xref: /linux/fs/nfs/flexfilelayout/flexfilelayout.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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/file.h>
15 #include <linux/sched/mm.h>
16 
17 #include <linux/sunrpc/metrics.h>
18 
19 #include "flexfilelayout.h"
20 #include "../nfs4session.h"
21 #include "../nfs4idmap.h"
22 #include "../internal.h"
23 #include "../delegation.h"
24 #include "../nfs4trace.h"
25 #include "../iostat.h"
26 #include "../nfs.h"
27 #include "../nfs42.h"
28 
29 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
30 
31 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
32 #define FF_LAYOUTRETURN_MAXERR 20
33 
34 enum nfs4_ff_op_type {
35 	NFS4_FF_OP_LAYOUTSTATS,
36 	NFS4_FF_OP_LAYOUTRETURN,
37 };
38 
39 static unsigned short io_maxretrans;
40 
41 static const struct pnfs_commit_ops ff_layout_commit_ops;
42 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
43 		struct nfs_pgio_header *hdr);
44 static int
45 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
46 			       struct nfs42_layoutstat_devinfo *devinfo,
47 			       int dev_limit, enum nfs4_ff_op_type type);
48 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
49 			      const struct nfs42_layoutstat_devinfo *devinfo,
50 			      struct nfs4_ff_layout_mirror *mirror);
51 
52 static struct pnfs_layout_hdr *
53 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
54 {
55 	struct nfs4_flexfile_layout *ffl;
56 
57 	ffl = kzalloc(sizeof(*ffl), gfp_flags);
58 	if (ffl) {
59 		pnfs_init_ds_commit_info(&ffl->commit_info);
60 		INIT_LIST_HEAD(&ffl->error_list);
61 		INIT_LIST_HEAD(&ffl->mirrors);
62 		ffl->last_report_time = ktime_get();
63 		ffl->commit_info.ops = &ff_layout_commit_ops;
64 		return &ffl->generic_hdr;
65 	} else
66 		return NULL;
67 }
68 
69 static void
70 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
71 {
72 	struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
73 	struct nfs4_ff_layout_ds_err *err, *n;
74 
75 	list_for_each_entry_safe(err, n, &ffl->error_list, list) {
76 		list_del(&err->list);
77 		kfree(err);
78 	}
79 	kfree_rcu(ffl, generic_hdr.plh_rcu);
80 }
81 
82 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
83 {
84 	__be32 *p;
85 
86 	p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
87 	if (unlikely(p == NULL))
88 		return -ENOBUFS;
89 	stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
90 	memcpy(stateid->data, p, NFS4_STATEID_SIZE);
91 	dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
92 		p[0], p[1], p[2], p[3]);
93 	return 0;
94 }
95 
96 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
97 {
98 	__be32 *p;
99 
100 	p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
101 	if (unlikely(!p))
102 		return -ENOBUFS;
103 	memcpy(devid, p, NFS4_DEVICEID4_SIZE);
104 	nfs4_print_deviceid(devid);
105 	return 0;
106 }
107 
108 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
109 {
110 	__be32 *p;
111 
112 	p = xdr_inline_decode(xdr, 4);
113 	if (unlikely(!p))
114 		return -ENOBUFS;
115 	fh->size = be32_to_cpup(p++);
116 	if (fh->size > NFS_MAXFHSIZE) {
117 		printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
118 		       fh->size);
119 		return -EOVERFLOW;
120 	}
121 	/* fh.data */
122 	p = xdr_inline_decode(xdr, fh->size);
123 	if (unlikely(!p))
124 		return -ENOBUFS;
125 	memcpy(&fh->data, p, fh->size);
126 	dprintk("%s: fh len %d\n", __func__, fh->size);
127 
128 	return 0;
129 }
130 
131 /*
132  * Currently only stringified uids and gids are accepted.
133  * I.e., kerberos is not supported to the DSes, so no pricipals.
134  *
135  * That means that one common function will suffice, but when
136  * principals are added, this should be split to accomodate
137  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
138  */
139 static int
140 decode_name(struct xdr_stream *xdr, u32 *id)
141 {
142 	__be32 *p;
143 	int len;
144 
145 	/* opaque_length(4)*/
146 	p = xdr_inline_decode(xdr, 4);
147 	if (unlikely(!p))
148 		return -ENOBUFS;
149 	len = be32_to_cpup(p++);
150 	if (len < 0)
151 		return -EINVAL;
152 
153 	dprintk("%s: len %u\n", __func__, len);
154 
155 	/* opaque body */
156 	p = xdr_inline_decode(xdr, len);
157 	if (unlikely(!p))
158 		return -ENOBUFS;
159 
160 	if (!nfs_map_string_to_numeric((char *)p, len, id))
161 		return -EINVAL;
162 
163 	return 0;
164 }
165 
166 static struct nfsd_file *
167 ff_local_open_fh(struct nfs_client *clp, const struct cred *cred,
168 		 struct nfs_fh *fh, fmode_t mode)
169 {
170 	if (mode & FMODE_WRITE) {
171 		/*
172 		 * Always request read and write access since this corresponds
173 		 * to a rw layout.
174 		 */
175 		mode |= FMODE_READ;
176 	}
177 
178 	return nfs_local_open_fh(clp, cred, fh, mode);
179 }
180 
181 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
182 		const struct nfs4_ff_layout_mirror *m2)
183 {
184 	int i, j;
185 
186 	if (m1->fh_versions_cnt != m2->fh_versions_cnt)
187 		return false;
188 	for (i = 0; i < m1->fh_versions_cnt; i++) {
189 		bool found_fh = false;
190 		for (j = 0; j < m2->fh_versions_cnt; j++) {
191 			if (nfs_compare_fh(&m1->fh_versions[i],
192 					&m2->fh_versions[j]) == 0) {
193 				found_fh = true;
194 				break;
195 			}
196 		}
197 		if (!found_fh)
198 			return false;
199 	}
200 	return true;
201 }
202 
203 static struct nfs4_ff_layout_mirror *
204 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
205 		struct nfs4_ff_layout_mirror *mirror)
206 {
207 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
208 	struct nfs4_ff_layout_mirror *pos;
209 	struct inode *inode = lo->plh_inode;
210 
211 	spin_lock(&inode->i_lock);
212 	list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
213 		if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
214 			continue;
215 		if (!ff_mirror_match_fh(mirror, pos))
216 			continue;
217 		if (refcount_inc_not_zero(&pos->ref)) {
218 			spin_unlock(&inode->i_lock);
219 			return pos;
220 		}
221 	}
222 	list_add(&mirror->mirrors, &ff_layout->mirrors);
223 	mirror->layout = lo;
224 	spin_unlock(&inode->i_lock);
225 	return mirror;
226 }
227 
228 static void
229 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
230 {
231 	struct inode *inode;
232 	if (mirror->layout == NULL)
233 		return;
234 	inode = mirror->layout->plh_inode;
235 	spin_lock(&inode->i_lock);
236 	list_del(&mirror->mirrors);
237 	spin_unlock(&inode->i_lock);
238 	mirror->layout = NULL;
239 }
240 
241 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
242 {
243 	struct nfs4_ff_layout_mirror *mirror;
244 
245 	mirror = kzalloc(sizeof(*mirror), gfp_flags);
246 	if (mirror != NULL) {
247 		spin_lock_init(&mirror->lock);
248 		refcount_set(&mirror->ref, 1);
249 		INIT_LIST_HEAD(&mirror->mirrors);
250 	}
251 	return mirror;
252 }
253 
254 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
255 {
256 	const struct cred *cred;
257 
258 	ff_layout_remove_mirror(mirror);
259 	kfree(mirror->fh_versions);
260 	cred = rcu_access_pointer(mirror->ro_cred);
261 	put_cred(cred);
262 	cred = rcu_access_pointer(mirror->rw_cred);
263 	put_cred(cred);
264 	nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
265 	kfree(mirror);
266 }
267 
268 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
269 {
270 	if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
271 		ff_layout_free_mirror(mirror);
272 }
273 
274 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
275 {
276 	u32 i;
277 
278 	for (i = 0; i < fls->mirror_array_cnt; i++)
279 		ff_layout_put_mirror(fls->mirror_array[i]);
280 }
281 
282 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
283 {
284 	if (fls) {
285 		ff_layout_free_mirror_array(fls);
286 		kfree(fls);
287 	}
288 }
289 
290 static bool
291 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
292 		struct pnfs_layout_segment *l2)
293 {
294 	const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
295 	const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
296 	u32 i;
297 
298 	if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
299 		return false;
300 	for (i = 0; i < fl1->mirror_array_cnt; i++) {
301 		if (fl1->mirror_array[i] != fl2->mirror_array[i])
302 			return false;
303 	}
304 	return true;
305 }
306 
307 static bool
308 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
309 		const struct pnfs_layout_range *l2)
310 {
311 	u64 end1, end2;
312 
313 	if (l1->iomode != l2->iomode)
314 		return l1->iomode != IOMODE_READ;
315 	end1 = pnfs_calc_offset_end(l1->offset, l1->length);
316 	end2 = pnfs_calc_offset_end(l2->offset, l2->length);
317 	if (end1 < l2->offset)
318 		return false;
319 	if (end2 < l1->offset)
320 		return true;
321 	return l2->offset <= l1->offset;
322 }
323 
324 static bool
325 ff_lseg_merge(struct pnfs_layout_segment *new,
326 		struct pnfs_layout_segment *old)
327 {
328 	u64 new_end, old_end;
329 
330 	if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
331 		return false;
332 	if (new->pls_range.iomode != old->pls_range.iomode)
333 		return false;
334 	old_end = pnfs_calc_offset_end(old->pls_range.offset,
335 			old->pls_range.length);
336 	if (old_end < new->pls_range.offset)
337 		return false;
338 	new_end = pnfs_calc_offset_end(new->pls_range.offset,
339 			new->pls_range.length);
340 	if (new_end < old->pls_range.offset)
341 		return false;
342 	if (!ff_lseg_match_mirrors(new, old))
343 		return false;
344 
345 	/* Mergeable: copy info from 'old' to 'new' */
346 	if (new_end < old_end)
347 		new_end = old_end;
348 	if (new->pls_range.offset < old->pls_range.offset)
349 		new->pls_range.offset = old->pls_range.offset;
350 	new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
351 			new_end);
352 	if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
353 		set_bit(NFS_LSEG_ROC, &new->pls_flags);
354 	return true;
355 }
356 
357 static void
358 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
359 		struct pnfs_layout_segment *lseg,
360 		struct list_head *free_me)
361 {
362 	pnfs_generic_layout_insert_lseg(lo, lseg,
363 			ff_lseg_range_is_after,
364 			ff_lseg_merge,
365 			free_me);
366 }
367 
368 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
369 {
370 	int i, j;
371 
372 	for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
373 		for (j = i + 1; j < fls->mirror_array_cnt; j++)
374 			if (fls->mirror_array[i]->efficiency <
375 			    fls->mirror_array[j]->efficiency)
376 				swap(fls->mirror_array[i],
377 				     fls->mirror_array[j]);
378 	}
379 }
380 
381 static struct pnfs_layout_segment *
382 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
383 		     struct nfs4_layoutget_res *lgr,
384 		     gfp_t gfp_flags)
385 {
386 	struct pnfs_layout_segment *ret;
387 	struct nfs4_ff_layout_segment *fls = NULL;
388 	struct xdr_stream stream;
389 	struct xdr_buf buf;
390 	struct page *scratch;
391 	u64 stripe_unit;
392 	u32 mirror_array_cnt;
393 	__be32 *p;
394 	int i, rc;
395 
396 	dprintk("--> %s\n", __func__);
397 	scratch = alloc_page(gfp_flags);
398 	if (!scratch)
399 		return ERR_PTR(-ENOMEM);
400 
401 	xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
