xref: /linux/fs/nfs/blocklayout/blocklayout.c (revision 8c749ce93ee69e789e46b3be98de9e0cbfcf8ed8)
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(int rw, struct bio *bio)
106 {
107 	if (bio) {
108 		get_parallel(bio->bi_private);
109 		dprintk("%s submitting %s bio %u@%llu\n", __func__,
110 			rw == READ ? "read" : "write", bio->bi_iter.bi_size,
111 			(unsigned long long)bio->bi_iter.bi_sector);
112 		submit_bio(rw, bio);
113 	}
114 	return NULL;
115 }
116 
117 static struct bio *
118 bl_alloc_init_bio(int npg, struct block_device *bdev, sector_t disk_sector,
119 		bio_end_io_t end_io, struct parallel_io *par)
120 {
121 	struct bio *bio;
122 
123 	npg = min(npg, BIO_MAX_PAGES);
124 	bio = bio_alloc(GFP_NOIO, npg);
125 	if (!bio && (current->flags & PF_MEMALLOC)) {
126 		while (!bio && (npg /= 2))
127 			bio = bio_alloc(GFP_NOIO, npg);
128 	}
129 
130 	if (bio) {
131 		bio->bi_iter.bi_sector = disk_sector;
132 		bio->bi_bdev = bdev;
133 		bio->bi_end_io = end_io;
134 		bio->bi_private = par;
135 	}
136 	return bio;
137 }
138 
139 static struct bio *
140 do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect,
141 		struct page *page, struct pnfs_block_dev_map *map,
142 		struct pnfs_block_extent *be, bio_end_io_t end_io,
143 		struct parallel_io *par, unsigned int offset, int *len)
144 {
145 	struct pnfs_block_dev *dev =
146 		container_of(be->be_device, struct pnfs_block_dev, node);
147 	u64 disk_addr, end;
148 
149 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
150 		npg, rw, (unsigned long long)isect, offset, *len);
151 
152 	/* translate to device offset */
153 	isect += be->be_v_offset;
154 	isect -= be->be_f_offset;
155 
156 	/* translate to physical disk offset */
157 	disk_addr = (u64)isect << SECTOR_SHIFT;
158 	if (disk_addr < map->start || disk_addr >= map->start + map->len) {
159 		if (!dev->map(dev, disk_addr, map))
160 			return ERR_PTR(-EIO);
161 		bio = bl_submit_bio(rw, bio);
162 	}
163 	disk_addr += map->disk_offset;
164 	disk_addr -= map->start;
165 
166 	/* limit length to what the device mapping allows */
167 	end = disk_addr + *len;
168 	if (end >= map->start + map->len)
169 		*len = map->start + map->len - disk_addr;
170 
171 retry:
172 	if (!bio) {
173 		bio = bl_alloc_init_bio(npg, map->bdev,
174 				disk_addr >> SECTOR_SHIFT, end_io, par);
175 		if (!bio)
176 			return ERR_PTR(-ENOMEM);
177 	}
178 	if (bio_add_page(bio, page, *len, offset) < *len) {
179 		bio = bl_submit_bio(rw, bio);
180 		goto retry;
181 	}
182 	return bio;
183 }
184 
185 static void bl_end_io_read(struct bio *bio)
186 {
187 	struct parallel_io *par = bio->bi_private;
188 
189 	if (bio->bi_error) {
190 		struct nfs_pgio_header *header = par->data;
191 
192 		if (!header->pnfs_error)
193 			header->pnfs_error = -EIO;
194 		pnfs_set_lo_fail(header->lseg);
195 	}
196 
197 	bio_put(bio);
198 	put_parallel(par);
199 }
200 
201 static void bl_read_cleanup(struct work_struct *work)
202 {
203 	struct rpc_task *task;
204 	struct nfs_pgio_header *hdr;
205 	dprintk("%s enter\n", __func__);
206 	task = container_of(work, struct rpc_task, u.tk_work);
207 	hdr = container_of(task, struct nfs_pgio_header, task);
208 	pnfs_ld_read_done(hdr);
209 }
210 
211 static void
212 bl_end_par_io_read(void *data)
213 {
214 	struct nfs_pgio_header *hdr = data;
215 
216 	hdr->task.tk_status = hdr->pnfs_error;
217 	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
218 	schedule_work(&hdr->task.u.tk_work);
