1/*
2 *  linux/fs/nfs/blocklayout/blocklayout.c
3 *
4 *  Module for the NFSv4.1 pNFS block layout driver.
5 *
6 *  Copyright (c) 2006 The Regents of the University of Michigan.
7 *  All rights reserved.
8 *
9 *  Andy Adamson <andros@citi.umich.edu>
10 *  Fred Isaman <iisaman@umich.edu>
11 *
12 * permission is granted to use, copy, create derivative works and
13 * redistribute this software and such derivative works for any purpose,
14 * so long as the name of the university of michigan is not used in
15 * any advertising or publicity pertaining to the use or distribution
16 * of this software without specific, written prior authorization.  if
17 * the above copyright notice or any other identification of the
18 * university of michigan is included in any copy of any portion of
19 * this software, then the disclaimer below must also be included.
20 *
21 * this software is provided as is, without representation from the
22 * university of michigan as to its fitness for any purpose, and without
23 * warranty by the university of michigan of any kind, either express
24 * or implied, including without limitation the implied warranties of
25 * merchantability and fitness for a particular purpose.  the regents
26 * of the university of michigan shall not be liable for any damages,
27 * including special, indirect, incidental, or consequential damages,
28 * with respect to any claim arising out or in connection with the use
29 * of the software, even if it has been or is hereafter advised of the
30 * possibility of such damages.
31 */
32
33#include <linux/module.h>
34#include <linux/init.h>
35#include <linux/mount.h>
36#include <linux/namei.h>
37#include <linux/bio.h>		/* struct bio */
38#include <linux/prefetch.h>
39#include <linux/pagevec.h>
40
41#include "../pnfs.h"
42#include "../nfs4session.h"
43#include "../internal.h"
44#include "blocklayout.h"
45
46#define NFSDBG_FACILITY	NFSDBG_PNFS_LD
47
48MODULE_LICENSE("GPL");
49MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
50MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
51
52static bool is_hole(struct pnfs_block_extent *be)
53{
54	switch (be->be_state) {
55	case PNFS_BLOCK_NONE_DATA:
56		return true;
57	case PNFS_BLOCK_INVALID_DATA:
58		return be->be_tag ? false : true;
59	default:
60		return false;
61	}
62}
63
64/* The data we are handed might be spread across several bios.  We need
65 * to track when the last one is finished.
66 */
67struct parallel_io {
68	struct kref refcnt;
69	void (*pnfs_callback) (void *data);
70	void *data;
71};
72
73static inline struct parallel_io *alloc_parallel(void *data)
74{
75	struct parallel_io *rv;
76
77	rv  = kmalloc(sizeof(*rv), GFP_NOFS);
78	if (rv) {
79		rv->data = data;
80		kref_init(&rv->refcnt);
81	}
82	return rv;
83}
84
85static inline void get_parallel(struct parallel_io *p)
86{
87	kref_get(&p->refcnt);
88}
89
90static void destroy_parallel(struct kref *kref)
91{
92	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
93
94	dprintk("%s enter\n", __func__);
95	p->pnfs_callback(p->data);
96	kfree(p);
97}
98
99static inline void put_parallel(struct parallel_io *p)
100{
101	kref_put(&p->refcnt, destroy_parallel);
102}
103
104static struct bio *
105bl_submit_bio(int rw, struct bio *bio)
106{
107	if (bio) {
108		get_parallel(bio->bi_private);
109		dprintk("%s submitting %s bio %u@%llu\n", __func__,
110			rw == READ ? "read" : "write", bio->bi_iter.bi_size,
111			(unsigned long long)bio->bi_iter.bi_sector);
112		submit_bio(rw, bio);
113	}
114	return NULL;
115}
116
117static struct bio *
118bl_alloc_init_bio(int npg, struct block_device *bdev, sector_t disk_sector,
119		void (*end_io)(struct bio *, int err), struct parallel_io *par)
120{
121	struct bio *bio;
122
123	npg = min(npg, BIO_MAX_PAGES);
124	bio = bio_alloc(GFP_NOIO, npg);
125	if (!bio && (current->flags & PF_MEMALLOC)) {
126		while (!bio && (npg /= 2))
127			bio = bio_alloc(GFP_NOIO, npg);
128	}
129
130	if (bio) {
