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
22void 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}
29EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
30
31/* Fake up some data that will cause nfs_commit_release to retry the writes. */
32void 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}
40EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes);
41
42void 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}
49EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done);
50
51void 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(lseg: data->lseg);
57 nfs_put_client(data->ds_clp);
58 nfs_commitdata_release(data);
59}
60EXPORT_SYMBOL_GPL(pnfs_generic_commit_release);
61
62static struct pnfs_layout_segment *
63pnfs_free_bucket_lseg(struct pnfs_commit_bucket *bucket)
64{
65 if (list_empty(head: &bucket->committing) && list_empty(head: &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 */
77void
78pnfs_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(nr: PG_COMMIT_TO_DS, addr: &req->wb_flags))
84 goto out;
85 cinfo->ds->nwritten--;
86 if (list_is_singular(head: &req->wb_list))
87 bucket = list_first_entry(&req->wb_list,
88 struct pnfs_commit_bucket, written);
89out:
90 nfs_request_remove_commit_list(req, cinfo);
91 if (bucket)
92 pnfs_put_lseg(lseg: pnfs_free_bucket_lseg(bucket));
93}
94EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
95
96struct pnfs_commit_array *
97pnfs_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), flags: gfp_flags);
103 if (!p)
104 return NULL;
105 p->nbuckets = n;
106 INIT_LIST_HEAD(list: &p->cinfo_list);
107 INIT_LIST_HEAD(list: &p->lseg_list);
108 p->lseg = NULL;
109 for (b = &p->buckets[0]; n != 0; b++, n--) {
110 INIT_LIST_HEAD(list: &b->written);
111 INIT_LIST_HEAD(list: &b->committing);
112 b->lseg = NULL;
113 b->direct_verf.committed = NFS_INVALID_STABLE_HOW;
114 }
115 return p;
116}
117EXPORT_SYMBOL_GPL(pnfs_alloc_commit_array);
118
119void
120pnfs_free_commit_array(struct pnfs_commit_array *p)
121{
122 kfree_rcu(p, rcu);
123}
124EXPORT_SYMBOL_GPL(pnfs_free_commit_array);
125
126static struct pnfs_commit_array *
127pnfs_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
139struct pnfs_commit_array *
140pnfs_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(r: &new->refcount, n: 1);
151 list_add_rcu(new: &new->cinfo_list, head: &fl_cinfo->commits);
152 list_add(new: &new->lseg_list, head: &lseg->pls_commits);
153 return new;
154}
155EXPORT_SYMBOL_GPL(pnfs_add_commit_array);
156
157static struct pnfs_commit_array *
158pnfs_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
175static void
176pnfs_release_commit_array_locked(struct pnfs_commit_array *array)
177{
178 list_del_rcu(entry: &array->cinfo_list);
179 list_del(entry: &array->lseg_list);
180 pnfs_free_commit_array(array);
181}
182
183static void
184pnfs_put_commit_array_locked(struct pnfs_commit_array *array)
185{
186 if (refcount_dec_and_test(r: &array->refcount))
187 pnfs_release_commit_array_locked(array);
188}
189
190static void
191pnfs_put_commit_array(struct pnfs_commit_array *array, struct inode *inode)
192{
193 if (refcount_dec_and_lock(r: &array->refcount, lock: &inode->i_lock)) {
194 pnfs_release_commit_array_locked(array);
195 spin_unlock(lock: &inode->i_lock);
196 }
197}
198
199static struct pnfs_commit_array *
200pnfs_get_commit_array(struct pnfs_commit_array *array)
201{
202 if (refcount_inc_not_zero(r: &array->refcount))
203 return array;
204 return NULL;
205}
206
207static void
208pnfs_remove_and_free_commit_array(struct pnfs_commit_array *array)
209{
210 array->lseg = NULL;
211 list_del_init(entry: &array->lseg_list);
212 pnfs_put_commit_array_locked(array);
213}
214
215void
216pnfs_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}
224EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_release_lseg);
225
226void
227pnfs_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}
234EXPORT_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 */
240static int
