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 enum nfs4_ff_op_type {
34 	NFS4_FF_OP_LAYOUTSTATS,
35 	NFS4_FF_OP_LAYOUTRETURN,
36 };
37 
38 static unsigned short io_maxretrans;
39 
40 static const struct pnfs_commit_ops ff_layout_commit_ops;
41 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
42 		struct nfs_pgio_header *hdr);
43 static int
44 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
45 			       struct nfs42_layoutstat_devinfo *devinfo,
46 			       int dev_limit, enum nfs4_ff_op_type type);
47 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
48 			      const struct nfs42_layoutstat_devinfo *devinfo,
49 			      struct nfs4_ff_layout_mirror *mirror);
50 
51 static struct pnfs_layout_hdr *
52 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
53 {
54 	struct nfs4_flexfile_layout *ffl;
55 
56 	ffl = kzalloc(sizeof(*ffl), gfp_flags);
57 	if (ffl) {
58 		pnfs_init_ds_commit_info(&ffl->commit_info);
59 		INIT_LIST_HEAD(&ffl->error_list);
60 		INIT_LIST_HEAD(&ffl->mirrors);
61 		ffl->last_report_time = ktime_get();
62 		ffl->commit_info.ops = &ff_layout_commit_ops;
63 		return &ffl->generic_hdr;
64 	} else
65 		return NULL;
66 }
67 
68 static void
69 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
70 {
71 	struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
72 	struct nfs4_ff_layout_ds_err *err, *n;
73 
74 	list_for_each_entry_safe(err, n, &ffl->error_list, list) {
75 		list_del(&err->list);
76 		kfree(err);
77 	}
78 	kfree_rcu(ffl, generic_hdr.plh_rcu);
79 }
80 
81 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
82 {
83 	__be32 *p;
84 
85 	p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
86 	if (unlikely(p == NULL))
87 		return -ENOBUFS;
88 	stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
89 	memcpy(stateid->data, p, NFS4_STATEID_SIZE);
90 	dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
91 		p[0], p[1], p[2], p[3]);
92 	return 0;
93 }
94 
95 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
96 {
97 	__be32 *p;
98 
99 	p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
100 	if (unlikely(!p))
101 		return -ENOBUFS;
102 	memcpy(devid, p, NFS4_DEVICEID4_SIZE);
103 	nfs4_print_deviceid(devid);
104 	return 0;
105 }
106 
107 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
108 {
109 	__be32 *p;
110 
111 	p = xdr_inline_decode(xdr, 4);
112 	if (unlikely(!p))
113 		return -ENOBUFS;
114 	fh->size = be32_to_cpup(p++);
115 	if (fh->size > NFS_MAXFHSIZE) {
116 		printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
117 		       fh->size);
118 		return -EOVERFLOW;
119 	}
120 	/* fh.data */
121 	p = xdr_inline_decode(xdr, fh->size);
122 	if (unlikely(!p))
123 		return -ENOBUFS;
124 	memcpy(&fh->data, p, fh->size);
125 	dprintk("%s: fh len %d\n", __func__, fh->size);
126 
127 	return 0;
128 }
129 
130 /*
131  * Currently only stringified uids and gids are accepted.
132  * I.e., kerberos is not supported to the DSes, so no pricipals.
133  *
134  * That means that one common function will suffice, but when
135  * principals are added, this should be split to accomodate
136  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
137  */
138 static int
139 decode_name(struct xdr_stream *xdr, u32 *id)
140 {
141 	__be32 *p;
142 	int len;
143 
144 	/* opaque_length(4)*/
145 	p = xdr_inline_decode(xdr, 4);
146 	if (unlikely(!p))
147 		return -ENOBUFS;
148 	len = be32_to_cpup(p++);
149 	if (len < 0)
150 		return -EINVAL;
151 
152 	dprintk("%s: len %u\n", __func__, len);
153 
154 	/* opaque body */
155 	p = xdr_inline_decode(xdr, len);
156 	if (unlikely(!p))
157 		return -ENOBUFS;
158 
159 	if (!nfs_map_string_to_numeric((char *)p, len, id))
160 		return -EINVAL;
161 
162 	return 0;
163 }
164 
165 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
166 		const struct nfs4_ff_layout_mirror *m2)
167 {
168 	int i, j;
169 
170 	if (m1->fh_versions_cnt != m2->fh_versions_cnt)
171 		return false;
172 	for (i = 0; i < m1->fh_versions_cnt; i++) {
173 		bool found_fh = false;
174 		for (j = 0; j < m2->fh_versions_cnt; j++) {
175 			if (nfs_compare_fh(&m1->fh_versions[i],
176 					&m2->fh_versions[j]) == 0) {
177 				found_fh = true;
178 				break;
179 			}
180 		}
181 		if (!found_fh)
182 			return false;
183 	}
184 	return true;
185 }
186 
187 static struct nfs4_ff_layout_mirror *
188 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
189 		struct nfs4_ff_layout_mirror *mirror)
190 {
191 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
192 	struct nfs4_ff_layout_mirror *pos;
193 	struct inode *inode = lo->plh_inode;
194 
195 	spin_lock(&inode->i_lock);
196 	list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
197 		if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
198 			continue;
199 		if (!ff_mirror_match_fh(mirror, pos))
200 			continue;
201 		if (refcount_inc_not_zero(&pos->ref)) {
202 			spin_unlock(&inode->i_lock);
203 			return pos;
204 		}
205 	}
206 	list_add(&mirror->mirrors, &ff_layout->mirrors);
207 	mirror->layout = lo;
208 	spin_unlock(&inode->i_lock);
209 	return mirror;
210 }
211 
212 static void
213 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
214 {
215 	struct inode *inode;
216 	if (mirror->layout == NULL)
217 		return;
218 	inode = mirror->layout->plh_inode;
219 	spin_lock(&inode->i_lock);
220 	list_del(&mirror->mirrors);
221 	spin_unlock(&inode->i_lock);
222 	mirror->layout = NULL;
223 }
224 
225 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
226 {
227 	struct nfs4_ff_layout_mirror *mirror;
228 
229 	mirror = kzalloc(sizeof(*mirror), gfp_flags);
230 	if (mirror != NULL) {
231 		spin_lock_init(&mirror->lock);
232 		refcount_set(&mirror->ref, 1);
233 		INIT_LIST_HEAD(&mirror->mirrors);
234 	}
235 	return mirror;
236 }
237 
238 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
239 {
240 	const struct cred	*cred;
241 
242 	ff_layout_remove_mirror(mirror);
243 	kfree(mirror->fh_versions);
244 	cred = rcu_access_pointer(mirror->ro_cred);
245 	put_cred(cred);
246 	cred = rcu_access_pointer(mirror->rw_cred);
247 	put_cred(cred);
248 	nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
249 	kfree(mirror);
250 }
251 
252 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
253 {
254 	if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
255 		ff_layout_free_mirror(mirror);
256 }
257 
258 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
259 {
260 	u32 i;
261 
262 	for (i = 0; i < fls->mirror_array_cnt; i++)
263 		ff_layout_put_mirror(fls->mirror_array[i]);
264 }
265 
266 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
267 {
268 	if (fls) {
269 		ff_layout_free_mirror_array(fls);
270 		kfree(fls);
271 	}
272 }
273 
274 static bool
275 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
276 		struct pnfs_layout_segment *l2)
277 {
278 	const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
279 	const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
280 	u32 i;
281 
282 	if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
283 		return false;
284 	for (i = 0; i < fl1->mirror_array_cnt; i++) {
285 		if (fl1->mirror_array[i] != fl2->mirror_array[i])
286 			return false;
287 	}
288 	return true;
289 }
290 
291 static bool
292 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
293 		const struct pnfs_layout_range *l2)
294 {
295 	u64 end1, end2;
296 
297 	if (l1->iomode != l2->iomode)
298 		return l1->iomode != IOMODE_READ;
299 	end1 = pnfs_calc_offset_end(l1->offset, l1->length);
300 	end2 = pnfs_calc_offset_end(l2->offset, l2->length);
301 	if (end1 < l2->offset)
302 		return false;
303 	if (end2 < l1->offset)
304 		return true;
305 	return l2->offset <= l1->offset;
306 }
307 
308 static bool
309 ff_lseg_merge(struct pnfs_layout_segment *new,
310 		struct pnfs_layout_segment *old)
311 {
312 	u64 new_end, old_end;
313 
314 	if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
315 		return false;
316 	if (new->pls_range.iomode != old->pls_range.iomode)
317 		return false;
318 	old_end = pnfs_calc_offset_end(old->pls_range.offset,
319 			old->pls_range.length);
320 	if (old_end < new->pls_range.offset)
321 		return false;
322 	new_end = pnfs_calc_offset_end(new->pls_range.offset,
323 			new->pls_range.length);
324 	if (new_end < old->pls_range.offset)
325 		return false;
326 	if (!ff_lseg_match_mirrors(new, old))
327 		return false;
328 
329 	/* Mergeable: copy info from 'old' to 'new' */
330 	if (new_end < old_end)
331 		new_end = old_end;
332 	if (new->pls_range.offset < old->pls_range.offset)
333 		new->pls_range.offset = old->pls_range.offset;
334 	new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
335 			new_end);
336 	if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
337 		set_bit(NFS_LSEG_ROC, &new->pls_flags);
338 	return true;
339 }
340 
341 static void
342 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
343 		struct pnfs_layout_segment *lseg,
344 		struct list_head *free_me)
345 {
346 	pnfs_generic_layout_insert_lseg(lo, lseg,
347 			ff_lseg_range_is_after,
348 			ff_lseg_merge,
349 			free_me);
350 }
351 
352 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
353 {
354 	int i, j;
355 
356 	for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
357 		for (j = i + 1; j < fls->mirror_array_cnt; j++)
358 			if (fls->mirror_array[i]->efficiency <
359 			    fls->mirror_array[j]->efficiency)
360 				swap(fls->mirror_array[i],
361 				     fls->mirror_array[j]);
362 	}
363 }
364 
365 static struct pnfs_layout_segment *
366 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
367 		     struct nfs4_layoutget_res *lgr,
368 		     gfp_t gfp_flags)
369 {
370 	struct pnfs_layout_segment *ret;
371 	struct nfs4_ff_layout_segment *fls = NULL;
372 	struct xdr_stream stream;
373 	struct xdr_buf buf;
374 	struct page *scratch;
375 	u64 stripe_unit;
376 	u32 mirror_array_cnt;
377 	__be32 *p;
378 	int i, rc;
379 
380 	dprintk("--> %s\n", __func__);
381 	scratch = alloc_page(gfp_flags);
382 	if (!scratch)
383 		return ERR_PTR(-ENOMEM);
384 
385 	xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
386 			      lgr->layoutp->len);
387 	xdr_set_scratch_page(&stream, scratch);
388 
389 	/* stripe unit and mirror_array_cnt */
390 	rc = -EIO;
391 	p = xdr_inline_decode(&stream, 8 + 4);
