xref: /openbmc/linux/fs/nfs/blocklayout/blocklayout.c (revision 05bcf503)
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/buffer_head.h>	/* various write calls */
39 #include <linux/prefetch.h>
40 #include <linux/pagevec.h>
41 
42 #include "../pnfs.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 void print_page(struct page *page)
53 {
54 	dprintk("PRINTPAGE page %p\n", page);
55 	dprintk("	PagePrivate %d\n", PagePrivate(page));
56 	dprintk("	PageUptodate %d\n", PageUptodate(page));
57 	dprintk("	PageError %d\n", PageError(page));
58 	dprintk("	PageDirty %d\n", PageDirty(page));
59 	dprintk("	PageReferenced %d\n", PageReferenced(page));
60 	dprintk("	PageLocked %d\n", PageLocked(page));
61 	dprintk("	PageWriteback %d\n", PageWriteback(page));
62 	dprintk("	PageMappedToDisk %d\n", PageMappedToDisk(page));
63 	dprintk("\n");
64 }
65 
66 /* Given the be associated with isect, determine if page data needs to be
67  * initialized.
68  */
69 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
70 {
71 	if (be->be_state == PNFS_BLOCK_NONE_DATA)
72 		return 1;
73 	else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
74 		return 0;
75 	else
76 		return !bl_is_sector_init(be->be_inval, isect);
77 }
78 
79 /* Given the be associated with isect, determine if page data can be
80  * written to disk.
81  */
82 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
83 {
84 	return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
85 		be->be_state == PNFS_BLOCK_INVALID_DATA);
86 }
87 
88 /* The data we are handed might be spread across several bios.  We need
89  * to track when the last one is finished.
90  */
91 struct parallel_io {
92 	struct kref refcnt;
93 	void (*pnfs_callback) (void *data, int num_se);
94 	void *data;
95 	int bse_count;
96 };
97 
98 static inline struct parallel_io *alloc_parallel(void *data)
99 {
100 	struct parallel_io *rv;
101 
102 	rv  = kmalloc(sizeof(*rv), GFP_NOFS);
103 	if (rv) {
104 		rv->data = data;
105 		kref_init(&rv->refcnt);
106 		rv->bse_count = 0;
107 	}
108 	return rv;
109 }
110 
111 static inline void get_parallel(struct parallel_io *p)
112 {
113 	kref_get(&p->refcnt);
114 }
115 
116 static void destroy_parallel(struct kref *kref)
117 {
118 	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
119 
120 	dprintk("%s enter\n", __func__);
121 	p->pnfs_callback(p->data, p->bse_count);
122 	kfree(p);
123 }
124 
125 static inline void put_parallel(struct parallel_io *p)
126 {
127 	kref_put(&p->refcnt, destroy_parallel);
128 }
129 
130 static struct bio *
131 bl_submit_bio(int rw, struct bio *bio)
132 {
133 	if (bio) {
134 		get_parallel(bio->bi_private);
135 		dprintk("%s submitting %s bio %u@%llu\n", __func__,
136 			rw == READ ? "read" : "write",
137 			bio->bi_size, (unsigned long long)bio->bi_sector);
138 		submit_bio(rw, bio);
139 	}
140 	return NULL;
141 }
142 
143 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
144 				     struct pnfs_block_extent *be,
145 				     void (*end_io)(struct bio *, int err),
146 				     struct parallel_io *par)
147 {
148 	struct bio *bio;
149 
150 	npg = min(npg, BIO_MAX_PAGES);
151 	bio = bio_alloc(GFP_NOIO, npg);
152 	if (!bio && (current->flags & PF_MEMALLOC)) {
153 		while (!bio && (npg /= 2))
154 			bio = bio_alloc(GFP_NOIO, npg);
155 	}
156 
157 	if (bio) {
158 		bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
159 		bio->bi_bdev = be->be_mdev;
160 		bio->bi_end_io = end_io;
161 		bio->bi_private = par;
162 	}
163 	return bio;
164 }
165 
166 static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw,
167 				      sector_t isect, struct page *page,
168 				      struct pnfs_block_extent *be,
169 				      void (*end_io)(struct bio *, int err),
170 				      struct parallel_io *par,
171 				      unsigned int offset, int len)
172 {
173 	isect = isect + (offset >> SECTOR_SHIFT);
174 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
175 		npg, rw, (unsigned long long)isect, offset, len);
176 retry:
177 	if (!bio) {
178 		bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
179 		if (!bio)
180 			return ERR_PTR(-ENOMEM);
181 	}
182 	if (bio_add_page(bio, page, len, offset) < len) {
183 		bio = bl_submit_bio(rw, bio);
184 		goto retry;
185 	}
186 	return bio;
187 }
188 
189 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
190 				      sector_t isect, struct page *page,
191 				      struct pnfs_block_extent *be,
192 				      void (*end_io)(struct bio *, int err),
193 				      struct parallel_io *par)
194 {
195 	return do_add_page_to_bio(bio, npg, rw, isect, page, be,
196 				  end_io, par, 0, PAGE_CACHE_SIZE);
197 }
198 
199 /* This is basically copied from mpage_end_io_read */
200 static void bl_end_io_read(struct bio *bio, int err)
201 {
202 	struct parallel_io *par = bio->bi_private;
203 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
204 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
205 
206 	do {
207 		struct page *page = bvec->bv_page;
208 
209 		if (--bvec >= bio->bi_io_vec)
210 			prefetchw(&bvec->bv_page->flags);
211 		if (uptodate)
212 			SetPageUptodate(page);
213 	} while (bvec >= bio->bi_io_vec);
214 	if (!uptodate) {
215 		struct nfs_read_data *rdata = par->data;
216 		struct nfs_pgio_header *header = rdata->header;
217 
218 		if (!header->pnfs_error)
219 			header->pnfs_error = -EIO;
220 		pnfs_set_lo_fail(header->lseg);
221 	}
222 	bio_put(bio);
223 	put_parallel(par);
224 }
225 
226 static void bl_read_cleanup(struct work_struct *work)
227 {
228 	struct rpc_task *task;
229 	struct nfs_read_data *rdata;
230 	dprintk("%s enter\n", __func__);
231 	task = container_of(work, struct rpc_task, u.tk_work);
232 	rdata = container_of(task, struct nfs_read_data, task);
233 	pnfs_ld_read_done(rdata);
234 }
235 
236 static void
237 bl_end_par_io_read(void *data, int unused)
238 {
239 	struct nfs_read_data *rdata = data;
240 
241 	rdata->task.tk_status = rdata->header->pnfs_error;
