nfs4layouts.c

  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_fsid(fh), 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_fsid(fh),
 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	return 0;
124}
125
126void 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#ifdef CONFIG_NFSD_FLEXFILELAYOUT
136	exp->ex_layout_types |= 1 << LAYOUT_FLEX_FILES;
137#endif
138#ifdef CONFIG_NFSD_BLOCKLAYOUT
139	if (sb->s_export_op->get_uuid &&
140	    sb->s_export_op->map_blocks &&
141	    sb->s_export_op->commit_blocks)
142		exp->ex_layout_types |= 1 << LAYOUT_BLOCK_VOLUME;
143#endif
144#ifdef CONFIG_NFSD_SCSILAYOUT
145	if (sb->s_export_op->map_blocks &&
146	    sb->s_export_op->commit_blocks &&
147	    sb->s_bdev &&
148	    sb->s_bdev->bd_disk->fops->pr_ops &&
149	    sb->s_bdev->bd_disk->fops->get_unique_id)
150		exp->ex_layout_types |= 1 << LAYOUT_SCSI;
151#endif
152}
153
154void nfsd4_close_layout(struct nfs4_layout_stateid *ls)
155{
156	struct nfsd_file *fl;
157
158	spin_lock(&ls->ls_stid.sc_file->fi_lock);
159	fl = ls->ls_file;
160	ls->ls_file = NULL;
161	spin_unlock(&ls->ls_stid.sc_file->fi_lock);
162
163	if (fl) {
164		if (!nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
165			kernel_setlease(fl->nf_file, F_UNLCK, NULL,
166					(void **)&ls);
167		nfsd_file_put(fl);
168	}
169}
170
171static void
172nfsd4_free_layout_stateid(struct nfs4_stid *stid)
173{
174	struct nfs4_layout_stateid *ls = layoutstateid(stid);
175	struct nfs4_client *clp = ls->ls_stid.sc_client;
176	struct nfs4_file *fp = ls->ls_stid.sc_file;
177
178	trace_nfsd_layoutstate_free(&ls->ls_stid.sc_stateid);
179
180	spin_lock(&clp->cl_lock);
181	list_del_init(&ls->ls_perclnt);
182	spin_unlock(&clp->cl_lock);
183
184	spin_lock(&fp->fi_lock);
185	list_del_init(&ls->ls_perfile);
186	spin_unlock(&fp->fi_lock);
187
188	nfsd4_close_layout(ls);
189
190	if (ls->ls_recalled)
191		atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
192
193	kmem_cache_free(nfs4_layout_stateid_cache, ls);
194}
195
196static int
197nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
198{
199	struct file_lease *fl;
200	int status;
201
202	if (nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
203		return 0;
204
205	fl = locks_alloc_lease();
206	if (!fl)
207		return -ENOMEM;
208	locks_init_lease(fl);
209	fl->fl_lmops = &nfsd4_layouts_lm_ops;
210	fl->c.flc_flags = FL_LAYOUT;
211	fl->c.flc_type = F_RDLCK;
212	fl->c.flc_owner = ls;
213	fl->c.flc_pid = current->tgid;
214	fl->c.flc_file = ls->ls_file->nf_file;
215
216	status = kernel_setlease(fl->c.flc_file, fl->c.flc_type, &fl, NULL);
217	if (status) {
218		locks_free_lease(fl);
219		return status;
220	}
221	BUG_ON(fl != NULL);
222	return 0;
223}
224
225static struct nfs4_layout_stateid *
226nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
227		struct nfs4_stid *parent, u32 layout_type)
228{
229	struct nfs4_client *clp = cstate->clp;
230	struct nfs4_file *fp = parent->sc_file;
231	struct nfs4_layout_stateid *ls;
232	struct nfs4_stid *stp;
233
234	stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache,
235					nfsd4_free_layout_stateid);
236	if (!stp)
237		return NULL;
238
239	get_nfs4_file(fp);
240	stp->sc_file = fp;
241
242	ls = layoutstateid(stp);
243	INIT_LIST_HEAD(&ls->ls_perclnt);
244	INIT_LIST_HEAD(&ls->ls_perfile);
245	spin_lock_init(&ls->ls_lock);
246	INIT_LIST_HEAD(&ls->ls_layouts);
247	mutex_init(&ls->ls_mutex);
248	ls->ls_layout_type = layout_type;
249	nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
250			NFSPROC4_CLNT_CB_LAYOUT);
251
252	if (parent->sc_type == SC_TYPE_DELEG)
253		ls->ls_file = nfsd_file_get(fp->fi_deleg_file);
254	else
255		ls->ls_file = find_any_file(fp);
256	BUG_ON(!ls->ls_file);
257
258	if (nfsd4_layout_setlease(ls)) {
259		nfsd_file_put(ls->ls_file);
260		put_nfs4_file(fp);
261		kmem_cache_free(nfs4_layout_stateid_cache, ls);
262		return NULL;
263	}
264
265	spin_lock(&clp->cl_lock);