402 			      lgr->layoutp->len);
403 	xdr_set_scratch_page(&stream, scratch);
404 
405 	/* stripe unit and mirror_array_cnt */
406 	rc = -EIO;
407 	p = xdr_inline_decode(&stream, 8 + 4);
408 	if (!p)
409 		goto out_err_free;
410 
411 	p = xdr_decode_hyper(p, &stripe_unit);
412 	mirror_array_cnt = be32_to_cpup(p++);
413 	dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
414 		stripe_unit, mirror_array_cnt);
415 
416 	if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
417 	    mirror_array_cnt == 0)
418 		goto out_err_free;
419 
420 	rc = -ENOMEM;
421 	fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
422 			gfp_flags);
423 	if (!fls)
424 		goto out_err_free;
425 
426 	fls->mirror_array_cnt = mirror_array_cnt;
427 	fls->stripe_unit = stripe_unit;
428 
429 	for (i = 0; i < fls->mirror_array_cnt; i++) {
430 		struct nfs4_ff_layout_mirror *mirror;
431 		struct cred *kcred;
432 		const struct cred __rcu *cred;
433 		kuid_t uid;
434 		kgid_t gid;
435 		u32 ds_count, fh_count, id;
436 		int j;
437 
438 		rc = -EIO;
439 		p = xdr_inline_decode(&stream, 4);
440 		if (!p)
441 			goto out_err_free;
442 		ds_count = be32_to_cpup(p);
443 
444 		/* FIXME: allow for striping? */
445 		if (ds_count != 1)
446 			goto out_err_free;
447 
448 		fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
449 		if (fls->mirror_array[i] == NULL) {
450 			rc = -ENOMEM;
451 			goto out_err_free;
452 		}
453 
454 		fls->mirror_array[i]->ds_count = ds_count;
455 
456 		/* deviceid */
457 		rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
458 		if (rc)
459 			goto out_err_free;
460 
461 		/* efficiency */
462 		rc = -EIO;
463 		p = xdr_inline_decode(&stream, 4);
464 		if (!p)
465 			goto out_err_free;
466 		fls->mirror_array[i]->efficiency = be32_to_cpup(p);
467 
468 		/* stateid */
469 		rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
470 		if (rc)
471 			goto out_err_free;
472 
473 		/* fh */
474 		rc = -EIO;
475 		p = xdr_inline_decode(&stream, 4);
476 		if (!p)
477 			goto out_err_free;
478 		fh_count = be32_to_cpup(p);
479 
480 		fls->mirror_array[i]->fh_versions =
481 			kcalloc(fh_count, sizeof(struct nfs_fh),
482 				gfp_flags);
483 		if (fls->mirror_array[i]->fh_versions == NULL) {
484 			rc = -ENOMEM;
485 			goto out_err_free;
486 		}
487 
488 		for (j = 0; j < fh_count; j++) {
489 			rc = decode_nfs_fh(&stream,
490 					   &fls->mirror_array[i]->fh_versions[j]);
491 			if (rc)
492 				goto out_err_free;
493 		}
494 
495 		fls->mirror_array[i]->fh_versions_cnt = fh_count;
496 
497 		/* user */
498 		rc = decode_name(&stream, &id);
499 		if (rc)
500 			goto out_err_free;
501 
502 		uid = make_kuid(&init_user_ns, id);
503 
504 		/* group */
505 		rc = decode_name(&stream, &id);
506 		if (rc)
507 			goto out_err_free;
508 
509 		gid = make_kgid(&init_user_ns, id);
510 
511 		if (gfp_flags & __GFP_FS)
512 			kcred = prepare_kernel_cred(&init_task);
513 		else {
514 			unsigned int nofs_flags = memalloc_nofs_save();
515 			kcred = prepare_kernel_cred(&init_task);
516 			memalloc_nofs_restore(nofs_flags);
517 		}
518 		rc = -ENOMEM;
519 		if (!kcred)
520 			goto out_err_free;
521 		kcred->fsuid = uid;
522 		kcred->fsgid = gid;
523 		cred = RCU_INITIALIZER(kcred);
524 
525 		if (lgr->range.iomode == IOMODE_READ)
526 			rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
527 		else
528 			rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
529 
530 		mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
531 		if (mirror != fls->mirror_array[i]) {
532 			/* swap cred ptrs so free_mirror will clean up old */
533 			if (lgr->range.iomode == IOMODE_READ) {
534 				cred = xchg(&mirror->ro_cred, cred);
535 				rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
536 			} else {
537 				cred = xchg(&mirror->rw_cred, cred);
538 				rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
539 			}
540 			ff_layout_free_mirror(fls->mirror_array[i]);
541 			fls->mirror_array[i] = mirror;
542 		}
543 
544 		dprintk("%s: iomode %s uid %u gid %u\n", __func__,
545 			lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
546 			from_kuid(&init_user_ns, uid),
547 			from_kgid(&init_user_ns, gid));
548 	}
549 
550 	p = xdr_inline_decode(&stream, 4);
551 	if (!p)
552 		goto out_sort_mirrors;
553 	fls->flags = be32_to_cpup(p);
554 
555 	p = xdr_inline_decode(&stream, 4);
556 	if (!p)
557 		goto out_sort_mirrors;
558 	for (i=0; i < fls->mirror_array_cnt; i++)
559 		fls->mirror_array[i]->report_interval = be32_to_cpup(p);
560 
561 out_sort_mirrors:
562 	ff_layout_sort_mirrors(fls);
563 	ret = &fls->generic_hdr;
564 	dprintk("<-- %s (success)\n", __func__);
565 out_free_page:
566 	__free_page(scratch);
567 	return ret;
568 out_err_free:
569 	_ff_layout_free_lseg(fls);
570 	ret = ERR_PTR(rc);
571 	dprintk("<-- %s (%d)\n", __func__, rc);
572 	goto out_free_page;
573 }
574 
575 static void
576 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
577 {
578 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
579 
580 	dprintk("--> %s\n", __func__);
581 
582 	if (lseg->pls_range.iomode == IOMODE_RW) {
583 		struct nfs4_flexfile_layout *ffl;
584 		struct inode *inode;
585 
586 		ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
587 		inode = ffl->generic_hdr.plh_inode;
588 		spin_lock(&inode->i_lock);
589 		pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
590 		spin_unlock(&inode->i_lock);
591 	}
592 	_ff_layout_free_lseg(fls);
593 }
594 
595 static void
596 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
597 {
598 	/* first IO request? */
599 	if (atomic_inc_return(&timer->n_ops) == 1) {
600 		timer->start_time = now;
601 	}
602 }
603 
604 static ktime_t
605 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
606 {
607 	ktime_t start;
608 
609 	if (atomic_dec_return(&timer->n_ops) < 0)
610 		WARN_ON_ONCE(1);
611 
612 	start = timer->start_time;
613 	timer->start_time = now;
614 	return ktime_sub(now, start);
615 }
616 
617 static bool
618 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
619 			    struct nfs4_ff_layoutstat *layoutstat,
620 			    ktime_t now)
621 {
622 	s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
623 	struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
624 
625 	nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
626 	if (!mirror->start_time)
627 		mirror->start_time = now;
628 	if (mirror->report_interval != 0)
629 		report_interval = (s64)mirror->report_interval * 1000LL;
630 	else if (layoutstats_timer != 0)
631 		report_interval = (s64)layoutstats_timer * 1000LL;
632 	if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
633 			report_interval) {
634 		ffl->last_report_time = now;
635 		return true;
636 	}
637 
638 	return false;
639 }
640 
641 static void
642 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
643 		__u64 requested)
644 {
645 	struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
646 
647 	iostat->ops_requested++;
648 	iostat->bytes_requested += requested;
649 }
650 
651 static void
652 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
653 		__u64 requested,
654 		__u64 completed,
655 		ktime_t time_completed,
656 		ktime_t time_started)
657 {
658 	struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
659 	ktime_t completion_time = ktime_sub(time_completed, time_started);
660 	ktime_t timer;
661 
662 	iostat->ops_completed++;
663 	iostat->bytes_completed += completed;
664 	iostat->bytes_not_delivered += requested - completed;
665 
666 	timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
667 	iostat->total_busy_time =
668 			ktime_add(iostat->total_busy_time, timer);
669 	iostat->aggregate_completion_time =
670 			ktime_add(iostat->aggregate_completion_time,
671 					completion_time);
672 }
673 
674 static void
675 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
676 		struct nfs4_ff_layout_mirror *mirror,
677 		__u64 requested, ktime_t now)
678 {
679 	bool report;
680 
681 	spin_lock(&mirror->lock);
682 	report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
683 	nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
684 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
685 	spin_unlock(&mirror->lock);
686 
687 	if (report)
688 		pnfs_report_layoutstat(inode, nfs_io_gfp_mask());
689 }
690 
691 static void
692 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
693 		struct nfs4_ff_layout_mirror *mirror,
694 		__u64 requested,
695 		__u64 completed)
696 {
697 	spin_lock(&mirror->lock);
698 	nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
699 			requested, completed,
700 			ktime_get(), task->tk_start);
701 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
702 	spin_unlock(&mirror->lock);
703 }
704 
705 static void
706 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
707 		struct nfs4_ff_layout_mirror *mirror,
708 		__u64 requested, ktime_t now)
709 {
710 	bool report;
711 
712 	spin_lock(&mirror->lock);
713 	report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
714 	nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
715 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
716 	spin_unlock(&mirror->lock);
717 
718 	if (report)
719 		pnfs_report_layoutstat(inode, nfs_io_gfp_mask());
720 }
721 
722 static void
723 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
724 		struct nfs4_ff_layout_mirror *mirror,
725 		__u64 requested,
726 		__u64 completed,
727 		enum nfs3_stable_how committed)
728 {
729 	if (committed == NFS_UNSTABLE)
730 		requested = completed = 0;
731 
732 	spin_lock(&mirror->lock);
733 	nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
734 			requested, completed, ktime_get(), task->tk_start);
735 	set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
736 	spin_unlock(&mirror->lock);
737 }
738 
739 static void
740 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)
741 {
742 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
743 
744 	if (devid)
745 		nfs4_mark_deviceid_unavailable(devid);
746 }
747 
748 static void
749 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)
750 {
751 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
752 
753 	if (devid)
754 		nfs4_mark_deviceid_available(devid);
755 }
756 
757 static struct nfs4_pnfs_ds *
758 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
759 			     u32 start_idx, u32 *best_idx,
760 			     bool check_device)
761 {
762 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
763 	struct nfs4_ff_layout_mirror *mirror;
764 	struct nfs4_pnfs_ds *ds;
765 	u32 idx;
766 
767 	/* mirrors are initially sorted by efficiency */
768 	for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
769 		mirror = FF_LAYOUT_COMP(lseg, idx);
770 		ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
771 		if (!ds)
772 			continue;
773 
774 		if (check_device &&
775 		    nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
776 			continue;
777 
778 		*best_idx = idx;
779 		return ds;
780 	}
781 
782 	return NULL;
783 }
784 
785 static struct nfs4_pnfs_ds *
786 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
