xref: /openbmc/linux/fs/nfs/blocklayout/blocklayout.c (revision 4f6cce39)
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->bi_bdev = bdev;
134 		bio->bi_end_io = end_io;
135 		bio->bi_private = par;
136 	}
137 	return bio;
138 }
139 
140 static struct bio *
141 do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect,
142 		struct page *page, struct pnfs_block_dev_map *map,
143 		struct pnfs_block_extent *be, bio_end_io_t end_io,
144 		struct parallel_io *par, unsigned int offset, int *len)
145 {
146 	struct pnfs_block_dev *dev =
147 		container_of(be->be_device, struct pnfs_block_dev, node);
148 	u64 disk_addr, end;
149 
150 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
151 		npg, rw, (unsigned long long)isect, offset, *len);
152 
153 	/* translate to device offset */
154 	isect += be->be_v_offset;
155 	isect -= be->be_f_offset;
156 
157 	/* translate to physical disk offset */
158 	disk_addr = (u64)isect << SECTOR_SHIFT;
159 	if (disk_addr < map->start || disk_addr >= map->start + map->len) {
160 		if (!dev->map(dev, disk_addr, map))
161 			return ERR_PTR(-EIO);
162 		bio = bl_submit_bio(bio);
163 	}
164 	disk_addr += map->disk_offset;
165 	disk_addr -= map->start;
166 
167 	/* limit length to what the device mapping allows */
168 	end = disk_addr + *len;
169 	if (end >= map->start + map->len)
170 		*len = map->start + map->len - disk_addr;
171 
172 retry:
173 	if (!bio) {
174 		bio = bl_alloc_init_bio(npg, map->bdev,
175 				disk_addr >> SECTOR_SHIFT, end_io, par);
176 		if (!bio)
177 			return ERR_PTR(-ENOMEM);
178 		bio_set_op_attrs(bio, rw, 0);
179 	}
180 	if (bio_add_page(bio, page, *len, offset) < *len) {
181 		bio = bl_submit_bio(bio);
182 		goto retry;
183 	}
184 	return bio;
185 }
186 
187 static void bl_end_io_read(struct bio *bio)
188 {
189 	struct parallel_io *par = bio->bi_private;
190 
191 	if (bio->bi_error) {
192 		struct nfs_pgio_header *header = par->data;
193 
194 		if (!header->pnfs_error)
195 			header->pnfs_error = -EIO;
196 		pnfs_set_lo_fail(header->lseg);
197 	}
198 
199 	bio_put(bio);
200 	put_parallel(par);
201 }
202 
203 static void bl_read_cleanup(struct work_struct *work)
204 {
205 	struct rpc_task *task;
206 	struct nfs_pgio_header *hdr;
207 	dprintk("%s enter\n", __func__);
208 	task = container_of(work, struct rpc_task, u.tk_work);
209 	hdr = container_of(task, struct nfs_pgio_header, task);
210 	pnfs_ld_read_done(hdr);
211 }
212 
213 static void
214 bl_end_par_io_read(void *data)
215 {
216 	struct nfs_pgio_header *hdr = data;
217 
218 	hdr->task.tk_status = hdr->pnfs_error;
219 	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
220 	schedule_work(&hdr->task.u.tk_work);
221 }
222 
223 static enum pnfs_try_status
224 bl_read_pagelist(struct nfs_pgio_header *header)
225 {
226 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
227 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
228 	struct bio *bio = NULL;
229 	struct pnfs_block_extent be;
230 	sector_t isect, extent_length = 0;
231 	struct parallel_io *par;
