xref: /linux/fs/nfs/pnfs_nfs.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Common NFS I/O  operations for the pnfs file based
4  * layout drivers.
5  *
6  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
7  *
8  * Tom Haynes <loghyr@primarydata.com>
9  */
10 
11 #include <linux/nfs_fs.h>
12 #include <linux/nfs_page.h>
13 #include <linux/sunrpc/addr.h>
14 #include <linux/module.h>
15 
16 #include "nfs4session.h"
17 #include "internal.h"
18 #include "pnfs.h"
19 
20 #define NFSDBG_FACILITY		NFSDBG_PNFS
21 
22 void pnfs_generic_rw_release(void *data)
23 {
24 	struct nfs_pgio_header *hdr = data;
25 
26 	nfs_put_client(hdr->ds_clp);
27 	hdr->mds_ops->rpc_release(data);
28 }
29 EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
30 
31 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
32 void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data)
33 {
34 	struct nfs_writeverf *verf = data->res.verf;
35 
36 	data->task.tk_status = 0;
37 	memset(&verf->verifier, 0, sizeof(verf->verifier));
38 	verf->committed = NFS_UNSTABLE;
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 static struct pnfs_layout_segment *
63 pnfs_free_bucket_lseg(struct pnfs_commit_bucket *bucket)
64 {
65 	if (list_empty(&bucket->committing) && list_empty(&bucket->written)) {
66 		struct pnfs_layout_segment *freeme = bucket->lseg;
67 		bucket->lseg = NULL;
68 		return freeme;
69 	}
70 	return NULL;
71 }
72 
73 /* The generic layer is about to remove the req from the commit list.
74  * If this will make the bucket empty, it will need to put the lseg reference.
75  * Note this must be called holding nfsi->commit_mutex
76  */
77 void
78 pnfs_generic_clear_request_commit(struct nfs_page *req,
79 				  struct nfs_commit_info *cinfo)
80 {
81 	struct pnfs_commit_bucket *bucket = NULL;
82 
83 	if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
84 		goto out;
85 	cinfo->ds->nwritten--;
86 	if (list_is_singular(&req->wb_list))
87 		bucket = list_first_entry(&req->wb_list,
88 					  struct pnfs_commit_bucket, written);
89 out:
90 	nfs_request_remove_commit_list(req, cinfo);
91 	if (bucket)
92 		pnfs_put_lseg(pnfs_free_bucket_lseg(bucket));
93 }
94 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
95 
96 struct pnfs_commit_array *
97 pnfs_alloc_commit_array(size_t n, gfp_t gfp_flags)
98 {
99 	struct pnfs_commit_array *p;
100 	struct pnfs_commit_bucket *b;
101 
102 	p = kmalloc(struct_size(p, buckets, n), gfp_flags);
103 	if (!p)
104 		return NULL;
105 	p->nbuckets = n;
106 	INIT_LIST_HEAD(&p->cinfo_list);
107 	INIT_LIST_HEAD(&p->lseg_list);
108 	p->lseg = NULL;
109 	for (b = &p->buckets[0]; n != 0; b++, n--) {
110 		INIT_LIST_HEAD(&b->written);
111 		INIT_LIST_HEAD(&b->committing);
112 		b->lseg = NULL;
113 		b->direct_verf.committed = NFS_INVALID_STABLE_HOW;
114 	}
115 	return p;
116 }
117 EXPORT_SYMBOL_GPL(pnfs_alloc_commit_array);
118 
119 void
120 pnfs_free_commit_array(struct pnfs_commit_array *p)
121 {
122 	kfree_rcu(p, rcu);
123 }
124 EXPORT_SYMBOL_GPL(pnfs_free_commit_array);
125 
126 static struct pnfs_commit_array *
127 pnfs_find_commit_array_by_lseg(struct pnfs_ds_commit_info *fl_cinfo,
128 		struct pnfs_layout_segment *lseg)
129 {
130 	struct pnfs_commit_array *array;
131 
132 	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
133 		if (array->lseg == lseg)
134 			return array;
135 	}
136 	return NULL;
137 }
138 
139 struct pnfs_commit_array *
140 pnfs_add_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
141 		struct pnfs_commit_array *new,
142 		struct pnfs_layout_segment *lseg)
143 {
144 	struct pnfs_commit_array *array;
145 
146 	array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
147 	if (array)
148 		return array;