787 				 u32 start_idx, u32 *best_idx)
788 {
789 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
790 }
791 
792 static struct nfs4_pnfs_ds *
793 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
794 				   u32 start_idx, u32 *best_idx)
795 {
796 	return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
797 }
798 
799 static struct nfs4_pnfs_ds *
800 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
801 				  u32 start_idx, u32 *best_idx)
802 {
803 	struct nfs4_pnfs_ds *ds;
804 
805 	ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
806 	if (ds)
807 		return ds;
808 	return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
809 }
810 
811 static struct nfs4_pnfs_ds *
812 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio,
813 			  u32 *best_idx)
814 {
815 	struct pnfs_layout_segment *lseg = pgio->pg_lseg;
816 	struct nfs4_pnfs_ds *ds;
817 
818 	ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
819 					       best_idx);
820 	if (ds || !pgio->pg_mirror_idx)
821 		return ds;
822 	return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
823 }
824 
825 static void
826 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
827 		      struct nfs_page *req,
828 		      bool strict_iomode)
829 {
830 	pnfs_put_lseg(pgio->pg_lseg);
831 	pgio->pg_lseg =
832 		pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
833 				   req_offset(req), req->wb_bytes, IOMODE_READ,
834 				   strict_iomode, nfs_io_gfp_mask());
835 	if (IS_ERR(pgio->pg_lseg)) {
836 		pgio->pg_error = PTR_ERR(pgio->pg_lseg);
837 		pgio->pg_lseg = NULL;
838 	}
839 }
840 
841 static void
842 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
843 			struct nfs_page *req)
844 {
845 	struct nfs_pgio_mirror *pgm;
846 	struct nfs4_ff_layout_mirror *mirror;
847 	struct nfs4_pnfs_ds *ds;
848 	u32 ds_idx;
849 
850 retry:
851 	pnfs_generic_pg_check_layout(pgio, req);
852 	/* Use full layout for now */
853 	if (!pgio->pg_lseg) {
854 		ff_layout_pg_get_read(pgio, req, false);
855 		if (!pgio->pg_lseg)
856 			goto out_nolseg;
857 	}
858 	if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
859 		ff_layout_pg_get_read(pgio, req, true);
860 		if (!pgio->pg_lseg)
861 			goto out_nolseg;
862 	}
863 
864 	ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
865 	if (!ds) {
866 		if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
867 			goto out_mds;
868 		pnfs_generic_pg_cleanup(pgio);
869 		/* Sleep for 1 second before retrying */
870 		ssleep(1);
871 		goto retry;
872 	}
873 
874 	mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
875 	pgm = &pgio->pg_mirrors[0];
876 	pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
877 
878 	pgio->pg_mirror_idx = ds_idx;
879 
880 	if (NFS_SERVER(pgio->pg_inode)->flags &
881 			(NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
882 		pgio->pg_maxretrans = io_maxretrans;
883 	return;
884 out_nolseg:
885 	if (pgio->pg_error < 0)
886 		return;
887 out_mds:
888 	trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
889 			0, NFS4_MAX_UINT64, IOMODE_READ,
890 			NFS_I(pgio->pg_inode)->layout,
891 			pgio->pg_lseg);
892 	pgio->pg_maxretrans = 0;
893 	nfs_pageio_reset_read_mds(pgio);
894 }
895 
896 static void
897 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
898 			struct nfs_page *req)
899 {
900 	struct nfs4_ff_layout_mirror *mirror;
901 	struct nfs_pgio_mirror *pgm;
902 	struct nfs4_pnfs_ds *ds;
903 	u32 i;
904 
905 retry:
906 	pnfs_generic_pg_check_layout(pgio, req);
907 	if (!pgio->pg_lseg) {
908 		pgio->pg_lseg =
909 			pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
910 					   req_offset(req), req->wb_bytes,
911 					   IOMODE_RW, false, nfs_io_gfp_mask());
912 		if (IS_ERR(pgio->pg_lseg)) {
913 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
914 			pgio->pg_lseg = NULL;
915 			return;
916 		}
917 	}
918 	/* If no lseg, fall back to write through mds */
919 	if (pgio->pg_lseg == NULL)
920 		goto out_mds;
921 
922 	/* Use a direct mapping of ds_idx to pgio mirror_idx */
923 	if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
924 		goto out_eagain;
925 
926 	for (i = 0; i < pgio->pg_mirror_count; i++) {
927 		mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
928 		ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
929 		if (!ds) {
930 			if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
931 				goto out_mds;
932 			pnfs_generic_pg_cleanup(pgio);
933 			/* Sleep for 1 second before retrying */
934 			ssleep(1);
935 			goto retry;
936 		}
937 		pgm = &pgio->pg_mirrors[i];
938 		pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
939 	}
940 
941 	if (NFS_SERVER(pgio->pg_inode)->flags &
942 			(NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
943 		pgio->pg_maxretrans = io_maxretrans;
944 	return;
945 out_eagain:
946 	pnfs_generic_pg_cleanup(pgio);
947 	pgio->pg_error = -EAGAIN;
948 	return;
949 out_mds:
950 	trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
951 			0, NFS4_MAX_UINT64, IOMODE_RW,
952 			NFS_I(pgio->pg_inode)->layout,
953 			pgio->pg_lseg);
954 	pgio->pg_maxretrans = 0;
955 	nfs_pageio_reset_write_mds(pgio);
956 	pgio->pg_error = -EAGAIN;
957 }
958 
959 static unsigned int
960 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
961 				    struct nfs_page *req)
962 {
963 	if (!pgio->pg_lseg) {
964 		pgio->pg_lseg =
965 			pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
966 					   req_offset(req), req->wb_bytes,
967 					   IOMODE_RW, false, nfs_io_gfp_mask());
968 		if (IS_ERR(pgio->pg_lseg)) {
969 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
970 			pgio->pg_lseg = NULL;
971 			goto out;
972 		}
973 	}
974 	if (pgio->pg_lseg)
975 		return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
976 
977 	trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
978 			0, NFS4_MAX_UINT64, IOMODE_RW,
979 			NFS_I(pgio->pg_inode)->layout,
980 			pgio->pg_lseg);
981 	/* no lseg means that pnfs is not in use, so no mirroring here */
982 	nfs_pageio_reset_write_mds(pgio);
983 out:
984 	return 1;
985 }
986 
987 static u32
988 ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
989 {
990 	u32 old = desc->pg_mirror_idx;
991 
992 	desc->pg_mirror_idx = idx;
993 	return old;
994 }
995 
996 static struct nfs_pgio_mirror *
997 ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
998 {
999 	return &desc->pg_mirrors[idx];
1000 }
1001 
1002 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1003 	.pg_init = ff_layout_pg_init_read,
1004 	.pg_test = pnfs_generic_pg_test,
1005 	.pg_doio = pnfs_generic_pg_readpages,
1006 	.pg_cleanup = pnfs_generic_pg_cleanup,
1007 };
1008 
1009 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1010 	.pg_init = ff_layout_pg_init_write,
1011 	.pg_test = pnfs_generic_pg_test,
1012 	.pg_doio = pnfs_generic_pg_writepages,
1013 	.pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1014 	.pg_cleanup = pnfs_generic_pg_cleanup,
1015 	.pg_get_mirror = ff_layout_pg_get_mirror_write,
1016 	.pg_set_mirror = ff_layout_pg_set_mirror_write,
1017 };
1018 
1019 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1020 {
1021 	struct rpc_task *task = &hdr->task;
1022 
1023 	pnfs_layoutcommit_inode(hdr->inode, false);
1024 
1025 	if (retry_pnfs) {
1026 		dprintk("%s Reset task %5u for i/o through pNFS "
1027 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1028 			hdr->task.tk_pid,
1029 			hdr->inode->i_sb->s_id,
1030 			(unsigned long long)NFS_FILEID(hdr->inode),
1031 			hdr->args.count,
1032 			(unsigned long long)hdr->args.offset);
1033 
1034 		hdr->completion_ops->reschedule_io(hdr);
1035 		return;
1036 	}
1037 
1038 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1039 		dprintk("%s Reset task %5u for i/o through MDS "
1040 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1041 			hdr->task.tk_pid,
1042 			hdr->inode->i_sb->s_id,
1043 			(unsigned long long)NFS_FILEID(hdr->inode),
1044 			hdr->args.count,
1045 			(unsigned long long)hdr->args.offset);
1046 
1047 		trace_pnfs_mds_fallback_write_done(hdr->inode,
1048 				hdr->args.offset, hdr->args.count,
1049 				IOMODE_RW, NFS_I(hdr->inode)->layout,
1050 				hdr->lseg);
1051 		task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1052 	}
1053 }
1054 
1055 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1056 {
1057 	u32 idx = hdr->pgio_mirror_idx + 1;
1058 	u32 new_idx = 0;
1059 
1060 	if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx))
1061 		ff_layout_send_layouterror(hdr->lseg);
1062 	else
1063 		pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1064 	pnfs_read_resend_pnfs(hdr, new_idx);
1065 }
1066 
1067 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1068 {
1069 	struct rpc_task *task = &hdr->task;
1070 
1071 	pnfs_layoutcommit_inode(hdr->inode, false);
1072 	pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1073 
1074 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1075 		dprintk("%s Reset task %5u for i/o through MDS "
1076 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1077 			hdr->task.tk_pid,
1078 			hdr->inode->i_sb->s_id,
1079 			(unsigned long long)NFS_FILEID(hdr->inode),
1080 			hdr->args.count,
1081 			(unsigned long long)hdr->args.offset);
1082 
1083 		trace_pnfs_mds_fallback_read_done(hdr->inode,
1084 				hdr->args.offset, hdr->args.count,
1085 				IOMODE_READ, NFS_I(hdr->inode)->layout,
1086 				hdr->lseg);
1087 		task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1088 	}
1089 }
1090 
1091 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1092 					   struct nfs4_state *state,
1093 					   struct nfs_client *clp,
1094 					   struct pnfs_layout_segment *lseg,
1095 					   u32 idx)
1096 {
1097 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
1098 	struct inode *inode = lo->plh_inode;
1099 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1100 	struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1101 
1102 	switch (task->tk_status) {
1103 	case -NFS4ERR_BADSESSION:
1104 	case -NFS4ERR_BADSLOT:
1105 	case -NFS4ERR_BAD_HIGH_SLOT:
1106 	case -NFS4ERR_DEADSESSION:
1107 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1108 	case -NFS4ERR_SEQ_FALSE_RETRY:
1109 	case -NFS4ERR_SEQ_MISORDERED:
1110 		dprintk("%s ERROR %d, Reset session. Exchangeid "
1111 			"flags 0x%x\n", __func__, task->tk_status,
1112 			clp->cl_exchange_flags);
1113 		nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1114 		break;
1115 	case -NFS4ERR_DELAY:
1116 	case -NFS4ERR_GRACE:
1117 		rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1118 		break;
1119 	case -NFS4ERR_RETRY_UNCACHED_REP:
1120 		break;
1121 	/* Invalidate Layout errors */
1122 	case -NFS4ERR_PNFS_NO_LAYOUT:
1123 	case -ESTALE:           /* mapped NFS4ERR_STALE */
1124 	case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1125 	case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1126 	case -NFS4ERR_FHEXPIRED:
1127 	case -NFS4ERR_WRONG_TYPE:
1128 		dprintk("%s Invalid layout error %d\n", __func__,
1129 			task->tk_status);
1130 		/*
1131 		 * Destroy layout so new i/o will get a new layout.
