Linux Audio

Check our new training course

Yocto / OpenEmbedded training

Feb 10-13, 2025
Register
Loading...
v6.2
  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 lock_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
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
155static void
156nfsd4_free_layout_stateid(struct nfs4_stid *stid)
157{
158	struct nfs4_layout_stateid *ls = layoutstateid(stid);
159	struct nfs4_client *clp = ls->ls_stid.sc_client;
160	struct nfs4_file *fp = ls->ls_stid.sc_file;
161
162	trace_nfsd_layoutstate_free(&ls->ls_stid.sc_stateid);
163
164	spin_lock(&clp->cl_lock);
165	list_del_init(&ls->ls_perclnt);
166	spin_unlock(&clp->cl_lock);
167
168	spin_lock(&fp->fi_lock);
169	list_del_init(&ls->ls_perfile);
170	spin_unlock(&fp->fi_lock);
171
172	if (!nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
173		vfs_setlease(ls->ls_file->nf_file, F_UNLCK, NULL, (void **)&ls);
174	nfsd_file_put(ls->ls_file);
175
176	if (ls->ls_recalled)
177		atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
178
179	kmem_cache_free(nfs4_layout_stateid_cache, ls);
180}
181
182static int
183nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
184{
185	struct file_lock *fl;
186	int status;
187
188	if (nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls)
189		return 0;
190
191	fl = locks_alloc_lock();
192	if (!fl)
193		return -ENOMEM;
194	locks_init_lock(fl);
195	fl->fl_lmops = &nfsd4_layouts_lm_ops;
196	fl->fl_flags = FL_LAYOUT;
197	fl->fl_type = F_RDLCK;
198	fl->fl_end = OFFSET_MAX;
199	fl->fl_owner = ls;
200	fl->fl_pid = current->tgid;
201	fl->fl_file = ls->ls_file->nf_file;
202
203	status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL);
204	if (status) {
205		locks_free_lock(fl);
206		return status;
207	}
208	BUG_ON(fl != NULL);
209	return 0;
210}
211
212static struct nfs4_layout_stateid *
213nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
214		struct nfs4_stid *parent, u32 layout_type)
215{
216	struct nfs4_client *clp = cstate->clp;
217	struct nfs4_file *fp = parent->sc_file;
218	struct nfs4_layout_stateid *ls;
219	struct nfs4_stid *stp;
220
221	stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache,
222					nfsd4_free_layout_stateid);
223	if (!stp)
224		return NULL;
225
226	get_nfs4_file(fp);
227	stp->sc_file = fp;
228
229	ls = layoutstateid(stp);
230	INIT_LIST_HEAD(&ls->ls_perclnt);
231	INIT_LIST_HEAD(&ls->ls_perfile);
232	spin_lock_init(&ls->ls_lock);
233	INIT_LIST_HEAD(&ls->ls_layouts);
234	mutex_init(&ls->ls_mutex);
235	ls->ls_layout_type = layout_type;
236	nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
237			NFSPROC4_CLNT_CB_LAYOUT);
238
239	if (parent->sc_type == NFS4_DELEG_STID)
240		ls->ls_file = nfsd_file_get(fp->fi_deleg_file);
241	else
242		ls->ls_file = find_any_file(fp);
243	BUG_ON(!ls->ls_file);
244
245	if (nfsd4_layout_setlease(ls)) {
246		nfsd_file_put(ls->ls_file);
247		put_nfs4_file(fp);
248		kmem_cache_free(nfs4_layout_stateid_cache, ls);
249		return NULL;
250	}
251
252	spin_lock(&clp->cl_lock);
253	stp->sc_type = NFS4_LAYOUT_STID;
254	list_add(&ls->ls_perclnt, &clp->cl_lo_states);
255	spin_unlock(&clp->cl_lock);
256
257	spin_lock(&fp->fi_lock);
258	list_add(&ls->ls_perfile, &fp->fi_lo_states);
259	spin_unlock(&fp->fi_lock);
260
261	trace_nfsd_layoutstate_alloc(&ls->ls_stid.sc_stateid);
262	return ls;
263}
264
265__be32
266nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
267		struct nfsd4_compound_state *cstate, stateid_t *stateid,
268		bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
269{
270	struct nfs4_layout_stateid *ls;
271	struct nfs4_stid *stid;
272	unsigned char typemask = NFS4_LAYOUT_STID;
273	__be32 status;
274
275	if (create)
276		typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
277
278	status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
279			net_generic(SVC_NET(rqstp), nfsd_net_id));
280	if (status)
281		goto out;
282
283	if (!fh_match(&cstate->current_fh.fh_handle,
284		      &stid->sc_file->fi_fhandle)) {
285		status = nfserr_bad_stateid;
286		goto out_put_stid;
287	}
288
289	if (stid->sc_type != NFS4_LAYOUT_STID) {
290		ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
291		nfs4_put_stid(stid);
292
293		status = nfserr_jukebox;
294		if (!ls)
295			goto out;
296		mutex_lock(&ls->ls_mutex);
297	} else {
298		ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
299
300		status = nfserr_bad_stateid;
301		mutex_lock(&ls->ls_mutex);
302		if (nfsd4_stateid_generation_after(stateid, &stid->sc_stateid))
303			goto out_unlock_stid;
304		if (layout_type != ls->ls_layout_type)
305			goto out_unlock_stid;
306	}
307
308	*lsp = ls;
309	return 0;
310
311out_unlock_stid:
312	mutex_unlock(&ls->ls_mutex);
313out_put_stid:
314	nfs4_put_stid(stid);
315out:
316	return status;
317}
318
319static void
320nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
321{
322	spin_lock(&ls->ls_lock);
323	if (ls->ls_recalled)
324		goto out_unlock;
325
326	ls->ls_recalled = true;
327	atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
328	if (list_empty(&ls->ls_layouts))
329		goto out_unlock;
330
 
