Loading...
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
21void 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}
28EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
29
30/* Fake up some data that will cause nfs_commit_release to retry the writes. */
31void 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}
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(data->lseg);
57 nfs_put_client(data->ds_clp);
58 nfs_commitdata_release(data);
59}
60EXPORT_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 */
66void
67pnfs_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 }
84out:
85 nfs_request_remove_commit_list(req, cinfo);
86 pnfs_put_lseg_locked(freeme);
87}
88EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
89
90static int
91pnfs_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
113static int
114pnfs_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 */
140int 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}
154EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
155
156/* Pull everything off the committing lists and dump into @dst. */
157void 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);
165restart:
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}
179EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
180
181static 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
205static unsigned int
206pnfs_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 data = nfs_commitdata_alloc();
221 if (!data)
222 break;
223 data->ds_commit_index = i;
224 list_add(&data->pages, list);
225 nreq++;
226 }
227
228 /* Clean up on error */
229 pnfs_generic_retry_commit(cinfo, i);
230 return nreq;
231}
232
233static inline
234void pnfs_fetch_commit_bucket_list(struct list_head *pages,
235 struct nfs_commit_data *data,
236 struct nfs_commit_info *cinfo)
237{
238 struct pnfs_commit_bucket *bucket;
239
240 bucket = &cinfo->ds->buckets[data->ds_commit_index];
241 spin_lock(&cinfo->inode->i_lock);
242 list_splice_init(&bucket->committing, pages);
243 data->lseg = bucket->clseg;
244 bucket->clseg = NULL;
245 spin_unlock(&cinfo->inode->i_lock);
246
247}
248
249/* Helper function for pnfs_generic_commit_pagelist to catch an empty
250 * page list. This can happen when two commits race.
251 *
252 * This must be called instead of nfs_init_commit - call one or the other, but
253 * not both!
254 */
255static bool
256pnfs_generic_commit_cancel_empty_pagelist(struct list_head *pages,
257 struct nfs_commit_data *data,
258 struct nfs_commit_info *cinfo)
259{
260 if (list_empty(pages)) {
261 if (atomic_dec_and_test(&cinfo->mds->rpcs_out))
262 wake_up_atomic_t(&cinfo->mds->rpcs_out);
263 /* don't call nfs_commitdata_release - it tries to put
264 * the open_context which is not acquired until nfs_init_commit
265 * which has not been called on @data */
266 WARN_ON_ONCE(data->context);
267 nfs_commit_free(data);
268 return true;
269 }
270
271 return false;
272}
273
274/* This follows nfs_commit_list pretty closely */
275int
276pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
277 int how, struct nfs_commit_info *cinfo,
278 int (*initiate_commit)(struct nfs_commit_data *data,
279 int how))
280{
281 struct nfs_commit_data *data, *tmp;
282 LIST_HEAD(list);
283 unsigned int nreq = 0;
284
285 if (!list_empty(mds_pages)) {
286 data = nfs_commitdata_alloc();
287 if (data != NULL) {
288 data->ds_commit_index = -1;
289 list_add(&data->pages, &list);
290 nreq++;
291 } else {
292 nfs_retry_commit(mds_pages, NULL, cinfo, 0);
293 pnfs_generic_retry_commit(cinfo, 0);
294 return -ENOMEM;
295 }
296 }
297
298 nreq += pnfs_generic_alloc_ds_commits(cinfo, &list);
299
300 if (nreq == 0)
301 goto out;
302
303 atomic_add(nreq, &cinfo->mds->rpcs_out);
304
305 list_for_each_entry_safe(data, tmp, &list, pages) {
306 list_del_init(&data->pages);
307 if (data->ds_commit_index < 0) {
308 /* another commit raced with us */
309 if (pnfs_generic_commit_cancel_empty_pagelist(mds_pages,
310 data, cinfo))
311 continue;
312
313 nfs_init_commit(data, mds_pages, NULL, cinfo);
314 nfs_initiate_commit(NFS_CLIENT(inode), data,
315 NFS_PROTO(data->inode),
316 data->mds_ops, how, 0);
317 } else {
318 LIST_HEAD(pages);
319
320 pnfs_fetch_commit_bucket_list(&pages, data, cinfo);
321
322 /* another commit raced with us */
323 if (pnfs_generic_commit_cancel_empty_pagelist(&pages,
324 data, cinfo))
325 continue;
326
327 nfs_init_commit(data, &pages, data->lseg, cinfo);
328 initiate_commit(data, how);
329 }
330 }
331out:
332 cinfo->ds->ncommitting = 0;
333 return PNFS_ATTEMPTED;
334}
335EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
336
337/*
338 * Data server cache
339 *
340 * Data servers can be mapped to different device ids.
341 * nfs4_pnfs_ds reference counting
342 * - set to 1 on allocation
343 * - incremented when a device id maps a data server already in the cache.
344 * - decremented when deviceid is removed from the cache.
345 */
346static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
347static LIST_HEAD(nfs4_data_server_cache);
348
349/* Debug routines */
350static void
351print_ds(struct nfs4_pnfs_ds *ds)
352{
353 if (ds == NULL) {
354 printk(KERN_WARNING "%s NULL device\n", __func__);
355 return;
356 }
357 printk(KERN_WARNING " ds %s\n"
358 " ref count %d\n"
359 " client %p\n"
360 " cl_exchange_flags %x\n",
361 ds->ds_remotestr,
362 atomic_read(&ds->ds_count), ds->ds_clp,
363 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
364}
365
366static bool
367same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
368{
369 struct sockaddr_in *a, *b;
370 struct sockaddr_in6 *a6, *b6;
371
372 if (addr1->sa_family != addr2->sa_family)
373 return false;
374
375 switch (addr1->sa_family) {
376 case AF_INET:
377 a = (struct sockaddr_in *)addr1;
378 b = (struct sockaddr_in *)addr2;
379
380 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
381 a->sin_port == b->sin_port)
382 return true;
383 break;
384
385 case AF_INET6:
386 a6 = (struct sockaddr_in6 *)addr1;
387 b6 = (struct sockaddr_in6 *)addr2;
388
389 /* LINKLOCAL addresses must have matching scope_id */
390 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
391 IPV6_ADDR_SCOPE_LINKLOCAL &&
392 a6->sin6_scope_id != b6->sin6_scope_id)
393 return false;
394
395 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
396 a6->sin6_port == b6->sin6_port)
397 return true;
398 break;
399
400 default:
401 dprintk("%s: unhandled address family: %u\n",
402 __func__, addr1->sa_family);
403 return false;
404 }
405
406 return false;
407}
408
409/*
410 * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
411 * declare a match.
