xref: /openbmc/linux/fs/nfs/blocklayout/blocklayout.c (revision e3d786a3)
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 struct bio *
119 bl_alloc_init_bio(int npg, struct block_device *bdev, sector_t disk_sector,
120 		bio_end_io_t end_io, struct parallel_io *par)
121 {
122 	struct bio *bio;
123 
124 	npg = min(npg, BIO_MAX_PAGES);
125 	bio = bio_alloc(GFP_NOIO, npg);
126 	if (!bio && (current->flags & PF_MEMALLOC)) {
127 		while (!bio && (npg /= 2))
128 			bio = bio_alloc(GFP_NOIO, npg);
129 	}
130 
131 	if (bio) {
132 		bio->bi_iter.bi_sector = disk_sector;
133 		bio_set_dev(bio, bdev);
134 		bio->bi_end_io = end_io;
135 		bio->bi_private = par;
136 	}
137 	return bio;
138 }
139 
140 static bool offset_in_map(u64 offset, struct pnfs_block_dev_map *map)
141 {
142 	return offset >= map->start && offset < map->start + map->len;
143 }
144 
145 static struct bio *
146 do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect,
147 		struct page *page, struct pnfs_block_dev_map *map,
148 		struct pnfs_block_extent *be, bio_end_io_t end_io,
149 		struct parallel_io *par, unsigned int offset, int *len)
150 {
151 	struct pnfs_block_dev *dev =
152 		container_of(be->be_device, struct pnfs_block_dev, node);
153 	u64 disk_addr, end;
154 
155 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
156 		npg, rw, (unsigned long long)isect, offset, *len);
157 
158 	/* translate to device offset */
159 	isect += be->be_v_offset;
160 	isect -= be->be_f_offset;
161 
162 	/* translate to physical disk offset */
163 	disk_addr = (u64)isect << SECTOR_SHIFT;
164 	if (!offset_in_map(disk_addr, map)) {
165 		if (!dev->map(dev, disk_addr, map) || !offset_in_map(disk_addr, map))
166 			return ERR_PTR(-EIO);
167 		bio = bl_submit_bio(bio);
168 	}
169 	disk_addr += map->disk_offset;
170 	disk_addr -= map->start;
171 
172 	/* limit length to what the device mapping allows */
173 	end = disk_addr + *len;
174 	if (end >= map->start + map->len)
175 		*len = map->start + map->len - disk_addr;
176 
177 retry:
178 	if (!bio) {
179 		bio = bl_alloc_init_bio(npg, map->bdev,
180 				disk_addr >> SECTOR_SHIFT, end_io, par);
181 		if (!bio)
182 			return ERR_PTR(-ENOMEM);
183 		bio_set_op_attrs(bio, rw, 0);
184 	}
185 	if (bio_add_page(bio, page, *len, offset) < *len) {
186 		bio = bl_submit_bio(bio);
187 		goto retry;
188 	}
189 	return bio;
190 }
191 
192 static void bl_mark_devices_unavailable(struct nfs_pgio_header *header, bool rw)
193 {
194 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
195 	size_t bytes_left = header->args.count;
196 	sector_t isect, extent_length = 0;
197 	struct pnfs_block_extent be;
198 
199 	isect = header->args.offset >> SECTOR_SHIFT;
200 	bytes_left += header->args.offset - (isect << SECTOR_SHIFT);
201 
202 	while (bytes_left > 0) {
203 		if (!ext_tree_lookup(bl, isect, &be, rw))
204 				return;
205 		extent_length = be.be_length - (isect - be.be_f_offset);
206 		nfs4_mark_deviceid_unavailable(be.be_device);
207 		isect += extent_length;
208 		if (bytes_left > extent_length << SECTOR_SHIFT)
209 			bytes_left -= extent_length << SECTOR_SHIFT;
210 		else
211 			bytes_left = 0;
212 	}
213 }
214 
215 static void bl_end_io_read(struct bio *bio)
216 {
217 	struct parallel_io *par = bio->bi_private;
218 
219 	if (bio->bi_status) {
220 		struct nfs_pgio_header *header = par->data;
221 
222 		if (!header->pnfs_error)
223 			header->pnfs_error = -EIO;
224 		pnfs_set_lo_fail(header->lseg);
225 		bl_mark_devices_unavailable(header, false);
226 	}
227 
228 	bio_put(bio);
229 	put_parallel(par);
230 }
231 
232 static void bl_read_cleanup(struct work_struct *work)
233 {
234 	struct rpc_task *task;
235 	struct nfs_pgio_header *hdr;
236 	dprintk("%s enter\n", __func__);
237 	task = container_of(work, struct rpc_task, u.tk_work);