149 	new->lseg = lseg;
150 	refcount_set(&new->refcount, 1);
151 	list_add_rcu(&new->cinfo_list, &fl_cinfo->commits);
152 	list_add(&new->lseg_list, &lseg->pls_commits);
153 	return new;
154 }
155 EXPORT_SYMBOL_GPL(pnfs_add_commit_array);
156 
157 static struct pnfs_commit_array *
158 pnfs_lookup_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
159 		struct pnfs_layout_segment *lseg)
160 {
161 	struct pnfs_commit_array *array;
162 
163 	rcu_read_lock();
164 	array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
165 	if (!array) {
166 		rcu_read_unlock();
167 		fl_cinfo->ops->setup_ds_info(fl_cinfo, lseg);
168 		rcu_read_lock();
169 		array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
170 	}
171 	rcu_read_unlock();
172 	return array;
173 }
174 
175 static void
176 pnfs_release_commit_array_locked(struct pnfs_commit_array *array)
177 {
178 	list_del_rcu(&array->cinfo_list);
179 	list_del(&array->lseg_list);
180 	pnfs_free_commit_array(array);
181 }
182 
183 static void
184 pnfs_put_commit_array_locked(struct pnfs_commit_array *array)
185 {
186 	if (refcount_dec_and_test(&array->refcount))
187 		pnfs_release_commit_array_locked(array);
188 }
189 
190 static void
191 pnfs_put_commit_array(struct pnfs_commit_array *array, struct inode *inode)
192 {
193 	if (refcount_dec_and_lock(&array->refcount, &inode->i_lock)) {
194 		pnfs_release_commit_array_locked(array);
195 		spin_unlock(&inode->i_lock);
196 	}
197 }
198 
199 static struct pnfs_commit_array *
200 pnfs_get_commit_array(struct pnfs_commit_array *array)
201 {
202 	if (refcount_inc_not_zero(&array->refcount))
203 		return array;
204 	return NULL;
205 }
206 
207 static void
208 pnfs_remove_and_free_commit_array(struct pnfs_commit_array *array)
209 {
210 	array->lseg = NULL;
211 	list_del_init(&array->lseg_list);
212 	pnfs_put_commit_array_locked(array);
213 }
214 
215 void
216 pnfs_generic_ds_cinfo_release_lseg(struct pnfs_ds_commit_info *fl_cinfo,
217 		struct pnfs_layout_segment *lseg)
218 {
219 	struct pnfs_commit_array *array, *tmp;
220 
221 	list_for_each_entry_safe(array, tmp, &lseg->pls_commits, lseg_list)
222 		pnfs_remove_and_free_commit_array(array);
223 }
224 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_release_lseg);
225 
226 void
227 pnfs_generic_ds_cinfo_destroy(struct pnfs_ds_commit_info *fl_cinfo)
228 {
229 	struct pnfs_commit_array *array, *tmp;
230 
231 	list_for_each_entry_safe(array, tmp, &fl_cinfo->commits, cinfo_list)
232 		pnfs_remove_and_free_commit_array(array);
233 }
234 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_destroy);
235 
236 /*
237  * Locks the nfs_page requests for commit and moves them to
238  * @bucket->committing.
239  */
240 static int
241 pnfs_bucket_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
242 				struct nfs_commit_info *cinfo,
243 				int max)
244 {
245 	struct list_head *src = &bucket->written;
246 	struct list_head *dst = &bucket->committing;
247 	int ret;
248 
249 	lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
250 	ret = nfs_scan_commit_list(src, dst, cinfo, max);
251 	if (ret) {
252 		cinfo->ds->nwritten -= ret;
253 		cinfo->ds->ncommitting += ret;
254 	}
255 	return ret;
256 }
257 
258 static int pnfs_bucket_scan_array(struct nfs_commit_info *cinfo,
259 				  struct pnfs_commit_bucket *buckets,
260 				  unsigned int nbuckets,
261 				  int max)
262 {
263 	unsigned int i;
264 	int rv = 0, cnt;
265 
266 	for (i = 0; i < nbuckets && max != 0; i++) {
267 		cnt = pnfs_bucket_scan_ds_commit_list(&buckets[i], cinfo, max);
268 		rv += cnt;
269 		max -= cnt;
270 	}
271 	return rv;
272 }
273 
274 /* Move reqs from written to committing lists, returning count
275  * of number moved.