412 */
413static bool
414_same_data_server_addrs_locked(const struct list_head *dsaddrs1,
415 const struct list_head *dsaddrs2)
416{
417 struct nfs4_pnfs_ds_addr *da1, *da2;
418 struct sockaddr *sa1, *sa2;
419 bool match = false;
420
421 list_for_each_entry(da1, dsaddrs1, da_node) {
422 sa1 = (struct sockaddr *)&da1->da_addr;
423 match = false;
424 list_for_each_entry(da2, dsaddrs2, da_node) {
425 sa2 = (struct sockaddr *)&da2->da_addr;
426 match = same_sockaddr(sa1, sa2);
427 if (match)
428 break;
429 }
430 if (!match)
431 break;
432 }
433 return match;
434}
435
436/*
437 * Lookup DS by addresses. nfs4_ds_cache_lock is held
438 */
439static struct nfs4_pnfs_ds *
440_data_server_lookup_locked(const struct list_head *dsaddrs)
441{
442 struct nfs4_pnfs_ds *ds;
443
444 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
445 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
446 return ds;
447 return NULL;
448}
449
450static void destroy_ds(struct nfs4_pnfs_ds *ds)
451{
452 struct nfs4_pnfs_ds_addr *da;
453
454 dprintk("--> %s\n", __func__);
455 ifdebug(FACILITY)
456 print_ds(ds);
457
458 nfs_put_client(ds->ds_clp);
459
460 while (!list_empty(&ds->ds_addrs)) {
461 da = list_first_entry(&ds->ds_addrs,
462 struct nfs4_pnfs_ds_addr,
463 da_node);
464 list_del_init(&da->da_node);
465 kfree(da->da_remotestr);
466 kfree(da);
467 }
468
469 kfree(ds->ds_remotestr);
470 kfree(ds);
471}
472
473void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
474{
475 if (atomic_dec_and_lock(&ds->ds_count,
476 &nfs4_ds_cache_lock)) {
477 list_del_init(&ds->ds_node);
478 spin_unlock(&nfs4_ds_cache_lock);
479 destroy_ds(ds);
480 }
481}
482EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
483
484/*
485 * Create a string with a human readable address and port to avoid
486 * complicated setup around many dprinks.
487 */
488static char *
489nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
490{
491 struct nfs4_pnfs_ds_addr *da;
492 char *remotestr;
493 size_t len;
494 char *p;
495
496 len = 3; /* '{', '}' and eol */
497 list_for_each_entry(da, dsaddrs, da_node) {
498 len += strlen(da->da_remotestr) + 1; /* string plus comma */
499 }
500
501 remotestr = kzalloc(len, gfp_flags);
502 if (!remotestr)
503 return NULL;
504
505 p = remotestr;
506 *(p++) = '{';
507 len--;
508 list_for_each_entry(da, dsaddrs, da_node) {
509 size_t ll = strlen(da->da_remotestr);
510
511 if (ll > len)
512 goto out_err;
513
514 memcpy(p, da->da_remotestr, ll);
515 p += ll;
516 len -= ll;
517
518 if (len < 1)
519 goto out_err;
520 (*p++) = ',';
521 len--;
522 }
523 if (len < 2)
524 goto out_err;
525 *(p++) = '}';
526 *p = '\0';
527 return remotestr;
528out_err:
529 kfree(remotestr);
530 return NULL;
531}
532
533/*
534 * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
535 * uncached and return cached struct nfs4_pnfs_ds.
536 */
537struct nfs4_pnfs_ds *
538nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
539{
540 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
541 char *remotestr;
542
543 if (list_empty(dsaddrs)) {
544 dprintk("%s: no addresses defined\n", __func__);
545 goto out;
546 }
547
548 ds = kzalloc(sizeof(*ds), gfp_flags);
549 if (!ds)
550 goto out;
551
552 /* this is only used for debugging, so it's ok if its NULL */
553 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
554
555 spin_lock(&nfs4_ds_cache_lock);
556 tmp_ds = _data_server_lookup_locked(dsaddrs);
557 if (tmp_ds == NULL) {
558 INIT_LIST_HEAD(&ds->ds_addrs);
559 list_splice_init(dsaddrs, &ds->ds_addrs);
560 ds->ds_remotestr = remotestr;
561 atomic_set(&ds->ds_count, 1);
562 INIT_LIST_HEAD(&ds->ds_node);
563 ds->ds_clp = NULL;
564 list_add(&ds->ds_node, &nfs4_data_server_cache);
565 dprintk("%s add new data server %s\n", __func__,
566 ds->ds_remotestr);
567 } else {
568 kfree(remotestr);
569 kfree(ds);
570 atomic_inc(&tmp_ds->ds_count);
571 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
572 __func__, tmp_ds->ds_remotestr,
573 atomic_read(&tmp_ds->ds_count));
574 ds = tmp_ds;
575 }
576 spin_unlock(&nfs4_ds_cache_lock);
577out:
578 return ds;
579}
580EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
581
582static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
583{
584 might_sleep();
585 wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
586 TASK_KILLABLE);
587}
588
589static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
590{
591 smp_mb__before_atomic();
592 clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
593 smp_mb__after_atomic();
594 wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
595}
596
597static struct nfs_client *(*get_v3_ds_connect)(
598 struct nfs_server *mds_srv,
599 const struct sockaddr *ds_addr,
600 int ds_addrlen,
601 int ds_proto,
602 unsigned int ds_timeo,
603 unsigned int ds_retrans);
604
605static bool load_v3_ds_connect(void)
606{
607 if (!get_v3_ds_connect) {
608 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
609 WARN_ON_ONCE(!get_v3_ds_connect);
610 }
611
612 return(get_v3_ds_connect != NULL);
613}
614
615void nfs4_pnfs_v3_ds_connect_unload(void)
616{
617 if (get_v3_ds_connect) {
618 symbol_put(nfs3_set_ds_client);
619 get_v3_ds_connect = NULL;
620 }
621}
622EXPORT_SYMBOL_GPL(nfs4_pnfs_v3_ds_connect_unload);
623
624static 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);
668out:
669 return status;
670}
671
672static 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);
741out:
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.