238 	hdr = container_of(task, struct nfs_pgio_header, task);
239 	pnfs_ld_read_done(hdr);
240 }
241 
242 static void
243 bl_end_par_io_read(void *data)
244 {
245 	struct nfs_pgio_header *hdr = data;
246 
247 	hdr->task.tk_status = hdr->pnfs_error;
248 	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
249 	schedule_work(&hdr->task.u.tk_work);
250 }
251 
252 static enum pnfs_try_status
253 bl_read_pagelist(struct nfs_pgio_header *header)
254 {
255 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
256 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
257 	struct bio *bio = NULL;
258 	struct pnfs_block_extent be;
259 	sector_t isect, extent_length = 0;
260 	struct parallel_io *par;
261 	loff_t f_offset = header->args.offset;
262 	size_t bytes_left = header->args.count;
263 	unsigned int pg_offset = header->args.pgbase, pg_len;
264 	struct page **pages = header->args.pages;
265 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
266 	const bool is_dio = (header->dreq != NULL);
267 	struct blk_plug plug;
268 	int i;
269 
270 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
271 		header->page_array.npages, f_offset,
272 		(unsigned int)header->args.count);
273 
274 	par = alloc_parallel(header);
275 	if (!par)
276 		return PNFS_NOT_ATTEMPTED;
277 	par->pnfs_callback = bl_end_par_io_read;
278 
279 	blk_start_plug(&plug);
280 
281 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
282 	/* Code assumes extents are page-aligned */
283 	for (i = pg_index; i < header->page_array.npages; i++) {
284 		if (extent_length <= 0) {
285 			/* We've used up the previous extent */
286 			bio = bl_submit_bio(bio);
287 
288 			/* Get the next one */
289 			if (!ext_tree_lookup(bl, isect, &be, false)) {
290 				header->pnfs_error = -EIO;
291 				goto out;
292 			}
293 			extent_length = be.be_length - (isect - be.be_f_offset);
294 		}
295 
296 		if (is_dio) {
297 			if (pg_offset + bytes_left > PAGE_SIZE)
298 				pg_len = PAGE_SIZE - pg_offset;
299 			else
300 				pg_len = bytes_left;
301 		} else {
302 			BUG_ON(pg_offset != 0);
303 			pg_len = PAGE_SIZE;
304 		}
305 
306 		if (is_hole(&be)) {
307 			bio = bl_submit_bio(bio);
308 			/* Fill hole w/ zeroes w/o accessing device */
309 			dprintk("%s Zeroing page for hole\n", __func__);
310 			zero_user_segment(pages[i], pg_offset, pg_len);
311 
312 			/* invalidate map */
313 			map.start = NFS4_MAX_UINT64;
314 		} else {
315 			bio = do_add_page_to_bio(bio,
316 						 header->page_array.npages - i,
317 						 READ,
318 						 isect, pages[i], &map, &be,
319 						 bl_end_io_read, par,
320 						 pg_offset, &pg_len);
321 			if (IS_ERR(bio)) {
322 				header->pnfs_error = PTR_ERR(bio);
323 				bio = NULL;
324 				goto out;
325 			}
326 		}
327 		isect += (pg_len >> SECTOR_SHIFT);
328 		extent_length -= (pg_len >> SECTOR_SHIFT);
329 		f_offset += pg_len;
330 		bytes_left -= pg_len;
331 		pg_offset = 0;
332 	}
333 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
334 		header->res.eof = 1;
335 		header->res.count = header->inode->i_size - header->args.offset;
336 	} else {
337 		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
338 	}
339 out:
340 	bl_submit_bio(bio);
341 	blk_finish_plug(&plug);
342 	put_parallel(par);
343 	return PNFS_ATTEMPTED;
344 }
345 
346 static void bl_end_io_write(struct bio *bio)
347 {
348 	struct parallel_io *par = bio->bi_private;
349 	struct nfs_pgio_header *header = par->data;
350 
351 	if (bio->bi_status) {
352 		if (!header->pnfs_error)
353 			header->pnfs_error = -EIO;
354 		pnfs_set_lo_fail(header->lseg);
355 		bl_mark_devices_unavailable(header, true);
356 	}
357 	bio_put(bio);
358 	put_parallel(par);
359 }
360 
361 /* Function scheduled for call during bl_end_par_io_write,
362  * it marks sectors as written and extends the commitlist.