 
331	trace_nfsd_layout_recall(&ls->ls_stid.sc_stateid);
332
333	refcount_inc(&ls->ls_stid.sc_count);
334	nfsd4_run_cb(&ls->ls_recall);
335
336out_unlock:
337	spin_unlock(&ls->ls_lock);
338}
339
340static inline u64
341layout_end(struct nfsd4_layout_seg *seg)
342{
343	u64 end = seg->offset + seg->length;
344	return end >= seg->offset ? end : NFS4_MAX_UINT64;
345}
346
347static void
348layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
349{
350	if (end == NFS4_MAX_UINT64)
351		lo->length = NFS4_MAX_UINT64;
352	else
353		lo->length = end - lo->offset;
354}
355
356static bool
357layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
358{
359	if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
360		return false;
361	if (layout_end(&lo->lo_seg) <= s->offset)
362		return false;
363	if (layout_end(s) <= lo->lo_seg.offset)
364		return false;
365	return true;
366}
367
368static bool
369layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
370{
371	if (lo->iomode != new->iomode)
372		return false;
373	if (layout_end(new) < lo->offset)
374		return false;
375	if (layout_end(lo) < new->offset)
376		return false;
377
378	lo->offset = min(lo->offset, new->offset);
379	layout_update_len(lo, max(layout_end(lo), layout_end(new)));
380	return true;
381}
382
383static __be32
384nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
385{
386	struct nfs4_file *fp = ls->ls_stid.sc_file;
387	struct nfs4_layout_stateid *l, *n;
388	__be32 nfserr = nfs_ok;
389
390	assert_spin_locked(&fp->fi_lock);
391
392	list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
393		if (l != ls) {
394			nfsd4_recall_file_layout(l);
395			nfserr = nfserr_recallconflict;
396		}
397	}
398
399	return nfserr;
400}
401
402__be32
403nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
404{
405	struct nfsd4_layout_seg *seg = &lgp->lg_seg;
406	struct nfs4_file *fp = ls->ls_stid.sc_file;
407	struct nfs4_layout *lp, *new = NULL;
408	__be32 nfserr;
409
410	spin_lock(&fp->fi_lock);
411	nfserr = nfsd4_recall_conflict(ls);
412	if (nfserr)
413		goto out;
414	spin_lock(&ls->ls_lock);
415	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
416		if (layouts_try_merge(&lp->lo_seg, seg))
417			goto done;
418	}
419	spin_unlock(&ls->ls_lock);
420	spin_unlock(&fp->fi_lock);
421
422	new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
423	if (!new)
424		return nfserr_jukebox;
425	memcpy(&new->lo_seg, seg, sizeof(new->lo_seg));
426	new->lo_state = ls;
427
428	spin_lock(&fp->fi_lock);
429	nfserr = nfsd4_recall_conflict(ls);
430	if (nfserr)
431		goto out;
432	spin_lock(&ls->ls_lock);
433	list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
434		if (layouts_try_merge(&lp->lo_seg, seg))
435			goto done;
436	}
437
438	refcount_inc(&ls->ls_stid.sc_count);
439	list_add_tail(&new->lo_perstate, &ls->ls_layouts);
440	new = NULL;
441done:
442	nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid);
443	spin_unlock(&ls->ls_lock);
444out:
445	spin_unlock(&fp->fi_lock);
446	if (new)
447		kmem_cache_free(nfs4_layout_cache, new);
448	return nfserr;
449}
450
451static void
452nfsd4_free_layouts(struct list_head *reaplist)
453{
454	while (!list_empty(reaplist)) {
455		struct nfs4_layout *lp = list_first_entry(reaplist,
456				struct nfs4_layout, lo_perstate);
457
458		list_del(&lp->lo_perstate);
459		nfs4_put_stid(&lp->lo_state->ls_stid);
460		kmem_cache_free(nfs4_layout_cache, lp);
461	}
462}
463
464static void
465nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
466		struct list_head *reaplist)
467{
468	struct nfsd4_layout_seg *lo = &lp->lo_seg;
469	u64 end = layout_end(lo);
470
471	if (seg->offset <= lo->offset) {
472		if (layout_end(seg) >= end) {
473			list_move_tail(&lp->lo_perstate, reaplist);
474			return;
475		}
476		lo->offset = layout_end(seg);
477	} else {
478		/* retain the whole layout segment on a split. */
479		if (layout_end(seg) < end) {
480			dprintk("%s: split not supported\n", __func__);
481			return;
482		}
483		end = seg->offset;
484	}
485
486	layout_update_len(lo, end);
487}
488
489__be32
490nfsd4_return_file_layouts(struct svc_rqst *rqstp,
491		struct nfsd4_compound_state *cstate,
492		struct nfsd4_layoutreturn *lrp)
493{
494	struct nfs4_layout_stateid *ls;
495	struct nfs4_layout *lp, *n;
496	LIST_HEAD(reaplist);
497	__be32 nfserr;
498	int found = 0;
499
500	nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
501						false, lrp->lr_layout_type,
