xref: /openbmc/linux/fs/nfsd/nfs4layouts.c (revision 74a22e8f)
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 
18 struct nfs4_layout {
19 	struct list_head		lo_perstate;
20 	struct nfs4_layout_stateid	*lo_state;
21 	struct nfsd4_layout_seg		lo_seg;
22 };
23 
24 static struct kmem_cache *nfs4_layout_cache;
25 static struct kmem_cache *nfs4_layout_stateid_cache;
26 
27 static const struct nfsd4_callback_ops nfsd4_cb_layout_ops;
28 static const struct lock_manager_operations nfsd4_layouts_lm_ops;
29 
30 const 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)
46 static u64 nfsd_devid_seq = 1;
47 static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE];
48 static DEFINE_SPINLOCK(nfsd_devid_lock);
49 
50 static inline u32 devid_hashfn(u64 idx)
51 {
52 	return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK;
53 }
54 
55 static void
56 nfsd4_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 
92 out_unlock:
93 	spin_unlock(&nfsd_devid_lock);
94 	kfree(map);
95 }
96 
97 struct nfsd4_deviceid_map *
98 nfsd4_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 
111 int
112 nfsd4_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 
127 void 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 && sb->s_bdev->bd_disk->fops->pr_ops &&
149 		blk_queue_scsi_passthrough(sb->s_bdev->bd_disk->queue))
150 		exp->ex_layout_types |= 1 << LAYOUT_SCSI;
151 #endif
152 }
153 
154 static void
155 nfsd4_free_layout_stateid(struct nfs4_stid *stid)
156 {
157 	struct nfs4_layout_stateid *ls = layoutstateid(stid);
158 	struct nfs4_client *clp = ls->ls_stid.sc_client;
159 	struct nfs4_file *fp = ls->ls_stid.sc_file;
160 
161 	trace_nfsd_layoutstate_free(&ls->ls_stid.sc_stateid);
162 
163 	spin_lock(&clp->cl_lock);
164 	list_del_init(&ls->ls_perclnt);
165 	spin_unlock(&clp->cl_lock);
166 
167 	spin_lock(&fp->fi_lock);
168 	list_del_init(&ls->ls_perfile);
169 	spin_unlock(&fp->fi_lock);
170 
171 	if (!nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
172 		vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls);
173 	fput(ls->ls_file);
174 
175 	if (ls->ls_recalled)
176 		atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
177 
178 	kmem_cache_free(nfs4_layout_stateid_cache, ls);
179 }
180 
181 static int
182 nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
183 {
184 	struct file_lock *fl;
185 	int status;
186 
187 	if (nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
188 		return 0;
189 
190 	fl = locks_alloc_lock();
191 	if (!fl)
192 		return -ENOMEM;
193 	locks_init_lock(fl);
194 	fl->fl_lmops = &nfsd4_layouts_lm_ops;
195 	fl->fl_flags = FL_LAYOUT;
196 	fl->fl_type = F_RDLCK;
197 	fl->fl_end = OFFSET_MAX;
198 	fl->fl_owner = ls;
199 	fl->fl_pid = current->tgid;
200 	fl->fl_file = ls->ls_file;
201 
202 	status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL);
203 	if (status) {
204 		locks_free_lock(fl);
205 		return status;
206 	}
207 	BUG_ON(fl != NULL);
208 	return 0;
209 }
210 
211 static struct nfs4_layout_stateid *
212 nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
213 		struct nfs4_stid *parent, u32 layout_type)
214 {
215 	struct nfs4_client *clp = cstate->clp;
216 	struct nfs4_file *fp = parent->sc_file;
217 	struct nfs4_layout_stateid *ls;
218 	struct nfs4_stid *stp;
219 
220 	stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache,
221 					nfsd4_free_layout_stateid);
222 	if (!stp)
223 		return NULL;
224 
225 	get_nfs4_file(fp);
226 	stp->sc_file = fp;
227 
228 	ls = layoutstateid(stp);
229 	INIT_LIST_HEAD(&ls->ls_perclnt);
230 	INIT_LIST_HEAD(&ls->ls_perfile);
231 	spin_lock_init(&ls->ls_lock);
232 	INIT_LIST_HEAD(&ls->ls_layouts);
233 	mutex_init(&ls->ls_mutex);
234 	ls->ls_layout_type = layout_type;
235 	nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
236 			NFSPROC4_CLNT_CB_LAYOUT);
237 
238 	if (parent->sc_type == NFS4_DELEG_STID)
239 		ls->ls_file = get_file(fp->fi_deleg_file);
240 	else
