blocklayout.c

   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(struct bio *bio)
 106{
 107	if (bio) {
 108		get_parallel(bio->bi_private);
 109		dprintk("%s submitting %s bio %u@%llu\n", __func__,
 110			bio_op(bio) == READ ? "read" : "write",
 111			bio->bi_iter.bi_size,
 112			(unsigned long long)bio->bi_iter.bi_sector);
 113		submit_bio(bio);
 114	}
 115	return NULL;
 116}
 117
 118static bool offset_in_map(u64 offset, struct pnfs_block_dev_map *map)
 119{
 120	return offset >= map->start && offset < map->start + map->len;
 121}
 122
 123static struct bio *
 124do_add_page_to_bio(struct bio *bio, int npg, enum req_op op, sector_t isect,
 125		struct page *page, struct pnfs_block_dev_map *map,
 126		struct pnfs_block_extent *be, bio_end_io_t end_io,
 127		struct parallel_io *par, unsigned int offset, int *len)
 128{
 129	struct pnfs_block_dev *dev =
 130		container_of(be->be_device, struct pnfs_block_dev, node);
 131	u64 disk_addr, end;
 132
 133	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
 134		npg, (__force u32)op, (unsigned long long)isect, offset, *len);
 135
 136	/* translate to device offset */
 137	isect += be->be_v_offset;
 138	isect -= be->be_f_offset;
 139
 140	/* translate to physical disk offset */
 141	disk_addr = (u64)isect << SECTOR_SHIFT;
 142	if (!offset_in_map(disk_addr, map)) {
 143		if (!dev->map(dev, disk_addr, map) || !offset_in_map(disk_addr, map))
 144			return ERR_PTR(-EIO);
 145		bio = bl_submit_bio(bio);
 146	}
 147	disk_addr += map->disk_offset;
 148	disk_addr -= map->start;
 149
 150	/* limit length to what the device mapping allows */
 151	end = disk_addr + *len;
 152	if (end >= map->disk_offset + map->len)
 153		*len = map->disk_offset + map->len - disk_addr;
 154
 155retry:
 156	if (!bio) {
 157		bio = bio_alloc(map->bdev, bio_max_segs(npg), op, GFP_NOIO);
 158		bio->bi_iter.bi_sector = disk_addr >> SECTOR_SHIFT;
 159		bio->bi_end_io = end_io;
 160		bio->bi_private = par;
 161	}
 162	if (bio_add_page(bio, page, *len, offset) < *len) {
 163		bio = bl_submit_bio(bio);
 164		goto retry;
 165	}
 166	return bio;
 167}
 168
 169static void bl_mark_devices_unavailable(struct nfs_pgio_header *header, bool rw)
 170{
 171	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
 172	size_t bytes_left = header->args.count;
 173	sector_t isect, extent_length = 0;
 174	struct pnfs_block_extent be;
 175
 176	isect = header->args.offset >> SECTOR_SHIFT;
 177	bytes_left += header->args.offset - (isect << SECTOR_SHIFT);
 178
 179	while (bytes_left > 0) {
 180		if (!ext_tree_lookup(bl, isect, &be, rw))
 181				return;
 182		extent_length = be.be_length - (isect - be.be_f_offset);
 183		nfs4_mark_deviceid_unavailable(be.be_device);
 184		isect += extent_length;
 185		if (bytes_left > extent_length << SECTOR_SHIFT)
 186			bytes_left -= extent_length << SECTOR_SHIFT;
 187		else
 188			bytes_left = 0;
 189	}
 190}
 191
 192static void bl_end_io_read(struct bio *bio)
 193{
 194	struct parallel_io *par = bio->bi_private;
 195
 196	if (bio->bi_status) {
 197		struct nfs_pgio_header *header = par->data;
 198
 199		if (!header->pnfs_error)
 200			header->pnfs_error = -EIO;
 201		pnfs_set_lo_fail(header->lseg);
 202		bl_mark_devices_unavailable(header, false);
 203	}
 204
 205	bio_put(bio);
 206	put_parallel(par);
 207}
 208
 209static void bl_read_cleanup(struct work_struct *work)
 210{
 211	struct rpc_task *task;
 212	struct nfs_pgio_header *hdr;
 213	dprintk("%s enter\n", __func__);
 214	task = container_of(work, struct rpc_task, u.tk_work);
 215	hdr = container_of(task, struct nfs_pgio_header, task);
 216	pnfs_ld_read_done(hdr);
 217}
 218
 219static void
 220bl_end_par_io_read(void *data)