502						&ls);
503	if (nfserr) {
504		trace_nfsd_layout_return_lookup_fail(&lrp->lr_sid);
505		return nfserr;
506	}
507
508	spin_lock(&ls->ls_lock);
509	list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
510		if (layouts_overlapping(lp, &lrp->lr_seg)) {
511			nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
512			found++;
513		}
514	}
515	if (!list_empty(&ls->ls_layouts)) {
516		if (found)
517			nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid);
518		lrp->lrs_present = 1;
519	} else {
520		trace_nfsd_layoutstate_unhash(&ls->ls_stid.sc_stateid);
521		nfs4_unhash_stid(&ls->ls_stid);
522		lrp->lrs_present = 0;
523	}
524	spin_unlock(&ls->ls_lock);
525
526	mutex_unlock(&ls->ls_mutex);
527	nfs4_put_stid(&ls->ls_stid);
528	nfsd4_free_layouts(&reaplist);
529	return nfs_ok;
530}
531
532__be32
533nfsd4_return_client_layouts(struct svc_rqst *rqstp,
534		struct nfsd4_compound_state *cstate,
535		struct nfsd4_layoutreturn *lrp)
536{
537	struct nfs4_layout_stateid *ls, *n;
538	struct nfs4_client *clp = cstate->clp;
539	struct nfs4_layout *lp, *t;
540	LIST_HEAD(reaplist);
541
542	lrp->lrs_present = 0;
543
544	spin_lock(&clp->cl_lock);
545	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
546		if (ls->ls_layout_type != lrp->lr_layout_type)
547			continue;
548
549		if (lrp->lr_return_type == RETURN_FSID &&
550		    !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
551				   &cstate->current_fh.fh_handle))
552			continue;
553
554		spin_lock(&ls->ls_lock);
555		list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
556			if (lrp->lr_seg.iomode == IOMODE_ANY ||
557			    lrp->lr_seg.iomode == lp->lo_seg.iomode)
558				list_move_tail(&lp->lo_perstate, &reaplist);
559		}
560		spin_unlock(&ls->ls_lock);
561	}
562	spin_unlock(&clp->cl_lock);
563
564	nfsd4_free_layouts(&reaplist);
565	return 0;
566}
567
568static void
569nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
570		struct list_head *reaplist)
571{
572	spin_lock(&ls->ls_lock);
573	list_splice_init(&ls->ls_layouts, reaplist);
574	spin_unlock(&ls->ls_lock);
575}
576
577void
578nfsd4_return_all_client_layouts(struct nfs4_client *clp)
579{
580	struct nfs4_layout_stateid *ls, *n;
581	LIST_HEAD(reaplist);
582
583	spin_lock(&clp->cl_lock);
584	list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
585		nfsd4_return_all_layouts(ls, &reaplist);
586	spin_unlock(&clp->cl_lock);
587
588	nfsd4_free_layouts(&reaplist);
589}
590
591void
592nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
593{
594	struct nfs4_layout_stateid *ls, *n;
595	LIST_HEAD(reaplist);
596
597	spin_lock(&fp->fi_lock);
598	list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
599		if (ls->ls_stid.sc_client == clp)
600			nfsd4_return_all_layouts(ls, &reaplist);
601	}
602	spin_unlock(&fp->fi_lock);
603
604	nfsd4_free_layouts(&reaplist);
605}
606
607static void
608nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
609{
610	struct nfs4_client *clp = ls->ls_stid.sc_client;
611	char addr_str[INET6_ADDRSTRLEN];
612	static char const nfsd_recall_failed[] = "/sbin/nfsd-recall-failed";
613	static char *envp[] = {
614		"HOME=/",
615		"TERM=linux",
616		"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
617		NULL
618	};
619	char *argv[8];
620	int error;
621
622	rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
623
624	printk(KERN_WARNING
625		"nfsd: client %s failed to respond to layout recall. "
626		"  Fencing..\n", addr_str);
627
628	argv[0] = (char *)nfsd_recall_failed;
629	argv[1] = addr_str;
630	argv[2] = ls->ls_file->nf_file->f_path.mnt->mnt_sb->s_id;
631	argv[3] = NULL;
632
633	error = call_usermodehelper(nfsd_recall_failed, argv, envp,
634				    UMH_WAIT_PROC);
635	if (error) {
636		printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
637			addr_str, error);
638	}
639}
640
641static void
642nfsd4_cb_layout_prepare(struct nfsd4_callback *cb)
643{
644	struct nfs4_layout_stateid *ls =
645		container_of(cb, struct nfs4_layout_stateid, ls_recall);
646
647	mutex_lock(&ls->ls_mutex);
648	nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid);
649	mutex_unlock(&ls->ls_mutex);
650}
651
652static int
653nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
654{
655	struct nfs4_layout_stateid *ls =
656		container_of(cb, struct nfs4_layout_stateid, ls_recall);
657	struct nfsd_net *nn;
658	ktime_t now, cutoff;
659	const struct nfsd4_layout_ops *ops;
 
