xref: /openbmc/linux/fs/nfs/blocklayout/blocklayout.c (revision 5e0266f0)
1 /*
2  *  linux/fs/nfs/blocklayout/blocklayout.c
3  *
4  *  Module for the NFSv4.1 pNFS block layout driver.
5  *
6  *  Copyright (c) 2006 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Andy Adamson <andros@citi.umich.edu>
10  *  Fred Isaman <iisaman@umich.edu>
11  *
12  * permission is granted to use, copy, create derivative works and
13  * redistribute this software and such derivative works for any purpose,
14  * so long as the name of the university of michigan is not used in
15  * any advertising or publicity pertaining to the use or distribution
16  * of this software without specific, written prior authorization.  if
17  * the above copyright notice or any other identification of the
18  * university of michigan is included in any copy of any portion of
19  * this software, then the disclaimer below must also be included.
20  *
21  * this software is provided as is, without representation from the
22  * university of michigan as to its fitness for any purpose, and without
23  * warranty by the university of michigan of any kind, either express
24  * or implied, including without limitation the implied warranties of
25  * merchantability and fitness for a particular purpose.  the regents
26  * of the university of michigan shall not be liable for any damages,
27  * including special, indirect, incidental, or consequential damages,
28  * with respect to any claim arising out or in connection with the use
29  * of the software, even if it has been or is hereafter advised of the
30  * possibility of such damages.
31  */
32 
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/bio.h>		/* struct bio */
38 #include <linux/prefetch.h>
39 #include <linux/pagevec.h>
40 
41 #include "../pnfs.h"
42 #include "../nfs4session.h"
43 #include "../internal.h"
44 #include "blocklayout.h"
45 
46 #define NFSDBG_FACILITY	NFSDBG_PNFS_LD
47 
48 MODULE_LICENSE("GPL");
49 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
50 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
51 
52 static bool is_hole(struct pnfs_block_extent *be)
53 {
54 	switch (be->be_state) {
55 	case PNFS_BLOCK_NONE_DATA:
56 		return true;
57 	case PNFS_BLOCK_INVALID_DATA:
58 		return be->be_tag ? false : true;
59 	default:
60 		return false;
61 	}
62 }
63 
64 /* The data we are handed might be spread across several bios.  We need
65  * to track when the last one is finished.
66  */
67 struct parallel_io {
68 	struct kref refcnt;
69 	void (*pnfs_callback) (void *data);
70 	void *data;
71 };
72 
73 static inline struct parallel_io *alloc_parallel(void *data)
74 {
75 	struct parallel_io *rv;
76 
77 	rv  = kmalloc(sizeof(*rv), GFP_NOFS);
78 	if (rv) {
79 		rv->data = data;
80 		kref_init(&rv->refcnt);
81 	}
82 	return rv;
83 }
84 
85 static inline void get_parallel(struct parallel_io *p)
86 {
87 	kref_get(&p->refcnt);
88 }
89 
90 static void destroy_parallel(struct kref *kref)
91 {
92 	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
93 
94 	dprintk("%s enter\n", __func__);
95 	p->pnfs_callback(p->data);
96 	kfree(p);
97 }
98 
99 static inline void put_parallel(struct parallel_io *p)
100 {
101 	kref_put(&p->refcnt, destroy_parallel);
102 }
103 
104 static struct bio *
105 bl_submit_bio(struct bio *bio)
106 {
107 	if (bio) {
108 		get_parallel(bio->bi_private);
109 		dprintk("%s submitting %s bio %u@%llu\n", __func__,
110 			bio_op(bio) == READ ? "read" : "write",
111 			bio->bi_iter.bi_size,
