xref: /openbmc/linux/fs/nfsd/nfs4layouts.c (revision 8730046c)
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 
17 struct nfs4_layout {
18 	struct list_head		lo_perstate;
19 	struct nfs4_layout_stateid	*lo_state;
20 	struct nfsd4_layout_seg		lo_seg;
21 };
22 
23 static struct kmem_cache *nfs4_layout_cache;
24 static struct kmem_cache *nfs4_layout_stateid_cache;
25 
26 static const struct nfsd4_callback_ops nfsd4_cb_layout_ops;
27 static const struct lock_manager_operations nfsd4_layouts_lm_ops;
28 
29 const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] =  {
30 #ifdef CONFIG_NFSD_FLEXFILELAYOUT
31 	[LAYOUT_FLEX_FILES]	= &ff_layout_ops,
32 #endif
33 #ifdef CONFIG_NFSD_BLOCKLAYOUT
34 	[LAYOUT_BLOCK_VOLUME]	= &bl_layout_ops,
35 #endif
36 #ifdef CONFIG_NFSD_SCSILAYOUT
37 	[LAYOUT_SCSI]		= &scsi_layout_ops,
38 #endif
39 };
40 
41 /* pNFS device ID to export fsid mapping */
42 #define DEVID_HASH_BITS	8
43 #define DEVID_HASH_SIZE	(1 << DEVID_HASH_BITS)
44 #define DEVID_HASH_MASK	(DEVID_HASH_SIZE - 1)
45 static u64 nfsd_devid_seq = 1;
46 static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE];
47 static DEFINE_SPINLOCK(nfsd_devid_lock);
48 
49 static inline u32 devid_hashfn(u64 idx)
50 {
51 	return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK;
52 }
53 
54 static void
55 nfsd4_alloc_devid_map(const struct svc_fh *fhp)
56 {
57 	const struct knfsd_fh *fh = &fhp->fh_handle;
58 	size_t fsid_len = key_len(fh->fh_fsid_type);
59 	struct nfsd4_deviceid_map *map, *old;
60 	int i;
61 
62 	map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL);
63 	if (!map)
64 		return;
65 
66 	map->fsid_type = fh->fh_fsid_type;
67 	memcpy(&map->fsid, fh->fh_fsid, fsid_len);
68 
69 	spin_lock(&nfsd_devid_lock);
70 	if (fhp->fh_export->ex_devid_map)
71 		goto out_unlock;
72 
73 	for (i = 0; i < DEVID_HASH_SIZE; i++) {
74 		list_for_each_entry(old, &nfsd_devid_hash[i], hash) {
75 			if (old->fsid_type != fh->fh_fsid_type)
76 				continue;
77 			if (memcmp(old->fsid, fh->fh_fsid,
78 					key_len(old->fsid_type)))
79 				continue;
80 
81 			fhp->fh_export->ex_devid_map = old;
82 			goto out_unlock;
83 		}
84 	}
85 
86 	map->idx = nfsd_devid_seq++;
87 	list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]);
88 	fhp->fh_export->ex_devid_map = map;
89 	map = NULL;
90 
91 out_unlock:
92 	spin_unlock(&nfsd_devid_lock);
93 	kfree(map);
94 }
95 
96 struct nfsd4_deviceid_map *
97 nfsd4_find_devid_map(int idx)
98 {
99 	struct nfsd4_deviceid_map *map, *ret = NULL;
100 
101 	rcu_read_lock();
102 	list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash)
103 		if (map->idx == idx)
104 			ret = map;
105 	rcu_read_unlock();
106 
107 	return ret;
108 }
109 
110 int
111 nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp,
112 		u32 device_generation)
113 {
114 	if (!fhp->fh_export->ex_devid_map) {
115 		nfsd4_alloc_devid_map(fhp);
116 		if (!fhp->fh_export->ex_devid_map)
117 			return -ENOMEM;
118 	}
119 
120 	id->fsid_idx = fhp->fh_export->ex_devid_map->idx;
121 	id->generation = device_generation;
122 	id->pad = 0;
123 	return 0;
124 }
125 
126 void nfsd4_setup_layout_type(struct svc_export *exp)
127 {
128 #if defined(CONFIG_NFSD_BLOCKLAYOUT) || defined(CONFIG_NFSD_SCSILAYOUT)
129 	struct super_block *sb = exp->ex_path.mnt->mnt_sb;
130 #endif
131 
132 	if (!(exp->ex_flags & NFSEXP_PNFS))
133 		return;
134 
135 	/*
136 	 * If flex file is configured, use it by default. Otherwise
137 	 * check if the file system supports exporting a block-like layout.
