1 /*
2  * Module for pnfs flexfile layout driver.
3  *
4  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
5  *
6  * Tao Peng <bergwolf@primarydata.com>
7  */
8 
9 #include <linux/nfs_fs.h>
10 #include <linux/nfs_page.h>
11 #include <linux/module.h>
12 
13 #include <linux/sunrpc/metrics.h>
14 
15 #include "flexfilelayout.h"
16 #include "../nfs4session.h"
17 #include "../nfs4idmap.h"
18 #include "../internal.h"
19 #include "../delegation.h"
20 #include "../nfs4trace.h"
21 #include "../iostat.h"
22 #include "../nfs.h"
23 
24 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
25 
26 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
27 
28 static struct pnfs_layout_hdr *
29 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
30 {
31 	struct nfs4_flexfile_layout *ffl;
32 
33 	ffl = kzalloc(sizeof(*ffl), gfp_flags);
34 	if (ffl) {
35 		INIT_LIST_HEAD(&ffl->error_list);
36 		return &ffl->generic_hdr;
37 	} else
38 		return NULL;
39 }
40 
41 static void
42 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
43 {
44 	struct nfs4_ff_layout_ds_err *err, *n;
45 
46 	list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
47 				 list) {
48 		list_del(&err->list);
49 		kfree(err);
50 	}
51 	kfree(FF_LAYOUT_FROM_HDR(lo));
52 }
53 
54 static int decode_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
55 {
56 	__be32 *p;
57 
58 	p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
59 	if (unlikely(p == NULL))
60 		return -ENOBUFS;
61 	memcpy(stateid, p, NFS4_STATEID_SIZE);
62 	dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
63 		p[0], p[1], p[2], p[3]);
64 	return 0;
65 }
66 
67 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
68 {
69 	__be32 *p;
70 
71 	p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
72 	if (unlikely(!p))
73 		return -ENOBUFS;
74 	memcpy(devid, p, NFS4_DEVICEID4_SIZE);
75 	nfs4_print_deviceid(devid);
76 	return 0;
77 }
78 
79 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
80 {
81 	__be32 *p;
82 
83 	p = xdr_inline_decode(xdr, 4);
84 	if (unlikely(!p))
85 		return -ENOBUFS;
86 	fh->size = be32_to_cpup(p++);
87 	if (fh->size > sizeof(struct nfs_fh)) {
88 		printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
89 		       fh->size);
90 		return -EOVERFLOW;
91 	}
92 	/* fh.data */
93 	p = xdr_inline_decode(xdr, fh->size);
94 	if (unlikely(!p))
95 		return -ENOBUFS;
96 	memcpy(&fh->data, p, fh->size);
97 	dprintk("%s: fh len %d\n", __func__, fh->size);
98 
99 	return 0;
100 }
101 
102 /*
103  * Currently only stringified uids and gids are accepted.
104  * I.e., kerberos is not supported to the DSes, so no pricipals.
105  *
106  * That means that one common function will suffice, but when
107  * principals are added, this should be split to accomodate
108  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
109  */
110 static int
111 decode_name(struct xdr_stream *xdr, u32 *id)
112 {
113 	__be32 *p;
114 	int len;
115 
116 	/* opaque_length(4)*/
117 	p = xdr_inline_decode(xdr, 4);
118 	if (unlikely(!p))
119 		return -ENOBUFS;
120 	len = be32_to_cpup(p++);
121 	if (len < 0)
122 		return -EINVAL;
123 
124 	dprintk("%s: len %u\n", __func__, len);
125 
126 	/* opaque body */
127 	p = xdr_inline_decode(xdr, len);
128 	if (unlikely(!p))
129 		return -ENOBUFS;
130 
131 	if (!nfs_map_string_to_numeric((char *)p, len, id))
132 		return -EINVAL;
133 
134 	return 0;
135 }
136 
137 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
138 {
139 	int i;
140 
141 	if (fls->mirror_array) {
142 		for (i = 0; i < fls->mirror_array_cnt; i++) {
143 			/* normally mirror_ds is freed in
144 			 * .free_deviceid_node but we still do it here
145 			 * for .alloc_lseg error path */
146 			if (fls->mirror_array[i]) {
147 				kfree(fls->mirror_array[i]->fh_versions);
148 				nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
149 				kfree(fls->mirror_array[i]);
150 			}
151 		}
152 		kfree(fls->mirror_array);
153 		fls->mirror_array = NULL;
154 	}
155 }
156 
157 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
158 {
159 	int ret = 0;
160 
161 	dprintk("--> %s\n", __func__);
162 
163 	/* FIXME: remove this check when layout segment support is added */
164 	if (lgr->range.offset != 0 ||
165 	    lgr->range.length != NFS4_MAX_UINT64) {
166 		dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
167 			__func__);
168 		ret = -EINVAL;
169 	}
170 
171 	dprintk("--> %s returns %d\n", __func__, ret);
172 	return ret;
173 }
174 
175 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
176 {
177 	if (fls) {
178 		ff_layout_free_mirror_array(fls);
179 		kfree(fls);
180 	}
181 }
182 
183 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
184 {
185 	struct nfs4_ff_layout_mirror *tmp;
186 	int i, j;
187 
188 	for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
189 		for (j = i + 1; j < fls->mirror_array_cnt; j++)
190 			if (fls->mirror_array[i]->efficiency <
191 			    fls->mirror_array[j]->efficiency) {
192 				tmp = fls->mirror_array[i];
193 				fls->mirror_array[i] = fls->mirror_array[j];
194 				fls->mirror_array[j] = tmp;
195 			}
196 	}
197 }
198 
199 static struct pnfs_layout_segment *
200 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
201 		     struct nfs4_layoutget_res *lgr,
202 		     gfp_t gfp_flags)
203 {
204 	struct pnfs_layout_segment *ret;
205 	struct nfs4_ff_layout_segment *fls = NULL;
206 	struct xdr_stream stream;
207 	struct xdr_buf buf;
208 	struct page *scratch;
209 	u64 stripe_unit;
210 	u32 mirror_array_cnt;
211 	__be32 *p;
212 	int i, rc;
213 
214 	dprintk("--> %s\n", __func__);
215 	scratch = alloc_page(gfp_flags);
216 	if (!scratch)
217 		return ERR_PTR(-ENOMEM);
218 
219 	xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
220 			      lgr->layoutp->len);