1132 		 * Layout will not be destroyed until all current lseg
1133 		 * references are put. Mark layout as invalid to resend failed
1134 		 * i/o and all i/o waiting on the slot table to the MDS until
1135 		 * layout is destroyed and a new valid layout is obtained.
1136 		 */
1137 		pnfs_destroy_layout(NFS_I(inode));
1138 		rpc_wake_up(&tbl->slot_tbl_waitq);
1139 		goto reset;
1140 	/* RPC connection errors */
1141 	case -ECONNREFUSED:
1142 	case -EHOSTDOWN:
1143 	case -EHOSTUNREACH:
1144 	case -ENETUNREACH:
1145 	case -EIO:
1146 	case -ETIMEDOUT:
1147 	case -EPIPE:
1148 	case -EPROTO:
1149 	case -ENODEV:
1150 		dprintk("%s DS connection error %d\n", __func__,
1151 			task->tk_status);
1152 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1153 				&devid->deviceid);
1154 		rpc_wake_up(&tbl->slot_tbl_waitq);
1155 		fallthrough;
1156 	default:
1157 		if (ff_layout_avoid_mds_available_ds(lseg))
1158 			return -NFS4ERR_RESET_TO_PNFS;
1159 reset:
1160 		dprintk("%s Retry through MDS. Error %d\n", __func__,
1161 			task->tk_status);
1162 		return -NFS4ERR_RESET_TO_MDS;
1163 	}
1164 	task->tk_status = 0;
1165 	return -EAGAIN;
1166 }
1167 
1168 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1169 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1170 					   struct pnfs_layout_segment *lseg,
1171 					   u32 idx)
1172 {
1173 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1174 
1175 	switch (task->tk_status) {
1176 	/* File access problems. Don't mark the device as unavailable */
1177 	case -EACCES:
1178 	case -ESTALE:
1179 	case -EISDIR:
1180 	case -EBADHANDLE:
1181 	case -ELOOP:
1182 	case -ENOSPC:
1183 		break;
1184 	case -EJUKEBOX:
1185 		nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1186 		goto out_retry;
1187 	default:
1188 		dprintk("%s DS connection error %d\n", __func__,
1189 			task->tk_status);
1190 		nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1191 				&devid->deviceid);
1192 	}
1193 	/* FIXME: Need to prevent infinite looping here. */
1194 	return -NFS4ERR_RESET_TO_PNFS;
1195 out_retry:
1196 	task->tk_status = 0;
1197 	rpc_restart_call_prepare(task);
1198 	rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1199 	return -EAGAIN;
1200 }
1201 
1202 static int ff_layout_async_handle_error(struct rpc_task *task,
1203 					struct nfs4_state *state,
1204 					struct nfs_client *clp,
1205 					struct pnfs_layout_segment *lseg,
1206 					u32 idx)
1207 {
1208 	int vers = clp->cl_nfs_mod->rpc_vers->number;
1209 
1210 	if (task->tk_status >= 0) {
1211 		ff_layout_mark_ds_reachable(lseg, idx);
1212 		return 0;
1213 	}
1214 
1215 	/* Handle the case of an invalid layout segment */
1216 	if (!pnfs_is_valid_lseg(lseg))
1217 		return -NFS4ERR_RESET_TO_PNFS;
1218 
1219 	switch (vers) {
1220 	case 3:
1221 		return ff_layout_async_handle_error_v3(task, lseg, idx);
1222 	case 4:
1223 		return ff_layout_async_handle_error_v4(task, state, clp,
1224 						       lseg, idx);
1225 	default:
1226 		/* should never happen */
1227 		WARN_ON_ONCE(1);
1228 		return 0;
1229 	}
1230 }
1231 
1232 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1233 					u32 idx, u64 offset, u64 length,
1234 					u32 *op_status, int opnum, int error)
1235 {
1236 	struct nfs4_ff_layout_mirror *mirror;
1237 	u32 status = *op_status;
1238 	int err;
1239 
1240 	if (status == 0) {
1241 		switch (error) {
1242 		case -ETIMEDOUT:
1243 		case -EPFNOSUPPORT:
1244 		case -EPROTONOSUPPORT:
1245 		case -EOPNOTSUPP:
1246 		case -EINVAL:
1247 		case -ECONNREFUSED:
1248 		case -ECONNRESET:
1249 		case -EHOSTDOWN:
1250 		case -EHOSTUNREACH:
1251 		case -ENETUNREACH:
1252 		case -EADDRINUSE:
1253 		case -ENOBUFS:
1254 		case -EPIPE:
1255 		case -EPERM:
1256 		case -EPROTO:
1257 		case -ENODEV:
1258 			*op_status = status = NFS4ERR_NXIO;
1259 			break;
1260 		case -EACCES:
1261 			*op_status = status = NFS4ERR_ACCESS;
1262 			break;
1263 		default:
1264 			return;
1265 		}
1266 	}
1267 
1268 	mirror = FF_LAYOUT_COMP(lseg, idx);
1269 	err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1270 				       mirror, offset, length, status, opnum,
1271 				       nfs_io_gfp_mask());
1272 
1273 	switch (status) {
1274 	case NFS4ERR_DELAY:
1275 	case NFS4ERR_GRACE:
1276 		break;
1277 	case NFS4ERR_NXIO:
1278 		ff_layout_mark_ds_unreachable(lseg, idx);
1279 		/*
1280 		 * Don't return the layout if this is a read and we still
1281 		 * have layouts to try
1282 		 */
1283 		if (opnum == OP_READ)
1284 			break;
1285 		fallthrough;
1286 	default:
1287 		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1288 						  lseg);
1289 	}
1290 
1291 	dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1292 }
1293 
1294 /* NFS_PROTO call done callback routines */
1295 static int ff_layout_read_done_cb(struct rpc_task *task,
1296 				struct nfs_pgio_header *hdr)
1297 {
1298 	int err;
1299 
1300 	if (task->tk_status < 0) {
1301 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1302 					    hdr->args.offset, hdr->args.count,
1303 					    &hdr->res.op_status, OP_READ,
1304 					    task->tk_status);
1305 		trace_ff_layout_read_error(hdr);
1306 	}
1307 
1308 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1309 					   hdr->ds_clp, hdr->lseg,
1310 					   hdr->pgio_mirror_idx);
1311 
1312 	trace_nfs4_pnfs_read(hdr, err);
1313 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1314 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1315 	switch (err) {
1316 	case -NFS4ERR_RESET_TO_PNFS:
1317 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1318 		return task->tk_status;
1319 	case -NFS4ERR_RESET_TO_MDS:
1320 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1321 		return task->tk_status;
1322 	case -EAGAIN:
1323 		goto out_eagain;
1324 	}
1325 
1326 	return 0;
1327 out_eagain:
1328 	rpc_restart_call_prepare(task);
1329 	return -EAGAIN;
1330 }
1331 
1332 static bool
1333 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1334 {
1335 	return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1336 }
1337 
1338 /*
1339  * We reference the rpc_cred of the first WRITE that triggers the need for
1340  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1341  * rfc5661 is not clear about which credential should be used.
1342  *
1343  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1344  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1345  * we always send layoutcommit after DS writes.
1346  */
1347 static void
1348 ff_layout_set_layoutcommit(struct inode *inode,
1349 		struct pnfs_layout_segment *lseg,
1350 		loff_t end_offset)
1351 {
1352 	if (!ff_layout_need_layoutcommit(lseg))
1353 		return;
1354 
1355 	pnfs_set_layoutcommit(inode, lseg, end_offset);
1356 	dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1357 		(unsigned long long) NFS_I(inode)->layout->plh_lwb);
1358 }
1359 
1360 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1361 		struct nfs_pgio_header *hdr)
1362 {
1363 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1364 		return;
1365 	nfs4_ff_layout_stat_io_start_read(hdr->inode,
1366 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1367 			hdr->args.count,
1368 			task->tk_start);
1369 }
1370 
1371 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1372 		struct nfs_pgio_header *hdr)
1373 {
1374 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1375 		return;
1376 	nfs4_ff_layout_stat_io_end_read(task,
1377 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1378 			hdr->args.count,
1379 			hdr->res.count);
1380 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1381 }
1382 
1383 static int ff_layout_read_prepare_common(struct rpc_task *task,
1384 					 struct nfs_pgio_header *hdr)
1385 {
1386 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1387 		rpc_exit(task, -EIO);
1388 		return -EIO;
1389 	}
1390 
1391 	if (!pnfs_is_valid_lseg(hdr->lseg)) {
1392 		rpc_exit(task, -EAGAIN);
1393 		return -EAGAIN;
1394 	}
1395 
1396 	ff_layout_read_record_layoutstats_start(task, hdr);
1397 	return 0;
1398 }
1399 
1400 /*
1401  * Call ops for the async read/write cases
1402  * In the case of dense layouts, the offset needs to be reset to its
1403  * original value.