 221{
 222	struct nfs_pgio_header *hdr = data;
 223
 224	hdr->task.tk_status = hdr->pnfs_error;
 225	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
 226	schedule_work(&hdr->task.u.tk_work);
 227}
 228
 229static enum pnfs_try_status
 230bl_read_pagelist(struct nfs_pgio_header *header)
 231{
 232	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
 233	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
 234	struct bio *bio = NULL;
 235	struct pnfs_block_extent be;
 236	sector_t isect, extent_length = 0;
 237	struct parallel_io *par;
 238	loff_t f_offset = header->args.offset;
 239	size_t bytes_left = header->args.count;
 240	unsigned int pg_offset = header->args.pgbase, pg_len;
 241	struct page **pages = header->args.pages;
 242	int pg_index = header->args.pgbase >> PAGE_SHIFT;
 243	const bool is_dio = (header->dreq != NULL);
 244	struct blk_plug plug;
 245	int i;
 246
 247	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
 248		header->page_array.npages, f_offset,
 249		(unsigned int)header->args.count);
 250
 251	par = alloc_parallel(header);
 252	if (!par)
 253		return PNFS_NOT_ATTEMPTED;
 254	par->pnfs_callback = bl_end_par_io_read;
 255
 256	blk_start_plug(&plug);
 257
 258	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
 259	/* Code assumes extents are page-aligned */
 260	for (i = pg_index; i < header->page_array.npages; i++) {
 261		if (extent_length <= 0) {
 262			/* We've used up the previous extent */
 263			bio = bl_submit_bio(bio);
 264
 265			/* Get the next one */
 266			if (!ext_tree_lookup(bl, isect, &be, false)) {
 267				header->pnfs_error = -EIO;
 268				goto out;
 269			}
 270			extent_length = be.be_length - (isect - be.be_f_offset);
 271		}
 272
 273		if (is_dio) {
 274			if (pg_offset + bytes_left > PAGE_SIZE)
 275				pg_len = PAGE_SIZE - pg_offset;
 276			else
 277				pg_len = bytes_left;
 278		} else {
 279			BUG_ON(pg_offset != 0);
 280			pg_len = PAGE_SIZE;
 281		}
 282
 283		if (is_hole(&be)) {
 284			bio = bl_submit_bio(bio);
 285			/* Fill hole w/ zeroes w/o accessing device */
 286			dprintk("%s Zeroing page for hole\n", __func__);
 287			zero_user_segment(pages[i], pg_offset, pg_len);
 288
 289			/* invalidate map */
 290			map.start = NFS4_MAX_UINT64;
 291		} else {
 292			bio = do_add_page_to_bio(bio,
 293						 header->page_array.npages - i,
 294						 REQ_OP_READ,
 295						 isect, pages[i], &map, &be,
 296						 bl_end_io_read, par,
 297						 pg_offset, &pg_len);
 298			if (IS_ERR(bio)) {
 299				header->pnfs_error = PTR_ERR(bio);
 300				bio = NULL;
 301				goto out;
 302			}
 303		}
 304		isect += (pg_len >> SECTOR_SHIFT);
 305		extent_length -= (pg_len >> SECTOR_SHIFT);
 306		f_offset += pg_len;
 307		bytes_left -= pg_len;
 308		pg_offset = 0;
 309	}
 310	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
 311		header->res.eof = 1;
 312		header->res.count = header->inode->i_size - header->args.offset;
 313	} else {
 314		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
 315	}
 316out:
 317	bl_submit_bio(bio);
 318	blk_finish_plug(&plug);
 319	put_parallel(par);
 320	return PNFS_ATTEMPTED;
 321}
 322
 323static void bl_end_io_write(struct bio *bio)
 324{
 325	struct parallel_io *par = bio->bi_private;
 326	struct nfs_pgio_header *header = par->data;
 327
 328	if (bio->bi_status) {
 329		if (!header->pnfs_error)
 330			header->pnfs_error = -EIO;
 331		pnfs_set_lo_fail(header->lseg);
 332		bl_mark_devices_unavailable(header, true);
 333	}
 334	bio_put(bio);
 335	put_parallel(par);
 336}
 337
 338/* Function scheduled for call during bl_end_par_io_write,
 339 * it marks sectors as written and extends the commitlist.
