1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Module for pnfs flexfile layout driver.
4  *
5  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
6  *
7  * Tao Peng <bergwolf@primarydata.com>
8  */
9 
10 #include <linux/nfs_fs.h>
11 #include <linux/nfs_mount.h>
12 #include <linux/nfs_page.h>
13 #include <linux/module.h>
14 #include <linux/sched/mm.h>
15 
16 #include <linux/sunrpc/metrics.h>
17 
18 #include "flexfilelayout.h"
19 #include "../nfs4session.h"
20 #include "../nfs4idmap.h"
21 #include "../internal.h"
22 #include "../delegation.h"
23 #include "../nfs4trace.h"
24 #include "../iostat.h"
25 #include "../nfs.h"
26 #include "../nfs42.h"
27 
28 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
29 
30 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
31 #define FF_LAYOUTRETURN_MAXERR 20
32 
33 static unsigned short io_maxretrans;
34 
35 static const struct pnfs_commit_ops ff_layout_commit_ops;
36 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
37 		struct nfs_pgio_header *hdr);
38 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
39 			       struct nfs42_layoutstat_devinfo *devinfo,
40 			       int dev_limit);
41 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
42 			      const struct nfs42_layoutstat_devinfo *devinfo,
43 			      struct nfs4_ff_layout_mirror *mirror);
44 
45 static struct pnfs_layout_hdr *
46 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
47 {
48 	struct nfs4_flexfile_layout *ffl;
49 
50 	ffl = kzalloc(sizeof(*ffl), gfp_flags);
51 	if (ffl) {
52 		pnfs_init_ds_commit_info(&ffl->commit_info);
53 		INIT_LIST_HEAD(&ffl->error_list);
54 		INIT_LIST_HEAD(&ffl->mirrors);
55 		ffl->last_report_time = ktime_get();
56 		ffl->commit_info.ops = &ff_layout_commit_ops;
57 		return &ffl->generic_hdr;
58 	} else
59 		return NULL;
60 }
61 
62 static void
63 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
64 {
65 	struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
66 	struct nfs4_ff_layout_ds_err *err, *n;
67 
68 	list_for_each_entry_safe(err, n, &ffl->error_list, list) {
69 		list_del(&err->list);
70 		kfree(err);
71 	}
72 	kfree_rcu(ffl, generic_hdr.plh_rcu);
73 }
74 
75 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
76 {
77 	__be32 *p;
78 
79 	p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
80 	if (unlikely(p == NULL))
81 		return -ENOBUFS;
82 	stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
83 	memcpy(stateid->data, p, NFS4_STATEID_SIZE);
84 	dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
85 		p[0], p[1], p[2], p[3]);
86 	return 0;
87 }
88 
89 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
90 {
91 	__be32 *p;
92 
93 	p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
94 	if (unlikely(!p))
95 		return -ENOBUFS;
96 	memcpy(devid, p, NFS4_DEVICEID4_SIZE);
97 	nfs4_print_deviceid(devid);
98 	return 0;
99 }
100 
101 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
102 {
103 	__be32 *p;
104 
105 	p = xdr_inline_decode(xdr, 4);
106 	if (unlikely(!p))
107 		return -ENOBUFS;
108 	fh->size = be32_to_cpup(p++);
109 	if (fh->size > sizeof(struct nfs_fh)) {
110 		printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
111 		       fh->size);
112 		return -EOVERFLOW;
113 	}
114 	/* fh.data */
115 	p = xdr_inline_decode(xdr, fh->size);
116 	if (unlikely(!p))
117 		return -ENOBUFS;
118 	memcpy(&fh->data, p, fh->size);
119 	dprintk("%s: fh len %d\n", __func__, fh->size);
120 
121 	return 0;
122 }
123 
124 /*
125  * Currently only stringified uids and gids are accepted.
126  * I.e., kerberos is not supported to the DSes, so no pricipals.
127  *
128  * That means that one common function will suffice, but when
129  * principals are added, this should be split to accomodate
130  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
131  */
132 static int
133 decode_name(struct xdr_stream *xdr, u32 *id)
134 {
135 	__be32 *p;
136 	int len;
137 
138 	/* opaque_length(4)*/
139 	p = xdr_inline_decode(xdr, 4);
140 	if (unlikely(!p))
141 		return -ENOBUFS;
142 	len = be32_to_cpup(p++);
143 	if (len < 0)
144 		return -EINVAL;
145 
146 	dprintk("%s: len %u\n", __func__, len);
147 
148 	/* opaque body */
149 	p = xdr_inline_decode(xdr, len);
150 	if (unlikely(!p))
151 		return -ENOBUFS;
152 
153 	if (!nfs_map_string_to_numeric((char *)p, len, id))
154 		return -EINVAL;
155 
156 	return 0;
157 }
158 
159 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
160 		const struct nfs4_ff_layout_mirror *m2)
161 {
162 	int i, j;
163 
164 	if (m1->fh_versions_cnt != m2->fh_versions_cnt)
165 		return false;
166 	for (i = 0; i < m1->fh_versions_cnt; i++) {
167 		bool found_fh = false;
168 		for (j = 0; j < m2->fh_versions_cnt; j++) {
169 			if (nfs_compare_fh(&m1->fh_versions[i],
170 					&m2->fh_versions[j]) == 0) {
171 				found_fh = true;
172 				break;
173 			}
174 		}
175 		if (!found_fh)
176 			return false;
177 	}
178 	return true;
179 }
180 
181 static struct nfs4_ff_layout_mirror *
182 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
183 		struct nfs4_ff_layout_mirror *mirror)
184 {
185 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
186 	struct nfs4_ff_layout_mirror *pos;
187 	struct inode *inode = lo->plh_inode;
188 
189 	spin_lock(&inode->i_lock);
190 	list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
191 		if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
192 			continue;
193 		if (!ff_mirror_match_fh(mirror, pos))
194 			continue;
195 		if (refcount_inc_not_zero(&pos->ref)) {
196 			spin_unlock(&inode->i_lock);
197 			return pos;
198 		}
199 	}
200 	list_add(&mirror->mirrors, &ff_layout->mirrors);
201 	mirror->layout = lo;
202 	spin_unlock(&inode->i_lock);
203 	return mirror;
204 }
205 
206 static void
207 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
208 {
209 	struct inode *inode;
210 	if (mirror->layout == NULL)
211 		return;
212 	inode = mirror->layout->plh_inode;
213 	spin_lock(&inode->i_lock);
214 	list_del(&mirror->mirrors);
215 	spin_unlock(&inode->i_lock);
216 	mirror->layout = NULL;
217 }
218 
219 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
220 {
221 	struct nfs4_ff_layout_mirror *mirror;
222 
223 	mirror = kzalloc(sizeof(*mirror), gfp_flags);
224 	if (mirror != NULL) {
225 		spin_lock_init(&mirror->lock);
226 		refcount_set(&mirror->ref, 1);
227 		INIT_LIST_HEAD(&mirror->mirrors);
228 	}
229 	return mirror;
230 }
231 
232 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
233 {
234 	const struct cred	*cred;
235 
236 	ff_layout_remove_mirror(mirror);
237 	kfree(mirror->fh_versions);
238 	cred = rcu_access_pointer(mirror->ro_cred);
239 	put_cred(cred);
240 	cred = rcu_access_pointer(mirror->rw_cred);
241 	put_cred(cred);
242 	nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
243 	kfree(mirror);
244 }
245 
246 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
247 {
248 	if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
249 		ff_layout_free_mirror(mirror);
250 }
251 
252 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
253 {
254 	u32 i;
255 
256 	for (i = 0; i < fls->mirror_array_cnt; i++)
257 		ff_layout_put_mirror(fls->mirror_array[i]);
258 }
259 
260 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
261 {
262 	if (fls) {
263 		ff_layout_free_mirror_array(fls);
264 		kfree(fls);
265 	}
266 }
267 
268 static bool
269 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
270 		struct pnfs_layout_segment *l2)
271 {
272 	const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
273 	const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
274 	u32 i;
275 
276 	if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
277 		return false;
278 	for (i = 0; i < fl1->mirror_array_cnt; i++) {
279 		if (fl1->mirror_array[i] != fl2->mirror_array[i])
280 			return false;
281 	}
282 	return true;
283 }
284 
285 static bool
286 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
287 		const struct pnfs_layout_range *l2)
288 {
289 	u64 end1, end2;
290 
291 	if (l1->iomode != l2->iomode)
292 		return l1->iomode != IOMODE_READ;
293 	end1 = pnfs_calc_offset_end(l1->offset, l1->length);
294 	end2 = pnfs_calc_offset_end(l2->offset, l2->length);
295 	if (end1 < l2->offset)
296 		return false;
297 	if (end2 < l1->offset)
298 		return true;
299 	return l2->offset <= l1->offset;
300 }
301 
302 static bool
303 ff_lseg_merge(struct pnfs_layout_segment *new,
304 		struct pnfs_layout_segment *old)
305 {
306 	u64 new_end, old_end;
307 
308 	if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
309 		return false;
310 	if (new->pls_range.iomode != old->pls_range.iomode)
311 		return false;
312 	old_end = pnfs_calc_offset_end(old->pls_range.offset,
313 			old->pls_range.length);
314 	if (old_end < new->pls_range.offset)
315 		return false;
316 	new_end = pnfs_calc_offset_end(new->pls_range.offset,
317 			new->pls_range.length);
318 	if (new_end < old->pls_range.offset)
319 		return false;
320 	if (!ff_lseg_match_mirrors(new, old))
321 		return false;
322 
323 	/* Mergeable: copy info from 'old' to 'new' */
324 	if (new_end < old_end)
325 		new_end = old_end;
326 	if (new->pls_range.offset < old->pls_range.offset)
327 		new->pls_range.offset = old->pls_range.offset;
328 	new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
329 			new_end);
330 	if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
331 		set_bit(NFS_LSEG_ROC, &new->pls_flags);
332 	return true;
333 }
334 
335 static void
336 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
337 		struct pnfs_layout_segment *lseg,
338 		struct list_head *free_me)
339 {
340 	pnfs_generic_layout_insert_lseg(lo, lseg,
341 			ff_lseg_range_is_after,
342 			ff_lseg_merge,
343 			free_me);
344 }
345 
346 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
347 {
348 	int i, j;
349 
