xref: /linux/fs/nfs/blocklayout/blocklayout.c (revision 36d179fd6bea35698d53444b7bd3025fa3788266)
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 
40 #include "../pnfs.h"
41 #include "../nfs4session.h"
42 #include "../internal.h"
43 #include "blocklayout.h"
44 
45 #define NFSDBG_FACILITY	NFSDBG_PNFS_LD
46 
47 MODULE_LICENSE("GPL");
48 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
49 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
50 
51 static bool is_hole(struct pnfs_block_extent *be)
52 {
53 	switch (be->be_state) {
54 	case PNFS_BLOCK_NONE_DATA:
55 		return true;
56 	case PNFS_BLOCK_INVALID_DATA:
57 		return be->be_tag ? false : true;
58 	default:
59 		return false;
60 	}
61 }
62 
63 /* The data we are handed might be spread across several bios.  We need
64  * to track when the last one is finished.
65  */
66 struct parallel_io {
67 	struct kref refcnt;
68 	void (*pnfs_callback) (void *data);
69 	void *data;
70 };
71 
72 static inline struct parallel_io *alloc_parallel(void *data)
73 {
74 	struct parallel_io *rv;
75 
76 	rv = kmalloc_obj(*rv, GFP_NOFS);
77 	if (rv) {
78 		rv->data = data;
79 		kref_init(&rv->refcnt);
80 	}
81 	return rv;
82 }
83 
84 static inline void get_parallel(struct parallel_io *p)
85 {
86 	kref_get(&p->refcnt);
87 }
88 
89 static void destroy_parallel(struct kref *kref)
90 {
91 	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
92 
93 	dprintk("%s enter\n", __func__);
94 	p->pnfs_callback(p->data);
95 	kfree(p);
96 }
97 
98 static inline void put_parallel(struct parallel_io *p)
99 {
100 	kref_put(&p->refcnt, destroy_parallel);
101 }
102 
103 static struct bio *
104 bl_submit_bio(struct bio *bio)
105 {
106 	if (bio) {
107 		get_parallel(bio->bi_private);
108 		dprintk("%s submitting %s bio %u@%llu\n", __func__,
109 			bio_op(bio) == READ ? "read" : "write",
110 			bio->bi_iter.bi_size,
111 			(unsigned long long)bio->bi_iter.bi_sector);
112 		submit_bio(bio);
113 	}
114 	return NULL;
115 }
116 
117 static bool offset_in_map(u64 offset, struct pnfs_block_dev_map *map)
118 {
119 	return offset >= map->start && offset < map->start + map->len;
120 }
121 
122 static struct bio *
123 do_add_page_to_bio(struct bio *bio, int npg, enum req_op op, sector_t isect,
124 		struct page *page, struct pnfs_block_dev_map *map,
125 		struct pnfs_block_extent *be, bio_end_io_t end_io,
126 		struct parallel_io *par, unsigned int offset, int *len)
127 {
128 	struct pnfs_block_dev *dev =
129 		container_of(be->be_device, struct pnfs_block_dev, node);
130 	u64 disk_addr, end;
131 
132 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
133 		npg, (__force u32)op, (unsigned long long)isect, offset, *len);
134 
135 	/* translate to device offset */
136 	isect += be->be_v_offset;
137 	isect -= be->be_f_offset;
138 
139 	/* translate to physical disk offset */
140 	disk_addr = (u64)isect << SECTOR_SHIFT;
141 	if (!offset_in_map(disk_addr, map)) {
142 		if (!dev->map(dev, disk_addr, map) || !offset_in_map(disk_addr, map))
143 			return ERR_PTR(-EIO);
144 		bio = bl_submit_bio(bio);
145 	}
146 	disk_addr += map->disk_offset;
147 	disk_addr -= map->start;
148 
149 	/* limit length to what the device mapping allows */
150 	end = disk_addr + *len;
151 	if (end >= map->disk_offset + map->len)
152 		*len = map->disk_offset + map->len - disk_addr;
153 
154 retry:
155 	if (!bio) {
156 		bio = bio_alloc(map->bdev, bio_max_segs(npg), op, GFP_NOIO);
157 		bio->bi_iter.bi_sector = disk_addr >> SECTOR_SHIFT;
158 		bio->bi_end_io = end_io;
159 		bio->bi_private = par;
160 	}
161 	if (bio_add_page(bio, page, *len, offset) < *len) {
162 		bio = bl_submit_bio(bio);