131		bio->bi_iter.bi_sector = disk_sector;
132		bio->bi_bdev = bdev;
133		bio->bi_end_io = end_io;
134		bio->bi_private = par;
135	}
136	return bio;
137}
138
139static struct bio *
140do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect,
141		struct page *page, struct pnfs_block_dev_map *map,
142		struct pnfs_block_extent *be,
143		void (*end_io)(struct bio *, int err),
144		struct parallel_io *par, unsigned int offset, int *len)
145{
146	struct pnfs_block_dev *dev =
147		container_of(be->be_device, struct pnfs_block_dev, node);
148	u64 disk_addr, end;
149
150	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
151		npg, rw, (unsigned long long)isect, offset, *len);
152
153	/* translate to device offset */
154	isect += be->be_v_offset;
155	isect -= be->be_f_offset;
156
157	/* translate to physical disk offset */
158	disk_addr = (u64)isect << SECTOR_SHIFT;
159	if (disk_addr < map->start || disk_addr >= map->start + map->len) {
160		if (!dev->map(dev, disk_addr, map))
161			return ERR_PTR(-EIO);
162		bio = bl_submit_bio(rw, bio);
163	}
164	disk_addr += map->disk_offset;
165	disk_addr -= map->start;
166
167	/* limit length to what the device mapping allows */
168	end = disk_addr + *len;
169	if (end >= map->start + map->len)
170		*len = map->start + map->len - disk_addr;
171
172retry:
173	if (!bio) {
174		bio = bl_alloc_init_bio(npg, map->bdev,
175				disk_addr >> SECTOR_SHIFT, end_io, par);
176		if (!bio)
177			return ERR_PTR(-ENOMEM);
178	}
179	if (bio_add_page(bio, page, *len, offset) < *len) {
180		bio = bl_submit_bio(rw, bio);
181		goto retry;
182	}
183	return bio;
184}
185
186static void bl_end_io_read(struct bio *bio, int err)
187{
188	struct parallel_io *par = bio->bi_private;
189
190	if (err) {
191		struct nfs_pgio_header *header = par->data;
192
193		if (!header->pnfs_error)
194			header->pnfs_error = -EIO;
195		pnfs_set_lo_fail(header->lseg);
196	}
197
198	bio_put(bio);
199	put_parallel(par);
200}
201
202static void bl_read_cleanup(struct work_struct *work)
203{
204	struct rpc_task *task;
205	struct nfs_pgio_header *hdr;
206	dprintk("%s enter\n", __func__);
207	task = container_of(work, struct rpc_task, u.tk_work);
208	hdr = container_of(task, struct nfs_pgio_header, task);
209	pnfs_ld_read_done(hdr);
210}
211
212static void
213bl_end_par_io_read(void *data)
214{
215	struct nfs_pgio_header *hdr = data;
216
217	hdr->task.tk_status = hdr->pnfs_error;
218	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
219	schedule_work(&hdr->task.u.tk_work);
220}
221
222static enum pnfs_try_status
223bl_read_pagelist(struct nfs_pgio_header *header)
224{
225	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
226	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
227	struct bio *bio = NULL;
228	struct pnfs_block_extent be;
229	sector_t isect, extent_length = 0;
230	struct parallel_io *par;
231	loff_t f_offset = header->args.offset;
232	size_t bytes_left = header->args.count;
233	unsigned int pg_offset, pg_len;
234	struct page **pages = header->args.pages;
235	int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
236	const bool is_dio = (header->dreq != NULL);
237	struct blk_plug plug;
238	int i;
239
240	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
241		header->page_array.npages, f_offset,
242		(unsigned int)header->args.count);
243
244	par = alloc_parallel(header);
245	if (!par)
246		return PNFS_NOT_ATTEMPTED;
247	par->pnfs_callback = bl_end_par_io_read;
248
249	blk_start_plug(&plug);
250
251	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
252	/* Code assumes extents are page-aligned */
253	for (i = pg_index; i < header->page_array.npages; i++) {
254		if (extent_length <= 0) {
255			/* We've used up the previous extent */
256			bio = bl_submit_bio(READ, bio);
257
258			/* Get the next one */
259			if (!ext_tree_lookup(bl, isect, &be, false)) {
260				header->pnfs_error = -EIO;
261				goto out;
262			}
263			extent_length = be.be_length - (isect - be.be_f_offset);
