xref: /linux/fs/nfs/blocklayout/blocklayout.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 "../nfs4session.h"
44 #include "../internal.h"
45 #include "blocklayout.h"
46 
47 #define NFSDBG_FACILITY	NFSDBG_PNFS_LD
48 
49 MODULE_LICENSE("GPL");
50 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
51 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
52 
53 static void print_page(struct page *page)
54 {
55 	dprintk("PRINTPAGE page %p\n", page);
56 	dprintk("	PagePrivate %d\n", PagePrivate(page));
57 	dprintk("	PageUptodate %d\n", PageUptodate(page));
58 	dprintk("	PageError %d\n", PageError(page));
59 	dprintk("	PageDirty %d\n", PageDirty(page));
60 	dprintk("	PageReferenced %d\n", PageReferenced(page));
61 	dprintk("	PageLocked %d\n", PageLocked(page));
62 	dprintk("	PageWriteback %d\n", PageWriteback(page));
63 	dprintk("	PageMappedToDisk %d\n", PageMappedToDisk(page));
64 	dprintk("\n");
65 }
66 
67 /* Given the be associated with isect, determine if page data needs to be
68  * initialized.
69  */
70 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
71 {
72 	if (be->be_state == PNFS_BLOCK_NONE_DATA)
73 		return 1;
74 	else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
75 		return 0;
76 	else
77 		return !bl_is_sector_init(be->be_inval, isect);
78 }
79 
80 /* Given the be associated with isect, determine if page data can be
81  * written to disk.
82  */
83 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
84 {
85 	return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
86 		be->be_state == PNFS_BLOCK_INVALID_DATA);
87 }
88 
89 /* The data we are handed might be spread across several bios.  We need
90  * to track when the last one is finished.
91  */
92 struct parallel_io {
93 	struct kref refcnt;
94 	void (*pnfs_callback) (void *data, int num_se);
95 	void *data;
96 	int bse_count;
97 };
98 
99 static inline struct parallel_io *alloc_parallel(void *data)
100 {
101 	struct parallel_io *rv;
102 
103 	rv  = kmalloc(sizeof(*rv), GFP_NOFS);
104 	if (rv) {
105 		rv->data = data;
106 		kref_init(&rv->refcnt);
107 		rv->bse_count = 0;
108 	}
109 	return rv;
110 }
111 
112 static inline void get_parallel(struct parallel_io *p)
113 {
114 	kref_get(&p->refcnt);
115 }
116 
117 static void destroy_parallel(struct kref *kref)
118 {
119 	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
120 
121 	dprintk("%s enter\n", __func__);
122 	p->pnfs_callback(p->data, p->bse_count);
123 	kfree(p);
124 }
125 
126 static inline void put_parallel(struct parallel_io *p)
127 {
128 	kref_put(&p->refcnt, destroy_parallel);
129 }
130 
131 static struct bio *
132 bl_submit_bio(int rw, struct bio *bio)
133 {
134 	if (bio) {
135 		get_parallel(bio->bi_private);
136 		dprintk("%s submitting %s bio %u@%llu\n", __func__,
137 			rw == READ ? "read" : "write",
138 			bio->bi_size, (unsigned long long)bio->bi_sector);
139 		submit_bio(rw, bio);
140 	}
141 	return NULL;
142 }
143 
144 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
145 				     struct pnfs_block_extent *be,
146 				     void (*end_io)(struct bio *, int err),
147 				     struct parallel_io *par)
148 {
149 	struct bio *bio;
150 
151 	npg = min(npg, BIO_MAX_PAGES);
152 	bio = bio_alloc(GFP_NOIO, npg);
153 	if (!bio && (current->flags & PF_MEMALLOC)) {
154 		while (!bio && (npg /= 2))
155 			bio = bio_alloc(GFP_NOIO, npg);
156 	}
157 
158 	if (bio) {
159 		bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
160 		bio->bi_bdev = be->be_mdev;
161 		bio->bi_end_io = end_io;
162 		bio->bi_private = par;
163 	}
164 	return bio;
165 }
166 
167 static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw,
168 				      sector_t isect, struct page *page,
169 				      struct pnfs_block_extent *be,
170 				      void (*end_io)(struct bio *, int err),
171 				      struct parallel_io *par,
172 				      unsigned int offset, int len)
173 {
174 	isect = isect + (offset >> SECTOR_SHIFT);
175 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
176 		npg, rw, (unsigned long long)isect, offset, len);
177 retry:
178 	if (!bio) {
179 		bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
180 		if (!bio)
181 			return ERR_PTR(-ENOMEM);
182 	}
183 	if (bio_add_page(bio, page, len, offset) < len) {
184 		bio = bl_submit_bio(rw, bio);
185 		goto retry;
186 	}
187 	return bio;
188 }
189 
190 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
191 				      sector_t isect, struct page *page,
192 				      struct pnfs_block_extent *be,
193 				      void (*end_io)(struct bio *, int err),
194 				      struct parallel_io *par)
195 {
196 	return do_add_page_to_bio(bio, npg, rw, isect, page, be,
197 				  end_io, par, 0, PAGE_CACHE_SIZE);
198 }
199 
200 /* This is basically copied from mpage_end_io_read */
201 static void bl_end_io_read(struct bio *bio, int err)
202 {
203 	struct parallel_io *par = bio->bi_private;
204 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
205 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
206 
207 	do {
208 		struct page *page = bvec->bv_page;
209 
210 		if (--bvec >= bio->bi_io_vec)
211 			prefetchw(&bvec->bv_page->flags);
212 		if (uptodate)
213 			SetPageUptodate(page);
214 	} while (bvec >= bio->bi_io_vec);
215 	if (!uptodate) {
216 		struct nfs_read_data *rdata = par->data;
217 		struct nfs_pgio_header *header = rdata->header;
218 
219 		if (!header->pnfs_error)
220 			header->pnfs_error = -EIO;
221 		pnfs_set_lo_fail(header->lseg);
222 	}
223 	bio_put(bio);
224 	put_parallel(par);
225 }
226 
227 static void bl_read_cleanup(struct work_struct *work)
228 {
229 	struct rpc_task *task;
230 	struct nfs_read_data *rdata;
231 	dprintk("%s enter\n", __func__);
232 	task = container_of(work, struct rpc_task, u.tk_work);
233 	rdata = container_of(task, struct nfs_read_data, task);
234 	pnfs_ld_read_done(rdata);
235 }
236 
237 static void
238 bl_end_par_io_read(void *data, int unused)
239 {
240 	struct nfs_read_data *rdata = data;
241 
