xref: /openbmc/linux/fs/nfs/blocklayout/blocklayout.c (revision afb46f79)
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", bio->bi_iter.bi_size,
138 			(unsigned long long)bio->bi_iter.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_iter.bi_sector = isect - be->be_f_offset +
160 			be->be_v_offset;
161 		bio->bi_bdev = be->be_mdev;
162 		bio->bi_end_io = end_io;
163 		bio->bi_private = par;
164 	}
165 	return bio;
166 }
167 
168 static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw,
169 				      sector_t isect, struct page *page,
170 				      struct pnfs_block_extent *be,
171 				      void (*end_io)(struct bio *, int err),
172 				      struct parallel_io *par,
173 				      unsigned int offset, int len)
174 {
175 	isect = isect + (offset >> SECTOR_SHIFT);
176 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
177 		npg, rw, (unsigned long long)isect, offset, len);
178 retry:
179 	if (!bio) {
180 		bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
181 		if (!bio)
182 			return ERR_PTR(-ENOMEM);
183 	}
184 	if (bio_add_page(bio, page, len, offset) < len) {
185 		bio = bl_submit_bio(rw, bio);
186 		goto retry;
187 	}
188 	return bio;
189 }
190 
191 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
192 				      sector_t isect, struct page *page,
193 				      struct pnfs_block_extent *be,
194 				      void (*end_io)(struct bio *, int err),
195 				      struct parallel_io *par)
196 {
197 	return do_add_page_to_bio(bio, npg, rw, isect, page, be,
198 				  end_io, par, 0, PAGE_CACHE_SIZE);
199 }
200 
201 /* This is basically copied from mpage_end_io_read */
202 static void bl_end_io_read(struct bio *bio, int err)
203 {
204 	struct parallel_io *par = bio->bi_private;
205 	struct bio_vec *bvec;
206 	int i;
207 
208 	if (!err)
209 		bio_for_each_segment_all(bvec, bio, i)
210 			SetPageUptodate(bvec->bv_page);
211 
212 	if (err) {
213 		struct nfs_read_data *rdata = par->data;
214 		struct nfs_pgio_header *header = rdata->header;
215 
216 		if (!header->pnfs_error)
217 			header->pnfs_error = -EIO;
218 		pnfs_set_lo_fail(header->lseg);
219 	}
220 	bio_put(bio);
221 	put_parallel(par);
222 }
223 
224 static void bl_read_cleanup(struct work_struct *work)
225 {
226 	struct rpc_task *task;
227 	struct nfs_read_data *rdata;
228 	dprintk("%s enter\n", __func__);
229 	task = container_of(work, struct rpc_task, u.tk_work);
230 	rdata = container_of(task, struct nfs_read_data, task);
231 	pnfs_ld_read_done(rdata);
232 }
233 
234 static void
235 bl_end_par_io_read(void *data, int unused)
236 {
237 	struct nfs_read_data *rdata = data;
238 
239 	rdata->task.tk_status = rdata->header->pnfs_error;
240 	INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
241 	schedule_work(&rdata->task.u.tk_work);
242 }
243 
244 static enum pnfs_try_status
245 bl_read_pagelist(struct nfs_read_data *rdata)
246 {
247 	struct nfs_pgio_header *header = rdata->header;
248 	int i, hole;
249 	struct bio *bio = NULL;
250 	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
251 	sector_t isect, extent_length = 0;
252 	struct parallel_io *par;
253 	loff_t f_offset = rdata->args.offset;
254 	size_t bytes_left = rdata->args.count;
255 	unsigned int pg_offset, pg_len;
256 	struct page **pages = rdata->args.pages;
257 	int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
258 	const bool is_dio = (header->dreq != NULL);
259 
260 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
261 	       rdata->pages.npages, f_offset, (unsigned int)rdata->args.count);
262 
263 	par = alloc_parallel(rdata);
264 	if (!par)
265 		goto use_mds;
266 	par->pnfs_callback = bl_end_par_io_read;
267 	/* At this point, we can no longer jump to use_mds */
268 
269 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
270 	/* Code assumes extents are page-aligned */
271 	for (i = pg_index; i < rdata->pages.npages; i++) {
272 		if (!extent_length) {
273 			/* We've used up the previous extent */
274 			bl_put_extent(be);
275 			bl_put_extent(cow_read);
276 			bio = bl_submit_bio(READ, bio);
277 			/* Get the next one */
278 			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
279 					     isect, &cow_read);
280 			if (!be) {
281 				header->pnfs_error = -EIO;
282 				goto out;
283 			}
284 			extent_length = be->be_length -
285 				(isect - be->be_f_offset);
286 			if (cow_read) {
287 				sector_t cow_length = cow_read->be_length -
288 					(isect - cow_read->be_f_offset);
289 				extent_length = min(extent_length, cow_length);
290 			}
291 		}
292 
293 		if (is_dio) {
294 			pg_offset = f_offset & ~PAGE_CACHE_MASK;
295 			if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
296 				pg_len = PAGE_CACHE_SIZE - pg_offset;
297 			else
298 				pg_len = bytes_left;
299 
300 			f_offset += pg_len;
301 			bytes_left -= pg_len;