163 		goto retry;
164 	}
165 	return bio;
166 }
167 
168 static void bl_mark_devices_unavailable(struct nfs_pgio_header *header, bool rw)
169 {
170 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
171 	size_t bytes_left = header->args.count;
172 	sector_t isect, extent_length = 0;
173 	struct pnfs_block_extent be;
174 
175 	isect = header->args.offset >> SECTOR_SHIFT;
176 	bytes_left += header->args.offset - (isect << SECTOR_SHIFT);
177 
178 	while (bytes_left > 0) {
179 		if (!ext_tree_lookup(bl, isect, &be, rw))
180 				return;
181 		extent_length = be.be_length - (isect - be.be_f_offset);
182 		nfs4_mark_deviceid_unavailable(be.be_device);
183 		isect += extent_length;
184 		if (bytes_left > extent_length << SECTOR_SHIFT)
185 			bytes_left -= extent_length << SECTOR_SHIFT;
186 		else
187 			bytes_left = 0;
188 	}
189 }
190 
191 static void bl_end_io_read(struct bio *bio)
192 {
193 	struct parallel_io *par = bio->bi_private;
194 
195 	if (bio->bi_status) {
196 		struct nfs_pgio_header *header = par->data;
197 
198 		if (!header->pnfs_error)
199 			header->pnfs_error = -EIO;
200 		pnfs_set_lo_fail(header->lseg);
201 		bl_mark_devices_unavailable(header, false);
202 	}
203 
204 	bio_put(bio);
205 	put_parallel(par);
206 }
207 
208 static void bl_read_cleanup(struct work_struct *work)
209 {
210 	struct rpc_task *task;
211 	struct nfs_pgio_header *hdr;
212 	dprintk("%s enter\n", __func__);
213 	task = container_of(work, struct rpc_task, u.tk_work);
214 	hdr = container_of(task, struct nfs_pgio_header, task);
215 	pnfs_ld_read_done(hdr);
216 }
217 
218 static void
219 bl_end_par_io_read(void *data)
220 {
221 	struct nfs_pgio_header *hdr = data;
222 
223 	hdr->task.tk_status = hdr->pnfs_error;
224 	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
225 	schedule_work(&hdr->task.u.tk_work);
226 }
227 
228 static enum pnfs_try_status
229 bl_read_pagelist(struct nfs_pgio_header *header)
230 {
231 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
232 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
233 	struct bio *bio = NULL;
234 	struct pnfs_block_extent be;
235 	sector_t isect, extent_length = 0;
236 	struct parallel_io *par;
237 	loff_t f_offset = header->args.offset;
238 	size_t bytes_left = header->args.count;
239 	unsigned int pg_offset = header->args.pgbase, pg_len;
240 	struct page **pages = header->args.pages;
241 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
242 	const bool is_dio = (header->dreq != NULL);
243 	struct blk_plug plug;
244 	int i;
245 
246 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
247 		header->page_array.npages, f_offset,
248 		(unsigned int)header->args.count);
249 
250 	par = alloc_parallel(header);
251 	if (!par)
252 		return PNFS_NOT_ATTEMPTED;
253 	par->pnfs_callback = bl_end_par_io_read;
254 
255 	blk_start_plug(&plug);
256 
257 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
258 	/* Code assumes extents are page-aligned */
259 	for (i = pg_index; i < header->page_array.npages; i++) {
260 		if (extent_length <= 0) {
261 			/* We've used up the previous extent */
262 			bio = bl_submit_bio(bio);
263 
264 			/* Get the next one */
265 			if (!ext_tree_lookup(bl, isect, &be, false)) {
266 				header->pnfs_error = -EIO;
267 				goto out;
268 			}
269 			extent_length = be.be_length - (isect - be.be_f_offset);
270 		}
271 
272 		if (is_dio) {
273 			if (pg_offset + bytes_left > PAGE_SIZE)
274 				pg_len = PAGE_SIZE - pg_offset;
275 			else
276 				pg_len = bytes_left;
277 		} else {
278 			BUG_ON(pg_offset != 0);
279 			pg_len = PAGE_SIZE;
280 		}
281 
282 		if (is_hole(&be)) {