363  */
364 static void bl_write_cleanup(struct work_struct *work)
365 {
366 	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
367 	struct nfs_pgio_header *hdr =
368 			container_of(task, struct nfs_pgio_header, task);
369 
370 	dprintk("%s enter\n", __func__);
371 
372 	if (likely(!hdr->pnfs_error)) {
373 		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
374 		u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
375 		u64 end = (hdr->args.offset + hdr->args.count +
376 			PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
377 		u64 lwb = hdr->args.offset + hdr->args.count;
378 
379 		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
380 					(end - start) >> SECTOR_SHIFT, lwb);
381 	}
382 
383 	pnfs_ld_write_done(hdr);
384 }
385 
386 /* Called when last of bios associated with a bl_write_pagelist call finishes */
387 static void bl_end_par_io_write(void *data)
388 {
389 	struct nfs_pgio_header *hdr = data;
390 
391 	hdr->task.tk_status = hdr->pnfs_error;
392 	hdr->verf.committed = NFS_FILE_SYNC;
393 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
394 	schedule_work(&hdr->task.u.tk_work);
395 }
396 
397 static enum pnfs_try_status
398 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
399 {
400 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
401 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
402 	struct bio *bio = NULL;
403 	struct pnfs_block_extent be;
404 	sector_t isect, extent_length = 0;
405 	struct parallel_io *par = NULL;
406 	loff_t offset = header->args.offset;
407 	size_t count = header->args.count;
408 	struct page **pages = header->args.pages;
409 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
410 	unsigned int pg_len;
411 	struct blk_plug plug;
412 	int i;
413 
414 	dprintk("%s enter, %zu@%lld\n", __func__, count, offset);
415 
416 	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
417 	 * We want to write each, and if there is an error set pnfs_error
418 	 * to have it redone using nfs.
419 	 */
420 	par = alloc_parallel(header);
421 	if (!par)
422 		return PNFS_NOT_ATTEMPTED;
423 	par->pnfs_callback = bl_end_par_io_write;
424 
425 	blk_start_plug(&plug);
426 
427 	/* we always write out the whole page */
428 	offset = offset & (loff_t)PAGE_MASK;
429 	isect = offset >> SECTOR_SHIFT;
430 
431 	for (i = pg_index; i < header->page_array.npages; i++) {
432 		if (extent_length <= 0) {
433 			/* We've used up the previous extent */
434 			bio = bl_submit_bio(bio);
435 			/* Get the next one */
436 			if (!ext_tree_lookup(bl, isect, &be, true)) {
437 				header->pnfs_error = -EINVAL;
438 				goto out;
439 			}
440 
441 			extent_length = be.be_length - (isect - be.be_f_offset);
442 		}
443 
444 		pg_len = PAGE_SIZE;
445 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
446 					 WRITE, isect, pages[i], &map, &be,
447 					 bl_end_io_write, par,
448 					 0, &pg_len);
449 		if (IS_ERR(bio)) {
450 			header->pnfs_error = PTR_ERR(bio);
451 			bio = NULL;
452 			goto out;
453 		}
454 
455 		offset += pg_len;
456 		count -= pg_len;
457 		isect += (pg_len >> SECTOR_SHIFT);
458 		extent_length -= (pg_len >> SECTOR_SHIFT);
459 	}
460 
461 	header->res.count = header->args.count;
462 out:
463 	bl_submit_bio(bio);
464 	blk_finish_plug(&plug);
465 	put_parallel(par);
466 	return PNFS_ATTEMPTED;
467 }
468 
469 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
470 {
471 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
472 	int err;
473 
474 	dprintk("%s enter\n", __func__);
475 
476 	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
477 	WARN_ON(err);
478 
479 	kfree(bl);
480 }
481 
482 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
483 		gfp_t gfp_flags, bool is_scsi_layout)
484 {
485 	struct pnfs_block_layout *bl;
486 
487 	dprintk("%s enter\n", __func__);
488 	bl = kzalloc(sizeof(*bl), gfp_flags);
489 	if (!bl)
490 		return NULL;
491 
492 	bl->bl_ext_rw = RB_ROOT;
493 	bl->bl_ext_ro = RB_ROOT;
494 	spin_lock_init(&bl->bl_ext_lock);
495 
496 	bl->bl_scsi_layout = is_scsi_layout;
497 	return &bl->bl_layout;
498 }
499 
500 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