242 	rdata->task.tk_status = rdata->header->pnfs_error;
243 	INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
244 	schedule_work(&rdata->task.u.tk_work);
245 }
246 
247 static enum pnfs_try_status
248 bl_read_pagelist(struct nfs_read_data *rdata)
249 {
250 	struct nfs_pgio_header *header = rdata->header;
251 	int i, hole;
252 	struct bio *bio = NULL;
253 	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
254 	sector_t isect, extent_length = 0;
255 	struct parallel_io *par;
256 	loff_t f_offset = rdata->args.offset;
257 	size_t bytes_left = rdata->args.count;
258 	unsigned int pg_offset, pg_len;
259 	struct page **pages = rdata->args.pages;
260 	int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
261 	const bool is_dio = (header->dreq != NULL);
262 
263 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
264 	       rdata->pages.npages, f_offset, (unsigned int)rdata->args.count);
265 
266 	par = alloc_parallel(rdata);
267 	if (!par)
268 		goto use_mds;
269 	par->pnfs_callback = bl_end_par_io_read;
270 	/* At this point, we can no longer jump to use_mds */
271 
272 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
273 	/* Code assumes extents are page-aligned */
274 	for (i = pg_index; i < rdata->pages.npages; i++) {
275 		if (!extent_length) {
276 			/* We've used up the previous extent */
277 			bl_put_extent(be);
278 			bl_put_extent(cow_read);
279 			bio = bl_submit_bio(READ, bio);
280 			/* Get the next one */
281 			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
282 					     isect, &cow_read);
283 			if (!be) {
284 				header->pnfs_error = -EIO;
285 				goto out;
286 			}
287 			extent_length = be->be_length -
288 				(isect - be->be_f_offset);
289 			if (cow_read) {
290 				sector_t cow_length = cow_read->be_length -
291 					(isect - cow_read->be_f_offset);
292 				extent_length = min(extent_length, cow_length);
293 			}
294 		}
295 
296 		if (is_dio) {
297 			pg_offset = f_offset & ~PAGE_CACHE_MASK;
298 			if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
299 				pg_len = PAGE_CACHE_SIZE - pg_offset;
300 			else
301 				pg_len = bytes_left;
302 
303 			f_offset += pg_len;
304 			bytes_left -= pg_len;
305 			isect += (pg_offset >> SECTOR_SHIFT);
306 		} else {
307 			pg_offset = 0;
308 			pg_len = PAGE_CACHE_SIZE;
309 		}
310 
311 		hole = is_hole(be, isect);
312 		if (hole && !cow_read) {
313 			bio = bl_submit_bio(READ, bio);
314 			/* Fill hole w/ zeroes w/o accessing device */
315 			dprintk("%s Zeroing page for hole\n", __func__);
316 			zero_user_segment(pages[i], pg_offset, pg_len);
317 			print_page(pages[i]);
318 			SetPageUptodate(pages[i]);
319 		} else {
320 			struct pnfs_block_extent *be_read;
321 
322 			be_read = (hole && cow_read) ? cow_read : be;
323 			bio = do_add_page_to_bio(bio, rdata->pages.npages - i,
324 						 READ,
325 						 isect, pages[i], be_read,
326 						 bl_end_io_read, par,
327 						 pg_offset, pg_len);
328 			if (IS_ERR(bio)) {
329 				header->pnfs_error = PTR_ERR(bio);
330 				bio = NULL;
331 				goto out;
332 			}
333 		}
334 		isect += (pg_len >> SECTOR_SHIFT);
335 		extent_length -= PAGE_CACHE_SECTORS;
336 	}
337 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
338 		rdata->res.eof = 1;
339 		rdata->res.count = header->inode->i_size - rdata->args.offset;
340 	} else {
341 		rdata->res.count = (isect << SECTOR_SHIFT) - rdata->args.offset;
342 	}
343 out:
344 	bl_put_extent(be);
345 	bl_put_extent(cow_read);
346 	bl_submit_bio(READ, bio);
347 	put_parallel(par);
348 	return PNFS_ATTEMPTED;
349 
350  use_mds:
351 	dprintk("Giving up and using normal NFS\n");
352 	return PNFS_NOT_ATTEMPTED;
353 }
354 
355 static void mark_extents_written(struct pnfs_block_layout *bl,
356 				 __u64 offset, __u32 count)
357 {
358 	sector_t isect, end;
359 	struct pnfs_block_extent *be;
360 	struct pnfs_block_short_extent *se;
361 
362 	dprintk("%s(%llu, %u)\n", __func__, offset, count);
363 	if (count == 0)
364 		return;
365 	isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
366 	end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
367 	end >>= SECTOR_SHIFT;
368 	while (isect < end) {
369 		sector_t len;
370 		be = bl_find_get_extent(bl, isect, NULL);
371 		BUG_ON(!be); /* FIXME */
372 		len = min(end, be->be_f_offset + be->be_length) - isect;
373 		if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
374 			se = bl_pop_one_short_extent(be->be_inval);
375 			BUG_ON(!se);
376 			bl_mark_for_commit(be, isect, len, se);
377 		}
378 		isect += len;
379 		bl_put_extent(be);
380 	}
381 }
382 
383 static void bl_end_io_write_zero(struct bio *bio, int err)
384 {
385 	struct parallel_io *par = bio->bi_private;
386 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
387 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
388 
389 	do {
390 		struct page *page = bvec->bv_page;
391 
392 		if (--bvec >= bio->bi_io_vec)
393 			prefetchw(&bvec->bv_page->flags);
394 		/* This is the zeroing page we added */
395 		end_page_writeback(page);
396 		page_cache_release(page);
397 	} while (bvec >= bio->bi_io_vec);
398 
399 	if (unlikely(!uptodate)) {
400 		struct nfs_write_data *data = par->data;
401 		struct nfs_pgio_header *header = data->header;
402 
403 		if (!header->pnfs_error)
404 			header->pnfs_error = -EIO;
405 		pnfs_set_lo_fail(header->lseg);
406 	}
407 	bio_put(bio);
408 	put_parallel(par);
409 }
410 
411 static void bl_end_io_write(struct bio *bio, int err)
412 {
413 	struct parallel_io *par = bio->bi_private;
414 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
415 	struct nfs_write_data *data = par->data;
416 	struct nfs_pgio_header *header = data->header;
417 
418 	if (!uptodate) {
419 		if (!header->pnfs_error)
420 			header->pnfs_error = -EIO;
421 		pnfs_set_lo_fail(header->lseg);
422 	}
423 	bio_put(bio);
424 	put_parallel(par);
425 }
426 
427 /* Function scheduled for call during bl_end_par_io_write,
428  * it marks sectors as written and extends the commitlist.