242 	INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
243 	schedule_work(&rdata->task.u.tk_work);
244 }
245 
246 static enum pnfs_try_status
247 bl_read_pagelist(struct nfs_read_data *rdata)
248 {
249 	struct nfs_pgio_header *header = rdata->header;
250 	int i, hole;
251 	struct bio *bio = NULL;
252 	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
253 	sector_t isect, extent_length = 0;
254 	struct parallel_io *par;
255 	loff_t f_offset = rdata->args.offset;
256 	size_t bytes_left = rdata->args.count;
257 	unsigned int pg_offset, pg_len;
258 	struct page **pages = rdata->args.pages;
259 	int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
260 	const bool is_dio = (header->dreq != NULL);
261 
262 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
263 	       rdata->pages.npages, f_offset, (unsigned int)rdata->args.count);
264 
265 	par = alloc_parallel(rdata);
266 	if (!par)
267 		goto use_mds;
268 	par->pnfs_callback = bl_end_par_io_read;
269 	/* At this point, we can no longer jump to use_mds */
270 
271 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
272 	/* Code assumes extents are page-aligned */
273 	for (i = pg_index; i < rdata->pages.npages; i++) {
274 		if (!extent_length) {
275 			/* We've used up the previous extent */
276 			bl_put_extent(be);
277 			bl_put_extent(cow_read);
278 			bio = bl_submit_bio(READ, bio);
279 			/* Get the next one */
280 			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
281 					     isect, &cow_read);
282 			if (!be) {
283 				header->pnfs_error = -EIO;
284 				goto out;
285 			}
286 			extent_length = be->be_length -
287 				(isect - be->be_f_offset);
288 			if (cow_read) {
289 				sector_t cow_length = cow_read->be_length -
290 					(isect - cow_read->be_f_offset);
291 				extent_length = min(extent_length, cow_length);
292 			}
293 		}
294 
295 		if (is_dio) {
296 			pg_offset = f_offset & ~PAGE_CACHE_MASK;
297 			if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
298 				pg_len = PAGE_CACHE_SIZE - pg_offset;
299 			else
300 				pg_len = bytes_left;
301 
302 			f_offset += pg_len;
303 			bytes_left -= pg_len;
304 			isect += (pg_offset >> SECTOR_SHIFT);
305 		} else {
306 			pg_offset = 0;
307 			pg_len = PAGE_CACHE_SIZE;
308 		}
309 
310 		hole = is_hole(be, isect);
311 		if (hole && !cow_read) {
312 			bio = bl_submit_bio(READ, bio);
313 			/* Fill hole w/ zeroes w/o accessing device */
314 			dprintk("%s Zeroing page for hole\n", __func__);
315 			zero_user_segment(pages[i], pg_offset, pg_len);
316 			print_page(pages[i]);
317 			SetPageUptodate(pages[i]);
318 		} else {
319 			struct pnfs_block_extent *be_read;
320 
321 			be_read = (hole && cow_read) ? cow_read : be;
322 			bio = do_add_page_to_bio(bio, rdata->pages.npages - i,
323 						 READ,
324 						 isect, pages[i], be_read,
325 						 bl_end_io_read, par,
326 						 pg_offset, pg_len);
327 			if (IS_ERR(bio)) {
328 				header->pnfs_error = PTR_ERR(bio);
329 				bio = NULL;
330 				goto out;
331 			}
332 		}
333 		isect += (pg_len >> SECTOR_SHIFT);
334 		extent_length -= PAGE_CACHE_SECTORS;
335 	}
336 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
337 		rdata->res.eof = 1;
338 		rdata->res.count = header->inode->i_size - rdata->args.offset;
339 	} else {
340 		rdata->res.count = (isect << SECTOR_SHIFT) - rdata->args.offset;
341 	}
342 out:
343 	bl_put_extent(be);
344 	bl_put_extent(cow_read);
345 	bl_submit_bio(READ, bio);
346 	put_parallel(par);
347 	return PNFS_ATTEMPTED;
348 
349  use_mds:
350 	dprintk("Giving up and using normal NFS\n");
351 	return PNFS_NOT_ATTEMPTED;
352 }
353 
354 static void mark_extents_written(struct pnfs_block_layout *bl,
355 				 __u64 offset, __u32 count)
356 {
357 	sector_t isect, end;
358 	struct pnfs_block_extent *be;
359 	struct pnfs_block_short_extent *se;
360 
361 	dprintk("%s(%llu, %u)\n", __func__, offset, count);
362 	if (count == 0)
363 		return;
364 	isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
365 	end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
366 	end >>= SECTOR_SHIFT;
367 	while (isect < end) {
368 		sector_t len;
369 		be = bl_find_get_extent(bl, isect, NULL);
370 		BUG_ON(!be); /* FIXME */
371 		len = min(end, be->be_f_offset + be->be_length) - isect;
372 		if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
373 			se = bl_pop_one_short_extent(be->be_inval);
374 			BUG_ON(!se);
375 			bl_mark_for_commit(be, isect, len, se);
376 		}
377 		isect += len;
378 		bl_put_extent(be);
379 	}
380 }
381 
382 static void bl_end_io_write_zero(struct bio *bio, int err)
383 {
384 	struct parallel_io *par = bio->bi_private;
385 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
386 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
387 
388 	do {
389 		struct page *page = bvec->bv_page;
390 
391 		if (--bvec >= bio->bi_io_vec)
392 			prefetchw(&bvec->bv_page->flags);
393 		/* This is the zeroing page we added */
394 		end_page_writeback(page);
395 		page_cache_release(page);
396 	} while (bvec >= bio->bi_io_vec);
397 
398 	if (unlikely(!uptodate)) {
399 		struct nfs_write_data *data = par->data;
400 		struct nfs_pgio_header *header = data->header;
401 
402 		if (!header->pnfs_error)
403 			header->pnfs_error = -EIO;
404 		pnfs_set_lo_fail(header->lseg);
405 	}
406 	bio_put(bio);
407 	put_parallel(par);
408 }
409 
410 static void bl_end_io_write(struct bio *bio, int err)
411 {
412 	struct parallel_io *par = bio->bi_private;
413 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
414 	struct nfs_write_data *data = par->data;
415 	struct nfs_pgio_header *header = data->header;
416 
417 	if (!uptodate) {
418 		if (!header->pnfs_error)
419 			header->pnfs_error = -EIO;
420 		pnfs_set_lo_fail(header->lseg);
421 	}
422 	bio_put(bio);
423 	put_parallel(par);
424 }
425 
426 /* Function scheduled for call during bl_end_par_io_write,
427  * it marks sectors as written and extends the commitlist.