276  */
277 int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo, int max)
278 {
279 	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
280 	struct pnfs_commit_array *array;
281 	int rv = 0, cnt;
282 
283 	rcu_read_lock();
284 	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
285 		if (!array->lseg || !pnfs_get_commit_array(array))
286 			continue;
287 		rcu_read_unlock();
288 		cnt = pnfs_bucket_scan_array(cinfo, array->buckets,
289 				array->nbuckets, max);
290 		rcu_read_lock();
291 		pnfs_put_commit_array(array, cinfo->inode);
292 		rv += cnt;
293 		max -= cnt;
294 		if (!max)
295 			break;
296 	}
297 	rcu_read_unlock();
298 	return rv;
299 }
300 EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
301 
302 static unsigned int
303 pnfs_bucket_recover_commit_reqs(struct list_head *dst,
304 			        struct pnfs_commit_bucket *buckets,
305 				unsigned int nbuckets,
306 				struct nfs_commit_info *cinfo)
307 {
308 	struct pnfs_commit_bucket *b;
309 	struct pnfs_layout_segment *freeme;
310 	unsigned int nwritten, ret = 0;
311 	unsigned int i;
312 
313 restart:
314 	for (i = 0, b = buckets; i < nbuckets; i++, b++) {
315 		nwritten = nfs_scan_commit_list(&b->written, dst, cinfo, 0);
316 		if (!nwritten)
317 			continue;
318 		ret += nwritten;
319 		freeme = pnfs_free_bucket_lseg(b);
320 		if (freeme) {
321 			pnfs_put_lseg(freeme);
322 			goto restart;
323 		}
324 	}
325 	return ret;
326 }
327 
328 /* Pull everything off the committing lists and dump into @dst.  */
329 void pnfs_generic_recover_commit_reqs(struct list_head *dst,
330 				      struct nfs_commit_info *cinfo)
331 {
332 	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
333 	struct pnfs_commit_array *array;
334 	unsigned int nwritten;
335 
336 	lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
337 	rcu_read_lock();
338 	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
339 		if (!array->lseg || !pnfs_get_commit_array(array))
340 			continue;
341 		rcu_read_unlock();
342 		nwritten = pnfs_bucket_recover_commit_reqs(dst,
343 							   array->buckets,
344 							   array->nbuckets,
345 							   cinfo);
346 		rcu_read_lock();
347 		pnfs_put_commit_array(array, cinfo->inode);
348 		fl_cinfo->nwritten -= nwritten;
349 	}
350 	rcu_read_unlock();
351 }
352 EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
353 
354 static struct nfs_page *
355 pnfs_bucket_search_commit_reqs(struct pnfs_commit_bucket *buckets,
356 		unsigned int nbuckets, struct page *page)
357 {
358 	struct nfs_page *req;
359 	struct pnfs_commit_bucket *b;
360 	unsigned int i;
361 
362 	/* Linearly search the commit lists for each bucket until a matching
363 	 * request is found */
364 	for (i = 0, b = buckets; i < nbuckets; i++, b++) {
365 		list_for_each_entry(req, &b->written, wb_list) {
366 			if (req->wb_page == page)
367 				return req->wb_head;
368 		}
369 		list_for_each_entry(req, &b->committing, wb_list) {
370 			if (req->wb_page == page)
371 				return req->wb_head;
372 		}
373 	}
374 	return NULL;
375 }
376 
377 /* pnfs_generic_search_commit_reqs - Search lists in @cinfo for the head reqest
378  *				   for @page
379  * @cinfo - commit info for current inode
380  * @page - page to search for matching head request
381  *
382  * Returns a the head request if one is found, otherwise returns NULL.
383  */
384 struct nfs_page *
385 pnfs_generic_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
386 {
387 	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
388 	struct pnfs_commit_array *array;
389 	struct nfs_page *req;
390 
391 	list_for_each_entry(array, &fl_cinfo->commits, cinfo_list) {
392 		req = pnfs_bucket_search_commit_reqs(array->buckets,
393 				array->nbuckets, page);
394 		if (req)
395 			return req;
396 	}
397 	return NULL;
398 }
399 EXPORT_SYMBOL_GPL(pnfs_generic_search_commit_reqs);
400 
401 static struct pnfs_layout_segment *
402 pnfs_bucket_get_committing(struct list_head *head,
403 			   struct pnfs_commit_bucket *bucket,
404 			   struct nfs_commit_info *cinfo)
405 {
406 	struct pnfs_layout_segment *lseg;
407 	struct list_head *pos;
408 
409 	list_for_each(pos, &bucket->committing)
410 		cinfo->ds->ncommitting--;
411 	list_splice_init(&bucket->committing, head);
412 	lseg = pnfs_free_bucket_lseg(bucket);
413 	if (!lseg)
414 		lseg = pnfs_get_lseg(bucket->lseg);
415 	return lseg;
416 }
417 
418 static struct nfs_commit_data *
419 pnfs_bucket_fetch_commitdata(struct pnfs_commit_bucket *bucket,
420 			     struct nfs_commit_info *cinfo)
421 {
422 	struct nfs_commit_data *data = nfs_commitdata_alloc(false);
423 
424 	if (!data)
425 		return NULL;
426 	data->lseg = pnfs_bucket_get_committing(&data->pages, bucket, cinfo);
427 	return data;
428 }
429 
430 static void pnfs_generic_retry_commit(struct pnfs_commit_bucket *buckets,
431 				      unsigned int nbuckets,
432 				      struct nfs_commit_info *cinfo,
433 				      unsigned int idx)
434 {
435 	struct pnfs_commit_bucket *bucket;
436 	struct pnfs_layout_segment *freeme;
437 	LIST_HEAD(pages);
438 
439 	for (bucket = buckets; idx < nbuckets; bucket++, idx++) {
440 		if (list_empty(&bucket->committing))
441 			continue;
442 		mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
443 		freeme = pnfs_bucket_get_committing(&pages, bucket, cinfo);
444 		mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