138 	 * If the block device supports reservations prefer the SCSI layout,
139 	 * otherwise advertise the block layout.
140 	 */
141 #ifdef CONFIG_NFSD_FLEXFILELAYOUT
142 	exp->ex_layout_types |= 1 << LAYOUT_FLEX_FILES;
143 #endif
144 #ifdef CONFIG_NFSD_BLOCKLAYOUT
145 	/* overwrite flex file layout selection if needed */
146 	if (sb->s_export_op->get_uuid &&
147 	    sb->s_export_op->map_blocks &&
148 	    sb->s_export_op->commit_blocks)
149 		exp->ex_layout_types |= 1 << LAYOUT_BLOCK_VOLUME;
150 #endif
151 #ifdef CONFIG_NFSD_SCSILAYOUT
152 	/* overwrite block layout selection if needed */
153 	if (sb->s_export_op->map_blocks &&
154 	    sb->s_export_op->commit_blocks &&
155 	    sb->s_bdev && sb->s_bdev->bd_disk->fops->pr_ops)
156 		exp->ex_layout_types |= 1 << LAYOUT_SCSI;
157 #endif
158 }
159 
160 static void
161 nfsd4_free_layout_stateid(struct nfs4_stid *stid)
162 {
163 	struct nfs4_layout_stateid *ls = layoutstateid(stid);
164 	struct nfs4_client *clp = ls->ls_stid.sc_client;
165 	struct nfs4_file *fp = ls->ls_stid.sc_file;
166 
167 	trace_layoutstate_free(&ls->ls_stid.sc_stateid);
168 
169 	spin_lock(&clp->cl_lock);
170 	list_del_init(&ls->ls_perclnt);
171 	spin_unlock(&clp->cl_lock);
172 
173 	spin_lock(&fp->fi_lock);
174 	list_del_init(&ls->ls_perfile);
175 	spin_unlock(&fp->fi_lock);
176 
177 	if (!nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
178 		vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls);
179 	fput(ls->ls_file);
180 
181 	if (ls->ls_recalled)
182 		atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
183 
184 	kmem_cache_free(nfs4_layout_stateid_cache, ls);
185 }
186 
187 static int
188 nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
189 {
190 	struct file_lock *fl;
191 	int status;
192 
193 	if (nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
194 		return 0;
195 
196 	fl = locks_alloc_lock();
197 	if (!fl)
198 		return -ENOMEM;
199 	locks_init_lock(fl);
200 	fl->fl_lmops = &nfsd4_layouts_lm_ops;
201 	fl->fl_flags = FL_LAYOUT;
202 	fl->fl_type = F_RDLCK;
203 	fl->fl_end = OFFSET_MAX;
204 	fl->fl_owner = ls;
205 	fl->fl_pid = current->tgid;
206 	fl->fl_file = ls->ls_file;
207 
208 	status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL);
209 	if (status) {
210 		locks_free_lock(fl);
211 		return status;
212 	}
213 	BUG_ON(fl != NULL);
214 	return 0;
215 }
216 
217 static struct nfs4_layout_stateid *
218 nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
219 		struct nfs4_stid *parent, u32 layout_type)
220 {
221 	struct nfs4_client *clp = cstate->clp;
222 	struct nfs4_file *fp = parent->sc_file;
223 	struct nfs4_layout_stateid *ls;
224 	struct nfs4_stid *stp;
225 
226 	stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache);
227 	if (!stp)
228 		return NULL;
229 	stp->sc_free = nfsd4_free_layout_stateid;
230 	get_nfs4_file(fp);
231 	stp->sc_file = fp;
232 
233 	ls = layoutstateid(stp);
234 	INIT_LIST_HEAD(&ls->ls_perclnt);
235 	INIT_LIST_HEAD(&ls->ls_perfile);
236 	spin_lock_init(&ls->ls_lock);
237 	INIT_LIST_HEAD(&ls->ls_layouts);
238 	mutex_init(&ls->ls_mutex);
239 	ls->ls_layout_type = layout_type;
240 	nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
241 			NFSPROC4_CLNT_CB_LAYOUT);
242 
243 	if (parent->sc_type == NFS4_DELEG_STID)
244 		ls->ls_file = get_file(fp->fi_deleg_file);
245 	else
246 		ls->ls_file = find_any_file(fp);
247 	BUG_ON(!ls->ls_file);
248 
249 	if (nfsd4_layout_setlease(ls)) {
250 		fput(ls->ls_file);
251 		put_nfs4_file(fp);
252 		kmem_cache_free(nfs4_layout_stateid_cache, ls);
253 		return NULL;
254 	}
255 
256 	spin_lock(&clp->cl_lock);
257 	stp->sc_type = NFS4_LAYOUT_STID;
258 	list_add(&ls->ls_perclnt, &clp->cl_lo_states);
259 	spin_unlock(&clp->cl_lock);
260 
261 	spin_lock(&fp->fi_lock);