428  */
429 static void bl_write_cleanup(struct work_struct *work)
430 {
431 	struct rpc_task *task;
432 	struct nfs_write_data *wdata;
433 	dprintk("%s enter\n", __func__);
434 	task = container_of(work, struct rpc_task, u.tk_work);
435 	wdata = container_of(task, struct nfs_write_data, task);
436 	if (likely(!wdata->header->pnfs_error)) {
437 		/* Marks for LAYOUTCOMMIT */
438 		mark_extents_written(BLK_LSEG2EXT(wdata->header->lseg),
439 				     wdata->args.offset, wdata->args.count);
440 	}
441 	pnfs_ld_write_done(wdata);
442 }
443 
444 /* Called when last of bios associated with a bl_write_pagelist call finishes */
445 static void bl_end_par_io_write(void *data, int num_se)
446 {
447 	struct nfs_write_data *wdata = data;
448 
449 	if (unlikely(wdata->header->pnfs_error)) {
450 		bl_free_short_extents(&BLK_LSEG2EXT(wdata->header->lseg)->bl_inval,
451 					num_se);
452 	}
453 
454 	wdata->task.tk_status = wdata->header->pnfs_error;
455 	wdata->verf.committed = NFS_FILE_SYNC;
456 	INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
457 	schedule_work(&wdata->task.u.tk_work);
458 }
459 
460 /* FIXME STUB - mark intersection of layout and page as bad, so is not
461  * used again.
462  */
463 static void mark_bad_read(void)
464 {
465 	return;
466 }
467 
468 /*
469  * map_block:  map a requested I/0 block (isect) into an offset in the LVM
470  * block_device
471  */
472 static void
473 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
474 {
475 	dprintk("%s enter be=%p\n", __func__, be);
476 
477 	set_buffer_mapped(bh);
478 	bh->b_bdev = be->be_mdev;
479 	bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
480 	    (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
481 
482 	dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
483 		__func__, (unsigned long long)isect, (long)bh->b_blocknr,
484 		bh->b_size);
485 	return;
486 }
487 
488 static void
489 bl_read_single_end_io(struct bio *bio, int error)
490 {
491 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
492 	struct page *page = bvec->bv_page;
493 
494 	/* Only one page in bvec */
495 	unlock_page(page);
496 }
497 
498 static int
499 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
500 		    unsigned int offset, unsigned int len)
501 {
502 	struct bio *bio;
503 	struct page *shadow_page;
504 	sector_t isect;
505 	char *kaddr, *kshadow_addr;
506 	int ret = 0;
507 
508 	dprintk("%s: offset %u len %u\n", __func__, offset, len);
509 
510 	shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
511 	if (shadow_page == NULL)
512 		return -ENOMEM;
513 
514 	bio = bio_alloc(GFP_NOIO, 1);
515 	if (bio == NULL)
516 		return -ENOMEM;
517 
518 	isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
519 		(offset / SECTOR_SIZE);
520 
521 	bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
522 	bio->bi_bdev = be->be_mdev;
523 	bio->bi_end_io = bl_read_single_end_io;
524 
525 	lock_page(shadow_page);
526 	if (bio_add_page(bio, shadow_page,
527 			 SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
528 		unlock_page(shadow_page);
529 		bio_put(bio);
530 		return -EIO;
531 	}
532 
533 	submit_bio(READ, bio);
534 	wait_on_page_locked(shadow_page);
535 	if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
536 		ret = -EIO;
537 	} else {
538 		kaddr = kmap_atomic(page);
539 		kshadow_addr = kmap_atomic(shadow_page);
540 		memcpy(kaddr + offset, kshadow_addr + offset, len);
541 		kunmap_atomic(kshadow_addr);
542 		kunmap_atomic(kaddr);
543 	}
544 	__free_page(shadow_page);
545 	bio_put(bio);
546 
547 	return ret;
548 }
549 
550 static int
551 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
552 			  unsigned int dirty_offset, unsigned int dirty_len,
553 			  bool full_page)
554 {
555 	int ret = 0;
556 	unsigned int start, end;
557 
558 	if (full_page) {
559 		start = 0;
560 		end = PAGE_CACHE_SIZE;
561 	} else {
562 		start = round_down(dirty_offset, SECTOR_SIZE);
563 		end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
564 	}
565 
566 	dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
567 	if (!be) {
568 		zero_user_segments(page, start, dirty_offset,
569 				   dirty_offset + dirty_len, end);
570 		if (start == 0 && end == PAGE_CACHE_SIZE &&
571 		    trylock_page(page)) {
572 			SetPageUptodate(page);
573 			unlock_page(page);
574 		}
575 		return ret;
576 	}
577 
578 	if (start != dirty_offset)
579 		ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
580 
581 	if (!ret && (dirty_offset + dirty_len < end))
582 		ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
583 					  end - dirty_offset - dirty_len);
584 
585 	return ret;
586 }
587 
588 /* Given an unmapped page, zero it or read in page for COW, page is locked
589  * by caller.