445 		nfs_retry_commit(&pages, freeme, cinfo, idx);
446 		pnfs_put_lseg(freeme);
447 	}
448 }
449 
450 static unsigned int
451 pnfs_bucket_alloc_ds_commits(struct list_head *list,
452 			     struct pnfs_commit_bucket *buckets,
453 			     unsigned int nbuckets,
454 			     struct nfs_commit_info *cinfo)
455 {
456 	struct pnfs_commit_bucket *bucket;
457 	struct nfs_commit_data *data;
458 	unsigned int i;
459 	unsigned int nreq = 0;
460 
461 	for (i = 0, bucket = buckets; i < nbuckets; i++, bucket++) {
462 		if (list_empty(&bucket->committing))
463 			continue;
464 		mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
465 		if (!list_empty(&bucket->committing)) {
466 			data = pnfs_bucket_fetch_commitdata(bucket, cinfo);
467 			if (!data)
468 				goto out_error;
469 			data->ds_commit_index = i;
470 			list_add_tail(&data->list, list);
471 			atomic_inc(&cinfo->mds->rpcs_out);
472 			nreq++;
473 		}
474 		mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
475 	}
476 	return nreq;
477 out_error:
478 	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
479 	/* Clean up on error */
480 	pnfs_generic_retry_commit(buckets, nbuckets, cinfo, i);
481 	return nreq;
482 }
483 
484 static unsigned int
485 pnfs_alloc_ds_commits_list(struct list_head *list,
486 			   struct pnfs_ds_commit_info *fl_cinfo,
487 			   struct nfs_commit_info *cinfo)
488 {
489 	struct pnfs_commit_array *array;
490 	unsigned int ret = 0;
491 
492 	rcu_read_lock();
493 	list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
494 		if (!array->lseg || !pnfs_get_commit_array(array))
495 			continue;
496 		rcu_read_unlock();
497 		ret += pnfs_bucket_alloc_ds_commits(list, array->buckets,
498 				array->nbuckets, cinfo);
499 		rcu_read_lock();
500 		pnfs_put_commit_array(array, cinfo->inode);
501 	}
502 	rcu_read_unlock();
503 	return ret;
504 }
505 
506 /* This follows nfs_commit_list pretty closely */
507 int
508 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
509 			     int how, struct nfs_commit_info *cinfo,
510 			     int (*initiate_commit)(struct nfs_commit_data *data,
511 						    int how))
512 {
513 	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
514 	struct nfs_commit_data *data, *tmp;
515 	LIST_HEAD(list);
516 	unsigned int nreq = 0;
517 
518 	if (!list_empty(mds_pages)) {
519 		data = nfs_commitdata_alloc(true);
520 		data->ds_commit_index = -1;
521 		list_splice_init(mds_pages, &data->pages);
522 		list_add_tail(&data->list, &list);
523 		atomic_inc(&cinfo->mds->rpcs_out);
524 		nreq++;
525 	}
526 
527 	nreq += pnfs_alloc_ds_commits_list(&list, fl_cinfo, cinfo);
528 	if (nreq == 0)
529 		goto out;
530 
531 	list_for_each_entry_safe(data, tmp, &list, list) {
532 		list_del(&data->list);
533 		if (data->ds_commit_index < 0) {
534 			nfs_init_commit(data, NULL, NULL, cinfo);
535 			nfs_initiate_commit(NFS_CLIENT(inode), data,
536 					    NFS_PROTO(data->inode),
537 					    data->mds_ops, how,
538 					    RPC_TASK_CRED_NOREF);
539 		} else {
540 			nfs_init_commit(data, NULL, data->lseg, cinfo);
541 			initiate_commit(data, how);
542 		}
543 	}
544 out:
545 	return PNFS_ATTEMPTED;
546 }
547 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
548 
549 /*
550  * Data server cache
551  *
552  * Data servers can be mapped to different device ids.
553  * nfs4_pnfs_ds reference counting
554  *   - set to 1 on allocation
555  *   - incremented when a device id maps a data server already in the cache.
556  *   - decremented when deviceid is removed from the cache.
557  */
558 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
559 static LIST_HEAD(nfs4_data_server_cache);
560 
561 /* Debug routines */
562 static void
563 print_ds(struct nfs4_pnfs_ds *ds)
564 {
565 	if (ds == NULL) {
566 		printk(KERN_WARNING "%s NULL device\n", __func__);
567 		return;
568 	}
569 	printk(KERN_WARNING "        ds %s\n"
570 		"        ref count %d\n"
571 		"        client %p\n"
572 		"        cl_exchange_flags %x\n",
573 		ds->ds_remotestr,
574 		refcount_read(&ds->ds_count), ds->ds_clp,
575 		ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
576 }
577 
578 static bool
579 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
580 {
581 	struct sockaddr_in *a, *b;
582 	struct sockaddr_in6 *a6, *b6;
583 
584 	if (addr1->sa_family != addr2->sa_family)
585 		return false;
586 
587 	switch (addr1->sa_family) {
588 	case AF_INET:
589 		a = (struct sockaddr_in *)addr1;
590 		b = (struct sockaddr_in *)addr2;
591 
592 		if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
593 		    a->sin_port == b->sin_port)
594 			return true;
595 		break;
596 
597 	case AF_INET6:
598 		a6 = (struct sockaddr_in6 *)addr1;
599 		b6 = (struct sockaddr_in6 *)addr2;
600 
601 		/* LINKLOCAL addresses must have matching scope_id */
602 		if (ipv6_addr_src_scope(&a6->sin6_addr) ==
603 		    IPV6_ADDR_SCOPE_LINKLOCAL &&
604 		    a6->sin6_scope_id != b6->sin6_scope_id)
605 			return false;
606 
607 		if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
608 		    a6->sin6_port == b6->sin6_port)
609 			return true;
610 		break;
611 
612 	default:
613 		dprintk("%s: unhandled address family: %u\n",
614 			__func__, addr1->sa_family);
615 		return false;
616 	}
617 
618 	return false;
619 }
620 
621 /*
622  * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
623  * declare a match.