232 	loff_t f_offset = header->args.offset;
233 	size_t bytes_left = header->args.count;
234 	unsigned int pg_offset = header->args.pgbase, pg_len;
235 	struct page **pages = header->args.pages;
236 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
237 	const bool is_dio = (header->dreq != NULL);
238 	struct blk_plug plug;
239 	int i;
240 
241 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
242 		header->page_array.npages, f_offset,
243 		(unsigned int)header->args.count);
244 
245 	par = alloc_parallel(header);
246 	if (!par)
247 		return PNFS_NOT_ATTEMPTED;
248 	par->pnfs_callback = bl_end_par_io_read;
249 
250 	blk_start_plug(&plug);
251 
252 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
253 	/* Code assumes extents are page-aligned */
254 	for (i = pg_index; i < header->page_array.npages; i++) {
255 		if (extent_length <= 0) {
256 			/* We've used up the previous extent */
257 			bio = bl_submit_bio(bio);
258 
259 			/* Get the next one */
260 			if (!ext_tree_lookup(bl, isect, &be, false)) {
261 				header->pnfs_error = -EIO;
262 				goto out;
263 			}
264 			extent_length = be.be_length - (isect - be.be_f_offset);
265 		}
266 
267 		if (is_dio) {
268 			if (pg_offset + bytes_left > PAGE_SIZE)
269 				pg_len = PAGE_SIZE - pg_offset;
270 			else
271 				pg_len = bytes_left;
272 		} else {
273 			BUG_ON(pg_offset != 0);
274 			pg_len = PAGE_SIZE;
275 		}
276 
277 		if (is_hole(&be)) {
278 			bio = bl_submit_bio(bio);
279 			/* Fill hole w/ zeroes w/o accessing device */
280 			dprintk("%s Zeroing page for hole\n", __func__);
281 			zero_user_segment(pages[i], pg_offset, pg_len);
282 
283 			/* invalidate map */
284 			map.start = NFS4_MAX_UINT64;
285 		} else {
286 			bio = do_add_page_to_bio(bio,
287 						 header->page_array.npages - i,
288 						 READ,
289 						 isect, pages[i], &map, &be,
290 						 bl_end_io_read, par,
291 						 pg_offset, &pg_len);
292 			if (IS_ERR(bio)) {
293 				header->pnfs_error = PTR_ERR(bio);
294 				bio = NULL;
295 				goto out;
296 			}
297 		}
298 		isect += (pg_len >> SECTOR_SHIFT);
299 		extent_length -= (pg_len >> SECTOR_SHIFT);
300 		f_offset += pg_len;
301 		bytes_left -= pg_len;
302 		pg_offset = 0;
303 	}
304 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
305 		header->res.eof = 1;
306 		header->res.count = header->inode->i_size - header->args.offset;
307 	} else {
308 		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
309 	}
310 out:
311 	bl_submit_bio(bio);
312 	blk_finish_plug(&plug);
313 	put_parallel(par);
314 	return PNFS_ATTEMPTED;
315 }
316 
317 static void bl_end_io_write(struct bio *bio)
318 {
319 	struct parallel_io *par = bio->bi_private;
320 	struct nfs_pgio_header *header = par->data;
321 
322 	if (bio->bi_error) {
323 		if (!header->pnfs_error)
324 			header->pnfs_error = -EIO;
325 		pnfs_set_lo_fail(header->lseg);
326 	}
327 	bio_put(bio);
328 	put_parallel(par);
329 }
330 
331 /* Function scheduled for call during bl_end_par_io_write,
332  * it marks sectors as written and extends the commitlist.