241 		ls->ls_file = find_any_file(fp);
242 	BUG_ON(!ls->ls_file);
243 
244 	if (nfsd4_layout_setlease(ls)) {
245 		fput(ls->ls_file);
246 		put_nfs4_file(fp);
247 		kmem_cache_free(nfs4_layout_stateid_cache, ls);
248 		return NULL;
249 	}
250 
251 	spin_lock(&clp->cl_lock);
252 	stp->sc_type = NFS4_LAYOUT_STID;
253 	list_add(&ls->ls_perclnt, &clp->cl_lo_states);
254 	spin_unlock(&clp->cl_lock);
255 
256 	spin_lock(&fp->fi_lock);
257 	list_add(&ls->ls_perfile, &fp->fi_lo_states);
258 	spin_unlock(&fp->fi_lock);
259 
260 	trace_nfsd_layoutstate_alloc(&ls->ls_stid.sc_stateid);
261 	return ls;
262 }
263 
264 __be32
265 nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
266 		struct nfsd4_compound_state *cstate, stateid_t *stateid,
267 		bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
268 {
269 	struct nfs4_layout_stateid *ls;
270 	struct nfs4_stid *stid;
271 	unsigned char typemask = NFS4_LAYOUT_STID;
272 	__be32 status;
273 
274 	if (create)
275 		typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
276 
277 	status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
278 			net_generic(SVC_NET(rqstp), nfsd_net_id));
279 	if (status)
280 		goto out;
281 
282 	if (!fh_match(&cstate->current_fh.fh_handle,
283 		      &stid->sc_file->fi_fhandle)) {
284 		status = nfserr_bad_stateid;
285 		goto out_put_stid;
286 	}
287 
288 	if (stid->sc_type != NFS4_LAYOUT_STID) {
289 		ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
290 		nfs4_put_stid(stid);
291 
292 		status = nfserr_jukebox;
293 		if (!ls)
294 			goto out;
295 		mutex_lock(&ls->ls_mutex);
296 	} else {
297 		ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
298 
299 		status = nfserr_bad_stateid;
300 		mutex_lock(&ls->ls_mutex);
301 		if (nfsd4_stateid_generation_after(stateid, &stid->sc_stateid))
302 			goto out_unlock_stid;
303 		if (layout_type != ls->ls_layout_type)
304 			goto out_unlock_stid;
305 	}
306 
307 	*lsp = ls;
308 	return 0;
309 
310 out_unlock_stid:
311 	mutex_unlock(&ls->ls_mutex);
312 out_put_stid:
313 	nfs4_put_stid(stid);
314 out:
315 	return status;
316 }
317 
318 static void
319 nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
320 {
321 	spin_lock(&ls->ls_lock);
322 	if (ls->ls_recalled)
323 		goto out_unlock;
324 
325 	ls->ls_recalled = true;
326 	atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
327 	if (list_empty(&ls->ls_layouts))
328 		goto out_unlock;
329 
330 	trace_nfsd_layout_recall(&ls->ls_stid.sc_stateid);
331 
332 	refcount_inc(&ls->ls_stid.sc_count);
333 	nfsd4_run_cb(&ls->ls_recall);
334 
335 out_unlock:
336 	spin_unlock(&ls->ls_lock);
337 }
338 
339 static inline u64
340 layout_end(struct nfsd4_layout_seg *seg)
341 {
342 	u64 end = seg->offset + seg->length;
343 	return end >= seg->offset ? end : NFS4_MAX_UINT64;
344 }
345 
346 static void
347 layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
348 {
349 	if (end == NFS4_MAX_UINT64)
350 		lo->length = NFS4_MAX_UINT64;
351 	else
352 		lo->length = end - lo->offset;
353 }
354 
355 static bool
356 layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
357 {
358 	if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
359 		return false;
360 	if (layout_end(&lo->lo_seg) <= s->offset)
361 		return false;
362 	if (layout_end(s) <= lo->lo_seg.offset)
363 		return false;
364 	return true;
365 }
366 
367 static bool
368 layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
369 {
370 	if (lo->iomode != new->iomode)
371 		return false;
372 	if (layout_end(new) < lo->offset)
373 		return false;
374 	if (layout_end(lo) < new->offset)
375 		return false;
376 
377 	lo->offset = min(lo->offset, new->offset);
378 	layout_update_len(lo, max(layout_end(lo), layout_end(new)));
379 	return true;
380 }
381 
382 static __be32
383 nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
384 {
385 	struct nfs4_file *fp = ls->ls_stid.sc_file;
386 	struct nfs4_layout_stateid *l, *n;
387 	__be32 nfserr = nfs_ok;
388 