660
661	trace_nfsd_cb_layout_done(&ls->ls_stid.sc_stateid, task);
662	switch (task->tk_status) {
663	case 0:
664	case -NFS4ERR_DELAY:
665		/*
666		 * Anything left? If not, then call it done. Note that we don't
667		 * take the spinlock since this is an optimization and nothing
668		 * should get added until the cb counter goes to zero.
669		 */
670		if (list_empty(&ls->ls_layouts))
671			return 1;
672
673		/* Poll the client until it's done with the layout */
674		now = ktime_get();
675		nn = net_generic(ls->ls_stid.sc_client->net, nfsd_net_id);
676
677		/* Client gets 2 lease periods to return it */
678		cutoff = ktime_add_ns(task->tk_start,
679					 (u64)nn->nfsd4_lease * NSEC_PER_SEC * 2);
680
681		if (ktime_before(now, cutoff)) {
682			rpc_delay(task, HZ/100); /* 10 mili-seconds */
683			return 0;
684		}
685		fallthrough;
686	default:
687		/*
688		 * Unknown error or non-responding client, we'll need to fence.
689		 */
690		trace_nfsd_layout_recall_fail(&ls->ls_stid.sc_stateid);
691
692		ops = nfsd4_layout_ops[ls->ls_layout_type];
693		if (ops->fence_client)
694			ops->fence_client(ls);
695		else
696			nfsd4_cb_layout_fail(ls);
 