1404  */
1405 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1406 {
1407 	struct nfs_pgio_header *hdr = data;
1408 
1409 	if (ff_layout_read_prepare_common(task, hdr))
1410 		return;
1411 
1412 	rpc_call_start(task);
1413 }
1414 
1415 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1416 {
1417 	struct nfs_pgio_header *hdr = data;
1418 
1419 	if (nfs4_setup_sequence(hdr->ds_clp,
1420 				&hdr->args.seq_args,
1421 				&hdr->res.seq_res,
1422 				task))
1423 		return;
1424 
1425 	ff_layout_read_prepare_common(task, hdr);
1426 }
1427 
1428 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1429 {
1430 	struct nfs_pgio_header *hdr = data;
1431 
1432 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1433 	    task->tk_status == 0) {
1434 		nfs4_sequence_done(task, &hdr->res.seq_res);
1435 		return;
1436 	}
1437 
1438 	/* Note this may cause RPC to be resent */
1439 	hdr->mds_ops->rpc_call_done(task, hdr);
1440 }
1441 
1442 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1443 {
1444 	struct nfs_pgio_header *hdr = data;
1445 
1446 	ff_layout_read_record_layoutstats_done(task, hdr);
1447 	rpc_count_iostats_metrics(task,
1448 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1449 }
1450 
1451 static void ff_layout_read_release(void *data)
1452 {
1453 	struct nfs_pgio_header *hdr = data;
1454 
1455 	ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1456 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1457 		ff_layout_resend_pnfs_read(hdr);
1458 	else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1459 		ff_layout_reset_read(hdr);
1460 	pnfs_generic_rw_release(data);
1461 }
1462 
1463 
1464 static int ff_layout_write_done_cb(struct rpc_task *task,
1465 				struct nfs_pgio_header *hdr)
1466 {
1467 	loff_t end_offs = 0;
1468 	int err;
1469 
1470 	if (task->tk_status < 0) {
1471 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1472 					    hdr->args.offset, hdr->args.count,
1473 					    &hdr->res.op_status, OP_WRITE,
1474 					    task->tk_status);
1475 		trace_ff_layout_write_error(hdr);
1476 	}
1477 
1478 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1479 					   hdr->ds_clp, hdr->lseg,
1480 					   hdr->pgio_mirror_idx);
1481 
1482 	trace_nfs4_pnfs_write(hdr, err);
1483 	clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1484 	clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1485 	switch (err) {
1486 	case -NFS4ERR_RESET_TO_PNFS:
1487 		set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1488 		return task->tk_status;
1489 	case -NFS4ERR_RESET_TO_MDS:
1490 		set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1491 		return task->tk_status;
1492 	case -EAGAIN:
1493 		return -EAGAIN;
1494 	}
1495 
1496 	if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1497 	    hdr->res.verf->committed == NFS_DATA_SYNC)
1498 		end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1499 
1500 	/* Note: if the write is unstable, don't set end_offs until commit */
1501 	ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1502 
1503 	/* zero out fattr since we don't care DS attr at all */
1504 	hdr->fattr.valid = 0;
1505 	if (task->tk_status >= 0)
1506 		nfs_writeback_update_inode(hdr);
1507 
1508 	return 0;
1509 }
1510 
1511 static int ff_layout_commit_done_cb(struct rpc_task *task,
1512 				     struct nfs_commit_data *data)
1513 {
1514 	int err;
1515 
1516 	if (task->tk_status < 0) {
1517 		ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1518 					    data->args.offset, data->args.count,
1519 					    &data->res.op_status, OP_COMMIT,
1520 					    task->tk_status);
1521 		trace_ff_layout_commit_error(data);
1522 	}
1523 
1524 	err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1525 					   data->lseg, data->ds_commit_index);
1526 
1527 	trace_nfs4_pnfs_commit_ds(data, err);
1528 	switch (err) {
1529 	case -NFS4ERR_RESET_TO_PNFS:
1530 		pnfs_generic_prepare_to_resend_writes(data);
1531 		return -EAGAIN;
1532 	case -NFS4ERR_RESET_TO_MDS:
1533 		pnfs_generic_prepare_to_resend_writes(data);
1534 		return -EAGAIN;
1535 	case -EAGAIN:
1536 		rpc_restart_call_prepare(task);
1537 		return -EAGAIN;
1538 	}
1539 
1540 	ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1541 
1542 	return 0;
1543 }
1544 
1545 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1546 		struct nfs_pgio_header *hdr)
1547 {
1548 	if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1549 		return;
1550 	nfs4_ff_layout_stat_io_start_write(hdr->inode,
1551 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1552 			hdr->args.count,
1553 			task->tk_start);
1554 }
1555 
1556 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1557 		struct nfs_pgio_header *hdr)
1558 {
1559 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1560 		return;
1561 	nfs4_ff_layout_stat_io_end_write(task,
1562 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1563 			hdr->args.count, hdr->res.count,
1564 			hdr->res.verf->committed);
1565 	set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1566 }
1567 
1568 static int ff_layout_write_prepare_common(struct rpc_task *task,
1569 					  struct nfs_pgio_header *hdr)
1570 {
1571 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1572 		rpc_exit(task, -EIO);
1573 		return -EIO;
1574 	}
1575 
1576 	if (!pnfs_is_valid_lseg(hdr->lseg)) {
1577 		rpc_exit(task, -EAGAIN);
1578 		return -EAGAIN;
1579 	}
1580 
1581 	ff_layout_write_record_layoutstats_start(task, hdr);
1582 	return 0;
1583 }
1584 
1585 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1586 {
1587 	struct nfs_pgio_header *hdr = data;
1588 
1589 	if (ff_layout_write_prepare_common(task, hdr))
1590 		return;
1591 
1592 	rpc_call_start(task);
1593 }
1594 
1595 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1596 {
1597 	struct nfs_pgio_header *hdr = data;
1598 
1599 	if (nfs4_setup_sequence(hdr->ds_clp,
1600 				&hdr->args.seq_args,
1601 				&hdr->res.seq_res,
1602 				task))
1603 		return;
1604 
1605 	ff_layout_write_prepare_common(task, hdr);
1606 }
1607 
1608 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1609 {
1610 	struct nfs_pgio_header *hdr = data;
1611 
1612 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1613 	    task->tk_status == 0) {
1614 		nfs4_sequence_done(task, &hdr->res.seq_res);
1615 		return;
1616 	}
1617 
1618 	/* Note this may cause RPC to be resent */
1619 	hdr->mds_ops->rpc_call_done(task, hdr);
1620 }
1621 
1622 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1623 {
1624 	struct nfs_pgio_header *hdr = data;
1625 
1626 	ff_layout_write_record_layoutstats_done(task, hdr);
1627 	rpc_count_iostats_metrics(task,
1628 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1629 }
1630 
1631 static void ff_layout_write_release(void *data)
1632 {
1633 	struct nfs_pgio_header *hdr = data;
1634 
1635 	ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1636 	if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1637 		ff_layout_send_layouterror(hdr->lseg);
1638 		ff_layout_reset_write(hdr, true);
1639 	} else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1640 		ff_layout_reset_write(hdr, false);
1641 	pnfs_generic_rw_release(data);
1642 }
1643 
1644 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1645 		struct nfs_commit_data *cdata)
1646 {
1647 	if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1648 		return;
1649 	nfs4_ff_layout_stat_io_start_write(cdata->inode,
1650 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1651 			0, task->tk_start);
1652 }
1653 
1654 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1655 		struct nfs_commit_data *cdata)
1656 {
1657 	struct nfs_page *req;
1658 	__u64 count = 0;
1659 
1660 	if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1661 		return;
1662 
1663 	if (task->tk_status == 0) {
1664 		list_for_each_entry(req, &cdata->pages, wb_list)
1665 			count += req->wb_bytes;
1666 	}
1667 	nfs4_ff_layout_stat_io_end_write(task,
1668 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1669 			count, count, NFS_FILE_SYNC);
1670 	set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1671 }
1672 
1673 static int ff_layout_commit_prepare_common(struct rpc_task *task,
1674 					   struct nfs_commit_data *cdata)
1675 {
1676 	if (!pnfs_is_valid_lseg(cdata->lseg)) {
1677 		rpc_exit(task, -EAGAIN);
1678 		return -EAGAIN;
1679 	}
1680 
1681 	ff_layout_commit_record_layoutstats_start(task, cdata);
1682 	return 0;
1683 }
1684 
1685 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1686 {
1687 	if (ff_layout_commit_prepare_common(task, data))
1688 		return;
1689 
1690 	rpc_call_start(task);
1691 }
1692 
1693 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1694 {
1695 	struct nfs_commit_data *wdata = data;
1696 
1697 	if (nfs4_setup_sequence(wdata->ds_clp,
1698 				&wdata->args.seq_args,
1699 				&wdata->res.seq_res,
1700 				task))
1701 		return;
1702 	ff_layout_commit_prepare_common(task, data);
1703 }
1704 
1705 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1706 {
1707 	pnfs_generic_write_commit_done(task, data);
1708 }
1709 
1710 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1711 {
1712 	struct nfs_commit_data *cdata = data;
1713 
1714 	ff_layout_commit_record_layoutstats_done(task, cdata);
1715 	rpc_count_iostats_metrics(task,
1716 	    &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1717 }
1718 
1719 static void ff_layout_commit_release(void *data)
1720 {
1721 	struct nfs_commit_data *cdata = data;
1722 
1723 	ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1724 	pnfs_generic_commit_release(data);
1725 }
1726 
1727 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1728 	.rpc_call_prepare = ff_layout_read_prepare_v3,
1729 	.rpc_call_done = ff_layout_read_call_done,
1730 	.rpc_count_stats = ff_layout_read_count_stats,
1731 	.rpc_release = ff_layout_read_release,
1732 };
1733 
1734 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1735 	.rpc_call_prepare = ff_layout_read_prepare_v4,
1736 	.rpc_call_done = ff_layout_read_call_done,
1737 	.rpc_count_stats = ff_layout_read_count_stats,
1738 	.rpc_release = ff_layout_read_release,
1739 };
1740 
1741 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1742 	.rpc_call_prepare = ff_layout_write_prepare_v3,
1743 	.rpc_call_done = ff_layout_write_call_done,
1744 	.rpc_count_stats = ff_layout_write_count_stats,
1745 	.rpc_release = ff_layout_write_release,
1746 };
1747 
1748 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1749 	.rpc_call_prepare = ff_layout_write_prepare_v4,
1750 	.rpc_call_done = ff_layout_write_call_done,
1751 	.rpc_count_stats = ff_layout_write_count_stats,
1752 	.rpc_release = ff_layout_write_release,
1753 };
1754 
1755 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1756 	.rpc_call_prepare = ff_layout_commit_prepare_v3,
1757 	.rpc_call_done = ff_layout_commit_done,
1758 	.rpc_count_stats = ff_layout_commit_count_stats,
1759 	.rpc_release = ff_layout_commit_release,
1760 };
1761 
1762 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1763 	.rpc_call_prepare = ff_layout_commit_prepare_v4,
1764 	.rpc_call_done = ff_layout_commit_done,
1765 	.rpc_count_stats = ff_layout_commit_count_stats,
1766 	.rpc_release = ff_layout_commit_release,
1767 };
1768 
1769 static enum pnfs_try_status
1770 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1771 {
1772 	struct pnfs_layout_segment *lseg = hdr->lseg;
1773 	struct nfs4_pnfs_ds *ds;
1774 	struct rpc_clnt *ds_clnt;
1775 	struct nfsd_file *localio;
1776 	struct nfs4_ff_layout_mirror *mirror;
1777 	const struct cred *ds_cred;
1778 	loff_t offset = hdr->args.offset;
1779 	u32 idx = hdr->pgio_mirror_idx;
1780 	int vers;
1781 	struct nfs_fh *fh;
1782 
1783 	dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1784 		__func__, hdr->inode->i_ino,
1785 		hdr->args.pgbase, (size_t)hdr->args.count, offset);
1786 
1787 	mirror = FF_LAYOUT_COMP(lseg, idx);
1788 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1789 	if (!ds)
1790 		goto out_failed;
1791 
1792 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1793 						   hdr->inode);
1794 	if (IS_ERR(ds_clnt))
1795 		goto out_failed;
1796 
1797 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1798 	if (!ds_cred)
1799 		goto out_failed;
1800 
1801 	vers = nfs4_ff_layout_ds_version(mirror);
1802 
1803 	dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1804 		ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1805 
1806 	hdr->pgio_done_cb = ff_layout_read_done_cb;
1807 	refcount_inc(&ds->ds_clp->cl_count);
1808 	hdr->ds_clp = ds->ds_clp;
1809 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1810 	if (fh)
1811 		hdr->args.fh = fh;
1812 
1813 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1814 
1815 	/*
1816 	 * Note that if we ever decide to split across DSes,
1817 	 * then we may need to handle dense-like offsets.