221 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
222 
223 	/* stripe unit and mirror_array_cnt */
224 	rc = -EIO;
225 	p = xdr_inline_decode(&stream, 8 + 4);
226 	if (!p)
227 		goto out_err_free;
228 
229 	p = xdr_decode_hyper(p, &stripe_unit);
230 	mirror_array_cnt = be32_to_cpup(p++);
231 	dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
232 		stripe_unit, mirror_array_cnt);
233 
234 	if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
235 	    mirror_array_cnt == 0)
236 		goto out_err_free;
237 
238 	rc = -ENOMEM;
239 	fls = kzalloc(sizeof(*fls), gfp_flags);
240 	if (!fls)
241 		goto out_err_free;
242 
243 	fls->mirror_array_cnt = mirror_array_cnt;
244 	fls->stripe_unit = stripe_unit;
245 	fls->mirror_array = kcalloc(fls->mirror_array_cnt,
246 				    sizeof(fls->mirror_array[0]), gfp_flags);
247 	if (fls->mirror_array == NULL)
248 		goto out_err_free;
249 
250 	for (i = 0; i < fls->mirror_array_cnt; i++) {
251 		struct nfs4_deviceid devid;
252 		struct nfs4_deviceid_node *idnode;
253 		u32 ds_count;
254 		u32 fh_count;
255 		int j;
256 
257 		rc = -EIO;
258 		p = xdr_inline_decode(&stream, 4);
259 		if (!p)
260 			goto out_err_free;
261 		ds_count = be32_to_cpup(p);
262 
263 		/* FIXME: allow for striping? */
264 		if (ds_count != 1)
265 			goto out_err_free;
266 
267 		fls->mirror_array[i] =
268 			kzalloc(sizeof(struct nfs4_ff_layout_mirror),
269 				gfp_flags);
270 		if (fls->mirror_array[i] == NULL) {
271 			rc = -ENOMEM;
272 			goto out_err_free;
273 		}
274 
275 		spin_lock_init(&fls->mirror_array[i]->lock);
276 		fls->mirror_array[i]->ds_count = ds_count;
277 
278 		/* deviceid */
279 		rc = decode_deviceid(&stream, &devid);
280 		if (rc)
281 			goto out_err_free;
282 
283 		idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
284 						&devid, lh->plh_lc_cred,
285 						gfp_flags);
286 		/*
287 		 * upon success, mirror_ds is allocated by previous
288 		 * getdeviceinfo, or newly by .alloc_deviceid_node
289 		 * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
290 		 */
291 		if (idnode)
292 			fls->mirror_array[i]->mirror_ds =
293 				FF_LAYOUT_MIRROR_DS(idnode);
294 		else
295 			goto out_err_free;
296 
297 		/* efficiency */
298 		rc = -EIO;
299 		p = xdr_inline_decode(&stream, 4);
300 		if (!p)
301 			goto out_err_free;
302 		fls->mirror_array[i]->efficiency = be32_to_cpup(p);
303 
304 		/* stateid */
305 		rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
306 		if (rc)
307 			goto out_err_free;
308 
309 		/* fh */
310 		p = xdr_inline_decode(&stream, 4);
311 		if (!p)
312 			goto out_err_free;
313 		fh_count = be32_to_cpup(p);
314 
315 		fls->mirror_array[i]->fh_versions =
316 			kzalloc(fh_count * sizeof(struct nfs_fh),
317 				gfp_flags);
318 		if (fls->mirror_array[i]->fh_versions == NULL) {
319 			rc = -ENOMEM;
320 			goto out_err_free;
321 		}
322 
323 		for (j = 0; j < fh_count; j++) {
324 			rc = decode_nfs_fh(&stream,
325 					   &fls->mirror_array[i]->fh_versions[j]);
326 			if (rc)
327 				goto out_err_free;
328 		}
329 
330 		fls->mirror_array[i]->fh_versions_cnt = fh_count;
331 
332 		/* user */
333 		rc = decode_name(&stream, &fls->mirror_array[i]->uid);
334 		if (rc)
335 			goto out_err_free;
336 
337 		/* group */
338 		rc = decode_name(&stream, &fls->mirror_array[i]->gid);
339 		if (rc)
340 			goto out_err_free;
341 
342 		dprintk("%s: uid %d gid %d\n", __func__,
343 			fls->mirror_array[i]->uid,
344 			fls->mirror_array[i]->gid);
345 	}
346 
347 	ff_layout_sort_mirrors(fls);
348 	rc = ff_layout_check_layout(lgr);
349 	if (rc)
350 		goto out_err_free;
351 
352 	ret = &fls->generic_hdr;
353 	dprintk("<-- %s (success)\n", __func__);
354 out_free_page:
355 	__free_page(scratch);
356 	return ret;
357 out_err_free:
358 	_ff_layout_free_lseg(fls);
359 	ret = ERR_PTR(rc);
360 	dprintk("<-- %s (%d)\n", __func__, rc);
361 	goto out_free_page;
362 }
363 
364 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
365 {
366 	struct pnfs_layout_segment *lseg;
367 
368 	list_for_each_entry(lseg, &layout->plh_segs, pls_list)
369 		if (lseg->pls_range.iomode == IOMODE_RW)
370 			return true;
371 
372 	return false;
373 }
374 
375 static void
376 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
377 {
378 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
379 	int i;
380 
381 	dprintk("--> %s\n", __func__);
382 
383 	for (i = 0; i < fls->mirror_array_cnt; i++) {
384 		if (fls->mirror_array[i]) {
385 			nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
386 			fls->mirror_array[i]->mirror_ds = NULL;
387 			if (fls->mirror_array[i]->cred) {
388 				put_rpccred(fls->mirror_array[i]->cred);
389 				fls->mirror_array[i]->cred = NULL;
390 			}
391 		}
392 	}
393 
394 	if (lseg->pls_range.iomode == IOMODE_RW) {
395 		struct nfs4_flexfile_layout *ffl;
396 		struct inode *inode;
397 
398 		ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
399 		inode = ffl->generic_hdr.plh_inode;
400 		spin_lock(&inode->i_lock);
401 		if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
402 			ffl->commit_info.nbuckets = 0;
403 			kfree(ffl->commit_info.buckets);
404 			ffl->commit_info.buckets = NULL;
405 		}
406 		spin_unlock(&inode->i_lock);
407 	}
408 	_ff_layout_free_lseg(fls);
409 }
410 
411 /* Return 1 until we have multiple lsegs support */
412 static int
413 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
414 {
415 	return 1;
416 }
417 
418 static int
419 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
420 			    struct nfs_commit_info *cinfo,
421 			    gfp_t gfp_flags)
422 {
423 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
424 	struct pnfs_commit_bucket *buckets;
425 	int size;
426 
427 	if (cinfo->ds->nbuckets != 0) {
428 		/* This assumes there is only one RW lseg per file.