 340 */
 341static void bl_write_cleanup(struct work_struct *work)
 342{
 343	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
 344	struct nfs_pgio_header *hdr =
 345			container_of(task, struct nfs_pgio_header, task);
 346
 347	dprintk("%s enter\n", __func__);
 348
 349	if (likely(!hdr->pnfs_error)) {
 350		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
 351		u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
 352		u64 end = (hdr->args.offset + hdr->args.count +
 353			PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
 354		u64 lwb = hdr->args.offset + hdr->args.count;
 355
 356		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
 357					(end - start) >> SECTOR_SHIFT, lwb);
 358	}
 359
 360	pnfs_ld_write_done(hdr);
 361}
 362
 363/* Called when last of bios associated with a bl_write_pagelist call finishes */
 364static void bl_end_par_io_write(void *data)
 365{
 366	struct nfs_pgio_header *hdr = data;
 367
 368	hdr->task.tk_status = hdr->pnfs_error;
 369	hdr->verf.committed = NFS_FILE_SYNC;
 370	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
 371	schedule_work(&hdr->task.u.tk_work);
 372}
 373
 374static enum pnfs_try_status
 375bl_write_pagelist(struct nfs_pgio_header *header, int sync)
 376{
 377	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
 378	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
 379	struct bio *bio = NULL;
 380	struct pnfs_block_extent be;
 381	sector_t isect, extent_length = 0;
 382	struct parallel_io *par = NULL;
 383	loff_t offset = header->args.offset;
 384	size_t count = header->args.count;
 385	struct page **pages = header->args.pages;
 386	int pg_index = header->args.pgbase >> PAGE_SHIFT;
 387	unsigned int pg_len;
 388	struct blk_plug plug;
 389	int i;
 390
 391	dprintk("%s enter, %zu@%lld\n", __func__, count, offset);
 392
 393	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
 394	 * We want to write each, and if there is an error set pnfs_error
 395	 * to have it redone using nfs.
 396	 */
 397	par = alloc_parallel(header);
 398	if (!par)
 399		return PNFS_NOT_ATTEMPTED;
 400	par->pnfs_callback = bl_end_par_io_write;
 401
 402	blk_start_plug(&plug);
 403
 404	/* we always write out the whole page */
 405	offset = offset & (loff_t)PAGE_MASK;
 406	isect = offset >> SECTOR_SHIFT;
 407
 408	for (i = pg_index; i < header->page_array.npages; i++) {
 409		if (extent_length <= 0) {
 410			/* We've used up the previous extent */
 411			bio = bl_submit_bio(bio);
 412			/* Get the next one */
 413			if (!ext_tree_lookup(bl, isect, &be, true)) {
 414				header->pnfs_error = -EINVAL;
 415				goto out;
 416			}
 417
 418			extent_length = be.be_length - (isect - be.be_f_offset);
 419		}
 420
 421		pg_len = PAGE_SIZE;
 422		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
 423					 REQ_OP_WRITE, isect, pages[i], &map,
 424					 &be, bl_end_io_write, par, 0, &pg_len);
 425		if (IS_ERR(bio)) {
 426			header->pnfs_error = PTR_ERR(bio);
 427			bio = NULL;
 428			goto out;
 429		}
 430
 431		offset += pg_len;
 432		count -= pg_len;
 433		isect += (pg_len >> SECTOR_SHIFT);