302 			isect += (pg_offset >> SECTOR_SHIFT);
303 		} else {
304 			pg_offset = 0;
305 			pg_len = PAGE_CACHE_SIZE;
306 		}
307 
308 		hole = is_hole(be, isect);
309 		if (hole && !cow_read) {
310 			bio = bl_submit_bio(READ, bio);
311 			/* Fill hole w/ zeroes w/o accessing device */
312 			dprintk("%s Zeroing page for hole\n", __func__);
313 			zero_user_segment(pages[i], pg_offset, pg_len);
314 			print_page(pages[i]);
315 			SetPageUptodate(pages[i]);
316 		} else {
317 			struct pnfs_block_extent *be_read;
318 
319 			be_read = (hole && cow_read) ? cow_read : be;
320 			bio = do_add_page_to_bio(bio, rdata->pages.npages - i,
321 						 READ,
322 						 isect, pages[i], be_read,
323 						 bl_end_io_read, par,
324 						 pg_offset, pg_len);
325 			if (IS_ERR(bio)) {
326 				header->pnfs_error = PTR_ERR(bio);
327 				bio = NULL;
328 				goto out;
329 			}
330 		}
331 		isect += (pg_len >> SECTOR_SHIFT);
332 		extent_length -= PAGE_CACHE_SECTORS;
333 	}
334 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
335 		rdata->res.eof = 1;
336 		rdata->res.count = header->inode->i_size - rdata->args.offset;
337 	} else {
338 		rdata->res.count = (isect << SECTOR_SHIFT) - rdata->args.offset;
339 	}
340 out:
341 	bl_put_extent(be);
342 	bl_put_extent(cow_read);
343 	bl_submit_bio(READ, bio);
344 	put_parallel(par);
345 	return PNFS_ATTEMPTED;
346 
347  use_mds:
348 	dprintk("Giving up and using normal NFS\n");
349 	return PNFS_NOT_ATTEMPTED;
350 }
351 
352 static void mark_extents_written(struct pnfs_block_layout *bl,
353 				 __u64 offset, __u32 count)
354 {
355 	sector_t isect, end;
356 	struct pnfs_block_extent *be;
357 	struct pnfs_block_short_extent *se;
358 
359 	dprintk("%s(%llu, %u)\n", __func__, offset, count);
360 	if (count == 0)
361 		return;
362 	isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
363 	end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
364 	end >>= SECTOR_SHIFT;
365 	while (isect < end) {
366 		sector_t len;
367 		be = bl_find_get_extent(bl, isect, NULL);
368 		BUG_ON(!be); /* FIXME */
369 		len = min(end, be->be_f_offset + be->be_length) - isect;
370 		if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
371 			se = bl_pop_one_short_extent(be->be_inval);
372 			BUG_ON(!se);
373 			bl_mark_for_commit(be, isect, len, se);
374 		}
375 		isect += len;
376 		bl_put_extent(be);
377 	}
378 }
379 
380 static void bl_end_io_write_zero(struct bio *bio, int err)
381 {
382 	struct parallel_io *par = bio->bi_private;
383 	struct bio_vec *bvec;
384 	int i;
385 
386 	bio_for_each_segment_all(bvec, bio, i) {
387 		/* This is the zeroing page we added */
388 		end_page_writeback(bvec->bv_page);
389 		page_cache_release(bvec->bv_page);
390 	}
391 
392 	if (unlikely(err)) {
393 		struct nfs_write_data *data = par->data;
394 		struct nfs_pgio_header *header = data->header;
395 
396 		if (!header->pnfs_error)
397 			header->pnfs_error = -EIO;
398 		pnfs_set_lo_fail(header->lseg);
399 	}
400 	bio_put(bio);
401 	put_parallel(par);
402 }
403 
404 static void bl_end_io_write(struct bio *bio, int err)
405 {
406 	struct parallel_io *par = bio->bi_private;
407 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
408 	struct nfs_write_data *data = par->data;
409 	struct nfs_pgio_header *header = data->header;
410 
411 	if (!uptodate) {
412 		if (!header->pnfs_error)
413 			header->pnfs_error = -EIO;
414 		pnfs_set_lo_fail(header->lseg);
415 	}
416 	bio_put(bio);
417 	put_parallel(par);
418 }
419 
420 /* Function scheduled for call during bl_end_par_io_write,
421  * it marks sectors as written and extends the commitlist.
422  */
423 static void bl_write_cleanup(struct work_struct *work)
424 {
425 	struct rpc_task *task;
426 	struct nfs_write_data *wdata;
427 	dprintk("%s enter\n", __func__);
428 	task = container_of(work, struct rpc_task, u.tk_work);
429 	wdata = container_of(task, struct nfs_write_data, task);
430 	if (likely(!wdata->header->pnfs_error)) {
431 		/* Marks for LAYOUTCOMMIT */
432 		mark_extents_written(BLK_LSEG2EXT(wdata->header->lseg),
433 				     wdata->args.offset, wdata->args.count);
434 	}
435 	pnfs_ld_write_done(wdata);
436 }
437 
438 /* Called when last of bios associated with a bl_write_pagelist call finishes */
439 static void bl_end_par_io_write(void *data, int num_se)
440 {
441 	struct nfs_write_data *wdata = data;
442 
443 	if (unlikely(wdata->header->pnfs_error)) {
444 		bl_free_short_extents(&BLK_LSEG2EXT(wdata->header->lseg)->bl_inval,
445 					num_se);
446 	}
447 
448 	wdata->task.tk_status = wdata->header->pnfs_error;
449 	wdata->verf.committed = NFS_FILE_SYNC;
450 	INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
451 	schedule_work(&wdata->task.u.tk_work);
452 }
453 
454 /* FIXME STUB - mark intersection of layout and page as bad, so is not
455  * used again.