429 		 * To support multiple lseg per file, we need to
430 		 * change struct pnfs_commit_bucket to allow dynamic
431 		 * increasing nbuckets.
432 		 */
433 		return 0;
434 	}
435 
436 	size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
437 
438 	buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
439 			  gfp_flags);
440 	if (!buckets)
441 		return -ENOMEM;
442 	else {
443 		int i;
444 
445 		spin_lock(cinfo->lock);
446 		if (cinfo->ds->nbuckets != 0)
447 			kfree(buckets);
448 		else {
449 			cinfo->ds->buckets = buckets;
450 			cinfo->ds->nbuckets = size;
451 			for (i = 0; i < size; i++) {
452 				INIT_LIST_HEAD(&buckets[i].written);
453 				INIT_LIST_HEAD(&buckets[i].committing);
454 				/* mark direct verifier as unset */
455 				buckets[i].direct_verf.committed =
456 					NFS_INVALID_STABLE_HOW;
457 			}
458 		}
459 		spin_unlock(cinfo->lock);
460 		return 0;
461 	}
462 }
463 
464 static struct nfs4_pnfs_ds *
465 ff_layout_choose_best_ds_for_read(struct nfs_pageio_descriptor *pgio,
466 				  int *best_idx)
467 {
468 	struct nfs4_ff_layout_segment *fls;
469 	struct nfs4_pnfs_ds *ds;
470 	int idx;
471 
472 	fls = FF_LAYOUT_LSEG(pgio->pg_lseg);
473 	/* mirrors are sorted by efficiency */
474 	for (idx = 0; idx < fls->mirror_array_cnt; idx++) {
475 		ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, idx, false);
476 		if (ds) {
477 			*best_idx = idx;
478 			return ds;
479 		}
480 	}
481 
482 	return NULL;
483 }
484 
485 static void
486 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
487 			struct nfs_page *req)
488 {
489 	struct nfs_pgio_mirror *pgm;
490 	struct nfs4_ff_layout_mirror *mirror;
491 	struct nfs4_pnfs_ds *ds;
492 	int ds_idx;
493 
494 	/* Use full layout for now */
495 	if (!pgio->pg_lseg)
496 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
497 						   req->wb_context,
498 						   0,
499 						   NFS4_MAX_UINT64,
500 						   IOMODE_READ,
501 						   GFP_KERNEL);
502 	/* If no lseg, fall back to read through mds */
503 	if (pgio->pg_lseg == NULL)
504 		goto out_mds;
505 
506 	ds = ff_layout_choose_best_ds_for_read(pgio, &ds_idx);
507 	if (!ds)
508 		goto out_mds;
509 	mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
510 
511 	pgio->pg_mirror_idx = ds_idx;
512 
513 	/* read always uses only one mirror - idx 0 for pgio layer */
514 	pgm = &pgio->pg_mirrors[0];
515 	pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
516 
517 	return;
518 out_mds:
519 	pnfs_put_lseg(pgio->pg_lseg);
520 	pgio->pg_lseg = NULL;
521 	nfs_pageio_reset_read_mds(pgio);
522 }
523 
524 static void
525 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
526 			struct nfs_page *req)
527 {
528 	struct nfs4_ff_layout_mirror *mirror;
529 	struct nfs_pgio_mirror *pgm;
530 	struct nfs_commit_info cinfo;
531 	struct nfs4_pnfs_ds *ds;
532 	int i;
533 	int status;
534 
535 	if (!pgio->pg_lseg)
536 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
537 						   req->wb_context,
538 						   0,
539 						   NFS4_MAX_UINT64,
540 						   IOMODE_RW,
541 						   GFP_NOFS);
542 	/* If no lseg, fall back to write through mds */
543 	if (pgio->pg_lseg == NULL)
544 		goto out_mds;
545 
546 	nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
547 	status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
548 	if (status < 0)
549 		goto out_mds;
550 
551 	/* Use a direct mapping of ds_idx to pgio mirror_idx */
552 	if (WARN_ON_ONCE(pgio->pg_mirror_count !=
553 	    FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
554 		goto out_mds;
555 
556 	for (i = 0; i < pgio->pg_mirror_count; i++) {
557 		ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
558 		if (!ds)
559 			goto out_mds;
560 		pgm = &pgio->pg_mirrors[i];
561 		mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
562 		pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
563 	}
564 
565 	return;
566 
567 out_mds:
568 	pnfs_put_lseg(pgio->pg_lseg);
569 	pgio->pg_lseg = NULL;
570 	nfs_pageio_reset_write_mds(pgio);
571 }
572 
573 static unsigned int
574 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
575 				    struct nfs_page *req)
576 {
577 	if (!pgio->pg_lseg)
578 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
579 						   req->wb_context,
580 						   0,
581 						   NFS4_MAX_UINT64,
582 						   IOMODE_RW,
583 						   GFP_NOFS);
584 	if (pgio->pg_lseg)
585 		return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
586 
587 	/* no lseg means that pnfs is not in use, so no mirroring here */
588 	pnfs_put_lseg(pgio->pg_lseg);
589 	pgio->pg_lseg = NULL;
590 	nfs_pageio_reset_write_mds(pgio);
591 	return 1;
592 }
593 
594 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
595 	.pg_init = ff_layout_pg_init_read,
596 	.pg_test = pnfs_generic_pg_test,
597 	.pg_doio = pnfs_generic_pg_readpages,
598 	.pg_cleanup = pnfs_generic_pg_cleanup,
599 };
600 
601 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