624  */
625 static bool
626 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
627 			       const struct list_head *dsaddrs2)
628 {
629 	struct nfs4_pnfs_ds_addr *da1, *da2;
630 	struct sockaddr *sa1, *sa2;
631 	bool match = false;
632 
633 	list_for_each_entry(da1, dsaddrs1, da_node) {
634 		sa1 = (struct sockaddr *)&da1->da_addr;
635 		match = false;
636 		list_for_each_entry(da2, dsaddrs2, da_node) {
637 			sa2 = (struct sockaddr *)&da2->da_addr;
638 			match = same_sockaddr(sa1, sa2);
639 			if (match)
640 				break;
641 		}
642 		if (!match)
643 			break;
644 	}
645 	return match;
646 }
647 
648 /*
649  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
650  */
651 static struct nfs4_pnfs_ds *
652 _data_server_lookup_locked(const struct list_head *dsaddrs)
653 {
654 	struct nfs4_pnfs_ds *ds;
655 
656 	list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
657 		if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
658 			return ds;
659 	return NULL;
660 }
661 
662 static struct nfs4_pnfs_ds_addr *nfs4_pnfs_ds_addr_alloc(gfp_t gfp_flags)
663 {
664 	struct nfs4_pnfs_ds_addr *da = kzalloc(sizeof(*da), gfp_flags);
665 	if (da)
666 		INIT_LIST_HEAD(&da->da_node);
667 	return da;
668 }
669 
670 static void nfs4_pnfs_ds_addr_free(struct nfs4_pnfs_ds_addr *da)
671 {
672 	kfree(da->da_remotestr);
673 	kfree(da->da_netid);
674 	kfree(da);
675 }
676 
677 static void destroy_ds(struct nfs4_pnfs_ds *ds)
678 {
679 	struct nfs4_pnfs_ds_addr *da;
680 
681 	dprintk("--> %s\n", __func__);
682 	ifdebug(FACILITY)
683 		print_ds(ds);
684 
685 	nfs_put_client(ds->ds_clp);
686 
687 	while (!list_empty(&ds->ds_addrs)) {
688 		da = list_first_entry(&ds->ds_addrs,
689 				      struct nfs4_pnfs_ds_addr,
690 				      da_node);
691 		list_del_init(&da->da_node);
692 		nfs4_pnfs_ds_addr_free(da);
693 	}
694 
695 	kfree(ds->ds_remotestr);
696 	kfree(ds);
697 }
698 
699 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
700 {
701 	if (refcount_dec_and_lock(&ds->ds_count,
702 				&nfs4_ds_cache_lock)) {
703 		list_del_init(&ds->ds_node);
704 		spin_unlock(&nfs4_ds_cache_lock);
705 		destroy_ds(ds);
706 	}
707 }
708 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
709 
710 /*
711  * Create a string with a human readable address and port to avoid
712  * complicated setup around many dprinks.
713  */
714 static char *
715 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
716 {
717 	struct nfs4_pnfs_ds_addr *da;
718 	char *remotestr;
719 	size_t len;
720 	char *p;
721 
722 	len = 3;        /* '{', '}' and eol */
723 	list_for_each_entry(da, dsaddrs, da_node) {
724 		len += strlen(da->da_remotestr) + 1;    /* string plus comma */
725 	}
726 
727 	remotestr = kzalloc(len, gfp_flags);
728 	if (!remotestr)
729 		return NULL;
730 
731 	p = remotestr;
732 	*(p++) = '{';
733 	len--;
734 	list_for_each_entry(da, dsaddrs, da_node) {
735 		size_t ll = strlen(da->da_remotestr);
736 
737 		if (ll > len)
738 			goto out_err;
739 
740 		memcpy(p, da->da_remotestr, ll);
741 		p += ll;
742 		len -= ll;
743 
744 		if (len < 1)
745 			goto out_err;
746 		(*p++) = ',';
747 		len--;
748 	}
749 	if (len < 2)
750 		goto out_err;
751 	*(p++) = '}';
752 	*p = '\0';
753 	return remotestr;
754 out_err:
755 	kfree(remotestr);
756 	return NULL;
757 }
758 
759 /*
760  * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
761  * uncached and return cached struct nfs4_pnfs_ds.
