Loading...
1/*
2 * Copyright (c) 2014 Christoph Hellwig.
3 */
4#include <linux/blkdev.h>
5#include <linux/kmod.h>
6#include <linux/file.h>
7#include <linux/jhash.h>
8#include <linux/sched.h>
9#include <linux/sunrpc/addr.h>
10
11#include "pnfs.h"
12#include "netns.h"
13#include "trace.h"
14
15#define NFSDDBG_FACILITY NFSDDBG_PNFS
16
17struct nfs4_layout {
18 struct list_head lo_perstate;
19 struct nfs4_layout_stateid *lo_state;
20 struct nfsd4_layout_seg lo_seg;
21};
22
23static struct kmem_cache *nfs4_layout_cache;
24static struct kmem_cache *nfs4_layout_stateid_cache;
25
26static const struct nfsd4_callback_ops nfsd4_cb_layout_ops;
27static const struct lock_manager_operations nfsd4_layouts_lm_ops;
28
29const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] = {
30#ifdef CONFIG_NFSD_BLOCKLAYOUT
31 [LAYOUT_BLOCK_VOLUME] = &bl_layout_ops,
32#endif
33#ifdef CONFIG_NFSD_SCSILAYOUT
34 [LAYOUT_SCSI] = &scsi_layout_ops,
35#endif
36};
37
38/* pNFS device ID to export fsid mapping */
39#define DEVID_HASH_BITS 8
40#define DEVID_HASH_SIZE (1 << DEVID_HASH_BITS)
41#define DEVID_HASH_MASK (DEVID_HASH_SIZE - 1)
42static u64 nfsd_devid_seq = 1;
43static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE];
44static DEFINE_SPINLOCK(nfsd_devid_lock);
45
46static inline u32 devid_hashfn(u64 idx)
47{
48 return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK;
49}
50
51static void
52nfsd4_alloc_devid_map(const struct svc_fh *fhp)
53{
54 const struct knfsd_fh *fh = &fhp->fh_handle;
55 size_t fsid_len = key_len(fh->fh_fsid_type);
56 struct nfsd4_deviceid_map *map, *old;
57 int i;
58
59 map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL);
60 if (!map)
61 return;
62
63 map->fsid_type = fh->fh_fsid_type;
64 memcpy(&map->fsid, fh->fh_fsid, fsid_len);
65
66 spin_lock(&nfsd_devid_lock);
67 if (fhp->fh_export->ex_devid_map)
68 goto out_unlock;
69
70 for (i = 0; i < DEVID_HASH_SIZE; i++) {
71 list_for_each_entry(old, &nfsd_devid_hash[i], hash) {
72 if (old->fsid_type != fh->fh_fsid_type)
73 continue;
74 if (memcmp(old->fsid, fh->fh_fsid,
75 key_len(old->fsid_type)))
76 continue;
77
78 fhp->fh_export->ex_devid_map = old;
79 goto out_unlock;
80 }
81 }
82
83 map->idx = nfsd_devid_seq++;
84 list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]);
85 fhp->fh_export->ex_devid_map = map;
86 map = NULL;
87
88out_unlock:
89 spin_unlock(&nfsd_devid_lock);
90 kfree(map);
91}
92
93struct nfsd4_deviceid_map *
94nfsd4_find_devid_map(int idx)
95{
96 struct nfsd4_deviceid_map *map, *ret = NULL;
97
98 rcu_read_lock();
99 list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash)
100 if (map->idx == idx)
101 ret = map;
102 rcu_read_unlock();
103
104 return ret;
105}
106
107int
108nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp,
109 u32 device_generation)
110{
111 if (!fhp->fh_export->ex_devid_map) {
112 nfsd4_alloc_devid_map(fhp);
113 if (!fhp->fh_export->ex_devid_map)
114 return -ENOMEM;
115 }
116
117 id->fsid_idx = fhp->fh_export->ex_devid_map->idx;
118 id->generation = device_generation;
119 id->pad = 0;
120 return 0;
121}
122
123void nfsd4_setup_layout_type(struct svc_export *exp)
124{
125 struct super_block *sb = exp->ex_path.mnt->mnt_sb;
126
127 if (!(exp->ex_flags & NFSEXP_PNFS))
128 return;
129
130 /*
131 * Check if the file system supports exporting a block-like layout.
132 * If the block device supports reservations prefer the SCSI layout,
133 * otherwise advertise the block layout.
134 */
135#ifdef CONFIG_NFSD_BLOCKLAYOUT
136 if (sb->s_export_op->get_uuid &&
137 sb->s_export_op->map_blocks &&
138 sb->s_export_op->commit_blocks)
139 exp->ex_layout_type = LAYOUT_BLOCK_VOLUME;
140#endif
141#ifdef CONFIG_NFSD_SCSILAYOUT
142 /* overwrite block layout selection if needed */
143 if (sb->s_export_op->map_blocks &&
144 sb->s_export_op->commit_blocks &&
145 sb->s_bdev && sb->s_bdev->bd_disk->fops->pr_ops)
146 exp->ex_layout_type = LAYOUT_SCSI;
147#endif
148}
149
150static void
151nfsd4_free_layout_stateid(struct nfs4_stid *stid)
152{
153 struct nfs4_layout_stateid *ls = layoutstateid(stid);
154 struct nfs4_client *clp = ls->ls_stid.sc_client;
155 struct nfs4_file *fp = ls->ls_stid.sc_file;
156
157 trace_layoutstate_free(&ls->ls_stid.sc_stateid);
158
159 spin_lock(&clp->cl_lock);
160 list_del_init(&ls->ls_perclnt);
161 spin_unlock(&clp->cl_lock);
162
163 spin_lock(&fp->fi_lock);
164 list_del_init(&ls->ls_perfile);
165 spin_unlock(&fp->fi_lock);
166
167 vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls);
168 fput(ls->ls_file);
169
170 if (ls->ls_recalled)
171 atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
172
173 kmem_cache_free(nfs4_layout_stateid_cache, ls);
174}
175
176static int
177nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
178{
179 struct file_lock *fl;
180 int status;
181
182 fl = locks_alloc_lock();
183 if (!fl)
184 return -ENOMEM;
185 locks_init_lock(fl);
186 fl->fl_lmops = &nfsd4_layouts_lm_ops;
187 fl->fl_flags = FL_LAYOUT;
188 fl->fl_type = F_RDLCK;
189 fl->fl_end = OFFSET_MAX;
190 fl->fl_owner = ls;
191 fl->fl_pid = current->tgid;
192 fl->fl_file = ls->ls_file;
193
194 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL);
195 if (status) {
196 locks_free_lock(fl);
197 return status;
198 }
199 BUG_ON(fl != NULL);
200 return 0;
201}
202
203static struct nfs4_layout_stateid *
204nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
205 struct nfs4_stid *parent, u32 layout_type)
206{
207 struct nfs4_client *clp = cstate->clp;
208 struct nfs4_file *fp = parent->sc_file;
209 struct nfs4_layout_stateid *ls;
210 struct nfs4_stid *stp;
211
212 stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache);
213 if (!stp)
214 return NULL;
215 stp->sc_free = nfsd4_free_layout_stateid;
216 get_nfs4_file(fp);
217 stp->sc_file = fp;
218
219 ls = layoutstateid(stp);