602 	.pg_init = ff_layout_pg_init_write,
603 	.pg_test = pnfs_generic_pg_test,
604 	.pg_doio = pnfs_generic_pg_writepages,
605 	.pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
606 	.pg_cleanup = pnfs_generic_pg_cleanup,
607 };
608 
609 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
610 {
611 	struct rpc_task *task = &hdr->task;
612 
613 	pnfs_layoutcommit_inode(hdr->inode, false);
614 
615 	if (retry_pnfs) {
616 		dprintk("%s Reset task %5u for i/o through pNFS "
617 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
618 			hdr->task.tk_pid,
619 			hdr->inode->i_sb->s_id,
620 			(unsigned long long)NFS_FILEID(hdr->inode),
621 			hdr->args.count,
622 			(unsigned long long)hdr->args.offset);
623 
624 		if (!hdr->dreq) {
625 			struct nfs_open_context *ctx;
626 
627 			ctx = nfs_list_entry(hdr->pages.next)->wb_context;
628 			set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
629 			hdr->completion_ops->error_cleanup(&hdr->pages);
630 		} else {
631 			nfs_direct_set_resched_writes(hdr->dreq);
632 			/* fake unstable write to let common nfs resend pages */
633 			hdr->verf.committed = NFS_UNSTABLE;
634 			hdr->good_bytes = 0;
635 		}
636 		return;
637 	}
638 
639 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
640 		dprintk("%s Reset task %5u for i/o through MDS "
641 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
642 			hdr->task.tk_pid,
643 			hdr->inode->i_sb->s_id,
644 			(unsigned long long)NFS_FILEID(hdr->inode),
645 			hdr->args.count,
646 			(unsigned long long)hdr->args.offset);
647 
648 		task->tk_status = pnfs_write_done_resend_to_mds(hdr);
649 	}
650 }
651 
652 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
653 {
654 	struct rpc_task *task = &hdr->task;
655 
656 	pnfs_layoutcommit_inode(hdr->inode, false);
657 
658 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
659 		dprintk("%s Reset task %5u for i/o through MDS "
660 			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
661 			hdr->task.tk_pid,
662 			hdr->inode->i_sb->s_id,
663 			(unsigned long long)NFS_FILEID(hdr->inode),
664 			hdr->args.count,
665 			(unsigned long long)hdr->args.offset);
666 
667 		task->tk_status = pnfs_read_done_resend_to_mds(hdr);
668 	}
669 }
670 
671 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
672 					   struct nfs4_state *state,
673 					   struct nfs_client *clp,
674 					   struct pnfs_layout_segment *lseg,
675 					   int idx)
676 {
677 	struct pnfs_layout_hdr *lo = lseg->pls_layout;
678 	struct inode *inode = lo->plh_inode;
679 	struct nfs_server *mds_server = NFS_SERVER(inode);
680 
681 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
682 	struct nfs_client *mds_client = mds_server->nfs_client;
683 	struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
684 
685 	if (task->tk_status >= 0)
686 		return 0;
687 
688 	switch (task->tk_status) {
689 	/* MDS state errors */
690 	case -NFS4ERR_DELEG_REVOKED:
691 	case -NFS4ERR_ADMIN_REVOKED:
692 	case -NFS4ERR_BAD_STATEID:
693 		if (state == NULL)
694 			break;
695 		nfs_remove_bad_delegation(state->inode);
696 	case -NFS4ERR_OPENMODE:
697 		if (state == NULL)
698 			break;
699 		if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
700 			goto out_bad_stateid;
701 		goto wait_on_recovery;
702 	case -NFS4ERR_EXPIRED:
703 		if (state != NULL) {
704 			if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
705 				goto out_bad_stateid;
706 		}
707 		nfs4_schedule_lease_recovery(mds_client);
708 		goto wait_on_recovery;
709 	/* DS session errors */
710 	case -NFS4ERR_BADSESSION:
711 	case -NFS4ERR_BADSLOT:
712 	case -NFS4ERR_BAD_HIGH_SLOT:
713 	case -NFS4ERR_DEADSESSION:
714 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
715 	case -NFS4ERR_SEQ_FALSE_RETRY:
716 	case -NFS4ERR_SEQ_MISORDERED:
717 		dprintk("%s ERROR %d, Reset session. Exchangeid "
718 			"flags 0x%x\n", __func__, task->tk_status,
719 			clp->cl_exchange_flags);
720 		nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
721 		break;
722 	case -NFS4ERR_DELAY:
723 	case -NFS4ERR_GRACE:
724 		rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
725 		break;
726 	case -NFS4ERR_RETRY_UNCACHED_REP:
727 		break;
728 	/* Invalidate Layout errors */
729 	case -NFS4ERR_PNFS_NO_LAYOUT:
730 	case -ESTALE:           /* mapped NFS4ERR_STALE */
731 	case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
732 	case -EISDIR:           /* mapped NFS4ERR_ISDIR */
733 	case -NFS4ERR_FHEXPIRED:
734 	case -NFS4ERR_WRONG_TYPE:
735 		dprintk("%s Invalid layout error %d\n", __func__,
736 			task->tk_status);
737 		/*
738 		 * Destroy layout so new i/o will get a new layout.
739 		 * Layout will not be destroyed until all current lseg
740 		 * references are put. Mark layout as invalid to resend failed
741 		 * i/o and all i/o waiting on the slot table to the MDS until
742 		 * layout is destroyed and a new valid layout is obtained.