429  */
430 static void bl_write_cleanup(struct work_struct *work)
431 {
432 	struct rpc_task *task;
433 	struct nfs_write_data *wdata;
434 	dprintk("%s enter\n", __func__);
435 	task = container_of(work, struct rpc_task, u.tk_work);
436 	wdata = container_of(task, struct nfs_write_data, task);
437 	if (likely(!wdata->header->pnfs_error)) {
438 		/* Marks for LAYOUTCOMMIT */
439 		mark_extents_written(BLK_LSEG2EXT(wdata->header->lseg),
440 				     wdata->args.offset, wdata->args.count);
441 	}
442 	pnfs_ld_write_done(wdata);
443 }
444 
445 /* Called when last of bios associated with a bl_write_pagelist call finishes */
446 static void bl_end_par_io_write(void *data, int num_se)
447 {
448 	struct nfs_write_data *wdata = data;
449 
450 	if (unlikely(wdata->header->pnfs_error)) {
451 		bl_free_short_extents(&BLK_LSEG2EXT(wdata->header->lseg)->bl_inval,
452 					num_se);
453 	}
454 
455 	wdata->task.tk_status = wdata->header->pnfs_error;
456 	wdata->verf.committed = NFS_FILE_SYNC;
457 	INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
458 	schedule_work(&wdata->task.u.tk_work);
459 }
460 
461 /* FIXME STUB - mark intersection of layout and page as bad, so is not
462  * used again.
463  */
464 static void mark_bad_read(void)
465 {
466 	return;
467 }
468 
469 /*
470  * map_block:  map a requested I/0 block (isect) into an offset in the LVM
471  * block_device
472  */
473 static void
474 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
475 {
476 	dprintk("%s enter be=%p\n", __func__, be);
477 
478 	set_buffer_mapped(bh);
479 	bh->b_bdev = be->be_mdev;
480 	bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
481 	    (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
482 
483 	dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
484 		__func__, (unsigned long long)isect, (long)bh->b_blocknr,
485 		bh->b_size);
486 	return;
487 }
488 
489 static void
490 bl_read_single_end_io(struct bio *bio, int error)
491 {
492 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
493 	struct page *page = bvec->bv_page;
494 
495 	/* Only one page in bvec */
496 	unlock_page(page);
497 }
498 
499 static int
500 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
501 		    unsigned int offset, unsigned int len)
502 {
503 	struct bio *bio;
504 	struct page *shadow_page;
505 	sector_t isect;
506 	char *kaddr, *kshadow_addr;
507 	int ret = 0;
508 
509 	dprintk("%s: offset %u len %u\n", __func__, offset, len);
510 
511 	shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
512 	if (shadow_page == NULL)
513 		return -ENOMEM;
514 
515 	bio = bio_alloc(GFP_NOIO, 1);
516 	if (bio == NULL)
517 		return -ENOMEM;
518 
519 	isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
520 		(offset / SECTOR_SIZE);
521 
522 	bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
523 	bio->bi_bdev = be->be_mdev;
524 	bio->bi_end_io = bl_read_single_end_io;
525 
526 	lock_page(shadow_page);
527 	if (bio_add_page(bio, shadow_page,
528 			 SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
529 		unlock_page(shadow_page);
530 		bio_put(bio);
531 		return -EIO;
532 	}
533 
534 	submit_bio(READ, bio);
535 	wait_on_page_locked(shadow_page);
536 	if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
537 		ret = -EIO;
538 	} else {
539 		kaddr = kmap_atomic(page);
540 		kshadow_addr = kmap_atomic(shadow_page);
541 		memcpy(kaddr + offset, kshadow_addr + offset, len);
542 		kunmap_atomic(kshadow_addr);
543 		kunmap_atomic(kaddr);
544 	}
545 	__free_page(shadow_page);
546 	bio_put(bio);
547 
548 	return ret;
549 }
550 
551 static int
552 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
553 			  unsigned int dirty_offset, unsigned int dirty_len,
554 			  bool full_page)
555 {
556 	int ret = 0;
557 	unsigned int start, end;
558 
559 	if (full_page) {
560 		start = 0;
561 		end = PAGE_CACHE_SIZE;
562 	} else {
563 		start = round_down(dirty_offset, SECTOR_SIZE);
564 		end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
565 	}
566 
567 	dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
568 	if (!be) {
569 		zero_user_segments(page, start, dirty_offset,
570 				   dirty_offset + dirty_len, end);
571 		if (start == 0 && end == PAGE_CACHE_SIZE &&
572 		    trylock_page(page)) {
573 			SetPageUptodate(page);
574 			unlock_page(page);
575 		}
576 		return ret;
577 	}
578 
579 	if (start != dirty_offset)
580 		ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
581 
582 	if (!ret && (dirty_offset + dirty_len < end))
583 		ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
584 					  end - dirty_offset - dirty_len);
585 
586 	return ret;
587 }
588 
589 /* Given an unmapped page, zero it or read in page for COW, page is locked
590  * by caller.