762  */
763 struct nfs4_pnfs_ds *
764 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
765 {
766 	struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
767 	char *remotestr;
768 
769 	if (list_empty(dsaddrs)) {
770 		dprintk("%s: no addresses defined\n", __func__);
771 		goto out;
772 	}
773 
774 	ds = kzalloc(sizeof(*ds), gfp_flags);
775 	if (!ds)
776 		goto out;
777 
778 	/* this is only used for debugging, so it's ok if its NULL */
779 	remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
780 
781 	spin_lock(&nfs4_ds_cache_lock);
782 	tmp_ds = _data_server_lookup_locked(dsaddrs);
783 	if (tmp_ds == NULL) {
784 		INIT_LIST_HEAD(&ds->ds_addrs);
785 		list_splice_init(dsaddrs, &ds->ds_addrs);
786 		ds->ds_remotestr = remotestr;
787 		refcount_set(&ds->ds_count, 1);
788 		INIT_LIST_HEAD(&ds->ds_node);
789 		ds->ds_clp = NULL;
790 		list_add(&ds->ds_node, &nfs4_data_server_cache);
791 		dprintk("%s add new data server %s\n", __func__,
792 			ds->ds_remotestr);
793 	} else {
794 		kfree(remotestr);
795 		kfree(ds);
796 		refcount_inc(&tmp_ds->ds_count);
797 		dprintk("%s data server %s found, inc'ed ds_count to %d\n",
798 			__func__, tmp_ds->ds_remotestr,
799 			refcount_read(&tmp_ds->ds_count));
800 		ds = tmp_ds;
801 	}
802 	spin_unlock(&nfs4_ds_cache_lock);
803 out:
804 	return ds;
805 }
806 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
807 
808 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
809 {
810 	might_sleep();
811 	wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
812 			TASK_KILLABLE);
813 }
814 
815 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
816 {
817 	smp_mb__before_atomic();
818 	clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
819 	smp_mb__after_atomic();
820 	wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
821 }
822 
823 static struct nfs_client *(*get_v3_ds_connect)(
824 			struct nfs_server *mds_srv,
825 			const struct sockaddr *ds_addr,
826 			int ds_addrlen,
827 			int ds_proto,
828 			unsigned int ds_timeo,
829 			unsigned int ds_retrans);
830 
831 static bool load_v3_ds_connect(void)
832 {
833 	if (!get_v3_ds_connect) {
834 		get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
835 		WARN_ON_ONCE(!get_v3_ds_connect);
836 	}
837 
838 	return(get_v3_ds_connect != NULL);
839 }
840 
841 void nfs4_pnfs_v3_ds_connect_unload(void)
842 {
843 	if (get_v3_ds_connect) {
844 		symbol_put(nfs3_set_ds_client);
845 		get_v3_ds_connect = NULL;
846 	}
847 }
848 
849 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
850 				 struct nfs4_pnfs_ds *ds,
851 				 unsigned int timeo,
852 				 unsigned int retrans)
853 {
854 	struct nfs_client *clp = ERR_PTR(-EIO);
855 	struct nfs4_pnfs_ds_addr *da;
856 	int status = 0;
857 
858 	dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
859 
860 	if (!load_v3_ds_connect())
861 		goto out;
862 
863 	list_for_each_entry(da, &ds->ds_addrs, da_node) {
864 		dprintk("%s: DS %s: trying address %s\n",
865 			__func__, ds->ds_remotestr, da->da_remotestr);
866 
867 		if (!IS_ERR(clp)) {
868 			struct xprt_create xprt_args = {
869 				.ident = da->da_transport,
870 				.net = clp->cl_net,
871 				.dstaddr = (struct sockaddr *)&da->da_addr,
872 				.addrlen = da->da_addrlen,
873 				.servername = clp->cl_hostname,
874 			};
875 
876 			if (da->da_transport != clp->cl_proto)
877 				continue;
878 			if (da->da_addr.ss_family != clp->cl_addr.ss_family)
879 				continue;
880 			/* Add this address as an alias */
881 			rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
882 					rpc_clnt_test_and_add_xprt, NULL);
883 			continue;
884 		}
885 		clp = get_v3_ds_connect(mds_srv,
886 				(struct sockaddr *)&da->da_addr,
887 				da->da_addrlen, da->da_transport,
888 				timeo, retrans);
889 		if (IS_ERR(clp))
890 			continue;
891 		clp->cl_rpcclient->cl_softerr = 0;
892 		clp->cl_rpcclient->cl_softrtry = 0;
893 	}
894 
895 	if (IS_ERR(clp)) {
896 		status = PTR_ERR(clp);
897 		goto out;
898 	}
899 
900 	smp_wmb();
901 	ds->ds_clp = clp;