456  */
457 static void mark_bad_read(void)
458 {
459 	return;
460 }
461 
462 /*
463  * map_block:  map a requested I/0 block (isect) into an offset in the LVM
464  * block_device
465  */
466 static void
467 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
468 {
469 	dprintk("%s enter be=%p\n", __func__, be);
470 
471 	set_buffer_mapped(bh);
472 	bh->b_bdev = be->be_mdev;
473 	bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
474 	    (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
475 
476 	dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
477 		__func__, (unsigned long long)isect, (long)bh->b_blocknr,
478 		bh->b_size);
479 	return;
480 }
481 
482 static void
483 bl_read_single_end_io(struct bio *bio, int error)
484 {
485 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
486 	struct page *page = bvec->bv_page;
487 
488 	/* Only one page in bvec */
489 	unlock_page(page);
490 }
491 
492 static int
493 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
494 		    unsigned int offset, unsigned int len)
495 {
496 	struct bio *bio;
497 	struct page *shadow_page;
498 	sector_t isect;
499 	char *kaddr, *kshadow_addr;
500 	int ret = 0;
501 
502 	dprintk("%s: offset %u len %u\n", __func__, offset, len);
503 
504 	shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
505 	if (shadow_page == NULL)
506 		return -ENOMEM;
507 
508 	bio = bio_alloc(GFP_NOIO, 1);
509 	if (bio == NULL)
510 		return -ENOMEM;
511 
512 	isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
513 		(offset / SECTOR_SIZE);
514 
515 	bio->bi_iter.bi_sector = isect - be->be_f_offset + be->be_v_offset;
516 	bio->bi_bdev = be->be_mdev;
517 	bio->bi_end_io = bl_read_single_end_io;
518 
519 	lock_page(shadow_page);
520 	if (bio_add_page(bio, shadow_page,
521 			 SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
522 		unlock_page(shadow_page);
523 		bio_put(bio);
524 		return -EIO;
525 	}
526 
527 	submit_bio(READ, bio);
528 	wait_on_page_locked(shadow_page);
529 	if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
530 		ret = -EIO;
531 	} else {
532 		kaddr = kmap_atomic(page);
533 		kshadow_addr = kmap_atomic(shadow_page);
534 		memcpy(kaddr + offset, kshadow_addr + offset, len);
535 		kunmap_atomic(kshadow_addr);
536 		kunmap_atomic(kaddr);
537 	}
538 	__free_page(shadow_page);
539 	bio_put(bio);
540 
541 	return ret;
542 }
543 
544 static int
545 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
546 			  unsigned int dirty_offset, unsigned int dirty_len,
547 			  bool full_page)
548 {
549 	int ret = 0;
550 	unsigned int start, end;
551 
552 	if (full_page) {
553 		start = 0;
554 		end = PAGE_CACHE_SIZE;
555 	} else {
556 		start = round_down(dirty_offset, SECTOR_SIZE);
557 		end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
558 	}
559 
560 	dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
561 	if (!be) {
562 		zero_user_segments(page, start, dirty_offset,
563 				   dirty_offset + dirty_len, end);
564 		if (start == 0 && end == PAGE_CACHE_SIZE &&
565 		    trylock_page(page)) {
566 			SetPageUptodate(page);
567 			unlock_page(page);
568 		}
569 		return ret;
570 	}
571 
572 	if (start != dirty_offset)
573 		ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
574 
575 	if (!ret && (dirty_offset + dirty_len < end))
576 		ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
577 					  end - dirty_offset - dirty_len);
578 
579 	return ret;
580 }
581 
582 /* Given an unmapped page, zero it or read in page for COW, page is locked
583  * by caller.
584  */
585 static int
586 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
587 {
588 	struct buffer_head *bh = NULL;
589 	int ret = 0;
590 	sector_t isect;
591 
592 	dprintk("%s enter, %p\n", __func__, page);
593 	BUG_ON(PageUptodate(page));
594 	if (!cow_read) {
595 		zero_user_segment(page, 0, PAGE_SIZE);
596 		SetPageUptodate(page);
597 		goto cleanup;
598 	}
599 
600 	bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
601 	if (!bh) {
602 		ret = -ENOMEM;
603 		goto cleanup;
604 	}
605 
606 	isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
607 	map_block(bh, isect, cow_read);
608 	if (!bh_uptodate_or_lock(bh))
609 		ret = bh_submit_read(bh);
610 	if (ret)
611 		goto cleanup;
612 	SetPageUptodate(page);
613 
614 cleanup:
615 	if (bh)
616 		free_buffer_head(bh);
617 	if (ret) {
618 		/* Need to mark layout with bad read...should now
619 		 * just use nfs4 for reads and writes.
620 		 */
621 		mark_bad_read();
622 	}
623 	return ret;
624 }
625 
626 /* Find or create a zeroing page marked being writeback.
627  * Return ERR_PTR on error, NULL to indicate skip this page and page itself
628  * to indicate write out.