591  */
592 static int
593 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
594 {
595 	struct buffer_head *bh = NULL;
596 	int ret = 0;
597 	sector_t isect;
598 
599 	dprintk("%s enter, %p\n", __func__, page);
600 	BUG_ON(PageUptodate(page));
601 	if (!cow_read) {
602 		zero_user_segment(page, 0, PAGE_SIZE);
603 		SetPageUptodate(page);
604 		goto cleanup;
605 	}
606 
607 	bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
608 	if (!bh) {
609 		ret = -ENOMEM;
610 		goto cleanup;
611 	}
612 
613 	isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
614 	map_block(bh, isect, cow_read);
615 	if (!bh_uptodate_or_lock(bh))
616 		ret = bh_submit_read(bh);
617 	if (ret)
618 		goto cleanup;
619 	SetPageUptodate(page);
620 
621 cleanup:
622 	if (bh)
623 		free_buffer_head(bh);
624 	if (ret) {
625 		/* Need to mark layout with bad read...should now
626 		 * just use nfs4 for reads and writes.
627 		 */
628 		mark_bad_read();
629 	}
630 	return ret;
631 }
632 
633 /* Find or create a zeroing page marked being writeback.
634  * Return ERR_PTR on error, NULL to indicate skip this page and page itself
635  * to indicate write out.
636  */
637 static struct page *
638 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
639 			struct pnfs_block_extent *cow_read)
640 {
641 	struct page *page;
642 	int locked = 0;
643 	page = find_get_page(inode->i_mapping, index);
644 	if (page)
645 		goto check_page;
646 
647 	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
648 	if (unlikely(!page)) {
649 		dprintk("%s oom\n", __func__);
650 		return ERR_PTR(-ENOMEM);
651 	}
652 	locked = 1;
653 
654 check_page:
655 	/* PageDirty: Other will write this out
656 	 * PageWriteback: Other is writing this out
657 	 * PageUptodate: It was read before
658 	 */
659 	if (PageDirty(page) || PageWriteback(page)) {
660 		print_page(page);
661 		if (locked)
662 			unlock_page(page);
663 		page_cache_release(page);
664 		return NULL;
665 	}
666 
667 	if (!locked) {
668 		lock_page(page);
669 		locked = 1;
670 		goto check_page;
671 	}
672 	if (!PageUptodate(page)) {
673 		/* New page, readin or zero it */
674 		init_page_for_write(page, cow_read);
675 	}
676 	set_page_writeback(page);
677 	unlock_page(page);
678 
679 	return page;
680 }
681 
682 static enum pnfs_try_status
683 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
684 {
685 	struct nfs_pgio_header *header = wdata->header;
686 	int i, ret, npg_zero, pg_index, last = 0;
687 	struct bio *bio = NULL;
688 	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
689 	sector_t isect, last_isect = 0, extent_length = 0;
690 	struct parallel_io *par = NULL;
691 	loff_t offset = wdata->args.offset;
692 	size_t count = wdata->args.count;
693 	unsigned int pg_offset, pg_len, saved_len;
694 	struct page **pages = wdata->args.pages;
695 	struct page *page;
696 	pgoff_t index;
697 	u64 temp;
698 	int npg_per_block =
699 	    NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
700 
701 	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
702 
703 	if (header->dreq != NULL &&
704 	    (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
705 	     !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
706 		dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
707 		goto out_mds;
708 	}
709 	/* At this point, wdata->pages is a (sequential) list of nfs_pages.
710 	 * We want to write each, and if there is an error set pnfs_error
711 	 * to have it redone using nfs.