241pnfs_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
258static 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(bucket: &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 */
277int 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, buckets: array->buckets,
289 nbuckets: array->nbuckets, max);
290 rcu_read_lock();
291 pnfs_put_commit_array(array, inode: cinfo->inode);
292 rv += cnt;
293 max -= cnt;
294 if (!max)
295 break;
296 }
297 rcu_read_unlock();
298 return rv;
299}
300EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
301
302static unsigned int
303pnfs_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
313restart:
314 for (i = 0, b = buckets; i < nbuckets; i++, b++) {
315 nwritten = nfs_scan_commit_list(src: &b->written, dst, cinfo, max: 0);
316 if (!nwritten)
317 continue;
318 ret += nwritten;
319 freeme = pnfs_free_bucket_lseg(bucket: b);
320 if (freeme) {
321 pnfs_put_lseg(lseg: freeme);
322 goto restart;
323 }
324 }
325 return ret;
326}
327
328/* Pull everything off the committing lists and dump into @dst. */
329void 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 buckets: array->buckets,
344 nbuckets: array->nbuckets,
345 cinfo);
346 rcu_read_lock();
347 pnfs_put_commit_array(array, inode: cinfo->inode);
348 fl_cinfo->nwritten -= nwritten;
349 }
350 rcu_read_unlock();
351}
352EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
353
354static struct nfs_page *
355pnfs_bucket_search_commit_reqs(struct pnfs_commit_bucket *buckets,
356 unsigned int nbuckets, struct folio *folio)
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 (nfs_page_to_folio(req) == folio)
367 return req->wb_head;
368 }
369 list_for_each_entry(req, &b->committing, wb_list) {
370 if (nfs_page_to_folio(req) == folio)
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 request
378 * for @folio
379 * @cinfo - commit info for current inode
380 * @folio - page to search for matching head request
381 *
382 * Return: the head request if one is found, otherwise %NULL.
383 */
384struct nfs_page *pnfs_generic_search_commit_reqs(struct nfs_commit_info *cinfo,
385 struct folio *folio)
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(buckets: array->buckets,
393 nbuckets: array->nbuckets, folio);
394 if (req)
395 return req;
396 }
397 return NULL;
398}
399EXPORT_SYMBOL_GPL(pnfs_generic_search_commit_reqs);
400
401static struct pnfs_layout_segment *
402pnfs_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(list: &bucket->committing, head);
412 lseg = pnfs_free_bucket_lseg(bucket);
413 if (!lseg)
414 lseg = pnfs_get_lseg(lseg: bucket->lseg);
415 return lseg;
416}
417
418static struct nfs_commit_data *
419pnfs_bucket_fetch_commitdata(struct pnfs_commit_bucket *bucket,
420 struct nfs_commit_info *cinfo)
421{
422 struct nfs_commit_data *data = nfs_commitdata_alloc();
423
424 if (!data)
425 return NULL;
426 data->lseg = pnfs_bucket_get_committing(head: &data->pages, bucket, cinfo);
427 return data;
428}
429
430static 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(head: &bucket->committing))
441 continue;
442 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
443 freeme = pnfs_bucket_get_committing(head: &pages, bucket, cinfo);
444 mutex_unlock(lock: &NFS_I(inode: cinfo->inode)->commit_mutex);
445 nfs_retry_commit(page_list: &pages, lseg: freeme, cinfo, ds_commit_idx: idx);
446 pnfs_put_lseg(lseg: freeme);
447 }
448}
449
450static unsigned int
451pnfs_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(head: &bucket->committing))
463 continue;
464 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
465 if (!list_empty(head: &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(new: &data->list, head: list);
471 nreq++;
472 }
473 mutex_unlock(lock: &NFS_I(inode: cinfo->inode)->commit_mutex);
474 }
475 return nreq;
476out_error:
477 mutex_unlock(lock: &NFS_I(inode: cinfo->inode)->commit_mutex);
478 /* Clean up on error */
479 pnfs_generic_retry_commit(buckets, nbuckets, cinfo, idx: i);
480 return nreq;
481}
482
483static unsigned int