220 INIT_LIST_HEAD(&ls->ls_perclnt);
221 INIT_LIST_HEAD(&ls->ls_perfile);
222 spin_lock_init(&ls->ls_lock);
223 INIT_LIST_HEAD(&ls->ls_layouts);
224 mutex_init(&ls->ls_mutex);
225 ls->ls_layout_type = layout_type;
226 nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
227 NFSPROC4_CLNT_CB_LAYOUT);
228
229 if (parent->sc_type == NFS4_DELEG_STID)
230 ls->ls_file = get_file(fp->fi_deleg_file);
231 else
232 ls->ls_file = find_any_file(fp);
233 BUG_ON(!ls->ls_file);
234
235 if (nfsd4_layout_setlease(ls)) {
236 fput(ls->ls_file);
237 put_nfs4_file(fp);
238 kmem_cache_free(nfs4_layout_stateid_cache, ls);
239 return NULL;
240 }
241
242 spin_lock(&clp->cl_lock);
243 stp->sc_type = NFS4_LAYOUT_STID;
244 list_add(&ls->ls_perclnt, &clp->cl_lo_states);
245 spin_unlock(&clp->cl_lock);
246
247 spin_lock(&fp->fi_lock);
248 list_add(&ls->ls_perfile, &fp->fi_lo_states);
249 spin_unlock(&fp->fi_lock);
250
251 trace_layoutstate_alloc(&ls->ls_stid.sc_stateid);
252 return ls;
253}
254
255__be32
256nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
257 struct nfsd4_compound_state *cstate, stateid_t *stateid,
258 bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
259{
260 struct nfs4_layout_stateid *ls;
261 struct nfs4_stid *stid;
262 unsigned char typemask = NFS4_LAYOUT_STID;
263 __be32 status;
264
265 if (create)
266 typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
267
268 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
269 net_generic(SVC_NET(rqstp), nfsd_net_id));
270 if (status)
271 goto out;
272
273 if (!fh_match(&cstate->current_fh.fh_handle,
274 &stid->sc_file->fi_fhandle)) {
275 status = nfserr_bad_stateid;
276 goto out_put_stid;
277 }
278
279 if (stid->sc_type != NFS4_LAYOUT_STID) {
280 ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
281 nfs4_put_stid(stid);
282
283 status = nfserr_jukebox;
284 if (!ls)
285 goto out;
286 mutex_lock(&ls->ls_mutex);
287 } else {
288 ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
289
290 status = nfserr_bad_stateid;
291 mutex_lock(&ls->ls_mutex);
292 if (stateid->si_generation > stid->sc_stateid.si_generation)
293 goto out_unlock_stid;
294 if (layout_type != ls->ls_layout_type)
295 goto out_unlock_stid;
296 }
297
298 *lsp = ls;
299 return 0;
300
301out_unlock_stid:
302 mutex_unlock(&ls->ls_mutex);
303out_put_stid:
304 nfs4_put_stid(stid);
305out:
306 return status;
307}
308
309static void
310nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
311{
312 spin_lock(&ls->ls_lock);
313 if (ls->ls_recalled)
314 goto out_unlock;
315
316 ls->ls_recalled = true;
317 atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
318 if (list_empty(&ls->ls_layouts))
319 goto out_unlock;
320
321 trace_layout_recall(&ls->ls_stid.sc_stateid);
322
323 atomic_inc(&ls->ls_stid.sc_count);
324 nfsd4_run_cb(&ls->ls_recall);
325
326out_unlock:
327 spin_unlock(&ls->ls_lock);
328}
329
330static inline u64
331layout_end(struct nfsd4_layout_seg *seg)
332{
333 u64 end = seg->offset + seg->length;
334 return end >= seg->offset ? end : NFS4_MAX_UINT64;
335}
336
337static void
338layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
339{
340 if (end == NFS4_MAX_UINT64)
341 lo->length = NFS4_MAX_UINT64;
342 else
343 lo->length = end - lo->offset;
344}
345
346static bool
347layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
348{
349 if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
350 return false;
351 if (layout_end(&lo->lo_seg) <= s->offset)
352 return false;
353 if (layout_end(s) <= lo->lo_seg.offset)
354 return false;
355 return true;
356}
357
358static bool
359layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
360{
361 if (lo->iomode != new->iomode)
362 return false;
363 if (layout_end(new) < lo->offset)
364 return false;
365 if (layout_end(lo) < new->offset)
366 return false;
367
368 lo->offset = min(lo->offset, new->offset);
369 layout_update_len(lo, max(layout_end(lo), layout_end(new)));
370 return true;
371}
372
373static __be32
374nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
375{
376 struct nfs4_file *fp = ls->ls_stid.sc_file;
377 struct nfs4_layout_stateid *l, *n;
378 __be32 nfserr = nfs_ok;
379
380 assert_spin_locked(&fp->fi_lock);
381
382 list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
383 if (l != ls) {
384 nfsd4_recall_file_layout(l);
385 nfserr = nfserr_recallconflict;
386 }
387 }
388
389 return nfserr;
390}
391
392__be32
393nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
394{
395 struct nfsd4_layout_seg *seg = &lgp->lg_seg;
396 struct nfs4_file *fp = ls->ls_stid.sc_file;
397 struct nfs4_layout *lp, *new = NULL;
398 __be32 nfserr;
399
400 spin_lock(&fp->fi_lock);
401 nfserr = nfsd4_recall_conflict(ls);
402 if (nfserr)
403 goto out;
404 spin_lock(&ls->ls_lock);
405 list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
406 if (layouts_try_merge(&lp->lo_seg, seg))
407 goto done;
408 }
409 spin_unlock(&ls->ls_lock);
410 spin_unlock(&fp->fi_lock);
411
412 new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
413 if (!new)
414 return nfserr_jukebox;
415 memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg));
416 new->lo_state = ls;
417
418 spin_lock(&fp->fi_lock);
419 nfserr = nfsd4_recall_conflict(ls);
420 if (nfserr)
421 goto out;
422 spin_lock(&ls->ls_lock);
423 list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
424 if (layouts_try_merge(&lp->lo_seg, seg))
425 goto done;
426 }
427
428 atomic_inc(&ls->ls_stid.sc_count);
429 list_add_tail(&new->lo_perstate, &ls->ls_layouts);
430 new = NULL;
431done:
432 nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid);
433 spin_unlock(&ls->ls_lock);
434out:
435 spin_unlock(&fp->fi_lock);
436 if (new)
437 kmem_cache_free(nfs4_layout_cache, new);
438 return nfserr;
439}
440
441static void
442nfsd4_free_layouts(struct list_head *reaplist)
