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_buffer(&stream, page_address(scratch), PAGE_SIZE);
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, int 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, int 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 			     int start_idx, int *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 	bool fail_return = false;
744 	int idx;
745 
746 	/* mirrors are initially sorted by efficiency */
747 	for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
748 		if (idx+1 == fls->mirror_array_cnt)
749 			fail_return = !check_device;
750 
751 		mirror = FF_LAYOUT_COMP(lseg, idx);
752 		ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
753 		if (!ds)
754 			continue;
755 
756 		if (check_device &&
757 		    nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
758 			continue;
759 
760 		*best_idx = idx;
761 		return ds;
762 	}
763 
764 	return NULL;
765 }
766 
767 static struct nfs4_pnfs_ds *
768 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
769 				 int start_idx, int *best_idx)
770 {
771 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
772 }
773 
774 static struct nfs4_pnfs_ds *
775 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
776 				   int start_idx, int *best_idx)
777 {
778 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
779 }
780 
781 static struct nfs4_pnfs_ds *
782 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
783 				  int start_idx, int *best_idx)
784 {
785 	struct nfs4_pnfs_ds *ds;
786 
787 	ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
788 	if (ds)
789 		return ds;
790 	return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
791 }
792 
793 static void
794 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
795 		      struct nfs_page *req,
796 		      bool strict_iomode)
797 {
798 	pnfs_put_lseg(pgio->pg_lseg);
799 	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
800 					   nfs_req_openctx(req),
801 					   req_offset(req),
802 					   req->wb_bytes,
803 					   IOMODE_READ,
804 					   strict_iomode,
805 					   GFP_KERNEL);
806 	if (IS_ERR(pgio->pg_lseg)) {
807 		pgio->pg_error = PTR_ERR(pgio->pg_lseg);
808 		pgio->pg_lseg = NULL;
809 	}
810 }
811 
812 static void
813 ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
814 			  struct nfs_page *req)
815 {
816 	pnfs_generic_pg_check_layout(pgio);
817 	pnfs_generic_pg_check_range(pgio, req);
818 }
819 
820 static void
821 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
822 			struct nfs_page *req)
823 {
824 	struct nfs_pgio_mirror *pgm;
825 	struct nfs4_ff_layout_mirror *mirror;
826 	struct nfs4_pnfs_ds *ds;
827 	int ds_idx;
828 
829 retry:
830 	ff_layout_pg_check_layout(pgio, req);
831 	/* Use full layout for now */
832 	if (!pgio->pg_lseg) {
833 		ff_layout_pg_get_read(pgio, req, false);
834 		if (!pgio->pg_lseg)
835 			goto out_nolseg;
836 	}
837 	if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
838 		ff_layout_pg_get_read(pgio, req, true);
839 		if (!pgio->pg_lseg)
840 			goto out_nolseg;
841 	}
842 
843 	ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
844 	if (!ds) {
845 		if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
846 			goto out_mds;
847 		pnfs_put_lseg(pgio->pg_lseg);
848 		pgio->pg_lseg = NULL;
849 		/* Sleep for 1 second before retrying */
850 		ssleep(1);
851 		goto retry;
852 	}
853 
854 	mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
855 
856 	pgio->pg_mirror_idx = ds_idx;
857 
858 	/* read always uses only one mirror - idx 0 for pgio layer */
859 	pgm = &pgio->pg_mirrors[0];
860 	pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
861 
862 	if (NFS_SERVER(pgio->pg_inode)->flags &
863 			(NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
864 		pgio->pg_maxretrans = io_maxretrans;
865 	return;
866 out_nolseg:
867 	if (pgio->pg_error < 0)
868 		return;
869 out_mds:
870 	trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
871 			0, NFS4_MAX_UINT64, IOMODE_READ,
872 			NFS_I(pgio->pg_inode)->layout,
873 			pgio->pg_lseg);
874 	pnfs_put_lseg(pgio->pg_lseg);
875 	pgio->pg_lseg = NULL;
876 	pgio->pg_maxretrans = 0;
877 	nfs_pageio_reset_read_mds(pgio);
878 }
879 
880 static void
881 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
882 			struct nfs_page *req)
883 {
884 	struct nfs4_ff_layout_mirror *mirror;
885 	struct nfs_pgio_mirror *pgm;
886 	struct nfs4_pnfs_ds *ds;
887 	int i;
888 
889 retry:
890 	ff_layout_pg_check_layout(pgio, req);
891 	if (!pgio->pg_lseg) {
892 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
893 						   nfs_req_openctx(req),
894 						   req_offset(req),
895 						   req->wb_bytes,
896 						   IOMODE_RW,
897 						   false,
898 						   GFP_NOFS);
899 		if (IS_ERR(pgio->pg_lseg)) {
900 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
901 			pgio->pg_lseg = NULL;
902 			return;
903 		}
904 	}
905 	/* If no lseg, fall back to write through mds */
906 	if (pgio->pg_lseg == NULL)
907 		goto out_mds;
908 
909 	/* Use a direct mapping of ds_idx to pgio mirror_idx */
910 	if (WARN_ON_ONCE(pgio->pg_mirror_count !=
911 	    FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
912 		goto out_mds;
913 
914 	for (i = 0; i < pgio->pg_mirror_count; i++) {
915 		mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
916 		ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
917 		if (!ds) {
918 			if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
919 				goto out_mds;
920 			pnfs_put_lseg(pgio->pg_lseg);
921 			pgio->pg_lseg = NULL;
922 			/* Sleep for 1 second before retrying */
923 			ssleep(1);
924 			goto retry;
925 		}
926 		pgm = &pgio->pg_mirrors[i];
927 		pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
928 	}
929 
930 	if (NFS_SERVER(pgio->pg_inode)->flags &
931 			(NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
932 		pgio->pg_maxretrans = io_maxretrans;
933 	return;
934 
935 out_mds:
936 	trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
937 			0, NFS4_MAX_UINT64, IOMODE_RW,
938 			NFS_I(pgio->pg_inode)->layout,
939 			pgio->pg_lseg);
940 	pnfs_put_lseg(pgio->pg_lseg);
941 	pgio->pg_lseg = NULL;
942 	pgio->pg_maxretrans = 0;
943 	nfs_pageio_reset_write_mds(pgio);
944 }
945 
946 static unsigned int
947 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
948 				    struct nfs_page *req)
949 {
950 	if (!pgio->pg_lseg) {
951 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
952 						   nfs_req_openctx(req),