219 }
220 
221 static enum pnfs_try_status
222 bl_read_pagelist(struct nfs_pgio_header *header)
223 {
224 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
225 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
226 	struct bio *bio = NULL;
227 	struct pnfs_block_extent be;
228 	sector_t isect, extent_length = 0;
229 	struct parallel_io *par;
230 	loff_t f_offset = header->args.offset;
231 	size_t bytes_left = header->args.count;
232 	unsigned int pg_offset = header->args.pgbase, pg_len;
233 	struct page **pages = header->args.pages;
234 	int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
235 	const bool is_dio = (header->dreq != NULL);
236 	struct blk_plug plug;
237 	int i;
238 
239 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
240 		header->page_array.npages, f_offset,
241 		(unsigned int)header->args.count);
242 
243 	par = alloc_parallel(header);
244 	if (!par)
245 		return PNFS_NOT_ATTEMPTED;
246 	par->pnfs_callback = bl_end_par_io_read;
247 
248 	blk_start_plug(&plug);
249 
250 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
251 	/* Code assumes extents are page-aligned */
252 	for (i = pg_index; i < header->page_array.npages; i++) {
253 		if (extent_length <= 0) {
254 			/* We've used up the previous extent */
255 			bio = bl_submit_bio(READ, bio);
256 
257 			/* Get the next one */
258 			if (!ext_tree_lookup(bl, isect, &be, false)) {
259 				header->pnfs_error = -EIO;
260 				goto out;
261 			}
262 			extent_length = be.be_length - (isect - be.be_f_offset);
263 		}
264 
265 		if (is_dio) {
266 			if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
267 				pg_len = PAGE_CACHE_SIZE - pg_offset;
268 			else
269 				pg_len = bytes_left;
270 		} else {
271 			BUG_ON(pg_offset != 0);
272 			pg_len = PAGE_CACHE_SIZE;
273 		}
274 
275 		if (is_hole(&be)) {
276 			bio = bl_submit_bio(READ, bio);
277 			/* Fill hole w/ zeroes w/o accessing device */
278 			dprintk("%s Zeroing page for hole\n", __func__);
279 			zero_user_segment(pages[i], pg_offset, pg_len);
280 
281 			/* invalidate map */
282 			map.start = NFS4_MAX_UINT64;
283 		} else {
284 			bio = do_add_page_to_bio(bio,
285 						 header->page_array.npages - i,
286 						 READ,
287 						 isect, pages[i], &map, &be,
288 						 bl_end_io_read, par,
289 						 pg_offset, &pg_len);
290 			if (IS_ERR(bio)) {
291 				header->pnfs_error = PTR_ERR(bio);
292 				bio = NULL;
293 				goto out;
294 			}
295 		}
296 		isect += (pg_len >> SECTOR_SHIFT);
297 		extent_length -= (pg_len >> SECTOR_SHIFT);
298 		f_offset += pg_len;
299 		bytes_left -= pg_len;
300 		pg_offset = 0;
301 	}
302 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
303 		header->res.eof = 1;
304 		header->res.count = header->inode->i_size - header->args.offset;
305 	} else {
306 		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
307 	}
308 out:
309 	bl_submit_bio(READ, bio);
310 	blk_finish_plug(&plug);
311 	put_parallel(par);
312 	return PNFS_ATTEMPTED;
313 }
314 
315 static void bl_end_io_write(struct bio *bio)
316 {
317 	struct parallel_io *par = bio->bi_private;
318 	struct nfs_pgio_header *header = par->data;
319 
320 	if (bio->bi_error) {
321 		if (!header->pnfs_error)
322 			header->pnfs_error = -EIO;
323 		pnfs_set_lo_fail(header->lseg);
324 	}
325 	bio_put(bio);
326 	put_parallel(par);
327 }
328 
329 /* Function scheduled for call during bl_end_par_io_write,
330  * it marks sectors as written and extends the commitlist.