262 	list_add(&ls->ls_perfile, &fp->fi_lo_states);
263 	spin_unlock(&fp->fi_lock);
264 
265 	trace_layoutstate_alloc(&ls->ls_stid.sc_stateid);
266 	return ls;
267 }
268 
269 __be32
270 nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
271 		struct nfsd4_compound_state *cstate, stateid_t *stateid,
272 		bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
273 {
274 	struct nfs4_layout_stateid *ls;
275 	struct nfs4_stid *stid;
276 	unsigned char typemask = NFS4_LAYOUT_STID;
277 	__be32 status;
278 
279 	if (create)
280 		typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
281 
282 	status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
283 			net_generic(SVC_NET(rqstp), nfsd_net_id));
284 	if (status)
285 		goto out;
286 
287 	if (!fh_match(&cstate->current_fh.fh_handle,
288 		      &stid->sc_file->fi_fhandle)) {
289 		status = nfserr_bad_stateid;
290 		goto out_put_stid;
291 	}
292 
293 	if (stid->sc_type != NFS4_LAYOUT_STID) {
294 		ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
295 		nfs4_put_stid(stid);
296 
297 		status = nfserr_jukebox;
298 		if (!ls)
299 			goto out;
300 		mutex_lock(&ls->ls_mutex);
301 	} else {
302 		ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
303 
304 		status = nfserr_bad_stateid;
305 		mutex_lock(&ls->ls_mutex);
306 		if (nfsd4_stateid_generation_after(stateid, &stid->sc_stateid))
307 			goto out_unlock_stid;
308 		if (layout_type != ls->ls_layout_type)
309 			goto out_unlock_stid;
310 	}
311 
312 	*lsp = ls;
313 	return 0;
314 
315 out_unlock_stid:
316 	mutex_unlock(&ls->ls_mutex);
317 out_put_stid:
318 	nfs4_put_stid(stid);
319 out:
320 	return status;
321 }
322 
323 static void
324 nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
325 {
326 	spin_lock(&ls->ls_lock);
327 	if (ls->ls_recalled)
328 		goto out_unlock;
329 
330 	ls->ls_recalled = true;
331 	atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
332 	if (list_empty(&ls->ls_layouts))
333 		goto out_unlock;
334 
335 	trace_layout_recall(&ls->ls_stid.sc_stateid);
336 
337 	atomic_inc(&ls->ls_stid.sc_count);
338 	nfsd4_run_cb(&ls->ls_recall);
339 
340 out_unlock:
341 	spin_unlock(&ls->ls_lock);
342 }
343 
344 static inline u64
345 layout_end(struct nfsd4_layout_seg *seg)
346 {
347 	u64 end = seg->offset + seg->length;
348 	return end >= seg->offset ? end : NFS4_MAX_UINT64;
349 }
350 
351 static void
352 layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
353 {
354 	if (end == NFS4_MAX_UINT64)
355 		lo->length = NFS4_MAX_UINT64;
356 	else
357 		lo->length = end - lo->offset;
358 }
359 
360 static bool
361 layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
362 {
363 	if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
364 		return false;
365 	if (layout_end(&lo->lo_seg) <= s->offset)
366 		return false;
367 	if (layout_end(s) <= lo->lo_seg.offset)
368 		return false;
369 	return true;
370 }
371 
372 static bool
373 layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
374 {
375 	if (lo->iomode != new->iomode)
376 		return false;
377 	if (layout_end(new) < lo->offset)
378 		return false;
379 	if (layout_end(lo) < new->offset)
380 		return false;
381 
382 	lo->offset = min(lo->offset, new->offset);
383 	layout_update_len(lo, max(layout_end(lo), layout_end(new)));
384 	return true;
385 }
386 
387 static __be32
388 nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
389 {
390 	struct nfs4_file *fp = ls->ls_stid.sc_file;
391 	struct nfs4_layout_stateid *l, *n;
392 	__be32 nfserr = nfs_ok;
393 
394 	assert_spin_locked(&fp->fi_lock);
395 
396 	list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
397 		if (l != ls) {
398 			nfsd4_recall_file_layout(l);
399 			nfserr = nfserr_recallconflict;
400 		}
401 	}
402 
403 	return nfserr;
404 }
405 