283 			bio = bl_submit_bio(bio);
284 			/* Fill hole w/ zeroes w/o accessing device */
285 			dprintk("%s Zeroing page for hole\n", __func__);
286 			zero_user_segment(pages[i], pg_offset, pg_len);
287 
288 			/* invalidate map */
289 			map.start = NFS4_MAX_UINT64;
290 		} else {
291 			bio = do_add_page_to_bio(bio,
292 						 header->page_array.npages - i,
293 						 REQ_OP_READ,
294 						 isect, pages[i], &map, &be,
295 						 bl_end_io_read, par,
296 						 pg_offset, &pg_len);
297 			if (IS_ERR(bio)) {
298 				header->pnfs_error = PTR_ERR(bio);
299 				bio = NULL;
300 				goto out;
301 			}
302 		}
303 		isect += (pg_len >> SECTOR_SHIFT);
304 		extent_length -= (pg_len >> SECTOR_SHIFT);
305 		f_offset += pg_len;
306 		bytes_left -= pg_len;
307 		pg_offset = 0;
308 	}
309 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
310 		header->res.eof = 1;
311 		header->res.count = header->inode->i_size - header->args.offset;
312 	} else {
313 		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
314 	}
315 out:
316 	bl_submit_bio(bio);
317 	blk_finish_plug(&plug);
318 	put_parallel(par);
319 	return PNFS_ATTEMPTED;
320 }
321 
322 static void bl_end_io_write(struct bio *bio)
323 {
324 	struct parallel_io *par = bio->bi_private;
325 	struct nfs_pgio_header *header = par->data;
326 
327 	if (bio->bi_status) {
328 		if (!header->pnfs_error)
329 			header->pnfs_error = -EIO;
330 		pnfs_set_lo_fail(header->lseg);
331 		bl_mark_devices_unavailable(header, true);
332 	}
333 	bio_put(bio);
334 	put_parallel(par);
335 }
336 
337 /* Function scheduled for call during bl_end_par_io_write,
338  * it marks sectors as written and extends the commitlist.
339  */
340 static void bl_write_cleanup(struct work_struct *work)
341 {
342 	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
343 	struct nfs_pgio_header *hdr =
344 			container_of(task, struct nfs_pgio_header, task);
345 
346 	dprintk("%s enter\n", __func__);
347 
348 	if (likely(!hdr->pnfs_error)) {
349 		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
350 		u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
351 		u64 end = (hdr->args.offset + hdr->args.count +
352 			PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
353 		u64 lwb = hdr->args.offset + hdr->args.count;
354 
355 		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
356 					(end - start) >> SECTOR_SHIFT, lwb);
357 	}
358 
359 	pnfs_ld_write_done(hdr);
360 }
361 
362 /* Called when last of bios associated with a bl_write_pagelist call finishes */
363 static void bl_end_par_io_write(void *data)
364 {
365 	struct nfs_pgio_header *hdr = data;
366 
367 	hdr->task.tk_status = hdr->pnfs_error;
368 	hdr->verf.committed = NFS_FILE_SYNC;
369 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
370 	schedule_work(&hdr->task.u.tk_work);
371 }
372 
373 static enum pnfs_try_status
374 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
375 {
376 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
377 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
378 	struct bio *bio = NULL;
379 	struct pnfs_block_extent be;
380 	sector_t isect, extent_length = 0;
381 	struct parallel_io *par = NULL;
382 	loff_t offset = header->args.offset;
383 	struct page **pages = header->args.pages;
384 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
385 	unsigned int pg_len;
386 	struct blk_plug plug;
387 	int i;
388 
389 	dprintk("%s enter, %u@%lld\n", __func__, header->args.count, offset);
390 
391 	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
392 	 * We want to write each, and if there is an error set pnfs_error
393 	 * to have it redone using nfs.