392 	if (!p)
393 		goto out_err_free;
394 
395 	p = xdr_decode_hyper(p, &stripe_unit);
396 	mirror_array_cnt = be32_to_cpup(p++);
397 	dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
398 		stripe_unit, mirror_array_cnt);
399 
400 	if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
401 	    mirror_array_cnt == 0)
402 		goto out_err_free;
403 
404 	rc = -ENOMEM;
405 	fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
406 			gfp_flags);
407 	if (!fls)
408 		goto out_err_free;
409 
410 	fls->mirror_array_cnt = mirror_array_cnt;
411 	fls->stripe_unit = stripe_unit;
412 
413 	for (i = 0; i < fls->mirror_array_cnt; i++) {
414 		struct nfs4_ff_layout_mirror *mirror;
415 		struct cred *kcred;
416 		const struct cred __rcu *cred;
417 		kuid_t uid;
418 		kgid_t gid;
419 		u32 ds_count, fh_count, id;
420 		int j;
421 
422 		rc = -EIO;
423 		p = xdr_inline_decode(&stream, 4);
424 		if (!p)
425 			goto out_err_free;
426 		ds_count = be32_to_cpup(p);
427 
428 		/* FIXME: allow for striping? */
429 		if (ds_count != 1)
430 			goto out_err_free;
431 
432 		fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
433 		if (fls->mirror_array[i] == NULL) {
434 			rc = -ENOMEM;
435 			goto out_err_free;
436 		}
437 
438 		fls->mirror_array[i]->ds_count = ds_count;
439 
440 		/* deviceid */
441 		rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
442 		if (rc)
443 			goto out_err_free;
444 
445 		/* efficiency */
446 		rc = -EIO;
447 		p = xdr_inline_decode(&stream, 4);
448 		if (!p)
449 			goto out_err_free;
450 		fls->mirror_array[i]->efficiency = be32_to_cpup(p);
451 
452 		/* stateid */
453 		rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
454 		if (rc)
455 			goto out_err_free;
456 
457 		/* fh */
458 		rc = -EIO;
459 		p = xdr_inline_decode(&stream, 4);
460 		if (!p)
461 			goto out_err_free;
462 		fh_count = be32_to_cpup(p);
463 
464 		fls->mirror_array[i]->fh_versions =
465 			kcalloc(fh_count, sizeof(struct nfs_fh),
466 				gfp_flags);
467 		if (fls->mirror_array[i]->fh_versions == NULL) {
468 			rc = -ENOMEM;
469 			goto out_err_free;
470 		}
471 
472 		for (j = 0; j < fh_count; j++) {
473 			rc = decode_nfs_fh(&stream,
474 					   &fls->mirror_array[i]->fh_versions[j]);
475 			if (rc)
476 				goto out_err_free;
477 		}
478 
479 		fls->mirror_array[i]->fh_versions_cnt = fh_count;
480 
481 		/* user */
482 		rc = decode_name(&stream, &id);
483 		if (rc)
484 			goto out_err_free;
485 
486 		uid = make_kuid(&init_user_ns, id);
487 
488 		/* group */
489 		rc = decode_name(&stream, &id);
490 		if (rc)
491 			goto out_err_free;
492 
493 		gid = make_kgid(&init_user_ns, id);
494 
495 		if (gfp_flags & __GFP_FS)
496 			kcred = prepare_kernel_cred(&init_task);
497 		else {
498 			unsigned int nofs_flags = memalloc_nofs_save();
499 			kcred = prepare_kernel_cred(&init_task);
500 			memalloc_nofs_restore(nofs_flags);
501 		}
502 		rc = -ENOMEM;
503 		if (!kcred)
504 			goto out_err_free;
505 		kcred->fsuid = uid;
506 		kcred->fsgid = gid;
507 		cred = RCU_INITIALIZER(kcred);
508 
509 		if (lgr->range.iomode == IOMODE_READ)
510 			rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
511 		else
512 			rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
513 
514 		mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
515 		if (mirror != fls->mirror_array[i]) {
516 			/* swap cred ptrs so free_mirror will clean up old */
517 			if (lgr->range.iomode == IOMODE_READ) {
518 				cred = xchg(&mirror->ro_cred, cred);
519 				rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
520 			} else {
521 				cred = xchg(&mirror->rw_cred, cred);
522 				rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
523 			}
524 			ff_layout_free_mirror(fls->mirror_array[i]);
525 			fls->mirror_array[i] = mirror;
526 		}
527 
528 		dprintk("%s: iomode %s uid %u gid %u\n", __func__,
529 			lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
530 			from_kuid(&init_user_ns, uid),
531 			from_kgid(&init_user_ns, gid));
532 	}
533 
534 	p = xdr_inline_decode(&stream, 4);
535 	if (!p)
536 		goto out_sort_mirrors;
537 	fls->flags = be32_to_cpup(p);
538 
539 	p = xdr_inline_decode(&stream, 4);
540 	if (!p)
541 		goto out_sort_mirrors;
542 	for (i=0; i < fls->mirror_array_cnt; i++)
543 		fls->mirror_array[i]->report_interval = be32_to_cpup(p);
544 
545 out_sort_mirrors:
546 	ff_layout_sort_mirrors(fls);
547 	ret = &fls->generic_hdr;
548 	dprintk("<-- %s (success)\n", __func__);
549 out_free_page:
550 	__free_page(scratch);
551 	return ret;
552 out_err_free:
553 	_ff_layout_free_lseg(fls);
554 	ret = ERR_PTR(rc);
555 	dprintk("<-- %s (%d)\n", __func__, rc);
556 	goto out_free_page;
557 }
558 
559 static void
560 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
561 {
562 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
563 
564 	dprintk("--> %s\n", __func__);
565 
566 	if (lseg->pls_range.iomode == IOMODE_RW) {
567 		struct nfs4_flexfile_layout *ffl;
568 		struct inode *inode;
569 
570 		ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
571 		inode = ffl->generic_hdr.plh_inode;
572 		spin_lock(&inode->i_lock);
573 		pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
574 		spin_unlock(&inode->i_lock);
575 	}
576 	_ff_layout_free_lseg(fls);
577 }
578 
579 static void
580 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
581 {
582 	/* first IO request? */
583 	if (atomic_inc_return(&timer->n_ops) == 1) {
584 		timer->start_time = now;
585 	}
586 }
587 
588 static ktime_t
589 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
590 {
591 	ktime_t start;
592 
593 	if (atomic_dec_return(&timer->n_ops) < 0)
594 		WARN_ON_ONCE(1);
595 
596 	start = timer->start_time;
597 	timer->start_time = now;
598 	return ktime_sub(now, start);
599 }
600 
601 static bool
602 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
603 			    struct nfs4_ff_layoutstat *layoutstat,
604 			    ktime_t now)
605 {
606 	s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
607 	struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
608 
609 	nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
610 	if (!mirror->start_time)
611 		mirror->start_time = now;
612 	if (mirror->report_interval != 0)
613 		report_interval = (s64)mirror->report_interval * 1000LL;
614 	else if (layoutstats_timer != 0)
615 		report_interval = (s64)layoutstats_timer * 1000LL;
616 	if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
617 			report_interval) {
618 		ffl->last_report_time = now;
619 		return true;
620 	}
621 
622 	return false;
623 }
624 
625 static void
626 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
627 		__u64 requested)
628 {
629 	struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
630 
631 	iostat->ops_requested++;
632 	iostat->bytes_requested += requested;
633 }
634 
635 static void
636 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
637 		__u64 requested,
638 		__u64 completed,
639 		ktime_t time_completed,
640 		ktime_t time_started)
641 {
642 	struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
643 	ktime_t completion_time = ktime_sub(time_completed, time_started);
644 	ktime_t timer;
645 
646 	iostat->ops_completed++;
647 	iostat->bytes_completed += completed;
648 	iostat->bytes_not_delivered += requested - completed;
649 
650 	timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
651 	iostat->total_busy_time =
652 			ktime_add(iostat->total_busy_time, timer);
653 	iostat->aggregate_completion_time =
654 			ktime_add(iostat->aggregate_completion_time,
655 					completion_time);
656 }
657 
658 static void
659 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
660 		struct nfs4_ff_layout_mirror *mirror,
661 		__u64 requested, ktime_t now)
662 {
663 	bool report;
664 
665 	spin_lock(&mirror->lock);
666 	report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
667 	nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
668 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
669 	spin_unlock(&mirror->lock);
670 
671 	if (report)
672 		pnfs_report_layoutstat(inode, nfs_io_gfp_mask());
673 }
674 
675 static void
676 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
677 		struct nfs4_ff_layout_mirror *mirror,
678 		__u64 requested,
679 		__u64 completed)
680 {
681 	spin_lock(&mirror->lock);
682 	nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
683 			requested, completed,
684 			ktime_get(), task->tk_start);
685 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
686 	spin_unlock(&mirror->lock);
687 }
688 
689 static void
690 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
691 		struct nfs4_ff_layout_mirror *mirror,
692 		__u64 requested, ktime_t now)
693 {
694 	bool report;
695 
696 	spin_lock(&mirror->lock);
697 	report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
698 	nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
699 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
700 	spin_unlock(&mirror->lock);
701 
702 	if (report)
703 		pnfs_report_layoutstat(inode, nfs_io_gfp_mask());
704 }
705 
706 static void
707 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
708 		struct nfs4_ff_layout_mirror *mirror,
709 		__u64 requested,
710 		__u64 completed,
711 		enum nfs3_stable_how committed)
712 {
713 	if (committed == NFS_UNSTABLE)
714 		requested = completed = 0;
715 
716 	spin_lock(&mirror->lock);
717 	nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
718 			requested, completed, ktime_get(), task->tk_start);
719 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
720 	spin_unlock(&mirror->lock);
721 }
722 
723 static void
724 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)
725 {
726 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
727 
728 	if (devid)
729 		nfs4_mark_deviceid_unavailable(devid);
730 }
731 
732 static void
733 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)
734 {
735 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
736 
737 	if (devid)
738 		nfs4_mark_deviceid_available(devid);
739 }
740 
741 static struct nfs4_pnfs_ds *
742 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
743 			     u32 start_idx, u32 *best_idx,
744 			     bool check_device)
745 {
746 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
747 	struct nfs4_ff_layout_mirror *mirror;