749 */
750void 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 if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
755 int err = 0;
756
757 if (version == 3) {
758 err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
759 retrans);
760 } else if (version == 4) {
761 err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
762 retrans, minor_version);
763 } else {
764 dprintk("%s: unsupported DS version %d\n", __func__,
765 version);
766 err = -EPROTONOSUPPORT;
767 }
768
769 if (err)
770 nfs4_mark_deviceid_unavailable(devid);
771 nfs4_clear_ds_conn_bit(ds);
772 } else {
773 nfs4_wait_ds_connect(ds);
774 }
775}
776EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
777
778/*
779 * Currently only supports ipv4, ipv6 and one multi-path address.
780 */
781struct nfs4_pnfs_ds_addr *
782nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
783{
784 struct nfs4_pnfs_ds_addr *da = NULL;
785 char *buf, *portstr;
786 __be16 port;
787 int nlen, rlen;
788 int tmp[2];
789 __be32 *p;
790 char *netid, *match_netid;
791 size_t len, match_netid_len;
792 char *startsep = "";
793 char *endsep = "";
794
795
796 /* r_netid */
797 p = xdr_inline_decode(xdr, 4);
798 if (unlikely(!p))
799 goto out_err;
800 nlen = be32_to_cpup(p++);
801
802 p = xdr_inline_decode(xdr, nlen);
803 if (unlikely(!p))
804 goto out_err;
805
806 netid = kmalloc(nlen+1, gfp_flags);
807 if (unlikely(!netid))
808 goto out_err;
809
810 netid[nlen] = '\0';
811 memcpy(netid, p, nlen);
812
813 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
814 p = xdr_inline_decode(xdr, 4);
815 if (unlikely(!p))
816 goto out_free_netid;
817 rlen = be32_to_cpup(p);
818
819 p = xdr_inline_decode(xdr, rlen);
820 if (unlikely(!p))
821 goto out_free_netid;
822
823 /* port is ".ABC.DEF", 8 chars max */
824 if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
825 dprintk("%s: Invalid address, length %d\n", __func__,
826 rlen);
827 goto out_free_netid;
828 }
829 buf = kmalloc(rlen + 1, gfp_flags);
830 if (!buf) {
831 dprintk("%s: Not enough memory\n", __func__);
832 goto out_free_netid;
833 }
834 buf[rlen] = '\0';
835 memcpy(buf, p, rlen);
836
837 /* replace port '.' with '-' */
838 portstr = strrchr(buf, '.');
839 if (!portstr) {
840 dprintk("%s: Failed finding expected dot in port\n",
841 __func__);
842 goto out_free_buf;
843 }
844 *portstr = '-';
845
846 /* find '.' between address and port */
847 portstr = strrchr(buf, '.');
848 if (!portstr) {
849 dprintk("%s: Failed finding expected dot between address and "
850 "port\n", __func__);
851 goto out_free_buf;
852 }
853 *portstr = '\0';
854
855 da = kzalloc(sizeof(*da), gfp_flags);
856 if (unlikely(!da))
857 goto out_free_buf;
858
859 INIT_LIST_HEAD(&da->da_node);
860
861 if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
862 sizeof(da->da_addr))) {
863 dprintk("%s: error parsing address %s\n", __func__, buf);
864 goto out_free_da;
865 }
866
867 portstr++;
868 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
869 port = htons((tmp[0] << 8) | (tmp[1]));
870
871 switch (da->da_addr.ss_family) {
872 case AF_INET:
873 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
874 da->da_addrlen = sizeof(struct sockaddr_in);
875 match_netid = "tcp";
876 match_netid_len = 3;
877 break;
878
879 case AF_INET6:
880 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
881 da->da_addrlen = sizeof(struct sockaddr_in6);
882 match_netid = "tcp6";
883 match_netid_len = 4;
884 startsep = "[";
885 endsep = "]";
886 break;
887
888 default:
889 dprintk("%s: unsupported address family: %u\n",
890 __func__, da->da_addr.ss_family);
891 goto out_free_da;
892 }
893
894 if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
895 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
896 __func__, netid, match_netid);
897 goto out_free_da;
898 }
899
900 /* save human readable address */
901 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
902 da->da_remotestr = kzalloc(len, gfp_flags);
903
904 /* NULL is ok, only used for dprintk */
905 if (da->da_remotestr)
906 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
907 buf, endsep, ntohs(port));
908
909 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
910 kfree(buf);
911 kfree(netid);
912 return da;
913
914out_free_da:
915 kfree(da);
916out_free_buf:
917 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