331  */
332 static void bl_write_cleanup(struct work_struct *work)
333 {
334 	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
335 	struct nfs_pgio_header *hdr =
336 			container_of(task, struct nfs_pgio_header, task);
337 
338 	dprintk("%s enter\n", __func__);
339 
340 	if (likely(!hdr->pnfs_error)) {
341 		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
342 		u64 start = hdr->args.offset & (loff_t)PAGE_CACHE_MASK;
343 		u64 end = (hdr->args.offset + hdr->args.count +
344 			PAGE_CACHE_SIZE - 1) & (loff_t)PAGE_CACHE_MASK;
345 
346 		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
347 					(end - start) >> SECTOR_SHIFT);
348 	}
349 
350 	pnfs_ld_write_done(hdr);
351 }
352 
353 /* Called when last of bios associated with a bl_write_pagelist call finishes */
354 static void bl_end_par_io_write(void *data)
355 {
356 	struct nfs_pgio_header *hdr = data;
357 
358 	hdr->task.tk_status = hdr->pnfs_error;
359 	hdr->verf.committed = NFS_FILE_SYNC;
360 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
361 	schedule_work(&hdr->task.u.tk_work);
362 }
363 
364 static enum pnfs_try_status
365 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
366 {
367 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
368 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
369 	struct bio *bio = NULL;
370 	struct pnfs_block_extent be;
371 	sector_t isect, extent_length = 0;
372 	struct parallel_io *par = NULL;
373 	loff_t offset = header->args.offset;
374 	size_t count = header->args.count;
375 	struct page **pages = header->args.pages;
376 	int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
377 	unsigned int pg_len;
378 	struct blk_plug plug;
379 	int i;
380 
381 	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
382 
383 	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
384 	 * We want to write each, and if there is an error set pnfs_error
385 	 * to have it redone using nfs.
386 	 */
387 	par = alloc_parallel(header);
388 	if (!par)
389 		return PNFS_NOT_ATTEMPTED;
390 	par->pnfs_callback = bl_end_par_io_write;
391 
392 	blk_start_plug(&plug);
393 
394 	/* we always write out the whole page */
395 	offset = offset & (loff_t)PAGE_CACHE_MASK;
396 	isect = offset >> SECTOR_SHIFT;
397 
398 	for (i = pg_index; i < header->page_array.npages; i++) {
399 		if (extent_length <= 0) {
400 			/* We've used up the previous extent */
401 			bio = bl_submit_bio(WRITE, bio);
402 			/* Get the next one */
403 			if (!ext_tree_lookup(bl, isect, &be, true)) {
404 				header->pnfs_error = -EINVAL;
405 				goto out;
406 			}
407 
408 			extent_length = be.be_length - (isect - be.be_f_offset);
409 		}
410 
411 		pg_len = PAGE_CACHE_SIZE;
412 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
413 					 WRITE, isect, pages[i], &map, &be,
414 					 bl_end_io_write, par,
415 					 0, &pg_len);
416 		if (IS_ERR(bio)) {
417 			header->pnfs_error = PTR_ERR(bio);
418 			bio = NULL;
419 			goto out;
420 		}
421 
422 		offset += pg_len;
423 		count -= pg_len;
424 		isect += (pg_len >> SECTOR_SHIFT);
425 		extent_length -= (pg_len >> SECTOR_SHIFT);
426 	}
427 
428 	header->res.count = header->args.count;
429 out:
430 	bl_submit_bio(WRITE, bio);
431 	blk_finish_plug(&plug);
432 	put_parallel(par);
433 	return PNFS_ATTEMPTED;
434 }
435 
436 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
437 {
438 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
439 	int err;
440 
441 	dprintk("%s enter\n", __func__);
442 
443 	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
444 	WARN_ON(err);
445 
446 	kfree(bl);
447 }
448 
449 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
450 						   gfp_t gfp_flags)
451 {
452 	struct pnfs_block_layout *bl;
453 
454 	dprintk("%s enter\n", __func__);
455 	bl = kzalloc(sizeof(*bl), gfp_flags);
456 	if (!bl)
457 		return NULL;
458 
459 	bl->bl_ext_rw = RB_ROOT;
460 	bl->bl_ext_ro = RB_ROOT;
461 	spin_lock_init(&bl->bl_ext_lock);
462 
463 	return &bl->bl_layout;
464 }
465 
466 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
467 {
468 	dprintk("%s enter\n", __func__);
469 	kfree(lseg);
470 }
471 
472 /* Tracks info needed to ensure extents in layout obey constraints of spec */
473 struct layout_verification {
474 	u32 mode;	/* R or RW */
475 	u64 start;	/* Expected start of next non-COW extent */
476 	u64 inval;	/* Start of INVAL coverage */
477 	u64 cowread;	/* End of COW read coverage */
478 };
479 
480 /* Verify the extent meets the layout requirements of the pnfs-block draft,
481  * section 2.3.1.