748 	struct nfs4_pnfs_ds *ds;
749 	u32 idx;
750 
751 	/* mirrors are initially sorted by efficiency */
752 	for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
753 		mirror = FF_LAYOUT_COMP(lseg, idx);
754 		ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
755 		if (!ds)
756 			continue;
757 
758 		if (check_device &&
759 		    nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
760 			continue;
761 
762 		*best_idx = idx;
763 		return ds;
764 	}
765 
766 	return NULL;
767 }
768 
769 static struct nfs4_pnfs_ds *
770 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
771 				 u32 start_idx, u32 *best_idx)
772 {
773 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
774 }
775 
776 static struct nfs4_pnfs_ds *
777 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
778 				   u32 start_idx, u32 *best_idx)
779 {
780 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
781 }
782 
783 static struct nfs4_pnfs_ds *
784 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
785 				  u32 start_idx, u32 *best_idx)
786 {
787 	struct nfs4_pnfs_ds *ds;
788 
789 	ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
790 	if (ds)
791 		return ds;
792 	return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
793 }
794 
795 static struct nfs4_pnfs_ds *
796 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio,
797 			  u32 *best_idx)
798 {
799 	struct pnfs_layout_segment *lseg = pgio->pg_lseg;
800 	struct nfs4_pnfs_ds *ds;
801 
802 	ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
803 					       best_idx);
804 	if (ds || !pgio->pg_mirror_idx)
805 		return ds;
806 	return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
807 }
808 
809 static void
810 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
811 		      struct nfs_page *req,
812 		      bool strict_iomode)
813 {
814 	pnfs_put_lseg(pgio->pg_lseg);
815 	pgio->pg_lseg =
816 		pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
817 				   req_offset(req), req->wb_bytes, IOMODE_READ,
818 				   strict_iomode, nfs_io_gfp_mask());
819 	if (IS_ERR(pgio->pg_lseg)) {
820 		pgio->pg_error = PTR_ERR(pgio->pg_lseg);
821 		pgio->pg_lseg = NULL;
822 	}
823 }
824 
825 static void
826 ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
827 			  struct nfs_page *req)
828 {
829 	pnfs_generic_pg_check_layout(pgio);
830 	pnfs_generic_pg_check_range(pgio, req);
831 }
832 
833 static void
834 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
835 			struct nfs_page *req)
836 {
837 	struct nfs_pgio_mirror *pgm;
838 	struct nfs4_ff_layout_mirror *mirror;
839 	struct nfs4_pnfs_ds *ds;
840 	u32 ds_idx;
841 
842 retry:
843 	ff_layout_pg_check_layout(pgio, req);
844 	/* Use full layout for now */
845 	if (!pgio->pg_lseg) {
846 		ff_layout_pg_get_read(pgio, req, false);
847 		if (!pgio->pg_lseg)
848 			goto out_nolseg;
849 	}
850 	if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
851 		ff_layout_pg_get_read(pgio, req, true);
852 		if (!pgio->pg_lseg)
853 			goto out_nolseg;
854 	}
855 
856 	ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
857 	if (!ds) {
858 		if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
859 			goto out_mds;
860 		pnfs_generic_pg_cleanup(pgio);
861 		/* Sleep for 1 second before retrying */
862 		ssleep(1);
863 		goto retry;
864 	}
865 
866 	mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
867 	pgm = &pgio->pg_mirrors[0];
868 	pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
869 
870 	pgio->pg_mirror_idx = ds_idx;
871 
872 	if (NFS_SERVER(pgio->pg_inode)->flags &
873 			(NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
874 		pgio->pg_maxretrans = io_maxretrans;
875 	return;
876 out_nolseg:
877 	if (pgio->pg_error < 0)
878 		return;
879 out_mds:
880 	trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
881 			0, NFS4_MAX_UINT64, IOMODE_READ,
882 			NFS_I(pgio->pg_inode)->layout,
883 			pgio->pg_lseg);
884 	pgio->pg_maxretrans = 0;
885 	nfs_pageio_reset_read_mds(pgio);
886 }
887 
888 static void
889 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
890 			struct nfs_page *req)
891 {
892 	struct nfs4_ff_layout_mirror *mirror;
893 	struct nfs_pgio_mirror *pgm;
894 	struct nfs4_pnfs_ds *ds;
895 	u32 i;
896 
897 retry:
898 	ff_layout_pg_check_layout(pgio, req);
899 	if (!pgio->pg_lseg) {
900 		pgio->pg_lseg =
901 			pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
902 					   req_offset(req), req->wb_bytes,
903 					   IOMODE_RW, false, nfs_io_gfp_mask());
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 =
957 			pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
958 					   req_offset(req), req->wb_bytes,
959 					   IOMODE_RW, false, nfs_io_gfp_mask());
960 		if (IS_ERR(pgio->pg_lseg)) {
961 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
962 			pgio->pg_lseg = NULL;
963 			goto out;
964 		}
965 	}
966 	if (pgio->pg_lseg)
967 		return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
968 
969 	trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
970 			0, NFS4_MAX_UINT64, IOMODE_RW,
971 			NFS_I(pgio->pg_inode)->layout,
972 			pgio->pg_lseg);
973 	/* no lseg means that pnfs is not in use, so no mirroring here */
974 	nfs_pageio_reset_write_mds(pgio);
975 out:
976 	return 1;
977 }
978 
979 static u32
980 ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
981 {
982 	u32 old = desc->pg_mirror_idx;
983 
984 	desc->pg_mirror_idx = idx;
985 	return old;
986 }
987 
988 static struct nfs_pgio_mirror *
989 ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
990 {
991 	return &desc->pg_mirrors[idx];
992 }
993 
994 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
995 	.pg_init = ff_layout_pg_init_read,
996 	.pg_test = pnfs_generic_pg_test,
997 	.pg_doio = pnfs_generic_pg_readpages,
998 	.pg_cleanup = pnfs_generic_pg_cleanup,
999 };
1000 
1001 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1002 	.pg_init = ff_layout_pg_init_write,
1003 	.pg_test = pnfs_generic_pg_test,
1004 	.pg_doio = pnfs_generic_pg_writepages,
1005 	.pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1006 	.pg_cleanup = pnfs_generic_pg_cleanup,
1007 	.pg_get_mirror = ff_layout_pg_get_mirror_write,
1008 	.pg_set_mirror = ff_layout_pg_set_mirror_write,
1009 };
1010 
1011 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1012 {
1013 	struct rpc_task *task = &hdr->task;
1014 
1015 	pnfs_layoutcommit_inode(hdr->inode, false);
1016 
1017 	if (retry_pnfs) {
1018 		dprintk("%s Reset task %5u for i/o through pNFS "
1019 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1020 			hdr->task.tk_pid,
1021 			hdr->inode->i_sb->s_id,
1022 			(unsigned long long)NFS_FILEID(hdr->inode),
1023 			hdr->args.count,
1024 			(unsigned long long)hdr->args.offset);
1025 
1026 		hdr->completion_ops->reschedule_io(hdr);
1027 		return;
1028 	}
1029 
1030 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1031 		dprintk("%s Reset task %5u for i/o through MDS "
1032 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1033 			hdr->task.tk_pid,
1034 			hdr->inode->i_sb->s_id,
1035 			(unsigned long long)NFS_FILEID(hdr->inode),
1036 			hdr->args.count,
1037 			(unsigned long long)hdr->args.offset);
1038 
1039 		trace_pnfs_mds_fallback_write_done(hdr->inode,
1040 				hdr->args.offset, hdr->args.count,
1041 				IOMODE_RW, NFS_I(hdr->inode)->layout,
1042 				hdr->lseg);
1043 		task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1044 	}
1045 }
1046 
1047 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1048 {
1049 	u32 idx = hdr->pgio_mirror_idx + 1;
1050 	u32 new_idx = 0;
1051 
1052 	if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx))
1053 		ff_layout_send_layouterror(hdr->lseg);
1054 	else
1055 		pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1056 	pnfs_read_resend_pnfs(hdr, new_idx);
1057 }
1058 
1059 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1060 {
1061 	struct rpc_task *task = &hdr->task;
1062 
1063 	pnfs_layoutcommit_inode(hdr->inode, false);
1064 	pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1065 
1066 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1067 		dprintk("%s Reset task %5u for i/o through MDS "
1068 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1069 			hdr->task.tk_pid,
1070 			hdr->inode->i_sb->s_id,
1071 			(unsigned long long)NFS_FILEID(hdr->inode),
1072 			hdr->args.count,
1073 			(unsigned long long)hdr->args.offset);
1074 
1075 		trace_pnfs_mds_fallback_read_done(hdr->inode,
1076 				hdr->args.offset, hdr->args.count,
1077 				IOMODE_READ, NFS_I(hdr->inode)->layout,
1078 				hdr->lseg);
1079 		task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1080 	}
1081 }
1082 
1083 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1084 					   struct nfs4_state *state,
1085 					   struct nfs_client *clp,
1086 					   struct pnfs_layout_segment *lseg,
1087 					   u32 idx)
1088 {
1089 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
1090 	struct inode *inode = lo->plh_inode;
1091 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1092 	struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1093 
1094 	switch (task->tk_status) {
1095 	case -NFS4ERR_BADSESSION:
1096 	case -NFS4ERR_BADSLOT:
1097 	case -NFS4ERR_BAD_HIGH_SLOT:
1098 	case -NFS4ERR_DEADSESSION:
1099 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1100 	case -NFS4ERR_SEQ_FALSE_RETRY:
1101 	case -NFS4ERR_SEQ_MISORDERED:
1102 		dprintk("%s ERROR %d, Reset session. Exchangeid "
1103 			"flags 0x%x\n", __func__, task->tk_status,
1104 			clp->cl_exchange_flags);
1105 		nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1106 		break;
1107 	case -NFS4ERR_DELAY:
1108 	case -NFS4ERR_GRACE:
1109 		rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1110 		break;
1111 	case -NFS4ERR_RETRY_UNCACHED_REP:
1112 		break;
1113 	/* Invalidate Layout errors */
1114 	case -NFS4ERR_PNFS_NO_LAYOUT:
1115 	case -ESTALE:           /* mapped NFS4ERR_STALE */
1116 	case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1117 	case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1118 	case -NFS4ERR_FHEXPIRED:
1119 	case -NFS4ERR_WRONG_TYPE:
1120 		dprintk("%s Invalid layout error %d\n", __func__,
1121 			task->tk_status);
1122 		/*
1123 		 * Destroy layout so new i/o will get a new layout.