264		}
265
266		pg_offset = f_offset & ~PAGE_CACHE_MASK;
267		if (is_dio) {
268			if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
269				pg_len = PAGE_CACHE_SIZE - pg_offset;
270			else
271				pg_len = bytes_left;
272		} else {
273			BUG_ON(pg_offset != 0);
274			pg_len = PAGE_CACHE_SIZE;
275		}
276
277		isect += (pg_offset >> SECTOR_SHIFT);
278		extent_length -= (pg_offset >> SECTOR_SHIFT);
279
280		if (is_hole(&be)) {
281			bio = bl_submit_bio(READ, bio);
282			/* Fill hole w/ zeroes w/o accessing device */
283			dprintk("%s Zeroing page for hole\n", __func__);
284			zero_user_segment(pages[i], pg_offset, pg_len);
285
286			/* invalidate map */
287			map.start = NFS4_MAX_UINT64;
288		} else {
289			bio = do_add_page_to_bio(bio,
290						 header->page_array.npages - i,
291						 READ,
292						 isect, pages[i], &map, &be,
293						 bl_end_io_read, par,
294						 pg_offset, &pg_len);
295			if (IS_ERR(bio)) {
296				header->pnfs_error = PTR_ERR(bio);
297				bio = NULL;
298				goto out;
299			}
300		}
301		isect += (pg_len >> SECTOR_SHIFT);
302		extent_length -= (pg_len >> SECTOR_SHIFT);
303		f_offset += pg_len;
304		bytes_left -= pg_len;
305	}
306	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
307		header->res.eof = 1;
308		header->res.count = header->inode->i_size - header->args.offset;
309	} else {
310		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
311	}
312out:
313	bl_submit_bio(READ, bio);
314	blk_finish_plug(&plug);
315	put_parallel(par);
316	return PNFS_ATTEMPTED;
317}
318
319static void bl_end_io_write(struct bio *bio, int err)
320{
321	struct parallel_io *par = bio->bi_private;
322	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
323	struct nfs_pgio_header *header = par->data;
324
325	if (!uptodate) {
326		if (!header->pnfs_error)
327			header->pnfs_error = -EIO;
328		pnfs_set_lo_fail(header->lseg);
329	}
330	bio_put(bio);
331	put_parallel(par);
332}
333
334/* Function scheduled for call during bl_end_par_io_write,
335 * it marks sectors as written and extends the commitlist.
336 */
337static void bl_write_cleanup(struct work_struct *work)
338{
339	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
340	struct nfs_pgio_header *hdr =
341			container_of(task, struct nfs_pgio_header, task);
342
343	dprintk("%s enter\n", __func__);
344
345	if (likely(!hdr->pnfs_error)) {
346		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
347		u64 start = hdr->args.offset & (loff_t)PAGE_CACHE_MASK;
348		u64 end = (hdr->args.offset + hdr->args.count +
349			PAGE_CACHE_SIZE - 1) & (loff_t)PAGE_CACHE_MASK;
350
351		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
352					(end - start) >> SECTOR_SHIFT);
353	}
354
355	pnfs_ld_write_done(hdr);
356}
357
358/* Called when last of bios associated with a bl_write_pagelist call finishes */
359static void bl_end_par_io_write(void *data)
360{
361	struct nfs_pgio_header *hdr = data;
362
363	hdr->task.tk_status = hdr->pnfs_error;
364	hdr->verf.committed = NFS_FILE_SYNC;
365	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
366	schedule_work(&hdr->task.u.tk_work);
367}
368
369static enum pnfs_try_status
370bl_write_pagelist(struct nfs_pgio_header *header, int sync)
371{
372	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
373	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
374	struct bio *bio = NULL;
375	struct pnfs_block_extent be;
376	sector_t isect, extent_length = 0;
377	struct parallel_io *par = NULL;
378	loff_t offset = header->args.offset;
379	size_t count = header->args.count;
380	struct page **pages = header->args.pages;
381	int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
382	unsigned int pg_len;
383	struct blk_plug plug;
384	int i;
385
386	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
387
388	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
389	 * We want to write each, and if there is an error set pnfs_error
390	 * to have it redone using nfs.