112 			(unsigned long long)bio->bi_iter.bi_sector);
113 		submit_bio(bio);
114 	}
115 	return NULL;
116 }
117 
118 static bool offset_in_map(u64 offset, struct pnfs_block_dev_map *map)
119 {
120 	return offset >= map->start && offset < map->start + map->len;
121 }
122 
123 static struct bio *
124 do_add_page_to_bio(struct bio *bio, int npg, enum req_op op, sector_t isect,
125 		struct page *page, struct pnfs_block_dev_map *map,
126 		struct pnfs_block_extent *be, bio_end_io_t end_io,
127 		struct parallel_io *par, unsigned int offset, int *len)
128 {
129 	struct pnfs_block_dev *dev =
130 		container_of(be->be_device, struct pnfs_block_dev, node);
131 	u64 disk_addr, end;
132 
133 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
134 		npg, (__force u32)op, (unsigned long long)isect, offset, *len);
135 
136 	/* translate to device offset */
137 	isect += be->be_v_offset;
138 	isect -= be->be_f_offset;
139 
140 	/* translate to physical disk offset */
141 	disk_addr = (u64)isect << SECTOR_SHIFT;
142 	if (!offset_in_map(disk_addr, map)) {
143 		if (!dev->map(dev, disk_addr, map) || !offset_in_map(disk_addr, map))
144 			return ERR_PTR(-EIO);
145 		bio = bl_submit_bio(bio);
146 	}
147 	disk_addr += map->disk_offset;
148 	disk_addr -= map->start;
149 
150 	/* limit length to what the device mapping allows */
151 	end = disk_addr + *len;
152 	if (end >= map->start + map->len)
153 		*len = map->start + map->len - disk_addr;
154 
155 retry:
156 	if (!bio) {
157 		bio = bio_alloc(map->bdev, bio_max_segs(npg), op, GFP_NOIO);
158 		bio->bi_iter.bi_sector = disk_addr >> SECTOR_SHIFT;
159 		bio->bi_end_io = end_io;
160 		bio->bi_private = par;
161 	}
162 	if (bio_add_page(bio, page, *len, offset) < *len) {
163 		bio = bl_submit_bio(bio);
164 		goto retry;
165 	}
166 	return bio;
167 }
168 
169 static void bl_mark_devices_unavailable(struct nfs_pgio_header *header, bool rw)
170 {
171 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
172 	size_t bytes_left = header->args.count;
173 	sector_t isect, extent_length = 0;
174 	struct pnfs_block_extent be;
175 
176 	isect = header->args.offset >> SECTOR_SHIFT;
177 	bytes_left += header->args.offset - (isect << SECTOR_SHIFT);
178 
179 	while (bytes_left > 0) {
180 		if (!ext_tree_lookup(bl, isect, &be, rw))
181 				return;
182 		extent_length = be.be_length - (isect - be.be_f_offset);
183 		nfs4_mark_deviceid_unavailable(be.be_device);
184 		isect += extent_length;
185 		if (bytes_left > extent_length << SECTOR_SHIFT)
186 			bytes_left -= extent_length << SECTOR_SHIFT;
187 		else
188 			bytes_left = 0;
189 	}
190 }
191 
192 static void bl_end_io_read(struct bio *bio)
193 {
194 	struct parallel_io *par = bio->bi_private;
195 
196 	if (bio->bi_status) {
197 		struct nfs_pgio_header *header = par->data;
198 
199 		if (!header->pnfs_error)
200 			header->pnfs_error = -EIO;
201 		pnfs_set_lo_fail(header->lseg);
202 		bl_mark_devices_unavailable(header, false);
203 	}
204 
205 	bio_put(bio);
206 	put_parallel(par);
207 }
208 
209 static void bl_read_cleanup(struct work_struct *work)
210 {
211 	struct rpc_task *task;
212 	struct nfs_pgio_header *hdr;
213 	dprintk("%s enter\n", __func__);
214 	task = container_of(work, struct rpc_task, u.tk_work);
215 	hdr = container_of(task, struct nfs_pgio_header, task);
216 	pnfs_ld_read_done(hdr);
217 }
218 
219 static void
220 bl_end_par_io_read(void *data)
221 {
222 	struct nfs_pgio_header *hdr = data;
223 
224 	hdr->task.tk_status = hdr->pnfs_error;
225 	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