743 		 */
744 		pnfs_destroy_layout(NFS_I(inode));
745 		rpc_wake_up(&tbl->slot_tbl_waitq);
746 		goto reset;
747 	/* RPC connection errors */
748 	case -ECONNREFUSED:
749 	case -EHOSTDOWN:
750 	case -EHOSTUNREACH:
751 	case -ENETUNREACH:
752 	case -EIO:
753 	case -ETIMEDOUT:
754 	case -EPIPE:
755 		dprintk("%s DS connection error %d\n", __func__,
756 			task->tk_status);
757 		nfs4_mark_deviceid_unavailable(devid);
758 		rpc_wake_up(&tbl->slot_tbl_waitq);
759 		/* fall through */
760 	default:
761 		if (ff_layout_has_available_ds(lseg))
762 			return -NFS4ERR_RESET_TO_PNFS;
763 reset:
764 		dprintk("%s Retry through MDS. Error %d\n", __func__,
765 			task->tk_status);
766 		return -NFS4ERR_RESET_TO_MDS;
767 	}
768 out:
769 	task->tk_status = 0;
770 	return -EAGAIN;
771 out_bad_stateid:
772 	task->tk_status = -EIO;
773 	return 0;
774 wait_on_recovery:
775 	rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
776 	if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
777 		rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
778 	goto out;
779 }
780 
781 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
782 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
783 					   struct pnfs_layout_segment *lseg,
784 					   int idx)
785 {
786 	struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
787 
788 	if (task->tk_status >= 0)
789 		return 0;
790 
791 	if (task->tk_status != -EJUKEBOX) {
792 		dprintk("%s DS connection error %d\n", __func__,
793 			task->tk_status);
794 		nfs4_mark_deviceid_unavailable(devid);
795 		if (ff_layout_has_available_ds(lseg))
796 			return -NFS4ERR_RESET_TO_PNFS;
797 		else
798 			return -NFS4ERR_RESET_TO_MDS;
799 	}
800 
801 	if (task->tk_status == -EJUKEBOX)
802 		nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
803 	task->tk_status = 0;
804 	rpc_restart_call(task);
805 	rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
806 	return -EAGAIN;
807 }
808 
809 static int ff_layout_async_handle_error(struct rpc_task *task,
810 					struct nfs4_state *state,
811 					struct nfs_client *clp,
812 					struct pnfs_layout_segment *lseg,
813 					int idx)
814 {
815 	int vers = clp->cl_nfs_mod->rpc_vers->number;
816 
817 	switch (vers) {
818 	case 3:
819 		return ff_layout_async_handle_error_v3(task, lseg, idx);
820 	case 4:
821 		return ff_layout_async_handle_error_v4(task, state, clp,
822 						       lseg, idx);
823 	default:
824 		/* should never happen */
825 		WARN_ON_ONCE(1);
826 		return 0;
827 	}
828 }
829 
830 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
831 					int idx, u64 offset, u64 length,
832 					u32 status, int opnum)
833 {
834 	struct nfs4_ff_layout_mirror *mirror;
835 	int err;
836 
837 	mirror = FF_LAYOUT_COMP(lseg, idx);
838 	err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
839 				       mirror, offset, length, status, opnum,
840 				       GFP_NOIO);
841 	dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
842 }
843 
844 /* NFS_PROTO call done callback routines */
845 
846 static int ff_layout_read_done_cb(struct rpc_task *task,
847 				struct nfs_pgio_header *hdr)
848 {
849 	struct inode *inode;
850 	int err;
851 
852 	trace_nfs4_pnfs_read(hdr, task->tk_status);
853 	if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
854 		hdr->res.op_status = NFS4ERR_NXIO;
855 	if (task->tk_status < 0 && hdr->res.op_status)
856 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
857 					    hdr->args.offset, hdr->args.count,
858 					    hdr->res.op_status, OP_READ);
859 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
860 					   hdr->ds_clp, hdr->lseg,
861 					   hdr->pgio_mirror_idx);
862 
863 	switch (err) {
864 	case -NFS4ERR_RESET_TO_PNFS:
865 		set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
866 			&hdr->lseg->pls_layout->plh_flags);
867 		pnfs_read_resend_pnfs(hdr);
868 		return task->tk_status;
869 	case -NFS4ERR_RESET_TO_MDS:
870 		inode = hdr->lseg->pls_layout->plh_inode;
871 		pnfs_error_mark_layout_for_return(inode, hdr->lseg);
872 		ff_layout_reset_read(hdr);
873 		return task->tk_status;
874 	case -EAGAIN:
875 		rpc_restart_call_prepare(task);
876 		return -EAGAIN;
877 	}
878 
879 	return 0;
880 }
881 
882 /*
883  * We reference the rpc_cred of the first WRITE that triggers the need for
884  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
885  * rfc5661 is not clear about which credential should be used.
886  *
887  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
888  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
889  * we always send layoutcommit after DS writes.
890  */
891 static void
892 ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
893 {
894 	pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
895 			hdr->mds_offset + hdr->res.count);
896 	dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
897 		(unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
898 }
899 
900 static bool
901 ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
902 {
903 	/* No mirroring for now */
904 	struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
905 
906 	return ff_layout_test_devid_unavailable(node);
907 }
908 
909 static int ff_layout_read_prepare_common(struct rpc_task *task,
910 					 struct nfs_pgio_header *hdr)
911 {
912 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
913 		rpc_exit(task, -EIO);
914 		return -EIO;
915 	}
916 	if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
917 		dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
918 		if (ff_layout_has_available_ds(hdr->lseg))
919 			pnfs_read_resend_pnfs(hdr);
920 		else
921 			ff_layout_reset_read(hdr);
922 		rpc_exit(task, 0);
923 		return -EAGAIN;
924 	}
925 	hdr->pgio_done_cb = ff_layout_read_done_cb;
926 
927 	return 0;
928 }
929 
930 /*
931  * Call ops for the async read/write cases
932  * In the case of dense layouts, the offset needs to be reset to its
933  * original value.