 434		extent_length -= (pg_len >> SECTOR_SHIFT);
 435	}
 436
 437	header->res.count = header->args.count;
 438out:
 439	bl_submit_bio(bio);
 440	blk_finish_plug(&plug);
 441	put_parallel(par);
 442	return PNFS_ATTEMPTED;
 443}
 444
 445static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
 446{
 447	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
 448	int err;
 449
 450	dprintk("%s enter\n", __func__);
 451
 452	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
 453	WARN_ON(err);
 454
 455	kfree_rcu(bl, bl_layout.plh_rcu);
 456}
 457
 458static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
 459		gfp_t gfp_flags, bool is_scsi_layout)
 460{
 461	struct pnfs_block_layout *bl;
 462
 463	dprintk("%s enter\n", __func__);
 464	bl = kzalloc(sizeof(*bl), gfp_flags);
 465	if (!bl)
 466		return NULL;
 467
 468	bl->bl_ext_rw = RB_ROOT;
 469	bl->bl_ext_ro = RB_ROOT;
 470	spin_lock_init(&bl->bl_ext_lock);
 471
 472	bl->bl_scsi_layout = is_scsi_layout;
 473	return &bl->bl_layout;
 474}
 475
 476static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
 477						   gfp_t gfp_flags)
 478{
 479	return __bl_alloc_layout_hdr(inode, gfp_flags, false);
 480}
 481
 482static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
 483						   gfp_t gfp_flags)
 484{
 485	return __bl_alloc_layout_hdr(inode, gfp_flags, true);
 486}
 487
 488static void bl_free_lseg(struct pnfs_layout_segment *lseg)
 489{
 490	dprintk("%s enter\n", __func__);
 491	kfree(lseg);
 492}
 493
 494/* Tracks info needed to ensure extents in layout obey constraints of spec */
 495struct layout_verification {
 496	u32 mode;	/* R or RW */
 497	u64 start;	/* Expected start of next non-COW extent */
 498	u64 inval;	/* Start of INVAL coverage */
 499	u64 cowread;	/* End of COW read coverage */
 500};
 501
 502/* Verify the extent meets the layout requirements of the pnfs-block draft,
 503 * section 2.3.1.
 504 */
 505static int verify_extent(struct pnfs_block_extent *be,
 506			 struct layout_verification *lv)
 507{
 508	if (lv->mode == IOMODE_READ) {
 509		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
 510		    be->be_state == PNFS_BLOCK_INVALID_DATA)
 511			return -EIO;
 512		if (be->be_f_offset != lv->start)
 513			return -EIO;
 514		lv->start += be->be_length;
 515		return 0;
 516	}
 517	/* lv->mode == IOMODE_RW */
 518	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
 519		if (be->be_f_offset != lv->start)
 520			return -EIO;
 521		if (lv->cowread > lv->start)
 522			return -EIO;
 523		lv->start += be->be_length;
 524		lv->inval = lv->start;
 525		return 0;
 526	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 527		if (be->be_f_offset != lv->start)
 528			return -EIO;
 529		lv->start += be->be_length;
 530		return 0;
 531	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
 532		if (be->be_f_offset > lv->start)
 533			return -EIO;
 534		if (be->be_f_offset < lv->inval)
 535			return -EIO;
 536		if (be->be_f_offset < lv->cowread)
 537			return -EIO;
 538		/* It looks like you might want to min this with lv->start,
 539		 * but you really don't.