391	 */
392	par = alloc_parallel(header);
393	if (!par)
394		return PNFS_NOT_ATTEMPTED;
395	par->pnfs_callback = bl_end_par_io_write;
396
397	blk_start_plug(&plug);
398
399	/* we always write out the whole page */
400	offset = offset & (loff_t)PAGE_CACHE_MASK;
401	isect = offset >> SECTOR_SHIFT;
402
403	for (i = pg_index; i < header->page_array.npages; i++) {
404		if (extent_length <= 0) {
405			/* We've used up the previous extent */
406			bio = bl_submit_bio(WRITE, bio);
407			/* Get the next one */
408			if (!ext_tree_lookup(bl, isect, &be, true)) {
409				header->pnfs_error = -EINVAL;
410				goto out;
411			}
412
413			extent_length = be.be_length - (isect - be.be_f_offset);
414		}
415
416		pg_len = PAGE_CACHE_SIZE;
417		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
418					 WRITE, isect, pages[i], &map, &be,
419					 bl_end_io_write, par,
420					 0, &pg_len);
421		if (IS_ERR(bio)) {
422			header->pnfs_error = PTR_ERR(bio);
423			bio = NULL;
424			goto out;
425		}
426
427		offset += pg_len;
428		count -= pg_len;
429		isect += (pg_len >> SECTOR_SHIFT);
430		extent_length -= (pg_len >> SECTOR_SHIFT);
431	}
432
433	header->res.count = header->args.count;
434out:
435	bl_submit_bio(WRITE, bio);
436	blk_finish_plug(&plug);
437	put_parallel(par);
438	return PNFS_ATTEMPTED;
439}
440
441static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
442{
443	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
444	int err;
445
446	dprintk("%s enter\n", __func__);
447
448	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
449	WARN_ON(err);
450
451	kfree(bl);
452}
453
454static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
455						   gfp_t gfp_flags)
456{
457	struct pnfs_block_layout *bl;
458
459	dprintk("%s enter\n", __func__);
460	bl = kzalloc(sizeof(*bl), gfp_flags);
461	if (!bl)
462		return NULL;
463
464	bl->bl_ext_rw = RB_ROOT;
465	bl->bl_ext_ro = RB_ROOT;
466	spin_lock_init(&bl->bl_ext_lock);
467
468	return &bl->bl_layout;
469}
470
471static void bl_free_lseg(struct pnfs_layout_segment *lseg)
472{
473	dprintk("%s enter\n", __func__);
474	kfree(lseg);
475}
476
477/* Tracks info needed to ensure extents in layout obey constraints of spec */
478struct layout_verification {
479	u32 mode;	/* R or RW */
480	u64 start;	/* Expected start of next non-COW extent */
481	u64 inval;	/* Start of INVAL coverage */
482	u64 cowread;	/* End of COW read coverage */
483};
484
485/* Verify the extent meets the layout requirements of the pnfs-block draft,
486 * section 2.3.1.
