,
647 				.addrlen = da->da_addrlen,
648 				.servername = clp->cl_hostname,
649 			};
650 			/* Add this address as an alias */
651 			rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
652 					rpc_clnt_test_and_add_xprt, NULL);
653 		} else
654 			clp = get_v3_ds_connect(mds_srv,
655 					(struct sockaddr *)&da->da_addr,
656 					da->da_addrlen, IPPROTO_TCP,
657 					timeo, retrans);
658 	}
659 
660 	if (IS_ERR(clp)) {
661 		status = PTR_ERR(clp);
662 		goto out;
663 	}
664 
665 	smp_wmb();
666 	ds->ds_clp = clp;
667 	dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
668 out:
669 	return status;
670 }
671 
672 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
673 				 struct nfs4_pnfs_ds *ds,
674 				 unsigned int timeo,
675 				 unsigned int retrans,
676 				 u32 minor_version)
677 {
678 	struct nfs_client *clp = ERR_PTR(-EIO);
679 	struct nfs4_pnfs_ds_addr *da;
680 	int status = 0;
681 
682 	dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
683 
684 	list_for_each_entry(da, &ds->ds_addrs, da_node) {
685 		dprintk("%s: DS %s: trying address %s\n",
686 			__func__, ds->ds_remotestr, da->da_remotestr);
687 
688 		if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) {
689 			struct xprt_create xprt_args = {
690 				.ident = XPRT_TRANSPORT_TCP,
691 				.net = clp->cl_net,
692 				.dstaddr = (struct sockaddr *)&da->da_addr,
693 				.addrlen = da->da_addrlen,
694 				.servername = clp->cl_hostname,
695 			};
696 			struct nfs4_add_xprt_data xprtdata = {
697 				.clp = clp,
698 				.cred = nfs4_get_clid_cred(clp),
699 			};
700 			struct rpc_add_xprt_test rpcdata = {
701 				.add_xprt_test = clp->cl_mvops->session_trunk,
702 				.data = &xprtdata,
703 			};
704 
705 			/**
706 			* Test this address for session trunking and
707 			* add as an alias
708 			*/
709 			rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
710 					  rpc_clnt_setup_test_and_add_xprt,
711 					  &rpcdata);
712 			if (xprtdata.cred)
713 				put_rpccred(xprtdata.cred);
714 		} else {
715 			clp = nfs4_set_ds_client(mds_srv,
716 						(struct sockaddr *)&da->da_addr,
717 						da->da_addrlen, IPPROTO_TCP,
718 						timeo, retrans, minor_version);
719 			if (IS_ERR(clp))
720 				continue;
721 
722 			status = nfs4_init_ds_session(clp,
723 					mds_srv->nfs_client->cl_lease_time);
724 			if (status) {
725 				nfs_put_client(clp);
726 				clp = ERR_PTR(-EIO);
727 				continue;
728 			}
729 
730 		}
731 	}
732 
733 	if (IS_ERR(clp)) {
734 		status = PTR_ERR(clp);
735 		goto out;
736 	}
737 
738 	smp_wmb();
739 	ds->ds_clp = clp;
740 	dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
741 out:
742 	return status;
743 }
744 
745 /*
746  * Create an rpc connection to the nfs4_pnfs_ds data server.
747  * Currently only supports IPv4 and IPv6 addresses.
748  * If connection fails, make devid unavailable and return a -errno.
749  */
750 int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
751 			  struct nfs4_deviceid_node *devid, unsigned int timeo,
752 			  unsigned int retrans, u32 version, u32 minor_version)
753 {
754 	int err;
755 
756 again:
757 	err = 0;
758 	if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
759 		if (version == 3) {
760 			err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
761 						       retrans);
762 		} else if (version == 4) {
763 			err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
764 						       retrans, minor_version);
765 		} else {
766 			dprintk("%s: unsupported DS version %d\n", __func__,
767 				version);
768 			err = -EPROTONOSUPPORT;
769 		}
770 
771 		nfs4_clear_ds_conn_bit(ds);
772 	} else {
773 		nfs4_wait_ds_connect(ds);
774 
775 		/* what was waited on didn't connect AND didn't mark unavail */
776 		if (!ds->ds_clp && !nfs4_test_deviceid_unavailable(devid))
777 			goto again;
778 	}
779 
780 	/*
781 	 * At this point the ds->ds_clp should be ready, but it might have
782 	 * hit an error.
783 	 */
784 	if (!err) {
785 		if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) {
786 			WARN_ON_ONCE(ds->ds_clp ||
787 				!nfs4_test_deviceid_unavailable(devid));
788 			return -EINVAL;
789 		}
790 		err = nfs_client_init_status(ds->ds_clp);
791 	}
792 
793 	return err;
794 }
795 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
796 
797 /*
798  * Currently only supports ipv4, ipv6 and one multi-path address.