226 	schedule_work(&hdr->task.u.tk_work);
227 }
228 
229 static enum pnfs_try_status
230 bl_read_pagelist(struct nfs_pgio_header *header)
231 {
232 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
233 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
234 	struct bio *bio = NULL;
235 	struct pnfs_block_extent be;
236 	sector_t isect, extent_length = 0;
237 	struct parallel_io *par;
238 	loff_t f_offset = header->args.offset;
239 	size_t bytes_left = header->args.count;
240 	unsigned int pg_offset = header->args.pgbase, pg_len;
241 	struct page **pages = header->args.pages;
242 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
243 	const bool is_dio = (header->dreq != NULL);
244 	struct blk_plug plug;
245 	int i;
246 
247 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
248 		header->page_array.npages, f_offset,
249 		(unsigned int)header->args.count);
250 
251 	par = alloc_parallel(header);
252 	if (!par)
253 		return PNFS_NOT_ATTEMPTED;
254 	par->pnfs_callback = bl_end_par_io_read;
255 
256 	blk_start_plug(&plug);
257 
258 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
259 	/* Code assumes extents are page-aligned */
260 	for (i = pg_index; i < header->page_array.npages; i++) {
261 		if (extent_length <= 0) {
262 			/* We've used up the previous extent */
263 			bio = bl_submit_bio(bio);
264 
265 			/* Get the next one */
266 			if (!ext_tree_lookup(bl, isect, &be, false)) {
267 				header->pnfs_error = -EIO;
268 				goto out;
269 			}
270 			extent_length = be.be_length - (isect - be.be_f_offset);
271 		}
272 
273 		if (is_dio) {
274 			if (pg_offset + bytes_left > PAGE_SIZE)
275 				pg_len = PAGE_SIZE - pg_offset;
276 			else
277 				pg_len = bytes_left;
278 		} else {
279 			BUG_ON(pg_offset != 0);
280 			pg_len = PAGE_SIZE;
281 		}
282 
283 		if (is_hole(&be)) {
284 			bio = bl_submit_bio(bio);
285 			/* Fill hole w/ zeroes w/o accessing device */
286 			dprintk("%s Zeroing page for hole\n", __func__);
287 			zero_user_segment(pages[i], pg_offset, pg_len);
288 
289 			/* invalidate map */
290 			map.start = NFS4_MAX_UINT64;
291 		} else {
292 			bio = do_add_page_to_bio(bio,
293 						 header->page_array.npages - i,
294 						 REQ_OP_READ,
295 						 isect, pages[i], &map, &be,
296 						 bl_end_io_read, par,
297 						 pg_offset, &pg_len);
298 			if (IS_ERR(bio)) {
299 				header->pnfs_error = PTR_ERR(bio);
300 				bio = NULL;
301 				goto out;
302 			}
303 		}
304 		isect += (pg_len >> SECTOR_SHIFT);
305 		extent_length -= (pg_len >> SECTOR_SHIFT);
306 		f_offset += pg_len;
307 		bytes_left -= pg_len;
308 		pg_offset = 0;
309 	}
310 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
311 		header->res.eof = 1;
312 		header->res.count = header->inode->i_size - header->args.offset;
313 	} else {
314 		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
315 	}
316 out:
317 	bl_submit_bio(bio);
318 	blk_finish_plug(&plug);
319 	put_parallel(par);
320 	return PNFS_ATTEMPTED;
321 }
322 
323 static void bl_end_io_write(struct bio *bio)
324 {
325 	struct parallel_io *par = bio->bi_private;
326 	struct nfs_pgio_header *header = par->data;
327 
328 	if (bio->bi_status) {
329 		if (!header->pnfs_error)
330 			header->pnfs_error = -EIO;
331 		pnfs_set_lo_fail(header->lseg);
332 		bl_mark_devices_unavailable(header, true);
333 	}
334 	bio_put(bio);
335 	put_parallel(par);
336 }
337 
338 /* Function scheduled for call during bl_end_par_io_write,
339  * it marks sectors as written and extends the commitlist.