443{
444 while (!list_empty(reaplist)) {
445 struct nfs4_layout *lp = list_first_entry(reaplist,
446 struct nfs4_layout, lo_perstate);
447
448 list_del(&lp->lo_perstate);
449 nfs4_put_stid(&lp->lo_state->ls_stid);
450 kmem_cache_free(nfs4_layout_cache, lp);
451 }
452}
453
454static void
455nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
456 struct list_head *reaplist)
457{
458 struct nfsd4_layout_seg *lo = &lp->lo_seg;
459 u64 end = layout_end(lo);
460
461 if (seg->offset <= lo->offset) {
462 if (layout_end(seg) >= end) {
463 list_move_tail(&lp->lo_perstate, reaplist);
464 return;
465 }
466 lo->offset = layout_end(seg);
467 } else {
468 /* retain the whole layout segment on a split. */
469 if (layout_end(seg) < end) {
470 dprintk("%s: split not supported\n", __func__);
471 return;
472 }
473 end = seg->offset;
474 }
475
476 layout_update_len(lo, end);
477}
478
479__be32
480nfsd4_return_file_layouts(struct svc_rqst *rqstp,
481 struct nfsd4_compound_state *cstate,
482 struct nfsd4_layoutreturn *lrp)
483{
484 struct nfs4_layout_stateid *ls;
485 struct nfs4_layout *lp, *n;
486 LIST_HEAD(reaplist);
487 __be32 nfserr;
488 int found = 0;
489
490 nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
491 false, lrp->lr_layout_type,
492 &ls);
493 if (nfserr) {
494 trace_layout_return_lookup_fail(&lrp->lr_sid);
495 return nfserr;
496 }
497
498 spin_lock(&ls->ls_lock);
499 list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
500 if (layouts_overlapping(lp, &lrp->lr_seg)) {
501 nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
502 found++;
503 }
504 }
505 if (!list_empty(&ls->ls_layouts)) {
506 if (found)
507 nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid);
508 lrp->lrs_present = 1;
509 } else {
510 trace_layoutstate_unhash(&ls->ls_stid.sc_stateid);
511 nfs4_unhash_stid(&ls->ls_stid);
512 lrp->lrs_present = 0;
513 }
514 spin_unlock(&ls->ls_lock);
515
516 mutex_unlock(&ls->ls_mutex);
517 nfs4_put_stid(&ls->ls_stid);
518 nfsd4_free_layouts(&reaplist);
519 return nfs_ok;
520}
521
522__be32
523nfsd4_return_client_layouts(struct svc_rqst *rqstp,
524 struct nfsd4_compound_state *cstate,
525 struct nfsd4_layoutreturn *lrp)
526{
527 struct nfs4_layout_stateid *ls, *n;
528 struct nfs4_client *clp = cstate->clp;
529 struct nfs4_layout *lp, *t;
530 LIST_HEAD(reaplist);
531
532 lrp->lrs_present = 0;
533
534 spin_lock(&clp->cl_lock);
535 list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
536 if (ls->ls_layout_type != lrp->lr_layout_type)
537 continue;
538
539 if (lrp->lr_return_type == RETURN_FSID &&
540 !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
541 &cstate->current_fh.fh_handle))
542 continue;
543
544 spin_lock(&ls->ls_lock);
545 list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
546 if (lrp->lr_seg.iomode == IOMODE_ANY ||
547 lrp->lr_seg.iomode == lp->lo_seg.iomode)
548 list_move_tail(&lp->lo_perstate, &reaplist);
549 }
550 spin_unlock(&ls->ls_lock);
551 }
552 spin_unlock(&clp->cl_lock);
553
554 nfsd4_free_layouts(&reaplist);
555 return 0;
556}
557
558static void
559nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
560 struct list_head *reaplist)
561{
562 spin_lock(&ls->ls_lock);
563 list_splice_init(&ls->ls_layouts, reaplist);
564 spin_unlock(&ls->ls_lock);
565}
566
567void
568nfsd4_return_all_client_layouts(struct nfs4_client *clp)
569{
570 struct nfs4_layout_stateid *ls, *n;
571 LIST_HEAD(reaplist);
572
573 spin_lock(&clp->cl_lock);
574 list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
575 nfsd4_return_all_layouts(ls, &reaplist);
576 spin_unlock(&clp->cl_lock);
577
578 nfsd4_free_layouts(&reaplist);
579}
580
581void
582nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
583{
584 struct nfs4_layout_stateid *ls, *n;
585 LIST_HEAD(reaplist);
586
587 spin_lock(&fp->fi_lock);
588 list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
589 if (ls->ls_stid.sc_client == clp)
590 nfsd4_return_all_layouts(ls, &reaplist);
591 }
592 spin_unlock(&fp->fi_lock);
593
594 nfsd4_free_layouts(&reaplist);
595}
596
597static void
598nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
599{
600 struct nfs4_client *clp = ls->ls_stid.sc_client;
601 char addr_str[INET6_ADDRSTRLEN];
602 static char *envp[] = {
603 "HOME=/",
604 "TERM=linux",
605 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
606 NULL
607 };
608 char *argv[8];
609 int error;
610
611 rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
612
613 printk(KERN_WARNING
614 "nfsd: client %s failed to respond to layout recall. "
615 " Fencing..\n", addr_str);
616
617 argv[0] = "/sbin/nfsd-recall-failed";
618 argv[1] = addr_str;
619 argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id;
620 argv[3] = NULL;
621
622 error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
623 if (error) {
624 printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
625 addr_str, error);
626 }
627}
628
629static void
630nfsd4_cb_layout_prepare(struct nfsd4_callback *cb)
631{
632 struct nfs4_layout_stateid *ls =
633 container_of(cb, struct nfs4_layout_stateid, ls_recall);
634
635 mutex_lock(&ls->ls_mutex);
636 nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid);
637 mutex_unlock(&ls->ls_mutex);
638}
639
640static int
641nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
642{
643 struct nfs4_layout_stateid *ls =
644 container_of(cb, struct nfs4_layout_stateid, ls_recall);
645 struct nfsd_net *nn;
646 ktime_t now, cutoff;
647 const struct nfsd4_layout_ops *ops;
648 LIST_HEAD(reaplist);
649
650
651 switch (task->tk_status) {
652 case 0:
653 case -NFS4ERR_DELAY:
654 /*
655 * Anything left? If not, then call it done. Note that we don't
656 * take the spinlock since this is an optimization and nothing
657 * should get added until the cb counter goes to zero.