1124 		 * Layout will not be destroyed until all current lseg
1125 		 * references are put. Mark layout as invalid to resend failed
1126 		 * i/o and all i/o waiting on the slot table to the MDS until
1127 		 * layout is destroyed and a new valid layout is obtained.
1128 		 */
1129 		pnfs_destroy_layout(NFS_I(inode));
1130 		rpc_wake_up(&tbl->slot_tbl_waitq);
1131 		goto reset;
1132 	/* RPC connection errors */
1133 	case -ECONNREFUSED:
1134 	case -EHOSTDOWN:
1135 	case -EHOSTUNREACH:
1136 	case -ENETUNREACH:
1137 	case -EIO:
1138 	case -ETIMEDOUT:
1139 	case -EPIPE:
1140 	case -EPROTO:
1141 	case -ENODEV:
1142 		dprintk("%s DS connection error %d\n", __func__,
1143 			task->tk_status);
1144 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1145 				&devid->deviceid);
1146 		rpc_wake_up(&tbl->slot_tbl_waitq);
1147 		fallthrough;
1148 	default:
1149 		if (ff_layout_avoid_mds_available_ds(lseg))
1150 			return -NFS4ERR_RESET_TO_PNFS;
1151 reset:
1152 		dprintk("%s Retry through MDS. Error %d\n", __func__,
1153 			task->tk_status);
1154 		return -NFS4ERR_RESET_TO_MDS;
1155 	}
1156 	task->tk_status = 0;
1157 	return -EAGAIN;
1158 }
1159 
1160 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1161 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1162 					   struct pnfs_layout_segment *lseg,
1163 					   u32 idx)
1164 {
1165 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1166 
1167 	switch (task->tk_status) {
1168 	/* File access problems. Don't mark the device as unavailable */
1169 	case -EACCES:
1170 	case -ESTALE:
1171 	case -EISDIR:
1172 	case -EBADHANDLE:
1173 	case -ELOOP:
1174 	case -ENOSPC:
1175 		break;
1176 	case -EJUKEBOX:
1177 		nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1178 		goto out_retry;
1179 	default:
1180 		dprintk("%s DS connection error %d\n", __func__,
1181 			task->tk_status);
1182 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1183 				&devid->deviceid);
1184 	}
1185 	/* FIXME: Need to prevent infinite looping here. */
1186 	return -NFS4ERR_RESET_TO_PNFS;
1187 out_retry:
1188 	task->tk_status = 0;
1189 	rpc_restart_call_prepare(task);
1190 	rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1191 	return -EAGAIN;
1192 }
1193 
1194 static int ff_layout_async_handle_error(struct rpc_task *task,
1195 					struct nfs4_state *state,
1196 					struct nfs_client *clp,
1197 					struct pnfs_layout_segment *lseg,
1198 					u32 idx)
1199 {
1200 	int vers = clp->cl_nfs_mod->rpc_vers->number;
1201 
1202 	if (task->tk_status >= 0) {
1203 		ff_layout_mark_ds_reachable(lseg, idx);
1204 		return 0;
1205 	}
1206 
1207 	/* Handle the case of an invalid layout segment */
1208 	if (!pnfs_is_valid_lseg(lseg))
1209 		return -NFS4ERR_RESET_TO_PNFS;
1210 
1211 	switch (vers) {
1212 	case 3:
1213 		return ff_layout_async_handle_error_v3(task, lseg, idx);
1214 	case 4:
1215 		return ff_layout_async_handle_error_v4(task, state, clp,
1216 						       lseg, idx);
1217 	default:
1218 		/* should never happen */
1219 		WARN_ON_ONCE(1);
1220 		return 0;
1221 	}
1222 }
1223 
1224 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1225 					u32 idx, u64 offset, u64 length,
1226 					u32 *op_status, int opnum, int error)
1227 {
1228 	struct nfs4_ff_layout_mirror *mirror;
1229 	u32 status = *op_status;
1230 	int err;
1231 
1232 	if (status == 0) {
1233 		switch (error) {
1234 		case -ETIMEDOUT:
1235 		case -EPFNOSUPPORT:
1236 		case -EPROTONOSUPPORT:
1237 		case -EOPNOTSUPP:
1238 		case -ECONNREFUSED:
1239 		case -ECONNRESET:
1240 		case -EHOSTDOWN:
1241 		case -EHOSTUNREACH:
1242 		case -ENETUNREACH:
1243 		case -EADDRINUSE:
1244 		case -ENOBUFS:
1245 		case -EPIPE:
1246 		case -EPERM:
1247 		case -EPROTO:
1248 		case -ENODEV:
1249 			*op_status = status = NFS4ERR_NXIO;
1250 			break;
1251 		case -EACCES:
1252 			*op_status = status = NFS4ERR_ACCESS;
1253 			break;
1254 		default:
1255 			return;
1256 		}
1257 	}
1258 
1259 	mirror = FF_LAYOUT_COMP(lseg, idx);
1260 	err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1261 				       mirror, offset, length, status, opnum,
1262 				       nfs_io_gfp_mask());
1263 
1264 	switch (status) {
1265 	case NFS4ERR_DELAY:
1266 	case NFS4ERR_GRACE:
1267 		break;
1268 	case NFS4ERR_NXIO:
1269 		ff_layout_mark_ds_unreachable(lseg, idx);
1270 		/*
1271 		 * Don't return the layout if this is a read and we still
1272 		 * have layouts to try
1273 		 */
1274 		if (opnum == OP_READ)
1275 			break;
1276 		fallthrough;
1277 	default:
1278 		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1279 						  lseg);
1280 	}
1281 
1282 	dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1283 }
1284 
1285 /* NFS_PROTO call done callback routines */
1286 static int ff_layout_read_done_cb(struct rpc_task *task,
1287 				struct nfs_pgio_header *hdr)
1288 {
1289 	int err;
1290 
1291 	if (task->tk_status < 0) {
1292 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1293 					    hdr->args.offset, hdr->args.count,
1294 					    &hdr->res.op_status, OP_READ,
1295 					    task->tk_status);
1296 		trace_ff_layout_read_error(hdr);
1297 	}
1298 
1299 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1300 					   hdr->ds_clp, hdr->lseg,
1301 					   hdr->pgio_mirror_idx);
1302 
1303 	trace_nfs4_pnfs_read(hdr, err);
1304 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1305 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1306 	switch (err) {
1307 	case -NFS4ERR_RESET_TO_PNFS:
1308 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1309 		return task->tk_status;
1310 	case -NFS4ERR_RESET_TO_MDS:
1311 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1312 		return task->tk_status;
1313 	case -EAGAIN:
1314 		goto out_eagain;
1315 	}
1316 
1317 	return 0;
1318 out_eagain:
1319 	rpc_restart_call_prepare(task);
1320 	return -EAGAIN;
1321 }
1322 
1323 static bool
1324 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1325 {
1326 	return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1327 }
1328 
1329 /*
1330  * We reference the rpc_cred of the first WRITE that triggers the need for
1331  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1332  * rfc5661 is not clear about which credential should be used.
1333  *
1334  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1335  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1336  * we always send layoutcommit after DS writes.
1337  */
1338 static void
1339 ff_layout_set_layoutcommit(struct inode *inode,
1340 		struct pnfs_layout_segment *lseg,
1341 		loff_t end_offset)
1342 {
1343 	if (!ff_layout_need_layoutcommit(lseg))
1344 		return;
1345 
1346 	pnfs_set_layoutcommit(inode, lseg, end_offset);
1347 	dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1348 		(unsigned long long) NFS_I(inode)->layout->plh_lwb);
1349 }
1350 
1351 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1352 		struct nfs_pgio_header *hdr)
1353 {
1354 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1355 		return;
1356 	nfs4_ff_layout_stat_io_start_read(hdr->inode,
1357 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1358 			hdr->args.count,
1359 			task->tk_start);
1360 }
1361 
1362 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1363 		struct nfs_pgio_header *hdr)
1364 {
1365 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1366 		return;
1367 	nfs4_ff_layout_stat_io_end_read(task,
1368 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1369 			hdr->args.count,
1370 			hdr->res.count);
1371 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1372 }
1373 
1374 static int ff_layout_read_prepare_common(struct rpc_task *task,
1375 					 struct nfs_pgio_header *hdr)
1376 {
1377 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1378 		rpc_exit(task, -EIO);
1379 		return -EIO;
1380 	}
1381 
1382 	if (!pnfs_is_valid_lseg(hdr->lseg)) {
1383 		rpc_exit(task, -EAGAIN);
1384 		return -EAGAIN;
1385 	}
1386 
1387 	ff_layout_read_record_layoutstats_start(task, hdr);
1388 	return 0;
1389 }
1390 
1391 /*
1392  * Call ops for the async read/write cases
1393  * In the case of dense layouts, the offset needs to be reset to its
1394  * original value.