712 	 */
713 	par = alloc_parallel(wdata);
714 	if (!par)
715 		goto out_mds;
716 	par->pnfs_callback = bl_end_par_io_write;
717 	/* At this point, have to be more careful with error handling */
718 
719 	isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
720 	be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
721 	if (!be || !is_writable(be, isect)) {
722 		dprintk("%s no matching extents!\n", __func__);
723 		goto out_mds;
724 	}
725 
726 	/* First page inside INVALID extent */
727 	if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
728 		if (likely(!bl_push_one_short_extent(be->be_inval)))
729 			par->bse_count++;
730 		else
731 			goto out_mds;
732 		temp = offset >> PAGE_CACHE_SHIFT;
733 		npg_zero = do_div(temp, npg_per_block);
734 		isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
735 				     (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
736 		extent_length = be->be_length - (isect - be->be_f_offset);
737 
738 fill_invalid_ext:
739 		dprintk("%s need to zero %d pages\n", __func__, npg_zero);
740 		for (;npg_zero > 0; npg_zero--) {
741 			if (bl_is_sector_init(be->be_inval, isect)) {
742 				dprintk("isect %llu already init\n",
743 					(unsigned long long)isect);
744 				goto next_page;
745 			}
746 			/* page ref released in bl_end_io_write_zero */
747 			index = isect >> PAGE_CACHE_SECTOR_SHIFT;
748 			dprintk("%s zero %dth page: index %lu isect %llu\n",
749 				__func__, npg_zero, index,
750 				(unsigned long long)isect);
751 			page = bl_find_get_zeroing_page(header->inode, index,
752 							cow_read);
753 			if (unlikely(IS_ERR(page))) {
754 				header->pnfs_error = PTR_ERR(page);
755 				goto out;
756 			} else if (page == NULL)
757 				goto next_page;
758 
759 			ret = bl_mark_sectors_init(be->be_inval, isect,
760 						       PAGE_CACHE_SECTORS);
761 			if (unlikely(ret)) {
762 				dprintk("%s bl_mark_sectors_init fail %d\n",
763 					__func__, ret);
764 				end_page_writeback(page);
765 				page_cache_release(page);
766 				header->pnfs_error = ret;
767 				goto out;
768 			}
769 			if (likely(!bl_push_one_short_extent(be->be_inval)))
770 				par->bse_count++;
771 			else {
772 				end_page_writeback(page);
773 				page_cache_release(page);
774 				header->pnfs_error = -ENOMEM;
775 				goto out;
776 			}
777 			/* FIXME: This should be done in bi_end_io */
778 			mark_extents_written(BLK_LSEG2EXT(header->lseg),
779 					     page->index << PAGE_CACHE_SHIFT,
780 					     PAGE_CACHE_SIZE);
781 
782 			bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
783 						 isect, page, be,
784 						 bl_end_io_write_zero, par);
785 			if (IS_ERR(bio)) {
786 				header->pnfs_error = PTR_ERR(bio);
787 				bio = NULL;
788 				goto out;
789 			}
790 next_page:
791 			isect += PAGE_CACHE_SECTORS;
792 			extent_length -= PAGE_CACHE_SECTORS;
793 		}
794 		if (last)
795 			goto write_done;
796 	}
797 	bio = bl_submit_bio(WRITE, bio);
798 
799 	/* Middle pages */
800 	pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
801 	for (i = pg_index; i < wdata->pages.npages; i++) {
802 		if (!extent_length) {
803 			/* We've used up the previous extent */
804 			bl_put_extent(be);
805 			bl_put_extent(cow_read);
806 			bio = bl_submit_bio(WRITE, bio);
807 			/* Get the next one */
808 			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
809 					     isect, &cow_read);
810 			if (!be || !is_writable(be, isect)) {
811 				header->pnfs_error = -EINVAL;
812 				goto out;
813 			}
814 			if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
815 				if (likely(!bl_push_one_short_extent(
816 								be->be_inval)))
817 					par->bse_count++;
818 				else {
819 					header->pnfs_error = -ENOMEM;
820 					goto out;
821 				}
822 			}
823 			extent_length = be->be_length -
824 			    (isect - be->be_f_offset);
825 		}
826 
827 		dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
828 		pg_offset = offset & ~PAGE_CACHE_MASK;
829 		if (pg_offset + count > PAGE_CACHE_SIZE)
830 			pg_len = PAGE_CACHE_SIZE - pg_offset;
831 		else
832 			pg_len = count;
833 
834 		saved_len = pg_len;
835 		if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
836 		    !bl_is_sector_init(be->be_inval, isect)) {
837 			ret = bl_read_partial_page_sync(pages[i], cow_read,
838 							pg_offset, pg_len, true);
839 			if (ret) {
840 				dprintk("%s bl_read_partial_page_sync fail %d\n",
841 					__func__, ret);
842 				header->pnfs_error = ret;
843 				goto out;
844 			}
845 
846 			ret = bl_mark_sectors_init(be->be_inval, isect,
847 						       PAGE_CACHE_SECTORS);
848 			if (unlikely(ret)) {
849 				dprintk("%s bl_mark_sectors_init fail %d\n",
850 					__func__, ret);
851 				header->pnfs_error = ret;
852 				goto out;
853 			}
854 
855 			/* Expand to full page write */
856 			pg_offset = 0;
857 			pg_len = PAGE_CACHE_SIZE;
858 		} else if  ((pg_offset & (SECTOR_SIZE - 1)) ||
859 			    (pg_len & (SECTOR_SIZE - 1))){
860 			/* ahh, nasty case. We have to do sync full sector
861 			 * read-modify-write cycles.
862 			 */
863 			unsigned int saved_offset = pg_offset;
864 			ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
865 							pg_len, false);
866 			pg_offset = round_down(pg_offset, SECTOR_SIZE);
867 			pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
868 				 - pg_offset;
869 		}
870 
871 
872 		bio = do_add_page_to_bio(bio, wdata->pages.npages - i, WRITE,
873 					 isect, pages[i], be,
874 					 bl_end_io_write, par,
875 					 pg_offset, pg_len);
876 		if (IS_ERR(bio)) {
877 			header->pnfs_error = PTR_ERR(bio);