394 	 */
395 	par = alloc_parallel(header);
396 	if (!par)
397 		return PNFS_NOT_ATTEMPTED;
398 	par->pnfs_callback = bl_end_par_io_write;
399 
400 	blk_start_plug(&plug);
401 
402 	/* we always write out the whole page */
403 	offset = offset & (loff_t)PAGE_MASK;
404 	isect = offset >> SECTOR_SHIFT;
405 
406 	for (i = pg_index; i < header->page_array.npages; i++) {
407 		if (extent_length <= 0) {
408 			/* We've used up the previous extent */
409 			bio = bl_submit_bio(bio);
410 			/* Get the next one */
411 			if (!ext_tree_lookup(bl, isect, &be, true)) {
412 				header->pnfs_error = -EINVAL;
413 				goto out;
414 			}
415 
416 			extent_length = be.be_length - (isect - be.be_f_offset);
417 		}
418 
419 		pg_len = PAGE_SIZE;
420 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
421 					 REQ_OP_WRITE, isect, pages[i], &map,
422 					 &be, bl_end_io_write, par, 0, &pg_len);
423 		if (IS_ERR(bio)) {
424 			header->pnfs_error = PTR_ERR(bio);
425 			bio = NULL;
426 			goto out;
427 		}
428 
429 		offset += pg_len;
430 		isect += (pg_len >> SECTOR_SHIFT);
431 		extent_length -= (pg_len >> SECTOR_SHIFT);
432 	}
433 
434 	header->res.count = header->args.count;
435 out:
436 	bl_submit_bio(bio);
437 	blk_finish_plug(&plug);
438 	put_parallel(par);
439 	return PNFS_ATTEMPTED;
440 }
441 
442 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
443 {
444 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
445 	int err;
446 
447 	dprintk("%s enter\n", __func__);
448 
449 	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
450 	WARN_ON(err);
451 
452 	kfree_rcu(bl, bl_layout.plh_rcu);
453 }
454 
455 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
456 		gfp_t gfp_flags, bool is_scsi_layout)
457 {
458 	struct pnfs_block_layout *bl;
459 
460 	dprintk("%s enter\n", __func__);
461 	bl = kzalloc_obj(*bl, gfp_flags);
462 	if (!bl)
463 		return NULL;
464 
465 	bl->bl_ext_rw = RB_ROOT;
466 	bl->bl_ext_ro = RB_ROOT;
467 	spin_lock_init(&bl->bl_ext_lock);
468 
469 	bl->bl_scsi_layout = is_scsi_layout;
470 	return &bl->bl_layout;
471 }
472 
473 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
474 						   gfp_t gfp_flags)
475 {
476 	return __bl_alloc_layout_hdr(inode, gfp_flags, false);
477 }
478 
479 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
480 						   gfp_t gfp_flags)
481 {
482 	return __bl_alloc_layout_hdr(inode, gfp_flags, true);
483 }
484 
485 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
486 {
487 	dprintk("%s enter\n", __func__);
488 	kfree(lseg);
489 }
490 
491 /* Tracks info needed to ensure extents in layout obey constraints of spec */
492 struct layout_verification {
493 	u32 mode;	/* R or RW */
494 	u64 start;	/* Expected start of next non-COW extent */
495 	u64 inval;	/* Start of INVAL coverage */
496 	u64 cowread;	/* End of COW read coverage */
497 };
498 
499 /* Verify the extent meets the layout requirements of the pnfs-block draft,
500  * section 2.3.1.
501  */
502 static int verify_extent(struct pnfs_block_extent *be,
503 			 struct layout_verification *lv)
504 {
505 	if (lv->mode == IOMODE_READ) {
506 		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
507 		    be->be_state == PNFS_BLOCK_INVALID_DATA)
508 			return -EIO;
509 		if (be->be_f_offset != lv->start)
510 			return -EIO;
511 		lv->start += be->be_length;
512 		return 0;
513 	}
514 	/* lv->mode == IOMODE_RW */
515 	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
516 		if (be->be_f_offset != lv->start)
517 			return -EIO;
518 		if (lv->cowread > lv->start)
519 			return -EIO;
520 		lv->start += be->be_length;
521 		lv->inval = lv->start;
522 		return 0;
523 	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
524 		if (be->be_f_offset != lv->start)
525 			return -EIO;
526 		lv->start += be->be_length;
527 		return 0;
528 	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
529 		if (be->be_f_offset > lv->start)
530 			return -EIO;
531 		if (be->be_f_offset < lv->inval)
532 			return -EIO;
533 		if (be->be_f_offset < lv->cowread)
534 			return -EIO;
535 		/* It looks like you might want to min this with lv->start,
536 		 * but you really don't.