482  */
483 static int verify_extent(struct pnfs_block_extent *be,
484 			 struct layout_verification *lv)
485 {
486 	if (lv->mode == IOMODE_READ) {
487 		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
488 		    be->be_state == PNFS_BLOCK_INVALID_DATA)
489 			return -EIO;
490 		if (be->be_f_offset != lv->start)
491 			return -EIO;
492 		lv->start += be->be_length;
493 		return 0;
494 	}
495 	/* lv->mode == IOMODE_RW */
496 	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
497 		if (be->be_f_offset != lv->start)
498 			return -EIO;
499 		if (lv->cowread > lv->start)
500 			return -EIO;
501 		lv->start += be->be_length;
502 		lv->inval = lv->start;
503 		return 0;
504 	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
505 		if (be->be_f_offset != lv->start)
506 			return -EIO;
507 		lv->start += be->be_length;
508 		return 0;
509 	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
510 		if (be->be_f_offset > lv->start)
511 			return -EIO;
512 		if (be->be_f_offset < lv->inval)
513 			return -EIO;
514 		if (be->be_f_offset < lv->cowread)
515 			return -EIO;
516 		/* It looks like you might want to min this with lv->start,
517 		 * but you really don't.
518 		 */
519 		lv->inval = lv->inval + be->be_length;
520 		lv->cowread = be->be_f_offset + be->be_length;
521 		return 0;
522 	} else
523 		return -EIO;
524 }
525 
526 static int decode_sector_number(__be32 **rp, sector_t *sp)
527 {
528 	uint64_t s;
529 
530 	*rp = xdr_decode_hyper(*rp, &s);
531 	if (s & 0x1ff) {
532 		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
533 		return -1;
534 	}
535 	*sp = s >> SECTOR_SHIFT;
536 	return 0;
537 }
538 
539 static int
540 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
541 		struct layout_verification *lv, struct list_head *extents,
542 		gfp_t gfp_mask)
543 {
544 	struct pnfs_block_extent *be;
545 	struct nfs4_deviceid id;
546 	int error;
547 	__be32 *p;
548 
549 	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
550 	if (!p)
551 		return -EIO;
552 
553 	be = kzalloc(sizeof(*be), GFP_NOFS);
554 	if (!be)
555 		return -ENOMEM;
556 
557 	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
558 	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
559 
560 	error = -EIO;
561 	be->be_device = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
562 						lo->plh_lc_cred, gfp_mask);
563 	if (!be->be_device)
564 		goto out_free_be;
565 
566 	/*
567 	 * The next three values are read in as bytes, but stored in the
568 	 * extent structure in 512-byte granularity.
569 	 */
570 	if (decode_sector_number(&p, &be->be_f_offset) < 0)
571 		goto out_put_deviceid;
572 	if (decode_sector_number(&p, &be->be_length) < 0)
573 		goto out_put_deviceid;
574 	if (decode_sector_number(&p, &be->be_v_offset) < 0)
575 		goto out_put_deviceid;
576 	be->be_state = be32_to_cpup(p++);
577 
578 	error = verify_extent(be, lv);
579 	if (error) {
580 		dprintk("%s: extent verification failed\n", __func__);
581 		goto out_put_deviceid;
582 	}
583 
584 	list_add_tail(&be->be_list, extents);
585 	return 0;
586 
587 out_put_deviceid:
588 	nfs4_put_deviceid_node(be->be_device);
589 out_free_be:
590 	kfree(be);
591 	return error;
592 }
593 
594 static struct pnfs_layout_segment *
595 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
596 		gfp_t gfp_mask)
597 {
598 	struct layout_verification lv = {
599 		.mode = lgr->range.iomode,
600 		.start = lgr->range.offset >> SECTOR_SHIFT,
601 		.inval = lgr->range.offset >> SECTOR_SHIFT,
602 		.cowread = lgr->range.offset >> SECTOR_SHIFT,
603 	};
604 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
605 	struct pnfs_layout_segment *lseg;
606 	struct xdr_buf buf;
607 	struct xdr_stream xdr;
608 	struct page *scratch;
609 	int status, i;
610 	uint32_t count;
611 	__be32 *p;
612 	LIST_HEAD(extents);
613 
614 	dprintk("---> %s\n", __func__);
615 
616 	lseg = kzalloc(sizeof(*lseg), gfp_mask);
617 	if (!lseg)
618 		return ERR_PTR(-ENOMEM);
619 
620 	status = -ENOMEM;
621 	scratch = alloc_page(gfp_mask);
622 	if (!scratch)
623 		goto out;
624 
625 	xdr_init_decode_pages(&xdr, &buf,
626 			lgr->layoutp->pages, lgr->layoutp->len);
627 	xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
628 
629 	status = -EIO;
630 	p = xdr_inline_decode(&xdr, 4);
631 	if (unlikely(!p))
632 		goto out_free_scratch;
633 
634 	count = be32_to_cpup(p++);
635 	dprintk("%s: number of extents %d\n", __func__, count);
636 
637 	/*
638 	 * Decode individual extents, putting them in temporary staging area
639 	 * until whole layout is decoded to make error recovery easier.