658 */
659 if (list_empty(&ls->ls_layouts))
660 return 1;
661
662 /* Poll the client until it's done with the layout */
663 now = ktime_get();
664 nn = net_generic(ls->ls_stid.sc_client->net, nfsd_net_id);
665
666 /* Client gets 2 lease periods to return it */
667 cutoff = ktime_add_ns(task->tk_start,
668 nn->nfsd4_lease * NSEC_PER_SEC * 2);
669
670 if (ktime_before(now, cutoff)) {
671 rpc_delay(task, HZ/100); /* 10 mili-seconds */
672 return 0;
673 }
674 /* Fallthrough */
675 case -NFS4ERR_NOMATCHING_LAYOUT:
676 trace_layout_recall_done(&ls->ls_stid.sc_stateid);
677 task->tk_status = 0;
678 return 1;
679 default:
680 /*
681 * Unknown error or non-responding client, we'll need to fence.
682 */
683 trace_layout_recall_fail(&ls->ls_stid.sc_stateid);
684
685 ops = nfsd4_layout_ops[ls->ls_layout_type];
686 if (ops->fence_client)
687 ops->fence_client(ls);
688 else
689 nfsd4_cb_layout_fail(ls);
690 return -1;
691 }
692}
693
694static void
695nfsd4_cb_layout_release(struct nfsd4_callback *cb)
696{
697 struct nfs4_layout_stateid *ls =
698 container_of(cb, struct nfs4_layout_stateid, ls_recall);
699 LIST_HEAD(reaplist);
700
701 trace_layout_recall_release(&ls->ls_stid.sc_stateid);
702
703 nfsd4_return_all_layouts(ls, &reaplist);
704 nfsd4_free_layouts(&reaplist);
705 nfs4_put_stid(&ls->ls_stid);
706}
707
708static const struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
709 .prepare = nfsd4_cb_layout_prepare,
710 .done = nfsd4_cb_layout_done,
711 .release = nfsd4_cb_layout_release,
712};
713
714static bool
715nfsd4_layout_lm_break(struct file_lock *fl)
716{
717 /*
718 * We don't want the locks code to timeout the lease for us;
719 * we'll remove it ourself if a layout isn't returned
720 * in time:
721 */
722 fl->fl_break_time = 0;
723 nfsd4_recall_file_layout(fl->fl_owner);
724 return false;
725}
726
727static int
728nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
729 struct list_head *dispose)
730{
731 BUG_ON(!(arg & F_UNLCK));
732 return lease_modify(onlist, arg, dispose);
733}
734
735static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
736 .lm_break = nfsd4_layout_lm_break,
737 .lm_change = nfsd4_layout_lm_change,
738};
739
740int
741nfsd4_init_pnfs(void)
742{
743 int i;
744
745 for (i = 0; i < DEVID_HASH_SIZE; i++)
746 INIT_LIST_HEAD(&nfsd_devid_hash[i]);
747
748 nfs4_layout_cache = kmem_cache_create("nfs4_layout",
749 sizeof(struct nfs4_layout), 0, 0, NULL);
750 if (!nfs4_layout_cache)
751 return -ENOMEM;
752
753 nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
754 sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
755 if (!nfs4_layout_stateid_cache) {
756 kmem_cache_destroy(nfs4_layout_cache);
757 return -ENOMEM;
758 }
759 return 0;
760}
761
762void
763nfsd4_exit_pnfs(void)
764{
765 int i;
766
767 kmem_cache_destroy(nfs4_layout_cache);
768 kmem_cache_destroy(nfs4_layout_stateid_cache);
769
770 for (i = 0; i < DEVID_HASH_SIZE; i++) {
771 struct nfsd4_deviceid_map *map, *n;
772
773 list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
774 kfree(map);
775 }
776}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2014 Christoph Hellwig.