953 						   0,
954 						   NFS4_MAX_UINT64,
955 						   IOMODE_RW,
956 						   false,
957 						   GFP_NOFS);
958 		if (IS_ERR(pgio->pg_lseg)) {
959 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
960 			pgio->pg_lseg = NULL;
961 			goto out;
962 		}
963 	}
964 	if (pgio->pg_lseg)
965 		return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
966 
967 	trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
968 			0, NFS4_MAX_UINT64, IOMODE_RW,
969 			NFS_I(pgio->pg_inode)->layout,
970 			pgio->pg_lseg);
971 	/* no lseg means that pnfs is not in use, so no mirroring here */
972 	nfs_pageio_reset_write_mds(pgio);
973 out:
974 	return 1;
975 }
976 
977 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
978 	.pg_init = ff_layout_pg_init_read,
979 	.pg_test = pnfs_generic_pg_test,
980 	.pg_doio = pnfs_generic_pg_readpages,
981 	.pg_cleanup = pnfs_generic_pg_cleanup,
982 };
983 
984 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
985 	.pg_init = ff_layout_pg_init_write,
986 	.pg_test = pnfs_generic_pg_test,
987 	.pg_doio = pnfs_generic_pg_writepages,
988 	.pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
989 	.pg_cleanup = pnfs_generic_pg_cleanup,
990 };
991 
992 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
993 {
994 	struct rpc_task *task = &hdr->task;
995 
996 	pnfs_layoutcommit_inode(hdr->inode, false);
997 
998 	if (retry_pnfs) {
999 		dprintk("%s Reset task %5u for i/o through pNFS "
1000 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1001 			hdr->task.tk_pid,
1002 			hdr->inode->i_sb->s_id,
1003 			(unsigned long long)NFS_FILEID(hdr->inode),
1004 			hdr->args.count,
1005 			(unsigned long long)hdr->args.offset);
1006 
1007 		hdr->completion_ops->reschedule_io(hdr);
1008 		return;
1009 	}
1010 
1011 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1012 		dprintk("%s Reset task %5u for i/o through MDS "
1013 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1014 			hdr->task.tk_pid,
1015 			hdr->inode->i_sb->s_id,
1016 			(unsigned long long)NFS_FILEID(hdr->inode),
1017 			hdr->args.count,
1018 			(unsigned long long)hdr->args.offset);
1019 
1020 		trace_pnfs_mds_fallback_write_done(hdr->inode,
1021 				hdr->args.offset, hdr->args.count,
1022 				IOMODE_RW, NFS_I(hdr->inode)->layout,
1023 				hdr->lseg);
1024 		task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1025 	}
1026 }
1027 
1028 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1029 {
1030 	struct rpc_task *task = &hdr->task;
1031 
1032 	pnfs_layoutcommit_inode(hdr->inode, false);
1033 
1034 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1035 		dprintk("%s Reset task %5u for i/o through MDS "
1036 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1037 			hdr->task.tk_pid,
1038 			hdr->inode->i_sb->s_id,
1039 			(unsigned long long)NFS_FILEID(hdr->inode),
1040 			hdr->args.count,
1041 			(unsigned long long)hdr->args.offset);
1042 
1043 		trace_pnfs_mds_fallback_read_done(hdr->inode,
1044 				hdr->args.offset, hdr->args.count,
1045 				IOMODE_READ, NFS_I(hdr->inode)->layout,
1046 				hdr->lseg);
1047 		task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1048 	}
1049 }
1050 
1051 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1052 					   struct nfs4_state *state,
1053 					   struct nfs_client *clp,
1054 					   struct pnfs_layout_segment *lseg,
1055 					   int idx)
1056 {
1057 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
1058 	struct inode *inode = lo->plh_inode;
1059 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1060 	struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1061 
1062 	switch (task->tk_status) {
1063 	case -NFS4ERR_BADSESSION:
1064 	case -NFS4ERR_BADSLOT:
1065 	case -NFS4ERR_BAD_HIGH_SLOT:
1066 	case -NFS4ERR_DEADSESSION:
1067 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1068 	case -NFS4ERR_SEQ_FALSE_RETRY:
1069 	case -NFS4ERR_SEQ_MISORDERED:
1070 		dprintk("%s ERROR %d, Reset session. Exchangeid "
1071 			"flags 0x%x\n", __func__, task->tk_status,
1072 			clp->cl_exchange_flags);
1073 		nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1074 		break;
1075 	case -NFS4ERR_DELAY:
1076 	case -NFS4ERR_GRACE:
1077 		rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1078 		break;
1079 	case -NFS4ERR_RETRY_UNCACHED_REP:
1080 		break;
1081 	/* Invalidate Layout errors */
1082 	case -NFS4ERR_PNFS_NO_LAYOUT:
1083 	case -ESTALE:           /* mapped NFS4ERR_STALE */
1084 	case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1085 	case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1086 	case -NFS4ERR_FHEXPIRED:
1087 	case -NFS4ERR_WRONG_TYPE:
1088 		dprintk("%s Invalid layout error %d\n", __func__,
1089 			task->tk_status);
1090 		/*
1091 		 * Destroy layout so new i/o will get a new layout.
1092 		 * Layout will not be destroyed until all current lseg
1093 		 * references are put. Mark layout as invalid to resend failed
1094 		 * i/o and all i/o waiting on the slot table to the MDS until
1095 		 * layout is destroyed and a new valid layout is obtained.
1096 		 */
1097 		pnfs_destroy_layout(NFS_I(inode));
1098 		rpc_wake_up(&tbl->slot_tbl_waitq);
1099 		goto reset;
1100 	/* RPC connection errors */
1101 	case -ECONNREFUSED:
1102 	case -EHOSTDOWN:
1103 	case -EHOSTUNREACH:
1104 	case -ENETUNREACH:
1105 	case -EIO:
1106 	case -ETIMEDOUT:
1107 	case -EPIPE:
1108 		dprintk("%s DS connection error %d\n", __func__,
1109 			task->tk_status);
1110 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1111 				&devid->deviceid);
1112 		rpc_wake_up(&tbl->slot_tbl_waitq);
1113 		/* fall through */
1114 	default:
1115 		if (ff_layout_avoid_mds_available_ds(lseg))
1116 			return -NFS4ERR_RESET_TO_PNFS;
1117 reset:
1118 		dprintk("%s Retry through MDS. Error %d\n", __func__,
1119 			task->tk_status);
1120 		return -NFS4ERR_RESET_TO_MDS;
1121 	}
1122 	task->tk_status = 0;
1123 	return -EAGAIN;
1124 }
1125 
1126 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1127 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1128 					   struct pnfs_layout_segment *lseg,
1129 					   int idx)
1130 {
1131 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1132 
1133 	switch (task->tk_status) {
1134 	/* File access problems. Don't mark the device as unavailable */
1135 	case -EACCES:
1136 	case -ESTALE:
1137 	case -EISDIR:
1138 	case -EBADHANDLE:
1139 	case -ELOOP:
1140 	case -ENOSPC:
1141 		break;
1142 	case -EJUKEBOX:
1143 		nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1144 		goto out_retry;
1145 	default:
1146 		dprintk("%s DS connection error %d\n", __func__,
1147 			task->tk_status);
1148 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1149 				&devid->deviceid);
1150 	}
1151 	/* FIXME: Need to prevent infinite looping here. */
1152 	return -NFS4ERR_RESET_TO_PNFS;
1153 out_retry:
1154 	task->tk_status = 0;
1155 	rpc_restart_call_prepare(task);
1156 	rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1157 	return -EAGAIN;
1158 }
1159 
1160 static int ff_layout_async_handle_error(struct rpc_task *task,
1161 					struct nfs4_state *state,
1162 					struct nfs_client *clp,
1163 					struct pnfs_layout_segment *lseg,
1164 					int idx)
1165 {
1166 	int vers = clp->cl_nfs_mod->rpc_vers->number;
1167 
1168 	if (task->tk_status >= 0) {
1169 		ff_layout_mark_ds_reachable(lseg, idx);
1170 		return 0;
1171 	}
1172 
1173 	/* Handle the case of an invalid layout segment */
1174 	if (!pnfs_is_valid_lseg(lseg))
1175 		return -NFS4ERR_RESET_TO_PNFS;
1176 
1177 	switch (vers) {
1178 	case 3:
1179 		return ff_layout_async_handle_error_v3(task, lseg, idx);
1180 	case 4:
1181 		return ff_layout_async_handle_error_v4(task, state, clp,
1182 						       lseg, idx);
1183 	default:
1184 		/* should never happen */
1185 		WARN_ON_ONCE(1);
1186 		return 0;
1187 	}
1188 }
1189 
1190 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1191 					int idx, u64 offset, u64 length,
1192 					u32 *op_status, int opnum, int error)
1193 {
1194 	struct nfs4_ff_layout_mirror *mirror;
1195 	u32 status = *op_status;
1196 	int err;
1197 
1198 	if (status == 0) {
1199 		switch (error) {
1200 		case -ETIMEDOUT:
1201 		case -EPFNOSUPPORT:
1202 		case -EPROTONOSUPPORT:
1203 		case -EOPNOTSUPP:
1204 		case -ECONNREFUSED:
1205 		case -ECONNRESET:
1206 		case -EHOSTDOWN:
1207 		case -EHOSTUNREACH:
1208 		case -ENETUNREACH:
1209 		case -EADDRINUSE:
1210 		case -ENOBUFS:
1211 		case -EPIPE:
1212 		case -EPERM:
1213 			*op_status = status = NFS4ERR_NXIO;
1214 			break;
1215 		case -EACCES:
1216 			*op_status = status = NFS4ERR_ACCESS;
1217 			break;
1218 		default:
1219 			return;
1220 		}
1221 	}
1222 
1223 	mirror = FF_LAYOUT_COMP(lseg, idx);
1224 	err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1225 				       mirror, offset, length, status, opnum,
1226 				       GFP_NOIO);
1227 
1228 	switch (status) {
1229 	case NFS4ERR_DELAY:
1230 	case NFS4ERR_GRACE:
1231 		break;
1232 	case NFS4ERR_NXIO:
1233 		ff_layout_mark_ds_unreachable(lseg, idx);
1234 		/* Fallthrough */
1235 	default:
1236 		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1237 						  lseg);
1238 	}
1239 
1240 	dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1241 }
1242 
1243 /* NFS_PROTO call done callback routines */
1244 static int ff_layout_read_done_cb(struct rpc_task *task,
1245 				struct nfs_pgio_header *hdr)
1246 {
1247 	int new_idx = hdr->pgio_mirror_idx;
1248 	int err;
1249 
1250 	if (task->tk_status < 0) {
1251 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1252 					    hdr->args.offset, hdr->args.count,
1253 					    &hdr->res.op_status, OP_READ,
1254 					    task->tk_status);
1255 		trace_ff_layout_read_error(hdr);
1256 	}
1257 
1258 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1259 					   hdr->ds_clp, hdr->lseg,
1260 					   hdr->pgio_mirror_idx);
1261 
1262 	trace_nfs4_pnfs_read(hdr, err);
1263 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1264 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1265 	switch (err) {
1266 	case -NFS4ERR_RESET_TO_PNFS:
1267 		if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1268 					hdr->pgio_mirror_idx + 1,
1269 					&new_idx))
1270 			goto out_layouterror;
1271 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1272 		return task->tk_status;
1273 	case -NFS4ERR_RESET_TO_MDS:
1274 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1275 		return task->tk_status;
1276 	case -EAGAIN:
1277 		goto out_eagain;
1278 	}
1279 
1280 	return 0;
1281 out_layouterror:
1282 	ff_layout_read_record_layoutstats_done(task, hdr);
1283 	ff_layout_send_layouterror(hdr->lseg);
1284 	hdr->pgio_mirror_idx = new_idx;
1285 out_eagain:
1286 	rpc_restart_call_prepare(task);
1287 	return -EAGAIN;
1288 }
1289 
1290 static bool
1291 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1292 {
1293 	return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1294 }
1295 
1296 /*
1297  * We reference the rpc_cred of the first WRITE that triggers the need for
1298  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1299  * rfc5661 is not clear about which credential should be used.
1300  *
1301  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1302  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1303  * we always send layoutcommit after DS writes.
1304  */
1305 static void
1306 ff_layout_set_layoutcommit(struct inode *inode,
1307 		struct pnfs_layout_segment *lseg,
1308 		loff_t end_offset)
1309 {
1310 	if (!ff_layout_need_layoutcommit(lseg))
1311 		return;
1312 
1313 	pnfs_set_layoutcommit(inode, lseg, end_offset);
1314 	dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1315 		(unsigned long long) NFS_I(inode)->layout->plh_lwb);
1316 }
1317 
1318 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1319 		struct nfs_pgio_header *hdr)
1320 {
1321 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1322 		return;
1323 	nfs4_ff_layout_stat_io_start_read(hdr->inode,
1324 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1325 			hdr->args.count,
1326 			task->tk_start);
1327 }
1328 
1329 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1330 		struct nfs_pgio_header *hdr)
1331 {
1332 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1333 		return;
1334 	nfs4_ff_layout_stat_io_end_read(task,
1335 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1336 			hdr->args.count,
1337 			hdr->res.count);
1338 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1339 }
1340 
1341 static int ff_layout_read_prepare_common(struct rpc_task *task,
1342 					 struct nfs_pgio_header *hdr)
1343 {
1344 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1345 		rpc_exit(task, -EIO);
1346 		return -EIO;
1347 	}
1348 
1349 	ff_layout_read_record_layoutstats_start(task, hdr);
1350 	return 0;
1351 }
1352 
1353 /*
1354  * Call ops for the async read/write cases
1355  * In the case of dense layouts, the offset needs to be reset to its
1356  * original value.