878 			bio = NULL;
879 			goto out;
880 		}
881 		offset += saved_len;
882 		count -= saved_len;
883 		isect += PAGE_CACHE_SECTORS;
884 		last_isect = isect;
885 		extent_length -= PAGE_CACHE_SECTORS;
886 	}
887 
888 	/* Last page inside INVALID extent */
889 	if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
890 		bio = bl_submit_bio(WRITE, bio);
891 		temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
892 		npg_zero = npg_per_block - do_div(temp, npg_per_block);
893 		if (npg_zero < npg_per_block) {
894 			last = 1;
895 			goto fill_invalid_ext;
896 		}
897 	}
898 
899 write_done:
900 	wdata->res.count = wdata->args.count;
901 out:
902 	bl_put_extent(be);
903 	bl_put_extent(cow_read);
904 	bl_submit_bio(WRITE, bio);
905 	put_parallel(par);
906 	return PNFS_ATTEMPTED;
907 out_mds:
908 	bl_put_extent(be);
909 	bl_put_extent(cow_read);
910 	kfree(par);
911 	return PNFS_NOT_ATTEMPTED;
912 }
913 
914 /* FIXME - range ignored */
915 static void
916 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
917 {
918 	int i;
919 	struct pnfs_block_extent *be;
920 
921 	spin_lock(&bl->bl_ext_lock);
922 	for (i = 0; i < EXTENT_LISTS; i++) {
923 		while (!list_empty(&bl->bl_extents[i])) {
924 			be = list_first_entry(&bl->bl_extents[i],
925 					      struct pnfs_block_extent,
926 					      be_node);
927 			list_del(&be->be_node);
928 			bl_put_extent(be);
929 		}
930 	}
931 	spin_unlock(&bl->bl_ext_lock);
932 }
933 
934 static void
935 release_inval_marks(struct pnfs_inval_markings *marks)
936 {
937 	struct pnfs_inval_tracking *pos, *temp;
938 	struct pnfs_block_short_extent *se, *stemp;
939 
940 	list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
941 		list_del(&pos->it_link);
942 		kfree(pos);
943 	}
944 
945 	list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
946 		list_del(&se->bse_node);
947 		kfree(se);
948 	}
949 	return;
950 }
951 
952 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
953 {
954 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
955 
956 	dprintk("%s enter\n", __func__);
957 	release_extents(bl, NULL);
958 	release_inval_marks(&bl->bl_inval);
959 	kfree(bl);
960 }
961 
962 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
963 						   gfp_t gfp_flags)
964 {
965 	struct pnfs_block_layout *bl;
966 
967 	dprintk("%s enter\n", __func__);
968 	bl = kzalloc(sizeof(*bl), gfp_flags);
969 	if (!bl)
970 		return NULL;
971 	spin_lock_init(&bl->bl_ext_lock);
972 	INIT_LIST_HEAD(&bl->bl_extents[0]);
973 	INIT_LIST_HEAD(&bl->bl_extents[1]);
974 	INIT_LIST_HEAD(&bl->bl_commit);
975 	INIT_LIST_HEAD(&bl->bl_committing);
976 	bl->bl_count = 0;
977 	bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
978 	BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
979 	return &bl->bl_layout;
980 }
981 
982 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
983 {
984 	dprintk("%s enter\n", __func__);
985 	kfree(lseg);
986 }
987 
988 /* We pretty much ignore lseg, and store all data layout wide, so we
989  * can correctly merge.
990  */
991 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
992 						 struct nfs4_layoutget_res *lgr,
993 						 gfp_t gfp_flags)
994 {
995 	struct pnfs_layout_segment *lseg;
996 	int status;
997 
998 	dprintk("%s enter\n", __func__);
999 	lseg = kzalloc(sizeof(*lseg), gfp_flags);
1000 	if (!lseg)
1001 		return ERR_PTR(-ENOMEM);
1002 	status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
1003 	if (status) {
1004 		/* We don't want to call the full-blown bl_free_lseg,
1005 		 * since on error extents were not touched.
1006 		 */
1007 		kfree(lseg);
1008 		return ERR_PTR(status);
1009 	}
1010 	return lseg;
1011 }
1012 
1013 static void
1014 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1015 		       const struct nfs4_layoutcommit_args *arg)
1016 {
1017 	dprintk("%s enter\n", __func__);
1018 	encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1019 }
1020 
1021 static void
1022 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1023 {
1024 	struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1025 
1026 	dprintk("%s enter\n", __func__);
1027 	clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1028 }
1029 
1030 static void free_blk_mountid(struct block_mount_id *mid)
1031 {
1032 	if (mid) {
1033 		struct pnfs_block_dev *dev, *tmp;
1034 
1035 		/* No need to take bm_lock as we are last user freeing bm_devlist */
1036 		list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1037 			list_del(&dev->bm_node);
1038 			bl_free_block_dev(dev);
1039 		}
1040 		kfree(mid);
1041 	}
1042 }
1043 
1044 /* This is mostly copied from the filelayout_get_device_info function.
1045  * It seems much of this should be at the generic pnfs level.
1046  */
1047 static struct pnfs_block_dev *
1048 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1049 			struct nfs4_deviceid *d_id)
1050 {
1051 	struct pnfs_device *dev;
1052 	struct pnfs_block_dev *rv;
1053 	u32 max_resp_sz;
1054 	int max_pages;
1055 	struct page **pages = NULL;
1056 	int i, rc;
1057 
1058 	/*
1059 	 * Use the session max response size as the basis for setting
1060 	 * GETDEVICEINFO's maxcount
1061 	 */
1062 	max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1063 	max_pages = nfs_page_array_len(0, max_resp_sz);
1064 	dprintk("%s max_resp_sz %u max_pages %d\n",
1065 		__func__, max_resp_sz, max_pages);
1066 
1067 	dev = kmalloc(sizeof(*dev), GFP_NOFS);
1068 	if (!dev) {
1069 		dprintk("%s kmalloc failed\n", __func__);
1070 		return ERR_PTR(-ENOMEM);
1071 	}
1072 
1073 	pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