333  */
334 static void bl_write_cleanup(struct work_struct *work)
335 {
336 	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
337 	struct nfs_pgio_header *hdr =
338 			container_of(task, struct nfs_pgio_header, task);
339 
340 	dprintk("%s enter\n", __func__);
341 
342 	if (likely(!hdr->pnfs_error)) {
343 		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
344 		u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
345 		u64 end = (hdr->args.offset + hdr->args.count +
346 			PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
347 		u64 lwb = hdr->args.offset + hdr->args.count;
348 
349 		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
350 					(end - start) >> SECTOR_SHIFT, lwb);
351 	}
352 
353 	pnfs_ld_write_done(hdr);
354 }
355 
356 /* Called when last of bios associated with a bl_write_pagelist call finishes */
357 static void bl_end_par_io_write(void *data)
358 {
359 	struct nfs_pgio_header *hdr = data;
360 
361 	hdr->task.tk_status = hdr->pnfs_error;
362 	hdr->verf.committed = NFS_FILE_SYNC;
363 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
364 	schedule_work(&hdr->task.u.tk_work);
365 }
366 
367 static enum pnfs_try_status
368 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
369 {
370 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
371 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
372 	struct bio *bio = NULL;
373 	struct pnfs_block_extent be;
374 	sector_t isect, extent_length = 0;
375 	struct parallel_io *par = NULL;
376 	loff_t offset = header->args.offset;
377 	size_t count = header->args.count;
378 	struct page **pages = header->args.pages;
379 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
380 	unsigned int pg_len;
381 	struct blk_plug plug;
382 	int i;
383 
384 	dprintk("%s enter, %zu@%lld\n", __func__, count, offset);
385 
386 	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
387 	 * We want to write each, and if there is an error set pnfs_error
388 	 * to have it redone using nfs.
389 	 */
390 	par = alloc_parallel(header);
391 	if (!par)
392 		return PNFS_NOT_ATTEMPTED;
393 	par->pnfs_callback = bl_end_par_io_write;
394 
395 	blk_start_plug(&plug);
396 
397 	/* we always write out the whole page */
398 	offset = offset & (loff_t)PAGE_MASK;
399 	isect = offset >> SECTOR_SHIFT;
400 
401 	for (i = pg_index; i < header->page_array.npages; i++) {
402 		if (extent_length <= 0) {
403 			/* We've used up the previous extent */
404 			bio = bl_submit_bio(bio);
405 			/* Get the next one */
406 			if (!ext_tree_lookup(bl, isect, &be, true)) {
407 				header->pnfs_error = -EINVAL;
408 				goto out;
409 			}
410 
411 			extent_length = be.be_length - (isect - be.be_f_offset);
412 		}
413 
414 		pg_len = PAGE_SIZE;
415 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
416 					 WRITE, isect, pages[i], &map, &be,
417 					 bl_end_io_write, par,
418 					 0, &pg_len);
419 		if (IS_ERR(bio)) {
420 			header->pnfs_error = PTR_ERR(bio);
421 			bio = NULL;
422 			goto out;
423 		}
424 
425 		offset += pg_len;
426 		count -= pg_len;
427 		isect += (pg_len >> SECTOR_SHIFT);
428 		extent_length -= (pg_len >> SECTOR_SHIFT);
429 	}
430 
431 	header->res.count = header->args.count;
432 out:
433 	bl_submit_bio(bio);
434 	blk_finish_plug(&plug);
435 	put_parallel(par);
436 	return PNFS_ATTEMPTED;
437 }
438 
439 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
440 {
441 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
442 	int err;
443 
444 	dprintk("%s enter\n", __func__);
445 
446 	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
447 	WARN_ON(err);
448 
449 	kfree(bl);
450 }
451 
452 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
453 		gfp_t gfp_flags, bool is_scsi_layout)
454 {
455 	struct pnfs_block_layout *bl;
456 
457 	dprintk("%s enter\n", __func__);
458 	bl = kzalloc(sizeof(*bl), gfp_flags);
459 	if (!bl)
460 		return NULL;
461 
462 	bl->bl_ext_rw = RB_ROOT;
463 	bl->bl_ext_ro = RB_ROOT;
464 	spin_lock_init(&bl->bl_ext_lock);
465 
466 	bl->bl_scsi_layout = is_scsi_layout;
467 	return &bl->bl_layout;
468 }
469 
470 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
471 						   gfp_t gfp_flags)
472 {
473 	return __bl_alloc_layout_hdr(inode, gfp_flags, false);
474 }
475 
476 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
477 						   gfp_t gfp_flags)
478 {
479 	return __bl_alloc_layout_hdr(inode, gfp_flags, true);
480 }
481 
482 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
483 {
484 	dprintk("%s enter\n", __func__);
485 	kfree(lseg);
486 }
487 
488 /* Tracks info needed to ensure extents in layout obey constraints of spec */
489 struct layout_verification {
490 	u32 mode;	/* R or RW */
491 	u64 start;	/* Expected start of next non-COW extent */
492 	u64 inval;	/* Start of INVAL coverage */
493 	u64 cowread;	/* End of COW read coverage */
494 };
495 
496 /* Verify the extent meets the layout requirements of the pnfs-block draft,
497  * section 2.3.1.