484pnfs_alloc_ds_commits_list(struct list_head *list,
485 struct pnfs_ds_commit_info *fl_cinfo,
486 struct nfs_commit_info *cinfo)
487{
488 struct pnfs_commit_array *array;
489 unsigned int ret = 0;
490
491 rcu_read_lock();
492 list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
493 if (!array->lseg || !pnfs_get_commit_array(array))
494 continue;
495 rcu_read_unlock();
496 ret += pnfs_bucket_alloc_ds_commits(list, buckets: array->buckets,
497 nbuckets: array->nbuckets, cinfo);
498 rcu_read_lock();
499 pnfs_put_commit_array(array, inode: cinfo->inode);
500 }
501 rcu_read_unlock();
502 return ret;
503}
504
505/* This follows nfs_commit_list pretty closely */
506int
507pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
508 int how, struct nfs_commit_info *cinfo,
509 int (*initiate_commit)(struct nfs_commit_data *data,
510 int how))
511{
512 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
513 struct nfs_commit_data *data, *tmp;
514 LIST_HEAD(list);
515 unsigned int nreq = 0;
516
517 if (!list_empty(head: mds_pages)) {
518 data = nfs_commitdata_alloc();
519 if (!data) {
520 nfs_retry_commit(page_list: mds_pages, NULL, cinfo, ds_commit_idx: -1);
521 return -ENOMEM;
522 }
523 data->ds_commit_index = -1;
524 list_splice_init(list: mds_pages, head: &data->pages);
525 list_add_tail(new: &data->list, head: &list);
526 nreq++;
527 }
528
529 nreq += pnfs_alloc_ds_commits_list(list: &list, fl_cinfo, cinfo);
530 if (nreq == 0)
531 goto out;
532
533 list_for_each_entry_safe(data, tmp, &list, list) {
534 list_del(entry: &data->list);
535 if (data->ds_commit_index < 0) {
536 nfs_init_commit(data, NULL, NULL, cinfo);
537 nfs_initiate_commit(clnt: NFS_CLIENT(inode), data,
538 nfs_ops: NFS_PROTO(inode: data->inode),
539 call_ops: data->mds_ops, how,
540 RPC_TASK_CRED_NOREF);
541 } else {
542 nfs_init_commit(data, NULL, lseg: data->lseg, cinfo);
543 initiate_commit(data, how);
544 }
545 }
546out:
547 return PNFS_ATTEMPTED;
548}
549EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
550
551/*
552 * Data server cache
553 *
554 * Data servers can be mapped to different device ids.
555 * nfs4_pnfs_ds reference counting
556 * - set to 1 on allocation
557 * - incremented when a device id maps a data server already in the cache.
558 * - decremented when deviceid is removed from the cache.
559 */
560static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
561static LIST_HEAD(nfs4_data_server_cache);
562
563/* Debug routines */
564static void
565print_ds(struct nfs4_pnfs_ds *ds)
566{
567 if (ds == NULL) {
568 printk(KERN_WARNING "%s NULL device\n", __func__);
569 return;
570 }
571 printk(KERN_WARNING " ds %s\n"
572 " ref count %d\n"
573 " client %p\n"
574 " cl_exchange_flags %x\n",
575 ds->ds_remotestr,
576 refcount_read(&ds->ds_count), ds->ds_clp,
577 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
578}
579
580static bool
581same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
582{
583 struct sockaddr_in *a, *b;
584 struct sockaddr_in6 *a6, *b6;
585
586 if (addr1->sa_family != addr2->sa_family)
587 return false;
588
589 switch (addr1->sa_family) {
590 case AF_INET:
591 a = (struct sockaddr_in *)addr1;
592 b = (struct sockaddr_in *)addr2;
593
594 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
595 a->sin_port == b->sin_port)
596 return true;
597 break;
598
599 case AF_INET6:
600 a6 = (struct sockaddr_in6 *)addr1;
601 b6 = (struct sockaddr_in6 *)addr2;
602
603 /* LINKLOCAL addresses must have matching scope_id */
604 if (ipv6_addr_src_scope(addr: &a6->sin6_addr) ==
605 IPV6_ADDR_SCOPE_LINKLOCAL &&
606 a6->sin6_scope_id != b6->sin6_scope_id)
607 return false;
608
609 if (ipv6_addr_equal(a1: &a6->sin6_addr, a2: &b6->sin6_addr) &&
610 a6->sin6_port == b6->sin6_port)
611 return true;
612 break;
613
614 default:
615 dprintk("%s: unhandled address family: %u\n",
616 __func__, addr1->sa_family);
617 return false;
618 }
619
620 return false;
621}
622
623/*
624 * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
625 * declare a match.