4 */
5#include <linux/blkdev.h>
6#include <linux/kmod.h>
7#include <linux/file.h>
8#include <linux/jhash.h>
9#include <linux/sched.h>
10#include <linux/sunrpc/addr.h>
11
12#include "pnfs.h"
13#include "netns.h"
14#include "trace.h"
15
16#define NFSDDBG_FACILITY NFSDDBG_PNFS
17
18struct nfs4_layout {
19 struct list_head lo_perstate;
20 struct nfs4_layout_stateid *lo_state;
21 struct nfsd4_layout_seg lo_seg;
22};
23
24static struct kmem_cache *nfs4_layout_cache;
25static struct kmem_cache *nfs4_layout_stateid_cache;
26
27static const struct nfsd4_callback_ops nfsd4_cb_layout_ops;
28static const struct lease_manager_operations nfsd4_layouts_lm_ops;
29
30const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] = {
31#ifdef CONFIG_NFSD_FLEXFILELAYOUT
32 [LAYOUT_FLEX_FILES] = &ff_layout_ops,
33#endif
34#ifdef CONFIG_NFSD_BLOCKLAYOUT
35 [LAYOUT_BLOCK_VOLUME] = &bl_layout_ops,
36#endif
37#ifdef CONFIG_NFSD_SCSILAYOUT
38 [LAYOUT_SCSI] = &scsi_layout_ops,
39#endif
40};
41
42/* pNFS device ID to export fsid mapping */
43#define DEVID_HASH_BITS 8
44#define DEVID_HASH_SIZE (1 << DEVID_HASH_BITS)
45#define DEVID_HASH_MASK (DEVID_HASH_SIZE - 1)
46static u64 nfsd_devid_seq = 1;
47static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE];
48static DEFINE_SPINLOCK(nfsd_devid_lock);
49
50static inline u32 devid_hashfn(u64 idx)
51{
52 return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK;
53}
54
55static void
56nfsd4_alloc_devid_map(const struct svc_fh *fhp)
57{
58 const struct knfsd_fh *fh = &fhp->fh_handle;
59 size_t fsid_len = key_len(fh->fh_fsid_type);
60 struct nfsd4_deviceid_map *map, *old;
61 int i;
62
63 map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL);
64 if (!map)
65 return;
66
67 map->fsid_type = fh->fh_fsid_type;
68 memcpy(&map->fsid, fh->fh_fsid, fsid_len);
69
70 spin_lock(&nfsd_devid_lock);
71 if (fhp->fh_export->ex_devid_map)
72 goto out_unlock;
73
74 for (i = 0; i < DEVID_HASH_SIZE; i++) {
75 list_for_each_entry(old, &nfsd_devid_hash[i], hash) {
76 if (old->fsid_type != fh->fh_fsid_type)
77 continue;
78 if (memcmp(old->fsid, fh->fh_fsid,
79 key_len(old->fsid_type)))
80 continue;
81
82 fhp->fh_export->ex_devid_map = old;
83 goto out_unlock;
84 }
85 }
86
87 map->idx = nfsd_devid_seq++;
88 list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]);
89 fhp->fh_export->ex_devid_map = map;
90 map = NULL;
91
92out_unlock:
93 spin_unlock(&nfsd_devid_lock);
94 kfree(map);
95}
96
97struct nfsd4_deviceid_map *
98nfsd4_find_devid_map(int idx)
99{
100 struct nfsd4_deviceid_map *map, *ret = NULL;
101
102 rcu_read_lock();
103 list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash)
104 if (map->idx == idx)
105 ret = map;
106 rcu_read_unlock();
107
108 return ret;
109}
110
111int
112nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp,
113 u32 device_generation)
114{
115 if (!fhp->fh_export->ex_devid_map) {
116 nfsd4_alloc_devid_map(fhp);
117 if (!fhp->fh_export->ex_devid_map)
118 return -ENOMEM;
119 }
120
121 id->fsid_idx = fhp->fh_export->ex_devid_map->idx;
122 id->generation = device_generation;
123 id->pad = 0;
124 return 0;
125}
126
127void nfsd4_setup_layout_type(struct svc_export *exp)
128{
129#if defined(CONFIG_NFSD_BLOCKLAYOUT) || defined(CONFIG_NFSD_SCSILAYOUT)
130 struct super_block *sb = exp->ex_path.mnt->mnt_sb;
131#endif
132
133 if (!(exp->ex_flags & NFSEXP_PNFS))
134 return;
135
136#ifdef CONFIG_NFSD_FLEXFILELAYOUT
137 exp->ex_layout_types |= 1 << LAYOUT_FLEX_FILES;
138#endif
139#ifdef CONFIG_NFSD_BLOCKLAYOUT
140 if (sb->s_export_op->get_uuid &&
141 sb->s_export_op->map_blocks &&
142 sb->s_export_op->commit_blocks)
143 exp->ex_layout_types |= 1 << LAYOUT_BLOCK_VOLUME;
144#endif
145#ifdef CONFIG_NFSD_SCSILAYOUT
146 if (sb->s_export_op->map_blocks &&
147 sb->s_export_op->commit_blocks &&
148 sb->s_bdev &&
149 sb->s_bdev->bd_disk->fops->pr_ops &&
150 sb->s_bdev->bd_disk->fops->get_unique_id)
151 exp->ex_layout_types |= 1 << LAYOUT_SCSI;
152#endif
153}
154
155void nfsd4_close_layout(struct nfs4_layout_stateid *ls)
156{
157 struct nfsd_file *fl;
158
159 spin_lock(&ls->ls_stid.sc_file->fi_lock);
160 fl = ls->ls_file;
161 ls->ls_file = NULL;
162 spin_unlock(&ls->ls_stid.sc_file->fi_lock);
163
164 if (fl) {
165 if (!nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
166 kernel_setlease(fl->nf_file, F_UNLCK, NULL,
167 (void **)&ls);
168 nfsd_file_put(fl);
169 }
170}
171
172static void
173nfsd4_free_layout_stateid(struct nfs4_stid *stid)
174{
175 struct nfs4_layout_stateid *ls = layoutstateid(stid);
176 struct nfs4_client *clp = ls->ls_stid.sc_client;
177 struct nfs4_file *fp = ls->ls_stid.sc_file;
178
179 trace_nfsd_layoutstate_free(&ls->ls_stid.sc_stateid);
180
181 spin_lock(&clp->cl_lock);
182 list_del_init(&ls->ls_perclnt);
183 spin_unlock(&clp->cl_lock);
184
185 spin_lock(&fp->fi_lock);
186 list_del_init(&ls->ls_perfile);
187 spin_unlock(&fp->fi_lock);
188
189 nfsd4_close_layout(ls);
190
191 if (ls->ls_recalled)
192 atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
193
194 kmem_cache_free(nfs4_layout_stateid_cache, ls);
195}
196
197static int
198nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
199{
200 struct file_lease *fl;
201 int status;
202
203 if (nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
204 return 0;
205
206 fl = locks_alloc_lease();