1074 	if (pages == NULL) {
1075 		kfree(dev);
1076 		return ERR_PTR(-ENOMEM);
1077 	}
1078 	for (i = 0; i < max_pages; i++) {
1079 		pages[i] = alloc_page(GFP_NOFS);
1080 		if (!pages[i]) {
1081 			rv = ERR_PTR(-ENOMEM);
1082 			goto out_free;
1083 		}
1084 	}
1085 
1086 	memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1087 	dev->layout_type = LAYOUT_BLOCK_VOLUME;
1088 	dev->pages = pages;
1089 	dev->pgbase = 0;
1090 	dev->pglen = PAGE_SIZE * max_pages;
1091 	dev->mincount = 0;
1092 
1093 	dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1094 	rc = nfs4_proc_getdeviceinfo(server, dev);
1095 	dprintk("%s getdevice info returns %d\n", __func__, rc);
1096 	if (rc) {
1097 		rv = ERR_PTR(rc);
1098 		goto out_free;
1099 	}
1100 
1101 	rv = nfs4_blk_decode_device(server, dev);
1102  out_free:
1103 	for (i = 0; i < max_pages; i++)
1104 		__free_page(pages[i]);
1105 	kfree(pages);
1106 	kfree(dev);
1107 	return rv;
1108 }
1109 
1110 static int
1111 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1112 {
1113 	struct block_mount_id *b_mt_id = NULL;
1114 	struct pnfs_devicelist *dlist = NULL;
1115 	struct pnfs_block_dev *bdev;
1116 	LIST_HEAD(block_disklist);
1117 	int status, i;
1118 
1119 	dprintk("%s enter\n", __func__);
1120 
1121 	if (server->pnfs_blksize == 0) {
1122 		dprintk("%s Server did not return blksize\n", __func__);
1123 		return -EINVAL;
1124 	}
1125 	b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1126 	if (!b_mt_id) {
1127 		status = -ENOMEM;
1128 		goto out_error;
1129 	}
1130 	/* Initialize nfs4 block layout mount id */
1131 	spin_lock_init(&b_mt_id->bm_lock);
1132 	INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1133 
1134 	dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1135 	if (!dlist) {
1136 		status = -ENOMEM;
1137 		goto out_error;
1138 	}
1139 	dlist->eof = 0;
1140 	while (!dlist->eof) {
1141 		status = nfs4_proc_getdevicelist(server, fh, dlist);
1142 		if (status)
1143 			goto out_error;
1144 		dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1145 			__func__, dlist->num_devs, dlist->eof);
1146 		for (i = 0; i < dlist->num_devs; i++) {
1147 			bdev = nfs4_blk_get_deviceinfo(server, fh,
1148 						       &dlist->dev_id[i]);
1149 			if (IS_ERR(bdev)) {
1150 				status = PTR_ERR(bdev);
1151 				goto out_error;
1152 			}
1153 			spin_lock(&b_mt_id->bm_lock);
1154 			list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1155 			spin_unlock(&b_mt_id->bm_lock);
1156 		}
1157 	}
1158 	dprintk("%s SUCCESS\n", __func__);
1159 	server->pnfs_ld_data = b_mt_id;
1160 
1161  out_return:
1162 	kfree(dlist);
1163 	return status;
1164 
1165  out_error:
1166 	free_blk_mountid(b_mt_id);
1167 	goto out_return;
1168 }
1169 
1170 static int
1171 bl_clear_layoutdriver(struct nfs_server *server)
1172 {
1173 	struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1174 
1175 	dprintk("%s enter\n", __func__);
1176 	free_blk_mountid(b_mt_id);
1177 	dprintk("%s RETURNS\n", __func__);
1178 	return 0;
1179 }
1180 
1181 static bool
1182 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1183 {
1184 	return IS_ALIGNED(req->wb_offset, alignment) &&
1185 	       IS_ALIGNED(req->wb_bytes, alignment);
1186 }
1187 
1188 static void
1189 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1190 {
1191 	if (pgio->pg_dreq != NULL &&
1192 	    !is_aligned_req(req, SECTOR_SIZE))
1193 		nfs_pageio_reset_read_mds(pgio);
1194 	else
1195 		pnfs_generic_pg_init_read(pgio, req);
1196 }
1197 
1198 static bool
1199 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1200 		struct nfs_page *req)
1201 {
1202 	if (pgio->pg_dreq != NULL &&
1203 	    !is_aligned_req(req, SECTOR_SIZE))
1204 		return false;
1205 
1206 	return pnfs_generic_pg_test(pgio, prev, req);
1207 }
1208 
1209 /*
1210  * Return the number of contiguous bytes for a given inode
1211  * starting at page frame idx.
1212  */
1213 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
1214 {
1215 	struct address_space *mapping = inode->i_mapping;
1216 	pgoff_t end;
1217 
1218 	/* Optimize common case that writes from 0 to end of file */
1219 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
1220 	if (end != NFS_I(inode)->npages) {
1221 		rcu_read_lock();
1222 		end = radix_tree_next_hole(&mapping->page_tree, idx + 1, ULONG_MAX);
1223 		rcu_read_unlock();
1224 	}
1225 
1226 	if (!end)
1227 		return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
1228 	else
1229 		return (end - idx) << PAGE_CACHE_SHIFT;
1230 }
1231 
1232 static void
1233 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1234 {
1235 	if (pgio->pg_dreq != NULL &&
1236 	    !is_aligned_req(req, PAGE_CACHE_SIZE)) {
1237 		nfs_pageio_reset_write_mds(pgio);
1238 	} else {
1239 		u64 wb_size;
1240 		if (pgio->pg_dreq == NULL)
1241 			wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
1242 						      req->wb_index);
1243 		else
1244 			wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1245 
1246 		pnfs_generic_pg_init_write(pgio, req, wb_size);
1247 	}
1248 }
1249 
1250 static bool
1251 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1252 		 struct nfs_page *req)
1253 {
1254 	if (pgio->pg_dreq != NULL &&
1255 	    !is_aligned_req(req, PAGE_CACHE_SIZE))
1256 		return false;
1257 
1258 	return pnfs_generic_pg_test(pgio, prev, req);
1259 }
1260 
1261 static const struct nfs_pageio_ops bl_pg_read_ops = {
1262 	.pg_init = bl_pg_init_read,
1263 	.pg_test = bl_pg_test_read,
1264 	.pg_doio = pnfs_generic_pg_readpages,
1265 };
1266 