487 */
488static int verify_extent(struct pnfs_block_extent *be,
489			 struct layout_verification *lv)
490{
491	if (lv->mode == IOMODE_READ) {
492		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
493		    be->be_state == PNFS_BLOCK_INVALID_DATA)
494			return -EIO;
495		if (be->be_f_offset != lv->start)
496			return -EIO;
497		lv->start += be->be_length;
498		return 0;
499	}
500	/* lv->mode == IOMODE_RW */
501	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
502		if (be->be_f_offset != lv->start)
503			return -EIO;
504		if (lv->cowread > lv->start)
505			return -EIO;
506		lv->start += be->be_length;
507		lv->inval = lv->start;
508		return 0;
509	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
510		if (be->be_f_offset != lv->start)
511			return -EIO;
512		lv->start += be->be_length;
513		return 0;
514	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
515		if (be->be_f_offset > lv->start)
516			return -EIO;
517		if (be->be_f_offset < lv->inval)
518			return -EIO;
519		if (be->be_f_offset < lv->cowread)
520			return -EIO;
521		/* It looks like you might want to min this with lv->start,
522		 * but you really don't.
523		 */
524		lv->inval = lv->inval + be->be_length;
525		lv->cowread = be->be_f_offset + be->be_length;
526		return 0;
527	} else
528		return -EIO;
529}
530
531static int decode_sector_number(__be32 **rp, sector_t *sp)
532{
533	uint64_t s;
534
535	*rp = xdr_decode_hyper(*rp, &s);
536	if (s & 0x1ff) {
537		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
538		return -1;
539	}
540	*sp = s >> SECTOR_SHIFT;
541	return 0;
542}
543
544static int
545bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
546		struct layout_verification *lv, struct list_head *extents,
547		gfp_t gfp_mask)
548{
549	struct pnfs_block_extent *be;
550	struct nfs4_deviceid id;
551	int error;
552	__be32 *p;
553
554	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
555	if (!p)
556		return -EIO;
557
558	be = kzalloc(sizeof(*be), GFP_NOFS);
559	if (!be)
560		return -ENOMEM;
561
562	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
563	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
564
565	error = -EIO;
566	be->be_device = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
567						lo->plh_lc_cred, gfp_mask);
568	if (!be->be_device)
569		goto out_free_be;
570
571	/*
572	 * The next three values are read in as bytes, but stored in the
573	 * extent structure in 512-byte granularity.
574	 */
575	if (decode_sector_number(&p, &be->be_f_offset) < 0)
576		goto out_put_deviceid;
577	if (decode_sector_number(&p, &be->be_length) < 0)
578		goto out_put_deviceid;
579	if (decode_sector_number(&p, &be->be_v_offset) < 0)
580		goto out_put_deviceid;
581	be->be_state = be32_to_cpup(p++);
582
583	error = verify_extent(be, lv);
584	if (error) {
585		dprintk("%s: extent verification failed\n", __func__);
586		goto out_put_deviceid;
587	}
588
589	list_add_tail(&be->be_list, extents);
590	return 0;
591
592out_put_deviceid:
593	nfs4_put_deviceid_node(be->be_device);
594out_free_be:
595	kfree(be);
596	return error;
597}
598
599static struct pnfs_layout_segment *
600bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
601		gfp_t gfp_mask)
602{
603	struct layout_verification lv = {
604		.mode = lgr->range.iomode,
605		.start = lgr->range.offset >> SECTOR_SHIFT,
606		.inval = lgr->range.offset >> SECTOR_SHIFT,
607		.cowread = lgr->range.offset >> SECTOR_SHIFT,
608	};
609	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
610	struct pnfs_layout_segment *lseg;
611	struct xdr_buf buf;
612	struct xdr_stream xdr;
613	struct page *scratch;
614	int status, i;
615	uint32_t count;
616	__be32 *p;
617	LIST_HEAD(extents);
618
619	dprintk("---> %s\n", __func__);
620
621	lseg = kzalloc(sizeof(*lseg), gfp_mask);
622	if (!lseg)
623		return ERR_PTR(-ENOMEM);
624
625	status = -ENOMEM;
626	scratch = alloc_page(gfp_mask);
627	if (!scratch)
628		goto out;
629
630	xdr_init_decode_pages(&xdr, &buf,
631			lgr->layoutp->pages, lgr->layoutp->len);
632	xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
633
634	status = -EIO;
635	p = xdr_inline_decode(&xdr, 4);
636	if (unlikely(!p))
637		goto out_free_scratch;
638
639	count = be32_to_cpup(p++);
640	dprintk("%s: number of extents %d\n", __func__, count);
641
642	/*
643	 * Decode individual extents, putting them in temporary staging area
644	 * until whole layout is decoded to make error recovery easier.