1395  */
1396 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1397 {
1398 	struct nfs_pgio_header *hdr = data;
1399 
1400 	if (ff_layout_read_prepare_common(task, hdr))
1401 		return;
1402 
1403 	rpc_call_start(task);
1404 }
1405 
1406 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1407 {
1408 	struct nfs_pgio_header *hdr = data;
1409 
1410 	if (nfs4_setup_sequence(hdr->ds_clp,
1411 				&hdr->args.seq_args,
1412 				&hdr->res.seq_res,
1413 				task))
1414 		return;
1415 
1416 	ff_layout_read_prepare_common(task, hdr);
1417 }
1418 
1419 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1420 {
1421 	struct nfs_pgio_header *hdr = data;
1422 
1423 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1424 	    task->tk_status == 0) {
1425 		nfs4_sequence_done(task, &hdr->res.seq_res);
1426 		return;
1427 	}
1428 
1429 	/* Note this may cause RPC to be resent */
1430 	hdr->mds_ops->rpc_call_done(task, hdr);
1431 }
1432 
1433 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1434 {
1435 	struct nfs_pgio_header *hdr = data;
1436 
1437 	ff_layout_read_record_layoutstats_done(task, hdr);
1438 	rpc_count_iostats_metrics(task,
1439 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1440 }
1441 
1442 static void ff_layout_read_release(void *data)
1443 {
1444 	struct nfs_pgio_header *hdr = data;
1445 
1446 	ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1447 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1448 		ff_layout_resend_pnfs_read(hdr);
1449 	else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1450 		ff_layout_reset_read(hdr);
1451 	pnfs_generic_rw_release(data);
1452 }
1453 
1454 
1455 static int ff_layout_write_done_cb(struct rpc_task *task,
1456 				struct nfs_pgio_header *hdr)
1457 {
1458 	loff_t end_offs = 0;
1459 	int err;
1460 
1461 	if (task->tk_status < 0) {
1462 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1463 					    hdr->args.offset, hdr->args.count,
1464 					    &hdr->res.op_status, OP_WRITE,
1465 					    task->tk_status);
1466 		trace_ff_layout_write_error(hdr);
1467 	}
1468 
1469 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1470 					   hdr->ds_clp, hdr->lseg,
1471 					   hdr->pgio_mirror_idx);
1472 
1473 	trace_nfs4_pnfs_write(hdr, err);
1474 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1475 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1476 	switch (err) {
1477 	case -NFS4ERR_RESET_TO_PNFS:
1478 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1479 		return task->tk_status;
1480 	case -NFS4ERR_RESET_TO_MDS:
1481 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1482 		return task->tk_status;
1483 	case -EAGAIN:
1484 		return -EAGAIN;
1485 	}
1486 
1487 	if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1488 	    hdr->res.verf->committed == NFS_DATA_SYNC)
1489 		end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1490 
1491 	/* Note: if the write is unstable, don't set end_offs until commit */
1492 	ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1493 
1494 	/* zero out fattr since we don't care DS attr at all */
1495 	hdr->fattr.valid = 0;
1496 	if (task->tk_status >= 0)
1497 		nfs_writeback_update_inode(hdr);
1498 
1499 	return 0;
1500 }
1501 
1502 static int ff_layout_commit_done_cb(struct rpc_task *task,
1503 				     struct nfs_commit_data *data)
1504 {
1505 	int err;
1506 
1507 	if (task->tk_status < 0) {
1508 		ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1509 					    data->args.offset, data->args.count,
1510 					    &data->res.op_status, OP_COMMIT,
1511 					    task->tk_status);
1512 		trace_ff_layout_commit_error(data);
1513 	}
1514 
1515 	err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1516 					   data->lseg, data->ds_commit_index);
1517 
1518 	trace_nfs4_pnfs_commit_ds(data, err);
1519 	switch (err) {
1520 	case -NFS4ERR_RESET_TO_PNFS:
1521 		pnfs_generic_prepare_to_resend_writes(data);
1522 		return -EAGAIN;
1523 	case -NFS4ERR_RESET_TO_MDS:
1524 		pnfs_generic_prepare_to_resend_writes(data);
1525 		return -EAGAIN;
1526 	case -EAGAIN:
1527 		rpc_restart_call_prepare(task);
1528 		return -EAGAIN;
1529 	}
1530 
1531 	ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1532 
1533 	return 0;
1534 }
1535 
1536 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1537 		struct nfs_pgio_header *hdr)
1538 {
1539 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1540 		return;
1541 	nfs4_ff_layout_stat_io_start_write(hdr->inode,
1542 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1543 			hdr->args.count,
1544 			task->tk_start);
1545 }
1546 
1547 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1548 		struct nfs_pgio_header *hdr)
1549 {
1550 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1551 		return;
1552 	nfs4_ff_layout_stat_io_end_write(task,
1553 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1554 			hdr->args.count, hdr->res.count,
1555 			hdr->res.verf->committed);
1556 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1557 }
1558 
1559 static int ff_layout_write_prepare_common(struct rpc_task *task,
1560 					  struct nfs_pgio_header *hdr)
1561 {
1562 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1563 		rpc_exit(task, -EIO);
1564 		return -EIO;
1565 	}
1566 
1567 	if (!pnfs_is_valid_lseg(hdr->lseg)) {
1568 		rpc_exit(task, -EAGAIN);
1569 		return -EAGAIN;
1570 	}
1571 
1572 	ff_layout_write_record_layoutstats_start(task, hdr);
1573 	return 0;
1574 }
1575 
1576 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1577 {
1578 	struct nfs_pgio_header *hdr = data;
1579 
1580 	if (ff_layout_write_prepare_common(task, hdr))
1581 		return;
1582 
1583 	rpc_call_start(task);
1584 }
1585 
1586 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1587 {
1588 	struct nfs_pgio_header *hdr = data;
1589 
1590 	if (nfs4_setup_sequence(hdr->ds_clp,
1591 				&hdr->args.seq_args,
1592 				&hdr->res.seq_res,
1593 				task))
1594 		return;
1595 
1596 	ff_layout_write_prepare_common(task, hdr);
1597 }
1598 
1599 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1600 {
1601 	struct nfs_pgio_header *hdr = data;
1602 
1603 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1604 	    task->tk_status == 0) {
1605 		nfs4_sequence_done(task, &hdr->res.seq_res);
1606 		return;
1607 	}
1608 
1609 	/* Note this may cause RPC to be resent */
1610 	hdr->mds_ops->rpc_call_done(task, hdr);
1611 }
1612 
1613 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1614 {
1615 	struct nfs_pgio_header *hdr = data;
1616 
1617 	ff_layout_write_record_layoutstats_done(task, hdr);
1618 	rpc_count_iostats_metrics(task,
1619 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1620 }
1621 
1622 static void ff_layout_write_release(void *data)
1623 {
1624 	struct nfs_pgio_header *hdr = data;
1625 
1626 	ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1627 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1628 		ff_layout_send_layouterror(hdr->lseg);
1629 		ff_layout_reset_write(hdr, true);
1630 	} else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1631 		ff_layout_reset_write(hdr, false);
1632 	pnfs_generic_rw_release(data);
1633 }
1634 
1635 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1636 		struct nfs_commit_data *cdata)
1637 {
1638 	if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1639 		return;
1640 	nfs4_ff_layout_stat_io_start_write(cdata->inode,
1641 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1642 			0, task->tk_start);
1643 }
1644 
1645 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1646 		struct nfs_commit_data *cdata)
1647 {
1648 	struct nfs_page *req;
1649 	__u64 count = 0;
1650 
1651 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1652 		return;
1653 
1654 	if (task->tk_status == 0) {
1655 		list_for_each_entry(req, &cdata->pages, wb_list)
1656 			count += req->wb_bytes;
1657 	}
1658 	nfs4_ff_layout_stat_io_end_write(task,
1659 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1660 			count, count, NFS_FILE_SYNC);
1661 	set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1662 }
1663 
1664 static int ff_layout_commit_prepare_common(struct rpc_task *task,
1665 					   struct nfs_commit_data *cdata)
1666 {
1667 	if (!pnfs_is_valid_lseg(cdata->lseg)) {
1668 		rpc_exit(task, -EAGAIN);
1669 		return -EAGAIN;
1670 	}
1671 
1672 	ff_layout_commit_record_layoutstats_start(task, cdata);
1673 	return 0;
1674 }
1675 
1676 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1677 {
1678 	if (ff_layout_commit_prepare_common(task, data))
1679 		return;
1680 
1681 	rpc_call_start(task);
1682 }
1683 
1684 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1685 {
1686 	struct nfs_commit_data *wdata = data;
1687 
1688 	if (nfs4_setup_sequence(wdata->ds_clp,
1689 				&wdata->args.seq_args,
1690 				&wdata->res.seq_res,
1691 				task))
1692 		return;
1693 	ff_layout_commit_prepare_common(task, data);
1694 }
1695 
1696 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1697 {
1698 	pnfs_generic_write_commit_done(task, data);
1699 }
1700 
1701 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1702 {
1703 	struct nfs_commit_data *cdata = data;
1704 
1705 	ff_layout_commit_record_layoutstats_done(task, cdata);
1706 	rpc_count_iostats_metrics(task,
1707 	    &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1708 }
1709 
1710 static void ff_layout_commit_release(void *data)
1711 {
1712 	struct nfs_commit_data *cdata = data;
1713 
1714 	ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1715 	pnfs_generic_commit_release(data);
1716 }
1717 
1718 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1719 	.rpc_call_prepare = ff_layout_read_prepare_v3,
1720 	.rpc_call_done = ff_layout_read_call_done,
1721 	.rpc_count_stats = ff_layout_read_count_stats,
1722 	.rpc_release = ff_layout_read_release,
1723 };
1724 
1725 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1726 	.rpc_call_prepare = ff_layout_read_prepare_v4,
1727 	.rpc_call_done = ff_layout_read_call_done,
1728 	.rpc_count_stats = ff_layout_read_count_stats,
1729 	.rpc_release = ff_layout_read_release,
1730 };
1731 
1732 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1733 	.rpc_call_prepare = ff_layout_write_prepare_v3,
1734 	.rpc_call_done = ff_layout_write_call_done,
1735 	.rpc_count_stats = ff_layout_write_count_stats,
1736 	.rpc_release = ff_layout_write_release,
1737 };
1738 
1739 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1740 	.rpc_call_prepare = ff_layout_write_prepare_v4,
1741 	.rpc_call_done = ff_layout_write_call_done,
1742 	.rpc_count_stats = ff_layout_write_count_stats,
1743 	.rpc_release = ff_layout_write_release,
1744 };
1745 
1746 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1747 	.rpc_call_prepare = ff_layout_commit_prepare_v3,
1748 	.rpc_call_done = ff_layout_commit_done,
1749 	.rpc_count_stats = ff_layout_commit_count_stats,
1750 	.rpc_release = ff_layout_commit_release,
1751 };
1752 
1753 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1754 	.rpc_call_prepare = ff_layout_commit_prepare_v4,
1755 	.rpc_call_done = ff_layout_commit_done,
1756 	.rpc_count_stats = ff_layout_commit_count_stats,
1757 	.rpc_release = ff_layout_commit_release,
1758 };
1759 
1760 static enum pnfs_try_status
1761 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1762 {
1763 	struct pnfs_layout_segment *lseg = hdr->lseg;
1764 	struct nfs4_pnfs_ds *ds;
1765 	struct rpc_clnt *ds_clnt;
1766 	struct nfs4_ff_layout_mirror *mirror;
1767 	const struct cred *ds_cred;
1768 	loff_t offset = hdr->args.offset;
1769 	u32 idx = hdr->pgio_mirror_idx;
1770 	int vers;
1771 	struct nfs_fh *fh;
1772 
1773 	dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1774 		__func__, hdr->inode->i_ino,
1775 		hdr->args.pgbase, (size_t)hdr->args.count, offset);
1776 
1777 	mirror = FF_LAYOUT_COMP(lseg, idx);
1778 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1779 	if (!ds)
1780 		goto out_failed;
1781 
1782 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1783 						   hdr->inode);
1784 	if (IS_ERR(ds_clnt))
1785 		goto out_failed;
1786 
1787 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1788 	if (!ds_cred)
1789 		goto out_failed;
1790 
1791 	vers = nfs4_ff_layout_ds_version(mirror);
1792 
1793 	dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1794 		ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1795 
1796 	hdr->pgio_done_cb = ff_layout_read_done_cb;
1797 	refcount_inc(&ds->ds_clp->cl_count);
1798 	hdr->ds_clp = ds->ds_clp;
1799 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1800 	if (fh)
1801 		hdr->args.fh = fh;
1802 
1803 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1804 
1805 	/*
1806 	 * Note that if we ever decide to split across DSes,
1807 	 * then we may need to handle dense-like offsets.