537 		 */
538 		lv->inval = lv->inval + be->be_length;
539 		lv->cowread = be->be_f_offset + be->be_length;
540 		return 0;
541 	} else
542 		return -EIO;
543 }
544 
545 static int decode_sector_number(__be32 **rp, sector_t *sp)
546 {
547 	uint64_t s;
548 
549 	*rp = xdr_decode_hyper(*rp, &s);
550 	if (s & 0x1ff) {
551 		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
552 		return -1;
553 	}
554 	*sp = s >> SECTOR_SHIFT;
555 	return 0;
556 }
557 
558 static struct nfs4_deviceid_node *
559 bl_find_get_deviceid(struct nfs_server *server,
560 		const struct nfs4_deviceid *id, const struct cred *cred,
561 		gfp_t gfp_mask)
562 {
563 	struct nfs4_deviceid_node *node;
564 	int err = -ENODEV;
565 
566 retry:
567 	node = nfs4_find_get_deviceid(server, id, cred, gfp_mask);
568 	if (!node)
569 		return ERR_PTR(-ENODEV);
570 
571 	/*
572 	 * Devices that are marked unavailable are left in the cache with a
573 	 * timeout to avoid sending GETDEVINFO after every LAYOUTGET, or
574 	 * constantly attempting to register the device.  Once marked as
575 	 * unavailable they must be deleted and never reused.
576 	 */
577 	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags)) {
578 		unsigned long end = jiffies;
579 		unsigned long start = end - PNFS_DEVICE_RETRY_TIMEOUT;
580 
581 		if (!time_in_range(node->timestamp_unavailable, start, end)) {
582 			/* Uncork subsequent GETDEVINFO operations for this device */
583 			nfs4_delete_deviceid(node->ld, node->nfs_client, id);
584 			goto retry;
585 		}
586 		goto out_put;
587 	}
588 
589 	if (!bl_register_dev(container_of(node, struct pnfs_block_dev, node))) {
590 		/*
591 		 * If we cannot register, treat this device as transient:
592 		 * Make a negative cache entry for the device
593 		 */
594 		nfs4_mark_deviceid_unavailable(node);
595 		goto out_put;
596 	}
597 
598 	return node;
599 
600 out_put:
601 	nfs4_put_deviceid_node(node);
602 	return ERR_PTR(err);
603 }
604 
605 static int
606 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
607 		struct layout_verification *lv, struct list_head *extents,
608 		gfp_t gfp_mask)
609 {
610 	struct pnfs_block_extent *be;
611 	struct nfs4_deviceid id;
612 	int error;
613 	__be32 *p;
614 
615 	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
616 	if (!p)
617 		return -EIO;
618 
619 	be = kzalloc_obj(*be, GFP_NOFS);
620 	if (!be)
621 		return -ENOMEM;
622 
623 	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
624 	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
625 
626 	be->be_device = bl_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
627 						lo->plh_lc_cred, gfp_mask);
628 	if (IS_ERR(be->be_device)) {
629 		error = PTR_ERR(be->be_device);
630 		goto out_free_be;
631 	}
632 
633 	/*
634 	 * The next three values are read in as bytes, but stored in the
635 	 * extent structure in 512-byte granularity.