590  */
591 static int
592 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
593 {
594 	struct buffer_head *bh = NULL;
595 	int ret = 0;
596 	sector_t isect;
597 
598 	dprintk("%s enter, %p\n", __func__, page);
599 	BUG_ON(PageUptodate(page));
600 	if (!cow_read) {
601 		zero_user_segment(page, 0, PAGE_SIZE);
602 		SetPageUptodate(page);
603 		goto cleanup;
604 	}
605 
606 	bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
607 	if (!bh) {
608 		ret = -ENOMEM;
609 		goto cleanup;
610 	}
611 
612 	isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
613 	map_block(bh, isect, cow_read);
614 	if (!bh_uptodate_or_lock(bh))
615 		ret = bh_submit_read(bh);
616 	if (ret)
617 		goto cleanup;
618 	SetPageUptodate(page);
619 
620 cleanup:
621 	if (bh)
622 		free_buffer_head(bh);
623 	if (ret) {
624 		/* Need to mark layout with bad read...should now
625 		 * just use nfs4 for reads and writes.
626 		 */
627 		mark_bad_read();
628 	}
629 	return ret;
630 }
631 
632 /* Find or create a zeroing page marked being writeback.
633  * Return ERR_PTR on error, NULL to indicate skip this page and page itself
634  * to indicate write out.
635  */
636 static struct page *
637 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
638 			struct pnfs_block_extent *cow_read)
639 {
640 	struct page *page;
641 	int locked = 0;
642 	page = find_get_page(inode->i_mapping, index);
643 	if (page)
644 		goto check_page;
645 
646 	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
647 	if (unlikely(!page)) {
648 		dprintk("%s oom\n", __func__);
649 		return ERR_PTR(-ENOMEM);
650 	}
651 	locked = 1;
652 
653 check_page:
654 	/* PageDirty: Other will write this out
655 	 * PageWriteback: Other is writing this out
656 	 * PageUptodate: It was read before
657 	 */
658 	if (PageDirty(page) || PageWriteback(page)) {
659 		print_page(page);
660 		if (locked)
661 			unlock_page(page);
662 		page_cache_release(page);
663 		return NULL;
664 	}
665 
666 	if (!locked) {
667 		lock_page(page);
668 		locked = 1;
669 		goto check_page;
670 	}
671 	if (!PageUptodate(page)) {
672 		/* New page, readin or zero it */
673 		init_page_for_write(page, cow_read);
674 	}
675 	set_page_writeback(page);
676 	unlock_page(page);
677 
678 	return page;
679 }
680 
681 static enum pnfs_try_status
682 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
683 {
684 	struct nfs_pgio_header *header = wdata->header;
685 	int i, ret, npg_zero, pg_index, last = 0;
686 	struct bio *bio = NULL;
687 	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
688 	sector_t isect, last_isect = 0, extent_length = 0;
689 	struct parallel_io *par = NULL;
690 	loff_t offset = wdata->args.offset;
691 	size_t count = wdata->args.count;
692 	unsigned int pg_offset, pg_len, saved_len;
693 	struct page **pages = wdata->args.pages;
694 	struct page *page;
695 	pgoff_t index;
696 	u64 temp;
697 	int npg_per_block =
698 	    NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
699 
700 	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
701 
702 	if (header->dreq != NULL &&
703 	    (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
704 	     !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
705 		dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
706 		goto out_mds;
707 	}
708 	/* At this point, wdata->pages is a (sequential) list of nfs_pages.
709 	 * We want to write each, and if there is an error set pnfs_error
710 	 * to have it redone using nfs.