918 kfree(buf);
919out_free_netid:
920 kfree(netid);
921out_err:
922 return NULL;
923}
924EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
925
926void
927pnfs_layout_mark_request_commit(struct nfs_page *req,
928 struct pnfs_layout_segment *lseg,
929 struct nfs_commit_info *cinfo,
930 u32 ds_commit_idx)
931{
932 struct list_head *list;
933 struct pnfs_commit_bucket *buckets;
934
935 spin_lock(&cinfo->inode->i_lock);
936 buckets = cinfo->ds->buckets;
937 list = &buckets[ds_commit_idx].written;
938 if (list_empty(list)) {
939 if (!pnfs_is_valid_lseg(lseg)) {
940 spin_unlock(&cinfo->inode->i_lock);
941 cinfo->completion_ops->resched_write(cinfo, req);
942 return;
943 }
944 /* Non-empty buckets hold a reference on the lseg. That ref
945 * is normally transferred to the COMMIT call and released
946 * there. It could also be released if the last req is pulled
947 * off due to a rewrite, in which case it will be done in
948 * pnfs_common_clear_request_commit
949 */
950 WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL);
951 buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg);
952 }
953 set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
954 cinfo->ds->nwritten++;
955
956 nfs_request_add_commit_list_locked(req, list, cinfo);
957 spin_unlock(&cinfo->inode->i_lock);
958 nfs_mark_page_unstable(req->wb_page, cinfo);
959}
960EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
961
962int
963pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
964{
965 int ret;
966
967 if (!pnfs_layoutcommit_outstanding(inode))
968 return 0;
969 ret = nfs_commit_inode(inode, FLUSH_SYNC);
970 if (ret < 0)
971 return ret;
972 if (datasync)
973 return 0;
974 return pnfs_layoutcommit_inode(inode, true);
975}
976EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
977
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_page *first = nfs_list_entry(data->pages.next);
35
36 data->task.tk_status = 0;
37 memcpy(&data->verf.verifier, &first->wb_verf,
38 sizeof(data->verf.verifier));
39 data->verf.verifier.data[0]++; /* ensure verifier mismatch */
40}
41EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes);
42
43void pnfs_generic_write_commit_done(struct rpc_task *task, void *data)
44{
45 struct nfs_commit_data *wdata = data;
46
47 /* Note this may cause RPC to be resent */
48 wdata->mds_ops->rpc_call_done(task, data);
49}
50EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done);
51
52void pnfs_generic_commit_release(void *calldata)
53{
54 struct nfs_commit_data *data = calldata;
55
56 data->completion_ops->completion(data);
57 pnfs_put_lseg(data->lseg);
58 nfs_put_client(data->ds_clp);
59 nfs_commitdata_release(data);
60}
61EXPORT_SYMBOL_GPL(pnfs_generic_commit_release);
62
63/* The generic layer is about to remove the req from the commit list.
64 * If this will make the bucket empty, it will need to put the lseg reference.
65 * Note this must be called holding nfsi->commit_mutex
66 */
67void
68pnfs_generic_clear_request_commit(struct nfs_page *req,
69 struct nfs_commit_info *cinfo)
70{
71 struct pnfs_layout_segment *freeme = NULL;
72
73 if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
74 goto out;
75 cinfo->ds->nwritten--;
76 if (list_is_singular(&req->wb_list)) {
77 struct pnfs_commit_bucket *bucket;
78
79 bucket = list_first_entry(&req->wb_list,
80 struct pnfs_commit_bucket,
81 written);
82 freeme = bucket->wlseg;
83 bucket->wlseg = NULL;
84 }
85out:
86 nfs_request_remove_commit_list(req, cinfo);
87 pnfs_put_lseg(freeme);
88}
89EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
90
91static int
92pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
93 struct nfs_commit_info *cinfo,
94 int max)
95{
96 struct list_head *src = &bucket->written;
97 struct list_head *dst = &bucket->committing;
98 int ret;
99
100 lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
101 ret = nfs_scan_commit_list(src, dst, cinfo, max);
102 if (ret) {
103 cinfo->ds->nwritten -= ret;
104 cinfo->ds->ncommitting += ret;
105 if (bucket->clseg == NULL)
106 bucket->clseg = pnfs_get_lseg(bucket->wlseg);
107 if (list_empty(src)) {
108 pnfs_put_lseg(bucket->wlseg);
109 bucket->wlseg = NULL;
110 }
111 }
112 return ret;
113}
114
115/* Move reqs from written to committing lists, returning count
116 * of number moved.