266	stp->sc_type = SC_TYPE_LAYOUT;
267	list_add(&ls->ls_perclnt, &clp->cl_lo_states);
268	spin_unlock(&clp->cl_lock);
269
270	spin_lock(&fp->fi_lock);
271	list_add(&ls->ls_perfile, &fp->fi_lo_states);
272	spin_unlock(&fp->fi_lock);
273
274	trace_nfsd_layoutstate_alloc(&ls->ls_stid.sc_stateid);
275	return ls;
276}
277
278__be32
279nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
280		struct nfsd4_compound_state *cstate, stateid_t *stateid,
281		bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
282{
283	struct nfs4_layout_stateid *ls;
284	struct nfs4_stid *stid;
285	unsigned short typemask = SC_TYPE_LAYOUT;
286	__be32 status;
287
288	if (create)
289		typemask |= (SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG);
290
291	status = nfsd4_lookup_stateid(cstate, stateid, typemask, 0, &stid,
292			net_generic(SVC_NET(rqstp), nfsd_net_id));
293	if (status)
294		goto out;
295
296	if (!fh_match(&cstate->current_fh.fh_handle,
297		      &stid->sc_file->fi_fhandle)) {
298		status = nfserr_bad_stateid;
299		goto out_put_stid;
300	}
301
302	if (stid->sc_type != SC_TYPE_LAYOUT) {
303		ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
304		nfs4_put_stid(stid);
305
306		status = nfserr_jukebox;
307		if (!ls)
308			goto out;
309		mutex_lock(&ls->ls_mutex);
310	} else {
311		ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
312
313		status = nfserr_bad_stateid;
314		mutex_lock(&ls->ls_mutex);
315		if (nfsd4_stateid_generation_after(stateid, &stid->sc_stateid))
316			goto out_unlock_stid;
317		if (layout_type != ls->ls_layout_type)
318			goto out_unlock_stid;
319	}
320
321	*lsp = ls;
322	return 0;
323
324out_unlock_stid:
325	mutex_unlock(&ls->ls_mutex);
326out_put_stid:
327	nfs4_put_stid(stid);
328out:
329	return status;
330}
331
332static void
333nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
334{
335	spin_lock(&ls->ls_lock);
336	if (ls->ls_recalled)
337		goto out_unlock;
338
339	if (list_empty(&ls->ls_layouts))
340		goto out_unlock;
341
342	ls->ls_recalled = true;
343	atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
344	trace_nfsd_layout_recall(&ls->ls_stid.sc_stateid);
345
346	if (!test_and_set_bit(NFSD4_CALLBACK_RUNNING, &ls->ls_recall.cb_flags)) {
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
767/**
768 *  nfsd4_layout_lm_open_conflict - see if the given file points to an inode that has
769 *				    an existing open that would conflict with the
770 *				    desired lease.
771 * @filp:	file to check
772 * @arg:	type of lease that we're trying to acquire
773 *
774 * The kernel will call into this operation to determine whether there
775 * are conflicting opens that may prevent the layout from being granted.
776 * For nfsd, that check is done at a higher level, so this trivially
777 * returns 0.
778 */
779static int
780nfsd4_layout_lm_open_conflict(struct file *filp, int arg)
781{
782	return 0;
783}
784
785static const struct lease_manager_operations nfsd4_layouts_lm_ops = {
786	.lm_break		= nfsd4_layout_lm_break,
787	.lm_change		= nfsd4_layout_lm_change,
788	.lm_open_conflict	= nfsd4_layout_lm_open_conflict,
789};
790
791int
792nfsd4_init_pnfs(void)
793{
794	int i;
795
796	for (i = 0; i < DEVID_HASH_SIZE; i++)
797		INIT_LIST_HEAD(&nfsd_devid_hash[i]);
798
799	nfs4_layout_cache = KMEM_CACHE(nfs4_layout, 0);
800	if (!nfs4_layout_cache)
801		return -ENOMEM;
802
803	nfs4_layout_stateid_cache = KMEM_CACHE(nfs4_layout_stateid, 0);
804	if (!nfs4_layout_stateid_cache) {
805		kmem_cache_destroy(nfs4_layout_cache);
806		return -ENOMEM;
807	}
808	return 0;
809}
810
811void
812nfsd4_exit_pnfs(void)
813{
814	int i;
815
816	kmem_cache_destroy(nfs4_layout_cache);
817	kmem_cache_destroy(nfs4_layout_stateid_cache);
818
819	for (i = 0; i < DEVID_HASH_SIZE; i++) {
820		struct nfsd4_deviceid_map *map, *n;
821
822		list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
823			kfree(map);
824	}
825}