350 	for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
351 		for (j = i + 1; j < fls->mirror_array_cnt; j++)
352 			if (fls->mirror_array[i]->efficiency <
353 			    fls->mirror_array[j]->efficiency)
354 				swap(fls->mirror_array[i],
355 				     fls->mirror_array[j]);
356 	}
357 }
358 
359 static struct pnfs_layout_segment *
360 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
361 		     struct nfs4_layoutget_res *lgr,
362 		     gfp_t gfp_flags)
363 {
364 	struct pnfs_layout_segment *ret;
365 	struct nfs4_ff_layout_segment *fls = NULL;
366 	struct xdr_stream stream;
367 	struct xdr_buf buf;
368 	struct page *scratch;
369 	u64 stripe_unit;
370 	u32 mirror_array_cnt;
371 	__be32 *p;
372 	int i, rc;
373 
374 	dprintk("--> %s\n", __func__);
375 	scratch = alloc_page(gfp_flags);
376 	if (!scratch)
377 		return ERR_PTR(-ENOMEM);
378 
379 	xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
380 			      lgr->layoutp->len);
381 	xdr_set_scratch_page(&stream, scratch);
382 
383 	/* stripe unit and mirror_array_cnt */
384 	rc = -EIO;
385 	p = xdr_inline_decode(&stream, 8 + 4);
386 	if (!p)
387 		goto out_err_free;
388 
389 	p = xdr_decode_hyper(p, &stripe_unit);
390 	mirror_array_cnt = be32_to_cpup(p++);
391 	dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
392 		stripe_unit, mirror_array_cnt);
393 
394 	if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
395 	    mirror_array_cnt == 0)
396 		goto out_err_free;
397 
398 	rc = -ENOMEM;
399 	fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
400 			gfp_flags);
401 	if (!fls)
402 		goto out_err_free;
403 
404 	fls->mirror_array_cnt = mirror_array_cnt;
405 	fls->stripe_unit = stripe_unit;
406 
407 	for (i = 0; i < fls->mirror_array_cnt; i++) {
408 		struct nfs4_ff_layout_mirror *mirror;
409 		struct cred *kcred;
410 		const struct cred __rcu *cred;
411 		kuid_t uid;
412 		kgid_t gid;
413 		u32 ds_count, fh_count, id;
414 		int j;
415 
416 		rc = -EIO;
417 		p = xdr_inline_decode(&stream, 4);
418 		if (!p)
419 			goto out_err_free;
420 		ds_count = be32_to_cpup(p);
421 
422 		/* FIXME: allow for striping? */
423 		if (ds_count != 1)
424 			goto out_err_free;
425 
426 		fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
427 		if (fls->mirror_array[i] == NULL) {
428 			rc = -ENOMEM;
429 			goto out_err_free;
430 		}
431 
432 		fls->mirror_array[i]->ds_count = ds_count;
433 
434 		/* deviceid */
435 		rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
436 		if (rc)
437 			goto out_err_free;
438 
439 		/* efficiency */
440 		rc = -EIO;
441 		p = xdr_inline_decode(&stream, 4);
442 		if (!p)
443 			goto out_err_free;
444 		fls->mirror_array[i]->efficiency = be32_to_cpup(p);
445 
446 		/* stateid */
447 		rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
448 		if (rc)
449 			goto out_err_free;
450 
451 		/* fh */
452 		rc = -EIO;
453 		p = xdr_inline_decode(&stream, 4);
454 		if (!p)
455 			goto out_err_free;
456 		fh_count = be32_to_cpup(p);
457 
458 		fls->mirror_array[i]->fh_versions =
459 			kcalloc(fh_count, sizeof(struct nfs_fh),
460 				gfp_flags);
461 		if (fls->mirror_array[i]->fh_versions == NULL) {
462 			rc = -ENOMEM;
463 			goto out_err_free;
464 		}
465 
466 		for (j = 0; j < fh_count; j++) {
467 			rc = decode_nfs_fh(&stream,
468 					   &fls->mirror_array[i]->fh_versions[j]);
469 			if (rc)
470 				goto out_err_free;
471 		}
472 
473 		fls->mirror_array[i]->fh_versions_cnt = fh_count;
474 
475 		/* user */
476 		rc = decode_name(&stream, &id);
477 		if (rc)
478 			goto out_err_free;
479 
480 		uid = make_kuid(&init_user_ns, id);
481 
482 		/* group */
483 		rc = decode_name(&stream, &id);
484 		if (rc)
485 			goto out_err_free;
486 
487 		gid = make_kgid(&init_user_ns, id);
488 
489 		if (gfp_flags & __GFP_FS)
490 			kcred = prepare_kernel_cred(NULL);
491 		else {
492 			unsigned int nofs_flags = memalloc_nofs_save();
493 			kcred = prepare_kernel_cred(NULL);
494 			memalloc_nofs_restore(nofs_flags);
495 		}
496 		rc = -ENOMEM;
497 		if (!kcred)
498 			goto out_err_free;
499 		kcred->fsuid = uid;
500 		kcred->fsgid = gid;
501 		cred = RCU_INITIALIZER(kcred);
502 
503 		if (lgr->range.iomode == IOMODE_READ)
504 			rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
505 		else
506 			rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
507 
508 		mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
509 		if (mirror != fls->mirror_array[i]) {
510 			/* swap cred ptrs so free_mirror will clean up old */
511 			if (lgr->range.iomode == IOMODE_READ) {
512 				cred = xchg(&mirror->ro_cred, cred);
513 				rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
514 			} else {
515 				cred = xchg(&mirror->rw_cred, cred);
516 				rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
517 			}
518 			ff_layout_free_mirror(fls->mirror_array[i]);
519 			fls->mirror_array[i] = mirror;
520 		}
521 
522 		dprintk("%s: iomode %s uid %u gid %u\n", __func__,
523 			lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
524 			from_kuid(&init_user_ns, uid),
525 			from_kgid(&init_user_ns, gid));
526 	}
527 
528 	p = xdr_inline_decode(&stream, 4);
529 	if (!p)
530 		goto out_sort_mirrors;
531 	fls->flags = be32_to_cpup(p);
532 
533 	p = xdr_inline_decode(&stream, 4);
534 	if (!p)
535 		goto out_sort_mirrors;
536 	for (i=0; i < fls->mirror_array_cnt; i++)
537 		fls->mirror_array[i]->report_interval = be32_to_cpup(p);
538 
539 out_sort_mirrors:
540 	ff_layout_sort_mirrors(fls);
541 	ret = &fls->generic_hdr;
542 	dprintk("<-- %s (success)\n", __func__);
543 out_free_page:
544 	__free_page(scratch);
545 	return ret;
546 out_err_free:
547 	_ff_layout_free_lseg(fls);
548 	ret = ERR_PTR(rc);
549 	dprintk("<-- %s (%d)\n", __func__, rc);
550 	goto out_free_page;
551 }
552 
553 static void
554 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
555 {
556 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
557 
558 	dprintk("--> %s\n", __func__);
559 
560 	if (lseg->pls_range.iomode == IOMODE_RW) {
561 		struct nfs4_flexfile_layout *ffl;
562 		struct inode *inode;
563 
564 		ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
565 		inode = ffl->generic_hdr.plh_inode;
566 		spin_lock(&inode->i_lock);
567 		pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
568 		spin_unlock(&inode->i_lock);
569 	}
570 	_ff_layout_free_lseg(fls);
571 }
572 
573 static void
574 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
575 {
576 	/* first IO request? */
577 	if (atomic_inc_return(&timer->n_ops) == 1) {
578 		timer->start_time = now;
579 	}
580 }
581 
582 static ktime_t
583 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
584 {
585 	ktime_t start;
586 
587 	if (atomic_dec_return(&timer->n_ops) < 0)
588 		WARN_ON_ONCE(1);
589 
590 	start = timer->start_time;
591 	timer->start_time = now;
592 	return ktime_sub(now, start);
593 }
594 
595 static bool
596 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
597 			    struct nfs4_ff_layoutstat *layoutstat,
598 			    ktime_t now)
599 {
600 	s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
601 	struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
602 
603 	nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
604 	if (!mirror->start_time)
605 		mirror->start_time = now;
606 	if (mirror->report_interval != 0)
607 		report_interval = (s64)mirror->report_interval * 1000LL;
608 	else if (layoutstats_timer != 0)
609 		report_interval = (s64)layoutstats_timer * 1000LL;
610 	if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
611 			report_interval) {
612 		ffl->last_report_time = now;
613 		return true;
614 	}
615 
616 	return false;
617 }
618 
619 static void
620 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
621 		__u64 requested)
622 {
623 	struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
624 
625 	iostat->ops_requested++;
626 	iostat->bytes_requested += requested;
627 }
628 
629 static void
630 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
631 		__u64 requested,
632 		__u64 completed,
633 		ktime_t time_completed,
634 		ktime_t time_started)
635 {
636 	struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
637 	ktime_t completion_time = ktime_sub(time_completed, time_started);
638 	ktime_t timer;
639 
640 	iostat->ops_completed++;
641 	iostat->bytes_completed += completed;
642 	iostat->bytes_not_delivered += requested - completed;
643 
644 	timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
645 	iostat->total_busy_time =
646 			ktime_add(iostat->total_busy_time, timer);
647 	iostat->aggregate_completion_time =
648 			ktime_add(iostat->aggregate_completion_time,
649 					completion_time);
650 }
651 
652 static void
653 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
654 		struct nfs4_ff_layout_mirror *mirror,
655 		__u64 requested, ktime_t now)
656 {
657 	bool report;
658 
659 	spin_lock(&mirror->lock);
660 	report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
661 	nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
662 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
663 	spin_unlock(&mirror->lock);
664 
665 	if (report)
666 		pnfs_report_layoutstat(inode, GFP_KERNEL);
667 }
668 
669 static void
670 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
671 		struct nfs4_ff_layout_mirror *mirror,
672 		__u64 requested,
673 		__u64 completed)
674 {
675 	spin_lock(&mirror->lock);
676 	nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
677 			requested, completed,
678 			ktime_get(), task->tk_start);
679 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
680 	spin_unlock(&mirror->lock);
681 }
682 
683 static void
684 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
685 		struct nfs4_ff_layout_mirror *mirror,
686 		__u64 requested, ktime_t now)
687 {
688 	bool report;
689 
690 	spin_lock(&mirror->lock);