406 __be32
407 nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
408 {
409 	struct nfsd4_layout_seg *seg = &lgp->lg_seg;
410 	struct nfs4_file *fp = ls->ls_stid.sc_file;
411 	struct nfs4_layout *lp, *new = NULL;
412 	__be32 nfserr;
413 
414 	spin_lock(&fp->fi_lock);
415 	nfserr = nfsd4_recall_conflict(ls);
416 	if (nfserr)
417 		goto out;
418 	spin_lock(&ls->ls_lock);
419 	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
420 		if (layouts_try_merge(&lp->lo_seg, seg))
421 			goto done;
422 	}
423 	spin_unlock(&ls->ls_lock);
424 	spin_unlock(&fp->fi_lock);
425 
426 	new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
427 	if (!new)
428 		return nfserr_jukebox;
429 	memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg));
430 	new->lo_state = ls;
431 
432 	spin_lock(&fp->fi_lock);
433 	nfserr = nfsd4_recall_conflict(ls);
434 	if (nfserr)
435 		goto out;
436 	spin_lock(&ls->ls_lock);
437 	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
438 		if (layouts_try_merge(&lp->lo_seg, seg))
439 			goto done;
440 	}
441 
442 	atomic_inc(&ls->ls_stid.sc_count);
443 	list_add_tail(&new->lo_perstate, &ls->ls_layouts);
444 	new = NULL;
445 done:
446 	nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid);
447 	spin_unlock(&ls->ls_lock);
448 out:
449 	spin_unlock(&fp->fi_lock);
450 	if (new)
451 		kmem_cache_free(nfs4_layout_cache, new);
452 	return nfserr;
453 }
454 
455 static void
456 nfsd4_free_layouts(struct list_head *reaplist)
457 {
458 	while (!list_empty(reaplist)) {
459 		struct nfs4_layout *lp = list_first_entry(reaplist,
460 				struct nfs4_layout, lo_perstate);
461 
462 		list_del(&lp->lo_perstate);
463 		nfs4_put_stid(&lp->lo_state->ls_stid);
464 		kmem_cache_free(nfs4_layout_cache, lp);
465 	}
466 }
467 
468 static void
469 nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
470 		struct list_head *reaplist)
471 {
472 	struct nfsd4_layout_seg *lo = &lp->lo_seg;
473 	u64 end = layout_end(lo);
474 
475 	if (seg->offset <= lo->offset) {
476 		if (layout_end(seg) >= end) {
477 			list_move_tail(&lp->lo_perstate, reaplist);
478 			return;
479 		}
480 		lo->offset = layout_end(seg);
481 	} else {
482 		/* retain the whole layout segment on a split. */
483 		if (layout_end(seg) < end) {
484 			dprintk("%s: split not supported\n", __func__);
485 			return;
486 		}
487 		end = seg->offset;
488 	}
489 
490 	layout_update_len(lo, end);
491 }
492 
493 __be32
494 nfsd4_return_file_layouts(struct svc_rqst *rqstp,
495 		struct nfsd4_compound_state *cstate,
496 		struct nfsd4_layoutreturn *lrp)
497 {
498 	struct nfs4_layout_stateid *ls;
499 	struct nfs4_layout *lp, *n;
500 	LIST_HEAD(reaplist);
501 	__be32 nfserr;
502 	int found = 0;
503 
504 	nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
505 						false, lrp->lr_layout_type,
506 						&ls);
507 	if (nfserr) {
508 		trace_layout_return_lookup_fail(&lrp->lr_sid);
509 		return nfserr;
510 	}
511 
512 	spin_lock(&ls->ls_lock);
513 	list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
514 		if (layouts_overlapping(lp, &lrp->lr_seg)) {
515 			nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
516 			found++;
517 		}
518 	}
519 	if (!list_empty(&ls->ls_layouts)) {
520 		if (found)
521 			nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid);
522 		lrp->lrs_present = 1;
523 	} else {
524 		trace_layoutstate_unhash(&ls->ls_stid.sc_stateid);
525 		nfs4_unhash_stid(&ls->ls_stid);
526 		lrp->lrs_present = 0;
527 	}
528 	spin_unlock(&ls->ls_lock);
529 
530 	mutex_unlock(&ls->ls_mutex);
531 	nfs4_put_stid(&ls->ls_stid);
532 	nfsd4_free_layouts(&reaplist);
533 	return nfs_ok;
534 }
535 
536 __be32
537 nfsd4_return_client_layouts(struct svc_rqst *rqstp,
538 		struct nfsd4_compound_state *cstate,
539 		struct nfsd4_layoutreturn *lrp)