711 	 */
712 	par = alloc_parallel(wdata);
713 	if (!par)
714 		goto out_mds;
715 	par->pnfs_callback = bl_end_par_io_write;
716 	/* At this point, have to be more careful with error handling */
717 
718 	isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
719 	be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
720 	if (!be || !is_writable(be, isect)) {
721 		dprintk("%s no matching extents!\n", __func__);
722 		goto out_mds;
723 	}
724 
725 	/* First page inside INVALID extent */
726 	if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
727 		if (likely(!bl_push_one_short_extent(be->be_inval)))
728 			par->bse_count++;
729 		else
730 			goto out_mds;
731 		temp = offset >> PAGE_CACHE_SHIFT;
732 		npg_zero = do_div(temp, npg_per_block);
733 		isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
734 				     (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
735 		extent_length = be->be_length - (isect - be->be_f_offset);
736 
737 fill_invalid_ext:
738 		dprintk("%s need to zero %d pages\n", __func__, npg_zero);
739 		for (;npg_zero > 0; npg_zero--) {
740 			if (bl_is_sector_init(be->be_inval, isect)) {
741 				dprintk("isect %llu already init\n",
742 					(unsigned long long)isect);
743 				goto next_page;
744 			}
745 			/* page ref released in bl_end_io_write_zero */
746 			index = isect >> PAGE_CACHE_SECTOR_SHIFT;
747 			dprintk("%s zero %dth page: index %lu isect %llu\n",
748 				__func__, npg_zero, index,
749 				(unsigned long long)isect);
750 			page = bl_find_get_zeroing_page(header->inode, index,
751 							cow_read);
752 			if (unlikely(IS_ERR(page))) {
753 				header->pnfs_error = PTR_ERR(page);
754 				goto out;
755 			} else if (page == NULL)
756 				goto next_page;
757 
758 			ret = bl_mark_sectors_init(be->be_inval, isect,
759 						       PAGE_CACHE_SECTORS);
760 			if (unlikely(ret)) {
761 				dprintk("%s bl_mark_sectors_init fail %d\n",
762 					__func__, ret);
763 				end_page_writeback(page);
764 				page_cache_release(page);
765 				header->pnfs_error = ret;
766 				goto out;
767 			}
768 			if (likely(!bl_push_one_short_extent(be->be_inval)))
769 				par->bse_count++;
770 			else {
771 				end_page_writeback(page);
772 				page_cache_release(page);
773 				header->pnfs_error = -ENOMEM;
774 				goto out;
775 			}
776 			/* FIXME: This should be done in bi_end_io */
777 			mark_extents_written(BLK_LSEG2EXT(header->lseg),
778 					     page->index << PAGE_CACHE_SHIFT,
779 					     PAGE_CACHE_SIZE);
780 
781 			bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
782 						 isect, page, be,
783 						 bl_end_io_write_zero, par);
784 			if (IS_ERR(bio)) {
785 				header->pnfs_error = PTR_ERR(bio);
786 				bio = NULL;
787 				goto out;
788 			}
789 next_page:
790 			isect += PAGE_CACHE_SECTORS;
791 			extent_length -= PAGE_CACHE_SECTORS;
792 		}
793 		if (last)
794 			goto write_done;
795 	}
796 	bio = bl_submit_bio(WRITE, bio);
797 
798 	/* Middle pages */
799 	pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
800 	for (i = pg_index; i < wdata->pages.npages; i++) {
801 		if (!extent_length) {
802 			/* We've used up the previous extent */
803 			bl_put_extent(be);
804 			bl_put_extent(cow_read);
805 			bio = bl_submit_bio(WRITE, bio);
806 			/* Get the next one */
807 			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
808 					     isect, &cow_read);
809 			if (!be || !is_writable(be, isect)) {
810 				header->pnfs_error = -EINVAL;
811 				goto out;
812 			}
813 			if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
814 				if (likely(!bl_push_one_short_extent(
815 								be->be_inval)))
816 					par->bse_count++;
817 				else {
818 					header->pnfs_error = -ENOMEM;
819 					goto out;
820 				}
821 			}
822 			extent_length = be->be_length -
823 			    (isect - be->be_f_offset);
824 		}
825 
826 		dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
827 		pg_offset = offset & ~PAGE_CACHE_MASK;
828 		if (pg_offset + count > PAGE_CACHE_SIZE)
829 			pg_len = PAGE_CACHE_SIZE - pg_offset;
830 		else
831 			pg_len = count;
832 
833 		saved_len = pg_len;
834 		if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
835 		    !bl_is_sector_init(be->be_inval, isect)) {
836 			ret = bl_read_partial_page_sync(pages[i], cow_read,
837 							pg_offset, pg_len, true);
838 			if (ret) {
839 				dprintk("%s bl_read_partial_page_sync fail %d\n",
840 					__func__, ret);
841 				header->pnfs_error = ret;
842 				goto out;
843 			}
844 
845 			ret = bl_mark_sectors_init(be->be_inval, isect,
846 						       PAGE_CACHE_SECTORS);
847 			if (unlikely(ret)) {
848 				dprintk("%s bl_mark_sectors_init fail %d\n",
849 					__func__, ret);
850 				header->pnfs_error = ret;
851 				goto out;
852 			}
853 
854 			/* Expand to full page write */
855 			pg_offset = 0;
856 			pg_len = PAGE_CACHE_SIZE;
857 		} else if  ((pg_offset & (SECTOR_SIZE - 1)) ||
858 			    (pg_len & (SECTOR_SIZE - 1))){
859 			/* ahh, nasty case. We have to do sync full sector
860 			 * read-modify-write cycles.
861 			 */
862 			unsigned int saved_offset = pg_offset;
863 			ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
864 							pg_len, false);
865 			pg_offset = round_down(pg_offset, SECTOR_SIZE);
866 			pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
867 				 - pg_offset;
868 		}
869 
870 
871 		bio = do_add_page_to_bio(bio, wdata->pages.npages - i, WRITE,
872 					 isect, pages[i], be,
873 					 bl_end_io_write, par,
874 					 pg_offset, pg_len);
875 		if (IS_ERR(bio)) {
876 			header->pnfs_error = PTR_ERR(bio);