340  */
341 static void bl_write_cleanup(struct work_struct *work)
342 {
343 	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
344 	struct nfs_pgio_header *hdr =
345 			container_of(task, struct nfs_pgio_header, task);
346 
347 	dprintk("%s enter\n", __func__);
348 
349 	if (likely(!hdr->pnfs_error)) {
350 		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
351 		u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
352 		u64 end = (hdr->args.offset + hdr->args.count +
353 			PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
354 		u64 lwb = hdr->args.offset + hdr->args.count;
355 
356 		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
357 					(end - start) >> SECTOR_SHIFT, lwb);
358 	}
359 
360 	pnfs_ld_write_done(hdr);
361 }
362 
363 /* Called when last of bios associated with a bl_write_pagelist call finishes */
364 static void bl_end_par_io_write(void *data)
365 {
366 	struct nfs_pgio_header *hdr = data;
367 
368 	hdr->task.tk_status = hdr->pnfs_error;
369 	hdr->verf.committed = NFS_FILE_SYNC;
370 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
371 	schedule_work(&hdr->task.u.tk_work);
372 }
373 
374 static enum pnfs_try_status
375 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
376 {
377 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
378 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
379 	struct bio *bio = NULL;
380 	struct pnfs_block_extent be;
381 	sector_t isect, extent_length = 0;
382 	struct parallel_io *par = NULL;
383 	loff_t offset = header->args.offset;
384 	size_t count = header->args.count;
385 	struct page **pages = header->args.pages;
386 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
387 	unsigned int pg_len;
388 	struct blk_plug plug;
389 	int i;
390 
391 	dprintk("%s enter, %zu@%lld\n", __func__, count, offset);
392 
393 	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
394 	 * We want to write each, and if there is an error set pnfs_error
395 	 * to have it redone using nfs.
396 	 */
397 	par = alloc_parallel(header);
398 	if (!par)
399 		return PNFS_NOT_ATTEMPTED;
400 	par->pnfs_callback = bl_end_par_io_write;
401 
402 	blk_start_plug(&plug);
403 
404 	/* we always write out the whole page */
405 	offset = offset & (loff_t)PAGE_MASK;
406 	isect = offset >> SECTOR_SHIFT;
407 
408 	for (i = pg_index; i < header->page_array.npages; i++) {
409 		if (extent_length <= 0) {
410 			/* We've used up the previous extent */
411 			bio = bl_submit_bio(bio);
412 			/* Get the next one */
413 			if (!ext_tree_lookup(bl, isect, &be, true)) {
414 				header->pnfs_error = -EINVAL;
415 				goto out;
416 			}
417 
418 			extent_length = be.be_length - (isect - be.be_f_offset);
419 		}
420 
421 		pg_len = PAGE_SIZE;
422 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
423 					 REQ_OP_WRITE, isect, pages[i], &map,
424 					 &be, bl_end_io_write, par, 0, &pg_len);
425 		if (IS_ERR(bio)) {
426 			header->pnfs_error = PTR_ERR(bio);
427 			bio = NULL;
428 			goto out;
429 		}
430 
431 		offset += pg_len;
432 		count -= pg_len;
433 		isect += (pg_len >> SECTOR_SHIFT);
434 		extent_length -= (pg_len >> SECTOR_SHIFT);
435 	}
436 
437 	header->res.count = header->args.count;
438 out:
439 	bl_submit_bio(bio);
440 	blk_finish_plug(&plug);
441 	put_parallel(par);
442 	return PNFS_ATTEMPTED;
443 }
444 
445 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
446 {
447 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
448 	int err;
449 
450 	dprintk("%s enter\n", __func__);
451 
452 	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
453 	WARN_ON(err);
454 
455 	kfree_rcu(bl, bl_layout.plh_rcu);
456 }
457 
458 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
459 		gfp_t gfp_flags, bool is_scsi_layout)
460 {
461 	struct pnfs_block_layout *bl;
462 
463 	dprintk("%s enter\n", __func__);
464 	bl = kzalloc(sizeof(*bl), gfp_flags);
465 	if (!bl)
466 		return NULL;
467 
468 	bl->bl_ext_rw = RB_ROOT;
469 	bl->bl_ext_ro = RB_ROOT;
470 	spin_lock_init(&bl->bl_ext_lock);
471 
472 	bl->bl_scsi_layout = is_scsi_layout;
473 	return &bl->bl_layout;
474 }
475 
476 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
477 						   gfp_t gfp_flags)
478 {
479 	return __bl_alloc_layout_hdr(inode, gfp_flags, false);
480 }
481 
482 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
483 						   gfp_t gfp_flags)
484 {
485 	return __bl_alloc_layout_hdr(inode, gfp_flags, true);
486 }
487 
488 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
489 {
490 	dprintk("%s enter\n", __func__);
491 	kfree(lseg);
492 }
493 
494 /* Tracks info needed to ensure extents in layout obey constraints of spec */
495 struct layout_verification {
496 	u32 mode;	/* R or RW */
497 	u64 start;	/* Expected start of next non-COW extent */
498 	u64 inval;	/* Start of INVAL coverage */
499 	u64 cowread;	/* End of COW read coverage */
500 };
501 
502 /* Verify the extent meets the layout requirements of the pnfs-block draft,
503  * section 2.3.1.