498  */
499 static int verify_extent(struct pnfs_block_extent *be,
500 			 struct layout_verification *lv)
501 {
502 	if (lv->mode == IOMODE_READ) {
503 		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
504 		    be->be_state == PNFS_BLOCK_INVALID_DATA)
505 			return -EIO;
506 		if (be->be_f_offset != lv->start)
507 			return -EIO;
508 		lv->start += be->be_length;
509 		return 0;
510 	}
511 	/* lv->mode == IOMODE_RW */
512 	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
513 		if (be->be_f_offset != lv->start)
514 			return -EIO;
515 		if (lv->cowread > lv->start)
516 			return -EIO;
517 		lv->start += be->be_length;
518 		lv->inval = lv->start;
519 		return 0;
520 	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
521 		if (be->be_f_offset != lv->start)
522 			return -EIO;
523 		lv->start += be->be_length;
524 		return 0;
525 	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
526 		if (be->be_f_offset > lv->start)
527 			return -EIO;
528 		if (be->be_f_offset < lv->inval)
529 			return -EIO;
530 		if (be->be_f_offset < lv->cowread)
531 			return -EIO;
532 		/* It looks like you might want to min this with lv->start,
533 		 * but you really don't.
534 		 */
535 		lv->inval = lv->inval + be->be_length;
536 		lv->cowread = be->be_f_offset + be->be_length;
537 		return 0;
538 	} else
539 		return -EIO;
540 }
541 
542 static int decode_sector_number(__be32 **rp, sector_t *sp)
543 {
544 	uint64_t s;
545 
546 	*rp = xdr_decode_hyper(*rp, &s);
547 	if (s & 0x1ff) {
548 		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
549 		return -1;
550 	}
551 	*sp = s >> SECTOR_SHIFT;
552 	return 0;
553 }
554 
555 static int
556 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
557 		struct layout_verification *lv, struct list_head *extents,
558 		gfp_t gfp_mask)
559 {
560 	struct pnfs_block_extent *be;
561 	struct nfs4_deviceid id;
562 	int error;
563 	__be32 *p;
564 
565 	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
566 	if (!p)
567 		return -EIO;
568 
569 	be = kzalloc(sizeof(*be), GFP_NOFS);
570 	if (!be)
571 		return -ENOMEM;
572 
573 	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
574 	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
575 
576 	error = -EIO;
577 	be->be_device = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
578 						lo->plh_lc_cred, gfp_mask);
579 	if (!be->be_device)
580 		goto out_free_be;
581 
582 	/*
583 	 * The next three values are read in as bytes, but stored in the
584 	 * extent structure in 512-byte granularity.