540 {
541 	struct nfs4_layout_stateid *ls, *n;
542 	struct nfs4_client *clp = cstate->clp;
543 	struct nfs4_layout *lp, *t;
544 	LIST_HEAD(reaplist);
545 
546 	lrp->lrs_present = 0;
547 
548 	spin_lock(&clp->cl_lock);
549 	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
550 		if (ls->ls_layout_type != lrp->lr_layout_type)
551 			continue;
552 
553 		if (lrp->lr_return_type == RETURN_FSID &&
554 		    !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
555 				   &cstate->current_fh.fh_handle))
556 			continue;
557 
558 		spin_lock(&ls->ls_lock);
559 		list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
560 			if (lrp->lr_seg.iomode == IOMODE_ANY ||
561 			    lrp->lr_seg.iomode == lp->lo_seg.iomode)
562 				list_move_tail(&lp->lo_perstate, &reaplist);
563 		}
564 		spin_unlock(&ls->ls_lock);
565 	}
566 	spin_unlock(&clp->cl_lock);
567 
568 	nfsd4_free_layouts(&reaplist);
569 	return 0;
570 }
571 
572 static void
573 nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
574 		struct list_head *reaplist)
575 {
576 	spin_lock(&ls->ls_lock);
577 	list_splice_init(&ls->ls_layouts, reaplist);
578 	spin_unlock(&ls->ls_lock);
579 }
580 
581 void
582 nfsd4_return_all_client_layouts(struct nfs4_client *clp)
583 {
584 	struct nfs4_layout_stateid *ls, *n;
585 	LIST_HEAD(reaplist);
586 
587 	spin_lock(&clp->cl_lock);
588 	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
589 		nfsd4_return_all_layouts(ls, &reaplist);
590 	spin_unlock(&clp->cl_lock);
591 
592 	nfsd4_free_layouts(&reaplist);
593 }
594 
595 void
596 nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
597 {
598 	struct nfs4_layout_stateid *ls, *n;
599 	LIST_HEAD(reaplist);
600 
601 	spin_lock(&fp->fi_lock);
602 	list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
603 		if (ls->ls_stid.sc_client == clp)
604 			nfsd4_return_all_layouts(ls, &reaplist);
605 	}
606 	spin_unlock(&fp->fi_lock);
607 
608 	nfsd4_free_layouts(&reaplist);
609 }
610 
611 static void
612 nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
613 {
614 	struct nfs4_client *clp = ls->ls_stid.sc_client;
615 	char addr_str[INET6_ADDRSTRLEN];
616 	static char *envp[] = {
617 		"HOME=/",
618 		"TERM=linux",
619 		"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
620 		NULL
621 	};
622 	char *argv[8];
623 	int error;
624 
625 	rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
626 
627 	printk(KERN_WARNING
628 		"nfsd: client %s failed to respond to layout recall. "
629 		"  Fencing..\n", addr_str);
630 
631 	argv[0] = "/sbin/nfsd-recall-failed";
632 	argv[1] = addr_str;
633 	argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id;
634 	argv[3] = NULL;
635 
636 	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
637 	if (error) {
638 		printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
639 			addr_str, error);
640 	}
641 }
642 
643 static void
644 nfsd4_cb_layout_prepare(struct nfsd4_callback *cb)
645 {
646 	struct nfs4_layout_stateid *ls =
647 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
648 
649 	mutex_lock(&ls->ls_mutex);
650 	nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid);
651 	mutex_unlock(&ls->ls_mutex);
652 }
653 
654 static int
655 nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
656 {
657 	struct nfs4_layout_stateid *ls =
658 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
659 	struct nfsd_net *nn;
660 	ktime_t now, cutoff;
661 	const struct nfsd4_layout_ops *ops;
662 	LIST_HEAD(reaplist);
663 
664 
665 	switch (task->tk_status) {
666 	case 0:
667 	case -NFS4ERR_DELAY:
668 		/*
669 		 * Anything left? If not, then call it done. Note that we don't
670 		 * take the spinlock since this is an optimization and nothing
671 		 * should get added until the cb counter goes to zero.