389 	assert_spin_locked(&fp->fi_lock);
390 
391 	list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
392 		if (l != ls) {
393 			nfsd4_recall_file_layout(l);
394 			nfserr = nfserr_recallconflict;
395 		}
396 	}
397 
398 	return nfserr;
399 }
400 
401 __be32
402 nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
403 {
404 	struct nfsd4_layout_seg *seg = &lgp->lg_seg;
405 	struct nfs4_file *fp = ls->ls_stid.sc_file;
406 	struct nfs4_layout *lp, *new = NULL;
407 	__be32 nfserr;
408 
409 	spin_lock(&fp->fi_lock);
410 	nfserr = nfsd4_recall_conflict(ls);
411 	if (nfserr)
412 		goto out;
413 	spin_lock(&ls->ls_lock);
414 	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
415 		if (layouts_try_merge(&lp->lo_seg, seg))
416 			goto done;
417 	}
418 	spin_unlock(&ls->ls_lock);
419 	spin_unlock(&fp->fi_lock);
420 
421 	new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
422 	if (!new)
423 		return nfserr_jukebox;
424 	memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg));
425 	new->lo_state = ls;
426 
427 	spin_lock(&fp->fi_lock);
428 	nfserr = nfsd4_recall_conflict(ls);
429 	if (nfserr)
430 		goto out;
431 	spin_lock(&ls->ls_lock);
432 	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
433 		if (layouts_try_merge(&lp->lo_seg, seg))
434 			goto done;
435 	}
436 
437 	refcount_inc(&ls->ls_stid.sc_count);
438 	list_add_tail(&new->lo_perstate, &ls->ls_layouts);
439 	new = NULL;
440 done:
441 	nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid);
442 	spin_unlock(&ls->ls_lock);
443 out:
444 	spin_unlock(&fp->fi_lock);
445 	if (new)
446 		kmem_cache_free(nfs4_layout_cache, new);
447 	return nfserr;
448 }
449 
450 static void
451 nfsd4_free_layouts(struct list_head *reaplist)
452 {
453 	while (!list_empty(reaplist)) {
454 		struct nfs4_layout *lp = list_first_entry(reaplist,
455 				struct nfs4_layout, lo_perstate);
456 
457 		list_del(&lp->lo_perstate);
458 		nfs4_put_stid(&lp->lo_state->ls_stid);
459 		kmem_cache_free(nfs4_layout_cache, lp);
460 	}
461 }
462 
463 static void
464 nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
465 		struct list_head *reaplist)
466 {
467 	struct nfsd4_layout_seg *lo = &lp->lo_seg;
468 	u64 end = layout_end(lo);
469 
470 	if (seg->offset <= lo->offset) {
471 		if (layout_end(seg) >= end) {
472 			list_move_tail(&lp->lo_perstate, reaplist);
473 			return;
474 		}
475 		lo->offset = layout_end(seg);
476 	} else {
477 		/* retain the whole layout segment on a split. */
478 		if (layout_end(seg) < end) {
479 			dprintk("%s: split not supported\n", __func__);
480 			return;
481 		}
482 		end = seg->offset;
483 	}
484 
485 	layout_update_len(lo, end);
486 }
487 
488 __be32
489 nfsd4_return_file_layouts(struct svc_rqst *rqstp,
490 		struct nfsd4_compound_state *cstate,
491 		struct nfsd4_layoutreturn *lrp)
492 {
493 	struct nfs4_layout_stateid *ls;
494 	struct nfs4_layout *lp, *n;
495 	LIST_HEAD(reaplist);
496 	__be32 nfserr;
497 	int found = 0;
498 
499 	nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
500 						false, lrp->lr_layout_type,
501 						&ls);
502 	if (nfserr) {
503 		trace_nfsd_layout_return_lookup_fail(&lrp->lr_sid);
504 		return nfserr;
505 	}
506 
507 	spin_lock(&ls->ls_lock);
508 	list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
509 		if (layouts_overlapping(lp, &lrp->lr_seg)) {
510 			nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
511 			found++;
512 		}
513 	}
514 	if (!list_empty(&ls->ls_layouts)) {
515 		if (found)
516 			nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid);
517 		lrp->lrs_present = 1;
518 	} else {
519 		trace_nfsd_layoutstate_unhash(&ls->ls_stid.sc_stateid);
520 		nfs4_unhash_stid(&ls->ls_stid);
521 		lrp->lrs_present = 0;
522 	}
523 	spin_unlock(&ls->ls_lock);
524 
525 	mutex_unlock(&ls->ls_mutex);
526 	nfs4_put_stid(&ls->ls_stid);
527 	nfsd4_free_layouts(&reaplist);
528 	return nfs_ok;
529 }
530 
531 __be32
532 nfsd4_return_client_layouts(struct svc_rqst *rqstp,