585 	 */
586 	if (decode_sector_number(&p, &be->be_f_offset) < 0)
587 		goto out_put_deviceid;
588 	if (decode_sector_number(&p, &be->be_length) < 0)
589 		goto out_put_deviceid;
590 	if (decode_sector_number(&p, &be->be_v_offset) < 0)
591 		goto out_put_deviceid;
592 	be->be_state = be32_to_cpup(p++);
593 
594 	error = verify_extent(be, lv);
595 	if (error) {
596 		dprintk("%s: extent verification failed\n", __func__);
597 		goto out_put_deviceid;
598 	}
599 
600 	list_add_tail(&be->be_list, extents);
601 	return 0;
602 
603 out_put_deviceid:
604 	nfs4_put_deviceid_node(be->be_device);
605 out_free_be:
606 	kfree(be);
607 	return error;
608 }
609 
610 static struct pnfs_layout_segment *
611 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
612 		gfp_t gfp_mask)
613 {
614 	struct layout_verification lv = {
615 		.mode = lgr->range.iomode,
616 		.start = lgr->range.offset >> SECTOR_SHIFT,
617 		.inval = lgr->range.offset >> SECTOR_SHIFT,
618 		.cowread = lgr->range.offset >> SECTOR_SHIFT,
619 	};
620 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
621 	struct pnfs_layout_segment *lseg;
622 	struct xdr_buf buf;
623 	struct xdr_stream xdr;
624 	struct page *scratch;
625 	int status, i;
626 	uint32_t count;
627 	__be32 *p;
628 	LIST_HEAD(extents);
629 
630 	dprintk("---> %s\n", __func__);
631 
632 	lseg = kzalloc(sizeof(*lseg), gfp_mask);
633 	if (!lseg)
634 		return ERR_PTR(-ENOMEM);
635 
636 	status = -ENOMEM;
637 	scratch = alloc_page(gfp_mask);
638 	if (!scratch)
639 		goto out;
640 
641 	xdr_init_decode_pages(&xdr, &buf,
642 			lgr->layoutp->pages, lgr->layoutp->len);
643 	xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
644 
645 	status = -EIO;
646 	p = xdr_inline_decode(&xdr, 4);
647 	if (unlikely(!p))
648 		goto out_free_scratch;
649 
650 	count = be32_to_cpup(p++);
651 	dprintk("%s: number of extents %d\n", __func__, count);
652 
653 	/*
654 	 * Decode individual extents, putting them in temporary staging area
655 	 * until whole layout is decoded to make error recovery easier.
656 	 */
657 	for (i = 0; i < count; i++) {
658 		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
659 		if (status)
660 			goto process_extents;
661 	}
662 
663 	if (lgr->range.offset + lgr->range.length !=
664 			lv.start << SECTOR_SHIFT) {
665 		dprintk("%s Final length mismatch\n", __func__);
666 		status = -EIO;
667 		goto process_extents;
668 	}
669 
670 	if (lv.start < lv.cowread) {
671 		dprintk("%s Final uncovered COW extent\n", __func__);
672 		status = -EIO;
673 	}
674 
675 process_extents:
676 	while (!list_empty(&extents)) {
677 		struct pnfs_block_extent *be =
678 			list_first_entry(&extents, struct pnfs_block_extent,
679 					 be_list);
680 		list_del(&be->be_list);
681 
682 		if (!status)
683 			status = ext_tree_insert(bl, be);
684 
685 		if (status) {
686 			nfs4_put_deviceid_node(be->be_device);
687 			kfree(be);
688 		}
689 	}
690 
691 out_free_scratch:
692 	__free_page(scratch);
693 out:
694 	dprintk("%s returns %d\n", __func__, status);
695 	if (status) {
696 		kfree(lseg);
697 		return ERR_PTR(status);
698 	}
699 	return lseg;
700 }
701 
702 static void
703 bl_return_range(struct pnfs_layout_hdr *lo,
704 		struct pnfs_layout_range *range)
705 {
706 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
707 	sector_t offset = range->offset >> SECTOR_SHIFT, end;
708 
709 	if (range->offset % 8) {
710 		dprintk("%s: offset %lld not block size aligned\n",
711 			__func__, range->offset);
712 		return;
713 	}
714 
715 	if (range->length != NFS4_MAX_UINT64) {
716 		if (range->length % 8) {
717 			dprintk("%s: length %lld not block size aligned\n",
718 				__func__, range->length);
719 			return;
720 		}
721 
722 		end = offset + (range->length >> SECTOR_SHIFT);