934  */
935 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
936 {
937 	struct nfs_pgio_header *hdr = data;
938 
939 	if (ff_layout_read_prepare_common(task, hdr))
940 		return;
941 
942 	rpc_call_start(task);
943 }
944 
945 static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
946 				    struct nfs4_sequence_args *args,
947 				    struct nfs4_sequence_res *res,
948 				    struct rpc_task *task)
949 {
950 	if (ds_clp->cl_session)
951 		return nfs41_setup_sequence(ds_clp->cl_session,
952 					   args,
953 					   res,
954 					   task);
955 	return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
956 				   args,
957 				   res,
958 				   task);
959 }
960 
961 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
962 {
963 	struct nfs_pgio_header *hdr = data;
964 
965 	if (ff_layout_read_prepare_common(task, hdr))
966 		return;
967 
968 	if (ff_layout_setup_sequence(hdr->ds_clp,
969 				     &hdr->args.seq_args,
970 				     &hdr->res.seq_res,
971 				     task))
972 		return;
973 
974 	if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
975 			hdr->args.lock_context, FMODE_READ) == -EIO)
976 		rpc_exit(task, -EIO); /* lost lock, terminate I/O */
977 }
978 
979 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
980 {
981 	struct nfs_pgio_header *hdr = data;
982 
983 	dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
984 
985 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
986 	    task->tk_status == 0) {
987 		nfs4_sequence_done(task, &hdr->res.seq_res);
988 		return;
989 	}
990 
991 	/* Note this may cause RPC to be resent */
992 	hdr->mds_ops->rpc_call_done(task, hdr);
993 }
994 
995 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
996 {
997 	struct nfs_pgio_header *hdr = data;
998 
999 	rpc_count_iostats_metrics(task,
1000 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1001 }
1002 
1003 static int ff_layout_write_done_cb(struct rpc_task *task,
1004 				struct nfs_pgio_header *hdr)
1005 {
1006 	struct inode *inode;
1007 	int err;
1008 
1009 	trace_nfs4_pnfs_write(hdr, task->tk_status);
1010 	if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
1011 		hdr->res.op_status = NFS4ERR_NXIO;
1012 	if (task->tk_status < 0 && hdr->res.op_status)
1013 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1014 					    hdr->args.offset, hdr->args.count,
1015 					    hdr->res.op_status, OP_WRITE);
1016 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
1017 					   hdr->ds_clp, hdr->lseg,
1018 					   hdr->pgio_mirror_idx);
1019 
1020 	switch (err) {
1021 	case -NFS4ERR_RESET_TO_PNFS:
1022 	case -NFS4ERR_RESET_TO_MDS:
1023 		inode = hdr->lseg->pls_layout->plh_inode;
1024 		pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1025 		if (err == -NFS4ERR_RESET_TO_PNFS) {
1026 			pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
1027 			ff_layout_reset_write(hdr, true);
1028 		} else {
1029 			pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
1030 			ff_layout_reset_write(hdr, false);
1031 		}
1032 		return task->tk_status;
1033 	case -EAGAIN:
1034 		rpc_restart_call_prepare(task);
1035 		return -EAGAIN;
1036 	}
1037 
1038 	if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1039 	    hdr->res.verf->committed == NFS_DATA_SYNC)
1040 		ff_layout_set_layoutcommit(hdr);
1041 
1042 	return 0;
1043 }
1044 
1045 static int ff_layout_commit_done_cb(struct rpc_task *task,
1046 				     struct nfs_commit_data *data)
1047 {
1048 	struct inode *inode;
1049 	int err;
1050 
1051 	trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1052 	if (task->tk_status == -ETIMEDOUT && !data->res.op_status)
1053 		data->res.op_status = NFS4ERR_NXIO;
1054 	if (task->tk_status < 0 && data->res.op_status)
1055 		ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1056 					    data->args.offset, data->args.count,
1057 					    data->res.op_status, OP_COMMIT);
1058 	err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1059 					   data->lseg, data->ds_commit_index);
1060 
1061 	switch (err) {
1062 	case -NFS4ERR_RESET_TO_PNFS:
1063 	case -NFS4ERR_RESET_TO_MDS:
1064 		inode = data->lseg->pls_layout->plh_inode;
1065 		pnfs_error_mark_layout_for_return(inode, data->lseg);
1066 		if (err == -NFS4ERR_RESET_TO_PNFS)
1067 			pnfs_set_retry_layoutget(data->lseg->pls_layout);
1068 		else
1069 			pnfs_clear_retry_layoutget(data->lseg->pls_layout);
1070 		pnfs_generic_prepare_to_resend_writes(data);
1071 		return -EAGAIN;
1072 	case -EAGAIN:
1073 		rpc_restart_call_prepare(task);
1074 		return -EAGAIN;
1075 	}
1076 
1077 	if (data->verf.committed == NFS_UNSTABLE)
1078 		pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
1079 
1080 	return 0;
1081 }
1082 
1083 static int ff_layout_write_prepare_common(struct rpc_task *task,
1084 					  struct nfs_pgio_header *hdr)
1085 {
1086 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1087 		rpc_exit(task, -EIO);
1088 		return -EIO;
1089 	}
1090 
1091 	if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1092 		bool retry_pnfs;
1093 
1094 		retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
1095 		dprintk("%s task %u reset io to %s\n", __func__,
1096 			task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
1097 		ff_layout_reset_write(hdr, retry_pnfs);
1098 		rpc_exit(task, 0);
1099 		return -EAGAIN;
1100 	}
1101 
1102 	return 0;
1103 }
1104 
1105 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1106 {
1107 	struct nfs_pgio_header *hdr = data;
1108 
1109 	if (ff_layout_write_prepare_common(task, hdr))
1110 		return;
1111 
1112 	rpc_call_start(task);
1113 }
1114 
1115 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1116 {
1117 	struct nfs_pgio_header *hdr = data;
1118 
1119 	if (ff_layout_write_prepare_common(task, hdr))
1120 		return;
1121 
1122 	if (ff_layout_setup_sequence(hdr->ds_clp,
1123 				     &hdr->args.seq_args,
1124 				     &hdr->res.seq_res,
1125 				     task))
1126 		return;
1127 
1128 	if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1129 			hdr->args.lock_context, FMODE_WRITE) == -EIO)
1130 		rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1131 }
1132 
1133 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1134 {
1135 	struct nfs_pgio_header *hdr = data;
1136 
1137 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1138 	    task->tk_status == 0) {
1139 		nfs4_sequence_done(task, &hdr->res.seq_res);
1140 		return;
1141 	}
1142 
1143 	/* Note this may cause RPC to be resent */
1144 	hdr->mds_ops->rpc_call_done(task, hdr);
1145 }
1146 