877 			bio = NULL;
878 			goto out;
879 		}
880 		offset += saved_len;
881 		count -= saved_len;
882 		isect += PAGE_CACHE_SECTORS;
883 		last_isect = isect;
884 		extent_length -= PAGE_CACHE_SECTORS;
885 	}
886 
887 	/* Last page inside INVALID extent */
888 	if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
889 		bio = bl_submit_bio(WRITE, bio);
890 		temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
891 		npg_zero = npg_per_block - do_div(temp, npg_per_block);
892 		if (npg_zero < npg_per_block) {
893 			last = 1;
894 			goto fill_invalid_ext;
895 		}
896 	}
897 
898 write_done:
899 	wdata->res.count = wdata->args.count;
900 out:
901 	bl_put_extent(be);
902 	bl_put_extent(cow_read);
903 	bl_submit_bio(WRITE, bio);
904 	put_parallel(par);
905 	return PNFS_ATTEMPTED;
906 out_mds:
907 	bl_put_extent(be);
908 	bl_put_extent(cow_read);
909 	kfree(par);
910 	return PNFS_NOT_ATTEMPTED;
911 }
912 
913 /* FIXME - range ignored */
914 static void
915 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
916 {
917 	int i;
918 	struct pnfs_block_extent *be;
919 
920 	spin_lock(&bl->bl_ext_lock);
921 	for (i = 0; i < EXTENT_LISTS; i++) {
922 		while (!list_empty(&bl->bl_extents[i])) {
923 			be = list_first_entry(&bl->bl_extents[i],
924 					      struct pnfs_block_extent,
925 					      be_node);
926 			list_del(&be->be_node);
927 			bl_put_extent(be);
928 		}
929 	}
930 	spin_unlock(&bl->bl_ext_lock);
931 }
932 
933 static void
934 release_inval_marks(struct pnfs_inval_markings *marks)
935 {
936 	struct pnfs_inval_tracking *pos, *temp;
937 	struct pnfs_block_short_extent *se, *stemp;
938 
939 	list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
940 		list_del(&pos->it_link);
941 		kfree(pos);
942 	}
943 
944 	list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
945 		list_del(&se->bse_node);
946 		kfree(se);
947 	}
948 	return;
949 }
950 
951 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
952 {
953 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
954 
955 	dprintk("%s enter\n", __func__);
956 	release_extents(bl, NULL);
957 	release_inval_marks(&bl->bl_inval);
958 	kfree(bl);
959 }
960 
961 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
962 						   gfp_t gfp_flags)
963 {
964 	struct pnfs_block_layout *bl;
965 
966 	dprintk("%s enter\n", __func__);
967 	bl = kzalloc(sizeof(*bl), gfp_flags);
968 	if (!bl)
969 		return NULL;
970 	spin_lock_init(&bl->bl_ext_lock);
971 	INIT_LIST_HEAD(&bl->bl_extents[0]);
972 	INIT_LIST_HEAD(&bl->bl_extents[1]);
973 	INIT_LIST_HEAD(&bl->bl_commit);
974 	INIT_LIST_HEAD(&bl->bl_committing);
975 	bl->bl_count = 0;
976 	bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
977 	BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
978 	return &bl->bl_layout;
979 }
980 
981 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
982 {
983 	dprintk("%s enter\n", __func__);
984 	kfree(lseg);
985 }
986 
987 /* We pretty much ignore lseg, and store all data layout wide, so we
988  * can correctly merge.
989  */
990 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
991 						 struct nfs4_layoutget_res *lgr,
992 						 gfp_t gfp_flags)
993 {
994 	struct pnfs_layout_segment *lseg;
995 	int status;
996 
997 	dprintk("%s enter\n", __func__);
998 	lseg = kzalloc(sizeof(*lseg), gfp_flags);
999 	if (!lseg)
1000 		return ERR_PTR(-ENOMEM);
1001 	status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
1002 	if (status) {
1003 		/* We don't want to call the full-blown bl_free_lseg,
1004 		 * since on error extents were not touched.
1005 		 */
1006 		kfree(lseg);
1007 		return ERR_PTR(status);
1008 	}
1009 	return lseg;
1010 }
1011 
1012 static void
1013 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1014 		       const struct nfs4_layoutcommit_args *arg)
1015 {
1016 	dprintk("%s enter\n", __func__);
1017 	encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1018 }
1019 
1020 static void
1021 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1022 {
1023 	struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1024 
1025 	dprintk("%s enter\n", __func__);
1026 	clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1027 }
1028 
1029 static void free_blk_mountid(struct block_mount_id *mid)
1030 {
1031 	if (mid) {
1032 		struct pnfs_block_dev *dev, *tmp;
1033 
1034 		/* No need to take bm_lock as we are last user freeing bm_devlist */
1035 		list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1036 			list_del(&dev->bm_node);
1037 			bl_free_block_dev(dev);
1038 		}
1039 		kfree(mid);
1040 	}
1041 }
1042 
1043 /* This is mostly copied from the filelayout_get_device_info function.
1044  * It seems much of this should be at the generic pnfs level.
1045  */
1046 static struct pnfs_block_dev *
1047 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1048 			struct nfs4_deviceid *d_id)
1049 {
1050 	struct pnfs_device *dev;
1051 	struct pnfs_block_dev *rv;
1052 	u32 max_resp_sz;
1053 	int max_pages;
1054 	struct page **pages = NULL;
1055 	int i, rc;
1056 
1057 	/*
1058 	 * Use the session max response size as the basis for setting
1059 	 * GETDEVICEINFO's maxcount
1060 	 */
1061 	max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1062 	max_pages = nfs_page_array_len(0, max_resp_sz);
1063 	dprintk("%s max_resp_sz %u max_pages %d\n",
1064 		__func__, max_resp_sz, max_pages);
1065 
1066 	dev = kmalloc(sizeof(*dev), GFP_NOFS);
1067 	if (!dev) {
1068 		dprintk("%s kmalloc failed\n", __func__);
1069 		return ERR_PTR(-ENOMEM);
1070 	}
1071 
1072 	pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