504  */
505 static int verify_extent(struct pnfs_block_extent *be,
506 			 struct layout_verification *lv)
507 {
508 	if (lv->mode == IOMODE_READ) {
509 		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
510 		    be->be_state == PNFS_BLOCK_INVALID_DATA)
511 			return -EIO;
512 		if (be->be_f_offset != lv->start)
513 			return -EIO;
514 		lv->start += be->be_length;
515 		return 0;
516 	}
517 	/* lv->mode == IOMODE_RW */
518 	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
519 		if (be->be_f_offset != lv->start)
520 			return -EIO;
521 		if (lv->cowread > lv->start)
522 			return -EIO;
523 		lv->start += be->be_length;
524 		lv->inval = lv->start;
525 		return 0;
526 	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
527 		if (be->be_f_offset != lv->start)
528 			return -EIO;
529 		lv->start += be->be_length;
530 		return 0;
531 	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
532 		if (be->be_f_offset > lv->start)
533 			return -EIO;
534 		if (be->be_f_offset < lv->inval)
535 			return -EIO;
536 		if (be->be_f_offset < lv->cowread)
537 			return -EIO;
538 		/* It looks like you might want to min this with lv->start,
539 		 * but you really don't.
540 		 */
541 		lv->inval = lv->inval + be->be_length;
542 		lv->cowread = be->be_f_offset + be->be_length;
543 		return 0;
544 	} else
545 		return -EIO;
546 }
547 
548 static int decode_sector_number(__be32 **rp, sector_t *sp)
549 {
550 	uint64_t s;
551 
552 	*rp = xdr_decode_hyper(*rp, &s);
553 	if (s & 0x1ff) {
554 		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
555 		return -1;
556 	}
557 	*sp = s >> SECTOR_SHIFT;
558 	return 0;
559 }
560 
561 static struct nfs4_deviceid_node *
562 bl_find_get_deviceid(struct nfs_server *server,
563 		const struct nfs4_deviceid *id, const struct cred *cred,
564 		gfp_t gfp_mask)
565 {
566 	struct nfs4_deviceid_node *node;
567 	unsigned long start, end;
568 
569 retry:
570 	node = nfs4_find_get_deviceid(server, id, cred, gfp_mask);
571 	if (!node)
572 		return ERR_PTR(-ENODEV);
573 
574 	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags) == 0)
575 		return node;
576 
577 	end = jiffies;
578 	start = end - PNFS_DEVICE_RETRY_TIMEOUT;
579 	if (!time_in_range(node->timestamp_unavailable, start, end)) {
580 		nfs4_delete_deviceid(node->ld, node->nfs_client, id);
581 		goto retry;
582 	}
583 	return ERR_PTR(-ENODEV);
584 }
585 
586 static int
587 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
588 		struct layout_verification *lv, struct list_head *extents,
589 		gfp_t gfp_mask)
590 {
591 	struct pnfs_block_extent *be;
592 	struct nfs4_deviceid id;
593 	int error;
594 	__be32 *p;
595 
596 	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
597 	if (!p)
598 		return -EIO;
599 
600 	be = kzalloc(sizeof(*be), GFP_NOFS);
601 	if (!be)
602 		return -ENOMEM;
603 
604 	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
605 	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
606 
607 	be->be_device = bl_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
608 						lo->plh_lc_cred, gfp_mask);
609 	if (IS_ERR(be->be_device)) {
610 		error = PTR_ERR(be->be_device);
611 		goto out_free_be;
612 	}
613 
614 	/*
615 	 * The next three values are read in as bytes, but stored in the
616 	 * extent structure in 512-byte granularity.