645	 */
646	for (i = 0; i < count; i++) {
647		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
648		if (status)
649			goto process_extents;
650	}
651
652	if (lgr->range.offset + lgr->range.length !=
653			lv.start << SECTOR_SHIFT) {
654		dprintk("%s Final length mismatch\n", __func__);
655		status = -EIO;
656		goto process_extents;
657	}
658
659	if (lv.start < lv.cowread) {
660		dprintk("%s Final uncovered COW extent\n", __func__);
661		status = -EIO;
662	}
663
664process_extents:
665	while (!list_empty(&extents)) {
666		struct pnfs_block_extent *be =
667			list_first_entry(&extents, struct pnfs_block_extent,
668					 be_list);
669		list_del(&be->be_list);
670
671		if (!status)
672			status = ext_tree_insert(bl, be);
673
674		if (status) {
675			nfs4_put_deviceid_node(be->be_device);
676			kfree(be);
677		}
678	}
679
680out_free_scratch:
681	__free_page(scratch);
682out:
683	dprintk("%s returns %d\n", __func__, status);
684	if (status) {
685		kfree(lseg);
686		return ERR_PTR(status);
687	}
688	return lseg;
689}
690
691static void
692bl_return_range(struct pnfs_layout_hdr *lo,
693		struct pnfs_layout_range *range)
694{
695	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
696	sector_t offset = range->offset >> SECTOR_SHIFT, end;
697
698	if (range->offset % 8) {
699		dprintk("%s: offset %lld not block size aligned\n",
700			__func__, range->offset);
701		return;
702	}
703
704	if (range->length != NFS4_MAX_UINT64) {
705		if (range->length % 8) {
706			dprintk("%s: length %lld not block size aligned\n",
707				__func__, range->length);
708			return;
709		}
710
711		end = offset + (range->length >> SECTOR_SHIFT);
712	} else {
713		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
714	}
715
716	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
717}
718
719static int
720bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
721{
722	return ext_tree_prepare_commit(arg);
723}
724
725static void
726bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
727{
728	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
729}
730
731static int
732bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
733{
734	dprintk("%s enter\n", __func__);
735
736	if (server->pnfs_blksize == 0) {
737		dprintk("%s Server did not return blksize\n", __func__);
738		return -EINVAL;
739	}
740	if (server->pnfs_blksize > PAGE_SIZE) {
741		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
742			__func__, server->pnfs_blksize);
743		return -EINVAL;
744	}
745
746	return 0;
747}
748
749static bool
750is_aligned_req(struct nfs_pageio_descriptor *pgio,
751		struct nfs_page *req, unsigned int alignment)
752{
753	/*
754	 * Always accept buffered writes, higher layers take care of the
755	 * right alignment.
756	 */
757	if (pgio->pg_dreq == NULL)
758		return true;
759
760	if (!IS_ALIGNED(req->wb_offset, alignment))
761		return false;
762
763	if (IS_ALIGNED(req->wb_bytes, alignment))
764		return true;
765
766	if (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode)) {
767		/*
768		 * If the write goes up to the inode size, just write
769		 * the full page.  Data past the inode size is
770		 * guaranteed to be zeroed by the higher level client
771		 * code, and this behaviour is mandated by RFC 5663
772		 * section 2.3.2.
773		 */
774		return true;
775	}
776
777	return false;
778}
779
780static void
781bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
782{
783	if (!is_aligned_req(pgio, req, SECTOR_SIZE)) {
784		nfs_pageio_reset_read_mds(pgio);
785		return;
786	}
787
788	pnfs_generic_pg_init_read(pgio, req);
789}
790
791/*
792 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
793 * of bytes (maximum @req->wb_bytes) that can be coalesced.