799  */
800 struct nfs4_pnfs_ds_addr *
801 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
802 {
803 	struct nfs4_pnfs_ds_addr *da = NULL;
804 	char *buf, *portstr;
805 	__be16 port;
806 	int nlen, rlen;
807 	int tmp[2];
808 	__be32 *p;
809 	char *netid, *match_netid;
810 	size_t len, match_netid_len;
811 	char *startsep = "";
812 	char *endsep = "";
813 
814 
815 	/* r_netid */
816 	p = xdr_inline_decode(xdr, 4);
817 	if (unlikely(!p))
818 		goto out_err;
819 	nlen = be32_to_cpup(p++);
820 
821 	p = xdr_inline_decode(xdr, nlen);
822 	if (unlikely(!p))
823 		goto out_err;
824 
825 	netid = kmalloc(nlen+1, gfp_flags);
826 	if (unlikely(!netid))
827 		goto out_err;
828 
829 	netid[nlen] = '\0';
830 	memcpy(netid, p, nlen);
831 
832 	/* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
833 	p = xdr_inline_decode(xdr, 4);
834 	if (unlikely(!p))
835 		goto out_free_netid;
836 	rlen = be32_to_cpup(p);
837 
838 	p = xdr_inline_decode(xdr, rlen);
839 	if (unlikely(!p))
840 		goto out_free_netid;
841 
842 	/* port is ".ABC.DEF", 8 chars max */
843 	if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
844 		dprintk("%s: Invalid address, length %d\n", __func__,
845 			rlen);
846 		goto out_free_netid;
847 	}
848 	buf = kmalloc(rlen + 1, gfp_flags);
849 	if (!buf) {
850 		dprintk("%s: Not enough memory\n", __func__);
851 		goto out_free_netid;
852 	}
853 	buf[rlen] = '\0';
854 	memcpy(buf, p, rlen);
855 
856 	/* replace port '.' with '-' */
857 	portstr = strrchr(buf, '.');
858 	if (!portstr) {
859 		dprintk("%s: Failed finding expected dot in port\n",
860 			__func__);
861 		goto out_free_buf;
862 	}
863 	*portstr = '-';
864 
865 	/* find '.' between address and port */
866 	portstr = strrchr(buf, '.');
867 	if (!portstr) {
868 		dprintk("%s: Failed finding expected dot between address and "
869 			"port\n", __func__);
870 		goto out_free_buf;
871 	}
872 	*portstr = '\0';
873 
874 	da = kzalloc(sizeof(*da), gfp_flags);
875 	if (unlikely(!da))
876 		goto out_free_buf;
877 
878 	INIT_LIST_HEAD(&da->da_node);
879 
880 	if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
881 		      sizeof(da->da_addr))) {
882 		dprintk("%s: error parsing address %s\n", __func__, buf);
883 		goto out_free_da;
884 	}
885 
886 	portstr++;
887 	sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
888 	port = htons((tmp[0] << 8) | (tmp[1]));
889 
890 	switch (da->da_addr.ss_family) {
891 	case AF_INET:
892 		((struct sockaddr_in *)&da->da_addr)->sin_port = port;
893 		da->da_addrlen = sizeof(struct sockaddr_in);
894 		match_netid = "tcp";
895 		match_netid_len = 3;
896 		break;
897 
898 	case AF_INET6:
899 		((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
900 		da->da_addrlen = sizeof(struct sockaddr_in6);
901 		match_netid = "tcp6";
902 		match_netid_len = 4;
903 		startsep = "[";
904 		endsep = "]";
905 		break;
906 
907 	default:
908 		dprintk("%s: unsupported address family: %u\n",
909 			__func__, da->da_addr.ss_family);
910 		goto out_free_da;
911 	}
912 
913 	if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
914 		dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
915 			__func__, netid, match_netid);
916 		goto out_free_da;
917 	}
918 
919 	/* save human readable address */
920 	len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
921 	da->da_remotestr = kzalloc(len, gfp_flags);
922 
923 	/* NULL is ok, only used for dprintk */
924 	if (da->da_remotestr)
925 		snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
926 			 buf, endsep, ntohs(port));
927 
928 	dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
929 	kfree(buf);
930 	kfree(netid);
931 	return da;
932 
933 out_free_da:
934 	kfree(da);
935 out_free_buf:
936 	dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
937 	kfree(buf);
938 out_free_netid:
939 	kfree(netid);
940 out_err:
941 	return NULL;
942 }
943 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
944 
945 void
946 pnfs_layout_mark_request_commit(struct nfs_page *req,
947 				struct pnfs_layout_segment *lseg,
948 				struct nfs_commit_info *cinfo,
949 				u32 ds_commit_idx)
950 {
951 	struct list_head *list;
952 	struct pnfs_commit_bucket *buckets;
953 
954 	spin_lock(&cinfo->inode->i_lock);
955 	buckets = cinfo->ds->buckets;
956 	list = &buckets[ds_commit_idx].written;
957 	if (list_empty(list)) {
958 		if (!pnfs_is_valid_lseg(lseg)) {
959 			spin_unlock(&cinfo->inode->i_lock);
960 			cinfo->completion_ops->resched_write(cinfo, req);
961 			return;
962 		}
963 		/* Non-empty buckets hold a reference on the lseg.  That ref
964 		 * is normally transferred to the COMMIT call and released
965 		 * there.  It could also be released if the last req is pulled
966 		 * off due to a rewrite, in which case it will be done in
967 		 * pnfs_common_clear_request_commit
968 		 */
969 		WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL);
970 		buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg);
971 	}
972 	set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
973 	cinfo->ds->nwritten++;
974 
975 	nfs_request_add_commit_list_locked(req, list, cinfo);
976 	spin_unlock(&cinfo->inode->i_lock);
977 	nfs_mark_page_unstable(req->wb_page, cinfo);
978 }
979 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
980 
981 int
982 pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
983 {
984 	int ret;
985 
986 	if (!pnfs_layoutcommit_outstanding(inode))
987 		return 0;
988 	ret = nfs_commit_inode(inode, FLUSH_SYNC);
989 	if (ret < 0)
990 		return ret;
991 	if (datasync)
992 		return 0;
993 	return pnfs_layoutcommit_inode(inode, true);
994 }
995 EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
996 
997