 540		 */
 541		lv->inval = lv->inval + be->be_length;
 542		lv->cowread = be->be_f_offset + be->be_length;
 543		return 0;
 544	} else
 545		return -EIO;
 546}
 547
 548static int decode_sector_number(__be32 **rp, sector_t *sp)
 549{
 550	uint64_t s;
 551
 552	*rp = xdr_decode_hyper(*rp, &s);
 553	if (s & 0x1ff) {
 554		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
 555		return -1;
 556	}
 557	*sp = s >> SECTOR_SHIFT;
 558	return 0;
 559}
 560
 561static struct nfs4_deviceid_node *
 562bl_find_get_deviceid(struct nfs_server *server,
 563		const struct nfs4_deviceid *id, const struct cred *cred,
 564		gfp_t gfp_mask)
 565{
 566	struct nfs4_deviceid_node *node;
 567	int err = -ENODEV;
 568
 569retry:
 570	node = nfs4_find_get_deviceid(server, id, cred, gfp_mask);
 571	if (!node)
 572		return ERR_PTR(-ENODEV);
 573
 574	/*
 575	 * Devices that are marked unavailable are left in the cache with a
 576	 * timeout to avoid sending GETDEVINFO after every LAYOUTGET, or
 577	 * constantly attempting to register the device.  Once marked as
 578	 * unavailable they must be deleted and never reused.
 579	 */
 580	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags)) {
 581		unsigned long end = jiffies;
 582		unsigned long start = end - PNFS_DEVICE_RETRY_TIMEOUT;
 583
 584		if (!time_in_range(node->timestamp_unavailable, start, end)) {
 585			/* Uncork subsequent GETDEVINFO operations for this device */
 586			nfs4_delete_deviceid(node->ld, node->nfs_client, id);
 587			goto retry;
 588		}
 589		goto out_put;
 590	}
 591
 592	if (!bl_register_dev(container_of(node, struct pnfs_block_dev, node))) {
 593		/*
 594		 * If we cannot register, treat this device as transient:
 595		 * Make a negative cache entry for the device
 596		 */
 597		nfs4_mark_deviceid_unavailable(node);
 598		goto out_put;
 599	}
 600
 601	return node;
 602
 603out_put:
 604	nfs4_put_deviceid_node(node);
 605	return ERR_PTR(err);
 606}
 607
 608static int
 609bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
 610		struct layout_verification *lv, struct list_head *extents,
 611		gfp_t gfp_mask)
 612{
 613	struct pnfs_block_extent *be;
 614	struct nfs4_deviceid id;
 615	int error;
 616	__be32 *p;
 617
 618	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
 619	if (!p)
 620		return -EIO;
 621
 622	be = kzalloc(sizeof(*be), GFP_NOFS);
 623	if (!be)
 624		return -ENOMEM;
 625
 626	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
 627	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
 628
 629	be->be_device = bl_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
 630						lo->plh_lc_cred, gfp_mask);
 631	if (IS_ERR(be->be_device)) {
 632		error = PTR_ERR(be->be_device);
 633		goto out_free_be;
 634	}
 635
 636	/*
 637	 * The next three values are read in as bytes, but stored in the
 638	 * extent structure in 512-byte granularity.
 639	 */
 640	error = -EIO;
 641	if (decode_sector_number(&p, &be->be_f_offset) < 0)
 642		goto out_put_deviceid;
 643	if (decode_sector_number(&p, &be->be_length) < 0)
 644		goto out_put_deviceid;
 645	if (decode_sector_number(&p, &be->be_v_offset) < 0)
 646		goto out_put_deviceid;
 647	be->be_state = be32_to_cpup(p++);
 648
 649	error = verify_extent(be, lv);
 650	if (error) {
 651		dprintk("%s: extent verification failed\n", __func__);
 652		goto out_put_deviceid;
 653	}
 654
 655	list_add_tail(&be->be_list, extents);
 656	return 0;
 657
 658out_put_deviceid:
 659	nfs4_put_deviceid_node(be->be_device);
 660out_free_be:
 661	kfree(be);
 662	return error;
 663}
 664
 665static struct pnfs_layout_segment *
 666bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
 667		gfp_t gfp_mask)
 668{
 669	struct layout_verification lv = {
 670		.mode = lgr->range.iomode,
 671		.start = lgr->range.offset >> SECTOR_SHIFT,
 672		.inval = lgr->range.offset >> SECTOR_SHIFT,
 673		.cowread = lgr->range.offset >> SECTOR_SHIFT,
 674	};
 675	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