691 	report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
692 	nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
693 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
694 	spin_unlock(&mirror->lock);
695 
696 	if (report)
697 		pnfs_report_layoutstat(inode, GFP_NOIO);
698 }
699 
700 static void
701 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
702 		struct nfs4_ff_layout_mirror *mirror,
703 		__u64 requested,
704 		__u64 completed,
705 		enum nfs3_stable_how committed)
706 {
707 	if (committed == NFS_UNSTABLE)
708 		requested = completed = 0;
709 
710 	spin_lock(&mirror->lock);
711 	nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
712 			requested, completed, ktime_get(), task->tk_start);
713 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
714 	spin_unlock(&mirror->lock);
715 }
716 
717 static void
718 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)
719 {
720 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
721 
722 	if (devid)
723 		nfs4_mark_deviceid_unavailable(devid);
724 }
725 
726 static void
727 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)
728 {
729 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
730 
731 	if (devid)
732 		nfs4_mark_deviceid_available(devid);
733 }
734 
735 static struct nfs4_pnfs_ds *
736 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
737 			     u32 start_idx, u32 *best_idx,
738 			     bool check_device)
739 {
740 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
741 	struct nfs4_ff_layout_mirror *mirror;
742 	struct nfs4_pnfs_ds *ds;
743 	u32 idx;
744 
745 	/* mirrors are initially sorted by efficiency */
746 	for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
747 		mirror = FF_LAYOUT_COMP(lseg, idx);
748 		ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
749 		if (!ds)
750 			continue;
751 
752 		if (check_device &&
753 		    nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
754 			continue;
755 
756 		*best_idx = idx;
757 		return ds;
758 	}
759 
760 	return NULL;
761 }
762 
763 static struct nfs4_pnfs_ds *
764 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
765 				 u32 start_idx, u32 *best_idx)
766 {
767 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
768 }
769 
770 static struct nfs4_pnfs_ds *
771 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
772 				   u32 start_idx, u32 *best_idx)
773 {
774 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
775 }
776 
777 static struct nfs4_pnfs_ds *
778 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
779 				  u32 start_idx, u32 *best_idx)
780 {
781 	struct nfs4_pnfs_ds *ds;
782 
783 	ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
784 	if (ds)
785 		return ds;
786 	return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
787 }
788 
789 static struct nfs4_pnfs_ds *
790 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio,
791 			  u32 *best_idx)
792 {
793 	struct pnfs_layout_segment *lseg = pgio->pg_lseg;
794 	struct nfs4_pnfs_ds *ds;
795 
796 	ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
797 					       best_idx);
798 	if (ds || !pgio->pg_mirror_idx)
799 		return ds;
800 	return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
801 }
802 
803 static void
804 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
805 		      struct nfs_page *req,
806 		      bool strict_iomode)
807 {
808 	pnfs_put_lseg(pgio->pg_lseg);
809 	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
810 					   nfs_req_openctx(req),
811 					   req_offset(req),
812 					   req->wb_bytes,
813 					   IOMODE_READ,
814 					   strict_iomode,
815 					   GFP_KERNEL);
816 	if (IS_ERR(pgio->pg_lseg)) {
817 		pgio->pg_error = PTR_ERR(pgio->pg_lseg);
818 		pgio->pg_lseg = NULL;
819 	}
820 }
821 
822 static void
823 ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
824 			  struct nfs_page *req)
825 {
826 	pnfs_generic_pg_check_layout(pgio);
827 	pnfs_generic_pg_check_range(pgio, req);
828 }
829 
830 static void
831 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
832 			struct nfs_page *req)
833 {
834 	struct nfs_pgio_mirror *pgm;
835 	struct nfs4_ff_layout_mirror *mirror;
836 	struct nfs4_pnfs_ds *ds;
837 	u32 ds_idx;
838 
839 retry:
840 	ff_layout_pg_check_layout(pgio, req);
841 	/* Use full layout for now */
842 	if (!pgio->pg_lseg) {
843 		ff_layout_pg_get_read(pgio, req, false);
844 		if (!pgio->pg_lseg)
845 			goto out_nolseg;
846 	}
847 	if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
848 		ff_layout_pg_get_read(pgio, req, true);
849 		if (!pgio->pg_lseg)
850 			goto out_nolseg;
851 	}
852 
853 	ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
854 	if (!ds) {
855 		if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
856 			goto out_mds;
857 		pnfs_generic_pg_cleanup(pgio);
858 		/* Sleep for 1 second before retrying */
859 		ssleep(1);
860 		goto retry;
861 	}
862 
863 	mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
864 	pgm = &pgio->pg_mirrors[0];
865 	pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
866 
867 	pgio->pg_mirror_idx = ds_idx;
868 
869 	if (NFS_SERVER(pgio->pg_inode)->flags &
870 			(NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
871 		pgio->pg_maxretrans = io_maxretrans;
872 	return;
873 out_nolseg:
874 	if (pgio->pg_error < 0)
875 		return;
876 out_mds:
877 	trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
878 			0, NFS4_MAX_UINT64, IOMODE_READ,
879 			NFS_I(pgio->pg_inode)->layout,
880 			pgio->pg_lseg);
881 	pgio->pg_maxretrans = 0;
882 	nfs_pageio_reset_read_mds(pgio);
883 }
884 
885 static void
886 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
887 			struct nfs_page *req)
888 {
889 	struct nfs4_ff_layout_mirror *mirror;
890 	struct nfs_pgio_mirror *pgm;
891 	struct nfs4_pnfs_ds *ds;
892 	u32 i;
893 
894 retry:
895 	ff_layout_pg_check_layout(pgio, req);
896 	if (!pgio->pg_lseg) {
897 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
898 						   nfs_req_openctx(req),
899 						   req_offset(req),
900 						   req->wb_bytes,
901 						   IOMODE_RW,
902 						   false,
903 						   GFP_NOFS);
904 		if (IS_ERR(pgio->pg_lseg)) {
905 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
906 			pgio->pg_lseg = NULL;
907 			return;
908 		}
909 	}
910 	/* If no lseg, fall back to write through mds */
911 	if (pgio->pg_lseg == NULL)
912 		goto out_mds;
913 
914 	/* Use a direct mapping of ds_idx to pgio mirror_idx */
915 	if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
916 		goto out_eagain;
917 
918 	for (i = 0; i < pgio->pg_mirror_count; i++) {
919 		mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
920 		ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
921 		if (!ds) {
922 			if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
923 				goto out_mds;
924 			pnfs_generic_pg_cleanup(pgio);
925 			/* Sleep for 1 second before retrying */
926 			ssleep(1);
927 			goto retry;
928 		}
929 		pgm = &pgio->pg_mirrors[i];
930 		pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
931 	}
932 
933 	if (NFS_SERVER(pgio->pg_inode)->flags &
934 			(NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
935 		pgio->pg_maxretrans = io_maxretrans;
936 	return;
937 out_eagain:
938 	pnfs_generic_pg_cleanup(pgio);
939 	pgio->pg_error = -EAGAIN;
940 	return;
941 out_mds:
942 	trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
943 			0, NFS4_MAX_UINT64, IOMODE_RW,
944 			NFS_I(pgio->pg_inode)->layout,
945 			pgio->pg_lseg);
946 	pgio->pg_maxretrans = 0;
947 	nfs_pageio_reset_write_mds(pgio);
948 	pgio->pg_error = -EAGAIN;
949 }
950 
951 static unsigned int
952 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
953 				    struct nfs_page *req)
954 {
955 	if (!pgio->pg_lseg) {
956 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
957 						   nfs_req_openctx(req),
958 						   req_offset(req),
959 						   req->wb_bytes,
960 						   IOMODE_RW,
961 						   false,
962 						   GFP_NOFS);
963 		if (IS_ERR(pgio->pg_lseg)) {
964 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
965 			pgio->pg_lseg = NULL;
966 			goto out;
967 		}
968 	}
969 	if (pgio->pg_lseg)
970 		return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
971 
972 	trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
973 			0, NFS4_MAX_UINT64, IOMODE_RW,
974 			NFS_I(pgio->pg_inode)->layout,
975 			pgio->pg_lseg);
976 	/* no lseg means that pnfs is not in use, so no mirroring here */
977 	nfs_pageio_reset_write_mds(pgio);
978 out:
979 	return 1;
980 }
981 
982 static u32
983 ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
984 {
985 	u32 old = desc->pg_mirror_idx;
986 
987 	desc->pg_mirror_idx = idx;
988 	return old;
989 }
990 
991 static struct nfs_pgio_mirror *
992 ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
993 {
994 	return &desc->pg_mirrors[idx];
995 }
996 
997 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
998 	.pg_init = ff_layout_pg_init_read,
999 	.pg_test = pnfs_generic_pg_test,
1000 	.pg_doio = pnfs_generic_pg_readpages,
1001 	.pg_cleanup = pnfs_generic_pg_cleanup,
1002 };
1003 
1004 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1005 	.pg_init = ff_layout_pg_init_write,
1006 	.pg_test = pnfs_generic_pg_test,
1007 	.pg_doio = pnfs_generic_pg_writepages,
1008 	.pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1009 	.pg_cleanup = pnfs_generic_pg_cleanup,
1010 	.pg_get_mirror = ff_layout_pg_get_mirror_write,
1011 	.pg_set_mirror = ff_layout_pg_set_mirror_write,
1012 };
1013 
1014 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1015 {
1016 	struct rpc_task *task = &hdr->task;
1017 
1018 	pnfs_layoutcommit_inode(hdr->inode, false);
1019 
1020 	if (retry_pnfs) {
1021 		dprintk("%s Reset task %5u for i/o through pNFS "
1022 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1023 			hdr->task.tk_pid,