626 */
627static bool
628_same_data_server_addrs_locked(const struct list_head *dsaddrs1,
629 const struct list_head *dsaddrs2)
630{
631 struct nfs4_pnfs_ds_addr *da1, *da2;
632 struct sockaddr *sa1, *sa2;
633 bool match = false;
634
635 list_for_each_entry(da1, dsaddrs1, da_node) {
636 sa1 = (struct sockaddr *)&da1->da_addr;
637 match = false;
638 list_for_each_entry(da2, dsaddrs2, da_node) {
639 sa2 = (struct sockaddr *)&da2->da_addr;
640 match = same_sockaddr(addr1: sa1, addr2: sa2);
641 if (match)
642 break;
643 }
644 if (!match)
645 break;
646 }
647 return match;
648}
649
650/*
651 * Lookup DS by addresses. nfs4_ds_cache_lock is held
652 */
653static struct nfs4_pnfs_ds *
654_data_server_lookup_locked(const struct list_head *dsaddrs)
655{
656 struct nfs4_pnfs_ds *ds;
657
658 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
659 if (_same_data_server_addrs_locked(dsaddrs1: &ds->ds_addrs, dsaddrs2: dsaddrs))
660 return ds;
661 return NULL;
662}
663
664static struct nfs4_pnfs_ds_addr *nfs4_pnfs_ds_addr_alloc(gfp_t gfp_flags)
665{
666 struct nfs4_pnfs_ds_addr *da = kzalloc(size: sizeof(*da), flags: gfp_flags);
667 if (da)
668 INIT_LIST_HEAD(list: &da->da_node);
669 return da;
670}
671
672static void nfs4_pnfs_ds_addr_free(struct nfs4_pnfs_ds_addr *da)
673{
674 kfree(objp: da->da_remotestr);
675 kfree(objp: da->da_netid);
676 kfree(objp: da);
677}
678
679static void destroy_ds(struct nfs4_pnfs_ds *ds)
680{
681 struct nfs4_pnfs_ds_addr *da;
682
683 dprintk("--> %s\n", __func__);
684 ifdebug(FACILITY)
685 print_ds(ds);
686
687 nfs_put_client(ds->ds_clp);
688
689 while (!list_empty(head: &ds->ds_addrs)) {
690 da = list_first_entry(&ds->ds_addrs,
691 struct nfs4_pnfs_ds_addr,
692 da_node);
693 list_del_init(entry: &da->da_node);
694 nfs4_pnfs_ds_addr_free(da);
695 }
696
697 kfree(objp: ds->ds_remotestr);
698 kfree(objp: ds);
699}
700
701void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
702{
703 if (refcount_dec_and_lock(r: &ds->ds_count,
704 lock: &nfs4_ds_cache_lock)) {
705 list_del_init(entry: &ds->ds_node);
706 spin_unlock(lock: &nfs4_ds_cache_lock);
707 destroy_ds(ds);
708 }
709}
710EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
711
712/*
713 * Create a string with a human readable address and port to avoid
714 * complicated setup around many dprinks.
715 */
716static char *
717nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
718{
719 struct nfs4_pnfs_ds_addr *da;
720 char *remotestr;
721 size_t len;
722 char *p;
723
724 len = 3; /* '{', '}' and eol */
725 list_for_each_entry(da, dsaddrs, da_node) {
726 len += strlen(da->da_remotestr) + 1; /* string plus comma */
727 }
728
729 remotestr = kzalloc(size: len, flags: gfp_flags);
730 if (!remotestr)
731 return NULL;
732
733 p = remotestr;
734 *(p++) = '{';
735 len--;
736 list_for_each_entry(da, dsaddrs, da_node) {
737 size_t ll = strlen(da->da_remotestr);
738
739 if (ll > len)
740 goto out_err;
741
742 memcpy(p, da->da_remotestr, ll);
743 p += ll;
744 len -= ll;
745
746 if (len < 1)
747 goto out_err;
748 (*p++) = ',';
749 len--;
750 }
751 if (len < 2)
752 goto out_err;
753 *(p++) = '}';
754 *p = '\0';
755 return remotestr;
756out_err:
757 kfree(objp: remotestr);
758 return NULL;
759}
760
761/*
762 * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
763 * uncached and return cached struct nfs4_pnfs_ds.