1357  */
1358 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1359 {
1360 	struct nfs_pgio_header *hdr = data;
1361 
1362 	if (ff_layout_read_prepare_common(task, hdr))
1363 		return;
1364 
1365 	rpc_call_start(task);
1366 }
1367 
1368 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1369 {
1370 	struct nfs_pgio_header *hdr = data;
1371 
1372 	if (nfs4_setup_sequence(hdr->ds_clp,
1373 				&hdr->args.seq_args,
1374 				&hdr->res.seq_res,
1375 				task))
1376 		return;
1377 
1378 	ff_layout_read_prepare_common(task, hdr);
1379 }
1380 
1381 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1382 {
1383 	struct nfs_pgio_header *hdr = data;
1384 
1385 	dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1386 
1387 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1388 	    task->tk_status == 0) {
1389 		nfs4_sequence_done(task, &hdr->res.seq_res);
1390 		return;
1391 	}
1392 
1393 	/* Note this may cause RPC to be resent */
1394 	hdr->mds_ops->rpc_call_done(task, hdr);
1395 }
1396 
1397 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1398 {
1399 	struct nfs_pgio_header *hdr = data;
1400 
1401 	ff_layout_read_record_layoutstats_done(task, hdr);
1402 	rpc_count_iostats_metrics(task,
1403 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1404 }
1405 
1406 static void ff_layout_read_release(void *data)
1407 {
1408 	struct nfs_pgio_header *hdr = data;
1409 
1410 	ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1411 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1412 		ff_layout_send_layouterror(hdr->lseg);
1413 		pnfs_read_resend_pnfs(hdr);
1414 	} else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1415 		ff_layout_reset_read(hdr);
1416 	pnfs_generic_rw_release(data);
1417 }
1418 
1419 
1420 static int ff_layout_write_done_cb(struct rpc_task *task,
1421 				struct nfs_pgio_header *hdr)
1422 {
1423 	loff_t end_offs = 0;
1424 	int err;
1425 
1426 	if (task->tk_status < 0) {
1427 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1428 					    hdr->args.offset, hdr->args.count,
1429 					    &hdr->res.op_status, OP_WRITE,
1430 					    task->tk_status);
1431 		trace_ff_layout_write_error(hdr);
1432 	}
1433 
1434 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1435 					   hdr->ds_clp, hdr->lseg,
1436 					   hdr->pgio_mirror_idx);
1437 
1438 	trace_nfs4_pnfs_write(hdr, err);
1439 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1440 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1441 	switch (err) {
1442 	case -NFS4ERR_RESET_TO_PNFS:
1443 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1444 		return task->tk_status;
1445 	case -NFS4ERR_RESET_TO_MDS:
1446 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1447 		return task->tk_status;
1448 	case -EAGAIN:
1449 		return -EAGAIN;
1450 	}
1451 
1452 	if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1453 	    hdr->res.verf->committed == NFS_DATA_SYNC)
1454 		end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1455 
1456 	/* Note: if the write is unstable, don't set end_offs until commit */
1457 	ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1458 
1459 	/* zero out fattr since we don't care DS attr at all */
1460 	hdr->fattr.valid = 0;
1461 	if (task->tk_status >= 0)
1462 		nfs_writeback_update_inode(hdr);
1463 
1464 	return 0;
1465 }
1466 
1467 static int ff_layout_commit_done_cb(struct rpc_task *task,
1468 				     struct nfs_commit_data *data)
1469 {
1470 	int err;
1471 
1472 	if (task->tk_status < 0) {
1473 		ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1474 					    data->args.offset, data->args.count,
1475 					    &data->res.op_status, OP_COMMIT,
1476 					    task->tk_status);
1477 		trace_ff_layout_commit_error(data);
1478 	}
1479 
1480 	err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1481 					   data->lseg, data->ds_commit_index);
1482 
1483 	trace_nfs4_pnfs_commit_ds(data, err);
1484 	switch (err) {
1485 	case -NFS4ERR_RESET_TO_PNFS:
1486 		pnfs_generic_prepare_to_resend_writes(data);
1487 		return -EAGAIN;
1488 	case -NFS4ERR_RESET_TO_MDS:
1489 		pnfs_generic_prepare_to_resend_writes(data);
1490 		return -EAGAIN;
1491 	case -EAGAIN:
1492 		rpc_restart_call_prepare(task);
1493 		return -EAGAIN;
1494 	}
1495 
1496 	ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1497 
1498 	return 0;
1499 }
1500 
1501 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1502 		struct nfs_pgio_header *hdr)
1503 {
1504 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1505 		return;
1506 	nfs4_ff_layout_stat_io_start_write(hdr->inode,
1507 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1508 			hdr->args.count,
1509 			task->tk_start);
1510 }
1511 
1512 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1513 		struct nfs_pgio_header *hdr)
1514 {
1515 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1516 		return;
1517 	nfs4_ff_layout_stat_io_end_write(task,
1518 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1519 			hdr->args.count, hdr->res.count,
1520 			hdr->res.verf->committed);
1521 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1522 }
1523 
1524 static int ff_layout_write_prepare_common(struct rpc_task *task,
1525 					  struct nfs_pgio_header *hdr)
1526 {
1527 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1528 		rpc_exit(task, -EIO);
1529 		return -EIO;
1530 	}
1531 
1532 	ff_layout_write_record_layoutstats_start(task, hdr);
1533 	return 0;
1534 }
1535 
1536 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1537 {
1538 	struct nfs_pgio_header *hdr = data;
1539 
1540 	if (ff_layout_write_prepare_common(task, hdr))
1541 		return;
1542 
1543 	rpc_call_start(task);
1544 }
1545 
1546 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1547 {
1548 	struct nfs_pgio_header *hdr = data;
1549 
1550 	if (nfs4_setup_sequence(hdr->ds_clp,
1551 				&hdr->args.seq_args,
1552 				&hdr->res.seq_res,
1553 				task))
1554 		return;
1555 
1556 	ff_layout_write_prepare_common(task, hdr);
1557 }
1558 
1559 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1560 {
1561 	struct nfs_pgio_header *hdr = data;
1562 
1563 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1564 	    task->tk_status == 0) {
1565 		nfs4_sequence_done(task, &hdr->res.seq_res);
1566 		return;
1567 	}
1568 
1569 	/* Note this may cause RPC to be resent */
1570 	hdr->mds_ops->rpc_call_done(task, hdr);
1571 }
1572 
1573 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1574 {
1575 	struct nfs_pgio_header *hdr = data;
1576 
1577 	ff_layout_write_record_layoutstats_done(task, hdr);
1578 	rpc_count_iostats_metrics(task,
1579 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1580 }
1581 
1582 static void ff_layout_write_release(void *data)
1583 {
1584 	struct nfs_pgio_header *hdr = data;
1585 
1586 	ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1587 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1588 		ff_layout_send_layouterror(hdr->lseg);
1589 		ff_layout_reset_write(hdr, true);
1590 	} else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1591 		ff_layout_reset_write(hdr, false);
1592 	pnfs_generic_rw_release(data);
1593 }
1594 
1595 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1596 		struct nfs_commit_data *cdata)
1597 {
1598 	if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1599 		return;
1600 	nfs4_ff_layout_stat_io_start_write(cdata->inode,
1601 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1602 			0, task->tk_start);