617 	 */
618 	error = -EIO;
619 	if (decode_sector_number(&p, &be->be_f_offset) < 0)
620 		goto out_put_deviceid;
621 	if (decode_sector_number(&p, &be->be_length) < 0)
622 		goto out_put_deviceid;
623 	if (decode_sector_number(&p, &be->be_v_offset) < 0)
624 		goto out_put_deviceid;
625 	be->be_state = be32_to_cpup(p++);
626 
627 	error = verify_extent(be, lv);
628 	if (error) {
629 		dprintk("%s: extent verification failed\n", __func__);
630 		goto out_put_deviceid;
631 	}
632 
633 	list_add_tail(&be->be_list, extents);
634 	return 0;
635 
636 out_put_deviceid:
637 	nfs4_put_deviceid_node(be->be_device);
638 out_free_be:
639 	kfree(be);
640 	return error;
641 }
642 
643 static struct pnfs_layout_segment *
644 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
645 		gfp_t gfp_mask)
646 {
647 	struct layout_verification lv = {
648 		.mode = lgr->range.iomode,
649 		.start = lgr->range.offset >> SECTOR_SHIFT,
650 		.inval = lgr->range.offset >> SECTOR_SHIFT,
651 		.cowread = lgr->range.offset >> SECTOR_SHIFT,
652 	};
653 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
654 	struct pnfs_layout_segment *lseg;
655 	struct xdr_buf buf;
656 	struct xdr_stream xdr;
657 	struct page *scratch;
658 	int status, i;
659 	uint32_t count;
660 	__be32 *p;
661 	LIST_HEAD(extents);
662 
663 	dprintk("---> %s\n", __func__);
664 
665 	lseg = kzalloc(sizeof(*lseg), gfp_mask);
666 	if (!lseg)
667 		return ERR_PTR(-ENOMEM);
668 
669 	status = -ENOMEM;
670 	scratch = alloc_page(gfp_mask);
671 	if (!scratch)
672 		goto out;
673 
674 	xdr_init_decode_pages(&xdr, &buf,
675 			lgr->layoutp->pages, lgr->layoutp->len);
676 	xdr_set_scratch_page(&xdr, scratch);
677 
678 	status = -EIO;
679 	p = xdr_inline_decode(&xdr, 4);
680 	if (unlikely(!p))
681 		goto out_free_scratch;
682 
683 	count = be32_to_cpup(p++);
684 	dprintk("%s: number of extents %d\n", __func__, count);
685 
686 	/*
687 	 * Decode individual extents, putting them in temporary staging area
688 	 * until whole layout is decoded to make error recovery easier.
689 	 */
690 	for (i = 0; i < count; i++) {
691 		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
692 		if (status)
693 			goto process_extents;
694 	}
695 
696 	if (lgr->range.offset + lgr->range.length !=
697 			lv.start << SECTOR_SHIFT) {
698 		dprintk("%s Final length mismatch\n", __func__);
699 		status = -EIO;
700 		goto process_extents;
701 	}
702 
703 	if (lv.start < lv.cowread) {
704 		dprintk("%s Final uncovered COW extent\n", __func__);
705 		status = -EIO;
706 	}
707 
708 process_extents:
709 	while (!list_empty(&extents)) {
710 		struct pnfs_block_extent *be =
711 			list_first_entry(&extents, struct pnfs_block_extent,
712 					 be_list);
713 		list_del(&be->be_list);
714 
715 		if (!status)
716 			status = ext_tree_insert(bl, be);
717 
718 		if (status) {
719 			nfs4_put_deviceid_node(be->be_device);
720 			kfree(be);
721 		}
722 	}
723 
724 out_free_scratch:
725 	__free_page(scratch);
726 out:
727 	dprintk("%s returns %d\n", __func__, status);
728 	switch (status) {
729 	case -ENODEV:
730 		/* Our extent block devices are unavailable */
731 		set_bit(NFS_LSEG_UNAVAILABLE, &lseg->pls_flags);
732 		fallthrough;
733 	case 0:
734 		return lseg;
735 	default:
736 		kfree(lseg);
737 		return ERR_PTR(status);
738 	}
739 }
740 
741 static void
742 bl_return_range(struct pnfs_layout_hdr *lo,
743 		struct pnfs_layout_range *range)
744 {
745 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
746 	sector_t offset = range->offset >> SECTOR_SHIFT, end;
747 
748 	if (range->offset % 8) {
749 		dprintk("%s: offset %lld not block size aligned\n",
750 			__func__, range->offset);
751 		return;
752 	}
753 
754 	if (range->length != NFS4_MAX_UINT64) {
755 		if (range->length % 8) {
756 			dprintk("%s: length %lld not block size aligned\n",
757 				__func__, range->length);
758 			return;
759 		}
760 
761 		end = offset + (range->length >> SECTOR_SHIFT);
762 	} else {
763 		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
764 	}
765 
766 	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
767 }
768 
769 static int
770 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
771 {
772 	return ext_tree_prepare_commit(arg);
773 }
774 
775 static void
776 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
777 {
778 	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
779 }
780 
781 static int
782 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
783 {
784 	dprintk("%s enter\n", __func__);
785 
786 	if (server->pnfs_blksize == 0) {
787 		dprintk("%s Server did not return blksize\n", __func__);
788 		return -EINVAL;
789 	}
790 	if (server->pnfs_blksize > PAGE_SIZE) {
791 		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
792 			__func__, server->pnfs_blksize);
793 		return -EINVAL;
794 	}
795 
796 	return 0;
797 }
798 
799 static bool
800 is_aligned_req(struct nfs_pageio_descriptor *pgio,
801 		struct nfs_page *req, unsigned int alignment, bool is_write)
802 {
803 	/*
804 	 * Always accept buffered writes, higher layers take care of the
805 	 * right alignment.