1024 			hdr->inode->i_sb->s_id,
1025 			(unsigned long long)NFS_FILEID(hdr->inode),
1026 			hdr->args.count,
1027 			(unsigned long long)hdr->args.offset);
1028 
1029 		hdr->completion_ops->reschedule_io(hdr);
1030 		return;
1031 	}
1032 
1033 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1034 		dprintk("%s Reset task %5u for i/o through MDS "
1035 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1036 			hdr->task.tk_pid,
1037 			hdr->inode->i_sb->s_id,
1038 			(unsigned long long)NFS_FILEID(hdr->inode),
1039 			hdr->args.count,
1040 			(unsigned long long)hdr->args.offset);
1041 
1042 		trace_pnfs_mds_fallback_write_done(hdr->inode,
1043 				hdr->args.offset, hdr->args.count,
1044 				IOMODE_RW, NFS_I(hdr->inode)->layout,
1045 				hdr->lseg);
1046 		task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1047 	}
1048 }
1049 
1050 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1051 {
1052 	u32 idx = hdr->pgio_mirror_idx + 1;
1053 	u32 new_idx = 0;
1054 
1055 	if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx))
1056 		ff_layout_send_layouterror(hdr->lseg);
1057 	else
1058 		pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1059 	pnfs_read_resend_pnfs(hdr, new_idx);
1060 }
1061 
1062 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1063 {
1064 	struct rpc_task *task = &hdr->task;
1065 
1066 	pnfs_layoutcommit_inode(hdr->inode, false);
1067 	pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1068 
1069 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1070 		dprintk("%s Reset task %5u for i/o through MDS "
1071 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1072 			hdr->task.tk_pid,
1073 			hdr->inode->i_sb->s_id,
1074 			(unsigned long long)NFS_FILEID(hdr->inode),
1075 			hdr->args.count,
1076 			(unsigned long long)hdr->args.offset);
1077 
1078 		trace_pnfs_mds_fallback_read_done(hdr->inode,
1079 				hdr->args.offset, hdr->args.count,
1080 				IOMODE_READ, NFS_I(hdr->inode)->layout,
1081 				hdr->lseg);
1082 		task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1083 	}
1084 }
1085 
1086 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1087 					   struct nfs4_state *state,
1088 					   struct nfs_client *clp,
1089 					   struct pnfs_layout_segment *lseg,
1090 					   u32 idx)
1091 {
1092 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
1093 	struct inode *inode = lo->plh_inode;
1094 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1095 	struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1096 
1097 	switch (task->tk_status) {
1098 	case -NFS4ERR_BADSESSION:
1099 	case -NFS4ERR_BADSLOT:
1100 	case -NFS4ERR_BAD_HIGH_SLOT:
1101 	case -NFS4ERR_DEADSESSION:
1102 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1103 	case -NFS4ERR_SEQ_FALSE_RETRY:
1104 	case -NFS4ERR_SEQ_MISORDERED:
1105 		dprintk("%s ERROR %d, Reset session. Exchangeid "
1106 			"flags 0x%x\n", __func__, task->tk_status,
1107 			clp->cl_exchange_flags);
1108 		nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1109 		break;
1110 	case -NFS4ERR_DELAY:
1111 	case -NFS4ERR_GRACE:
1112 		rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1113 		break;
1114 	case -NFS4ERR_RETRY_UNCACHED_REP:
1115 		break;
1116 	/* Invalidate Layout errors */
1117 	case -NFS4ERR_PNFS_NO_LAYOUT:
1118 	case -ESTALE:           /* mapped NFS4ERR_STALE */
1119 	case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1120 	case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1121 	case -NFS4ERR_FHEXPIRED:
1122 	case -NFS4ERR_WRONG_TYPE:
1123 		dprintk("%s Invalid layout error %d\n", __func__,
1124 			task->tk_status);
1125 		/*
1126 		 * Destroy layout so new i/o will get a new layout.
1127 		 * Layout will not be destroyed until all current lseg
1128 		 * references are put. Mark layout as invalid to resend failed
1129 		 * i/o and all i/o waiting on the slot table to the MDS until
1130 		 * layout is destroyed and a new valid layout is obtained.
1131 		 */
1132 		pnfs_destroy_layout(NFS_I(inode));
1133 		rpc_wake_up(&tbl->slot_tbl_waitq);
1134 		goto reset;
1135 	/* RPC connection errors */
1136 	case -ECONNREFUSED:
1137 	case -EHOSTDOWN:
1138 	case -EHOSTUNREACH:
1139 	case -ENETUNREACH:
1140 	case -EIO:
1141 	case -ETIMEDOUT:
1142 	case -EPIPE:
1143 		dprintk("%s DS connection error %d\n", __func__,
1144 			task->tk_status);
1145 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1146 				&devid->deviceid);
1147 		rpc_wake_up(&tbl->slot_tbl_waitq);
1148 		fallthrough;
1149 	default:
1150 		if (ff_layout_avoid_mds_available_ds(lseg))
1151 			return -NFS4ERR_RESET_TO_PNFS;
1152 reset:
1153 		dprintk("%s Retry through MDS. Error %d\n", __func__,
1154 			task->tk_status);
1155 		return -NFS4ERR_RESET_TO_MDS;
1156 	}
1157 	task->tk_status = 0;
1158 	return -EAGAIN;
1159 }
1160 
1161 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1162 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1163 					   struct pnfs_layout_segment *lseg,
1164 					   u32 idx)
1165 {
1166 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1167 
1168 	switch (task->tk_status) {
1169 	/* File access problems. Don't mark the device as unavailable */
1170 	case -EACCES:
1171 	case -ESTALE:
1172 	case -EISDIR:
1173 	case -EBADHANDLE:
1174 	case -ELOOP:
1175 	case -ENOSPC:
1176 		break;
1177 	case -EJUKEBOX:
1178 		nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1179 		goto out_retry;
1180 	default:
1181 		dprintk("%s DS connection error %d\n", __func__,
1182 			task->tk_status);
1183 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1184 				&devid->deviceid);
1185 	}
1186 	/* FIXME: Need to prevent infinite looping here. */
1187 	return -NFS4ERR_RESET_TO_PNFS;
1188 out_retry:
1189 	task->tk_status = 0;
1190 	rpc_restart_call_prepare(task);
1191 	rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1192 	return -EAGAIN;
1193 }
1194 
1195 static int ff_layout_async_handle_error(struct rpc_task *task,
1196 					struct nfs4_state *state,
1197 					struct nfs_client *clp,
1198 					struct pnfs_layout_segment *lseg,
1199 					u32 idx)
1200 {
1201 	int vers = clp->cl_nfs_mod->rpc_vers->number;
1202 
1203 	if (task->tk_status >= 0) {
1204 		ff_layout_mark_ds_reachable(lseg, idx);
1205 		return 0;
1206 	}
1207 
1208 	/* Handle the case of an invalid layout segment */
1209 	if (!pnfs_is_valid_lseg(lseg))
1210 		return -NFS4ERR_RESET_TO_PNFS;
1211 
1212 	switch (vers) {
1213 	case 3:
1214 		return ff_layout_async_handle_error_v3(task, lseg, idx);
1215 	case 4:
1216 		return ff_layout_async_handle_error_v4(task, state, clp,
1217 						       lseg, idx);
1218 	default:
1219 		/* should never happen */
1220 		WARN_ON_ONCE(1);
1221 		return 0;
1222 	}
1223 }
1224 
1225 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1226 					u32 idx, u64 offset, u64 length,
1227 					u32 *op_status, int opnum, int error)
1228 {
1229 	struct nfs4_ff_layout_mirror *mirror;
1230 	u32 status = *op_status;
1231 	int err;
1232 
1233 	if (status == 0) {
1234 		switch (error) {
1235 		case -ETIMEDOUT:
1236 		case -EPFNOSUPPORT:
1237 		case -EPROTONOSUPPORT:
1238 		case -EOPNOTSUPP:
1239 		case -ECONNREFUSED:
1240 		case -ECONNRESET:
1241 		case -EHOSTDOWN:
1242 		case -EHOSTUNREACH:
1243 		case -ENETUNREACH:
1244 		case -EADDRINUSE:
1245 		case -ENOBUFS:
1246 		case -EPIPE:
1247 		case -EPERM:
1248 			*op_status = status = NFS4ERR_NXIO;
1249 			break;
1250 		case -EACCES:
1251 			*op_status = status = NFS4ERR_ACCESS;
1252 			break;
1253 		default:
1254 			return;
1255 		}
1256 	}
1257 
1258 	mirror = FF_LAYOUT_COMP(lseg, idx);
1259 	err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1260 				       mirror, offset, length, status, opnum,
1261 				       GFP_NOIO);
1262 
1263 	switch (status) {
1264 	case NFS4ERR_DELAY:
1265 	case NFS4ERR_GRACE:
1266 		break;
1267 	case NFS4ERR_NXIO:
1268 		ff_layout_mark_ds_unreachable(lseg, idx);
1269 		/*
1270 		 * Don't return the layout if this is a read and we still
1271 		 * have layouts to try
1272 		 */
1273 		if (opnum == OP_READ)
1274 			break;
1275 		fallthrough;
1276 	default:
1277 		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1278 						  lseg);
1279 	}
1280 
1281 	dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1282 }
1283 
1284 /* NFS_PROTO call done callback routines */
1285 static int ff_layout_read_done_cb(struct rpc_task *task,
1286 				struct nfs_pgio_header *hdr)
1287 {
1288 	int err;
1289 
1290 	if (task->tk_status < 0) {
1291 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1292 					    hdr->args.offset, hdr->args.count,
1293 					    &hdr->res.op_status, OP_READ,
1294 					    task->tk_status);
1295 		trace_ff_layout_read_error(hdr);
1296 	}
1297 
1298 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1299 					   hdr->ds_clp, hdr->lseg,
1300 					   hdr->pgio_mirror_idx);
1301 
1302 	trace_nfs4_pnfs_read(hdr, err);
1303 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1304 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1305 	switch (err) {
1306 	case -NFS4ERR_RESET_TO_PNFS:
1307 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1308 		return task->tk_status;
1309 	case -NFS4ERR_RESET_TO_MDS:
1310 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1311 		return task->tk_status;
1312 	case -EAGAIN:
1313 		goto out_eagain;
1314 	}
1315 
1316 	return 0;
1317 out_eagain:
1318 	rpc_restart_call_prepare(task);
1319 	return -EAGAIN;
1320 }
1321 
1322 static bool
1323 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1324 {
1325 	return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1326 }
1327 
1328 /*
1329  * We reference the rpc_cred of the first WRITE that triggers the need for
1330  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1331  * rfc5661 is not clear about which credential should be used.