1603 }
1604 
1605 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1606 		struct nfs_commit_data *cdata)
1607 {
1608 	struct nfs_page *req;
1609 	__u64 count = 0;
1610 
1611 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1612 		return;
1613 
1614 	if (task->tk_status == 0) {
1615 		list_for_each_entry(req, &cdata->pages, wb_list)
1616 			count += req->wb_bytes;
1617 	}
1618 	nfs4_ff_layout_stat_io_end_write(task,
1619 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1620 			count, count, NFS_FILE_SYNC);
1621 	set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1622 }
1623 
1624 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1625 		struct nfs_commit_data *cdata)
1626 {
1627 	ff_layout_commit_record_layoutstats_start(task, cdata);
1628 }
1629 
1630 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1631 {
1632 	ff_layout_commit_prepare_common(task, data);
1633 	rpc_call_start(task);
1634 }
1635 
1636 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1637 {
1638 	struct nfs_commit_data *wdata = data;
1639 
1640 	if (nfs4_setup_sequence(wdata->ds_clp,
1641 				&wdata->args.seq_args,
1642 				&wdata->res.seq_res,
1643 				task))
1644 		return;
1645 	ff_layout_commit_prepare_common(task, data);
1646 }
1647 
1648 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1649 {
1650 	pnfs_generic_write_commit_done(task, data);
1651 }
1652 
1653 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1654 {
1655 	struct nfs_commit_data *cdata = data;
1656 
1657 	ff_layout_commit_record_layoutstats_done(task, cdata);
1658 	rpc_count_iostats_metrics(task,
1659 	    &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1660 }
1661 
1662 static void ff_layout_commit_release(void *data)
1663 {
1664 	struct nfs_commit_data *cdata = data;
1665 
1666 	ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1667 	pnfs_generic_commit_release(data);
1668 }
1669 
1670 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1671 	.rpc_call_prepare = ff_layout_read_prepare_v3,
1672 	.rpc_call_done = ff_layout_read_call_done,
1673 	.rpc_count_stats = ff_layout_read_count_stats,
1674 	.rpc_release = ff_layout_read_release,
1675 };
1676 
1677 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1678 	.rpc_call_prepare = ff_layout_read_prepare_v4,
1679 	.rpc_call_done = ff_layout_read_call_done,
1680 	.rpc_count_stats = ff_layout_read_count_stats,
1681 	.rpc_release = ff_layout_read_release,
1682 };
1683 
1684 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1685 	.rpc_call_prepare = ff_layout_write_prepare_v3,
1686 	.rpc_call_done = ff_layout_write_call_done,
1687 	.rpc_count_stats = ff_layout_write_count_stats,
1688 	.rpc_release = ff_layout_write_release,
1689 };
1690 
1691 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1692 	.rpc_call_prepare = ff_layout_write_prepare_v4,
1693 	.rpc_call_done = ff_layout_write_call_done,
1694 	.rpc_count_stats = ff_layout_write_count_stats,
1695 	.rpc_release = ff_layout_write_release,
1696 };
1697 
1698 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1699 	.rpc_call_prepare = ff_layout_commit_prepare_v3,
1700 	.rpc_call_done = ff_layout_commit_done,
1701 	.rpc_count_stats = ff_layout_commit_count_stats,
1702 	.rpc_release = ff_layout_commit_release,
1703 };
1704 
1705 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1706 	.rpc_call_prepare = ff_layout_commit_prepare_v4,
1707 	.rpc_call_done = ff_layout_commit_done,
1708 	.rpc_count_stats = ff_layout_commit_count_stats,
1709 	.rpc_release = ff_layout_commit_release,
1710 };
1711 
1712 static enum pnfs_try_status
1713 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1714 {
1715 	struct pnfs_layout_segment *lseg = hdr->lseg;
1716 	struct nfs4_pnfs_ds *ds;
1717 	struct rpc_clnt *ds_clnt;
1718 	struct nfs4_ff_layout_mirror *mirror;
1719 	const struct cred *ds_cred;
1720 	loff_t offset = hdr->args.offset;
1721 	u32 idx = hdr->pgio_mirror_idx;
1722 	int vers;
1723 	struct nfs_fh *fh;
1724 
1725 	dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1726 		__func__, hdr->inode->i_ino,
1727 		hdr->args.pgbase, (size_t)hdr->args.count, offset);
1728 
1729 	mirror = FF_LAYOUT_COMP(lseg, idx);
1730 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1731 	if (!ds)
1732 		goto out_failed;
1733 
1734 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1735 						   hdr->inode);
1736 	if (IS_ERR(ds_clnt))
1737 		goto out_failed;
1738 
1739 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1740 	if (!ds_cred)
1741 		goto out_failed;
1742 
1743 	vers = nfs4_ff_layout_ds_version(mirror);
1744 
1745 	dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1746 		ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1747 
1748 	hdr->pgio_done_cb = ff_layout_read_done_cb;
1749 	refcount_inc(&ds->ds_clp->cl_count);
1750 	hdr->ds_clp = ds->ds_clp;
1751 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1752 	if (fh)
1753 		hdr->args.fh = fh;
1754 
1755 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1756 
1757 	/*
1758 	 * Note that if we ever decide to split across DSes,
1759 	 * then we may need to handle dense-like offsets.
1760 	 */
1761 	hdr->args.offset = offset;
1762 	hdr->mds_offset = offset;
1763 
1764 	/* Perform an asynchronous read to ds */
1765 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1766 			  vers == 3 ? &ff_layout_read_call_ops_v3 :
1767 				      &ff_layout_read_call_ops_v4,
1768 			  0, RPC_TASK_SOFTCONN);
1769 	put_cred(ds_cred);
1770 	return PNFS_ATTEMPTED;
1771 
1772 out_failed:
1773 	if (ff_layout_avoid_mds_available_ds(lseg))
1774 		return PNFS_TRY_AGAIN;
1775 	trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1776 			hdr->args.offset, hdr->args.count,
1777 			IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1778 	return PNFS_NOT_ATTEMPTED;
1779 }
1780 
1781 /* Perform async writes. */
1782 static enum pnfs_try_status
1783 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1784 {
1785 	struct pnfs_layout_segment *lseg = hdr->lseg;
1786 	struct nfs4_pnfs_ds *ds;
1787 	struct rpc_clnt *ds_clnt;
1788 	struct nfs4_ff_layout_mirror *mirror;
1789 	const struct cred *ds_cred;
1790 	loff_t offset = hdr->args.offset;
1791 	int vers;
1792 	struct nfs_fh *fh;
1793 	int idx = hdr->pgio_mirror_idx;
1794 
1795 	mirror = FF_LAYOUT_COMP(lseg, idx);
1796 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1797 	if (!ds)
1798 		goto out_failed;
1799 
1800 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1801 						   hdr->inode);
1802 	if (IS_ERR(ds_clnt))
1803 		goto out_failed;
1804 
1805 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1806 	if (!ds_cred)
1807 		goto out_failed;
1808 
1809 	vers = nfs4_ff_layout_ds_version(mirror);
1810 
1811 	dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1812 		__func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1813 		offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1814 		vers);
1815 
1816 	hdr->pgio_done_cb = ff_layout_write_done_cb;
1817 	refcount_inc(&ds->ds_clp->cl_count);
1818 	hdr->ds_clp = ds->ds_clp;
1819 	hdr->ds_commit_idx = idx;
1820 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1821 	if (fh)
1822 		hdr->args.fh = fh;
1823 
1824 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1825 
1826 	/*
1827 	 * Note that if we ever decide to split across DSes,
1828 	 * then we may need to handle dense-like offsets.