 676	struct pnfs_layout_segment *lseg;
 677	struct xdr_buf buf;
 678	struct xdr_stream xdr;
 679	struct folio *scratch;
 680	int status, i;
 681	uint32_t count;
 682	__be32 *p;
 683	LIST_HEAD(extents);
 684
 685	dprintk("---> %s\n", __func__);
 686
 687	lseg = kzalloc(sizeof(*lseg), gfp_mask);
 688	if (!lseg)
 689		return ERR_PTR(-ENOMEM);
 690
 691	status = -ENOMEM;
 692	scratch = folio_alloc(gfp_mask, 0);
 693	if (!scratch)
 694		goto out;
 695
 696	xdr_init_decode_pages(&xdr, &buf,
 697			lgr->layoutp->pages, lgr->layoutp->len);
 698	xdr_set_scratch_folio(&xdr, scratch);
 699
 700	status = -EIO;
 701	p = xdr_inline_decode(&xdr, 4);
 702	if (unlikely(!p))
 703		goto out_free_scratch;
 704
 705	count = be32_to_cpup(p++);
 706	dprintk("%s: number of extents %d\n", __func__, count);
 707
 708	/*
 709	 * Decode individual extents, putting them in temporary staging area
 710	 * until whole layout is decoded to make error recovery easier.
 711	 */
 712	for (i = 0; i < count; i++) {
 713		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
 714		if (status)
 715			goto process_extents;
 716	}
 717
 718	if (lgr->range.offset + lgr->range.length !=
 719			lv.start << SECTOR_SHIFT) {
 720		dprintk("%s Final length mismatch\n", __func__);
 721		status = -EIO;
 722		goto process_extents;
 723	}
 724
 725	if (lv.start < lv.cowread) {
 726		dprintk("%s Final uncovered COW extent\n", __func__);
 727		status = -EIO;
 728	}
 729
 730process_extents:
 731	while (!list_empty(&extents)) {
 732		struct pnfs_block_extent *be =
 733			list_first_entry(&extents, struct pnfs_block_extent,
 734					 be_list);
 735		list_del(&be->be_list);
 736
 737		if (!status)
 738			status = ext_tree_insert(bl, be);
 739
 740		if (status) {
 741			nfs4_put_deviceid_node(be->be_device);
 742			kfree(be);
 743		}
 744	}
 745
 746out_free_scratch:
 747	folio_put(scratch);
 748out:
 749	dprintk("%s returns %d\n", __func__, status);
 750	switch (status) {
 751	case -ENODEV:
 752		/* Our extent block devices are unavailable */
 753		set_bit(NFS_LSEG_UNAVAILABLE, &lseg->pls_flags);
 754		fallthrough;
 755	case 0:
 756		return lseg;
 757	default:
 758		kfree(lseg);
 759		return ERR_PTR(status);
 760	}
 761}
 762
 763static void
 764bl_return_range(struct pnfs_layout_hdr *lo,
 765		struct pnfs_layout_range *range)
 766{
 767	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
 768	sector_t offset = range->offset >> SECTOR_SHIFT, end;
 769
 770	if (range->offset % 8) {
 771		dprintk("%s: offset %lld not block size aligned\n",
 772			__func__, range->offset);
 773		return;
 774	}
 775
 776	if (range->length != NFS4_MAX_UINT64) {
 777		if (range->length % 8) {
 778			dprintk("%s: length %lld not block size aligned\n",
 779				__func__, range->length);
 780			return;
 781		}
 782
 783		end = offset + (range->length >> SECTOR_SHIFT);
 784	} else {
 785		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
 786	}
 787
 788	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
 789}
 790
 791static int
 792bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
 793{
 794	return ext_tree_prepare_commit(arg);
 795}
 796
 797static void
 798bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
 799{
 800	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
 801}
 802
 803static int
 804bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
 805{
 806	dprintk("%s enter\n", __func__);
 807
 808	if (server->pnfs_blksize == 0) {
 809		dprintk("%s Server did not return blksize\n", __func__);
 810		return -EINVAL;
 811	}
 812	if (server->pnfs_blksize > PAGE_SIZE) {
 813		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
 814			__func__, server->pnfs_blksize);
 815		return -EINVAL;
 816	}
 817
 818	return 0;
 819}
 820
 821static bool
 822is_aligned_req(struct nfs_pageio_descriptor *pgio,
 823		struct nfs_page *req, unsigned int alignment, bool is_write)
 824{
 825	/*
 826	 * Always accept buffered writes, higher layers take care of the
 827	 * right alignment.