806 	 */
807 	if (pgio->pg_dreq == NULL)
808 		return true;
809 
810 	if (!IS_ALIGNED(req->wb_offset, alignment))
811 		return false;
812 
813 	if (IS_ALIGNED(req->wb_bytes, alignment))
814 		return true;
815 
816 	if (is_write &&
817 	    (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
818 		/*
819 		 * If the write goes up to the inode size, just write
820 		 * the full page.  Data past the inode size is
821 		 * guaranteed to be zeroed by the higher level client
822 		 * code, and this behaviour is mandated by RFC 5663
823 		 * section 2.3.2.
824 		 */
825 		return true;
826 	}
827 
828 	return false;
829 }
830 
831 static void
832 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
833 {
834 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
835 		nfs_pageio_reset_read_mds(pgio);
836 		return;
837 	}
838 
839 	pnfs_generic_pg_init_read(pgio, req);
840 
841 	if (pgio->pg_lseg &&
842 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
843 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
844 		pnfs_set_lo_fail(pgio->pg_lseg);
845 		nfs_pageio_reset_read_mds(pgio);
846 	}
847 }
848 
849 /*
850  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
851  * of bytes (maximum @req->wb_bytes) that can be coalesced.
852  */
853 static size_t
854 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
855 		struct nfs_page *req)
856 {
857 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
858 		return 0;
859 	return pnfs_generic_pg_test(pgio, prev, req);
860 }
861 
862 /*
863  * Return the number of contiguous bytes for a given inode
864  * starting at page frame idx.
865  */
866 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
867 {
868 	struct address_space *mapping = inode->i_mapping;
869 	pgoff_t end;
870 
871 	/* Optimize common case that writes from 0 to end of file */
872 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
873 	if (end != inode->i_mapping->nrpages) {
874 		rcu_read_lock();
875 		end = page_cache_next_miss(mapping, idx + 1, ULONG_MAX);
876 		rcu_read_unlock();
877 	}
878 
879 	if (!end)
880 		return i_size_read(inode) - (idx << PAGE_SHIFT);
881 	else
882 		return (end - idx) << PAGE_SHIFT;
883 }
884 
885 static void
886 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
887 {
888 	u64 wb_size;
889 
890 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
891 		nfs_pageio_reset_write_mds(pgio);
892 		return;
893 	}
894 
895 	if (pgio->pg_dreq == NULL)
896 		wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
897 					      req->wb_index);
898 	else
899 		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
900 
901 	pnfs_generic_pg_init_write(pgio, req, wb_size);
902 
903 	if (pgio->pg_lseg &&
904 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
905 
906 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
907 		pnfs_set_lo_fail(pgio->pg_lseg);
908 		nfs_pageio_reset_write_mds(pgio);
909 	}
910 }
911 
912 /*
913  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
914  * of bytes (maximum @req->wb_bytes) that can be coalesced.