1829 	 */
1830 	hdr->args.offset = offset;
1831 
1832 	/* Perform an asynchronous write */
1833 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1834 			  vers == 3 ? &ff_layout_write_call_ops_v3 :
1835 				      &ff_layout_write_call_ops_v4,
1836 			  sync, RPC_TASK_SOFTCONN);
1837 	put_cred(ds_cred);
1838 	return PNFS_ATTEMPTED;
1839 
1840 out_failed:
1841 	if (ff_layout_avoid_mds_available_ds(lseg))
1842 		return PNFS_TRY_AGAIN;
1843 	trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1844 			hdr->args.offset, hdr->args.count,
1845 			IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1846 	return PNFS_NOT_ATTEMPTED;
1847 }
1848 
1849 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1850 {
1851 	return i;
1852 }
1853 
1854 static struct nfs_fh *
1855 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1856 {
1857 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1858 
1859 	/* FIXME: Assume that there is only one NFS version available
1860 	 * for the DS.
1861 	 */
1862 	return &flseg->mirror_array[i]->fh_versions[0];
1863 }
1864 
1865 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1866 {
1867 	struct pnfs_layout_segment *lseg = data->lseg;
1868 	struct nfs4_pnfs_ds *ds;
1869 	struct rpc_clnt *ds_clnt;
1870 	struct nfs4_ff_layout_mirror *mirror;
1871 	const struct cred *ds_cred;
1872 	u32 idx;
1873 	int vers, ret;
1874 	struct nfs_fh *fh;
1875 
1876 	if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1877 	    test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1878 		goto out_err;
1879 
1880 	idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1881 	mirror = FF_LAYOUT_COMP(lseg, idx);
1882 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1883 	if (!ds)
1884 		goto out_err;
1885 
1886 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1887 						   data->inode);
1888 	if (IS_ERR(ds_clnt))
1889 		goto out_err;
1890 
1891 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1892 	if (!ds_cred)
1893 		goto out_err;
1894 
1895 	vers = nfs4_ff_layout_ds_version(mirror);
1896 
1897 	dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1898 		data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1899 		vers);
1900 	data->commit_done_cb = ff_layout_commit_done_cb;
1901 	data->cred = ds_cred;
1902 	refcount_inc(&ds->ds_clp->cl_count);
1903 	data->ds_clp = ds->ds_clp;
1904 	fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1905 	if (fh)
1906 		data->args.fh = fh;
1907 
1908 	ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1909 				   vers == 3 ? &ff_layout_commit_call_ops_v3 :
1910 					       &ff_layout_commit_call_ops_v4,
1911 				   how, RPC_TASK_SOFTCONN);
1912 	put_cred(ds_cred);
1913 	return ret;
1914 out_err:
1915 	pnfs_generic_prepare_to_resend_writes(data);
1916 	pnfs_generic_commit_release(data);
1917 	return -EAGAIN;
1918 }
1919 
1920 static int
1921 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1922 			   int how, struct nfs_commit_info *cinfo)
1923 {
1924 	return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1925 					    ff_layout_initiate_commit);
1926 }
1927 
1928 static struct pnfs_ds_commit_info *
1929 ff_layout_get_ds_info(struct inode *inode)
1930 {
1931 	struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1932 
1933 	if (layout == NULL)
1934 		return NULL;
1935 
1936 	return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1937 }
1938 
1939 static void
1940 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1941 		struct pnfs_layout_segment *lseg)
1942 {
1943 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1944 	struct inode *inode = lseg->pls_layout->plh_inode;
1945 	struct pnfs_commit_array *array, *new;
1946 
1947 	new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1948 	if (new) {
1949 		spin_lock(&inode->i_lock);
1950 		array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1951 		spin_unlock(&inode->i_lock);
1952 		if (array != new)
1953 			pnfs_free_commit_array(new);
1954 	}
1955 }
1956 
1957 static void
1958 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1959 		struct inode *inode)
1960 {
1961 	spin_lock(&inode->i_lock);
1962 	pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1963 	spin_unlock(&inode->i_lock);
1964 }
1965 
1966 static void
1967 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1968 {
1969 	nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1970 						  id_node));
1971 }
1972 
1973 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
1974 				  const struct nfs4_layoutreturn_args *args,
1975 				  const struct nfs4_flexfile_layoutreturn_args *ff_args)
1976 {
1977 	__be32 *start;
1978 
1979 	start = xdr_reserve_space(xdr, 4);
1980 	if (unlikely(!start))
1981 		return -E2BIG;
1982 
1983 	*start = cpu_to_be32(ff_args->num_errors);
1984 	/* This assume we always return _ALL_ layouts */
1985 	return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
1986 }
1987 
1988 static void
1989 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
1990 {
1991 	WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
1992 }
1993 
1994 static void
1995 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
1996 			    const nfs4_stateid *stateid,
1997 			    const struct nfs42_layoutstat_devinfo *devinfo)
1998 {
1999 	__be32 *p;
2000 
2001 	p = xdr_reserve_space(xdr, 8 + 8);
2002 	p = xdr_encode_hyper(p, devinfo->offset);
2003 	p = xdr_encode_hyper(p, devinfo->length);
2004 	encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2005 	p = xdr_reserve_space(xdr, 4*8);
2006 	p = xdr_encode_hyper(p, devinfo->read_count);
2007 	p = xdr_encode_hyper(p, devinfo->read_bytes);
2008 	p = xdr_encode_hyper(p, devinfo->write_count);
2009 	p = xdr_encode_hyper(p, devinfo->write_bytes);
2010 	encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2011 }
2012 
2013 static void
2014 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2015 			    const nfs4_stateid *stateid,
2016 			    const struct nfs42_layoutstat_devinfo *devinfo)
2017 {
2018 	ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2019 	ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2020 			devinfo->ld_private.data);
2021 }
2022 
2023 /* report nothing for now */
2024 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2025 		const struct nfs4_layoutreturn_args *args,
2026 		struct nfs4_flexfile_layoutreturn_args *ff_args)
2027 {
2028 	__be32 *p;
2029 	int i;
2030 
2031 	p = xdr_reserve_space(xdr, 4);
2032 	*p = cpu_to_be32(ff_args->num_dev);
2033 	for (i = 0; i < ff_args->num_dev; i++)
2034 		ff_layout_encode_ff_iostat(xdr,
2035 				&args->layout->plh_stateid,
2036 				&ff_args->devinfo[i]);
2037 }
2038 
2039 static void
2040 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2041 		unsigned int num_entries)
2042 {
2043 	unsigned int i;
2044 
2045 	for (i = 0; i < num_entries; i++) {
2046 		if (!devinfo[i].ld_private.ops)
2047 			continue;
2048 		if (!devinfo[i].ld_private.ops->free)
2049 			continue;
2050 		devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2051 	}
2052 }
2053 
2054 static struct nfs4_deviceid_node *
2055 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2056 			      struct pnfs_device *pdev, gfp_t gfp_flags)