723 	} else {
724 		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
725 	}
726 
727 	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
728 }
729 
730 static int
731 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
732 {
733 	return ext_tree_prepare_commit(arg);
734 }
735 
736 static void
737 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
738 {
739 	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
740 }
741 
742 static int
743 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
744 {
745 	dprintk("%s enter\n", __func__);
746 
747 	if (server->pnfs_blksize == 0) {
748 		dprintk("%s Server did not return blksize\n", __func__);
749 		return -EINVAL;
750 	}
751 	if (server->pnfs_blksize > PAGE_SIZE) {
752 		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
753 			__func__, server->pnfs_blksize);
754 		return -EINVAL;
755 	}
756 
757 	return 0;
758 }
759 
760 static bool
761 is_aligned_req(struct nfs_pageio_descriptor *pgio,
762 		struct nfs_page *req, unsigned int alignment, bool is_write)
763 {
764 	/*
765 	 * Always accept buffered writes, higher layers take care of the
766 	 * right alignment.
767 	 */
768 	if (pgio->pg_dreq == NULL)
769 		return true;
770 
771 	if (!IS_ALIGNED(req->wb_offset, alignment))
772 		return false;
773 
774 	if (IS_ALIGNED(req->wb_bytes, alignment))
775 		return true;
776 
777 	if (is_write &&
778 	    (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
779 		/*
780 		 * If the write goes up to the inode size, just write
781 		 * the full page.  Data past the inode size is
782 		 * guaranteed to be zeroed by the higher level client
783 		 * code, and this behaviour is mandated by RFC 5663
784 		 * section 2.3.2.
785 		 */
786 		return true;
787 	}
788 
789 	return false;
790 }
791 
792 static void
793 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
794 {
795 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
796 		nfs_pageio_reset_read_mds(pgio);
797 		return;
798 	}
799 
800 	pnfs_generic_pg_init_read(pgio, req);
801 }
802 
803 /*
804  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
805  * of bytes (maximum @req->wb_bytes) that can be coalesced.
806  */
807 static size_t
808 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
809 		struct nfs_page *req)
810 {
811 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
812 		return 0;
813 	return pnfs_generic_pg_test(pgio, prev, req);
814 }
815 
816 /*
817  * Return the number of contiguous bytes for a given inode
818  * starting at page frame idx.
819  */
820 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
821 {
822 	struct address_space *mapping = inode->i_mapping;
823 	pgoff_t end;
824 
825 	/* Optimize common case that writes from 0 to end of file */
826 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
827 	if (end != inode->i_mapping->nrpages) {
828 		rcu_read_lock();
829 		end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
830 		rcu_read_unlock();
831 	}
832 
833 	if (!end)
834 		return i_size_read(inode) - (idx << PAGE_SHIFT);
835 	else
836 		return (end - idx) << PAGE_SHIFT;
837 }
838 
839 static void
840 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
841 {
842 	u64 wb_size;
843 
844 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
845 		nfs_pageio_reset_write_mds(pgio);
846 		return;
847 	}
848 
849 	if (pgio->pg_dreq == NULL)
850 		wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
851 					      req->wb_index);
852 	else
853 		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
854 
855 	pnfs_generic_pg_init_write(pgio, req, wb_size);
856 }
857 
858 /*
859  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
860  * of bytes (maximum @req->wb_bytes) that can be coalesced.