1332  *
1333  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1334  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1335  * we always send layoutcommit after DS writes.
1336  */
1337 static void
1338 ff_layout_set_layoutcommit(struct inode *inode,
1339 		struct pnfs_layout_segment *lseg,
1340 		loff_t end_offset)
1341 {
1342 	if (!ff_layout_need_layoutcommit(lseg))
1343 		return;
1344 
1345 	pnfs_set_layoutcommit(inode, lseg, end_offset);
1346 	dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1347 		(unsigned long long) NFS_I(inode)->layout->plh_lwb);
1348 }
1349 
1350 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1351 		struct nfs_pgio_header *hdr)
1352 {
1353 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1354 		return;
1355 	nfs4_ff_layout_stat_io_start_read(hdr->inode,
1356 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1357 			hdr->args.count,
1358 			task->tk_start);
1359 }
1360 
1361 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1362 		struct nfs_pgio_header *hdr)
1363 {
1364 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1365 		return;
1366 	nfs4_ff_layout_stat_io_end_read(task,
1367 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1368 			hdr->args.count,
1369 			hdr->res.count);
1370 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1371 }
1372 
1373 static int ff_layout_read_prepare_common(struct rpc_task *task,
1374 					 struct nfs_pgio_header *hdr)
1375 {
1376 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1377 		rpc_exit(task, -EIO);
1378 		return -EIO;
1379 	}
1380 
1381 	ff_layout_read_record_layoutstats_start(task, hdr);
1382 	return 0;
1383 }
1384 
1385 /*
1386  * Call ops for the async read/write cases
1387  * In the case of dense layouts, the offset needs to be reset to its
1388  * original value.
1389  */
1390 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1391 {
1392 	struct nfs_pgio_header *hdr = data;
1393 
1394 	if (ff_layout_read_prepare_common(task, hdr))
1395 		return;
1396 
1397 	rpc_call_start(task);
1398 }
1399 
1400 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1401 {
1402 	struct nfs_pgio_header *hdr = data;
1403 
1404 	if (nfs4_setup_sequence(hdr->ds_clp,
1405 				&hdr->args.seq_args,
1406 				&hdr->res.seq_res,
1407 				task))
1408 		return;
1409 
1410 	ff_layout_read_prepare_common(task, hdr);
1411 }
1412 
1413 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1414 {
1415 	struct nfs_pgio_header *hdr = data;
1416 
1417 	dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1418 
1419 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1420 	    task->tk_status == 0) {
1421 		nfs4_sequence_done(task, &hdr->res.seq_res);
1422 		return;
1423 	}
1424 
1425 	/* Note this may cause RPC to be resent */
1426 	hdr->mds_ops->rpc_call_done(task, hdr);
1427 }
1428 
1429 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1430 {
1431 	struct nfs_pgio_header *hdr = data;
1432 
1433 	ff_layout_read_record_layoutstats_done(task, hdr);
1434 	rpc_count_iostats_metrics(task,
1435 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1436 }
1437 
1438 static void ff_layout_read_release(void *data)
1439 {
1440 	struct nfs_pgio_header *hdr = data;
1441 
1442 	ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1443 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1444 		ff_layout_resend_pnfs_read(hdr);
1445 	else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1446 		ff_layout_reset_read(hdr);
1447 	pnfs_generic_rw_release(data);
1448 }
1449 
1450 
1451 static int ff_layout_write_done_cb(struct rpc_task *task,
1452 				struct nfs_pgio_header *hdr)
1453 {
1454 	loff_t end_offs = 0;
1455 	int err;
1456 
1457 	if (task->tk_status < 0) {
1458 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1459 					    hdr->args.offset, hdr->args.count,
1460 					    &hdr->res.op_status, OP_WRITE,
1461 					    task->tk_status);
1462 		trace_ff_layout_write_error(hdr);
1463 	}
1464 
1465 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1466 					   hdr->ds_clp, hdr->lseg,
1467 					   hdr->pgio_mirror_idx);
1468 
1469 	trace_nfs4_pnfs_write(hdr, err);
1470 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1471 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1472 	switch (err) {
1473 	case -NFS4ERR_RESET_TO_PNFS:
1474 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1475 		return task->tk_status;
1476 	case -NFS4ERR_RESET_TO_MDS:
1477 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1478 		return task->tk_status;
1479 	case -EAGAIN:
1480 		return -EAGAIN;
1481 	}
1482 
1483 	if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1484 	    hdr->res.verf->committed == NFS_DATA_SYNC)
1485 		end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1486 
1487 	/* Note: if the write is unstable, don't set end_offs until commit */
1488 	ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1489 
1490 	/* zero out fattr since we don't care DS attr at all */
1491 	hdr->fattr.valid = 0;
1492 	if (task->tk_status >= 0)
1493 		nfs_writeback_update_inode(hdr);
1494 
1495 	return 0;
1496 }
1497 
1498 static int ff_layout_commit_done_cb(struct rpc_task *task,
1499 				     struct nfs_commit_data *data)
1500 {
1501 	int err;
1502 
1503 	if (task->tk_status < 0) {
1504 		ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1505 					    data->args.offset, data->args.count,
1506 					    &data->res.op_status, OP_COMMIT,
1507 					    task->tk_status);
1508 		trace_ff_layout_commit_error(data);
1509 	}
1510 
1511 	err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1512 					   data->lseg, data->ds_commit_index);
1513 
1514 	trace_nfs4_pnfs_commit_ds(data, err);
1515 	switch (err) {
1516 	case -NFS4ERR_RESET_TO_PNFS:
1517 		pnfs_generic_prepare_to_resend_writes(data);
1518 		return -EAGAIN;
1519 	case -NFS4ERR_RESET_TO_MDS:
1520 		pnfs_generic_prepare_to_resend_writes(data);
1521 		return -EAGAIN;
1522 	case -EAGAIN:
1523 		rpc_restart_call_prepare(task);
1524 		return -EAGAIN;
1525 	}
1526 
1527 	ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1528 
1529 	return 0;
1530 }
1531 
1532 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1533 		struct nfs_pgio_header *hdr)
1534 {
1535 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1536 		return;
1537 	nfs4_ff_layout_stat_io_start_write(hdr->inode,
1538 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1539 			hdr->args.count,
1540 			task->tk_start);
1541 }
1542 
1543 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1544 		struct nfs_pgio_header *hdr)
1545 {
1546 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1547 		return;
1548 	nfs4_ff_layout_stat_io_end_write(task,
1549 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1550 			hdr->args.count, hdr->res.count,
1551 			hdr->res.verf->committed);
1552 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1553 }
1554 
1555 static int ff_layout_write_prepare_common(struct rpc_task *task,
1556 					  struct nfs_pgio_header *hdr)
1557 {
1558 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1559 		rpc_exit(task, -EIO);
1560 		return -EIO;
1561 	}
1562 
1563 	ff_layout_write_record_layoutstats_start(task, hdr);
1564 	return 0;
1565 }
1566 
1567 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1568 {
1569 	struct nfs_pgio_header *hdr = data;
1570 
1571 	if (ff_layout_write_prepare_common(task, hdr))
1572 		return;
1573 
1574 	rpc_call_start(task);
1575 }
1576 
1577 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1578 {
1579 	struct nfs_pgio_header *hdr = data;
1580 
1581 	if (nfs4_setup_sequence(hdr->ds_clp,
1582 				&hdr->args.seq_args,
1583 				&hdr->res.seq_res,
1584 				task))
1585 		return;
1586 
1587 	ff_layout_write_prepare_common(task, hdr);
1588 }
1589 
1590 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1591 {
1592 	struct nfs_pgio_header *hdr = data;
1593 
1594 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1595 	    task->tk_status == 0) {
1596 		nfs4_sequence_done(task, &hdr->res.seq_res);
1597 		return;
1598 	}
1599 
1600 	/* Note this may cause RPC to be resent */
1601 	hdr->mds_ops->rpc_call_done(task, hdr);
1602 }
1603 
1604 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1605 {
1606 	struct nfs_pgio_header *hdr = data;
1607 
1608 	ff_layout_write_record_layoutstats_done(task, hdr);
1609 	rpc_count_iostats_metrics(task,
1610 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1611 }
1612 
1613 static void ff_layout_write_release(void *data)
1614 {
1615 	struct nfs_pgio_header *hdr = data;
1616 
1617 	ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1618 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1619 		ff_layout_send_layouterror(hdr->lseg);
1620 		ff_layout_reset_write(hdr, true);