629  */
630 static struct page *
631 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
632 			struct pnfs_block_extent *cow_read)
633 {
634 	struct page *page;
635 	int locked = 0;
636 	page = find_get_page(inode->i_mapping, index);
637 	if (page)
638 		goto check_page;
639 
640 	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
641 	if (unlikely(!page)) {
642 		dprintk("%s oom\n", __func__);
643 		return ERR_PTR(-ENOMEM);
644 	}
645 	locked = 1;
646 
647 check_page:
648 	/* PageDirty: Other will write this out
649 	 * PageWriteback: Other is writing this out
650 	 * PageUptodate: It was read before
651 	 */
652 	if (PageDirty(page) || PageWriteback(page)) {
653 		print_page(page);
654 		if (locked)
655 			unlock_page(page);
656 		page_cache_release(page);
657 		return NULL;
658 	}
659 
660 	if (!locked) {
661 		lock_page(page);
662 		locked = 1;
663 		goto check_page;
664 	}
665 	if (!PageUptodate(page)) {
666 		/* New page, readin or zero it */
667 		init_page_for_write(page, cow_read);
668 	}
669 	set_page_writeback(page);
670 	unlock_page(page);
671 
672 	return page;
673 }
674 
675 static enum pnfs_try_status
676 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
677 {
678 	struct nfs_pgio_header *header = wdata->header;
679 	int i, ret, npg_zero, pg_index, last = 0;
680 	struct bio *bio = NULL;
681 	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
682 	sector_t isect, last_isect = 0, extent_length = 0;
683 	struct parallel_io *par = NULL;
684 	loff_t offset = wdata->args.offset;
685 	size_t count = wdata->args.count;
686 	unsigned int pg_offset, pg_len, saved_len;
687 	struct page **pages = wdata->args.pages;
688 	struct page *page;
689 	pgoff_t index;
690 	u64 temp;
691 	int npg_per_block =
692 	    NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
693 
694 	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
695 
696 	if (header->dreq != NULL &&
697 	    (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
698 	     !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
699 		dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
700 		goto out_mds;
701 	}
702 	/* At this point, wdata->pages is a (sequential) list of nfs_pages.
703 	 * We want to write each, and if there is an error set pnfs_error
704 	 * to have it redone using nfs.
705 	 */
706 	par = alloc_parallel(wdata);
707 	if (!par)
708 		goto out_mds;
709 	par->pnfs_callback = bl_end_par_io_write;
710 	/* At this point, have to be more careful with error handling */
711 
712 	isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
713 	be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
714 	if (!be || !is_writable(be, isect)) {
715 		dprintk("%s no matching extents!\n", __func__);
716 		goto out_mds;
717 	}
718 
719 	/* First page inside INVALID extent */
720 	if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
721 		if (likely(!bl_push_one_short_extent(be->be_inval)))
722 			par->bse_count++;
723 		else
724 			goto out_mds;
725 		temp = offset >> PAGE_CACHE_SHIFT;
726 		npg_zero = do_div(temp, npg_per_block);
727 		isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
728 				     (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
729 		extent_length = be->be_length - (isect - be->be_f_offset);
730 
731 fill_invalid_ext:
732 		dprintk("%s need to zero %d pages\n", __func__, npg_zero);
733 		for (;npg_zero > 0; npg_zero--) {
734 			if (bl_is_sector_init(be->be_inval, isect)) {
735 				dprintk("isect %llu already init\n",
736 					(unsigned long long)isect);
737 				goto next_page;
738 			}
739 			/* page ref released in bl_end_io_write_zero */
740 			index = isect >> PAGE_CACHE_SECTOR_SHIFT;
741 			dprintk("%s zero %dth page: index %lu isect %llu\n",
742 				__func__, npg_zero, index,
743 				(unsigned long long)isect);
744 			page = bl_find_get_zeroing_page(header->inode, index,
745 							cow_read);
746 			if (unlikely(IS_ERR(page))) {
747 				header->pnfs_error = PTR_ERR(page);
748 				goto out;
749 			} else if (page == NULL)
750 				goto next_page;
751 
752 			ret = bl_mark_sectors_init(be->be_inval, isect,
753 						       PAGE_CACHE_SECTORS);
754 			if (unlikely(ret)) {
755 				dprintk("%s bl_mark_sectors_init fail %d\n",
756 					__func__, ret);
757 				end_page_writeback(page);
758 				page_cache_release(page);
759 				header->pnfs_error = ret;
760 				goto out;
761 			}
762 			if (likely(!bl_push_one_short_extent(be->be_inval)))
763 				par->bse_count++;
764 			else {
765 				end_page_writeback(page);
766 				page_cache_release(page);
767 				header->pnfs_error = -ENOMEM;
768 				goto out;
769 			}
770 			/* FIXME: This should be done in bi_end_io */
771 			mark_extents_written(BLK_LSEG2EXT(header->lseg),
772 					     page->index << PAGE_CACHE_SHIFT,
773 					     PAGE_CACHE_SIZE);
774 
775 			bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
776 						 isect, page, be,
777 						 bl_end_io_write_zero, par);
778 			if (IS_ERR(bio)) {
779 				header->pnfs_error = PTR_ERR(bio);
780 				bio = NULL;
781 				goto out;
782 			}
783 next_page:
784 			isect += PAGE_CACHE_SECTORS;
785 			extent_length -= PAGE_CACHE_SECTORS;
786 		}
787 		if (last)
788 			goto write_done;
789 	}
790 	bio = bl_submit_bio(WRITE, bio);
791 
792 	/* Middle pages */
793 	pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
794 	for (i = pg_index; i < wdata->pages.npages; i++) {
795 		if (!extent_length) {
796 			/* We've used up the previous extent */