1073 	if (pages == NULL) {
1074 		kfree(dev);
1075 		return ERR_PTR(-ENOMEM);
1076 	}
1077 	for (i = 0; i < max_pages; i++) {
1078 		pages[i] = alloc_page(GFP_NOFS);
1079 		if (!pages[i]) {
1080 			rv = ERR_PTR(-ENOMEM);
1081 			goto out_free;
1082 		}
1083 	}
1084 
1085 	memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1086 	dev->layout_type = LAYOUT_BLOCK_VOLUME;
1087 	dev->pages = pages;
1088 	dev->pgbase = 0;
1089 	dev->pglen = PAGE_SIZE * max_pages;
1090 	dev->mincount = 0;
1091 
1092 	dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1093 	rc = nfs4_proc_getdeviceinfo(server, dev);
1094 	dprintk("%s getdevice info returns %d\n", __func__, rc);
1095 	if (rc) {
1096 		rv = ERR_PTR(rc);
1097 		goto out_free;
1098 	}
1099 
1100 	rv = nfs4_blk_decode_device(server, dev);
1101  out_free:
1102 	for (i = 0; i < max_pages; i++)
1103 		__free_page(pages[i]);
1104 	kfree(pages);
1105 	kfree(dev);
1106 	return rv;
1107 }
1108 
1109 static int
1110 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1111 {
1112 	struct block_mount_id *b_mt_id = NULL;
1113 	struct pnfs_devicelist *dlist = NULL;
1114 	struct pnfs_block_dev *bdev;
1115 	LIST_HEAD(block_disklist);
1116 	int status, i;
1117 
1118 	dprintk("%s enter\n", __func__);
1119 
1120 	if (server->pnfs_blksize == 0) {
1121 		dprintk("%s Server did not return blksize\n", __func__);
1122 		return -EINVAL;
1123 	}
1124 	b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1125 	if (!b_mt_id) {
1126 		status = -ENOMEM;
1127 		goto out_error;
1128 	}
1129 	/* Initialize nfs4 block layout mount id */
1130 	spin_lock_init(&b_mt_id->bm_lock);
1131 	INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1132 
1133 	dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1134 	if (!dlist) {
1135 		status = -ENOMEM;
1136 		goto out_error;
1137 	}
1138 	dlist->eof = 0;
1139 	while (!dlist->eof) {
1140 		status = nfs4_proc_getdevicelist(server, fh, dlist);
1141 		if (status)
1142 			goto out_error;
1143 		dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1144 			__func__, dlist->num_devs, dlist->eof);
1145 		for (i = 0; i < dlist->num_devs; i++) {
1146 			bdev = nfs4_blk_get_deviceinfo(server, fh,
1147 						       &dlist->dev_id[i]);
1148 			if (IS_ERR(bdev)) {
1149 				status = PTR_ERR(bdev);
1150 				goto out_error;
1151 			}
1152 			spin_lock(&b_mt_id->bm_lock);
1153 			list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1154 			spin_unlock(&b_mt_id->bm_lock);
1155 		}
1156 	}
1157 	dprintk("%s SUCCESS\n", __func__);
1158 	server->pnfs_ld_data = b_mt_id;
1159 
1160  out_return:
1161 	kfree(dlist);
1162 	return status;
1163 
1164  out_error:
1165 	free_blk_mountid(b_mt_id);
1166 	goto out_return;
1167 }
1168 
1169 static int
1170 bl_clear_layoutdriver(struct nfs_server *server)
1171 {
1172 	struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1173 
1174 	dprintk("%s enter\n", __func__);
1175 	free_blk_mountid(b_mt_id);
1176 	dprintk("%s RETURNS\n", __func__);
1177 	return 0;
1178 }
1179 
1180 static bool
1181 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1182 {
1183 	return IS_ALIGNED(req->wb_offset, alignment) &&
1184 	       IS_ALIGNED(req->wb_bytes, alignment);
1185 }
1186 
1187 static void
1188 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1189 {
1190 	if (pgio->pg_dreq != NULL &&
1191 	    !is_aligned_req(req, SECTOR_SIZE))
1192 		nfs_pageio_reset_read_mds(pgio);
1193 	else
1194 		pnfs_generic_pg_init_read(pgio, req);
1195 }
1196 
1197 static bool
1198 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1199 		struct nfs_page *req)
1200 {
1201 	if (pgio->pg_dreq != NULL &&
1202 	    !is_aligned_req(req, SECTOR_SIZE))
1203 		return false;
1204 
1205 	return pnfs_generic_pg_test(pgio, prev, req);
1206 }
1207 
1208 /*
1209  * Return the number of contiguous bytes for a given inode
1210  * starting at page frame idx.
1211  */
1212 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
1213 {
1214 	struct address_space *mapping = inode->i_mapping;
1215 	pgoff_t end;
1216 
1217 	/* Optimize common case that writes from 0 to end of file */
1218 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
1219 	if (end != NFS_I(inode)->npages) {
1220 		rcu_read_lock();
1221 		end = radix_tree_next_hole(&mapping->page_tree, idx + 1, ULONG_MAX);
1222 		rcu_read_unlock();
1223 	}
1224 
1225 	if (!end)
1226 		return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
1227 	else
1228 		return (end - idx) << PAGE_CACHE_SHIFT;
1229 }
1230 
1231 static void
1232 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1233 {
1234 	if (pgio->pg_dreq != NULL &&
1235 	    !is_aligned_req(req, PAGE_CACHE_SIZE)) {
1236 		nfs_pageio_reset_write_mds(pgio);
1237 	} else {
1238 		u64 wb_size;
1239 		if (pgio->pg_dreq == NULL)
1240 			wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
1241 						      req->wb_index);
1242 		else
1243 			wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1244 
1245 		pnfs_generic_pg_init_write(pgio, req, wb_size);
1246 	}
1247 }
1248 
1249 static bool
1250 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1251 		 struct nfs_page *req)
1252 {
1253 	if (pgio->pg_dreq != NULL &&
1254 	    !is_aligned_req(req, PAGE_CACHE_SIZE))
1255 		return false;
1256 
1257 	return pnfs_generic_pg_test(pgio, prev, req);
1258 }
1259 
1260 static const struct nfs_pageio_ops bl_pg_read_ops = {
1261 	.pg_init = bl_pg_init_read,
1262 	.pg_test = bl_pg_test_read,
1263 	.pg_doio = pnfs_generic_pg_readpages,
1264 };
1265 