636 	 */
637 	error = -EIO;
638 	if (decode_sector_number(&p, &be->be_f_offset) < 0)
639 		goto out_put_deviceid;
640 	if (decode_sector_number(&p, &be->be_length) < 0)
641 		goto out_put_deviceid;
642 	if (decode_sector_number(&p, &be->be_v_offset) < 0)
643 		goto out_put_deviceid;
644 	be->be_state = be32_to_cpup(p++);
645 
646 	error = verify_extent(be, lv);
647 	if (error) {
648 		dprintk("%s: extent verification failed\n", __func__);
649 		goto out_put_deviceid;
650 	}
651 
652 	list_add_tail(&be->be_list, extents);
653 	return 0;
654 
655 out_put_deviceid:
656 	nfs4_put_deviceid_node(be->be_device);
657 out_free_be:
658 	kfree(be);
659 	return error;
660 }
661 
662 static struct pnfs_layout_segment *
663 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
664 		gfp_t gfp_mask)
665 {
666 	struct layout_verification lv = {
667 		.mode = lgr->range.iomode,
668 		.start = lgr->range.offset >> SECTOR_SHIFT,
669 		.inval = lgr->range.offset >> SECTOR_SHIFT,
670 		.cowread = lgr->range.offset >> SECTOR_SHIFT,
671 	};
672 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
673 	struct pnfs_layout_segment *lseg;
674 	struct xdr_buf buf;
675 	struct xdr_stream xdr;
676 	struct folio *scratch;
677 	int status, i;
678 	uint32_t count;
679 	__be32 *p;
680 	LIST_HEAD(extents);
681 
682 	dprintk("---> %s\n", __func__);
683 
684 	lseg = kzalloc_obj(*lseg, gfp_mask);
685 	if (!lseg)
686 		return ERR_PTR(-ENOMEM);
687 
688 	status = -ENOMEM;
689 	scratch = folio_alloc(gfp_mask, 0);
690 	if (!scratch)
691 		goto out;
692 
693 	xdr_init_decode_pages(&xdr, &buf,
694 			lgr->layoutp->pages, lgr->layoutp->len);
695 	xdr_set_scratch_folio(&xdr, scratch);
696 
697 	status = -EIO;
698 	p = xdr_inline_decode(&xdr, 4);
699 	if (unlikely(!p))
700 		goto out_free_scratch;
701 
702 	count = be32_to_cpup(p++);
703 	dprintk("%s: number of extents %d\n", __func__, count);
704 
705 	/*
706 	 * Decode individual extents, putting them in temporary staging area
707 	 * until whole layout is decoded to make error recovery easier.
708 	 */
709 	for (i = 0; i < count; i++) {
710 		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
711 		if (status)
712 			goto process_extents;
713 	}
714 
715 	if (lgr->range.offset + lgr->range.length !=
716 			lv.start << SECTOR_SHIFT) {
717 		dprintk("%s Final length mismatch\n", __func__);
718 		status = -EIO;
719 		goto process_extents;
720 	}
721 
722 	if (lv.start < lv.cowread) {
723 		dprintk("%s Final uncovered COW extent\n", __func__);
724 		status = -EIO;
725 	}
726 
727 process_extents:
728 	while (!list_empty(&extents)) {
729 		struct pnfs_block_extent *be =
730 			list_first_entry(&extents, struct pnfs_block_extent,
731 					 be_list);
732 		list_del(&be->be_list);
733 
734 		if (!status)
735 			status = ext_tree_insert(bl, be);
736 
737 		if (status) {
738 			nfs4_put_deviceid_node(be->be_device);
739 			kfree(be);
740 		}
741 	}
742 
743 out_free_scratch:
744 	folio_put(scratch);
745 out:
746 	dprintk("%s returns %d\n", __func__, status);
747 	switch (status) {
748 	case -ENODEV:
749 		/* Our extent block devices are unavailable */
750 		set_bit(NFS_LSEG_UNAVAILABLE, &lseg->pls_flags);
751 		fallthrough;
752 	case 0:
753 		return lseg;
754 	default:
755 		kfree(lseg);
756 		return ERR_PTR(status);
757 	}
758 }
759 
760 static void
761 bl_return_range(struct pnfs_layout_hdr *lo,
762 		struct pnfs_layout_range *range)
763 {
764 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
765 	sector_t offset = range->offset >> SECTOR_SHIFT, end;
766 
767 	if (range->offset % 8) {
768 		dprintk("%s: offset %lld not block size aligned\n",
769 			__func__, range->offset);
770 		return;
771 	}
772 
773 	if (range->length != NFS4_MAX_UINT64) {
774 		if (range->length % 8) {
775 			dprintk("%s: length %lld not block size aligned\n",
776 				__func__, range->length);
777 			return;
778 		}
779 
780 		end = offset + (range->length >> SECTOR_SHIFT);
781 	} else {
782 		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
783 	}
784 
785 	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
786 }
787 
788 static int
789 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
790 {
791 	return ext_tree_prepare_commit(arg);
792 }
793 
794 static void
795 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
796 {
797 	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
798 }
799 
800 static int
801 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
802 {
803 	dprintk("%s enter\n", __func__);
804 
805 	if (server->pnfs_blksize == 0) {
806 		dprintk("%s Server did not return blksize\n", __func__);
807 		return -EINVAL;
808 	}
809 	if (server->pnfs_blksize > PAGE_SIZE) {
810 		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
811 			__func__, server->pnfs_blksize);
812 		return -EINVAL;
813 	}
814 
815 	return 0;
816 }
817 
818 static bool
819 is_aligned_req(struct nfs_pageio_descriptor *pgio,
820 		struct nfs_page *req, unsigned int alignment, bool is_write)
821 {
822 	/*
823 	 * Always accept buffered writes, higher layers take care of the
824 	 * right alignment.