533 		struct nfsd4_compound_state *cstate,
534 		struct nfsd4_layoutreturn *lrp)
535 {
536 	struct nfs4_layout_stateid *ls, *n;
537 	struct nfs4_client *clp = cstate->clp;
538 	struct nfs4_layout *lp, *t;
539 	LIST_HEAD(reaplist);
540 
541 	lrp->lrs_present = 0;
542 
543 	spin_lock(&clp->cl_lock);
544 	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
545 		if (ls->ls_layout_type != lrp->lr_layout_type)
546 			continue;
547 
548 		if (lrp->lr_return_type == RETURN_FSID &&
549 		    !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
550 				   &cstate->current_fh.fh_handle))
551 			continue;
552 
553 		spin_lock(&ls->ls_lock);
554 		list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
555 			if (lrp->lr_seg.iomode == IOMODE_ANY ||
556 			    lrp->lr_seg.iomode == lp->lo_seg.iomode)
557 				list_move_tail(&lp->lo_perstate, &reaplist);
558 		}
559 		spin_unlock(&ls->ls_lock);
560 	}
561 	spin_unlock(&clp->cl_lock);
562 
563 	nfsd4_free_layouts(&reaplist);
564 	return 0;
565 }
566 
567 static void
568 nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
569 		struct list_head *reaplist)
570 {
571 	spin_lock(&ls->ls_lock);
572 	list_splice_init(&ls->ls_layouts, reaplist);
573 	spin_unlock(&ls->ls_lock);
574 }
575 
576 void
577 nfsd4_return_all_client_layouts(struct nfs4_client *clp)
578 {
579 	struct nfs4_layout_stateid *ls, *n;
580 	LIST_HEAD(reaplist);
581 
582 	spin_lock(&clp->cl_lock);
583 	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
584 		nfsd4_return_all_layouts(ls, &reaplist);
585 	spin_unlock(&clp->cl_lock);
586 
587 	nfsd4_free_layouts(&reaplist);
588 }
589 
590 void
591 nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
592 {
593 	struct nfs4_layout_stateid *ls, *n;
594 	LIST_HEAD(reaplist);
595 
596 	spin_lock(&fp->fi_lock);
597 	list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
598 		if (ls->ls_stid.sc_client == clp)
599 			nfsd4_return_all_layouts(ls, &reaplist);
600 	}
601 	spin_unlock(&fp->fi_lock);
602 
603 	nfsd4_free_layouts(&reaplist);
604 }
605 
606 static void
607 nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
608 {
609 	struct nfs4_client *clp = ls->ls_stid.sc_client;
610 	char addr_str[INET6_ADDRSTRLEN];
611 	static char const nfsd_recall_failed[] = "/sbin/nfsd-recall-failed";
612 	static char *envp[] = {
613 		"HOME=/",
614 		"TERM=linux",
615 		"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
616 		NULL
617 	};
618 	char *argv[8];
619 	int error;
620 
621 	rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
622 
623 	printk(KERN_WARNING
624 		"nfsd: client %s failed to respond to layout recall. "
625 		"  Fencing..\n", addr_str);
626 
627 	argv[0] = (char *)nfsd_recall_failed;
628 	argv[1] = addr_str;
629 	argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id;
630 	argv[3] = NULL;
631 
632 	error = call_usermodehelper(nfsd_recall_failed, argv, envp,
633 				    UMH_WAIT_PROC);
634 	if (error) {
635 		printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
636 			addr_str, error);
637 	}
638 }
639 
640 static void
641 nfsd4_cb_layout_prepare(struct nfsd4_callback *cb)
642 {
643 	struct nfs4_layout_stateid *ls =
644 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
645 
646 	mutex_lock(&ls->ls_mutex);
647 	nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid);
648 	mutex_unlock(&ls->ls_mutex);
649 }
650 
651 static int
652 nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
653 {
654 	struct nfs4_layout_stateid *ls =
655 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
656 	struct nfsd_net *nn;
657 	ktime_t now, cutoff;
658 	const struct nfsd4_layout_ops *ops;
659 
660 
661 	switch (task->tk_status) {
662 	case 0:
663 	case -NFS4ERR_DELAY:
664 		/*
665 		 * Anything left? If not, then call it done. Note that we don't
666 		 * take the spinlock since this is an optimization and nothing
667 		 * should get added until the cb counter goes to zero.