1147 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1148 {
1149 	struct nfs_pgio_header *hdr = data;
1150 
1151 	rpc_count_iostats_metrics(task,
1152 	    &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1153 }
1154 
1155 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1156 {
1157 	rpc_call_start(task);
1158 }
1159 
1160 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1161 {
1162 	struct nfs_commit_data *wdata = data;
1163 
1164 	ff_layout_setup_sequence(wdata->ds_clp,
1165 				 &wdata->args.seq_args,
1166 				 &wdata->res.seq_res,
1167 				 task);
1168 }
1169 
1170 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1171 {
1172 	struct nfs_commit_data *cdata = data;
1173 
1174 	rpc_count_iostats_metrics(task,
1175 	    &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1176 }
1177 
1178 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1179 	.rpc_call_prepare = ff_layout_read_prepare_v3,
1180 	.rpc_call_done = ff_layout_read_call_done,
1181 	.rpc_count_stats = ff_layout_read_count_stats,
1182 	.rpc_release = pnfs_generic_rw_release,
1183 };
1184 
1185 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1186 	.rpc_call_prepare = ff_layout_read_prepare_v4,
1187 	.rpc_call_done = ff_layout_read_call_done,
1188 	.rpc_count_stats = ff_layout_read_count_stats,
1189 	.rpc_release = pnfs_generic_rw_release,
1190 };
1191 
1192 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1193 	.rpc_call_prepare = ff_layout_write_prepare_v3,
1194 	.rpc_call_done = ff_layout_write_call_done,
1195 	.rpc_count_stats = ff_layout_write_count_stats,
1196 	.rpc_release = pnfs_generic_rw_release,
1197 };
1198 
1199 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1200 	.rpc_call_prepare = ff_layout_write_prepare_v4,
1201 	.rpc_call_done = ff_layout_write_call_done,
1202 	.rpc_count_stats = ff_layout_write_count_stats,
1203 	.rpc_release = pnfs_generic_rw_release,
1204 };
1205 
1206 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1207 	.rpc_call_prepare = ff_layout_commit_prepare_v3,
1208 	.rpc_call_done = pnfs_generic_write_commit_done,
1209 	.rpc_count_stats = ff_layout_commit_count_stats,
1210 	.rpc_release = pnfs_generic_commit_release,
1211 };
1212 
1213 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1214 	.rpc_call_prepare = ff_layout_commit_prepare_v4,
1215 	.rpc_call_done = pnfs_generic_write_commit_done,
1216 	.rpc_count_stats = ff_layout_commit_count_stats,
1217 	.rpc_release = pnfs_generic_commit_release,
1218 };
1219 
1220 static enum pnfs_try_status
1221 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1222 {
1223 	struct pnfs_layout_segment *lseg = hdr->lseg;
1224 	struct nfs4_pnfs_ds *ds;
1225 	struct rpc_clnt *ds_clnt;
1226 	struct rpc_cred *ds_cred;
1227 	loff_t offset = hdr->args.offset;
1228 	u32 idx = hdr->pgio_mirror_idx;
1229 	int vers;
1230 	struct nfs_fh *fh;
1231 
1232 	dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1233 		__func__, hdr->inode->i_ino,
1234 		hdr->args.pgbase, (size_t)hdr->args.count, offset);
1235 
1236 	ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1237 	if (!ds)
1238 		goto out_failed;
1239 
1240 	ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1241 						   hdr->inode);
1242 	if (IS_ERR(ds_clnt))
1243 		goto out_failed;
1244 
1245 	ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1246 	if (IS_ERR(ds_cred))
1247 		goto out_failed;
1248 
1249 	vers = nfs4_ff_layout_ds_version(lseg, idx);
1250 
1251 	dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1252 		ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1253 
1254 	atomic_inc(&ds->ds_clp->cl_count);
1255 	hdr->ds_clp = ds->ds_clp;
1256 	fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1257 	if (fh)
1258 		hdr->args.fh = fh;
1259 
1260 	/*
1261 	 * Note that if we ever decide to split across DSes,
1262 	 * then we may need to handle dense-like offsets.
1263 	 */
1264 	hdr->args.offset = offset;
1265 	hdr->mds_offset = offset;
1266 
1267 	/* Perform an asynchronous read to ds */
1268 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1269 			  vers == 3 ? &ff_layout_read_call_ops_v3 :
1270 				      &ff_layout_read_call_ops_v4,
1271 			  0, RPC_TASK_SOFTCONN);
1272 
1273 	return PNFS_ATTEMPTED;
1274 
1275 out_failed:
1276 	if (ff_layout_has_available_ds(lseg))
1277 		return PNFS_TRY_AGAIN;
1278 	return PNFS_NOT_ATTEMPTED;
1279 }
1280 
1281 /* Perform async writes. */
1282 static enum pnfs_try_status
1283 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1284 {
1285 	struct pnfs_layout_segment *lseg = hdr->lseg;
1286 	struct nfs4_pnfs_ds *ds;
1287 	struct rpc_clnt *ds_clnt;
1288 	struct rpc_cred *ds_cred;
1289 	loff_t offset = hdr->args.offset;
1290 	int vers;
1291 	struct nfs_fh *fh;
1292 	int idx = hdr->pgio_mirror_idx;
1293 
1294 	ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1295 	if (!ds)
1296 		return PNFS_NOT_ATTEMPTED;
1297 
1298 	ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1299 						   hdr->inode);
1300 	if (IS_ERR(ds_clnt))
1301 		return PNFS_NOT_ATTEMPTED;
1302 
1303 	ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1304 	if (IS_ERR(ds_cred))
1305 		return PNFS_NOT_ATTEMPTED;
1306 
1307 	vers = nfs4_ff_layout_ds_version(lseg, idx);
1308 
1309 	dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1310 		__func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1311 		offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1312 		vers);
1313 
1314 	hdr->pgio_done_cb = ff_layout_write_done_cb;
1315 	atomic_inc(&ds->ds_clp->cl_count);
1316 	hdr->ds_clp = ds->ds_clp;
1317 	hdr->ds_commit_idx = idx;
1318 	fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1319 	if (fh)
1320 		hdr->args.fh = fh;
1321 
1322 	/*
1323 	 * Note that if we ever decide to split across DSes,
1324 	 * then we may need to handle dense-like offsets.
1325 	 */
1326 	hdr->args.offset = offset;
1327 
1328 	/* Perform an asynchronous write */
1329 	nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1330 			  vers == 3 ? &ff_layout_write_call_ops_v3 :
1331 				      &ff_layout_write_call_ops_v4,
1332 			  sync, RPC_TASK_SOFTCONN);
1333 	return PNFS_ATTEMPTED;
1334 }
1335 
1336 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1337 {
1338 	return i;
1339 }
1340 
1341 static struct nfs_fh *
1342 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1343 {
1344 	struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1345 
1346 	/* FIXME: Assume that there is only one NFS version available
1347 	 * for the DS.