764 */
765struct nfs4_pnfs_ds *
766nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
767{
768 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
769 char *remotestr;
770
771 if (list_empty(head: dsaddrs)) {
772 dprintk("%s: no addresses defined\n", __func__);
773 goto out;
774 }
775
776 ds = kzalloc(size: sizeof(*ds), flags: gfp_flags);
777 if (!ds)
778 goto out;
779
780 /* this is only used for debugging, so it's ok if its NULL */
781 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
782
783 spin_lock(lock: &nfs4_ds_cache_lock);
784 tmp_ds = _data_server_lookup_locked(dsaddrs);
785 if (tmp_ds == NULL) {
786 INIT_LIST_HEAD(list: &ds->ds_addrs);
787 list_splice_init(list: dsaddrs, head: &ds->ds_addrs);
788 ds->ds_remotestr = remotestr;
789 refcount_set(r: &ds->ds_count, n: 1);
790 INIT_LIST_HEAD(list: &ds->ds_node);
791 ds->ds_clp = NULL;
792 list_add(new: &ds->ds_node, head: &nfs4_data_server_cache);
793 dprintk("%s add new data server %s\n", __func__,
794 ds->ds_remotestr);
795 } else {
796 kfree(objp: remotestr);
797 kfree(objp: ds);
798 refcount_inc(r: &tmp_ds->ds_count);
799 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
800 __func__, tmp_ds->ds_remotestr,
801 refcount_read(&tmp_ds->ds_count));
802 ds = tmp_ds;
803 }
804 spin_unlock(lock: &nfs4_ds_cache_lock);
805out:
806 return ds;
807}
808EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
809
810static int nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
811{
812 might_sleep();
813 return wait_on_bit(word: &ds->ds_state, NFS4DS_CONNECTING, TASK_KILLABLE);
814}
815
816static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
817{
818 smp_mb__before_atomic();
819 clear_and_wake_up_bit(NFS4DS_CONNECTING, word: &ds->ds_state);
820}
821
822static struct nfs_client *(*get_v3_ds_connect)(
823 struct nfs_server *mds_srv,
824 const struct sockaddr_storage *ds_addr,
825 int ds_addrlen,
826 int ds_proto,
827 unsigned int ds_timeo,
828 unsigned int ds_retrans);
829
830static bool load_v3_ds_connect(void)
831{
832 if (!get_v3_ds_connect) {
833 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
834 WARN_ON_ONCE(!get_v3_ds_connect);
835 }
836
837 return(get_v3_ds_connect != NULL);
838}
839
840void nfs4_pnfs_v3_ds_connect_unload(void)
841{
842 if (get_v3_ds_connect) {
843 symbol_put(nfs3_set_ds_client);
844 get_v3_ds_connect = NULL;
845 }
846}
847
848static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
849 struct nfs4_pnfs_ds *ds,
850 unsigned int timeo,
851 unsigned int retrans)
852{
853 struct nfs_client *clp = ERR_PTR(error: -EIO);
854 struct nfs4_pnfs_ds_addr *da;
855 unsigned long connect_timeout = timeo * (retrans + 1) * HZ / 10;
856 int status = 0;
857
858 dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
859
860 if (!load_v3_ds_connect())
861 return -EPROTONOSUPPORT;
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(ptr: 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 .connect_timeout = connect_timeout,
875 .reconnect_timeout = connect_timeout,
876 };
877
878 if (da->da_transport != clp->cl_proto)
879 continue;
880 if (da->da_addr.ss_family != clp->cl_addr.ss_family)
881 continue;
882 /* Add this address as an alias */
883 rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
884 setup: rpc_clnt_test_and_add_xprt, NULL);
885 continue;
886 }
887 clp = get_v3_ds_connect(mds_srv,
888 &da->da_addr,
889 da->da_addrlen, da->da_transport,
890 timeo, retrans);
891 if (IS_ERR(ptr: clp))
892 continue;
893 clp->cl_rpcclient->cl_softerr = 0;
894 clp->cl_rpcclient->cl_softrtry = 0;
895 }
896
897 if (IS_ERR(ptr: clp)) {
898 status = PTR_ERR(ptr: clp);
899 goto out;