1818 	 */
1819 	hdr->args.offset = offset;
1820 	hdr->mds_offset = offset;
1821 
1822 	/* Start IO accounting for local read */
1823 	localio = ff_local_open_fh(ds->ds_clp, ds_cred, fh, FMODE_READ);
1824 	if (localio) {
1825 		hdr->task.tk_start = ktime_get();
1826 		ff_layout_read_record_layoutstats_start(&hdr->task, hdr);
1827 	}
1828 
1829 	/* Perform an asynchronous read to ds */
1830 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1831 			  vers == 3 ? &ff_layout_read_call_ops_v3 :
1832 				      &ff_layout_read_call_ops_v4,
1833 			  0, RPC_TASK_SOFTCONN, localio);
1834 	put_cred(ds_cred);
1835 	return PNFS_ATTEMPTED;
1836 
1837 out_failed:
1838 	if (ff_layout_avoid_mds_available_ds(lseg))
1839 		return PNFS_TRY_AGAIN;
1840 	trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1841 			hdr->args.offset, hdr->args.count,
1842 			IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1843 	return PNFS_NOT_ATTEMPTED;
1844 }
1845 
1846 /* Perform async writes. */
1847 static enum pnfs_try_status
1848 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1849 {
1850 	struct pnfs_layout_segment *lseg = hdr->lseg;
1851 	struct nfs4_pnfs_ds *ds;
1852 	struct rpc_clnt *ds_clnt;
1853 	struct nfsd_file *localio;
1854 	struct nfs4_ff_layout_mirror *mirror;
1855 	const struct cred *ds_cred;
1856 	loff_t offset = hdr->args.offset;
1857 	int vers;
1858 	struct nfs_fh *fh;
1859 	u32 idx = hdr->pgio_mirror_idx;
1860 
1861 	mirror = FF_LAYOUT_COMP(lseg, idx);
1862 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1863 	if (!ds)
1864 		goto out_failed;
1865 
1866 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1867 						   hdr->inode);
1868 	if (IS_ERR(ds_clnt))
1869 		goto out_failed;
1870 
1871 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1872 	if (!ds_cred)
1873 		goto out_failed;
1874 
1875 	vers = nfs4_ff_layout_ds_version(mirror);
1876 
1877 	dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1878 		__func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1879 		offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1880 		vers);
1881 
1882 	hdr->pgio_done_cb = ff_layout_write_done_cb;
1883 	refcount_inc(&ds->ds_clp->cl_count);
1884 	hdr->ds_clp = ds->ds_clp;
1885 	hdr->ds_commit_idx = idx;
1886 	fh = nfs4_ff_layout_select_ds_fh(mirror);
1887 	if (fh)
1888 		hdr->args.fh = fh;
1889 
1890 	nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1891 
1892 	/*
1893 	 * Note that if we ever decide to split across DSes,
1894 	 * then we may need to handle dense-like offsets.
1895 	 */
1896 	hdr->args.offset = offset;
1897 
1898 	/* Start IO accounting for local write */
1899 	localio = ff_local_open_fh(ds->ds_clp, ds_cred, fh,
1900 				   FMODE_READ|FMODE_WRITE);
1901 	if (localio) {
1902 		hdr->task.tk_start = ktime_get();
1903 		ff_layout_write_record_layoutstats_start(&hdr->task, hdr);
1904 	}
1905 
1906 	/* Perform an asynchronous write */
1907 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1908 			  vers == 3 ? &ff_layout_write_call_ops_v3 :
1909 				      &ff_layout_write_call_ops_v4,
1910 			  sync, RPC_TASK_SOFTCONN, localio);
1911 	put_cred(ds_cred);
1912 	return PNFS_ATTEMPTED;
1913 
1914 out_failed:
1915 	if (ff_layout_avoid_mds_available_ds(lseg))
1916 		return PNFS_TRY_AGAIN;
1917 	trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1918 			hdr->args.offset, hdr->args.count,
1919 			IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1920 	return PNFS_NOT_ATTEMPTED;
1921 }
1922 
1923 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1924 {
1925 	return i;
1926 }
1927 
1928 static struct nfs_fh *
1929 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1930 {
1931 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1932 
1933 	/* FIXME: Assume that there is only one NFS version available
1934 	 * for the DS.
1935 	 */
1936 	return &flseg->mirror_array[i]->fh_versions[0];
1937 }
1938 
1939 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1940 {
1941 	struct pnfs_layout_segment *lseg = data->lseg;
1942 	struct nfs4_pnfs_ds *ds;
1943 	struct rpc_clnt *ds_clnt;
1944 	struct nfsd_file *localio;
1945 	struct nfs4_ff_layout_mirror *mirror;
1946 	const struct cred *ds_cred;
1947 	u32 idx;
1948 	int vers, ret;
1949 	struct nfs_fh *fh;
1950 
1951 	if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1952 	    test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1953 		goto out_err;
1954 
1955 	idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1956 	mirror = FF_LAYOUT_COMP(lseg, idx);
1957 	ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1958 	if (!ds)
1959 		goto out_err;
1960 
1961 	ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1962 						   data->inode);
1963 	if (IS_ERR(ds_clnt))
1964 		goto out_err;
1965 
1966 	ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1967 	if (!ds_cred)
1968 		goto out_err;
1969 
1970 	vers = nfs4_ff_layout_ds_version(mirror);
1971 
1972 	dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1973 		data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1974 		vers);
1975 	data->commit_done_cb = ff_layout_commit_done_cb;
1976 	data->cred = ds_cred;
1977 	refcount_inc(&ds->ds_clp->cl_count);
1978 	data->ds_clp = ds->ds_clp;
1979 	fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1980 	if (fh)
1981 		data->args.fh = fh;
1982 
1983 	/* Start IO accounting for local commit */
1984 	localio = ff_local_open_fh(ds->ds_clp, ds_cred, fh,
1985 				   FMODE_READ|FMODE_WRITE);
1986 	if (localio) {
1987 		data->task.tk_start = ktime_get();
1988 		ff_layout_commit_record_layoutstats_start(&data->task, data);
1989 	}
1990 
1991 	ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1992 				   vers == 3 ? &ff_layout_commit_call_ops_v3 :
1993 					       &ff_layout_commit_call_ops_v4,
1994 				   how, RPC_TASK_SOFTCONN, localio);
1995 	put_cred(ds_cred);
1996 	return ret;
1997 out_err:
1998 	pnfs_generic_prepare_to_resend_writes(data);
1999 	pnfs_generic_commit_release(data);
2000 	return -EAGAIN;
2001 }
2002 
2003 static int
2004 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
2005 			   int how, struct nfs_commit_info *cinfo)
2006 {
2007 	return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
2008 					    ff_layout_initiate_commit);
2009 }
2010 
2011 static bool ff_layout_match_rw(const struct rpc_task *task,
2012 			       const struct nfs_pgio_header *hdr,
2013 			       const struct pnfs_layout_segment *lseg)
2014 {
2015 	return hdr->lseg == lseg;
2016 }
2017 
2018 static bool ff_layout_match_commit(const struct rpc_task *task,
2019 				   const struct nfs_commit_data *cdata,
2020 				   const struct pnfs_layout_segment *lseg)
2021 {
2022 	return cdata->lseg == lseg;
2023 }
2024 
2025 static bool ff_layout_match_io(const struct rpc_task *task, const void *data)
2026 {
2027 	const struct rpc_call_ops *ops = task->tk_ops;
2028 
2029 	if (ops == &ff_layout_read_call_ops_v3 ||
2030 	    ops == &ff_layout_read_call_ops_v4 ||
2031 	    ops == &ff_layout_write_call_ops_v3 ||
2032 	    ops == &ff_layout_write_call_ops_v4)
2033 		return ff_layout_match_rw(task, task->tk_calldata, data);
2034 	if (ops == &ff_layout_commit_call_ops_v3 ||
2035 	    ops == &ff_layout_commit_call_ops_v4)
2036 		return ff_layout_match_commit(task, task->tk_calldata, data);
2037 	return false;
2038 }
2039 
2040 static void ff_layout_cancel_io(struct pnfs_layout_segment *lseg)
2041 {
2042 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
2043 	struct nfs4_ff_layout_mirror *mirror;
2044 	struct nfs4_ff_layout_ds *mirror_ds;
2045 	struct nfs4_pnfs_ds *ds;
2046 	struct nfs_client *ds_clp;
2047 	struct rpc_clnt *clnt;
2048 	u32 idx;
2049 
2050 	for (idx = 0; idx < flseg->mirror_array_cnt; idx++) {
2051 		mirror = flseg->mirror_array[idx];
2052 		mirror_ds = mirror->mirror_ds;
2053 		if (IS_ERR_OR_NULL(mirror_ds))
2054 			continue;
2055 		ds = mirror->mirror_ds->ds;
2056 		if (!ds)
2057 			continue;
2058 		ds_clp = ds->ds_clp;
2059 		if (!ds_clp)
2060 			continue;
2061 		clnt = ds_clp->cl_rpcclient;
2062 		if (!clnt)
2063 			continue;
2064 		if (!rpc_cancel_tasks(clnt, -EAGAIN, ff_layout_match_io, lseg))
2065 			continue;
2066 		rpc_clnt_disconnect(clnt);
2067 	}
2068 }
2069 
2070 static struct pnfs_ds_commit_info *
2071 ff_layout_get_ds_info(struct inode *inode)
2072 {
2073 	struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
2074 
2075 	if (layout == NULL)
2076 		return NULL;
2077 
2078 	return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2079 }
2080 
2081 static void
2082 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
2083 		struct pnfs_layout_segment *lseg)
2084 {