1348 	 */
1349 	return &flseg->mirror_array[i]->fh_versions[0];
1350 }
1351 
1352 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1353 {
1354 	struct pnfs_layout_segment *lseg = data->lseg;
1355 	struct nfs4_pnfs_ds *ds;
1356 	struct rpc_clnt *ds_clnt;
1357 	struct rpc_cred *ds_cred;
1358 	u32 idx;
1359 	int vers;
1360 	struct nfs_fh *fh;
1361 
1362 	idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1363 	ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1364 	if (!ds)
1365 		goto out_err;
1366 
1367 	ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1368 						   data->inode);
1369 	if (IS_ERR(ds_clnt))
1370 		goto out_err;
1371 
1372 	ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1373 	if (IS_ERR(ds_cred))
1374 		goto out_err;
1375 
1376 	vers = nfs4_ff_layout_ds_version(lseg, idx);
1377 
1378 	dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1379 		data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1380 		vers);
1381 	data->commit_done_cb = ff_layout_commit_done_cb;
1382 	data->cred = ds_cred;
1383 	atomic_inc(&ds->ds_clp->cl_count);
1384 	data->ds_clp = ds->ds_clp;
1385 	fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1386 	if (fh)
1387 		data->args.fh = fh;
1388 	return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1389 				   vers == 3 ? &ff_layout_commit_call_ops_v3 :
1390 					       &ff_layout_commit_call_ops_v4,
1391 				   how, RPC_TASK_SOFTCONN);
1392 out_err:
1393 	pnfs_generic_prepare_to_resend_writes(data);
1394 	pnfs_generic_commit_release(data);
1395 	return -EAGAIN;
1396 }
1397 
1398 static int
1399 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1400 			   int how, struct nfs_commit_info *cinfo)
1401 {
1402 	return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1403 					    ff_layout_initiate_commit);
1404 }
1405 
1406 static struct pnfs_ds_commit_info *
1407 ff_layout_get_ds_info(struct inode *inode)
1408 {
1409 	struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1410 
1411 	if (layout == NULL)
1412 		return NULL;
1413 
1414 	return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1415 }
1416 
1417 static void
1418 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1419 {
1420 	nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1421 						  id_node));
1422 }
1423 
1424 static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1425 				  struct xdr_stream *xdr,
1426 				  const struct nfs4_layoutreturn_args *args)
1427 {
1428 	struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1429 	__be32 *start;
1430 	int count = 0, ret = 0;
1431 
1432 	start = xdr_reserve_space(xdr, 4);
1433 	if (unlikely(!start))
1434 		return -E2BIG;
1435 
1436 	/* This assume we always return _ALL_ layouts */
1437 	spin_lock(&hdr->plh_inode->i_lock);
1438 	ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1439 	spin_unlock(&hdr->plh_inode->i_lock);
1440 
1441 	*start = cpu_to_be32(count);
1442 
1443 	return ret;
1444 }
1445 
1446 /* report nothing for now */
1447 static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1448 				     struct xdr_stream *xdr,
1449 				     const struct nfs4_layoutreturn_args *args)
1450 {
1451 	__be32 *p;
1452 
1453 	p = xdr_reserve_space(xdr, 4);
1454 	if (likely(p))
1455 		*p = cpu_to_be32(0);
1456 }
1457 
1458 static struct nfs4_deviceid_node *
1459 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1460 			      struct pnfs_device *pdev, gfp_t gfp_flags)
1461 {
1462 	struct nfs4_ff_layout_ds *dsaddr;
1463 
1464 	dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
1465 	if (!dsaddr)
1466 		return NULL;
1467 	return &dsaddr->id_node;
1468 }
1469 
1470 static void
1471 ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
1472 			      struct xdr_stream *xdr,
1473 			      const struct nfs4_layoutreturn_args *args)
1474 {
1475 	struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
1476 	__be32 *start;
1477 
1478 	dprintk("%s: Begin\n", __func__);
1479 	start = xdr_reserve_space(xdr, 4);
1480 	BUG_ON(!start);
1481 
1482 	if (ff_layout_encode_ioerr(flo, xdr, args))
1483 		goto out;
1484 
1485 	ff_layout_encode_iostats(flo, xdr, args);
1486 out:
1487 	*start = cpu_to_be32((xdr->p - start - 1) * 4);
1488 	dprintk("%s: Return\n", __func__);
1489 }
1490 
1491 static struct pnfs_layoutdriver_type flexfilelayout_type = {
1492 	.id			= LAYOUT_FLEX_FILES,
1493 	.name			= "LAYOUT_FLEX_FILES",
1494 	.owner			= THIS_MODULE,
1495 	.alloc_layout_hdr	= ff_layout_alloc_layout_hdr,
1496 	.free_layout_hdr	= ff_layout_free_layout_hdr,
1497 	.alloc_lseg		= ff_layout_alloc_lseg,
1498 	.free_lseg		= ff_layout_free_lseg,
1499 	.pg_read_ops		= &ff_layout_pg_read_ops,
1500 	.pg_write_ops		= &ff_layout_pg_write_ops,
1501 	.get_ds_info		= ff_layout_get_ds_info,
1502 	.free_deviceid_node	= ff_layout_free_deviceid_node,
1503 	.mark_request_commit	= pnfs_layout_mark_request_commit,
1504 	.clear_request_commit	= pnfs_generic_clear_request_commit,
1505 	.scan_commit_lists	= pnfs_generic_scan_commit_lists,
1506 	.recover_commit_reqs	= pnfs_generic_recover_commit_reqs,
1507 	.commit_pagelist	= ff_layout_commit_pagelist,
1508 	.read_pagelist		= ff_layout_read_pagelist,
1509 	.write_pagelist		= ff_layout_write_pagelist,
1510 	.alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
1511 	.encode_layoutreturn    = ff_layout_encode_layoutreturn,
1512 	.sync			= pnfs_nfs_generic_sync,
1513 };
1514 
1515 static int __init nfs4flexfilelayout_init(void)
1516 {
1517 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
1518 	       __func__);
1519 	return pnfs_register_layoutdriver(&flexfilelayout_type);
1520 }
1521 
1522 static void __exit nfs4flexfilelayout_exit(void)
1523 {
1524 	printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
1525 	       __func__);
1526 	pnfs_unregister_layoutdriver(&flexfilelayout_type);
1527 }
1528 
1529 MODULE_ALIAS("nfs-layouttype4-4");
1530 
1531 MODULE_LICENSE("GPL");
1532 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
1533 
1534 module_init(nfs4flexfilelayout_init);
1535 module_exit(nfs4flexfilelayout_exit);
1536