 828	 */
 829	if (pgio->pg_dreq == NULL)
 830		return true;
 831
 832	if (!IS_ALIGNED(req->wb_offset, alignment))
 833		return false;
 834
 835	if (IS_ALIGNED(req->wb_bytes, alignment))
 836		return true;
 837
 838	if (is_write &&
 839	    (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
 840		/*
 841		 * If the write goes up to the inode size, just write
 842		 * the full page.  Data past the inode size is
 843		 * guaranteed to be zeroed by the higher level client
 844		 * code, and this behaviour is mandated by RFC 5663
 845		 * section 2.3.2.
 846		 */
 847		return true;
 848	}
 849
 850	return false;
 851}
 852
 853static void
 854bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
 855{
 856	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
 857		nfs_pageio_reset_read_mds(pgio);
 858		return;
 859	}
 860
 861	pnfs_generic_pg_init_read(pgio, req);
 862
 863	if (pgio->pg_lseg &&
 864		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
 865		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
 866		pnfs_set_lo_fail(pgio->pg_lseg);
 867		nfs_pageio_reset_read_mds(pgio);
 868	}
 869}
 870
 871/*
 872 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
 873 * of bytes (maximum @req->wb_bytes) that can be coalesced.
 874 */
 875static size_t
 876bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
 877		struct nfs_page *req)
 878{
 879	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
 880		return 0;
 881	return pnfs_generic_pg_test(pgio, prev, req);
 882}
 883
 884/*
 885 * Return the number of contiguous bytes for a given inode
 886 * starting at page frame idx.
 887 */
 888static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
 889{
 890	struct address_space *mapping = inode->i_mapping;
 891	pgoff_t end;
 892
 893	/* Optimize common case that writes from 0 to end of file */
 894	end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
 895	if (end != inode->i_mapping->nrpages) {
 896		rcu_read_lock();
 897		end = page_cache_next_miss(mapping, idx + 1, ULONG_MAX);
 898		rcu_read_unlock();
 899	}
 900
 901	if (!end)
 902		return i_size_read(inode) - (idx << PAGE_SHIFT);
 903	else
 904		return (end - idx) << PAGE_SHIFT;
 905}
 906
 907static void
 908bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
 909{
 910	u64 wb_size;
 911
 912	if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
 913		nfs_pageio_reset_write_mds(pgio);
 914		return;
 915	}
 916
 917	if (pgio->pg_dreq == NULL)
 918		wb_size = pnfs_num_cont_bytes(pgio->pg_inode, req->wb_index);
 919	else
 920		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq, req_offset(req));
 921
 922	pnfs_generic_pg_init_write(pgio, req, wb_size);
 923
 924	if (pgio->pg_lseg &&
 925		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
 926
 927		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
 928		pnfs_set_lo_fail(pgio->pg_lseg);
 929		nfs_pageio_reset_write_mds(pgio);
 930	}
 931}
 932
 933/*
 934 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
 935 * of bytes (maximum @req->wb_bytes) that can be coalesced.