797 			bl_put_extent(be);
798 			bl_put_extent(cow_read);
799 			bio = bl_submit_bio(WRITE, bio);
800 			/* Get the next one */
801 			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
802 					     isect, &cow_read);
803 			if (!be || !is_writable(be, isect)) {
804 				header->pnfs_error = -EINVAL;
805 				goto out;
806 			}
807 			if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
808 				if (likely(!bl_push_one_short_extent(
809 								be->be_inval)))
810 					par->bse_count++;
811 				else {
812 					header->pnfs_error = -ENOMEM;
813 					goto out;
814 				}
815 			}
816 			extent_length = be->be_length -
817 			    (isect - be->be_f_offset);
818 		}
819 
820 		dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
821 		pg_offset = offset & ~PAGE_CACHE_MASK;
822 		if (pg_offset + count > PAGE_CACHE_SIZE)
823 			pg_len = PAGE_CACHE_SIZE - pg_offset;
824 		else
825 			pg_len = count;
826 
827 		saved_len = pg_len;
828 		if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
829 		    !bl_is_sector_init(be->be_inval, isect)) {
830 			ret = bl_read_partial_page_sync(pages[i], cow_read,
831 							pg_offset, pg_len, true);
832 			if (ret) {
833 				dprintk("%s bl_read_partial_page_sync fail %d\n",
834 					__func__, ret);
835 				header->pnfs_error = ret;
836 				goto out;
837 			}
838 
839 			ret = bl_mark_sectors_init(be->be_inval, isect,
840 						       PAGE_CACHE_SECTORS);
841 			if (unlikely(ret)) {
842 				dprintk("%s bl_mark_sectors_init fail %d\n",
843 					__func__, ret);
844 				header->pnfs_error = ret;
845 				goto out;
846 			}
847 
848 			/* Expand to full page write */
849 			pg_offset = 0;
850 			pg_len = PAGE_CACHE_SIZE;
851 		} else if  ((pg_offset & (SECTOR_SIZE - 1)) ||
852 			    (pg_len & (SECTOR_SIZE - 1))){
853 			/* ahh, nasty case. We have to do sync full sector
854 			 * read-modify-write cycles.
855 			 */
856 			unsigned int saved_offset = pg_offset;
857 			ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
858 							pg_len, false);
859 			pg_offset = round_down(pg_offset, SECTOR_SIZE);
860 			pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
861 				 - pg_offset;
862 		}
863 
864 
865 		bio = do_add_page_to_bio(bio, wdata->pages.npages - i, WRITE,
866 					 isect, pages[i], be,
867 					 bl_end_io_write, par,
868 					 pg_offset, pg_len);
869 		if (IS_ERR(bio)) {
870 			header->pnfs_error = PTR_ERR(bio);
871 			bio = NULL;
872 			goto out;
873 		}
874 		offset += saved_len;
875 		count -= saved_len;
876 		isect += PAGE_CACHE_SECTORS;
877 		last_isect = isect;
878 		extent_length -= PAGE_CACHE_SECTORS;
879 	}
880 
881 	/* Last page inside INVALID extent */
882 	if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
883 		bio = bl_submit_bio(WRITE, bio);
884 		temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
885 		npg_zero = npg_per_block - do_div(temp, npg_per_block);
886 		if (npg_zero < npg_per_block) {
887 			last = 1;
888 			goto fill_invalid_ext;
889 		}
890 	}
891 
892 write_done:
893 	wdata->res.count = wdata->args.count;
894 out:
895 	bl_put_extent(be);
896 	bl_put_extent(cow_read);
897 	bl_submit_bio(WRITE, bio);
898 	put_parallel(par);
899 	return PNFS_ATTEMPTED;
900 out_mds:
901 	bl_put_extent(be);
902 	bl_put_extent(cow_read);
903 	kfree(par);
904 	return PNFS_NOT_ATTEMPTED;
905 }
906 
907 /* FIXME - range ignored */
908 static void
909 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
910 {
911 	int i;
912 	struct pnfs_block_extent *be;
913 
914 	spin_lock(&bl->bl_ext_lock);
915 	for (i = 0; i < EXTENT_LISTS; i++) {
916 		while (!list_empty(&bl->bl_extents[i])) {
917 			be = list_first_entry(&bl->bl_extents[i],
918 					      struct pnfs_block_extent,
919 					      be_node);
920 			list_del(&be->be_node);
921 			bl_put_extent(be);
922 		}
923 	}
924 	spin_unlock(&bl->bl_ext_lock);
925 }
926 
927 static void
928 release_inval_marks(struct pnfs_inval_markings *marks)
929 {
930 	struct pnfs_inval_tracking *pos, *temp;
931 	struct pnfs_block_short_extent *se, *stemp;
932 
933 	list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
934 		list_del(&pos->it_link);
935 		kfree(pos);
936 	}
937 
938 	list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
939 		list_del(&se->bse_node);
940 		kfree(se);
941 	}
942 	return;
943 }
944 
945 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
946 {
947 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
948 
949 	dprintk("%s enter\n", __func__);
950 	release_extents(bl, NULL);
951 	release_inval_marks(&bl->bl_inval);
952 	kfree(bl);
953 }
954 
955 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
956 						   gfp_t gfp_flags)
957 {
958 	struct pnfs_block_layout *bl;
959 
960 	dprintk("%s enter\n", __func__);
961 	bl = kzalloc(sizeof(*bl), gfp_flags);
962 	if (!bl)
963 		return NULL;
964 	spin_lock_init(&bl->bl_ext_lock);
965 	INIT_LIST_HEAD(&bl->bl_extents[0]);
966 	INIT_LIST_HEAD(&bl->bl_extents[1]);
967 	INIT_LIST_HEAD(&bl->bl_commit);
968 	INIT_LIST_HEAD(&bl->bl_committing);
969 	bl->bl_count = 0;
970 	bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
971 	BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
972 	return &bl->bl_layout;
973 }
974 
975 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
976 {
977 	dprintk("%s enter\n", __func__);
978 	kfree(lseg);
979 }
980 
981 /* We pretty much ignore lseg, and store all data layout wide, so we
982  * can correctly merge.