825 	 */
826 	if (pgio->pg_dreq == NULL)
827 		return true;
828 
829 	if (!IS_ALIGNED(req->wb_offset, alignment))
830 		return false;
831 
832 	if (IS_ALIGNED(req->wb_bytes, alignment))
833 		return true;
834 
835 	if (is_write &&
836 	    (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
837 		/*
838 		 * If the write goes up to the inode size, just write
839 		 * the full page.  Data past the inode size is
840 		 * guaranteed to be zeroed by the higher level client
841 		 * code, and this behaviour is mandated by RFC 5663
842 		 * section 2.3.2.
843 		 */
844 		return true;
845 	}
846 
847 	return false;
848 }
849 
850 static void
851 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
852 {
853 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
854 		nfs_pageio_reset_read_mds(pgio);
855 		return;
856 	}
857 
858 	pnfs_generic_pg_init_read(pgio, req);
859 
860 	if (pgio->pg_lseg &&
861 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
862 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
863 		pnfs_set_lo_fail(pgio->pg_lseg);
864 		nfs_pageio_reset_read_mds(pgio);
865 	}
866 }
867 
868 /*
869  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
870  * of bytes (maximum @req->wb_bytes) that can be coalesced.
871  */
872 static size_t
873 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
874 		struct nfs_page *req)
875 {
876 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
877 		return 0;
878 	return pnfs_generic_pg_test(pgio, prev, req);
879 }
880 
881 /*
882  * Return the number of contiguous bytes for a given inode
883  * starting at page frame idx.
884  */
885 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
886 {
887 	struct address_space *mapping = inode->i_mapping;
888 	pgoff_t end;
889 
890 	/* Optimize common case that writes from 0 to end of file */
891 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
892 	if (end != inode->i_mapping->nrpages) {
893 		rcu_read_lock();
894 		end = page_cache_next_miss(mapping, idx + 1, ULONG_MAX);
895 		rcu_read_unlock();
896 	}
897 
898 	if (!end)
899 		return i_size_read(inode) - (idx << PAGE_SHIFT);
900 	else
901 		return (end - idx) << PAGE_SHIFT;
902 }
903 
904 static void
905 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
906 {
907 	u64 wb_size;
908 
909 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
910 		nfs_pageio_reset_write_mds(pgio);
911 		return;
912 	}
913 
914 	if (pgio->pg_dreq == NULL)
915 		wb_size = pnfs_num_cont_bytes(pgio->pg_inode, req->wb_index);
916 	else
917 		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq, req_offset(req));
918 
919 	pnfs_generic_pg_init_write(pgio, req, wb_size);
920 
921 	if (pgio->pg_lseg &&
922 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
923 
924 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
925 		pnfs_set_lo_fail(pgio->pg_lseg);
926 		nfs_pageio_reset_write_mds(pgio);
927 	}
928 }
929 
930 /*
931  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
932  * of bytes (maximum @req->wb_bytes) that can be coalesced.