1266 static const struct nfs_pageio_ops bl_pg_write_ops = {
1267 	.pg_init = bl_pg_init_write,
1268 	.pg_test = bl_pg_test_write,
1269 	.pg_doio = pnfs_generic_pg_writepages,
1270 };
1271 
1272 static struct pnfs_layoutdriver_type blocklayout_type = {
1273 	.id				= LAYOUT_BLOCK_VOLUME,
1274 	.name				= "LAYOUT_BLOCK_VOLUME",
1275 	.read_pagelist			= bl_read_pagelist,
1276 	.write_pagelist			= bl_write_pagelist,
1277 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
1278 	.free_layout_hdr		= bl_free_layout_hdr,
1279 	.alloc_lseg			= bl_alloc_lseg,
1280 	.free_lseg			= bl_free_lseg,
1281 	.encode_layoutcommit		= bl_encode_layoutcommit,
1282 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
1283 	.set_layoutdriver		= bl_set_layoutdriver,
1284 	.clear_layoutdriver		= bl_clear_layoutdriver,
1285 	.pg_read_ops			= &bl_pg_read_ops,
1286 	.pg_write_ops			= &bl_pg_write_ops,
1287 };
1288 
1289 static const struct rpc_pipe_ops bl_upcall_ops = {
1290 	.upcall		= rpc_pipe_generic_upcall,
1291 	.downcall	= bl_pipe_downcall,
1292 	.destroy_msg	= bl_pipe_destroy_msg,
1293 };
1294 
1295 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1296 					    struct rpc_pipe *pipe)
1297 {
1298 	struct dentry *dir, *dentry;
1299 
1300 	dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1301 	if (dir == NULL)
1302 		return ERR_PTR(-ENOENT);
1303 	dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1304 	dput(dir);
1305 	return dentry;
1306 }
1307 
1308 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1309 					  struct rpc_pipe *pipe)
1310 {
1311 	if (pipe->dentry)
1312 		rpc_unlink(pipe->dentry);
1313 }
1314 
1315 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1316 			   void *ptr)
1317 {
1318 	struct super_block *sb = ptr;
1319 	struct net *net = sb->s_fs_info;
1320 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1321 	struct dentry *dentry;
1322 	int ret = 0;
1323 
1324 	if (!try_module_get(THIS_MODULE))
1325 		return 0;
1326 
1327 	if (nn->bl_device_pipe == NULL) {
1328 		module_put(THIS_MODULE);
1329 		return 0;
1330 	}
1331 
1332 	switch (event) {
1333 	case RPC_PIPEFS_MOUNT:
1334 		dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1335 		if (IS_ERR(dentry)) {
1336 			ret = PTR_ERR(dentry);
1337 			break;
1338 		}
1339 		nn->bl_device_pipe->dentry = dentry;
1340 		break;
1341 	case RPC_PIPEFS_UMOUNT:
1342 		if (nn->bl_device_pipe->dentry)
1343 			nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1344 		break;
1345 	default:
1346 		ret = -ENOTSUPP;
1347 		break;
1348 	}
1349 	module_put(THIS_MODULE);
1350 	return ret;
1351 }
1352 
1353 static struct notifier_block nfs4blocklayout_block = {
1354 	.notifier_call = rpc_pipefs_event,
1355 };
1356 
1357 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1358 						   struct rpc_pipe *pipe)
1359 {
1360 	struct super_block *pipefs_sb;
1361 	struct dentry *dentry;
1362 
1363 	pipefs_sb = rpc_get_sb_net(net);
1364 	if (!pipefs_sb)
1365 		return NULL;
1366 	dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1367 	rpc_put_sb_net(net);
1368 	return dentry;
1369 }
1370 
1371 static void nfs4blocklayout_unregister_net(struct net *net,
1372 					   struct rpc_pipe *pipe)
1373 {
1374 	struct super_block *pipefs_sb;
1375 
1376 	pipefs_sb = rpc_get_sb_net(net);
1377 	if (pipefs_sb) {
1378 		nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1379 		rpc_put_sb_net(net);
1380 	}
1381 }
1382 
1383 static int nfs4blocklayout_net_init(struct net *net)
1384 {
1385 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1386 	struct dentry *dentry;
1387 
1388 	init_waitqueue_head(&nn->bl_wq);
1389 	nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1390 	if (IS_ERR(nn->bl_device_pipe))
1391 		return PTR_ERR(nn->bl_device_pipe);
1392 	dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1393 	if (IS_ERR(dentry)) {
1394 		rpc_destroy_pipe_data(nn->bl_device_pipe);
1395 		return PTR_ERR(dentry);
1396 	}
1397 	nn->bl_device_pipe->dentry = dentry;
1398 	return 0;
1399 }
1400 
1401 static void nfs4blocklayout_net_exit(struct net *net)
1402 {
1403 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1404 
1405 	nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1406 	rpc_destroy_pipe_data(nn->bl_device_pipe);
1407 	nn->bl_device_pipe = NULL;
1408 }
1409 
1410 static struct pernet_operations nfs4blocklayout_net_ops = {
1411 	.init = nfs4blocklayout_net_init,
1412 	.exit = nfs4blocklayout_net_exit,
1413 };
1414 
1415 static int __init nfs4blocklayout_init(void)
1416 {
1417 	int ret;
1418 
1419 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1420 
1421 	ret = pnfs_register_layoutdriver(&blocklayout_type);
1422 	if (ret)
1423 		goto out;
1424 
1425 	ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1426 	if (ret)
1427 		goto out_remove;
1428 	ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1429 	if (ret)
1430 		goto out_notifier;
1431 out:
1432 	return ret;
1433 
1434 out_notifier:
1435 	rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1436 out_remove:
1437 	pnfs_unregister_layoutdriver(&blocklayout_type);
1438 	return ret;
1439 }
1440 
1441 static void __exit nfs4blocklayout_exit(void)
1442 {
1443 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1444 	       __func__);
1445 
1446 	rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1447 	unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1448 	pnfs_unregister_layoutdriver(&blocklayout_type);
1449 }
1450 
1451 MODULE_ALIAS("nfs-layouttype4-3");
1452 
1453 module_init(nfs4blocklayout_init);
1454 module_exit(nfs4blocklayout_exit);
1455