861  */
862 static size_t
863 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
864 		 struct nfs_page *req)
865 {
866 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
867 		return 0;
868 	return pnfs_generic_pg_test(pgio, prev, req);
869 }
870 
871 static const struct nfs_pageio_ops bl_pg_read_ops = {
872 	.pg_init = bl_pg_init_read,
873 	.pg_test = bl_pg_test_read,
874 	.pg_doio = pnfs_generic_pg_readpages,
875 	.pg_cleanup = pnfs_generic_pg_cleanup,
876 };
877 
878 static const struct nfs_pageio_ops bl_pg_write_ops = {
879 	.pg_init = bl_pg_init_write,
880 	.pg_test = bl_pg_test_write,
881 	.pg_doio = pnfs_generic_pg_writepages,
882 	.pg_cleanup = pnfs_generic_pg_cleanup,
883 };
884 
885 static struct pnfs_layoutdriver_type blocklayout_type = {
886 	.id				= LAYOUT_BLOCK_VOLUME,
887 	.name				= "LAYOUT_BLOCK_VOLUME",
888 	.owner				= THIS_MODULE,
889 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
890 					  PNFS_READ_WHOLE_PAGE,
891 	.read_pagelist			= bl_read_pagelist,
892 	.write_pagelist			= bl_write_pagelist,
893 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
894 	.free_layout_hdr		= bl_free_layout_hdr,
895 	.alloc_lseg			= bl_alloc_lseg,
896 	.free_lseg			= bl_free_lseg,
897 	.return_range			= bl_return_range,
898 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
899 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
900 	.set_layoutdriver		= bl_set_layoutdriver,
901 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
902 	.free_deviceid_node		= bl_free_deviceid_node,
903 	.pg_read_ops			= &bl_pg_read_ops,
904 	.pg_write_ops			= &bl_pg_write_ops,
905 	.sync				= pnfs_generic_sync,
906 };
907 
908 static struct pnfs_layoutdriver_type scsilayout_type = {
909 	.id				= LAYOUT_SCSI,
910 	.name				= "LAYOUT_SCSI",
911 	.owner				= THIS_MODULE,
912 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
913 					  PNFS_READ_WHOLE_PAGE,
914 	.read_pagelist			= bl_read_pagelist,
915 	.write_pagelist			= bl_write_pagelist,
916 	.alloc_layout_hdr		= sl_alloc_layout_hdr,
917 	.free_layout_hdr		= bl_free_layout_hdr,
918 	.alloc_lseg			= bl_alloc_lseg,
919 	.free_lseg			= bl_free_lseg,
920 	.return_range			= bl_return_range,
921 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
922 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
923 	.set_layoutdriver		= bl_set_layoutdriver,
924 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
925 	.free_deviceid_node		= bl_free_deviceid_node,
926 	.pg_read_ops			= &bl_pg_read_ops,
927 	.pg_write_ops			= &bl_pg_write_ops,
928 	.sync				= pnfs_generic_sync,
929 };
930 
931 
932 static int __init nfs4blocklayout_init(void)
933 {
934 	int ret;
935 
936 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
937 
938 	ret = bl_init_pipefs();
939 	if (ret)
940 		goto out;
941 
942 	ret = pnfs_register_layoutdriver(&blocklayout_type);
943 	if (ret)
944 		goto out_cleanup_pipe;
945 
946 	ret = pnfs_register_layoutdriver(&scsilayout_type);
947 	if (ret)
948 		goto out_unregister_block;
949 	return 0;
950 
951 out_unregister_block:
952 	pnfs_unregister_layoutdriver(&blocklayout_type);
953 out_cleanup_pipe:
954 	bl_cleanup_pipefs();
955 out:
956 	return ret;
957 }
958 
959 static void __exit nfs4blocklayout_exit(void)
960 {
961 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
962 	       __func__);
963 
964 	pnfs_unregister_layoutdriver(&scsilayout_type);
965 	pnfs_unregister_layoutdriver(&blocklayout_type);
966 	bl_cleanup_pipefs();
967 }
968 
969 MODULE_ALIAS("nfs-layouttype4-3");
970 
971 module_init(nfs4blocklayout_init);
972 module_exit(nfs4blocklayout_exit);
973