672 		 */
673 		if (list_empty(&ls->ls_layouts))
674 			return 1;
675 
676 		/* Poll the client until it's done with the layout */
677 		now = ktime_get();
678 		nn = net_generic(ls->ls_stid.sc_client->net, nfsd_net_id);
679 
680 		/* Client gets 2 lease periods to return it */
681 		cutoff = ktime_add_ns(task->tk_start,
682 					 nn->nfsd4_lease * NSEC_PER_SEC * 2);
683 
684 		if (ktime_before(now, cutoff)) {
685 			rpc_delay(task, HZ/100); /* 10 mili-seconds */
686 			return 0;
687 		}
688 		/* Fallthrough */
689 	default:
690 		/*
691 		 * Unknown error or non-responding client, we'll need to fence.
692 		 */
693 		trace_layout_recall_fail(&ls->ls_stid.sc_stateid);
694 
695 		ops = nfsd4_layout_ops[ls->ls_layout_type];
696 		if (ops->fence_client)
697 			ops->fence_client(ls);
698 		else
699 			nfsd4_cb_layout_fail(ls);
700 		return -1;
701 	case -NFS4ERR_NOMATCHING_LAYOUT:
702 		trace_layout_recall_done(&ls->ls_stid.sc_stateid);
703 		task->tk_status = 0;
704 		return 1;
705 	}
706 }
707 
708 static void
709 nfsd4_cb_layout_release(struct nfsd4_callback *cb)
710 {
711 	struct nfs4_layout_stateid *ls =
712 		container_of(cb, struct nfs4_layout_stateid, ls_recall);
713 	LIST_HEAD(reaplist);
714 
715 	trace_layout_recall_release(&ls->ls_stid.sc_stateid);
716 
717 	nfsd4_return_all_layouts(ls, &reaplist);
718 	nfsd4_free_layouts(&reaplist);
719 	nfs4_put_stid(&ls->ls_stid);
720 }
721 
722 static const struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
723 	.prepare	= nfsd4_cb_layout_prepare,
724 	.done		= nfsd4_cb_layout_done,
725 	.release	= nfsd4_cb_layout_release,
726 };
727 
728 static bool
729 nfsd4_layout_lm_break(struct file_lock *fl)
730 {
731 	/*
732 	 * We don't want the locks code to timeout the lease for us;
733 	 * we'll remove it ourself if a layout isn't returned
734 	 * in time:
735 	 */
736 	fl->fl_break_time = 0;
737 	nfsd4_recall_file_layout(fl->fl_owner);
738 	return false;
739 }
740 
741 static int
742 nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
743 		struct list_head *dispose)
744 {
745 	BUG_ON(!(arg & F_UNLCK));
746 	return lease_modify(onlist, arg, dispose);
747 }
748 
749 static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
750 	.lm_break	= nfsd4_layout_lm_break,
751 	.lm_change	= nfsd4_layout_lm_change,
752 };
753 
754 int
755 nfsd4_init_pnfs(void)
756 {
757 	int i;
758 
759 	for (i = 0; i < DEVID_HASH_SIZE; i++)
760 		INIT_LIST_HEAD(&nfsd_devid_hash[i]);
761 
762 	nfs4_layout_cache = kmem_cache_create("nfs4_layout",
763 			sizeof(struct nfs4_layout), 0, 0, NULL);
764 	if (!nfs4_layout_cache)
765 		return -ENOMEM;
766 
767 	nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
768 			sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
769 	if (!nfs4_layout_stateid_cache) {
770 		kmem_cache_destroy(nfs4_layout_cache);
771 		return -ENOMEM;
772 	}
773 	return 0;
774 }
775 
776 void
777 nfsd4_exit_pnfs(void)
778 {
779 	int i;
780 
781 	kmem_cache_destroy(nfs4_layout_cache);
782 	kmem_cache_destroy(nfs4_layout_stateid_cache);
783 
784 	for (i = 0; i < DEVID_HASH_SIZE; i++) {
785 		struct nfsd4_deviceid_map *map, *n;
786 
787 		list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
788 			kfree(map);
789 	}
790 }
791