2057 {
2058 	struct nfs4_ff_layout_ds *dsaddr;
2059 
2060 	dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2061 	if (!dsaddr)
2062 		return NULL;
2063 	return &dsaddr->id_node;
2064 }
2065 
2066 static void
2067 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2068 		const void *voidargs,
2069 		const struct nfs4_xdr_opaque_data *ff_opaque)
2070 {
2071 	const struct nfs4_layoutreturn_args *args = voidargs;
2072 	struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2073 	struct xdr_buf tmp_buf = {
2074 		.head = {
2075 			[0] = {
2076 				.iov_base = page_address(ff_args->pages[0]),
2077 			},
2078 		},
2079 		.buflen = PAGE_SIZE,
2080 	};
2081 	struct xdr_stream tmp_xdr;
2082 	__be32 *start;
2083 
2084 	dprintk("%s: Begin\n", __func__);
2085 
2086 	xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2087 
2088 	ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2089 	ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2090 
2091 	start = xdr_reserve_space(xdr, 4);
2092 	*start = cpu_to_be32(tmp_buf.len);
2093 	xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2094 
2095 	dprintk("%s: Return\n", __func__);
2096 }
2097 
2098 static void
2099 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2100 {
2101 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2102 
2103 	if (!args->data)
2104 		return;
2105 	ff_args = args->data;
2106 	args->data = NULL;
2107 
2108 	ff_layout_free_ds_ioerr(&ff_args->errors);
2109 	ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2110 
2111 	put_page(ff_args->pages[0]);
2112 	kfree(ff_args);
2113 }
2114 
2115 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2116 	.encode = ff_layout_encode_layoutreturn,
2117 	.free = ff_layout_free_layoutreturn,
2118 };
2119 
2120 static int
2121 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2122 {
2123 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2124 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2125 
2126 	ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2127 	if (!ff_args)
2128 		goto out_nomem;
2129 	ff_args->pages[0] = alloc_page(GFP_KERNEL);
2130 	if (!ff_args->pages[0])
2131 		goto out_nomem_free;
2132 
2133 	INIT_LIST_HEAD(&ff_args->errors);
2134 	ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2135 			&args->range, &ff_args->errors,
2136 			FF_LAYOUTRETURN_MAXERR);
2137 
2138 	spin_lock(&args->inode->i_lock);
2139 	ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2140 			&ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2141 	spin_unlock(&args->inode->i_lock);
2142 
2143 	args->ld_private->ops = &layoutreturn_ops;
2144 	args->ld_private->data = ff_args;
2145 	return 0;
2146 out_nomem_free:
2147 	kfree(ff_args);
2148 out_nomem:
2149 	return -ENOMEM;
2150 }
2151 
2152 #ifdef CONFIG_NFS_V4_2
2153 void
2154 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2155 {
2156 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
2157 	struct nfs42_layout_error *errors;
2158 	LIST_HEAD(head);
2159 
2160 	if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2161 		return;
2162 	ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2163 	if (list_empty(&head))
2164 		return;
2165 
2166 	errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2167 			sizeof(*errors), GFP_NOFS);
2168 	if (errors != NULL) {
2169 		const struct nfs4_ff_layout_ds_err *pos;
2170 		size_t n = 0;
2171 
2172 		list_for_each_entry(pos, &head, list) {
2173 			errors[n].offset = pos->offset;
2174 			errors[n].length = pos->length;
2175 			nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2176 			errors[n].errors[0].dev_id = pos->deviceid;
2177 			errors[n].errors[0].status = pos->status;
2178 			errors[n].errors[0].opnum = pos->opnum;
2179 			n++;
2180 			if (!list_is_last(&pos->list, &head) &&
2181 			    n < NFS42_LAYOUTERROR_MAX)
2182 				continue;
2183 			if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2184 				break;
2185 			n = 0;
2186 		}
2187 		kfree(errors);
2188 	}
2189 	ff_layout_free_ds_ioerr(&head);
2190 }
2191 #else
2192 void
2193 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2194 {
2195 }
2196 #endif
2197 
2198 static int
2199 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2200 {
2201 	const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2202 
2203 	return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2204 }
2205 
2206 static size_t
2207 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2208 			  const int buflen)
2209 {
2210 	const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2211 	const struct in6_addr *addr = &sin6->sin6_addr;
2212 
2213 	/*
2214 	 * RFC 4291, Section 2.2.2
2215 	 *
2216 	 * Shorthanded ANY address
2217 	 */
2218 	if (ipv6_addr_any(addr))
2219 		return snprintf(buf, buflen, "::");
2220 
2221 	/*
2222 	 * RFC 4291, Section 2.2.2
2223 	 *
2224 	 * Shorthanded loopback address
2225 	 */
2226 	if (ipv6_addr_loopback(addr))
2227 		return snprintf(buf, buflen, "::1");
2228 
2229 	/*
2230 	 * RFC 4291, Section 2.2.3
2231 	 *
2232 	 * Special presentation address format for mapped v4
2233 	 * addresses.
2234 	 */
2235 	if (ipv6_addr_v4mapped(addr))
2236 		return snprintf(buf, buflen, "::ffff:%pI4",
2237 					&addr->s6_addr32[3]);
2238 
2239 	/*
2240 	 * RFC 4291, Section 2.2.1
2241 	 */
2242 	return snprintf(buf, buflen, "%pI6c", addr);
2243 }
2244 
2245 /* Derived from rpc_sockaddr2uaddr */
2246 static void
2247 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2248 {
2249 	struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2250 	char portbuf[RPCBIND_MAXUADDRPLEN];
2251 	char addrbuf[RPCBIND_MAXUADDRLEN];
2252 	char *netid;
2253 	unsigned short port;
2254 	int len, netid_len;
2255 	__be32 *p;
2256 
2257 	switch (sap->sa_family) {
2258 	case AF_INET:
2259 		if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2260 			return;
2261 		port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2262 		netid = "tcp";
2263 		netid_len = 3;
2264 		break;
2265 	case AF_INET6:
2266 		if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2267 			return;
2268 		port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2269 		netid = "tcp6";
2270 		netid_len = 4;
2271 		break;
2272 	default:
2273 		/* we only support tcp and tcp6 */
2274 		WARN_ON_ONCE(1);
2275 		return;
2276 	}
2277 
2278 	snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2279 	len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2280 
2281 	p = xdr_reserve_space(xdr, 4 + netid_len);
2282 	xdr_encode_opaque(p, netid, netid_len);
2283 
2284 	p = xdr_reserve_space(xdr, 4 + len);
2285 	xdr_encode_opaque(p, addrbuf, len);
2286 }
2287 
2288 static void
2289 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2290 			 ktime_t t)
2291 {
2292 	struct timespec64 ts;
2293 	__be32 *p;
2294 
2295 	p = xdr_reserve_space(xdr, 12);
2296 	ts = ktime_to_timespec64(t);
2297 	p = xdr_encode_hyper(p, ts.tv_sec);
2298 	*p++ = cpu_to_be32(ts.tv_nsec);
2299 }
2300 
2301 static void
2302 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2303 			    struct nfs4_ff_io_stat *stat)
2304 {
2305 	__be32 *p;
2306 
2307 	p = xdr_reserve_space(xdr, 5 * 8);
2308 	p = xdr_encode_hyper(p, stat->ops_requested);