900 }
901
902 smp_wmb();
903 WRITE_ONCE(ds->ds_clp, clp);
904 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
905out:
906 return status;
907}
908
909static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
910 struct nfs4_pnfs_ds *ds,
911 unsigned int timeo,
912 unsigned int retrans,
913 u32 minor_version)
914{
915 struct nfs_client *clp = ERR_PTR(error: -EIO);
916 struct nfs4_pnfs_ds_addr *da;
917 int status = 0;
918
919 dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
920
921 list_for_each_entry(da, &ds->ds_addrs, da_node) {
922 char servername[48];
923
924 dprintk("%s: DS %s: trying address %s\n",
925 __func__, ds->ds_remotestr, da->da_remotestr);
926
927 if (!IS_ERR(ptr: clp) && clp->cl_mvops->session_trunk) {
928 struct xprt_create xprt_args = {
929 .ident = da->da_transport,
930 .net = clp->cl_net,
931 .dstaddr = (struct sockaddr *)&da->da_addr,
932 .addrlen = da->da_addrlen,
933 .servername = clp->cl_hostname,
934 .xprtsec = clp->cl_xprtsec,
935 };
936 struct nfs4_add_xprt_data xprtdata = {
937 .clp = clp,
938 };
939 struct rpc_add_xprt_test rpcdata = {
940 .add_xprt_test = clp->cl_mvops->session_trunk,
941 .data = &xprtdata,
942 };
943
944 if (da->da_transport != clp->cl_proto &&
945 clp->cl_proto != XPRT_TRANSPORT_TCP_TLS)
946 continue;
947 if (da->da_transport == XPRT_TRANSPORT_TCP &&
948 mds_srv->nfs_client->cl_proto ==
949 XPRT_TRANSPORT_TCP_TLS) {
950 struct sockaddr *addr =
951 (struct sockaddr *)&da->da_addr;
952 struct sockaddr_in *sin =
953 (struct sockaddr_in *)&da->da_addr;
954 struct sockaddr_in6 *sin6 =
955 (struct sockaddr_in6 *)&da->da_addr;
956
957 /* for NFS with TLS we need to supply a correct
958 * servername of the trunked transport, not the
959 * servername of the main transport stored in
960 * clp->cl_hostname. And set the protocol to
961 * indicate to use TLS
962 */
963 servername[0] = '\0';
964 switch(addr->sa_family) {
965 case AF_INET:
966 snprintf(buf: servername, size: sizeof(servername),
967 fmt: "%pI4", &sin->sin_addr.s_addr);
968 break;
969 case AF_INET6:
970 snprintf(buf: servername, size: sizeof(servername),
971 fmt: "%pI6", &sin6->sin6_addr);
972 break;
973 default:
974 /* do not consider this address */
975 continue;
976 }
977 xprt_args.ident = XPRT_TRANSPORT_TCP_TLS;
978 xprt_args.servername = servername;
979 }
980 if (da->da_addr.ss_family != clp->cl_addr.ss_family)
981 continue;
982
983 /**
984 * Test this address for session trunking and
985 * add as an alias
986 */
987 xprtdata.cred = nfs4_get_clid_cred(clp);
988 rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
989 setup: rpc_clnt_setup_test_and_add_xprt,
990 data: &rpcdata);
991 if (xprtdata.cred)
992 put_cred(cred: xprtdata.cred);
993 } else {
994 if (da->da_transport == XPRT_TRANSPORT_TCP &&
995 mds_srv->nfs_client->cl_proto ==
996 XPRT_TRANSPORT_TCP_TLS)
997 da->da_transport = XPRT_TRANSPORT_TCP_TLS;
998 clp = nfs4_set_ds_client(mds_srv,
999 ds_addr: &da->da_addr,
1000 ds_addrlen: da->da_addrlen,
1001 ds_proto: da->da_transport, ds_timeo: timeo,
1002 ds_retrans: retrans, minor_version);
1003 if (IS_ERR(ptr: clp))
1004 continue;
1005
1006 status = nfs4_init_ds_session(clp,
1007 mds_srv->nfs_client->cl_lease_time);
1008 if (status) {
1009 nfs_put_client(clp);
1010 clp = ERR_PTR(error: -EIO);
1011 continue;
1012 }
1013
1014 }
1015 }
1016
1017 if (IS_ERR(ptr: clp)) {
1018 status = PTR_ERR(ptr: clp);
1019 goto out;
1020 }
1021
1022 smp_wmb();
1023 WRITE_ONCE(ds->ds_clp, clp);
1024 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
1025out:
1026 return status;
1027}
1028
1029/*
1030 * Create an rpc connection to the nfs4_pnfs_ds data server.