117 */
118int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo,
119 int max)
120{
121 int i, rv = 0, cnt;
122
123 lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
124 for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
125 cnt = pnfs_generic_scan_ds_commit_list(&cinfo->ds->buckets[i],
126 cinfo, max);
127 max -= cnt;
128 rv += cnt;
129 }
130 return rv;
131}
132EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
133
134/* Pull everything off the committing lists and dump into @dst. */
135void pnfs_generic_recover_commit_reqs(struct list_head *dst,
136 struct nfs_commit_info *cinfo)
137{
138 struct pnfs_commit_bucket *b;
139 struct pnfs_layout_segment *freeme;
140 int nwritten;
141 int i;
142
143 lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
144restart:
145 for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
146 nwritten = nfs_scan_commit_list(&b->written, dst, cinfo, 0);
147 if (!nwritten)
148 continue;
149 cinfo->ds->nwritten -= nwritten;
150 if (list_empty(&b->written)) {
151 freeme = b->wlseg;
152 b->wlseg = NULL;
153 pnfs_put_lseg(freeme);
154 goto restart;
155 }
156 }
157}
158EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
159
160static void pnfs_generic_retry_commit(struct nfs_commit_info *cinfo, int idx)
161{
162 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
163 struct pnfs_commit_bucket *bucket;
164 struct pnfs_layout_segment *freeme;
165 struct list_head *pos;
166 LIST_HEAD(pages);
167 int i;
168
169 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
170 for (i = idx; i < fl_cinfo->nbuckets; i++) {
171 bucket = &fl_cinfo->buckets[i];
172 if (list_empty(&bucket->committing))
173 continue;
174 freeme = bucket->clseg;
175 bucket->clseg = NULL;
176 list_for_each(pos, &bucket->committing)
177 cinfo->ds->ncommitting--;
178 list_splice_init(&bucket->committing, &pages);
179 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
180 nfs_retry_commit(&pages, freeme, cinfo, i);
181 pnfs_put_lseg(freeme);
182 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
183 }
184 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
185}
186
187static unsigned int
188pnfs_generic_alloc_ds_commits(struct nfs_commit_info *cinfo,
189 struct list_head *list)
190{
191 struct pnfs_ds_commit_info *fl_cinfo;
192 struct pnfs_commit_bucket *bucket;
193 struct nfs_commit_data *data;
194 int i;
195 unsigned int nreq = 0;
196
197 fl_cinfo = cinfo->ds;
198 bucket = fl_cinfo->buckets;
199 for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
200 if (list_empty(&bucket->committing))
201 continue;
202 data = nfs_commitdata_alloc(false);
203 if (!data)
204 break;
205 data->ds_commit_index = i;
206 list_add(&data->pages, list);
207 nreq++;
208 }
209
210 /* Clean up on error */
211 pnfs_generic_retry_commit(cinfo, i);
212 return nreq;
213}
214
215static inline
216void pnfs_fetch_commit_bucket_list(struct list_head *pages,
217 struct nfs_commit_data *data,
218 struct nfs_commit_info *cinfo)
219{
220 struct pnfs_commit_bucket *bucket;
221 struct list_head *pos;
222
223 bucket = &cinfo->ds->buckets[data->ds_commit_index];
224 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
225 list_for_each(pos, &bucket->committing)
226 cinfo->ds->ncommitting--;
227 list_splice_init(&bucket->committing, pages);
228 data->lseg = bucket->clseg;
229 bucket->clseg = NULL;
230 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
231
232}
233
234/* Helper function for pnfs_generic_commit_pagelist to catch an empty
235 * page list. This can happen when two commits race.
236 *
237 * This must be called instead of nfs_init_commit - call one or the other, but
238 * not both!
239 */
240static bool
241pnfs_generic_commit_cancel_empty_pagelist(struct list_head *pages,
242 struct nfs_commit_data *data,
243 struct nfs_commit_info *cinfo)
244{
245 if (list_empty(pages)) {
246 if (atomic_dec_and_test(&cinfo->mds->rpcs_out))
247 wake_up_var(&cinfo->mds->rpcs_out);
248 /* don't call nfs_commitdata_release - it tries to put
249 * the open_context which is not acquired until nfs_init_commit
250 * which has not been called on @data */
251 WARN_ON_ONCE(data->context);
252 nfs_commit_free(data);
253 return true;
254 }
255
256 return false;
257}
258
259/* This follows nfs_commit_list pretty closely */
260int
261pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
262 int how, struct nfs_commit_info *cinfo,
263 int (*initiate_commit)(struct nfs_commit_data *data,
264 int how))
265{
266 struct nfs_commit_data *data, *tmp;
267 LIST_HEAD(list);
268 unsigned int nreq = 0;
269
270 if (!list_empty(mds_pages)) {
271 data = nfs_commitdata_alloc(true);
272 data->ds_commit_index = -1;
273 list_add(&data->pages, &list);
274 nreq++;
275 }
276
277 nreq += pnfs_generic_alloc_ds_commits(cinfo, &list);
278
279 if (nreq == 0)
280 goto out;
281
282 atomic_add(nreq, &cinfo->mds->rpcs_out);
283
284 list_for_each_entry_safe(data, tmp, &list, pages) {
285 list_del_init(&data->pages);
286 if (data->ds_commit_index < 0) {
287 /* another commit raced with us */
288 if (pnfs_generic_commit_cancel_empty_pagelist(mds_pages,
289 data, cinfo))
290 continue;
291
292 nfs_init_commit(data, mds_pages, NULL, cinfo);
293 nfs_initiate_commit(NFS_CLIENT(inode), data,
294 NFS_PROTO(data->inode),
295 data->mds_ops, how, 0);
296 } else {
297 LIST_HEAD(pages);
298
299 pnfs_fetch_commit_bucket_list(&pages, data, cinfo);
300
301 /* another commit raced with us */
302 if (pnfs_generic_commit_cancel_empty_pagelist(&pages,
303 data, cinfo))
304 continue;
305
306 nfs_init_commit(data, &pages, data->lseg, cinfo);
307 initiate_commit(data, how);
308 }
309 }
310out:
311 return PNFS_ATTEMPTED;
312}
313EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
314
315/*
316 * Data server cache
317 *
318 * Data servers can be mapped to different device ids.
319 * nfs4_pnfs_ds reference counting
320 * - set to 1 on allocation
321 * - incremented when a device id maps a data server already in the cache.
322 * - decremented when deviceid is removed from the cache.
323 */
324static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
325static LIST_HEAD(nfs4_data_server_cache);
326
327/* Debug routines */
328static void
329print_ds(struct nfs4_pnfs_ds *ds)
330{
331 if (ds == NULL) {
332 printk(KERN_WARNING "%s NULL device\n", __func__);
333 return;
334 }
335 printk(KERN_WARNING " ds %s\n"
336 " ref count %d\n"
337 " client %p\n"
338 " cl_exchange_flags %x\n",
339 ds->ds_remotestr,
340 refcount_read(&ds->ds_count), ds->ds_clp,
341 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
342}
343
344static bool
345same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
346{
347 struct sockaddr_in *a, *b;
348 struct sockaddr_in6 *a6, *b6;
349
350 if (addr1->sa_family != addr2->sa_family)
351 return false;
352
353 switch (addr1->sa_family) {
354 case AF_INET:
355 a = (struct sockaddr_in *)addr1;
356 b = (struct sockaddr_in *)addr2;
357
358 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
359 a->sin_port == b->sin_port)
360 return true;
361 break;
362
363 case AF_INET6:
364 a6 = (struct sockaddr_in6 *)addr1;
365 b6 = (struct sockaddr_in6 *)addr2;
366
367 /* LINKLOCAL addresses must have matching scope_id */
368 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
369 IPV6_ADDR_SCOPE_LINKLOCAL &&
370 a6->sin6_scope_id != b6->sin6_scope_id)
371 return false;
372
373 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
374 a6->sin6_port == b6->sin6_port)
375 return true;
376 break;
377
378 default:
379 dprintk("%s: unhandled address family: %u\n",
380 __func__, addr1->sa_family);
381 return false;
382 }
383
384 return false;
385}
386
387/*
388 * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
389 * declare a match.