983  */
984 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
985 						 struct nfs4_layoutget_res *lgr,
986 						 gfp_t gfp_flags)
987 {
988 	struct pnfs_layout_segment *lseg;
989 	int status;
990 
991 	dprintk("%s enter\n", __func__);
992 	lseg = kzalloc(sizeof(*lseg), gfp_flags);
993 	if (!lseg)
994 		return ERR_PTR(-ENOMEM);
995 	status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
996 	if (status) {
997 		/* We don't want to call the full-blown bl_free_lseg,
998 		 * since on error extents were not touched.
999 		 */
1000 		kfree(lseg);
1001 		return ERR_PTR(status);
1002 	}
1003 	return lseg;
1004 }
1005 
1006 static void
1007 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1008 		       const struct nfs4_layoutcommit_args *arg)
1009 {
1010 	dprintk("%s enter\n", __func__);
1011 	encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1012 }
1013 
1014 static void
1015 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1016 {
1017 	struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1018 
1019 	dprintk("%s enter\n", __func__);
1020 	clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1021 }
1022 
1023 static void free_blk_mountid(struct block_mount_id *mid)
1024 {
1025 	if (mid) {
1026 		struct pnfs_block_dev *dev, *tmp;
1027 
1028 		/* No need to take bm_lock as we are last user freeing bm_devlist */
1029 		list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1030 			list_del(&dev->bm_node);
1031 			bl_free_block_dev(dev);
1032 		}
1033 		kfree(mid);
1034 	}
1035 }
1036 
1037 /* This is mostly copied from the filelayout_get_device_info function.
1038  * It seems much of this should be at the generic pnfs level.
1039  */
1040 static struct pnfs_block_dev *
1041 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1042 			struct nfs4_deviceid *d_id)
1043 {
1044 	struct pnfs_device *dev;
1045 	struct pnfs_block_dev *rv;
1046 	u32 max_resp_sz;
1047 	int max_pages;
1048 	struct page **pages = NULL;
1049 	int i, rc;
1050 
1051 	/*
1052 	 * Use the session max response size as the basis for setting
1053 	 * GETDEVICEINFO's maxcount
1054 	 */
1055 	max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1056 	max_pages = nfs_page_array_len(0, max_resp_sz);
1057 	dprintk("%s max_resp_sz %u max_pages %d\n",
1058 		__func__, max_resp_sz, max_pages);
1059 
1060 	dev = kmalloc(sizeof(*dev), GFP_NOFS);
1061 	if (!dev) {
1062 		dprintk("%s kmalloc failed\n", __func__);
1063 		return ERR_PTR(-ENOMEM);
1064 	}
1065 
1066 	pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
1067 	if (pages == NULL) {
1068 		kfree(dev);
1069 		return ERR_PTR(-ENOMEM);
1070 	}
1071 	for (i = 0; i < max_pages; i++) {
1072 		pages[i] = alloc_page(GFP_NOFS);
1073 		if (!pages[i]) {
1074 			rv = ERR_PTR(-ENOMEM);
1075 			goto out_free;
1076 		}
1077 	}
1078 
1079 	memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1080 	dev->layout_type = LAYOUT_BLOCK_VOLUME;
1081 	dev->pages = pages;
1082 	dev->pgbase = 0;
1083 	dev->pglen = PAGE_SIZE * max_pages;
1084 	dev->mincount = 0;
1085 	dev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead;
1086 
1087 	dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1088 	rc = nfs4_proc_getdeviceinfo(server, dev, NULL);
1089 	dprintk("%s getdevice info returns %d\n", __func__, rc);
1090 	if (rc) {
1091 		rv = ERR_PTR(rc);
1092 		goto out_free;
1093 	}
1094 
1095 	rv = nfs4_blk_decode_device(server, dev);
1096  out_free:
1097 	for (i = 0; i < max_pages; i++)
1098 		__free_page(pages[i]);
1099 	kfree(pages);
1100 	kfree(dev);
1101 	return rv;
1102 }
1103 
1104 static int
1105 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1106 {
1107 	struct block_mount_id *b_mt_id = NULL;
1108 	struct pnfs_devicelist *dlist = NULL;
1109 	struct pnfs_block_dev *bdev;
1110 	LIST_HEAD(block_disklist);
1111 	int status, i;
1112 
1113 	dprintk("%s enter\n", __func__);
1114 
1115 	if (server->pnfs_blksize == 0) {
1116 		dprintk("%s Server did not return blksize\n", __func__);
1117 		return -EINVAL;
1118 	}
1119 	b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1120 	if (!b_mt_id) {
1121 		status = -ENOMEM;
1122 		goto out_error;
1123 	}
1124 	/* Initialize nfs4 block layout mount id */
1125 	spin_lock_init(&b_mt_id->bm_lock);
1126 	INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1127 
1128 	dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1129 	if (!dlist) {
1130 		status = -ENOMEM;
1131 		goto out_error;
1132 	}
1133 	dlist->eof = 0;
1134 	while (!dlist->eof) {
1135 		status = nfs4_proc_getdevicelist(server, fh, dlist);
1136 		if (status)