1267 static const struct nfs_pageio_ops bl_pg_write_ops = {
1268 	.pg_init = bl_pg_init_write,
1269 	.pg_test = bl_pg_test_write,
1270 	.pg_doio = pnfs_generic_pg_writepages,
1271 };
1272 
1273 static struct pnfs_layoutdriver_type blocklayout_type = {
1274 	.id				= LAYOUT_BLOCK_VOLUME,
1275 	.name				= "LAYOUT_BLOCK_VOLUME",
1276 	.read_pagelist			= bl_read_pagelist,
1277 	.write_pagelist			= bl_write_pagelist,
1278 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
1279 	.free_layout_hdr		= bl_free_layout_hdr,
1280 	.alloc_lseg			= bl_alloc_lseg,
1281 	.free_lseg			= bl_free_lseg,
1282 	.encode_layoutcommit		= bl_encode_layoutcommit,
1283 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
1284 	.set_layoutdriver		= bl_set_layoutdriver,
1285 	.clear_layoutdriver		= bl_clear_layoutdriver,
1286 	.pg_read_ops			= &bl_pg_read_ops,
1287 	.pg_write_ops			= &bl_pg_write_ops,
1288 };
1289 
1290 static const struct rpc_pipe_ops bl_upcall_ops = {
1291 	.upcall		= rpc_pipe_generic_upcall,
1292 	.downcall	= bl_pipe_downcall,
1293 	.destroy_msg	= bl_pipe_destroy_msg,
1294 };
1295 
1296 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1297 					    struct rpc_pipe *pipe)
1298 {
1299 	struct dentry *dir, *dentry;
1300 
1301 	dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1302 	if (dir == NULL)
1303 		return ERR_PTR(-ENOENT);
1304 	dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1305 	dput(dir);
1306 	return dentry;
1307 }
1308 
1309 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1310 					  struct rpc_pipe *pipe)
1311 {
1312 	if (pipe->dentry)
1313 		rpc_unlink(pipe->dentry);
1314 }
1315 
1316 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1317 			   void *ptr)
1318 {
1319 	struct super_block *sb = ptr;
1320 	struct net *net = sb->s_fs_info;
1321 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1322 	struct dentry *dentry;
1323 	int ret = 0;
1324 
1325 	if (!try_module_get(THIS_MODULE))
1326 		return 0;
1327 
1328 	if (nn->bl_device_pipe == NULL) {
1329 		module_put(THIS_MODULE);
1330 		return 0;
1331 	}
1332 
1333 	switch (event) {
1334 	case RPC_PIPEFS_MOUNT:
1335 		dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1336 		if (IS_ERR(dentry)) {
1337 			ret = PTR_ERR(dentry);
1338 			break;
1339 		}
1340 		nn->bl_device_pipe->dentry = dentry;
1341 		break;
1342 	case RPC_PIPEFS_UMOUNT:
1343 		if (nn->bl_device_pipe->dentry)
1344 			nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1345 		break;
1346 	default:
1347 		ret = -ENOTSUPP;
1348 		break;
1349 	}
1350 	module_put(THIS_MODULE);
1351 	return ret;
1352 }
1353 
1354 static struct notifier_block nfs4blocklayout_block = {
1355 	.notifier_call = rpc_pipefs_event,
1356 };
1357 
1358 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1359 						   struct rpc_pipe *pipe)
1360 {
1361 	struct super_block *pipefs_sb;
1362 	struct dentry *dentry;
1363 
1364 	pipefs_sb = rpc_get_sb_net(net);
1365 	if (!pipefs_sb)
1366 		return NULL;
1367 	dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1368 	rpc_put_sb_net(net);
1369 	return dentry;
1370 }
1371 
1372 static void nfs4blocklayout_unregister_net(struct net *net,
1373 					   struct rpc_pipe *pipe)
1374 {
1375 	struct super_block *pipefs_sb;
1376 
1377 	pipefs_sb = rpc_get_sb_net(net);
1378 	if (pipefs_sb) {
1379 		nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1380 		rpc_put_sb_net(net);
1381 	}
1382 }
1383 
1384 static int nfs4blocklayout_net_init(struct net *net)
1385 {
1386 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1387 	struct dentry *dentry;
1388 
1389 	init_waitqueue_head(&nn->bl_wq);
1390 	nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1391 	if (IS_ERR(nn->bl_device_pipe))
1392 		return PTR_ERR(nn->bl_device_pipe);
1393 	dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1394 	if (IS_ERR(dentry)) {
1395 		rpc_destroy_pipe_data(nn->bl_device_pipe);
1396 		return PTR_ERR(dentry);
1397 	}
1398 	nn->bl_device_pipe->dentry = dentry;
1399 	return 0;
1400 }
1401 
1402 static void nfs4blocklayout_net_exit(struct net *net)
1403 {
1404 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1405 
1406 	nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1407 	rpc_destroy_pipe_data(nn->bl_device_pipe);
1408 	nn->bl_device_pipe = NULL;
1409 }
1410 
1411 static struct pernet_operations nfs4blocklayout_net_ops = {
1412 	.init = nfs4blocklayout_net_init,
1413 	.exit = nfs4blocklayout_net_exit,
1414 };
1415 
1416 static int __init nfs4blocklayout_init(void)
1417 {
1418 	int ret;
1419 
1420 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1421 
1422 	ret = pnfs_register_layoutdriver(&blocklayout_type);
1423 	if (ret)
1424 		goto out;
1425 
1426 	ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1427 	if (ret)
1428 		goto out_remove;
1429 	ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1430 	if (ret)
1431 		goto out_notifier;
1432 out:
1433 	return ret;
1434 
1435 out_notifier:
1436 	rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1437 out_remove:
1438 	pnfs_unregister_layoutdriver(&blocklayout_type);
1439 	return ret;
1440 }
1441 
1442 static void __exit nfs4blocklayout_exit(void)
1443 {
1444 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1445 	       __func__);
1446 
1447 	rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1448 	unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1449 	pnfs_unregister_layoutdriver(&blocklayout_type);
1450 }
1451 
1452 MODULE_ALIAS("nfs-layouttype4-3");
1453 
1454 module_init(nfs4blocklayout_init);
1455 module_exit(nfs4blocklayout_exit);
1456