390 */
391static bool
392_same_data_server_addrs_locked(const struct list_head *dsaddrs1,
393 const struct list_head *dsaddrs2)
394{
395 struct nfs4_pnfs_ds_addr *da1, *da2;
396 struct sockaddr *sa1, *sa2;
397 bool match = false;
398
399 list_for_each_entry(da1, dsaddrs1, da_node) {
400 sa1 = (struct sockaddr *)&da1->da_addr;
401 match = false;
402 list_for_each_entry(da2, dsaddrs2, da_node) {
403 sa2 = (struct sockaddr *)&da2->da_addr;
404 match = same_sockaddr(sa1, sa2);
405 if (match)
406 break;
407 }
408 if (!match)
409 break;
410 }
411 return match;
412}
413
414/*
415 * Lookup DS by addresses. nfs4_ds_cache_lock is held
416 */
417static struct nfs4_pnfs_ds *
418_data_server_lookup_locked(const struct list_head *dsaddrs)
419{
420 struct nfs4_pnfs_ds *ds;
421
422 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
423 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
424 return ds;
425 return NULL;
426}
427
428static void destroy_ds(struct nfs4_pnfs_ds *ds)
429{
430 struct nfs4_pnfs_ds_addr *da;
431
432 dprintk("--> %s\n", __func__);
433 ifdebug(FACILITY)
434 print_ds(ds);
435
436 nfs_put_client(ds->ds_clp);
437
438 while (!list_empty(&ds->ds_addrs)) {
439 da = list_first_entry(&ds->ds_addrs,
440 struct nfs4_pnfs_ds_addr,
441 da_node);
442 list_del_init(&da->da_node);
443 kfree(da->da_remotestr);
444 kfree(da);
445 }
446
447 kfree(ds->ds_remotestr);
448 kfree(ds);
449}
450
451void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
452{
453 if (refcount_dec_and_lock(&ds->ds_count,
454 &nfs4_ds_cache_lock)) {
455 list_del_init(&ds->ds_node);
456 spin_unlock(&nfs4_ds_cache_lock);
457 destroy_ds(ds);
458 }
459}
460EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
461
462/*
463 * Create a string with a human readable address and port to avoid
464 * complicated setup around many dprinks.
465 */
466static char *
467nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
468{
469 struct nfs4_pnfs_ds_addr *da;
470 char *remotestr;
471 size_t len;
472 char *p;
473
474 len = 3; /* '{', '}' and eol */
475 list_for_each_entry(da, dsaddrs, da_node) {
476 len += strlen(da->da_remotestr) + 1; /* string plus comma */
477 }
478
479 remotestr = kzalloc(len, gfp_flags);
480 if (!remotestr)
481 return NULL;
482
483 p = remotestr;
484 *(p++) = '{';
485 len--;
486 list_for_each_entry(da, dsaddrs, da_node) {
487 size_t ll = strlen(da->da_remotestr);
488
489 if (ll > len)
490 goto out_err;
491
492 memcpy(p, da->da_remotestr, ll);
493 p += ll;
494 len -= ll;
495
496 if (len < 1)
497 goto out_err;
498 (*p++) = ',';
499 len--;
500 }
501 if (len < 2)
502 goto out_err;
503 *(p++) = '}';
504 *p = '\0';
505 return remotestr;
506out_err:
507 kfree(remotestr);
508 return NULL;
509}
510
511/*
512 * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
513 * uncached and return cached struct nfs4_pnfs_ds.