1808 	 */
1809 	hdr->args.offset = offset;
1810 	hdr->mds_offset = offset;
1811 
1812 	/* Perform an asynchronous read to ds */
1813 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1814 			  vers == 3 ? &ff_layout_read_call_ops_v3 :
1815 				      &ff_layout_read_call_ops_v4,
1816 			  0, RPC_TASK_SOFTCONN);
1817 	put_cred(ds_cred);
1818 	return PNFS_ATTEMPTED;
1819 
1820 out_failed:
1821 	if (ff_layout_avoid_mds_available_ds(lseg))
1822 		return PNFS_TRY_AGAIN;
1823 	trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1824 			hdr->args.offset, hdr->args.count,
1825 			IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1826 	return PNFS_NOT_ATTEMPTED;
1827 }
1828 
1829 /* Perform async writes. */
1830 static enum pnfs_try_status
1831 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1832 {
1833 	struct pnfs_layout_segment *lseg = hdr->lseg;
1834 	struct nfs4_pnfs_ds *ds;
1835 	struct rpc_clnt *ds_clnt;
1836 	struct nfs4_ff_layout_mirror *mirror;
1837 	const struct cred *ds_cred;
1838 	loff_t offset = hdr->args.offset;
1839 	int vers;
1840 	struct nfs_fh *fh;
1841 	u32 idx = hdr->pgio_mirror_idx;
1842 
1843 	mirror = FF_LAYOUT_COMP(lseg, idx);
1844 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1845 	if (!ds)
1846 		goto out_failed;
1847 
1848 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1849 						   hdr->inode);
1850 	if (IS_ERR(ds_clnt))
1851 		goto out_failed;
1852 
1853 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1854 	if (!ds_cred)
1855 		goto out_failed;
1856 
1857 	vers = nfs4_ff_layout_ds_version(mirror);
1858 
1859 	dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1860 		__func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1861 		offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1862 		vers);
1863 
1864 	hdr->pgio_done_cb = ff_layout_write_done_cb;
1865 	refcount_inc(&ds->ds_clp->cl_count);
1866 	hdr->ds_clp = ds->ds_clp;
1867 	hdr->ds_commit_idx = idx;
1868 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1869 	if (fh)
1870 		hdr->args.fh = fh;
1871 
1872 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1873 
1874 	/*
1875 	 * Note that if we ever decide to split across DSes,
1876 	 * then we may need to handle dense-like offsets.
1877 	 */
1878 	hdr->args.offset = offset;
1879 
1880 	/* Perform an asynchronous write */
1881 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1882 			  vers == 3 ? &ff_layout_write_call_ops_v3 :
1883 				      &ff_layout_write_call_ops_v4,
1884 			  sync, RPC_TASK_SOFTCONN);
1885 	put_cred(ds_cred);
1886 	return PNFS_ATTEMPTED;
1887 
1888 out_failed:
1889 	if (ff_layout_avoid_mds_available_ds(lseg))
1890 		return PNFS_TRY_AGAIN;
1891 	trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1892 			hdr->args.offset, hdr->args.count,
1893 			IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1894 	return PNFS_NOT_ATTEMPTED;
1895 }
1896 
1897 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1898 {
1899 	return i;
1900 }
1901 
1902 static struct nfs_fh *
1903 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1904 {
1905 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1906 
1907 	/* FIXME: Assume that there is only one NFS version available
1908 	 * for the DS.
1909 	 */
1910 	return &flseg->mirror_array[i]->fh_versions[0];
1911 }
1912 
1913 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1914 {
1915 	struct pnfs_layout_segment *lseg = data->lseg;
1916 	struct nfs4_pnfs_ds *ds;
1917 	struct rpc_clnt *ds_clnt;
1918 	struct nfs4_ff_layout_mirror *mirror;
1919 	const struct cred *ds_cred;
1920 	u32 idx;
1921 	int vers, ret;
1922 	struct nfs_fh *fh;
1923 
1924 	if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1925 	    test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1926 		goto out_err;
1927 
1928 	idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1929 	mirror = FF_LAYOUT_COMP(lseg, idx);
1930 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1931 	if (!ds)
1932 		goto out_err;
1933 
1934 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1935 						   data->inode);
1936 	if (IS_ERR(ds_clnt))
1937 		goto out_err;
1938 
1939 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1940 	if (!ds_cred)
1941 		goto out_err;
1942 
1943 	vers = nfs4_ff_layout_ds_version(mirror);
1944 
1945 	dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1946 		data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1947 		vers);
1948 	data->commit_done_cb = ff_layout_commit_done_cb;
1949 	data->cred = ds_cred;
1950 	refcount_inc(&ds->ds_clp->cl_count);
1951 	data->ds_clp = ds->ds_clp;
1952 	fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1953 	if (fh)
1954 		data->args.fh = fh;
1955 
1956 	ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1957 				   vers == 3 ? &ff_layout_commit_call_ops_v3 :
1958 					       &ff_layout_commit_call_ops_v4,
1959 				   how, RPC_TASK_SOFTCONN);
1960 	put_cred(ds_cred);
1961 	return ret;
1962 out_err:
1963 	pnfs_generic_prepare_to_resend_writes(data);
1964 	pnfs_generic_commit_release(data);
1965 	return -EAGAIN;
1966 }
1967 
1968 static int
1969 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1970 			   int how, struct nfs_commit_info *cinfo)
1971 {
1972 	return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1973 					    ff_layout_initiate_commit);
1974 }
1975 
1976 static bool ff_layout_match_rw(const struct rpc_task *task,
1977 			       const struct nfs_pgio_header *hdr,
1978 			       const struct pnfs_layout_segment *lseg)
1979 {
1980 	return hdr->lseg == lseg;
1981 }
1982 
1983 static bool ff_layout_match_commit(const struct rpc_task *task,
1984 				   const struct nfs_commit_data *cdata,
1985 				   const struct pnfs_layout_segment *lseg)
1986 {
1987 	return cdata->lseg == lseg;
1988 }
1989 
1990 static bool ff_layout_match_io(const struct rpc_task *task, const void *data)
1991 {
1992 	const struct rpc_call_ops *ops = task->tk_ops;
1993 
1994 	if (ops == &ff_layout_read_call_ops_v3 ||
1995 	    ops == &ff_layout_read_call_ops_v4 ||
1996 	    ops == &ff_layout_write_call_ops_v3 ||
1997 	    ops == &ff_layout_write_call_ops_v4)
1998 		return ff_layout_match_rw(task, task->tk_calldata, data);
1999 	if (ops == &ff_layout_commit_call_ops_v3 ||
2000 	    ops == &ff_layout_commit_call_ops_v4)
2001 		return ff_layout_match_commit(task, task->tk_calldata, data);
2002 	return false;
2003 }
2004 
2005 static void ff_layout_cancel_io(struct pnfs_layout_segment *lseg)
2006 {
2007 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
2008 	struct nfs4_ff_layout_mirror *mirror;
2009 	struct nfs4_ff_layout_ds *mirror_ds;
2010 	struct nfs4_pnfs_ds *ds;
2011 	struct nfs_client *ds_clp;
2012 	struct rpc_clnt *clnt;
2013 	u32 idx;
2014 
2015 	for (idx = 0; idx < flseg->mirror_array_cnt; idx++) {
2016 		mirror = flseg->mirror_array[idx];
2017 		mirror_ds = mirror->mirror_ds;
2018 		if (!mirror_ds)
2019 			continue;
2020 		ds = mirror->mirror_ds->ds;
2021 		if (!ds)
2022 			continue;
2023 		ds_clp = ds->ds_clp;
2024 		if (!ds_clp)
2025 			continue;
2026 		clnt = ds_clp->cl_rpcclient;
2027 		if (!clnt)
2028 			continue;
2029 		if (!rpc_cancel_tasks(clnt, -EAGAIN, ff_layout_match_io, lseg))
2030 			continue;
2031 		rpc_clnt_disconnect(clnt);
2032 	}
2033 }
2034 
2035 static struct pnfs_ds_commit_info *
2036 ff_layout_get_ds_info(struct inode *inode)
2037 {
2038 	struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
2039 
2040 	if (layout == NULL)
2041 		return NULL;
2042 
2043 	return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2044 }
2045 
2046 static void
2047 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
2048 		struct pnfs_layout_segment *lseg)
2049 {
2050 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
2051 	struct inode *inode = lseg->pls_layout->plh_inode;
2052 	struct pnfs_commit_array *array, *new;
2053 
2054 	new = pnfs_alloc_commit_array(flseg->mirror_array_cnt,
2055 				      nfs_io_gfp_mask());
2056 	if (new) {
2057 		spin_lock(&inode->i_lock);
2058 		array = pnfs_add_commit_array(fl_cinfo, new, lseg);
2059 		spin_unlock(&inode->i_lock);
2060 		if (array != new)
2061 			pnfs_free_commit_array(new);
2062 	}
2063 }
2064 
2065 static void
2066 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