207 if (!fl)
208 return -ENOMEM;
209 locks_init_lease(fl);
210 fl->fl_lmops = &nfsd4_layouts_lm_ops;
211 fl->c.flc_flags = FL_LAYOUT;
212 fl->c.flc_type = F_RDLCK;
213 fl->c.flc_owner = ls;
214 fl->c.flc_pid = current->tgid;
215 fl->c.flc_file = ls->ls_file->nf_file;
216
217 status = kernel_setlease(fl->c.flc_file, fl->c.flc_type, &fl, NULL);
218 if (status) {
219 locks_free_lease(fl);
220 return status;
221 }
222 BUG_ON(fl != NULL);
223 return 0;
224}
225
226static struct nfs4_layout_stateid *
227nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
228 struct nfs4_stid *parent, u32 layout_type)
229{
230 struct nfs4_client *clp = cstate->clp;
231 struct nfs4_file *fp = parent->sc_file;
232 struct nfs4_layout_stateid *ls;
233 struct nfs4_stid *stp;
234
235 stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache,
236 nfsd4_free_layout_stateid);
237 if (!stp)
238 return NULL;
239
240 get_nfs4_file(fp);
241 stp->sc_file = fp;
242
243 ls = layoutstateid(stp);
244 INIT_LIST_HEAD(&ls->ls_perclnt);
245 INIT_LIST_HEAD(&ls->ls_perfile);
246 spin_lock_init(&ls->ls_lock);
247 INIT_LIST_HEAD(&ls->ls_layouts);
248 mutex_init(&ls->ls_mutex);
249 ls->ls_layout_type = layout_type;
250 nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
251 NFSPROC4_CLNT_CB_LAYOUT);
252
253 if (parent->sc_type == SC_TYPE_DELEG)
254 ls->ls_file = nfsd_file_get(fp->fi_deleg_file);
255 else
256 ls->ls_file = find_any_file(fp);
257 BUG_ON(!ls->ls_file);
258
259 if (nfsd4_layout_setlease(ls)) {
260 nfsd_file_put(ls->ls_file);
261 put_nfs4_file(fp);
262 kmem_cache_free(nfs4_layout_stateid_cache, ls);
263 return NULL;
264 }
265
266 spin_lock(&clp->cl_lock);
267 stp->sc_type = SC_TYPE_LAYOUT;
268 list_add(&ls->ls_perclnt, &clp->cl_lo_states);
269 spin_unlock(&clp->cl_lock);
270
271 spin_lock(&fp->fi_lock);
272 list_add(&ls->ls_perfile, &fp->fi_lo_states);
273 spin_unlock(&fp->fi_lock);
274
275 trace_nfsd_layoutstate_alloc(&ls->ls_stid.sc_stateid);
276 return ls;
277}
278
279__be32
280nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
281 struct nfsd4_compound_state *cstate, stateid_t *stateid,
282 bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
283{
284 struct nfs4_layout_stateid *ls;
285 struct nfs4_stid *stid;
286 unsigned short typemask = SC_TYPE_LAYOUT;
287 __be32 status;
288
289 if (create)
290 typemask |= (SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG);
291
292 status = nfsd4_lookup_stateid(cstate, stateid, typemask, 0, &stid,
293 net_generic(SVC_NET(rqstp), nfsd_net_id));
294 if (status)
295 goto out;
296
297 if (!fh_match(&cstate->current_fh.fh_handle,
298 &stid->sc_file->fi_fhandle)) {
299 status = nfserr_bad_stateid;
300 goto out_put_stid;
301 }
302
303 if (stid->sc_type != SC_TYPE_LAYOUT) {
304 ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
305 nfs4_put_stid(stid);
306
307 status = nfserr_jukebox;
308 if (!ls)
309 goto out;
310 mutex_lock(&ls->ls_mutex);
311 } else {
312 ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
313
314 status = nfserr_bad_stateid;
315 mutex_lock(&ls->ls_mutex);
316 if (nfsd4_stateid_generation_after(stateid, &stid->sc_stateid))
317 goto out_unlock_stid;
318 if (layout_type != ls->ls_layout_type)
319 goto out_unlock_stid;
320 }
321
322 *lsp = ls;
323 return 0;
324
325out_unlock_stid:
326 mutex_unlock(&ls->ls_mutex);
327out_put_stid:
328 nfs4_put_stid(stid);
329out:
330 return status;
331}
332
333static void
334nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
335{
336 spin_lock(&ls->ls_lock);
337 if (ls->ls_recalled)
338 goto out_unlock;
339
340 if (list_empty(&ls->ls_layouts))
341 goto out_unlock;
342
343 ls->ls_recalled = true;
344 atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
345 trace_nfsd_layout_recall(&ls->ls_stid.sc_stateid);
346
347 refcount_inc(&ls->ls_stid.sc_count);
348 nfsd4_run_cb(&ls->ls_recall);
349
350out_unlock:
351 spin_unlock(&ls->ls_lock);
352}
353
354static inline u64
355layout_end(struct nfsd4_layout_seg *seg)
356{
357 u64 end = seg->offset + seg->length;
358 return end >= seg->offset ? end : NFS4_MAX_UINT64;
359}
360
361static void
362layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
363{
364 if (end == NFS4_MAX_UINT64)
365 lo->length = NFS4_MAX_UINT64;
366 else
367 lo->length = end - lo->offset;
368}
369
370static bool
371layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
372{
373 if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
374 return false;
375 if (layout_end(&lo->lo_seg) <= s->offset)
376 return false;
377 if (layout_end(s) <= lo->lo_seg.offset)
378 return false;
379 return true;
380}
381
382static bool
383layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
384{
385 if (lo->iomode != new->iomode)
386 return false;
387 if (layout_end(new) < lo->offset)
388 return false;
389 if (layout_end(lo) < new->offset)
390 return false;
391
392 lo->offset = min(lo->offset, new->offset);
393 layout_update_len(lo, max(layout_end(lo), layout_end(new)));
394 return true;
395}
396
397static __be32
398nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
399{
400 struct nfs4_file *fp = ls->ls_stid.sc_file;
401 struct nfs4_layout_stateid *l, *n;
402 __be32 nfserr = nfs_ok;
403
404 assert_spin_locked(&fp->fi_lock);
405
406 list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
407 if (l != ls) {
408 nfsd4_recall_file_layout(l);