1621 	} else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1622 		ff_layout_reset_write(hdr, false);
1623 	pnfs_generic_rw_release(data);
1624 }
1625 
1626 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1627 		struct nfs_commit_data *cdata)
1628 {
1629 	if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1630 		return;
1631 	nfs4_ff_layout_stat_io_start_write(cdata->inode,
1632 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1633 			0, task->tk_start);
1634 }
1635 
1636 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1637 		struct nfs_commit_data *cdata)
1638 {
1639 	struct nfs_page *req;
1640 	__u64 count = 0;
1641 
1642 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1643 		return;
1644 
1645 	if (task->tk_status == 0) {
1646 		list_for_each_entry(req, &cdata->pages, wb_list)
1647 			count += req->wb_bytes;
1648 	}
1649 	nfs4_ff_layout_stat_io_end_write(task,
1650 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1651 			count, count, NFS_FILE_SYNC);
1652 	set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1653 }
1654 
1655 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1656 		struct nfs_commit_data *cdata)
1657 {
1658 	ff_layout_commit_record_layoutstats_start(task, cdata);
1659 }
1660 
1661 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1662 {
1663 	ff_layout_commit_prepare_common(task, data);
1664 	rpc_call_start(task);
1665 }
1666 
1667 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1668 {
1669 	struct nfs_commit_data *wdata = data;
1670 
1671 	if (nfs4_setup_sequence(wdata->ds_clp,
1672 				&wdata->args.seq_args,
1673 				&wdata->res.seq_res,
1674 				task))
1675 		return;
1676 	ff_layout_commit_prepare_common(task, data);
1677 }
1678 
1679 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1680 {
1681 	pnfs_generic_write_commit_done(task, data);
1682 }
1683 
1684 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1685 {
1686 	struct nfs_commit_data *cdata = data;
1687 
1688 	ff_layout_commit_record_layoutstats_done(task, cdata);
1689 	rpc_count_iostats_metrics(task,
1690 	    &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1691 }
1692 
1693 static void ff_layout_commit_release(void *data)
1694 {
1695 	struct nfs_commit_data *cdata = data;
1696 
1697 	ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1698 	pnfs_generic_commit_release(data);
1699 }
1700 
1701 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1702 	.rpc_call_prepare = ff_layout_read_prepare_v3,
1703 	.rpc_call_done = ff_layout_read_call_done,
1704 	.rpc_count_stats = ff_layout_read_count_stats,
1705 	.rpc_release = ff_layout_read_release,
1706 };
1707 
1708 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1709 	.rpc_call_prepare = ff_layout_read_prepare_v4,
1710 	.rpc_call_done = ff_layout_read_call_done,
1711 	.rpc_count_stats = ff_layout_read_count_stats,
1712 	.rpc_release = ff_layout_read_release,
1713 };
1714 
1715 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1716 	.rpc_call_prepare = ff_layout_write_prepare_v3,
1717 	.rpc_call_done = ff_layout_write_call_done,
1718 	.rpc_count_stats = ff_layout_write_count_stats,
1719 	.rpc_release = ff_layout_write_release,
1720 };
1721 
1722 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1723 	.rpc_call_prepare = ff_layout_write_prepare_v4,
1724 	.rpc_call_done = ff_layout_write_call_done,
1725 	.rpc_count_stats = ff_layout_write_count_stats,
1726 	.rpc_release = ff_layout_write_release,
1727 };
1728 
1729 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1730 	.rpc_call_prepare = ff_layout_commit_prepare_v3,
1731 	.rpc_call_done = ff_layout_commit_done,
1732 	.rpc_count_stats = ff_layout_commit_count_stats,
1733 	.rpc_release = ff_layout_commit_release,
1734 };
1735 
1736 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1737 	.rpc_call_prepare = ff_layout_commit_prepare_v4,
1738 	.rpc_call_done = ff_layout_commit_done,
1739 	.rpc_count_stats = ff_layout_commit_count_stats,
1740 	.rpc_release = ff_layout_commit_release,
1741 };
1742 
1743 static enum pnfs_try_status
1744 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1745 {
1746 	struct pnfs_layout_segment *lseg = hdr->lseg;
1747 	struct nfs4_pnfs_ds *ds;
1748 	struct rpc_clnt *ds_clnt;
1749 	struct nfs4_ff_layout_mirror *mirror;
1750 	const struct cred *ds_cred;
1751 	loff_t offset = hdr->args.offset;
1752 	u32 idx = hdr->pgio_mirror_idx;
1753 	int vers;
1754 	struct nfs_fh *fh;
1755 
1756 	dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1757 		__func__, hdr->inode->i_ino,
1758 		hdr->args.pgbase, (size_t)hdr->args.count, offset);
1759 
1760 	mirror = FF_LAYOUT_COMP(lseg, idx);
1761 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1762 	if (!ds)
1763 		goto out_failed;
1764 
1765 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1766 						   hdr->inode);
1767 	if (IS_ERR(ds_clnt))
1768 		goto out_failed;
1769 
1770 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1771 	if (!ds_cred)
1772 		goto out_failed;
1773 
1774 	vers = nfs4_ff_layout_ds_version(mirror);
1775 
1776 	dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1777 		ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1778 
1779 	hdr->pgio_done_cb = ff_layout_read_done_cb;
1780 	refcount_inc(&ds->ds_clp->cl_count);
1781 	hdr->ds_clp = ds->ds_clp;
1782 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1783 	if (fh)
1784 		hdr->args.fh = fh;
1785 
1786 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1787 
1788 	/*
1789 	 * Note that if we ever decide to split across DSes,
1790 	 * then we may need to handle dense-like offsets.
1791 	 */
1792 	hdr->args.offset = offset;
1793 	hdr->mds_offset = offset;
1794 
1795 	/* Perform an asynchronous read to ds */
1796 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1797 			  vers == 3 ? &ff_layout_read_call_ops_v3 :
1798 				      &ff_layout_read_call_ops_v4,
1799 			  0, RPC_TASK_SOFTCONN);
1800 	put_cred(ds_cred);
1801 	return PNFS_ATTEMPTED;
1802 
1803 out_failed:
1804 	if (ff_layout_avoid_mds_available_ds(lseg))
1805 		return PNFS_TRY_AGAIN;
1806 	trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1807 			hdr->args.offset, hdr->args.count,
1808 			IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1809 	return PNFS_NOT_ATTEMPTED;
1810 }
1811 
1812 /* Perform async writes. */
1813 static enum pnfs_try_status
1814 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1815 {
1816 	struct pnfs_layout_segment *lseg = hdr->lseg;
1817 	struct nfs4_pnfs_ds *ds;
1818 	struct rpc_clnt *ds_clnt;
1819 	struct nfs4_ff_layout_mirror *mirror;
1820 	const struct cred *ds_cred;
1821 	loff_t offset = hdr->args.offset;
1822 	int vers;
1823 	struct nfs_fh *fh;
1824 	u32 idx = hdr->pgio_mirror_idx;
1825 
1826 	mirror = FF_LAYOUT_COMP(lseg, idx);
1827 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1828 	if (!ds)
1829 		goto out_failed;
1830 
1831 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1832 						   hdr->inode);
1833 	if (IS_ERR(ds_clnt))
1834 		goto out_failed;
1835 
1836 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1837 	if (!ds_cred)
1838 		goto out_failed;
1839 
1840 	vers = nfs4_ff_layout_ds_version(mirror);
1841 
1842 	dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1843 		__func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1844 		offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1845 		vers);
1846 
1847 	hdr->pgio_done_cb = ff_layout_write_done_cb;
1848 	refcount_inc(&ds->ds_clp->cl_count);
1849 	hdr->ds_clp = ds->ds_clp;
1850 	hdr->ds_commit_idx = idx;
1851 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1852 	if (fh)
1853 		hdr->args.fh = fh;
1854 
1855 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1856 
1857 	/*
1858 	 * Note that if we ever decide to split across DSes,
1859 	 * then we may need to handle dense-like offsets.
1860 	 */
1861 	hdr->args.offset = offset;
1862 
1863 	/* Perform an asynchronous write */
1864 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1865 			  vers == 3 ? &ff_layout_write_call_ops_v3 :
1866 				      &ff_layout_write_call_ops_v4,
1867 			  sync, RPC_TASK_SOFTCONN);
1868 	put_cred(ds_cred);
1869 	return PNFS_ATTEMPTED;
1870 
1871 out_failed:
1872 	if (ff_layout_avoid_mds_available_ds(lseg))
1873 		return PNFS_TRY_AGAIN;
1874 	trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1875 			hdr->args.offset, hdr->args.count,
1876 			IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1877 	return PNFS_NOT_ATTEMPTED;
1878 }
1879 
1880 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1881 {
1882 	return i;
1883 }
1884 
1885 static struct nfs_fh *
1886 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1887 {
1888 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1889 
1890 	/* FIXME: Assume that there is only one NFS version available
1891 	 * for the DS.