902 	dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
903 out:
904 	return status;
905 }
906 
907 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
908 				 struct nfs4_pnfs_ds *ds,
909 				 unsigned int timeo,
910 				 unsigned int retrans,
911 				 u32 minor_version)
912 {
913 	struct nfs_client *clp = ERR_PTR(-EIO);
914 	struct nfs4_pnfs_ds_addr *da;
915 	int status = 0;
916 
917 	dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
918 
919 	list_for_each_entry(da, &ds->ds_addrs, da_node) {
920 		dprintk("%s: DS %s: trying address %s\n",
921 			__func__, ds->ds_remotestr, da->da_remotestr);
922 
923 		if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) {
924 			struct xprt_create xprt_args = {
925 				.ident = da->da_transport,
926 				.net = clp->cl_net,
927 				.dstaddr = (struct sockaddr *)&da->da_addr,
928 				.addrlen = da->da_addrlen,
929 				.servername = clp->cl_hostname,
930 			};
931 			struct nfs4_add_xprt_data xprtdata = {
932 				.clp = clp,
933 			};
934 			struct rpc_add_xprt_test rpcdata = {
935 				.add_xprt_test = clp->cl_mvops->session_trunk,
936 				.data = &xprtdata,
937 			};
938 
939 			if (da->da_transport != clp->cl_proto)
940 				continue;
941 			if (da->da_addr.ss_family != clp->cl_addr.ss_family)
942 				continue;
943 			/**
944 			* Test this address for session trunking and
945 			* add as an alias
946 			*/
947 			xprtdata.cred = nfs4_get_clid_cred(clp),
948 			rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
949 					  rpc_clnt_setup_test_and_add_xprt,
950 					  &rpcdata);
951 			if (xprtdata.cred)
952 				put_cred(xprtdata.cred);
953 		} else {
954 			clp = nfs4_set_ds_client(mds_srv,
955 						(struct sockaddr *)&da->da_addr,
956 						da->da_addrlen,
957 						da->da_transport, timeo,
958 						retrans, minor_version);
959 			if (IS_ERR(clp))
960 				continue;
961 
962 			status = nfs4_init_ds_session(clp,
963 					mds_srv->nfs_client->cl_lease_time);
964 			if (status) {
965 				nfs_put_client(clp);
966 				clp = ERR_PTR(-EIO);
967 				continue;
968 			}
969 
970 		}
971 	}
972 
973 	if (IS_ERR(clp)) {
974 		status = PTR_ERR(clp);
975 		goto out;
976 	}
977 
978 	smp_wmb();
979 	ds->ds_clp = clp;
980 	dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
981 out:
982 	return status;
983 }
984 
985 /*
986  * Create an rpc connection to the nfs4_pnfs_ds data server.
987  * Currently only supports IPv4 and IPv6 addresses.
988  * If connection fails, make devid unavailable and return a -errno.
989  */
990 int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
991 			  struct nfs4_deviceid_node *devid, unsigned int timeo,
992 			  unsigned int retrans, u32 version, u32 minor_version)
993 {
994 	int err;
995 
996 again:
997 	err = 0;
998 	if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
999 		if (version == 3) {
1000 			err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
1001 						       retrans);
1002 		} else if (version == 4) {
1003 			err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
1004 						       retrans, minor_version);
1005 		} else {
1006 			dprintk("%s: unsupported DS version %d\n", __func__,
1007 				version);
1008 			err = -EPROTONOSUPPORT;
1009 		}
1010 
1011 		nfs4_clear_ds_conn_bit(ds);
1012 	} else {
1013 		nfs4_wait_ds_connect(ds);
1014 
1015 		/* what was waited on didn't connect AND didn't mark unavail */
1016 		if (!ds->ds_clp && !nfs4_test_deviceid_unavailable(devid))
1017 			goto again;
1018 	}
1019 
1020 	/*
1021 	 * At this point the ds->ds_clp should be ready, but it might have
1022 	 * hit an error.
1023 	 */
1024 	if (!err) {
1025 		if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) {
1026 			WARN_ON_ONCE(ds->ds_clp ||
1027 				!nfs4_test_deviceid_unavailable(devid));
1028 			return -EINVAL;
1029 		}
1030 		err = nfs_client_init_status(ds->ds_clp);
1031 	}
1032 
1033 	return err;
1034 }
1035 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
1036 
1037 /*
1038  * Currently only supports ipv4, ipv6 and one multi-path address.