2309 	p = xdr_encode_hyper(p, stat->bytes_requested);
2310 	p = xdr_encode_hyper(p, stat->ops_completed);
2311 	p = xdr_encode_hyper(p, stat->bytes_completed);
2312 	p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2313 	ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2314 	ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2315 }
2316 
2317 static void
2318 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2319 			      const struct nfs42_layoutstat_devinfo *devinfo,
2320 			      struct nfs4_ff_layout_mirror *mirror)
2321 {
2322 	struct nfs4_pnfs_ds_addr *da;
2323 	struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2324 	struct nfs_fh *fh = &mirror->fh_versions[0];
2325 	__be32 *p;
2326 
2327 	da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2328 	dprintk("%s: DS %s: encoding address %s\n",
2329 		__func__, ds->ds_remotestr, da->da_remotestr);
2330 	/* netaddr4 */
2331 	ff_layout_encode_netaddr(xdr, da);
2332 	/* nfs_fh4 */
2333 	p = xdr_reserve_space(xdr, 4 + fh->size);
2334 	xdr_encode_opaque(p, fh->data, fh->size);
2335 	/* ff_io_latency4 read */
2336 	spin_lock(&mirror->lock);
2337 	ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2338 	/* ff_io_latency4 write */
2339 	ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2340 	spin_unlock(&mirror->lock);
2341 	/* nfstime4 */
2342 	ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2343 	/* bool */
2344 	p = xdr_reserve_space(xdr, 4);
2345 	*p = cpu_to_be32(false);
2346 }
2347 
2348 static void
2349 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2350 			     const struct nfs4_xdr_opaque_data *opaque)
2351 {
2352 	struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2353 			struct nfs42_layoutstat_devinfo, ld_private);
2354 	__be32 *start;
2355 
2356 	/* layoutupdate length */
2357 	start = xdr_reserve_space(xdr, 4);
2358 	ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2359 
2360 	*start = cpu_to_be32((xdr->p - start - 1) * 4);
2361 }
2362 
2363 static void
2364 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2365 {
2366 	struct nfs4_ff_layout_mirror *mirror = opaque->data;
2367 
2368 	ff_layout_put_mirror(mirror);
2369 }
2370 
2371 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2372 	.encode = ff_layout_encode_layoutstats,
2373 	.free	= ff_layout_free_layoutstats,
2374 };
2375 
2376 static int
2377 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2378 			       struct nfs42_layoutstat_devinfo *devinfo,
2379 			       int dev_limit)
2380 {
2381 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2382 	struct nfs4_ff_layout_mirror *mirror;
2383 	struct nfs4_deviceid_node *dev;
2384 	int i = 0;
2385 
2386 	list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2387 		if (i >= dev_limit)
2388 			break;
2389 		if (IS_ERR_OR_NULL(mirror->mirror_ds))
2390 			continue;
2391 		if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2392 			continue;
2393 		/* mirror refcount put in cleanup_layoutstats */
2394 		if (!refcount_inc_not_zero(&mirror->ref))
2395 			continue;
2396 		dev = &mirror->mirror_ds->id_node;
2397 		memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2398 		devinfo->offset = 0;
2399 		devinfo->length = NFS4_MAX_UINT64;
2400 		spin_lock(&mirror->lock);
2401 		devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2402 		devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2403 		devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2404 		devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2405 		spin_unlock(&mirror->lock);
2406 		devinfo->layout_type = LAYOUT_FLEX_FILES;
2407 		devinfo->ld_private.ops = &layoutstat_ops;
2408 		devinfo->ld_private.data = mirror;
2409 
2410 		devinfo++;
2411 		i++;
2412 	}
2413 	return i;
2414 }
2415 
2416 static int
2417 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2418 {
2419 	struct nfs4_flexfile_layout *ff_layout;
2420 	const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2421 
2422 	/* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2423 	args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2424 	if (!args->devinfo)
2425 		return -ENOMEM;
2426 
2427 	spin_lock(&args->inode->i_lock);
2428 	ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2429 	args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2430 			&args->devinfo[0], dev_count);
2431 	spin_unlock(&args->inode->i_lock);
2432 	if (!args->num_dev) {
2433 		kfree(args->devinfo);
2434 		args->devinfo = NULL;
2435 		return -ENOENT;
2436 	}
2437 
2438 	return 0;
2439 }
2440 
2441 static int
2442 ff_layout_set_layoutdriver(struct nfs_server *server,
2443 		const struct nfs_fh *dummy)
2444 {
2445 #if IS_ENABLED(CONFIG_NFS_V4_2)
2446 	server->caps |= NFS_CAP_LAYOUTSTATS;
2447 #endif
2448 	return 0;
2449 }
2450 
2451 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2452 	.setup_ds_info		= ff_layout_setup_ds_info,
2453 	.release_ds_info	= ff_layout_release_ds_info,
2454 	.mark_request_commit	= pnfs_layout_mark_request_commit,
2455 	.clear_request_commit	= pnfs_generic_clear_request_commit,
2456 	.scan_commit_lists	= pnfs_generic_scan_commit_lists,
2457 	.recover_commit_reqs	= pnfs_generic_recover_commit_reqs,
2458 	.commit_pagelist	= ff_layout_commit_pagelist,
2459 };
2460 
2461 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2462 	.id			= LAYOUT_FLEX_FILES,
2463 	.name			= "LAYOUT_FLEX_FILES",
2464 	.owner			= THIS_MODULE,
2465 	.flags			= PNFS_LAYOUTGET_ON_OPEN,
2466 	.max_layoutget_response	= 4096, /* 1 page or so... */
2467 	.set_layoutdriver	= ff_layout_set_layoutdriver,
2468 	.alloc_layout_hdr	= ff_layout_alloc_layout_hdr,
2469 	.free_layout_hdr	= ff_layout_free_layout_hdr,
2470 	.alloc_lseg		= ff_layout_alloc_lseg,
2471 	.free_lseg		= ff_layout_free_lseg,
2472 	.add_lseg		= ff_layout_add_lseg,
2473 	.pg_read_ops		= &ff_layout_pg_read_ops,
2474 	.pg_write_ops		= &ff_layout_pg_write_ops,
2475 	.get_ds_info		= ff_layout_get_ds_info,
2476 	.free_deviceid_node	= ff_layout_free_deviceid_node,
2477 	.read_pagelist		= ff_layout_read_pagelist,
2478 	.write_pagelist		= ff_layout_write_pagelist,
2479 	.alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2480 	.prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2481 	.sync			= pnfs_nfs_generic_sync,
2482 	.prepare_layoutstats	= ff_layout_prepare_layoutstats,
2483 };
2484 
2485 static int __init nfs4flexfilelayout_init(void)
2486 {
2487 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2488 	       __func__);
2489 	return pnfs_register_layoutdriver(&flexfilelayout_type);
2490 }
2491 
2492 static void __exit nfs4flexfilelayout_exit(void)
2493 {
2494 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2495 	       __func__);
2496 	pnfs_unregister_layoutdriver(&flexfilelayout_type);
2497 }
2498 
2499 MODULE_ALIAS("nfs-layouttype4-4");
2500 
2501 MODULE_LICENSE("GPL");
2502 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2503 
2504 module_init(nfs4flexfilelayout_init);
2505 module_exit(nfs4flexfilelayout_exit);
2506 
2507 module_param(io_maxretrans, ushort, 0644);
2508 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2509 			"retries an I/O request before returning an error. ");
2510