1137 			goto out_error;
1138 		dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1139 			__func__, dlist->num_devs, dlist->eof);
1140 		for (i = 0; i < dlist->num_devs; i++) {
1141 			bdev = nfs4_blk_get_deviceinfo(server, fh,
1142 						       &dlist->dev_id[i]);
1143 			if (IS_ERR(bdev)) {
1144 				status = PTR_ERR(bdev);
1145 				goto out_error;
1146 			}
1147 			spin_lock(&b_mt_id->bm_lock);
1148 			list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1149 			spin_unlock(&b_mt_id->bm_lock);
1150 		}
1151 	}
1152 	dprintk("%s SUCCESS\n", __func__);
1153 	server->pnfs_ld_data = b_mt_id;
1154 
1155  out_return:
1156 	kfree(dlist);
1157 	return status;
1158 
1159  out_error:
1160 	free_blk_mountid(b_mt_id);
1161 	goto out_return;
1162 }
1163 
1164 static int
1165 bl_clear_layoutdriver(struct nfs_server *server)
1166 {
1167 	struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1168 
1169 	dprintk("%s enter\n", __func__);
1170 	free_blk_mountid(b_mt_id);
1171 	dprintk("%s RETURNS\n", __func__);
1172 	return 0;
1173 }
1174 
1175 static bool
1176 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1177 {
1178 	return IS_ALIGNED(req->wb_offset, alignment) &&
1179 	       IS_ALIGNED(req->wb_bytes, alignment);
1180 }
1181 
1182 static void
1183 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1184 {
1185 	if (pgio->pg_dreq != NULL &&
1186 	    !is_aligned_req(req, SECTOR_SIZE))
1187 		nfs_pageio_reset_read_mds(pgio);
1188 	else
1189 		pnfs_generic_pg_init_read(pgio, req);
1190 }
1191 
1192 static bool
1193 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1194 		struct nfs_page *req)
1195 {
1196 	if (pgio->pg_dreq != NULL &&
1197 	    !is_aligned_req(req, SECTOR_SIZE))
1198 		return false;
1199 
1200 	return pnfs_generic_pg_test(pgio, prev, req);
1201 }
1202 
1203 /*
1204  * Return the number of contiguous bytes for a given inode
1205  * starting at page frame idx.
1206  */
1207 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
1208 {
1209 	struct address_space *mapping = inode->i_mapping;
1210 	pgoff_t end;
1211 
1212 	/* Optimize common case that writes from 0 to end of file */
1213 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
1214 	if (end != NFS_I(inode)->npages) {
1215 		rcu_read_lock();
1216 		end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
1217 		rcu_read_unlock();
1218 	}
1219 
1220 	if (!end)
1221 		return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
1222 	else
1223 		return (end - idx) << PAGE_CACHE_SHIFT;
1224 }
1225 
1226 static void
1227 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1228 {
1229 	if (pgio->pg_dreq != NULL &&
1230 	    !is_aligned_req(req, PAGE_CACHE_SIZE)) {
1231 		nfs_pageio_reset_write_mds(pgio);
1232 	} else {
1233 		u64 wb_size;
1234 		if (pgio->pg_dreq == NULL)
1235 			wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
1236 						      req->wb_index);
1237 		else
1238 			wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1239 
1240 		pnfs_generic_pg_init_write(pgio, req, wb_size);
1241 	}
1242 }
1243 
1244 static bool
1245 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1246 		 struct nfs_page *req)
1247 {
1248 	if (pgio->pg_dreq != NULL &&
1249 	    !is_aligned_req(req, PAGE_CACHE_SIZE))
1250 		return false;
1251 
1252 	return pnfs_generic_pg_test(pgio, prev, req);
1253 }
1254 
1255 static const struct nfs_pageio_ops bl_pg_read_ops = {
1256 	.pg_init = bl_pg_init_read,
1257 	.pg_test = bl_pg_test_read,
1258 	.pg_doio = pnfs_generic_pg_readpages,
1259 };
1260 
1261 static const struct nfs_pageio_ops bl_pg_write_ops = {
1262 	.pg_init = bl_pg_init_write,
1263 	.pg_test = bl_pg_test_write,
1264 	.pg_doio = pnfs_generic_pg_writepages,
1265 };
1266 
1267 static struct pnfs_layoutdriver_type blocklayout_type = {
1268 	.id				= LAYOUT_BLOCK_VOLUME,
1269 	.name				= "LAYOUT_BLOCK_VOLUME",
1270 	.owner				= THIS_MODULE,
1271 	.read_pagelist			= bl_read_pagelist,
1272 	.write_pagelist			= bl_write_pagelist,
1273 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
1274 	.free_layout_hdr		= bl_free_layout_hdr,
1275 	.alloc_lseg			= bl_alloc_lseg,
1276 	.free_lseg			= bl_free_lseg,
1277 	.encode_layoutcommit		= bl_encode_layoutcommit,
1278 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
1279 	.set_layoutdriver		= bl_set_layoutdriver,
1280 	.clear_layoutdriver		= bl_clear_layoutdriver,
1281 	.pg_read_ops			= &bl_pg_read_ops,
1282 	.pg_write_ops			= &bl_pg_write_ops,
1283 };
1284 
1285 static const struct rpc_pipe_ops bl_upcall_ops = {
1286 	.upcall		= rpc_pipe_generic_upcall,