933  */
934 static size_t
935 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
936 		 struct nfs_page *req)
937 {
938 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
939 		return 0;
940 	return pnfs_generic_pg_test(pgio, prev, req);
941 }
942 
943 static const struct nfs_pageio_ops bl_pg_read_ops = {
944 	.pg_init = bl_pg_init_read,
945 	.pg_test = bl_pg_test_read,
946 	.pg_doio = pnfs_generic_pg_readpages,
947 	.pg_cleanup = pnfs_generic_pg_cleanup,
948 };
949 
950 static const struct nfs_pageio_ops bl_pg_write_ops = {
951 	.pg_init = bl_pg_init_write,
952 	.pg_test = bl_pg_test_write,
953 	.pg_doio = pnfs_generic_pg_writepages,
954 	.pg_cleanup = pnfs_generic_pg_cleanup,
955 };
956 
957 static struct pnfs_layoutdriver_type blocklayout_type = {
958 	.id				= LAYOUT_BLOCK_VOLUME,
959 	.name				= "LAYOUT_BLOCK_VOLUME",
960 	.owner				= THIS_MODULE,
961 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
962 					  PNFS_LAYOUTRET_ON_ERROR |
963 					  PNFS_READ_WHOLE_PAGE,
964 	.read_pagelist			= bl_read_pagelist,
965 	.write_pagelist			= bl_write_pagelist,
966 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
967 	.free_layout_hdr		= bl_free_layout_hdr,
968 	.alloc_lseg			= bl_alloc_lseg,
969 	.free_lseg			= bl_free_lseg,
970 	.return_range			= bl_return_range,
971 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
972 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
973 	.set_layoutdriver		= bl_set_layoutdriver,
974 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
975 	.free_deviceid_node		= bl_free_deviceid_node,
976 	.pg_read_ops			= &bl_pg_read_ops,
977 	.pg_write_ops			= &bl_pg_write_ops,
978 	.sync				= pnfs_generic_sync,
979 };
980 
981 static struct pnfs_layoutdriver_type scsilayout_type = {
982 	.id				= LAYOUT_SCSI,
983 	.name				= "LAYOUT_SCSI",
984 	.owner				= THIS_MODULE,
985 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
986 					  PNFS_LAYOUTRET_ON_ERROR |
987 					  PNFS_READ_WHOLE_PAGE,
988 	.read_pagelist			= bl_read_pagelist,
989 	.write_pagelist			= bl_write_pagelist,
990 	.alloc_layout_hdr		= sl_alloc_layout_hdr,
991 	.free_layout_hdr		= bl_free_layout_hdr,
992 	.alloc_lseg			= bl_alloc_lseg,
993 	.free_lseg			= bl_free_lseg,
994 	.return_range			= bl_return_range,
995 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
996 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
997 	.set_layoutdriver		= bl_set_layoutdriver,
998 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
999 	.free_deviceid_node		= bl_free_deviceid_node,
1000 	.pg_read_ops			= &bl_pg_read_ops,
1001 	.pg_write_ops			= &bl_pg_write_ops,
1002 	.sync				= pnfs_generic_sync,
1003 };
1004 
1005 
1006 static int __init nfs4blocklayout_init(void)
1007 {
1008 	int ret;
1009 
1010 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1011 
1012 	ret = bl_init_pipefs();
1013 	if (ret)
1014 		goto out;
1015 
1016 	ret = pnfs_register_layoutdriver(&blocklayout_type);
1017 	if (ret)
1018 		goto out_cleanup_pipe;
1019 
1020 	ret = pnfs_register_layoutdriver(&scsilayout_type);
1021 	if (ret)
1022 		goto out_unregister_block;
1023 	return 0;
1024 
1025 out_unregister_block:
1026 	pnfs_unregister_layoutdriver(&blocklayout_type);
1027 out_cleanup_pipe:
1028 	bl_cleanup_pipefs();
1029 out:
1030 	return ret;
1031 }
1032 
1033 static void __exit nfs4blocklayout_exit(void)
1034 {
1035 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1036 	       __func__);
1037 
1038 	pnfs_unregister_layoutdriver(&scsilayout_type);
1039 	pnfs_unregister_layoutdriver(&blocklayout_type);
1040 	bl_cleanup_pipefs();
1041 }
1042 
1043 MODULE_ALIAS("nfs-layouttype4-3");
1044 MODULE_ALIAS("nfs-layouttype4-5");
1045 
1046 module_init(nfs4blocklayout_init);
1047 module_exit(nfs4blocklayout_exit);
1048