1892 	 */
1893 	return &flseg->mirror_array[i]->fh_versions[0];
1894 }
1895 
1896 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1897 {
1898 	struct pnfs_layout_segment *lseg = data->lseg;
1899 	struct nfs4_pnfs_ds *ds;
1900 	struct rpc_clnt *ds_clnt;
1901 	struct nfs4_ff_layout_mirror *mirror;
1902 	const struct cred *ds_cred;
1903 	u32 idx;
1904 	int vers, ret;
1905 	struct nfs_fh *fh;
1906 
1907 	if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1908 	    test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1909 		goto out_err;
1910 
1911 	idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1912 	mirror = FF_LAYOUT_COMP(lseg, idx);
1913 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1914 	if (!ds)
1915 		goto out_err;
1916 
1917 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1918 						   data->inode);
1919 	if (IS_ERR(ds_clnt))
1920 		goto out_err;
1921 
1922 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1923 	if (!ds_cred)
1924 		goto out_err;
1925 
1926 	vers = nfs4_ff_layout_ds_version(mirror);
1927 
1928 	dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1929 		data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1930 		vers);
1931 	data->commit_done_cb = ff_layout_commit_done_cb;
1932 	data->cred = ds_cred;
1933 	refcount_inc(&ds->ds_clp->cl_count);
1934 	data->ds_clp = ds->ds_clp;
1935 	fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1936 	if (fh)
1937 		data->args.fh = fh;
1938 
1939 	ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1940 				   vers == 3 ? &ff_layout_commit_call_ops_v3 :
1941 					       &ff_layout_commit_call_ops_v4,
1942 				   how, RPC_TASK_SOFTCONN);
1943 	put_cred(ds_cred);
1944 	return ret;
1945 out_err:
1946 	pnfs_generic_prepare_to_resend_writes(data);
1947 	pnfs_generic_commit_release(data);
1948 	return -EAGAIN;
1949 }
1950 
1951 static int
1952 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1953 			   int how, struct nfs_commit_info *cinfo)
1954 {
1955 	return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1956 					    ff_layout_initiate_commit);
1957 }
1958 
1959 static struct pnfs_ds_commit_info *
1960 ff_layout_get_ds_info(struct inode *inode)
1961 {
1962 	struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1963 
1964 	if (layout == NULL)
1965 		return NULL;
1966 
1967 	return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1968 }
1969 
1970 static void
1971 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1972 		struct pnfs_layout_segment *lseg)
1973 {
1974 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1975 	struct inode *inode = lseg->pls_layout->plh_inode;
1976 	struct pnfs_commit_array *array, *new;
1977 
1978 	new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1979 	if (new) {
1980 		spin_lock(&inode->i_lock);
1981 		array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1982 		spin_unlock(&inode->i_lock);
1983 		if (array != new)
1984 			pnfs_free_commit_array(new);
1985 	}
1986 }
1987 
1988 static void
1989 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1990 		struct inode *inode)
1991 {
1992 	spin_lock(&inode->i_lock);
1993 	pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1994 	spin_unlock(&inode->i_lock);
1995 }
1996 
1997 static void
1998 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1999 {
2000 	nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2001 						  id_node));
2002 }
2003 
2004 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2005 				  const struct nfs4_layoutreturn_args *args,
2006 				  const struct nfs4_flexfile_layoutreturn_args *ff_args)
2007 {
2008 	__be32 *start;
2009 
2010 	start = xdr_reserve_space(xdr, 4);
2011 	if (unlikely(!start))
2012 		return -E2BIG;
2013 
2014 	*start = cpu_to_be32(ff_args->num_errors);
2015 	/* This assume we always return _ALL_ layouts */
2016 	return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2017 }
2018 
2019 static void
2020 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2021 {
2022 	WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2023 }
2024 
2025 static void
2026 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2027 			    const nfs4_stateid *stateid,
2028 			    const struct nfs42_layoutstat_devinfo *devinfo)
2029 {
2030 	__be32 *p;
2031 
2032 	p = xdr_reserve_space(xdr, 8 + 8);
2033 	p = xdr_encode_hyper(p, devinfo->offset);
2034 	p = xdr_encode_hyper(p, devinfo->length);
2035 	encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2036 	p = xdr_reserve_space(xdr, 4*8);
2037 	p = xdr_encode_hyper(p, devinfo->read_count);
2038 	p = xdr_encode_hyper(p, devinfo->read_bytes);
2039 	p = xdr_encode_hyper(p, devinfo->write_count);
2040 	p = xdr_encode_hyper(p, devinfo->write_bytes);
2041 	encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2042 }
2043 
2044 static void
2045 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2046 			    const nfs4_stateid *stateid,
2047 			    const struct nfs42_layoutstat_devinfo *devinfo)
2048 {
2049 	ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2050 	ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2051 			devinfo->ld_private.data);
2052 }
2053 
2054 /* report nothing for now */
2055 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2056 		const struct nfs4_layoutreturn_args *args,
2057 		struct nfs4_flexfile_layoutreturn_args *ff_args)
2058 {
2059 	__be32 *p;
2060 	int i;
2061 
2062 	p = xdr_reserve_space(xdr, 4);
2063 	*p = cpu_to_be32(ff_args->num_dev);
2064 	for (i = 0; i < ff_args->num_dev; i++)
2065 		ff_layout_encode_ff_iostat(xdr,
2066 				&args->layout->plh_stateid,
2067 				&ff_args->devinfo[i]);
2068 }
2069 
2070 static void
2071 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2072 		unsigned int num_entries)
2073 {
2074 	unsigned int i;
2075 
2076 	for (i = 0; i < num_entries; i++) {
2077 		if (!devinfo[i].ld_private.ops)
2078 			continue;
2079 		if (!devinfo[i].ld_private.ops->free)
2080 			continue;
2081 		devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2082 	}
2083 }
2084 
2085 static struct nfs4_deviceid_node *
2086 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2087 			      struct pnfs_device *pdev, gfp_t gfp_flags)
2088 {
2089 	struct nfs4_ff_layout_ds *dsaddr;
2090 
2091 	dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2092 	if (!dsaddr)
2093 		return NULL;
2094 	return &dsaddr->id_node;
2095 }
2096 
2097 static void
2098 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2099 		const void *voidargs,
2100 		const struct nfs4_xdr_opaque_data *ff_opaque)
2101 {
2102 	const struct nfs4_layoutreturn_args *args = voidargs;
2103 	struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2104 	struct xdr_buf tmp_buf = {
2105 		.head = {
2106 			[0] = {
2107 				.iov_base = page_address(ff_args->pages[0]),
2108 			},
2109 		},
2110 		.buflen = PAGE_SIZE,
2111 	};
2112 	struct xdr_stream tmp_xdr;
2113 	__be32 *start;
2114 
2115 	dprintk("%s: Begin\n", __func__);
2116 
2117 	xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2118 
2119 	ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2120 	ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2121 
2122 	start = xdr_reserve_space(xdr, 4);
2123 	*start = cpu_to_be32(tmp_buf.len);
2124 	xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2125 
2126 	dprintk("%s: Return\n", __func__);
2127 }
2128 
2129 static void
2130 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2131 {
2132 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2133 
2134 	if (!args->data)
2135 		return;
2136 	ff_args = args->data;
2137 	args->data = NULL;
2138 
2139 	ff_layout_free_ds_ioerr(&ff_args->errors);
2140 	ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2141 
2142 	put_page(ff_args->pages[0]);
2143 	kfree(ff_args);
2144 }
2145 
2146 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2147 	.encode = ff_layout_encode_layoutreturn,
2148 	.free = ff_layout_free_layoutreturn,
2149 };
2150 
2151 static int
2152 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2153 {
2154 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2155 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2156 
2157 	ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2158 	if (!ff_args)
2159 		goto out_nomem;
2160 	ff_args->pages[0] = alloc_page(GFP_KERNEL);
2161 	if (!ff_args->pages[0])
2162 		goto out_nomem_free;
2163 
2164 	INIT_LIST_HEAD(&ff_args->errors);
2165 	ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2166 			&args->range, &ff_args->errors,
2167 			FF_LAYOUTRETURN_MAXERR);
2168 
2169 	spin_lock(&args->inode->i_lock);
2170 	ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2171 			&ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2172 	spin_unlock(&args->inode->i_lock);
2173 
2174 	args->ld_private->ops = &layoutreturn_ops;
2175 	args->ld_private->data = ff_args;
2176 	return 0;
2177 out_nomem_free:
2178 	kfree(ff_args);
2179 out_nomem:
2180 	return -ENOMEM;
2181 }
2182 
2183 #ifdef CONFIG_NFS_V4_2
2184 void
2185 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2186 {
2187 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
2188 	struct nfs42_layout_error *errors;
2189 	LIST_HEAD(head);
2190 
2191 	if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2192 		return;
2193 	ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2194 	if (list_empty(&head))
2195 		return;
2196 
2197 	errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2198 			sizeof(*errors), GFP_NOFS);
2199 	if (errors != NULL) {
2200 		const struct nfs4_ff_layout_ds_err *pos;
2201 		size_t n = 0;
2202 
2203 		list_for_each_entry(pos, &head, list) {
2204 			errors[n].offset = pos->offset;
2205 			errors[n].length = pos->length;
2206 			nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2207 			errors[n].errors[0].dev_id = pos->deviceid;
2208 			errors[n].errors[0].status = pos->status;
2209 			errors[n].errors[0].opnum = pos->opnum;
2210 			n++;
2211 			if (!list_is_last(&pos->list, &head) &&
2212 			    n < NFS42_LAYOUTERROR_MAX)
2213 				continue;
2214 			if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2215 				break;
2216 			n = 0;
2217 		}
2218 		kfree(errors);
2219 	}
2220 	ff_layout_free_ds_ioerr(&head);
2221 }
2222 #else
2223 void
2224 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2225 {
2226 }
2227 #endif
2228 
2229 static int
2230 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2231 {
2232 	const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2233 
2234 	return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2235 }
2236 
2237 static size_t
2238 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2239 			  const int buflen)
2240 {
2241 	const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2242 	const struct in6_addr *addr = &sin6->sin6_addr;
2243 
2244 	/*
2245 	 * RFC 4291, Section 2.2.2
2246 	 *
2247 	 * Shorthanded ANY address
2248 	 */
2249 	if (ipv6_addr_any(addr))
2250 		return snprintf(buf, buflen, "::");
2251 
2252 	/*
2253 	 * RFC 4291, Section 2.2.2
2254 	 *
2255 	 * Shorthanded loopback address
2256 	 */
2257 	if (ipv6_addr_loopback(addr))
2258 		return snprintf(buf, buflen, "::1");
2259 
2260 	/*
2261 	 * RFC 4291, Section 2.2.3
2262 	 *
2263 	 * Special presentation address format for mapped v4
2264 	 * addresses.