 
 
 
 
697		return 1;
698	case -NFS4ERR_NOMATCHING_LAYOUT:
699		trace_nfsd_layout_recall_done(&ls->ls_stid.sc_stateid);
700		task->tk_status = 0;
701		return 1;
702	}
703}
704
705static void
706nfsd4_cb_layout_release(struct nfsd4_callback *cb)
707{
708	struct nfs4_layout_stateid *ls =
709		container_of(cb, struct nfs4_layout_stateid, ls_recall);
710	LIST_HEAD(reaplist);
711
712	trace_nfsd_layout_recall_release(&ls->ls_stid.sc_stateid);
713
714	nfsd4_return_all_layouts(ls, &reaplist);
715	nfsd4_free_layouts(&reaplist);
716	nfs4_put_stid(&ls->ls_stid);
717}
718
719static const struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
720	.prepare	= nfsd4_cb_layout_prepare,
721	.done		= nfsd4_cb_layout_done,
722	.release	= nfsd4_cb_layout_release,
 
723};
724
725static bool
726nfsd4_layout_lm_break(struct file_lock *fl)
727{
728	/*
729	 * We don't want the locks code to timeout the lease for us;
730	 * we'll remove it ourself if a layout isn't returned
731	 * in time:
732	 */
733	fl->fl_break_time = 0;
734	nfsd4_recall_file_layout(fl->fl_owner);
735	return false;
736}
737
738static int
739nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
740		struct list_head *dispose)
741{
742	BUG_ON(!(arg & F_UNLCK));
743	return lease_modify(onlist, arg, dispose);
744}
745
746static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
747	.lm_break	= nfsd4_layout_lm_break,
748	.lm_change	= nfsd4_layout_lm_change,
749};
750
751int
752nfsd4_init_pnfs(void)
753{
754	int i;
755
756	for (i = 0; i < DEVID_HASH_SIZE; i++)
757		INIT_LIST_HEAD(&nfsd_devid_hash[i]);
758
759	nfs4_layout_cache = kmem_cache_create("nfs4_layout",
760			sizeof(struct nfs4_layout), 0, 0, NULL);
761	if (!nfs4_layout_cache)
762		return -ENOMEM;
763
764	nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
765			sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
766	if (!nfs4_layout_stateid_cache) {
767		kmem_cache_destroy(nfs4_layout_cache);
768		return -ENOMEM;
769	}
770	return 0;
771}
772
773void
774nfsd4_exit_pnfs(void)
775{
776	int i;
777
778	kmem_cache_destroy(nfs4_layout_cache);
779	kmem_cache_destroy(nfs4_layout_stateid_cache);
780
781	for (i = 0; i < DEVID_HASH_SIZE; i++) {
782		struct nfsd4_deviceid_map *map, *n;
783
784		list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
785			kfree(map);
786	}
787}
v6.13.7
  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}