668 		 */
669 		if (list_empty(&ls->ls_layouts))
670 			return 1;
671 
672 		/* Poll the client until it's done with the layout */
673 		now = ktime_get();
674 		nn = net_generic(ls->ls_stid.sc_client->net, nfsd_net_id);
675 
676 		/* Client gets 2 lease periods to return it */
677 		cutoff = ktime_add_ns(task->tk_start,
678 					 nn->nfsd4_lease * NSEC_PER_SEC * 2);
679 
680 		if (ktime_before(now, cutoff)) {
681 			rpc_delay(task, HZ/100); /* 10 mili-seconds */
682 			return 0;
683 		}
684 		/* Fallthrough */
685 	default:
686 		/*
687 		 * Unknown error or non-responding client, we'll need to fence.
688 		 */
689 		trace_nfsd_layout_recall_fail(&ls->ls_stid.sc_stateid);
690 
691 		ops = nfsd4_layout_ops[ls->ls_layout_type];
692 		if (ops->fence_client)
693 			ops->fence_client(ls);
694 		else
695 			nfsd4_cb_layout_fail(ls);
696 		return 1;
697 	case -NFS4ERR_NOMATCHING_LAYOUT:
698 		trace_nfsd_layout_recall_done(&ls->ls_stid.sc_stateid);
699 		task->tk_status = 0;
700 		return 1;
701 	}
702 }
703 
704 static void
705 nfsd4_cb_layout_release(struct nfsd4_callback *cb)
706 {
707 	struct nfs4_layout_stateid *ls =
708 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
709 	LIST_HEAD(reaplist);
710 
711 	trace_nfsd_layout_recall_release(&ls->ls_stid.sc_stateid);
712 
713 	nfsd4_return_all_layouts(ls, &reaplist);
714 	nfsd4_free_layouts(&reaplist);
715 	nfs4_put_stid(&ls->ls_stid);
716 }
717 
718 static const struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
719 	.prepare	= nfsd4_cb_layout_prepare,
720 	.done		= nfsd4_cb_layout_done,
721 	.release	= nfsd4_cb_layout_release,
722 };
723 
724 static bool
725 nfsd4_layout_lm_break(struct file_lock *fl)
726 {
727 	/*
728 	 * We don't want the locks code to timeout the lease for us;
729 	 * we'll remove it ourself if a layout isn't returned
730 	 * in time:
731 	 */
732 	fl->fl_break_time = 0;
733 	nfsd4_recall_file_layout(fl->fl_owner);
734 	return false;
735 }
736 
737 static int
738 nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
739 		struct list_head *dispose)
740 {
741 	BUG_ON(!(arg & F_UNLCK));
742 	return lease_modify(onlist, arg, dispose);
743 }
744 
745 static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
746 	.lm_break	= nfsd4_layout_lm_break,
747 	.lm_change	= nfsd4_layout_lm_change,
748 };
749 
750 int
751 nfsd4_init_pnfs(void)
752 {
753 	int i;
754 
755 	for (i = 0; i < DEVID_HASH_SIZE; i++)
756 		INIT_LIST_HEAD(&nfsd_devid_hash[i]);
757 
758 	nfs4_layout_cache = kmem_cache_create("nfs4_layout",
759 			sizeof(struct nfs4_layout), 0, 0, NULL);
760 	if (!nfs4_layout_cache)
761 		return -ENOMEM;
762 
763 	nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
764 			sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
765 	if (!nfs4_layout_stateid_cache) {
766 		kmem_cache_destroy(nfs4_layout_cache);
767 		return -ENOMEM;
768 	}
769 	return 0;
770 }
771 
772 void
773 nfsd4_exit_pnfs(void)
774 {
775 	int i;
776 
777 	kmem_cache_destroy(nfs4_layout_cache);
778 	kmem_cache_destroy(nfs4_layout_stateid_cache);
779 
780 	for (i = 0; i < DEVID_HASH_SIZE; i++) {
781 		struct nfsd4_deviceid_map *map, *n;
782 
783 		list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
784 			kfree(map);
785 	}
786 }
787