1031 * Currently only supports IPv4 and IPv6 addresses.
1032 * If connection fails, make devid unavailable and return a -errno.
1033 */
1034int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
1035 struct nfs4_deviceid_node *devid, unsigned int timeo,
1036 unsigned int retrans, u32 version, u32 minor_version)
1037{
1038 int err;
1039
1040 do {
1041 err = nfs4_wait_ds_connect(ds);
1042 if (err || ds->ds_clp)
1043 goto out;
1044 if (nfs4_test_deviceid_unavailable(node: devid))
1045 return -ENODEV;
1046 } while (test_and_set_bit(NFS4DS_CONNECTING, addr: &ds->ds_state) != 0);
1047
1048 if (ds->ds_clp)
1049 goto connect_done;
1050
1051 switch (version) {
1052 case 3:
1053 err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo, retrans);
1054 break;
1055 case 4:
1056 err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo, retrans,
1057 minor_version);
1058 break;
1059 default:
1060 dprintk("%s: unsupported DS version %d\n", __func__, version);
1061 err = -EPROTONOSUPPORT;
1062 }
1063
1064connect_done:
1065 nfs4_clear_ds_conn_bit(ds);
1066out:
1067 /*
1068 * At this point the ds->ds_clp should be ready, but it might have
1069 * hit an error.
1070 */
1071 if (!err) {
1072 if (!ds->ds_clp || !nfs_client_init_is_complete(clp: ds->ds_clp)) {
1073 WARN_ON_ONCE(ds->ds_clp ||
1074 !nfs4_test_deviceid_unavailable(devid));
1075 return -EINVAL;
1076 }
1077 err = nfs_client_init_status(clp: ds->ds_clp);
1078 }
1079
1080 return err;
1081}
1082EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
1083
1084/*
1085 * Currently only supports ipv4, ipv6 and one multi-path address.
1086 */
1087struct nfs4_pnfs_ds_addr *
1088nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
1089{
1090 struct nfs4_pnfs_ds_addr *da = NULL;
1091 char *buf, *portstr;
1092 __be16 port;
1093 ssize_t nlen, rlen;
1094 int tmp[2];
1095 char *netid;
1096 size_t len;
1097 char *startsep = "";
1098 char *endsep = "";
1099
1100
1101 /* r_netid */
1102 nlen = xdr_stream_decode_string_dup(xdr, str: &netid, XDR_MAX_NETOBJ,
1103 gfp_flags);
1104 if (unlikely(nlen < 0))
1105 goto out_err;
1106
1107 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
1108 /* port is ".ABC.DEF", 8 chars max */
1109 rlen = xdr_stream_decode_string_dup(xdr, str: &buf, INET6_ADDRSTRLEN +
1110 IPV6_SCOPE_ID_LEN + 8, gfp_flags);
1111 if (unlikely(rlen < 0))
1112 goto out_free_netid;
1113
1114 /* replace port '.' with '-' */
1115 portstr = strrchr(buf, '.');
1116 if (!portstr) {
1117 dprintk("%s: Failed finding expected dot in port\n",
1118 __func__);
1119 goto out_free_buf;
1120 }
1121 *portstr = '-';
1122
1123 /* find '.' between address and port */
1124 portstr = strrchr(buf, '.');
1125 if (!portstr) {
1126 dprintk("%s: Failed finding expected dot between address and "
1127 "port\n", __func__);
1128 goto out_free_buf;
1129 }
1130 *portstr = '\0';
1131
1132 da = nfs4_pnfs_ds_addr_alloc(gfp_flags);
1133 if (unlikely(!da))
1134 goto out_free_buf;
1135
1136 if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
1137 sizeof(da->da_addr))) {
1138 dprintk("%s: error parsing address %s\n", __func__, buf);
1139 goto out_free_da;
1140 }
1141
1142 portstr++;
1143 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
1144 port = htons((tmp[0] << 8) | (tmp[1]));
1145
1146 switch (da->da_addr.ss_family) {
1147 case AF_INET:
1148 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
1149 da->da_addrlen = sizeof(struct sockaddr_in);
1150 break;
1151
1152 case AF_INET6:
1153 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
1154 da->da_addrlen = sizeof(struct sockaddr_in6);
1155 startsep = "[";
1156 endsep = "]";
1157 break;
1158
1159 default:
1160 dprintk("%s: unsupported address family: %u\n",
1161 __func__, da->da_addr.ss_family);
1162 goto out_free_da;
1163 }
1164
1165 da->da_transport = xprt_find_transport_ident(netid);
1166 if (da->da_transport < 0) {
1167 dprintk("%s: ERROR: unknown r_netid \"%s\"\n",
1168 __func__, netid);
1169 goto out_free_da;
1170 }
1171
1172 da->da_netid = netid;
1173
1174 /* save human readable address */
1175 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
1176 da->da_remotestr = kzalloc(size: len, flags: gfp_flags);
1177
1178 /* NULL is ok, only used for dprintk */
1179 if (da->da_remotestr)
1180 snprintf(buf: da->da_remotestr, size: len, fmt: "%s%s%s:%u", startsep,
1181 buf, endsep, ntohs(port));
1182
1183 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
1184 kfree(objp: buf);
1185 return da;
1186
1187out_free_da:
1188 kfree(objp: da);
1189out_free_buf:
1190 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
1191 kfree(objp: buf);
1192out_free_netid:
1193 kfree(objp: netid);
1194out_err:
1195 return NULL;
1196}
1197EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
1198
1199void
1200pnfs_layout_mark_request_commit(struct nfs_page *req,
1201 struct pnfs_layout_segment *lseg,
1202 struct nfs_commit_info *cinfo,
1203 u32 ds_commit_idx)
1204{
1205 struct list_head *list;
1206 struct pnfs_commit_array *array;
1207 struct pnfs_commit_bucket *bucket;
1208
1209 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
1210 array = pnfs_lookup_commit_array(fl_cinfo: cinfo->ds, lseg);
1211 if (!array || !pnfs_is_valid_lseg(lseg))
1212 goto out_resched;
1213 bucket = &array->buckets[ds_commit_idx];
1214 list = &bucket->written;
1215 /* Non-empty buckets hold a reference on the lseg. That ref
1216 * is normally transferred to the COMMIT call and released
1217 * there. It could also be released if the last req is pulled
1218 * off due to a rewrite, in which case it will be done in
1219 * pnfs_common_clear_request_commit
1220 */
1221 if (!bucket->lseg)
1222 bucket->lseg = pnfs_get_lseg(lseg);
1223 set_bit(nr: PG_COMMIT_TO_DS, addr: &req->wb_flags);
1224 cinfo->ds->nwritten++;
1225
1226 nfs_request_add_commit_list_locked(req, dst: list, cinfo);
1227 mutex_unlock(lock: &NFS_I(inode: cinfo->inode)->commit_mutex);
1228 nfs_folio_mark_unstable(folio: nfs_page_to_folio(req), cinfo);
1229 return;
1230out_resched:
1231 mutex_unlock(lock: &NFS_I(inode: cinfo->inode)->commit_mutex);
1232 cinfo->completion_ops->resched_write(cinfo, req);
1233}
1234EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
1235
1236int
1237pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
1238{
1239 int ret;
1240
1241 if (!pnfs_layoutcommit_outstanding(inode))
1242 return 0;
1243 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1244 if (ret < 0)
1245 return ret;
1246 if (datasync)
1247 return 0;
1248 return pnfs_layoutcommit_inode(inode, sync: true);
1249}
1250EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
1251
1252

source code of linux/fs/nfs/pnfs_nfs.c