409 nfserr = nfserr_recallconflict;
410 }
411 }
412
413 return nfserr;
414}
415
416__be32
417nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
418{
419 struct nfsd4_layout_seg *seg = &lgp->lg_seg;
420 struct nfs4_file *fp = ls->ls_stid.sc_file;
421 struct nfs4_layout *lp, *new = NULL;
422 __be32 nfserr;
423
424 spin_lock(&fp->fi_lock);
425 nfserr = nfsd4_recall_conflict(ls);
426 if (nfserr)
427 goto out;
428 spin_lock(&ls->ls_lock);
429 list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
430 if (layouts_try_merge(&lp->lo_seg, seg))
431 goto done;
432 }
433 spin_unlock(&ls->ls_lock);
434 spin_unlock(&fp->fi_lock);
435
436 new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
437 if (!new)
438 return nfserr_jukebox;
439 memcpy(&new->lo_seg, seg, sizeof(new->lo_seg));
440 new->lo_state = ls;
441
442 spin_lock(&fp->fi_lock);
443 nfserr = nfsd4_recall_conflict(ls);
444 if (nfserr)
445 goto out;
446 spin_lock(&ls->ls_lock);
447 list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
448 if (layouts_try_merge(&lp->lo_seg, seg))
449 goto done;
450 }
451
452 refcount_inc(&ls->ls_stid.sc_count);
453 list_add_tail(&new->lo_perstate, &ls->ls_layouts);
454 new = NULL;
455done:
456 nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid);
457 spin_unlock(&ls->ls_lock);
458out:
459 spin_unlock(&fp->fi_lock);
460 if (new)
461 kmem_cache_free(nfs4_layout_cache, new);
462 return nfserr;
463}
464
465static void
466nfsd4_free_layouts(struct list_head *reaplist)
467{
468 while (!list_empty(reaplist)) {
469 struct nfs4_layout *lp = list_first_entry(reaplist,
470 struct nfs4_layout, lo_perstate);
471
472 list_del(&lp->lo_perstate);
473 nfs4_put_stid(&lp->lo_state->ls_stid);
474 kmem_cache_free(nfs4_layout_cache, lp);
475 }
476}
477
478static void
479nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
480 struct list_head *reaplist)
481{
482 struct nfsd4_layout_seg *lo = &lp->lo_seg;
483 u64 end = layout_end(lo);
484
485 if (seg->offset <= lo->offset) {
486 if (layout_end(seg) >= end) {
487 list_move_tail(&lp->lo_perstate, reaplist);
488 return;
489 }
490 lo->offset = layout_end(seg);
491 } else {
492 /* retain the whole layout segment on a split. */
493 if (layout_end(seg) < end) {
494 dprintk("%s: split not supported\n", __func__);
495 return;
496 }
497 end = seg->offset;
498 }
499
500 layout_update_len(lo, end);
501}
502
503__be32
504nfsd4_return_file_layouts(struct svc_rqst *rqstp,
505 struct nfsd4_compound_state *cstate,
506 struct nfsd4_layoutreturn *lrp)
507{
508 struct nfs4_layout_stateid *ls;
509 struct nfs4_layout *lp, *n;
510 LIST_HEAD(reaplist);
511 __be32 nfserr;
512 int found = 0;
513
514 nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
515 false, lrp->lr_layout_type,
516 &ls);
517 if (nfserr) {
518 trace_nfsd_layout_return_lookup_fail(&lrp->lr_sid);
519 return nfserr;
520 }
521
522 spin_lock(&ls->ls_lock);
523 list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
524 if (layouts_overlapping(lp, &lrp->lr_seg)) {
525 nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
526 found++;
527 }
528 }
529 if (!list_empty(&ls->ls_layouts)) {
530 if (found)
531 nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid);
532 lrp->lrs_present = true;
533 } else {
534 trace_nfsd_layoutstate_unhash(&ls->ls_stid.sc_stateid);
535 ls->ls_stid.sc_status |= SC_STATUS_CLOSED;
536 lrp->lrs_present = false;
537 }
538 spin_unlock(&ls->ls_lock);
539
540 mutex_unlock(&ls->ls_mutex);
541 nfs4_put_stid(&ls->ls_stid);
542 nfsd4_free_layouts(&reaplist);
543 return nfs_ok;
544}
545
546__be32
547nfsd4_return_client_layouts(struct svc_rqst *rqstp,
548 struct nfsd4_compound_state *cstate,
549 struct nfsd4_layoutreturn *lrp)
550{
551 struct nfs4_layout_stateid *ls, *n;
552 struct nfs4_client *clp = cstate->clp;
553 struct nfs4_layout *lp, *t;
554 LIST_HEAD(reaplist);
555
556 lrp->lrs_present = false;
557
558 spin_lock(&clp->cl_lock);
559 list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
560 if (ls->ls_layout_type != lrp->lr_layout_type)
561 continue;
562
563 if (lrp->lr_return_type == RETURN_FSID &&
564 !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
565 &cstate->current_fh.fh_handle))
566 continue;
567
568 spin_lock(&ls->ls_lock);
569 list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
570 if (lrp->lr_seg.iomode == IOMODE_ANY ||
571 lrp->lr_seg.iomode == lp->lo_seg.iomode)
572 list_move_tail(&lp->lo_perstate, &reaplist);
573 }
574 spin_unlock(&ls->ls_lock);
575 }
576 spin_unlock(&clp->cl_lock);
577
578 nfsd4_free_layouts(&reaplist);
579 return 0;
580}
581
582static void
583nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
584 struct list_head *reaplist)
585{
586 spin_lock(&ls->ls_lock);
587 list_splice_init(&ls->ls_layouts, reaplist);
588 spin_unlock(&ls->ls_lock);
589}
590
591void
592nfsd4_return_all_client_layouts(struct nfs4_client *clp)
593{
594 struct nfs4_layout_stateid *ls, *n;
595 LIST_HEAD(reaplist);
596
597 spin_lock(&clp->cl_lock);
598 list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
599 nfsd4_return_all_layouts(ls, &reaplist);
600 spin_unlock(&clp->cl_lock);
601
602 nfsd4_free_layouts(&reaplist);
603}
604
605void
606nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
607{
608 struct nfs4_layout_stateid *ls, *n;
609 LIST_HEAD(reaplist);
610