1039  */
1040 struct nfs4_pnfs_ds_addr *
1041 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
1042 {
1043 	struct nfs4_pnfs_ds_addr *da = NULL;
1044 	char *buf, *portstr;
1045 	__be16 port;
1046 	ssize_t nlen, rlen;
1047 	int tmp[2];
1048 	char *netid;
1049 	size_t len;
1050 	char *startsep = "";
1051 	char *endsep = "";
1052 
1053 
1054 	/* r_netid */
1055 	nlen = xdr_stream_decode_string_dup(xdr, &netid, XDR_MAX_NETOBJ,
1056 					    gfp_flags);
1057 	if (unlikely(nlen < 0))
1058 		goto out_err;
1059 
1060 	/* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
1061 	/* port is ".ABC.DEF", 8 chars max */
1062 	rlen = xdr_stream_decode_string_dup(xdr, &buf, INET6_ADDRSTRLEN +
1063 					    IPV6_SCOPE_ID_LEN + 8, gfp_flags);
1064 	if (unlikely(rlen < 0))
1065 		goto out_free_netid;
1066 
1067 	/* replace port '.' with '-' */
1068 	portstr = strrchr(buf, '.');
1069 	if (!portstr) {
1070 		dprintk("%s: Failed finding expected dot in port\n",
1071 			__func__);
1072 		goto out_free_buf;
1073 	}
1074 	*portstr = '-';
1075 
1076 	/* find '.' between address and port */
1077 	portstr = strrchr(buf, '.');
1078 	if (!portstr) {
1079 		dprintk("%s: Failed finding expected dot between address and "
1080 			"port\n", __func__);
1081 		goto out_free_buf;
1082 	}
1083 	*portstr = '\0';
1084 
1085 	da = nfs4_pnfs_ds_addr_alloc(gfp_flags);
1086 	if (unlikely(!da))
1087 		goto out_free_buf;
1088 
1089 	if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
1090 		      sizeof(da->da_addr))) {
1091 		dprintk("%s: error parsing address %s\n", __func__, buf);
1092 		goto out_free_da;
1093 	}
1094 
1095 	portstr++;
1096 	sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
1097 	port = htons((tmp[0] << 8) | (tmp[1]));
1098 
1099 	switch (da->da_addr.ss_family) {
1100 	case AF_INET:
1101 		((struct sockaddr_in *)&da->da_addr)->sin_port = port;
1102 		da->da_addrlen = sizeof(struct sockaddr_in);
1103 		break;
1104 
1105 	case AF_INET6:
1106 		((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
1107 		da->da_addrlen = sizeof(struct sockaddr_in6);
1108 		startsep = "[";
1109 		endsep = "]";
1110 		break;
1111 
1112 	default:
1113 		dprintk("%s: unsupported address family: %u\n",
1114 			__func__, da->da_addr.ss_family);
1115 		goto out_free_da;
1116 	}
1117 
1118 	da->da_transport = xprt_find_transport_ident(netid);
1119 	if (da->da_transport < 0) {
1120 		dprintk("%s: ERROR: unknown r_netid \"%s\"\n",
1121 			__func__, netid);
1122 		goto out_free_da;
1123 	}
1124 
1125 	da->da_netid = netid;
1126 
1127 	/* save human readable address */
1128 	len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
1129 	da->da_remotestr = kzalloc(len, gfp_flags);
1130 
1131 	/* NULL is ok, only used for dprintk */
1132 	if (da->da_remotestr)
1133 		snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
1134 			 buf, endsep, ntohs(port));
1135 
1136 	dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
1137 	kfree(buf);
1138 	return da;
1139 
1140 out_free_da:
1141 	kfree(da);
1142 out_free_buf:
1143 	dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
1144 	kfree(buf);
1145 out_free_netid:
1146 	kfree(netid);
1147 out_err:
1148 	return NULL;
1149 }
1150 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
1151 
1152 void
1153 pnfs_layout_mark_request_commit(struct nfs_page *req,
1154 				struct pnfs_layout_segment *lseg,
1155 				struct nfs_commit_info *cinfo,
1156 				u32 ds_commit_idx)
1157 {
1158 	struct list_head *list;
1159 	struct pnfs_commit_array *array;
1160 	struct pnfs_commit_bucket *bucket;
1161 
1162 	mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
1163 	array = pnfs_lookup_commit_array(cinfo->ds, lseg);
1164 	if (!array || !pnfs_is_valid_lseg(lseg))
1165 		goto out_resched;
1166 	bucket = &array->buckets[ds_commit_idx];
1167 	list = &bucket->written;
1168 	/* Non-empty buckets hold a reference on the lseg.  That ref
1169 	 * is normally transferred to the COMMIT call and released
1170 	 * there.  It could also be released if the last req is pulled
1171 	 * off due to a rewrite, in which case it will be done in
1172 	 * pnfs_common_clear_request_commit
1173 	 */
1174 	if (!bucket->lseg)
1175 		bucket->lseg = pnfs_get_lseg(lseg);
1176 	set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1177 	cinfo->ds->nwritten++;
1178 
1179 	nfs_request_add_commit_list_locked(req, list, cinfo);
1180 	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1181 	nfs_mark_page_unstable(req->wb_page, cinfo);
1182 	return;
1183 out_resched:
1184 	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1185 	cinfo->completion_ops->resched_write(cinfo, req);
1186 }
1187 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
1188 
1189 int
1190 pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
1191 {
1192 	int ret;
1193 
1194 	if (!pnfs_layoutcommit_outstanding(inode))
1195 		return 0;
1196 	ret = nfs_commit_inode(inode, FLUSH_SYNC);
1197 	if (ret < 0)
1198 		return ret;
1199 	if (datasync)
1200 		return 0;
1201 	return pnfs_layoutcommit_inode(inode, true);
1202 }
1203 EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
1204 
1205