794 */
795static size_t
796bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
797		struct nfs_page *req)
798{
799	if (!is_aligned_req(pgio, req, SECTOR_SIZE))
800		return 0;
801	return pnfs_generic_pg_test(pgio, prev, req);
802}
803
804/*
805 * Return the number of contiguous bytes for a given inode
806 * starting at page frame idx.
807 */
808static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
809{
810	struct address_space *mapping = inode->i_mapping;
811	pgoff_t end;
812
813	/* Optimize common case that writes from 0 to end of file */
814	end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
815	if (end != inode->i_mapping->nrpages) {
816		rcu_read_lock();
817		end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
818		rcu_read_unlock();
819	}
820
821	if (!end)
822		return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
823	else
824		return (end - idx) << PAGE_CACHE_SHIFT;
825}
826
827static void
828bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
829{
830	u64 wb_size;
831
832	if (!is_aligned_req(pgio, req, PAGE_SIZE)) {
833		nfs_pageio_reset_write_mds(pgio);
834		return;
835	}
836
837	if (pgio->pg_dreq == NULL)
838		wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
839					      req->wb_index);
840	else
841		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
842
843	pnfs_generic_pg_init_write(pgio, req, wb_size);
844}
845
846/*
847 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
848 * of bytes (maximum @req->wb_bytes) that can be coalesced.
849 */
850static size_t
851bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
852		 struct nfs_page *req)
853{
854	if (!is_aligned_req(pgio, req, PAGE_SIZE))
855		return 0;
856	return pnfs_generic_pg_test(pgio, prev, req);
857}
858
859static const struct nfs_pageio_ops bl_pg_read_ops = {
860	.pg_init = bl_pg_init_read,
861	.pg_test = bl_pg_test_read,
862	.pg_doio = pnfs_generic_pg_readpages,
863	.pg_cleanup = pnfs_generic_pg_cleanup,
864};
865
866static const struct nfs_pageio_ops bl_pg_write_ops = {
867	.pg_init = bl_pg_init_write,
868	.pg_test = bl_pg_test_write,
869	.pg_doio = pnfs_generic_pg_writepages,
870	.pg_cleanup = pnfs_generic_pg_cleanup,
871};
872
873static struct pnfs_layoutdriver_type blocklayout_type = {
874	.id				= LAYOUT_BLOCK_VOLUME,
875	.name				= "LAYOUT_BLOCK_VOLUME",
876	.owner				= THIS_MODULE,
877	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
878					  PNFS_READ_WHOLE_PAGE,
879	.read_pagelist			= bl_read_pagelist,
880	.write_pagelist			= bl_write_pagelist,
881	.alloc_layout_hdr		= bl_alloc_layout_hdr,
882	.free_layout_hdr		= bl_free_layout_hdr,
883	.alloc_lseg			= bl_alloc_lseg,
884	.free_lseg			= bl_free_lseg,
885	.return_range			= bl_return_range,
886	.prepare_layoutcommit		= bl_prepare_layoutcommit,
887	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
888	.set_layoutdriver		= bl_set_layoutdriver,
889	.alloc_deviceid_node		= bl_alloc_deviceid_node,
890	.free_deviceid_node		= bl_free_deviceid_node,
891	.pg_read_ops			= &bl_pg_read_ops,
892	.pg_write_ops			= &bl_pg_write_ops,
893	.sync				= pnfs_generic_sync,
894};
895
896static int __init nfs4blocklayout_init(void)
897{
898	int ret;
899
900	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
901
902	ret = pnfs_register_layoutdriver(&blocklayout_type);
903	if (ret)
904		goto out;
905	ret = bl_init_pipefs();
906	if (ret)
907		goto out_unregister;
908	return 0;
909
910out_unregister:
911	pnfs_unregister_layoutdriver(&blocklayout_type);
912out:
913	return ret;
914}
915
916static void __exit nfs4blocklayout_exit(void)
917{
918	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
919	       __func__);
920
921	bl_cleanup_pipefs();
922	pnfs_unregister_layoutdriver(&blocklayout_type);
923}
924
925MODULE_ALIAS("nfs-layouttype4-3");
926
927module_init(nfs4blocklayout_init);
928module_exit(nfs4blocklayout_exit);
929