2085 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
2086 	struct inode *inode = lseg->pls_layout->plh_inode;
2087 	struct pnfs_commit_array *array, *new;
2088 
2089 	new = pnfs_alloc_commit_array(flseg->mirror_array_cnt,
2090 				      nfs_io_gfp_mask());
2091 	if (new) {
2092 		spin_lock(&inode->i_lock);
2093 		array = pnfs_add_commit_array(fl_cinfo, new, lseg);
2094 		spin_unlock(&inode->i_lock);
2095 		if (array != new)
2096 			pnfs_free_commit_array(new);
2097 	}
2098 }
2099 
2100 static void
2101 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
2102 		struct inode *inode)
2103 {
2104 	spin_lock(&inode->i_lock);
2105 	pnfs_generic_ds_cinfo_destroy(fl_cinfo);
2106 	spin_unlock(&inode->i_lock);
2107 }
2108 
2109 static void
2110 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2111 {
2112 	nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2113 						  id_node));
2114 }
2115 
2116 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2117 				  const struct nfs4_layoutreturn_args *args,
2118 				  const struct nfs4_flexfile_layoutreturn_args *ff_args)
2119 {
2120 	__be32 *start;
2121 
2122 	start = xdr_reserve_space(xdr, 4);
2123 	if (unlikely(!start))
2124 		return -E2BIG;
2125 
2126 	*start = cpu_to_be32(ff_args->num_errors);
2127 	/* This assume we always return _ALL_ layouts */
2128 	return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2129 }
2130 
2131 static void
2132 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2133 			    const nfs4_stateid *stateid,
2134 			    const struct nfs42_layoutstat_devinfo *devinfo)
2135 {
2136 	__be32 *p;
2137 
2138 	p = xdr_reserve_space(xdr, 8 + 8);
2139 	p = xdr_encode_hyper(p, devinfo->offset);
2140 	p = xdr_encode_hyper(p, devinfo->length);
2141 	encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2142 	p = xdr_reserve_space(xdr, 4*8);
2143 	p = xdr_encode_hyper(p, devinfo->read_count);
2144 	p = xdr_encode_hyper(p, devinfo->read_bytes);
2145 	p = xdr_encode_hyper(p, devinfo->write_count);
2146 	p = xdr_encode_hyper(p, devinfo->write_bytes);
2147 	encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2148 }
2149 
2150 static void
2151 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2152 			    const nfs4_stateid *stateid,
2153 			    const struct nfs42_layoutstat_devinfo *devinfo)
2154 {
2155 	ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2156 	ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2157 			devinfo->ld_private.data);
2158 }
2159 
2160 /* report nothing for now */
2161 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2162 		const struct nfs4_layoutreturn_args *args,
2163 		struct nfs4_flexfile_layoutreturn_args *ff_args)
2164 {
2165 	__be32 *p;
2166 	int i;
2167 
2168 	p = xdr_reserve_space(xdr, 4);
2169 	*p = cpu_to_be32(ff_args->num_dev);
2170 	for (i = 0; i < ff_args->num_dev; i++)
2171 		ff_layout_encode_ff_iostat(xdr,
2172 				&args->layout->plh_stateid,
2173 				&ff_args->devinfo[i]);
2174 }
2175 
2176 static void
2177 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2178 		unsigned int num_entries)
2179 {
2180 	unsigned int i;
2181 
2182 	for (i = 0; i < num_entries; i++) {
2183 		if (!devinfo[i].ld_private.ops)
2184 			continue;
2185 		if (!devinfo[i].ld_private.ops->free)
2186 			continue;
2187 		devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2188 	}
2189 }
2190 
2191 static struct nfs4_deviceid_node *
2192 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2193 			      struct pnfs_device *pdev, gfp_t gfp_flags)
2194 {
2195 	struct nfs4_ff_layout_ds *dsaddr;
2196 
2197 	dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2198 	if (!dsaddr)
2199 		return NULL;
2200 	return &dsaddr->id_node;
2201 }
2202 
2203 static void
2204 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2205 		const void *voidargs,
2206 		const struct nfs4_xdr_opaque_data *ff_opaque)
2207 {
2208 	const struct nfs4_layoutreturn_args *args = voidargs;
2209 	struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2210 	struct xdr_buf tmp_buf = {
2211 		.head = {
2212 			[0] = {
2213 				.iov_base = page_address(ff_args->pages[0]),
2214 			},
2215 		},
2216 		.buflen = PAGE_SIZE,
2217 	};
2218 	struct xdr_stream tmp_xdr;
2219 	__be32 *start;
2220 
2221 	dprintk("%s: Begin\n", __func__);
2222 
2223 	xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2224 
2225 	ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2226 	ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2227 
2228 	start = xdr_reserve_space(xdr, 4);
2229 	*start = cpu_to_be32(tmp_buf.len);
2230 	xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2231 
2232 	dprintk("%s: Return\n", __func__);
2233 }
2234 
2235 static void
2236 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2237 {
2238 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2239 
2240 	if (!args->data)
2241 		return;
2242 	ff_args = args->data;
2243 	args->data = NULL;
2244 
2245 	ff_layout_free_ds_ioerr(&ff_args->errors);
2246 	ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2247 
2248 	put_page(ff_args->pages[0]);
2249 	kfree(ff_args);
2250 }
2251 
2252 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2253 	.encode = ff_layout_encode_layoutreturn,
2254 	.free = ff_layout_free_layoutreturn,
2255 };
2256 
2257 static int
2258 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2259 {
2260 	struct nfs4_flexfile_layoutreturn_args *ff_args;
2261 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2262 
2263 	ff_args = kmalloc(sizeof(*ff_args), nfs_io_gfp_mask());
2264 	if (!ff_args)
2265 		goto out_nomem;
2266 	ff_args->pages[0] = alloc_page(nfs_io_gfp_mask());
2267 	if (!ff_args->pages[0])
2268 		goto out_nomem_free;
2269 
2270 	INIT_LIST_HEAD(&ff_args->errors);
2271 	ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2272 			&args->range, &ff_args->errors,
2273 			FF_LAYOUTRETURN_MAXERR);
2274 
2275 	spin_lock(&args->inode->i_lock);
2276 	ff_args->num_dev = ff_layout_mirror_prepare_stats(
2277 		&ff_layout->generic_hdr, &ff_args->devinfo[0],
2278 		ARRAY_SIZE(ff_args->devinfo), NFS4_FF_OP_LAYOUTRETURN);
2279 	spin_unlock(&args->inode->i_lock);
2280 
2281 	args->ld_private->ops = &layoutreturn_ops;
2282 	args->ld_private->data = ff_args;
2283 	return 0;
2284 out_nomem_free:
2285 	kfree(ff_args);
2286 out_nomem:
2287 	return -ENOMEM;
2288 }
2289 
2290 #ifdef CONFIG_NFS_V4_2
2291 void
2292 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2293 {
2294 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
2295 	struct nfs42_layout_error *errors;
2296 	LIST_HEAD(head);
2297 
2298 	if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2299 		return;
2300 	ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2301 	if (list_empty(&head))
2302 		return;
2303 
2304 	errors = kmalloc_array(NFS42_LAYOUTERROR_MAX, sizeof(*errors),
2305 			       nfs_io_gfp_mask());
2306 	if (errors != NULL) {
2307 		const struct nfs4_ff_layout_ds_err *pos;
2308 		size_t n = 0;
2309 
2310 		list_for_each_entry(pos, &head, list) {
2311 			errors[n].offset = pos->offset;
2312 			errors[n].length = pos->length;
2313 			nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2314 			errors[n].errors[0].dev_id = pos->deviceid;
2315 			errors[n].errors[0].status = pos->status;
2316 			errors[n].errors[0].opnum = pos->opnum;
2317 			n++;
2318 			if (!list_is_last(&pos->list, &head) &&
2319 			    n < NFS42_LAYOUTERROR_MAX)
2320 				continue;
2321 			if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2322 				break;
2323 			n = 0;
2324 		}
2325 		kfree(errors);
2326 	}
2327 	ff_layout_free_ds_ioerr(&head);
2328 }
2329 #else
2330 void
2331 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2332 {
2333 }
2334 #endif
2335 
2336 static int
2337 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2338 {
2339 	const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2340 
2341 	return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2342 }
2343 
2344 static size_t
2345 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2346 			  const int buflen)
2347 {
2348 	const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2349 	const struct in6_addr *addr = &sin6->sin6_addr;
2350 
2351 	/*
2352 	 * RFC 4291, Section 2.2.2
2353 	 *
2354 	 * Shorthanded ANY address
2355 	 */
2356 	if (ipv6_addr_any(addr))
2357 		return snprintf(buf, buflen, "::");
2358 
2359 	/*
2360 	 * RFC 4291, Section 2.2.2
2361 	 *
2362 	 * Shorthanded loopback address
2363 	 */
2364 	if (ipv6_addr_loopback(addr))
2365 		return snprintf(buf, buflen, "::1");
2366 
2367 	/*
2368 	 * RFC 4291, Section 2.2.3
2369 	 *
2370 	 * Special presentation address format for mapped v4
2371 	 * addresses.