501 						   gfp_t gfp_flags)
502 {
503 	return __bl_alloc_layout_hdr(inode, gfp_flags, false);
504 }
505 
506 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
507 						   gfp_t gfp_flags)
508 {
509 	return __bl_alloc_layout_hdr(inode, gfp_flags, true);
510 }
511 
512 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
513 {
514 	dprintk("%s enter\n", __func__);
515 	kfree(lseg);
516 }
517 
518 /* Tracks info needed to ensure extents in layout obey constraints of spec */
519 struct layout_verification {
520 	u32 mode;	/* R or RW */
521 	u64 start;	/* Expected start of next non-COW extent */
522 	u64 inval;	/* Start of INVAL coverage */
523 	u64 cowread;	/* End of COW read coverage */
524 };
525 
526 /* Verify the extent meets the layout requirements of the pnfs-block draft,
527  * section 2.3.1.
528  */
529 static int verify_extent(struct pnfs_block_extent *be,
530 			 struct layout_verification *lv)
531 {
532 	if (lv->mode == IOMODE_READ) {
533 		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
534 		    be->be_state == PNFS_BLOCK_INVALID_DATA)
535 			return -EIO;
536 		if (be->be_f_offset != lv->start)
537 			return -EIO;
538 		lv->start += be->be_length;
539 		return 0;
540 	}
541 	/* lv->mode == IOMODE_RW */
542 	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
543 		if (be->be_f_offset != lv->start)
544 			return -EIO;
545 		if (lv->cowread > lv->start)
546 			return -EIO;
547 		lv->start += be->be_length;
548 		lv->inval = lv->start;
549 		return 0;
550 	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
551 		if (be->be_f_offset != lv->start)
552 			return -EIO;
553 		lv->start += be->be_length;
554 		return 0;
555 	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
556 		if (be->be_f_offset > lv->start)
557 			return -EIO;
558 		if (be->be_f_offset < lv->inval)
559 			return -EIO;
560 		if (be->be_f_offset < lv->cowread)
561 			return -EIO;
562 		/* It looks like you might want to min this with lv->start,
563 		 * but you really don't.
564 		 */
565 		lv->inval = lv->inval + be->be_length;
566 		lv->cowread = be->be_f_offset + be->be_length;
567 		return 0;
568 	} else
569 		return -EIO;
570 }
571 
572 static int decode_sector_number(__be32 **rp, sector_t *sp)
573 {
574 	uint64_t s;
575 
576 	*rp = xdr_decode_hyper(*rp, &s);
577 	if (s & 0x1ff) {
578 		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
579 		return -1;
580 	}
581 	*sp = s >> SECTOR_SHIFT;
582 	return 0;
583 }
584 
585 static struct nfs4_deviceid_node *
586 bl_find_get_deviceid(struct nfs_server *server,
587 		const struct nfs4_deviceid *id, struct rpc_cred *cred,
588 		gfp_t gfp_mask)
589 {
590 	struct nfs4_deviceid_node *node;
591 	unsigned long start, end;
592 
593 retry:
594 	node = nfs4_find_get_deviceid(server, id, cred, gfp_mask);
595 	if (!node)
596 		return ERR_PTR(-ENODEV);
597 
598 	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags) == 0)
599 		return node;
600 
601 	end = jiffies;
602 	start = end - PNFS_DEVICE_RETRY_TIMEOUT;
603 	if (!time_in_range(node->timestamp_unavailable, start, end)) {
604 		nfs4_delete_deviceid(node->ld, node->nfs_client, id);
605 		goto retry;
606 	}
607 	return ERR_PTR(-ENODEV);
608 }
609 
610 static int
611 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
612 		struct layout_verification *lv, struct list_head *extents,
613 		gfp_t gfp_mask)
614 {
615 	struct pnfs_block_extent *be;
616 	struct nfs4_deviceid id;
617 	int error;
618 	__be32 *p;
619 
620 	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
621 	if (!p)
622 		return -EIO;
623 
624 	be = kzalloc(sizeof(*be), GFP_NOFS);
625 	if (!be)
626 		return -ENOMEM;
627 
628 	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
629 	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
630 
631 	be->be_device = bl_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
632 						lo->plh_lc_cred, gfp_mask);
633 	if (IS_ERR(be->be_device)) {
634 		error = PTR_ERR(be->be_device);
635 		goto out_free_be;
636 	}
637 
638 	/*
639 	 * The next three values are read in as bytes, but stored in the
640 	 * extent structure in 512-byte granularity.