611 spin_lock(&fp->fi_lock);
612 list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
613 if (ls->ls_stid.sc_client == clp)
614 nfsd4_return_all_layouts(ls, &reaplist);
615 }
616 spin_unlock(&fp->fi_lock);
617
618 nfsd4_free_layouts(&reaplist);
619}
620
621static void
622nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls, struct nfsd_file *file)
623{
624 struct nfs4_client *clp = ls->ls_stid.sc_client;
625 char addr_str[INET6_ADDRSTRLEN];
626 static char const nfsd_recall_failed[] = "/sbin/nfsd-recall-failed";
627 static char *envp[] = {
628 "HOME=/",
629 "TERM=linux",
630 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
631 NULL
632 };
633 char *argv[8];
634 int error;
635
636 rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
637
638 printk(KERN_WARNING
639 "nfsd: client %s failed to respond to layout recall. "
640 " Fencing..\n", addr_str);
641
642 argv[0] = (char *)nfsd_recall_failed;
643 argv[1] = addr_str;
644 argv[2] = file->nf_file->f_path.mnt->mnt_sb->s_id;
645 argv[3] = NULL;
646
647 error = call_usermodehelper(nfsd_recall_failed, argv, envp,
648 UMH_WAIT_PROC);
649 if (error) {
650 printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
651 addr_str, error);
652 }
653}
654
655static void
656nfsd4_cb_layout_prepare(struct nfsd4_callback *cb)
657{
658 struct nfs4_layout_stateid *ls =
659 container_of(cb, struct nfs4_layout_stateid, ls_recall);
660
661 mutex_lock(&ls->ls_mutex);
662 nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid);
663 mutex_unlock(&ls->ls_mutex);
664}
665
666static int
667nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
668{
669 struct nfs4_layout_stateid *ls =
670 container_of(cb, struct nfs4_layout_stateid, ls_recall);
671 struct nfsd_net *nn;
672 ktime_t now, cutoff;
673 const struct nfsd4_layout_ops *ops;
674 struct nfsd_file *fl;
675
676 trace_nfsd_cb_layout_done(&ls->ls_stid.sc_stateid, task);
677 switch (task->tk_status) {
678 case 0:
679 case -NFS4ERR_DELAY:
680 /*
681 * Anything left? If not, then call it done. Note that we don't
682 * take the spinlock since this is an optimization and nothing
683 * should get added until the cb counter goes to zero.
684 */
685 if (list_empty(&ls->ls_layouts))
686 return 1;
687
688 /* Poll the client until it's done with the layout */
689 now = ktime_get();
690 nn = net_generic(ls->ls_stid.sc_client->net, nfsd_net_id);
691
692 /* Client gets 2 lease periods to return it */
693 cutoff = ktime_add_ns(task->tk_start,
694 (u64)nn->nfsd4_lease * NSEC_PER_SEC * 2);
695
696 if (ktime_before(now, cutoff)) {
697 rpc_delay(task, HZ/100); /* 10 mili-seconds */
698 return 0;
699 }
700 fallthrough;
701 default:
702 /*
703 * Unknown error or non-responding client, we'll need to fence.
704 */
705 trace_nfsd_layout_recall_fail(&ls->ls_stid.sc_stateid);
706 rcu_read_lock();
707 fl = nfsd_file_get(ls->ls_file);
708 rcu_read_unlock();
709 if (fl) {
710 ops = nfsd4_layout_ops[ls->ls_layout_type];
711 if (ops->fence_client)
712 ops->fence_client(ls, fl);
713 else
714 nfsd4_cb_layout_fail(ls, fl);
715 nfsd_file_put(fl);
716 }
717 return 1;
718 case -NFS4ERR_NOMATCHING_LAYOUT:
719 trace_nfsd_layout_recall_done(&ls->ls_stid.sc_stateid);
720 task->tk_status = 0;
721 return 1;
722 }
723}
724
725static void
726nfsd4_cb_layout_release(struct nfsd4_callback *cb)
727{
728 struct nfs4_layout_stateid *ls =
729 container_of(cb, struct nfs4_layout_stateid, ls_recall);
730 LIST_HEAD(reaplist);
731
732 trace_nfsd_layout_recall_release(&ls->ls_stid.sc_stateid);
733
734 nfsd4_return_all_layouts(ls, &reaplist);
735 nfsd4_free_layouts(&reaplist);
736 nfs4_put_stid(&ls->ls_stid);
737}
738
739static const struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
740 .prepare = nfsd4_cb_layout_prepare,
741 .done = nfsd4_cb_layout_done,
742 .release = nfsd4_cb_layout_release,
743 .opcode = OP_CB_LAYOUTRECALL,
744};
745
746static bool
747nfsd4_layout_lm_break(struct file_lease *fl)
748{
749 /*
750 * We don't want the locks code to timeout the lease for us;
751 * we'll remove it ourself if a layout isn't returned
752 * in time:
753 */
754 fl->fl_break_time = 0;
755 nfsd4_recall_file_layout(fl->c.flc_owner);
756 return false;
757}
758
759static int
760nfsd4_layout_lm_change(struct file_lease *onlist, int arg,
761 struct list_head *dispose)
762{
763 BUG_ON(!(arg & F_UNLCK));
764 return lease_modify(onlist, arg, dispose);
765}
766
767static const struct lease_manager_operations nfsd4_layouts_lm_ops = {
768 .lm_break = nfsd4_layout_lm_break,
769 .lm_change = nfsd4_layout_lm_change,
770};
771
772int
773nfsd4_init_pnfs(void)
774{
775 int i;
776
777 for (i = 0; i < DEVID_HASH_SIZE; i++)
778 INIT_LIST_HEAD(&nfsd_devid_hash[i]);
779
780 nfs4_layout_cache = KMEM_CACHE(nfs4_layout, 0);
781 if (!nfs4_layout_cache)
782 return -ENOMEM;
783
784 nfs4_layout_stateid_cache = KMEM_CACHE(nfs4_layout_stateid, 0);
785 if (!nfs4_layout_stateid_cache) {
786 kmem_cache_destroy(nfs4_layout_cache);
787 return -ENOMEM;
788 }
789 return 0;
790}
791
792void
793nfsd4_exit_pnfs(void)
794{
795 int i;
796
797 kmem_cache_destroy(nfs4_layout_cache);
798 kmem_cache_destroy(nfs4_layout_stateid_cache);
799
800 for (i = 0; i < DEVID_HASH_SIZE; i++) {
801 struct nfsd4_deviceid_map *map, *n;
802
803 list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
804 kfree(map);
805 }
806}