640 	 */
641 	for (i = 0; i < count; i++) {
642 		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
643 		if (status)
644 			goto process_extents;
645 	}
646 
647 	if (lgr->range.offset + lgr->range.length !=
648 			lv.start << SECTOR_SHIFT) {
649 		dprintk("%s Final length mismatch\n", __func__);
650 		status = -EIO;
651 		goto process_extents;
652 	}
653 
654 	if (lv.start < lv.cowread) {
655 		dprintk("%s Final uncovered COW extent\n", __func__);
656 		status = -EIO;
657 	}
658 
659 process_extents:
660 	while (!list_empty(&extents)) {
661 		struct pnfs_block_extent *be =
662 			list_first_entry(&extents, struct pnfs_block_extent,
663 					 be_list);
664 		list_del(&be->be_list);
665 
666 		if (!status)
667 			status = ext_tree_insert(bl, be);
668 
669 		if (status) {
670 			nfs4_put_deviceid_node(be->be_device);
671 			kfree(be);
672 		}
673 	}
674 
675 out_free_scratch:
676 	__free_page(scratch);
677 out:
678 	dprintk("%s returns %d\n", __func__, status);
679 	if (status) {
680 		kfree(lseg);
681 		return ERR_PTR(status);
682 	}
683 	return lseg;
684 }
685 
686 static void
687 bl_return_range(struct pnfs_layout_hdr *lo,
688 		struct pnfs_layout_range *range)
689 {
690 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
691 	sector_t offset = range->offset >> SECTOR_SHIFT, end;
692 
693 	if (range->offset % 8) {
694 		dprintk("%s: offset %lld not block size aligned\n",
695 			__func__, range->offset);
696 		return;
697 	}
698 
699 	if (range->length != NFS4_MAX_UINT64) {
700 		if (range->length % 8) {
701 			dprintk("%s: length %lld not block size aligned\n",
702 				__func__, range->length);
703 			return;
704 		}
705 
706 		end = offset + (range->length >> SECTOR_SHIFT);
707 	} else {
708 		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
709 	}
710 
711 	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
712 }
713 
714 static int
715 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
716 {
717 	return ext_tree_prepare_commit(arg);
718 }
719 
720 static void
721 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
722 {
723 	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
724 }
725 
726 static int
727 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
728 {
729 	dprintk("%s enter\n", __func__);
730 
731 	if (server->pnfs_blksize == 0) {
732 		dprintk("%s Server did not return blksize\n", __func__);
733 		return -EINVAL;
734 	}
735 	if (server->pnfs_blksize > PAGE_SIZE) {
736 		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
737 			__func__, server->pnfs_blksize);
738 		return -EINVAL;
739 	}
740 
741 	return 0;
742 }
743 
744 static bool
745 is_aligned_req(struct nfs_pageio_descriptor *pgio,
746 		struct nfs_page *req, unsigned int alignment)
747 {
748 	/*
749 	 * Always accept buffered writes, higher layers take care of the
750 	 * right alignment.
751 	 */
752 	if (pgio->pg_dreq == NULL)
753 		return true;
754 
755 	if (!IS_ALIGNED(req->wb_offset, alignment))
756 		return false;
757 
758 	if (IS_ALIGNED(req->wb_bytes, alignment))
759 		return true;
760 
761 	if (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode)) {
762 		/*
763 		 * If the write goes up to the inode size, just write
764 		 * the full page.  Data past the inode size is
765 		 * guaranteed to be zeroed by the higher level client
766 		 * code, and this behaviour is mandated by RFC 5663
767 		 * section 2.3.2.
768 		 */
769 		return true;
770 	}
771 
772 	return false;
773 }
774 
775 static void
776 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
777 {
778 	if (!is_aligned_req(pgio, req, SECTOR_SIZE)) {
779 		nfs_pageio_reset_read_mds(pgio);
780 		return;
781 	}
782 
783 	pnfs_generic_pg_init_read(pgio, req);
784 }
785 
786 /*
787  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
788  * of bytes (maximum @req->wb_bytes) that can be coalesced.
