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