641 	 */
642 	error = -EIO;
643 	if (decode_sector_number(&p, &be->be_f_offset) < 0)
644 		goto out_put_deviceid;
645 	if (decode_sector_number(&p, &be->be_length) < 0)
646 		goto out_put_deviceid;
647 	if (decode_sector_number(&p, &be->be_v_offset) < 0)
648 		goto out_put_deviceid;
649 	be->be_state = be32_to_cpup(p++);
650 
651 	error = verify_extent(be, lv);
652 	if (error) {
653 		dprintk("%s: extent verification failed\n", __func__);
654 		goto out_put_deviceid;
655 	}
656 
657 	list_add_tail(&be->be_list, extents);
658 	return 0;
659 
660 out_put_deviceid:
661 	nfs4_put_deviceid_node(be->be_device);
662 out_free_be:
663 	kfree(be);
664 	return error;
665 }
666 
667 static struct pnfs_layout_segment *
668 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
669 		gfp_t gfp_mask)
670 {
671 	struct layout_verification lv = {
672 		.mode = lgr->range.iomode,
673 		.start = lgr->range.offset >> SECTOR_SHIFT,
674 		.inval = lgr->range.offset >> SECTOR_SHIFT,
675 		.cowread = lgr->range.offset >> SECTOR_SHIFT,
676 	};
677 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
678 	struct pnfs_layout_segment *lseg;
679 	struct xdr_buf buf;
680 	struct xdr_stream xdr;
681 	struct page *scratch;
682 	int status, i;
683 	uint32_t count;
684 	__be32 *p;
685 	LIST_HEAD(extents);
686 
687 	dprintk("---> %s\n", __func__);
688 
689 	lseg = kzalloc(sizeof(*lseg), gfp_mask);
690 	if (!lseg)
691 		return ERR_PTR(-ENOMEM);
692 
693 	status = -ENOMEM;
694 	scratch = alloc_page(gfp_mask);
695 	if (!scratch)
696 		goto out;
697 
698 	xdr_init_decode_pages(&xdr, &buf,
699 			lgr->layoutp->pages, lgr->layoutp->len);
700 	xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
701 
702 	status = -EIO;
703 	p = xdr_inline_decode(&xdr, 4);
704 	if (unlikely(!p))
705 		goto out_free_scratch;
706 
707 	count = be32_to_cpup(p++);
708 	dprintk("%s: number of extents %d\n", __func__, count);
709 
710 	/*
711 	 * Decode individual extents, putting them in temporary staging area
712 	 * until whole layout is decoded to make error recovery easier.