2265 	 */
2266 	if (ipv6_addr_v4mapped(addr))
2267 		return snprintf(buf, buflen, "::ffff:%pI4",
2268 					&addr->s6_addr32[3]);
2269 
2270 	/*
2271 	 * RFC 4291, Section 2.2.1
2272 	 */
2273 	return snprintf(buf, buflen, "%pI6c", addr);
2274 }
2275 
2276 /* Derived from rpc_sockaddr2uaddr */
2277 static void
2278 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2279 {
2280 	struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2281 	char portbuf[RPCBIND_MAXUADDRPLEN];
2282 	char addrbuf[RPCBIND_MAXUADDRLEN];
2283 	unsigned short port;
2284 	int len, netid_len;
2285 	__be32 *p;
2286 
2287 	switch (sap->sa_family) {
2288 	case AF_INET:
2289 		if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2290 			return;
2291 		port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2292 		break;
2293 	case AF_INET6:
2294 		if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2295 			return;
2296 		port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2297 		break;
2298 	default:
2299 		WARN_ON_ONCE(1);
2300 		return;
2301 	}
2302 
2303 	snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2304 	len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2305 
2306 	netid_len = strlen(da->da_netid);
2307 	p = xdr_reserve_space(xdr, 4 + netid_len);
2308 	xdr_encode_opaque(p, da->da_netid, netid_len);
2309 
2310 	p = xdr_reserve_space(xdr, 4 + len);
2311 	xdr_encode_opaque(p, addrbuf, len);
2312 }
2313 
2314 static void
2315 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2316 			 ktime_t t)
2317 {
2318 	struct timespec64 ts;
2319 	__be32 *p;
2320 
2321 	p = xdr_reserve_space(xdr, 12);
2322 	ts = ktime_to_timespec64(t);
2323 	p = xdr_encode_hyper(p, ts.tv_sec);
2324 	*p++ = cpu_to_be32(ts.tv_nsec);
2325 }
2326 
2327 static void
2328 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2329 			    struct nfs4_ff_io_stat *stat)
2330 {
2331 	__be32 *p;
2332 
2333 	p = xdr_reserve_space(xdr, 5 * 8);
2334 	p = xdr_encode_hyper(p, stat->ops_requested);
2335 	p = xdr_encode_hyper(p, stat->bytes_requested);
2336 	p = xdr_encode_hyper(p, stat->ops_completed);
2337 	p = xdr_encode_hyper(p, stat->bytes_completed);
2338 	p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2339 	ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2340 	ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2341 }
2342 
2343 static void
2344 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2345 			      const struct nfs42_layoutstat_devinfo *devinfo,
2346 			      struct nfs4_ff_layout_mirror *mirror)
2347 {
2348 	struct nfs4_pnfs_ds_addr *da;
2349 	struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2350 	struct nfs_fh *fh = &mirror->fh_versions[0];
2351 	__be32 *p;
2352 
2353 	da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2354 	dprintk("%s: DS %s: encoding address %s\n",
2355 		__func__, ds->ds_remotestr, da->da_remotestr);
2356 	/* netaddr4 */
2357 	ff_layout_encode_netaddr(xdr, da);
2358 	/* nfs_fh4 */
2359 	p = xdr_reserve_space(xdr, 4 + fh->size);
2360 	xdr_encode_opaque(p, fh->data, fh->size);
2361 	/* ff_io_latency4 read */
2362 	spin_lock(&mirror->lock);
2363 	ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2364 	/* ff_io_latency4 write */
2365 	ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2366 	spin_unlock(&mirror->lock);
2367 	/* nfstime4 */
2368 	ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2369 	/* bool */
2370 	p = xdr_reserve_space(xdr, 4);
2371 	*p = cpu_to_be32(false);
2372 }
2373 
2374 static void
2375 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2376 			     const struct nfs4_xdr_opaque_data *opaque)
2377 {
2378 	struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2379 			struct nfs42_layoutstat_devinfo, ld_private);
2380 	__be32 *start;
2381 
2382 	/* layoutupdate length */
2383 	start = xdr_reserve_space(xdr, 4);
2384 	ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2385 
2386 	*start = cpu_to_be32((xdr->p - start - 1) * 4);
2387 }
2388 
2389 static void
2390 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2391 {
2392 	struct nfs4_ff_layout_mirror *mirror = opaque->data;
2393 
2394 	ff_layout_put_mirror(mirror);
2395 }
2396 
2397 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2398 	.encode = ff_layout_encode_layoutstats,
2399 	.free	= ff_layout_free_layoutstats,
2400 };
2401 
2402 static int
2403 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2404 			       struct nfs42_layoutstat_devinfo *devinfo,
2405 			       int dev_limit)
2406 {
2407 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2408 	struct nfs4_ff_layout_mirror *mirror;
2409 	struct nfs4_deviceid_node *dev;
2410 	int i = 0;
2411 
2412 	list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2413 		if (i >= dev_limit)
2414 			break;
2415 		if (IS_ERR_OR_NULL(mirror->mirror_ds))
2416 			continue;
2417 		if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2418 			continue;
2419 		/* mirror refcount put in cleanup_layoutstats */
2420 		if (!refcount_inc_not_zero(&mirror->ref))
2421 			continue;
2422 		dev = &mirror->mirror_ds->id_node;
2423 		memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2424 		devinfo->offset = 0;
2425 		devinfo->length = NFS4_MAX_UINT64;
2426 		spin_lock(&mirror->lock);
2427 		devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2428 		devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2429 		devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2430 		devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2431 		spin_unlock(&mirror->lock);
2432 		devinfo->layout_type = LAYOUT_FLEX_FILES;
2433 		devinfo->ld_private.ops = &layoutstat_ops;
2434 		devinfo->ld_private.data = mirror;
2435 
2436 		devinfo++;
2437 		i++;
2438 	}
2439 	return i;
2440 }
2441 
2442 static int
2443 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2444 {
2445 	struct nfs4_flexfile_layout *ff_layout;
2446 	const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2447 
2448 	/* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2449 	args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2450 	if (!args->devinfo)
2451 		return -ENOMEM;
2452 
2453 	spin_lock(&args->inode->i_lock);
2454 	ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2455 	args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2456 			&args->devinfo[0], dev_count);
2457 	spin_unlock(&args->inode->i_lock);
2458 	if (!args->num_dev) {
2459 		kfree(args->devinfo);
2460 		args->devinfo = NULL;
2461 		return -ENOENT;
2462 	}
2463 
2464 	return 0;
2465 }
2466 
2467 static int
2468 ff_layout_set_layoutdriver(struct nfs_server *server,
2469 		const struct nfs_fh *dummy)
2470 {
2471 #if IS_ENABLED(CONFIG_NFS_V4_2)
2472 	server->caps |= NFS_CAP_LAYOUTSTATS;
2473 #endif
2474 	return 0;
2475 }
2476 
2477 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2478 	.setup_ds_info		= ff_layout_setup_ds_info,
2479 	.release_ds_info	= ff_layout_release_ds_info,
2480 	.mark_request_commit	= pnfs_layout_mark_request_commit,
2481 	.clear_request_commit	= pnfs_generic_clear_request_commit,
2482 	.scan_commit_lists	= pnfs_generic_scan_commit_lists,
2483 	.recover_commit_reqs	= pnfs_generic_recover_commit_reqs,
2484 	.commit_pagelist	= ff_layout_commit_pagelist,
2485 };
2486 
2487 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2488 	.id			= LAYOUT_FLEX_FILES,
2489 	.name			= "LAYOUT_FLEX_FILES",
2490 	.owner			= THIS_MODULE,
2491 	.flags			= PNFS_LAYOUTGET_ON_OPEN,
2492 	.max_layoutget_response	= 4096, /* 1 page or so... */
2493 	.set_layoutdriver	= ff_layout_set_layoutdriver,
2494 	.alloc_layout_hdr	= ff_layout_alloc_layout_hdr,
2495 	.free_layout_hdr	= ff_layout_free_layout_hdr,
2496 	.alloc_lseg		= ff_layout_alloc_lseg,
2497 	.free_lseg		= ff_layout_free_lseg,
2498 	.add_lseg		= ff_layout_add_lseg,
2499 	.pg_read_ops		= &ff_layout_pg_read_ops,
2500 	.pg_write_ops		= &ff_layout_pg_write_ops,
2501 	.get_ds_info		= ff_layout_get_ds_info,
2502 	.free_deviceid_node	= ff_layout_free_deviceid_node,
2503 	.read_pagelist		= ff_layout_read_pagelist,
2504 	.write_pagelist		= ff_layout_write_pagelist,
2505 	.alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2506 	.prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2507 	.sync			= pnfs_nfs_generic_sync,
2508 	.prepare_layoutstats	= ff_layout_prepare_layoutstats,
2509 };
2510 
2511 static int __init nfs4flexfilelayout_init(void)
2512 {
2513 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2514 	       __func__);
2515 	return pnfs_register_layoutdriver(&flexfilelayout_type);
2516 }
2517 
2518 static void __exit nfs4flexfilelayout_exit(void)
2519 {
2520 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2521 	       __func__);
2522 	pnfs_unregister_layoutdriver(&flexfilelayout_type);
2523 }
2524 
2525 MODULE_ALIAS("nfs-layouttype4-4");
2526 
2527 MODULE_LICENSE("GPL");
2528 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2529 
2530 module_init(nfs4flexfilelayout_init);
2531 module_exit(nfs4flexfilelayout_exit);
2532 
2533 module_param(io_maxretrans, ushort, 0644);
2534 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2535 			"retries an I/O request before returning an error. ");
2536