2067 		struct inode *inode)
2068 {
2069 	spin_lock(&inode->i_lock);
2070 	pnfs_generic_ds_cinfo_destroy(fl_cinfo);
2071 	spin_unlock(&inode->i_lock);
2072 }
2073 
2074 static void
2075 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2076 {
2077 	nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2078 						  id_node));
2079 }
2080 
2081 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2082 				  const struct nfs4_layoutreturn_args *args,
2083 				  const struct nfs4_flexfile_layoutreturn_args *ff_args)
2084 {
2085 	__be32 *start;
2086 
2087 	start = xdr_reserve_space(xdr, 4);
2088 	if (unlikely(!start))
2089 		return -E2BIG;
2090 
2091 	*start = cpu_to_be32(ff_args->num_errors);
2092 	/* This assume we always return _ALL_ layouts */
2093 	return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2094 }
2095 
2096 static void
2097 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2098 {
2099 	WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2100 }
2101 
2102 static void
2103 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2104 			    const nfs4_stateid *stateid,
2105 			    const struct nfs42_layoutstat_devinfo *devinfo)
2106 {
2107 	__be32 *p;
2108 
2109 	p = xdr_reserve_space(xdr, 8 + 8);
2110 	p = xdr_encode_hyper(p, devinfo->offset);
2111 	p = xdr_encode_hyper(p, devinfo->length);
2112 	encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2113 	p = xdr_reserve_space(xdr, 4*8);
2114 	p = xdr_encode_hyper(p, devinfo->read_count);
2115 	p = xdr_encode_hyper(p, devinfo->read_bytes);
2116 	p = xdr_encode_hyper(p, devinfo->write_count);
2117 	p = xdr_encode_hyper(p, devinfo->write_bytes);
2118 	encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2119 }
2120 
2121 static void
2122 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2123 			    const nfs4_stateid *stateid,
2124 			    const struct nfs42_layoutstat_devinfo *devinfo)
2125 {
2126 	ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2127 	ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2128 			devinfo->ld_private.data);
2129 }
2130 
2131 /* report nothing for now */
2132 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2133 		const struct nfs4_layoutreturn_args *args,
2134 		struct nfs4_flexfile_layoutreturn_args *ff_args)
2135 {
2136 	__be32 *p;
2137 	int i;
2138 
2139 	p = xdr_reserve_space(xdr, 4);
2140 	*p = cpu_to_be32(ff_args->num_dev);
2141 	for (i = 0; i < ff_args->num_dev; i++)
2142 		ff_layout_encode_ff_iostat(xdr,
2143 				&args->layout->plh_stateid,
2144 				&ff_args->devinfo[i]);
2145 }
2146 
2147 static void
2148 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2149 		unsigned int num_entries)
2150 {
2151 	unsigned int i;
2152 
2153 	for (i = 0; i < num_entries; i++) {
2154 		if (!devinfo[i].ld_private.ops)
2155 			continue;
2156 		if (!devinfo[i].ld_private.ops->free)
2157 			continue;
2158 		devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2159 	}
2160 }
2161 
2162 static struct nfs4_deviceid_node *
2163 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2164 			      struct pnfs_device *pdev, gfp_t gfp_flags)
2165 {
2166 	struct nfs4_ff_layout_ds *dsaddr;
2167 
2168 	dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2169 	if (!dsaddr)
2170 		return NULL;
2171 	return &dsaddr->id_node;
2172 }
2173 
2174 static void
2175 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2176 		const void *voidargs,
2177 		const struct nfs4_xdr_opaque_data *ff_opaque)
2178 {
2179 	const struct nfs4_layoutreturn_args *args = voidargs;
2180 	struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2181 	struct xdr_buf tmp_buf = {
2182 		.head = {
2183 			[0] = {
2184 				.iov_base = page_address(ff_args->pages[0]),
2185 			},
2186 		},
2187 		.buflen = PAGE_SIZE,
2188 	};
2189 	struct xdr_stream tmp_xdr;
2190 	__be32 *start;
2191 
2192 	dprintk("%s: Begin\n", __func__);
2193 
2194 	xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2195 
2196 	ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2197 	ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2198 
2199 	start = xdr_reserve_space(xdr, 4);
2200 	*start = cpu_to_be32(tmp_buf.len);
2201 	xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2202 
2203 	dprintk("%s: Return\n", __func__);
2204 }
2205 
2206 static void
2207 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2208 {
2209 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2210 
2211 	if (!args->data)
2212 		return;
2213 	ff_args = args->data;
2214 	args->data = NULL;
2215 
2216 	ff_layout_free_ds_ioerr(&ff_args->errors);
2217 	ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2218 
2219 	put_page(ff_args->pages[0]);
2220 	kfree(ff_args);
2221 }
2222 
2223 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2224 	.encode = ff_layout_encode_layoutreturn,
2225 	.free = ff_layout_free_layoutreturn,
2226 };
2227 
2228 static int
2229 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2230 {
2231 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2232 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2233 
2234 	ff_args = kmalloc(sizeof(*ff_args), nfs_io_gfp_mask());
2235 	if (!ff_args)
2236 		goto out_nomem;
2237 	ff_args->pages[0] = alloc_page(nfs_io_gfp_mask());
2238 	if (!ff_args->pages[0])
2239 		goto out_nomem_free;
2240 
2241 	INIT_LIST_HEAD(&ff_args->errors);
2242 	ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2243 			&args->range, &ff_args->errors,
2244 			FF_LAYOUTRETURN_MAXERR);
2245 
2246 	spin_lock(&args->inode->i_lock);
2247 	ff_args->num_dev = ff_layout_mirror_prepare_stats(
2248 		&ff_layout->generic_hdr, &ff_args->devinfo[0],
2249 		ARRAY_SIZE(ff_args->devinfo), NFS4_FF_OP_LAYOUTRETURN);
2250 	spin_unlock(&args->inode->i_lock);
2251 
2252 	args->ld_private->ops = &layoutreturn_ops;
2253 	args->ld_private->data = ff_args;
2254 	return 0;
2255 out_nomem_free:
2256 	kfree(ff_args);
2257 out_nomem:
2258 	return -ENOMEM;
2259 }
2260 
2261 #ifdef CONFIG_NFS_V4_2
2262 void
2263 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2264 {
2265 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
2266 	struct nfs42_layout_error *errors;
2267 	LIST_HEAD(head);
2268 
2269 	if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2270 		return;
2271 	ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2272 	if (list_empty(&head))
2273 		return;
2274 
2275 	errors = kmalloc_array(NFS42_LAYOUTERROR_MAX, sizeof(*errors),
2276 			       nfs_io_gfp_mask());
2277 	if (errors != NULL) {
2278 		const struct nfs4_ff_layout_ds_err *pos;
2279 		size_t n = 0;
2280 
2281 		list_for_each_entry(pos, &head, list) {
2282 			errors[n].offset = pos->offset;
2283 			errors[n].length = pos->length;
2284 			nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2285 			errors[n].errors[0].dev_id = pos->deviceid;
2286 			errors[n].errors[0].status = pos->status;
2287 			errors[n].errors[0].opnum = pos->opnum;
2288 			n++;
2289 			if (!list_is_last(&pos->list, &head) &&
2290 			    n < NFS42_LAYOUTERROR_MAX)
2291 				continue;
2292 			if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2293 				break;
2294 			n = 0;
2295 		}
2296 		kfree(errors);
2297 	}
2298 	ff_layout_free_ds_ioerr(&head);
2299 }
2300 #else
2301 void
2302 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2303 {
2304 }
2305 #endif
2306 
2307 static int
2308 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2309 {
2310 	const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2311 
2312 	return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2313 }
2314 
2315 static size_t
2316 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2317 			  const int buflen)
2318 {
2319 	const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2320 	const struct in6_addr *addr = &sin6->sin6_addr;
2321 
2322 	/*
2323 	 * RFC 4291, Section 2.2.2
2324 	 *
2325 	 * Shorthanded ANY address
2326 	 */
2327 	if (ipv6_addr_any(addr))
2328 		return snprintf(buf, buflen, "::");
2329 
2330 	/*
2331 	 * RFC 4291, Section 2.2.2
2332 	 *
2333 	 * Shorthanded loopback address
2334 	 */
2335 	if (ipv6_addr_loopback(addr))
2336 		return snprintf(buf, buflen, "::1");
2337 
2338 	/*
2339 	 * RFC 4291, Section 2.2.3
2340 	 *
2341 	 * Special presentation address format for mapped v4
2342 	 * addresses.