713 	 */
714 	for (i = 0; i < count; i++) {
715 		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
716 		if (status)
717 			goto process_extents;
718 	}
719 
720 	if (lgr->range.offset + lgr->range.length !=
721 			lv.start << SECTOR_SHIFT) {
722 		dprintk("%s Final length mismatch\n", __func__);
723 		status = -EIO;
724 		goto process_extents;
725 	}
726 
727 	if (lv.start < lv.cowread) {
728 		dprintk("%s Final uncovered COW extent\n", __func__);
729 		status = -EIO;
730 	}
731 
732 process_extents:
733 	while (!list_empty(&extents)) {
734 		struct pnfs_block_extent *be =
735 			list_first_entry(&extents, struct pnfs_block_extent,
736 					 be_list);
737 		list_del(&be->be_list);
738 
739 		if (!status)
740 			status = ext_tree_insert(bl, be);
741 
742 		if (status) {
743 			nfs4_put_deviceid_node(be->be_device);
744 			kfree(be);
745 		}
746 	}
747 
748 out_free_scratch:
749 	__free_page(scratch);
750 out:
751 	dprintk("%s returns %d\n", __func__, status);
752 	switch (status) {
753 	case -ENODEV:
754 		/* Our extent block devices are unavailable */
755 		set_bit(NFS_LSEG_UNAVAILABLE, &lseg->pls_flags);
756 		/* Fall through */
757 	case 0:
758 		return lseg;
759 	default:
760 		kfree(lseg);
761 		return ERR_PTR(status);
762 	}
763 }
764 
765 static void
766 bl_return_range(struct pnfs_layout_hdr *lo,
767 		struct pnfs_layout_range *range)
768 {
769 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
770 	sector_t offset = range->offset >> SECTOR_SHIFT, end;
771 
772 	if (range->offset % 8) {
773 		dprintk("%s: offset %lld not block size aligned\n",
774 			__func__, range->offset);
775 		return;
776 	}
777 
778 	if (range->length != NFS4_MAX_UINT64) {
779 		if (range->length % 8) {
780 			dprintk("%s: length %lld not block size aligned\n",
781 				__func__, range->length);
782 			return;
783 		}
784 
785 		end = offset + (range->length >> SECTOR_SHIFT);
786 	} else {
787 		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
788 	}
789 
790 	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
791 }
792 
793 static int
794 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
795 {
796 	return ext_tree_prepare_commit(arg);
797 }
798 
799 static void
800 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
801 {
802 	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
803 }
804 
805 static int
806 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
807 {
808 	dprintk("%s enter\n", __func__);
809 
810 	if (server->pnfs_blksize == 0) {
811 		dprintk("%s Server did not return blksize\n", __func__);
812 		return -EINVAL;
813 	}
814 	if (server->pnfs_blksize > PAGE_SIZE) {
815 		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
816 			__func__, server->pnfs_blksize);
817 		return -EINVAL;
818 	}
819 
820 	return 0;
821 }
822 
823 static bool
824 is_aligned_req(struct nfs_pageio_descriptor *pgio,
825 		struct nfs_page *req, unsigned int alignment, bool is_write)
826 {
827 	/*
828 	 * Always accept buffered writes, higher layers take care of the
829 	 * right alignment.
830 	 */
831 	if (pgio->pg_dreq == NULL)
832 		return true;
833 
834 	if (!IS_ALIGNED(req->wb_offset, alignment))
835 		return false;
836 
837 	if (IS_ALIGNED(req->wb_bytes, alignment))
838 		return true;
839 
840 	if (is_write &&
841 	    (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
842 		/*
843 		 * If the write goes up to the inode size, just write
844 		 * the full page.  Data past the inode size is
845 		 * guaranteed to be zeroed by the higher level client
846 		 * code, and this behaviour is mandated by RFC 5663
847 		 * section 2.3.2.
848 		 */
849 		return true;
850 	}
851 
852 	return false;
853 }
854 
855 static void
856 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
857 {
858 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
859 		nfs_pageio_reset_read_mds(pgio);
860 		return;
861 	}
862 
863 	pnfs_generic_pg_init_read(pgio, req);
864 
865 	if (pgio->pg_lseg &&
866 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
867 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
868 		pnfs_set_lo_fail(pgio->pg_lseg);
869 		nfs_pageio_reset_read_mds(pgio);
870 	}
871 }
872 
873 /*
874  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
875  * of bytes (maximum @req->wb_bytes) that can be coalesced.
876  */
877 static size_t
878 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
879 		struct nfs_page *req)
880 {
881 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
882 		return 0;
883 	return pnfs_generic_pg_test(pgio, prev, req);
884 }
885 
886 /*
887  * Return the number of contiguous bytes for a given inode
888  * starting at page frame idx.
889  */
890 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
891 {
892 	struct address_space *mapping = inode->i_mapping;
893 	pgoff_t end;
894 
895 	/* Optimize common case that writes from 0 to end of file */
896 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
897 	if (end != inode->i_mapping->nrpages) {
898 		rcu_read_lock();
899 		end = page_cache_next_miss(mapping, idx + 1, ULONG_MAX);
900 		rcu_read_unlock();
901 	}
902 
903 	if (!end)
904 		return i_size_read(inode) - (idx << PAGE_SHIFT);
905 	else
906 		return (end - idx) << PAGE_SHIFT;
907 }
908 
909 static void
910 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
911 {
912 	u64 wb_size;
913 
914 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
915 		nfs_pageio_reset_write_mds(pgio);
916 		return;
917 	}
918 
919 	if (pgio->pg_dreq == NULL)
920 		wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
921 					      req->wb_index);
922 	else
923 		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
924 
925 	pnfs_generic_pg_init_write(pgio, req, wb_size);
926 
927 	if (pgio->pg_lseg &&
928 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
929 
930 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
931 		pnfs_set_lo_fail(pgio->pg_lseg);
932 		nfs_pageio_reset_write_mds(pgio);
933 	}
934 }
935 
936 /*
937  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
938  * of bytes (maximum @req->wb_bytes) that can be coalesced.