514 */
515struct nfs4_pnfs_ds *
516nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
517{
518 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
519 char *remotestr;
520
521 if (list_empty(dsaddrs)) {
522 dprintk("%s: no addresses defined\n", __func__);
523 goto out;
524 }
525
526 ds = kzalloc(sizeof(*ds), gfp_flags);
527 if (!ds)
528 goto out;
529
530 /* this is only used for debugging, so it's ok if its NULL */
531 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
532
533 spin_lock(&nfs4_ds_cache_lock);
534 tmp_ds = _data_server_lookup_locked(dsaddrs);
535 if (tmp_ds == NULL) {
536 INIT_LIST_HEAD(&ds->ds_addrs);
537 list_splice_init(dsaddrs, &ds->ds_addrs);
538 ds->ds_remotestr = remotestr;
539 refcount_set(&ds->ds_count, 1);
540 INIT_LIST_HEAD(&ds->ds_node);
541 ds->ds_clp = NULL;
542 list_add(&ds->ds_node, &nfs4_data_server_cache);
543 dprintk("%s add new data server %s\n", __func__,
544 ds->ds_remotestr);
545 } else {
546 kfree(remotestr);
547 kfree(ds);
548 refcount_inc(&tmp_ds->ds_count);
549 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
550 __func__, tmp_ds->ds_remotestr,
551 refcount_read(&tmp_ds->ds_count));
552 ds = tmp_ds;
553 }
554 spin_unlock(&nfs4_ds_cache_lock);
555out:
556 return ds;
557}
558EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
559
560static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
561{
562 might_sleep();
563 wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
564 TASK_KILLABLE);
565}
566
567static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
568{
569 smp_mb__before_atomic();
570 clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
571 smp_mb__after_atomic();
572 wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
573}
574
575static struct nfs_client *(*get_v3_ds_connect)(
576 struct nfs_server *mds_srv,
577 const struct sockaddr *ds_addr,
578 int ds_addrlen,
579 int ds_proto,
580 unsigned int ds_timeo,
581 unsigned int ds_retrans);
582
583static bool load_v3_ds_connect(void)
584{
585 if (!get_v3_ds_connect) {
586 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
587 WARN_ON_ONCE(!get_v3_ds_connect);
588 }
589
590 return(get_v3_ds_connect != NULL);
591}
592
593void nfs4_pnfs_v3_ds_connect_unload(void)
594{
595 if (get_v3_ds_connect) {
596 symbol_put(nfs3_set_ds_client);
597 get_v3_ds_connect = NULL;
598 }
599}
600
601static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
602 struct nfs4_pnfs_ds *ds,
603 unsigned int timeo,
604 unsigned int retrans)
605{
606 struct nfs_client *clp = ERR_PTR(-EIO);
607 struct nfs4_pnfs_ds_addr *da;
608 int status = 0;
609
610 dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
611
612 if (!load_v3_ds_connect())
613 goto out;
614
615 list_for_each_entry(da, &ds->ds_addrs, da_node) {
616 dprintk("%s: DS %s: trying address %s\n",
617 __func__, ds->ds_remotestr, da->da_remotestr);
618
619 if (!IS_ERR(clp)) {
620 struct xprt_create xprt_args = {
621 .ident = XPRT_TRANSPORT_TCP,
622 .net = clp->cl_net,
623 .dstaddr = (struct sockaddr *)&da->da_addr,
624 .addrlen = da->da_addrlen,
625 .servername = clp->cl_hostname,
626 };
627 /* Add this address as an alias */
628 rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
629 rpc_clnt_test_and_add_xprt, NULL);
630 continue;
631 }
632 clp = get_v3_ds_connect(mds_srv,
633 (struct sockaddr *)&da->da_addr,
634 da->da_addrlen, IPPROTO_TCP,
635 timeo, retrans);
636 if (IS_ERR(clp))
637 continue;
638 clp->cl_rpcclient->cl_softerr = 0;
639 clp->cl_rpcclient->cl_softrtry = 0;
640 }
641
642 if (IS_ERR(clp)) {
643 status = PTR_ERR(clp);
644 goto out;
645 }
646
647 smp_wmb();
648 ds->ds_clp = clp;
649 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
650out:
651 return status;
652}
653
654static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
655 struct nfs4_pnfs_ds *ds,
656 unsigned int timeo,
657 unsigned int retrans,
658 u32 minor_version)
659{
660 struct nfs_client *clp = ERR_PTR(-EIO);
661 struct nfs4_pnfs_ds_addr *da;
662 int status = 0;
663
664 dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
665
666 list_for_each_entry(da, &ds->ds_addrs, da_node) {
667 dprintk("%s: DS %s: trying address %s\n",
668 __func__, ds->ds_remotestr, da->da_remotestr);
669
670 if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) {
671 struct xprt_create xprt_args = {
672 .ident = XPRT_TRANSPORT_TCP,
673 .net = clp->cl_net,
674 .dstaddr = (struct sockaddr *)&da->da_addr,
675 .addrlen = da->da_addrlen,
676 .servername = clp->cl_hostname,
677 };
678 struct nfs4_add_xprt_data xprtdata = {
679 .clp = clp,
680 .cred = nfs4_get_clid_cred(clp),
681 };
682 struct rpc_add_xprt_test rpcdata = {
683 .add_xprt_test = clp->cl_mvops->session_trunk,
684 .data = &xprtdata,
685 };
686
687 /**
688 * Test this address for session trunking and
689 * add as an alias
690 */
691 rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
692 rpc_clnt_setup_test_and_add_xprt,
693 &rpcdata);
694 if (xprtdata.cred)
695 put_cred(xprtdata.cred);
696 } else {
697 clp = nfs4_set_ds_client(mds_srv,
698 (struct sockaddr *)&da->da_addr,
699 da->da_addrlen, IPPROTO_TCP,
700 timeo, retrans, minor_version);
701 if (IS_ERR(clp))
702 continue;
703
704 status = nfs4_init_ds_session(clp,
705 mds_srv->nfs_client->cl_lease_time);
706 if (status) {
707 nfs_put_client(clp);
708 clp = ERR_PTR(-EIO);
709 continue;
710 }
711
712 }
713 }
714
715 if (IS_ERR(clp)) {
716 status = PTR_ERR(clp);
717 goto out;
718 }
719
720 smp_wmb();
721 ds->ds_clp = clp;
722 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
723out:
724 return status;
725}
726
727/*
728 * Create an rpc connection to the nfs4_pnfs_ds data server.
729 * Currently only supports IPv4 and IPv6 addresses.
730 * If connection fails, make devid unavailable and return a -errno.
731 */
732int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
733 struct nfs4_deviceid_node *devid, unsigned int timeo,
734 unsigned int retrans, u32 version, u32 minor_version)
735{
736 int err;
737
738again:
739 err = 0;
740 if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
741 if (version == 3) {
742 err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
743 retrans);
744 } else if (version == 4) {
745 err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
746 retrans, minor_version);
747 } else {
748 dprintk("%s: unsupported DS version %d\n", __func__,
749 version);
750 err = -EPROTONOSUPPORT;
751 }
752
753 nfs4_clear_ds_conn_bit(ds);
754 } else {
755 nfs4_wait_ds_connect(ds);
756
757 /* what was waited on didn't connect AND didn't mark unavail */
758 if (!ds->ds_clp && !nfs4_test_deviceid_unavailable(devid))
759 goto again;
760 }
761
762 /*
763 * At this point the ds->ds_clp should be ready, but it might have
764 * hit an error.