789  */
790 static size_t
791 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
792 		struct nfs_page *req)
793 {
794 	if (!is_aligned_req(pgio, req, SECTOR_SIZE))
795 		return 0;
796 	return pnfs_generic_pg_test(pgio, prev, req);
797 }
798 
799 /*
800  * Return the number of contiguous bytes for a given inode
801  * starting at page frame idx.
802  */
803 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
804 {
805 	struct address_space *mapping = inode->i_mapping;
806 	pgoff_t end;
807 
808 	/* Optimize common case that writes from 0 to end of file */
809 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
810 	if (end != inode->i_mapping->nrpages) {
811 		rcu_read_lock();
812 		end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
813 		rcu_read_unlock();
814 	}
815 
816 	if (!end)
817 		return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
818 	else
819 		return (end - idx) << PAGE_CACHE_SHIFT;
820 }
821 
822 static void
823 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
824 {
825 	u64 wb_size;
826 
827 	if (!is_aligned_req(pgio, req, PAGE_SIZE)) {
828 		nfs_pageio_reset_write_mds(pgio);
829 		return;
830 	}
831 
832 	if (pgio->pg_dreq == NULL)
833 		wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
834 					      req->wb_index);
835 	else
836 		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
837 
838 	pnfs_generic_pg_init_write(pgio, req, wb_size);
839 }
840 
841 /*
842  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
843  * of bytes (maximum @req->wb_bytes) that can be coalesced.
844  */
845 static size_t
846 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
847 		 struct nfs_page *req)
848 {
849 	if (!is_aligned_req(pgio, req, PAGE_SIZE))
850 		return 0;
851 	return pnfs_generic_pg_test(pgio, prev, req);
852 }
853 
854 static const struct nfs_pageio_ops bl_pg_read_ops = {
855 	.pg_init = bl_pg_init_read,
856 	.pg_test = bl_pg_test_read,
857 	.pg_doio = pnfs_generic_pg_readpages,
858 	.pg_cleanup = pnfs_generic_pg_cleanup,
859 };
860 
861 static const struct nfs_pageio_ops bl_pg_write_ops = {
862 	.pg_init = bl_pg_init_write,
863 	.pg_test = bl_pg_test_write,
864 	.pg_doio = pnfs_generic_pg_writepages,
865 	.pg_cleanup = pnfs_generic_pg_cleanup,
866 };
867 
868 static struct pnfs_layoutdriver_type blocklayout_type = {
869 	.id				= LAYOUT_BLOCK_VOLUME,
870 	.name				= "LAYOUT_BLOCK_VOLUME",
871 	.owner				= THIS_MODULE,
872 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
873 					  PNFS_READ_WHOLE_PAGE,
874 	.read_pagelist			= bl_read_pagelist,
875 	.write_pagelist			= bl_write_pagelist,
876 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
877 	.free_layout_hdr		= bl_free_layout_hdr,
878 	.alloc_lseg			= bl_alloc_lseg,
879 	.free_lseg			= bl_free_lseg,
880 	.return_range			= bl_return_range,
881 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
882 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
883 	.set_layoutdriver		= bl_set_layoutdriver,
884 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
885 	.free_deviceid_node		= bl_free_deviceid_node,
886 	.pg_read_ops			= &bl_pg_read_ops,
887 	.pg_write_ops			= &bl_pg_write_ops,
888 	.sync				= pnfs_generic_sync,
889 };
890 
891 static int __init nfs4blocklayout_init(void)
892 {
893 	int ret;
894 
895 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
896 
897 	ret = pnfs_register_layoutdriver(&blocklayout_type);
898 	if (ret)
899 		goto out;
900 	ret = bl_init_pipefs();
901 	if (ret)
902 		goto out_unregister;
903 	return 0;
904 
905 out_unregister:
906 	pnfs_unregister_layoutdriver(&blocklayout_type);
907 out:
908 	return ret;
909 }
910 
911 static void __exit nfs4blocklayout_exit(void)
912 {
913 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
914 	       __func__);
915 
916 	bl_cleanup_pipefs();
917 	pnfs_unregister_layoutdriver(&blocklayout_type);
918 }
919 
920 MODULE_ALIAS("nfs-layouttype4-3");
921 
922 module_init(nfs4blocklayout_init);
923 module_exit(nfs4blocklayout_exit);
924