2372 	 */
2373 	if (ipv6_addr_v4mapped(addr))
2374 		return snprintf(buf, buflen, "::ffff:%pI4",
2375 					&addr->s6_addr32[3]);
2376 
2377 	/*
2378 	 * RFC 4291, Section 2.2.1
2379 	 */
2380 	return snprintf(buf, buflen, "%pI6c", addr);
2381 }
2382 
2383 /* Derived from rpc_sockaddr2uaddr */
2384 static void
2385 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2386 {
2387 	struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2388 	char portbuf[RPCBIND_MAXUADDRPLEN];
2389 	char addrbuf[RPCBIND_MAXUADDRLEN];
2390 	unsigned short port;
2391 	int len, netid_len;
2392 	__be32 *p;
2393 
2394 	switch (sap->sa_family) {
2395 	case AF_INET:
2396 		if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2397 			return;
2398 		port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2399 		break;
2400 	case AF_INET6:
2401 		if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2402 			return;
2403 		port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2404 		break;
2405 	default:
2406 		WARN_ON_ONCE(1);
2407 		return;
2408 	}
2409 
2410 	snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2411 	len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2412 
2413 	netid_len = strlen(da->da_netid);
2414 	p = xdr_reserve_space(xdr, 4 + netid_len);
2415 	xdr_encode_opaque(p, da->da_netid, netid_len);
2416 
2417 	p = xdr_reserve_space(xdr, 4 + len);
2418 	xdr_encode_opaque(p, addrbuf, len);
2419 }
2420 
2421 static void
2422 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2423 			 ktime_t t)
2424 {
2425 	struct timespec64 ts;
2426 	__be32 *p;
2427 
2428 	p = xdr_reserve_space(xdr, 12);
2429 	ts = ktime_to_timespec64(t);
2430 	p = xdr_encode_hyper(p, ts.tv_sec);
2431 	*p++ = cpu_to_be32(ts.tv_nsec);
2432 }
2433 
2434 static void
2435 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2436 			    struct nfs4_ff_io_stat *stat)
2437 {
2438 	__be32 *p;
2439 
2440 	p = xdr_reserve_space(xdr, 5 * 8);
2441 	p = xdr_encode_hyper(p, stat->ops_requested);
2442 	p = xdr_encode_hyper(p, stat->bytes_requested);
2443 	p = xdr_encode_hyper(p, stat->ops_completed);
2444 	p = xdr_encode_hyper(p, stat->bytes_completed);
2445 	p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2446 	ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2447 	ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2448 }
2449 
2450 static void
2451 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2452 			      const struct nfs42_layoutstat_devinfo *devinfo,
2453 			      struct nfs4_ff_layout_mirror *mirror)
2454 {
2455 	struct nfs4_pnfs_ds_addr *da;
2456 	struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2457 	struct nfs_fh *fh = &mirror->fh_versions[0];
2458 	__be32 *p;
2459 
2460 	da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2461 	dprintk("%s: DS %s: encoding address %s\n",
2462 		__func__, ds->ds_remotestr, da->da_remotestr);
2463 	/* netaddr4 */
2464 	ff_layout_encode_netaddr(xdr, da);
2465 	/* nfs_fh4 */
2466 	p = xdr_reserve_space(xdr, 4 + fh->size);
2467 	xdr_encode_opaque(p, fh->data, fh->size);
2468 	/* ff_io_latency4 read */
2469 	spin_lock(&mirror->lock);
2470 	ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2471 	/* ff_io_latency4 write */
2472 	ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2473 	spin_unlock(&mirror->lock);
2474 	/* nfstime4 */
2475 	ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2476 	/* bool */
2477 	p = xdr_reserve_space(xdr, 4);
2478 	*p = cpu_to_be32(false);
2479 }
2480 
2481 static void
2482 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2483 			     const struct nfs4_xdr_opaque_data *opaque)
2484 {
2485 	struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2486 			struct nfs42_layoutstat_devinfo, ld_private);
2487 	__be32 *start;
2488 
2489 	/* layoutupdate length */
2490 	start = xdr_reserve_space(xdr, 4);
2491 	ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2492 
2493 	*start = cpu_to_be32((xdr->p - start - 1) * 4);
2494 }
2495 
2496 static void
2497 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2498 {
2499 	struct nfs4_ff_layout_mirror *mirror = opaque->data;
2500 
2501 	ff_layout_put_mirror(mirror);
2502 }
2503 
2504 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2505 	.encode = ff_layout_encode_layoutstats,
2506 	.free	= ff_layout_free_layoutstats,
2507 };
2508 
2509 static int
2510 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2511 			       struct nfs42_layoutstat_devinfo *devinfo,
2512 			       int dev_limit, enum nfs4_ff_op_type type)
2513 {
2514 	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2515 	struct nfs4_ff_layout_mirror *mirror;
2516 	struct nfs4_deviceid_node *dev;
2517 	int i = 0;
2518 
2519 	list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2520 		if (i >= dev_limit)
2521 			break;
2522 		if (IS_ERR_OR_NULL(mirror->mirror_ds))
2523 			continue;
2524 		if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL,
2525 					&mirror->flags) &&
2526 		    type != NFS4_FF_OP_LAYOUTRETURN)
2527 			continue;
2528 		/* mirror refcount put in cleanup_layoutstats */
2529 		if (!refcount_inc_not_zero(&mirror->ref))
2530 			continue;
2531 		dev = &mirror->mirror_ds->id_node;
2532 		memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2533 		devinfo->offset = 0;
2534 		devinfo->length = NFS4_MAX_UINT64;
2535 		spin_lock(&mirror->lock);
2536 		devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2537 		devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2538 		devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2539 		devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2540 		spin_unlock(&mirror->lock);
2541 		devinfo->layout_type = LAYOUT_FLEX_FILES;
2542 		devinfo->ld_private.ops = &layoutstat_ops;
2543 		devinfo->ld_private.data = mirror;
2544 
2545 		devinfo++;
2546 		i++;
2547 	}
2548 	return i;
2549 }
2550 
2551 static int ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2552 {
2553 	struct pnfs_layout_hdr *lo;
2554 	struct nfs4_flexfile_layout *ff_layout;
2555 	const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2556 
2557 	/* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2558 	args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo),
2559 				      nfs_io_gfp_mask());
2560 	if (!args->devinfo)
2561 		return -ENOMEM;
2562 
2563 	spin_lock(&args->inode->i_lock);
2564 	lo = NFS_I(args->inode)->layout;
2565 	if (lo && pnfs_layout_is_valid(lo)) {
2566 		ff_layout = FF_LAYOUT_FROM_HDR(lo);
2567 		args->num_dev = ff_layout_mirror_prepare_stats(
2568 			&ff_layout->generic_hdr, &args->devinfo[0], dev_count,
2569 			NFS4_FF_OP_LAYOUTSTATS);
2570 	} else
2571 		args->num_dev = 0;
2572 	spin_unlock(&args->inode->i_lock);
2573 	if (!args->num_dev) {
2574 		kfree(args->devinfo);
2575 		args->devinfo = NULL;
2576 		return -ENOENT;
2577 	}
2578 
2579 	return 0;
2580 }
2581 
2582 static int
2583 ff_layout_set_layoutdriver(struct nfs_server *server,
2584 		const struct nfs_fh *dummy)
2585 {
2586 #if IS_ENABLED(CONFIG_NFS_V4_2)
2587 	server->caps |= NFS_CAP_LAYOUTSTATS | NFS_CAP_REBOOT_LAYOUTRETURN;
2588 #endif
2589 	return 0;
2590 }
2591 
2592 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2593 	.setup_ds_info		= ff_layout_setup_ds_info,
2594 	.release_ds_info	= ff_layout_release_ds_info,
2595 	.mark_request_commit	= pnfs_layout_mark_request_commit,
2596 	.clear_request_commit	= pnfs_generic_clear_request_commit,
2597 	.scan_commit_lists	= pnfs_generic_scan_commit_lists,
2598 	.recover_commit_reqs	= pnfs_generic_recover_commit_reqs,
2599 	.commit_pagelist	= ff_layout_commit_pagelist,
2600 };
2601 
2602 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2603 	.id			= LAYOUT_FLEX_FILES,
2604 	.name			= "LAYOUT_FLEX_FILES",
2605 	.owner			= THIS_MODULE,
2606 	.flags			= PNFS_LAYOUTGET_ON_OPEN,
2607 	.max_layoutget_response	= 4096, /* 1 page or so... */
2608 	.set_layoutdriver	= ff_layout_set_layoutdriver,
2609 	.alloc_layout_hdr	= ff_layout_alloc_layout_hdr,
2610 	.free_layout_hdr	= ff_layout_free_layout_hdr,
2611 	.alloc_lseg		= ff_layout_alloc_lseg,
2612 	.free_lseg		= ff_layout_free_lseg,
2613 	.add_lseg		= ff_layout_add_lseg,
2614 	.pg_read_ops		= &ff_layout_pg_read_ops,
2615 	.pg_write_ops		= &ff_layout_pg_write_ops,
2616 	.get_ds_info		= ff_layout_get_ds_info,
2617 	.free_deviceid_node	= ff_layout_free_deviceid_node,
2618 	.read_pagelist		= ff_layout_read_pagelist,
2619 	.write_pagelist		= ff_layout_write_pagelist,
2620 	.alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2621 	.prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2622 	.sync			= pnfs_nfs_generic_sync,
2623 	.prepare_layoutstats	= ff_layout_prepare_layoutstats,
2624 	.cancel_io		= ff_layout_cancel_io,
2625 };
2626 
2627 static int __init nfs4flexfilelayout_init(void)
2628 {
2629 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2630 	       __func__);
2631 	return pnfs_register_layoutdriver(&flexfilelayout_type);
2632 }
2633 
2634 static void __exit nfs4flexfilelayout_exit(void)
2635 {
2636 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2637 	       __func__);
2638 	pnfs_unregister_layoutdriver(&flexfilelayout_type);
2639 }
2640 
2641 MODULE_ALIAS("nfs-layouttype4-4");
2642 
2643 MODULE_LICENSE("GPL");
2644 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2645 
2646 module_init(nfs4flexfilelayout_init);
2647 module_exit(nfs4flexfilelayout_exit);
2648 
2649 module_param(io_maxretrans, ushort, 0644);
2650 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2651 			"retries an I/O request before returning an error. ");
2652