1287 	.downcall	= bl_pipe_downcall,
1288 	.destroy_msg	= bl_pipe_destroy_msg,
1289 };
1290 
1291 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1292 					    struct rpc_pipe *pipe)
1293 {
1294 	struct dentry *dir, *dentry;
1295 
1296 	dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1297 	if (dir == NULL)
1298 		return ERR_PTR(-ENOENT);
1299 	dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1300 	dput(dir);
1301 	return dentry;
1302 }
1303 
1304 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1305 					  struct rpc_pipe *pipe)
1306 {
1307 	if (pipe->dentry)
1308 		rpc_unlink(pipe->dentry);
1309 }
1310 
1311 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1312 			   void *ptr)
1313 {
1314 	struct super_block *sb = ptr;
1315 	struct net *net = sb->s_fs_info;
1316 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1317 	struct dentry *dentry;
1318 	int ret = 0;
1319 
1320 	if (!try_module_get(THIS_MODULE))
1321 		return 0;
1322 
1323 	if (nn->bl_device_pipe == NULL) {
1324 		module_put(THIS_MODULE);
1325 		return 0;
1326 	}
1327 
1328 	switch (event) {
1329 	case RPC_PIPEFS_MOUNT:
1330 		dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1331 		if (IS_ERR(dentry)) {
1332 			ret = PTR_ERR(dentry);
1333 			break;
1334 		}
1335 		nn->bl_device_pipe->dentry = dentry;
1336 		break;
1337 	case RPC_PIPEFS_UMOUNT:
1338 		if (nn->bl_device_pipe->dentry)
1339 			nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1340 		break;
1341 	default:
1342 		ret = -ENOTSUPP;
1343 		break;
1344 	}
1345 	module_put(THIS_MODULE);
1346 	return ret;
1347 }
1348 
1349 static struct notifier_block nfs4blocklayout_block = {
1350 	.notifier_call = rpc_pipefs_event,
1351 };
1352 
1353 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1354 						   struct rpc_pipe *pipe)
1355 {
1356 	struct super_block *pipefs_sb;
1357 	struct dentry *dentry;
1358 
1359 	pipefs_sb = rpc_get_sb_net(net);
1360 	if (!pipefs_sb)
1361 		return NULL;
1362 	dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1363 	rpc_put_sb_net(net);
1364 	return dentry;
1365 }
1366 
1367 static void nfs4blocklayout_unregister_net(struct net *net,
1368 					   struct rpc_pipe *pipe)
1369 {
1370 	struct super_block *pipefs_sb;
1371 
1372 	pipefs_sb = rpc_get_sb_net(net);
1373 	if (pipefs_sb) {
1374 		nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1375 		rpc_put_sb_net(net);
1376 	}
1377 }
1378 
1379 static int nfs4blocklayout_net_init(struct net *net)
1380 {
1381 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1382 	struct dentry *dentry;
1383 
1384 	init_waitqueue_head(&nn->bl_wq);
1385 	nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1386 	if (IS_ERR(nn->bl_device_pipe))
1387 		return PTR_ERR(nn->bl_device_pipe);
1388 	dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1389 	if (IS_ERR(dentry)) {
1390 		rpc_destroy_pipe_data(nn->bl_device_pipe);
1391 		return PTR_ERR(dentry);
1392 	}
1393 	nn->bl_device_pipe->dentry = dentry;
1394 	return 0;
1395 }
1396 
1397 static void nfs4blocklayout_net_exit(struct net *net)
1398 {
1399 	struct nfs_net *nn = net_generic(net, nfs_net_id);
1400 
1401 	nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1402 	rpc_destroy_pipe_data(nn->bl_device_pipe);
1403 	nn->bl_device_pipe = NULL;
1404 }
1405 
1406 static struct pernet_operations nfs4blocklayout_net_ops = {
1407 	.init = nfs4blocklayout_net_init,
1408 	.exit = nfs4blocklayout_net_exit,
1409 };
1410 
1411 static int __init nfs4blocklayout_init(void)
1412 {
1413 	int ret;
1414 
1415 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1416 
1417 	ret = pnfs_register_layoutdriver(&blocklayout_type);
1418 	if (ret)
1419 		goto out;
1420 
1421 	ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1422 	if (ret)
1423 		goto out_remove;
1424 	ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1425 	if (ret)
1426 		goto out_notifier;
1427 out:
1428 	return ret;
1429 
1430 out_notifier:
1431 	rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1432 out_remove:
1433 	pnfs_unregister_layoutdriver(&blocklayout_type);
1434 	return ret;
1435 }
1436 
1437 static void __exit nfs4blocklayout_exit(void)
1438 {
1439 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1440 	       __func__);
1441 
1442 	rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1443 	unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1444 	pnfs_unregister_layoutdriver(&blocklayout_type);
1445 }
1446 
1447 MODULE_ALIAS("nfs-layouttype4-3");
1448 
1449 module_init(nfs4blocklayout_init);
1450 module_exit(nfs4blocklayout_exit);
1451