765 */
766 if (!err) {
767 if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) {
768 WARN_ON_ONCE(ds->ds_clp ||
769 !nfs4_test_deviceid_unavailable(devid));
770 return -EINVAL;
771 }
772 err = nfs_client_init_status(ds->ds_clp);
773 }
774
775 return err;
776}
777EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
778
779/*
780 * Currently only supports ipv4, ipv6 and one multi-path address.
781 */
782struct nfs4_pnfs_ds_addr *
783nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
784{
785 struct nfs4_pnfs_ds_addr *da = NULL;
786 char *buf, *portstr;
787 __be16 port;
788 int nlen, rlen;
789 int tmp[2];
790 __be32 *p;
791 char *netid, *match_netid;
792 size_t len, match_netid_len;
793 char *startsep = "";
794 char *endsep = "";
795
796
797 /* r_netid */
798 p = xdr_inline_decode(xdr, 4);
799 if (unlikely(!p))
800 goto out_err;
801 nlen = be32_to_cpup(p++);
802
803 p = xdr_inline_decode(xdr, nlen);
804 if (unlikely(!p))
805 goto out_err;
806
807 netid = kmalloc(nlen+1, gfp_flags);
808 if (unlikely(!netid))
809 goto out_err;
810
811 netid[nlen] = '\0';
812 memcpy(netid, p, nlen);
813
814 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
815 p = xdr_inline_decode(xdr, 4);
816 if (unlikely(!p))
817 goto out_free_netid;
818 rlen = be32_to_cpup(p);
819
820 p = xdr_inline_decode(xdr, rlen);
821 if (unlikely(!p))
822 goto out_free_netid;
823
824 /* port is ".ABC.DEF", 8 chars max */
825 if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
826 dprintk("%s: Invalid address, length %d\n", __func__,
827 rlen);
828 goto out_free_netid;
829 }
830 buf = kmalloc(rlen + 1, gfp_flags);
831 if (!buf) {
832 dprintk("%s: Not enough memory\n", __func__);
833 goto out_free_netid;
834 }
835 buf[rlen] = '\0';
836 memcpy(buf, p, rlen);
837
838 /* replace port '.' with '-' */
839 portstr = strrchr(buf, '.');
840 if (!portstr) {
841 dprintk("%s: Failed finding expected dot in port\n",
842 __func__);
843 goto out_free_buf;
844 }
845 *portstr = '-';
846
847 /* find '.' between address and port */
848 portstr = strrchr(buf, '.');
849 if (!portstr) {
850 dprintk("%s: Failed finding expected dot between address and "
851 "port\n", __func__);
852 goto out_free_buf;
853 }
854 *portstr = '\0';
855
856 da = kzalloc(sizeof(*da), gfp_flags);
857 if (unlikely(!da))
858 goto out_free_buf;
859
860 INIT_LIST_HEAD(&da->da_node);
861
862 if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
863 sizeof(da->da_addr))) {
864 dprintk("%s: error parsing address %s\n", __func__, buf);
865 goto out_free_da;
866 }
867
868 portstr++;
869 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
870 port = htons((tmp[0] << 8) | (tmp[1]));
871
872 switch (da->da_addr.ss_family) {
873 case AF_INET:
874 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
875 da->da_addrlen = sizeof(struct sockaddr_in);
876 match_netid = "tcp";
877 match_netid_len = 3;
878 break;
879
880 case AF_INET6:
881 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
882 da->da_addrlen = sizeof(struct sockaddr_in6);
883 match_netid = "tcp6";
884 match_netid_len = 4;
885 startsep = "[";
886 endsep = "]";
887 break;
888
889 default:
890 dprintk("%s: unsupported address family: %u\n",
891 __func__, da->da_addr.ss_family);
892 goto out_free_da;
893 }
894
895 if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
896 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
897 __func__, netid, match_netid);
898 goto out_free_da;
899 }
900
901 /* save human readable address */
902 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
903 da->da_remotestr = kzalloc(len, gfp_flags);
904
905 /* NULL is ok, only used for dprintk */
906 if (da->da_remotestr)
907 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
908 buf, endsep, ntohs(port));
909
910 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
911 kfree(buf);
912 kfree(netid);
913 return da;
914
915out_free_da:
916 kfree(da);
917out_free_buf:
918 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
919 kfree(buf);
920out_free_netid:
921 kfree(netid);
922out_err:
923 return NULL;
924}
925EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
926
927void
928pnfs_layout_mark_request_commit(struct nfs_page *req,
929 struct pnfs_layout_segment *lseg,
930 struct nfs_commit_info *cinfo,
931 u32 ds_commit_idx)
932{
933 struct list_head *list;
934 struct pnfs_commit_bucket *buckets;
935
936 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
937 buckets = cinfo->ds->buckets;
938 list = &buckets[ds_commit_idx].written;
939 if (list_empty(list)) {
940 if (!pnfs_is_valid_lseg(lseg)) {
941 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
942 cinfo->completion_ops->resched_write(cinfo, req);
943 return;
944 }
945 /* Non-empty buckets hold a reference on the lseg. That ref
946 * is normally transferred to the COMMIT call and released
947 * there. It could also be released if the last req is pulled
948 * off due to a rewrite, in which case it will be done in
949 * pnfs_common_clear_request_commit
950 */
951 WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL);
952 buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg);
953 }
954 set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
955 cinfo->ds->nwritten++;
956
957 nfs_request_add_commit_list_locked(req, list, cinfo);
958 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
959 nfs_mark_page_unstable(req->wb_page, cinfo);
960}
961EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
962
963int
964pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
965{
966 int ret;
967
968 if (!pnfs_layoutcommit_outstanding(inode))
969 return 0;
970 ret = nfs_commit_inode(inode, FLUSH_SYNC);
971 if (ret < 0)
972 return ret;
973 if (datasync)
974 return 0;
975 return pnfs_layoutcommit_inode(inode, true);
976}
977EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
978