915  */
916 static size_t
917 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
918 		 struct nfs_page *req)
919 {
920 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
921 		return 0;
922 	return pnfs_generic_pg_test(pgio, prev, req);
923 }
924 
925 static const struct nfs_pageio_ops bl_pg_read_ops = {
926 	.pg_init = bl_pg_init_read,
927 	.pg_test = bl_pg_test_read,
928 	.pg_doio = pnfs_generic_pg_readpages,
929 	.pg_cleanup = pnfs_generic_pg_cleanup,
930 };
931 
932 static const struct nfs_pageio_ops bl_pg_write_ops = {
933 	.pg_init = bl_pg_init_write,
934 	.pg_test = bl_pg_test_write,
935 	.pg_doio = pnfs_generic_pg_writepages,
936 	.pg_cleanup = pnfs_generic_pg_cleanup,
937 };
938 
939 static struct pnfs_layoutdriver_type blocklayout_type = {
940 	.id				= LAYOUT_BLOCK_VOLUME,
941 	.name				= "LAYOUT_BLOCK_VOLUME",
942 	.owner				= THIS_MODULE,
943 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
944 					  PNFS_LAYOUTRET_ON_ERROR |
945 					  PNFS_READ_WHOLE_PAGE,
946 	.read_pagelist			= bl_read_pagelist,
947 	.write_pagelist			= bl_write_pagelist,
948 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
949 	.free_layout_hdr		= bl_free_layout_hdr,
950 	.alloc_lseg			= bl_alloc_lseg,
951 	.free_lseg			= bl_free_lseg,
952 	.return_range			= bl_return_range,
953 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
954 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
955 	.set_layoutdriver		= bl_set_layoutdriver,
956 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
957 	.free_deviceid_node		= bl_free_deviceid_node,
958 	.pg_read_ops			= &bl_pg_read_ops,
959 	.pg_write_ops			= &bl_pg_write_ops,
960 	.sync				= pnfs_generic_sync,
961 };
962 
963 static struct pnfs_layoutdriver_type scsilayout_type = {
964 	.id				= LAYOUT_SCSI,
965 	.name				= "LAYOUT_SCSI",
966 	.owner				= THIS_MODULE,
967 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
968 					  PNFS_LAYOUTRET_ON_ERROR |
969 					  PNFS_READ_WHOLE_PAGE,
970 	.read_pagelist			= bl_read_pagelist,
971 	.write_pagelist			= bl_write_pagelist,
972 	.alloc_layout_hdr		= sl_alloc_layout_hdr,
973 	.free_layout_hdr		= bl_free_layout_hdr,
974 	.alloc_lseg			= bl_alloc_lseg,
975 	.free_lseg			= bl_free_lseg,
976 	.return_range			= bl_return_range,
977 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
978 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
979 	.set_layoutdriver		= bl_set_layoutdriver,
980 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
981 	.free_deviceid_node		= bl_free_deviceid_node,
982 	.pg_read_ops			= &bl_pg_read_ops,
983 	.pg_write_ops			= &bl_pg_write_ops,
984 	.sync				= pnfs_generic_sync,
985 };
986 
987 
988 static int __init nfs4blocklayout_init(void)
989 {
990 	int ret;
991 
992 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
993 
994 	ret = bl_init_pipefs();
995 	if (ret)
996 		goto out;
997 
998 	ret = pnfs_register_layoutdriver(&blocklayout_type);
999 	if (ret)
1000 		goto out_cleanup_pipe;
1001 
1002 	ret = pnfs_register_layoutdriver(&scsilayout_type);
1003 	if (ret)
1004 		goto out_unregister_block;
1005 	return 0;
1006 
1007 out_unregister_block:
1008 	pnfs_unregister_layoutdriver(&blocklayout_type);
1009 out_cleanup_pipe:
1010 	bl_cleanup_pipefs();
1011 out:
1012 	return ret;
1013 }
1014 
1015 static void __exit nfs4blocklayout_exit(void)
1016 {
1017 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1018 	       __func__);
1019 
1020 	pnfs_unregister_layoutdriver(&scsilayout_type);
1021 	pnfs_unregister_layoutdriver(&blocklayout_type);
1022 	bl_cleanup_pipefs();
1023 }
1024 
1025 MODULE_ALIAS("nfs-layouttype4-3");
1026 MODULE_ALIAS("nfs-layouttype4-5");
1027 
1028 module_init(nfs4blocklayout_init);
1029 module_exit(nfs4blocklayout_exit);
1030