939  */
940 static size_t
941 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
942 		 struct nfs_page *req)
943 {
944 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
945 		return 0;
946 	return pnfs_generic_pg_test(pgio, prev, req);
947 }
948 
949 static const struct nfs_pageio_ops bl_pg_read_ops = {
950 	.pg_init = bl_pg_init_read,
951 	.pg_test = bl_pg_test_read,
952 	.pg_doio = pnfs_generic_pg_readpages,
953 	.pg_cleanup = pnfs_generic_pg_cleanup,
954 };
955 
956 static const struct nfs_pageio_ops bl_pg_write_ops = {
957 	.pg_init = bl_pg_init_write,
958 	.pg_test = bl_pg_test_write,
959 	.pg_doio = pnfs_generic_pg_writepages,
960 	.pg_cleanup = pnfs_generic_pg_cleanup,
961 };
962 
963 static struct pnfs_layoutdriver_type blocklayout_type = {
964 	.id				= LAYOUT_BLOCK_VOLUME,
965 	.name				= "LAYOUT_BLOCK_VOLUME",
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		= bl_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 static struct pnfs_layoutdriver_type scsilayout_type = {
988 	.id				= LAYOUT_SCSI,
989 	.name				= "LAYOUT_SCSI",
990 	.owner				= THIS_MODULE,
991 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
992 					  PNFS_LAYOUTRET_ON_ERROR |
993 					  PNFS_READ_WHOLE_PAGE,
994 	.read_pagelist			= bl_read_pagelist,
995 	.write_pagelist			= bl_write_pagelist,
996 	.alloc_layout_hdr		= sl_alloc_layout_hdr,
997 	.free_layout_hdr		= bl_free_layout_hdr,
998 	.alloc_lseg			= bl_alloc_lseg,
999 	.free_lseg			= bl_free_lseg,
1000 	.return_range			= bl_return_range,
1001 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
1002 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
1003 	.set_layoutdriver		= bl_set_layoutdriver,
1004 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
1005 	.free_deviceid_node		= bl_free_deviceid_node,
1006 	.pg_read_ops			= &bl_pg_read_ops,
1007 	.pg_write_ops			= &bl_pg_write_ops,
1008 	.sync				= pnfs_generic_sync,
1009 };
1010 
1011 
1012 static int __init nfs4blocklayout_init(void)
1013 {
1014 	int ret;
1015 
1016 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1017 
1018 	ret = bl_init_pipefs();
1019 	if (ret)
1020 		goto out;
1021 
1022 	ret = pnfs_register_layoutdriver(&blocklayout_type);
1023 	if (ret)
1024 		goto out_cleanup_pipe;
1025 
1026 	ret = pnfs_register_layoutdriver(&scsilayout_type);
1027 	if (ret)
1028 		goto out_unregister_block;
1029 	return 0;
1030 
1031 out_unregister_block:
1032 	pnfs_unregister_layoutdriver(&blocklayout_type);
1033 out_cleanup_pipe:
1034 	bl_cleanup_pipefs();
1035 out:
1036 	return ret;
1037 }
1038 
1039 static void __exit nfs4blocklayout_exit(void)
1040 {
1041 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1042 	       __func__);
1043 
1044 	pnfs_unregister_layoutdriver(&scsilayout_type);
1045 	pnfs_unregister_layoutdriver(&blocklayout_type);
1046 	bl_cleanup_pipefs();
1047 }
1048 
1049 MODULE_ALIAS("nfs-layouttype4-3");
1050 MODULE_ALIAS("nfs-layouttype4-5");
1051 
1052 module_init(nfs4blocklayout_init);
1053 module_exit(nfs4blocklayout_exit);
1054