2343 	 */
2344 	if (ipv6_addr_v4mapped(addr))
2345 		return snprintf(buf, buflen, "::ffff:%pI4",
2346 					&addr->s6_addr32[3]);
2347 
2348 	/*
2349 	 * RFC 4291, Section 2.2.1
2350 	 */
2351 	return snprintf(buf, buflen, "%pI6c", addr);
2352 }
2353 
2354 /* Derived from rpc_sockaddr2uaddr */
2355 static void
2356 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2357 {
2358 	struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2359 	char portbuf[RPCBIND_MAXUADDRPLEN];
2360 	char addrbuf[RPCBIND_MAXUADDRLEN];
2361 	unsigned short port;
2362 	int len, netid_len;
2363 	__be32 *p;
2364 
2365 	switch (sap->sa_family) {
2366 	case AF_INET:
2367 		if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2368 			return;
2369 		port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2370 		break;
2371 	case AF_INET6:
2372 		if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2373 			return;
2374 		port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2375 		break;
2376 	default:
2377 		WARN_ON_ONCE(1);
2378 		return;
2379 	}
2380 
2381 	snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2382 	len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2383 
2384 	netid_len = strlen(da->da_netid);
2385 	p = xdr_reserve_space(xdr, 4 + netid_len);
2386 	xdr_encode_opaque(p, da->da_netid, netid_len);
2387 
2388 	p = xdr_reserve_space(xdr, 4 + len);
2389 	xdr_encode_opaque(p, addrbuf, len);
2390 }
2391 
2392 static void
2393 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2394 			 ktime_t t)
2395 {
2396 	struct timespec64 ts;
2397 	__be32 *p;
2398 
2399 	p = xdr_reserve_space(xdr, 12);
2400 	ts = ktime_to_timespec64(t);
2401 	p = xdr_encode_hyper(p, ts.tv_sec);
2402 	*p++ = cpu_to_be32(ts.tv_nsec);
2403 }
2404 
2405 static void
2406 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2407 			    struct nfs4_ff_io_stat *stat)
2408 {
2409 	__be32 *p;
2410 
2411 	p = xdr_reserve_space(xdr, 5 * 8);
2412 	p = xdr_encode_hyper(p, stat->ops_requested);
2413 	p = xdr_encode_hyper(p, stat->bytes_requested);
2414 	p = xdr_encode_hyper(p, stat->ops_completed);
2415 	p = xdr_encode_hyper(p, stat->bytes_completed);
2416 	p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2417 	ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2418 	ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2419 }
2420 
2421 static void
2422 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2423 			      const struct nfs42_layoutstat_devinfo *devinfo,
2424 			      struct nfs4_ff_layout_mirror *mirror)
2425 {
2426 	struct nfs4_pnfs_ds_addr *da;
2427 	struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2428 	struct nfs_fh *fh = &mirror->fh_versions[0];
2429 	__be32 *p;
2430 
2431 	da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2432 	dprintk("%s: DS %s: encoding address %s\n",
2433 		__func__, ds->ds_remotestr, da->da_remotestr);
2434 	/* netaddr4 */
2435 	ff_layout_encode_netaddr(xdr, da);
2436 	/* nfs_fh4 */
2437 	p = xdr_reserve_space(xdr, 4 + fh->size);
2438 	xdr_encode_opaque(p, fh->data, fh->size);
2439 	/* ff_io_latency4 read */
2440 	spin_lock(&mirror->lock);
2441 	ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2442 	/* ff_io_latency4 write */
2443 	ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2444 	spin_unlock(&mirror->lock);
2445 	/* nfstime4 */
2446 	ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2447 	/* bool */
2448 	p = xdr_reserve_space(xdr, 4);
2449 	*p = cpu_to_be32(false);
2450 }
2451 
2452 static void
2453 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2454 			     const struct nfs4_xdr_opaque_data *opaque)
2455 {
2456 	struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2457 			struct nfs42_layoutstat_devinfo, ld_private);
2458 	__be32 *start;
2459 
2460 	/* layoutupdate length */
2461 	start = xdr_reserve_space(xdr, 4);
2462 	ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2463 
2464 	*start = cpu_to_be32((xdr->p - start - 1) * 4);
2465 }
2466 
2467 static void
2468 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2469 {
2470 	struct nfs4_ff_layout_mirror *mirror = opaque->data;
2471 
2472 	ff_layout_put_mirror(mirror);
2473 }
2474 
2475 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2476 	.encode = ff_layout_encode_layoutstats,
2477 	.free	= ff_layout_free_layoutstats,
2478 };
2479 
2480 static int
2481 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2482 			       struct nfs42_layoutstat_devinfo *devinfo,
2483 			       int dev_limit, enum nfs4_ff_op_type type)
2484 {
2485 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2486 	struct nfs4_ff_layout_mirror *mirror;
2487 	struct nfs4_deviceid_node *dev;
2488 	int i = 0;
2489 
2490 	list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2491 		if (i >= dev_limit)
2492 			break;
2493 		if (IS_ERR_OR_NULL(mirror->mirror_ds))
2494 			continue;
2495 		if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL,
2496 					&mirror->flags) &&
2497 		    type != NFS4_FF_OP_LAYOUTRETURN)
2498 			continue;
2499 		/* mirror refcount put in cleanup_layoutstats */
2500 		if (!refcount_inc_not_zero(&mirror->ref))
2501 			continue;
2502 		dev = &mirror->mirror_ds->id_node;
2503 		memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2504 		devinfo->offset = 0;
2505 		devinfo->length = NFS4_MAX_UINT64;
2506 		spin_lock(&mirror->lock);
2507 		devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2508 		devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2509 		devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2510 		devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2511 		spin_unlock(&mirror->lock);
2512 		devinfo->layout_type = LAYOUT_FLEX_FILES;
2513 		devinfo->ld_private.ops = &layoutstat_ops;
2514 		devinfo->ld_private.data = mirror;
2515 
2516 		devinfo++;
2517 		i++;
2518 	}
2519 	return i;
2520 }
2521 
2522 static int
2523 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2524 {
2525 	struct nfs4_flexfile_layout *ff_layout;
2526 	const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2527 
2528 	/* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2529 	args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo),
2530 				      nfs_io_gfp_mask());
2531 	if (!args->devinfo)
2532 		return -ENOMEM;
2533 
2534 	spin_lock(&args->inode->i_lock);
2535 	ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2536 	args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2537 						       &args->devinfo[0],
2538 						       dev_count,
2539 						       NFS4_FF_OP_LAYOUTSTATS);
2540 	spin_unlock(&args->inode->i_lock);
2541 	if (!args->num_dev) {
2542 		kfree(args->devinfo);
2543 		args->devinfo = NULL;
2544 		return -ENOENT;
2545 	}
2546 
2547 	return 0;
2548 }
2549 
2550 static int
2551 ff_layout_set_layoutdriver(struct nfs_server *server,
2552 		const struct nfs_fh *dummy)
2553 {
2554 #if IS_ENABLED(CONFIG_NFS_V4_2)
2555 	server->caps |= NFS_CAP_LAYOUTSTATS;
2556 #endif
2557 	return 0;
2558 }
2559 
2560 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2561 	.setup_ds_info		= ff_layout_setup_ds_info,
2562 	.release_ds_info	= ff_layout_release_ds_info,
2563 	.mark_request_commit	= pnfs_layout_mark_request_commit,
2564 	.clear_request_commit	= pnfs_generic_clear_request_commit,
2565 	.scan_commit_lists	= pnfs_generic_scan_commit_lists,
2566 	.recover_commit_reqs	= pnfs_generic_recover_commit_reqs,
2567 	.commit_pagelist	= ff_layout_commit_pagelist,
2568 };
2569 
2570 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2571 	.id			= LAYOUT_FLEX_FILES,
2572 	.name			= "LAYOUT_FLEX_FILES",
2573 	.owner			= THIS_MODULE,
2574 	.flags			= PNFS_LAYOUTGET_ON_OPEN,
2575 	.max_layoutget_response	= 4096, /* 1 page or so... */
2576 	.set_layoutdriver	= ff_layout_set_layoutdriver,
2577 	.alloc_layout_hdr	= ff_layout_alloc_layout_hdr,
2578 	.free_layout_hdr	= ff_layout_free_layout_hdr,
2579 	.alloc_lseg		= ff_layout_alloc_lseg,
2580 	.free_lseg		= ff_layout_free_lseg,
2581 	.add_lseg		= ff_layout_add_lseg,
2582 	.pg_read_ops		= &ff_layout_pg_read_ops,
2583 	.pg_write_ops		= &ff_layout_pg_write_ops,
2584 	.get_ds_info		= ff_layout_get_ds_info,
2585 	.free_deviceid_node	= ff_layout_free_deviceid_node,
2586 	.read_pagelist		= ff_layout_read_pagelist,
2587 	.write_pagelist		= ff_layout_write_pagelist,
2588 	.alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2589 	.prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2590 	.sync			= pnfs_nfs_generic_sync,
2591 	.prepare_layoutstats	= ff_layout_prepare_layoutstats,
2592 	.cancel_io		= ff_layout_cancel_io,
2593 };
2594 
2595 static int __init nfs4flexfilelayout_init(void)
2596 {
2597 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2598 	       __func__);
2599 	return pnfs_register_layoutdriver(&flexfilelayout_type);
2600 }
2601 
2602 static void __exit nfs4flexfilelayout_exit(void)
2603 {
2604 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2605 	       __func__);
2606 	pnfs_unregister_layoutdriver(&flexfilelayout_type);
2607 }
2608 
2609 MODULE_ALIAS("nfs-layouttype4-4");
2610 
2611 MODULE_LICENSE("GPL");
2612 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2613 
2614 module_init(nfs4flexfilelayout_init);
2615 module_exit(nfs4flexfilelayout_exit);
2616 
2617 module_param(io_maxretrans, ushort, 0644);
2618 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2619 			"retries an I/O request before returning an error. ");
2620