 936 */
 937static size_t
 938bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
 939		 struct nfs_page *req)
 940{
 941	if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
 942		return 0;
 943	return pnfs_generic_pg_test(pgio, prev, req);
 944}
 945
 946static const struct nfs_pageio_ops bl_pg_read_ops = {
 947	.pg_init = bl_pg_init_read,
 948	.pg_test = bl_pg_test_read,
 949	.pg_doio = pnfs_generic_pg_readpages,
 950	.pg_cleanup = pnfs_generic_pg_cleanup,
 951};
 952
 953static const struct nfs_pageio_ops bl_pg_write_ops = {
 954	.pg_init = bl_pg_init_write,
 955	.pg_test = bl_pg_test_write,
 956	.pg_doio = pnfs_generic_pg_writepages,
 957	.pg_cleanup = pnfs_generic_pg_cleanup,
 958};
 959
 960static struct pnfs_layoutdriver_type blocklayout_type = {
 961	.id				= LAYOUT_BLOCK_VOLUME,
 962	.name				= "LAYOUT_BLOCK_VOLUME",
 963	.owner				= THIS_MODULE,
 964	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
 965					  PNFS_LAYOUTRET_ON_ERROR |
 966					  PNFS_READ_WHOLE_PAGE,
 967	.read_pagelist			= bl_read_pagelist,
 968	.write_pagelist			= bl_write_pagelist,
 969	.alloc_layout_hdr		= bl_alloc_layout_hdr,
 970	.free_layout_hdr		= bl_free_layout_hdr,
 971	.alloc_lseg			= bl_alloc_lseg,
 972	.free_lseg			= bl_free_lseg,
 973	.return_range			= bl_return_range,
 974	.prepare_layoutcommit		= bl_prepare_layoutcommit,
 975	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
 976	.set_layoutdriver		= bl_set_layoutdriver,
 977	.alloc_deviceid_node		= bl_alloc_deviceid_node,
 978	.free_deviceid_node		= bl_free_deviceid_node,
 979	.pg_read_ops			= &bl_pg_read_ops,
 980	.pg_write_ops			= &bl_pg_write_ops,
 981	.sync				= pnfs_generic_sync,
 982};
 983
 984static struct pnfs_layoutdriver_type scsilayout_type = {
 985	.id				= LAYOUT_SCSI,
 986	.name				= "LAYOUT_SCSI",
 987	.owner				= THIS_MODULE,
 988	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
 989					  PNFS_LAYOUTRET_ON_ERROR |
 990					  PNFS_READ_WHOLE_PAGE,
 991	.read_pagelist			= bl_read_pagelist,
 992	.write_pagelist			= bl_write_pagelist,
 993	.alloc_layout_hdr		= sl_alloc_layout_hdr,
 994	.free_layout_hdr		= bl_free_layout_hdr,
 995	.alloc_lseg			= bl_alloc_lseg,
 996	.free_lseg			= bl_free_lseg,
 997	.return_range			= bl_return_range,
 998	.prepare_layoutcommit		= bl_prepare_layoutcommit,
 999	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
1000	.set_layoutdriver		= bl_set_layoutdriver,
1001	.alloc_deviceid_node		= bl_alloc_deviceid_node,
1002	.free_deviceid_node		= bl_free_deviceid_node,
1003	.pg_read_ops			= &bl_pg_read_ops,
1004	.pg_write_ops			= &bl_pg_write_ops,
1005	.sync				= pnfs_generic_sync,
1006};
1007
1008
1009static int __init nfs4blocklayout_init(void)
1010{
1011	int ret;
1012
1013	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1014
1015	ret = bl_init_pipefs();
1016	if (ret)
1017		goto out;
1018
1019	ret = pnfs_register_layoutdriver(&blocklayout_type);
1020	if (ret)
1021		goto out_cleanup_pipe;
1022
1023	ret = pnfs_register_layoutdriver(&scsilayout_type);
1024	if (ret)
1025		goto out_unregister_block;
1026	return 0;
1027
1028out_unregister_block:
1029	pnfs_unregister_layoutdriver(&blocklayout_type);
1030out_cleanup_pipe:
1031	bl_cleanup_pipefs();
1032out:
1033	return ret;
1034}
1035
1036static void __exit nfs4blocklayout_exit(void)
1037{
1038	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1039	       __func__);
1040
1041	pnfs_unregister_layoutdriver(&scsilayout_type);
1042	pnfs_unregister_layoutdriver(&blocklayout_type);
1043	bl_cleanup_pipefs();
1044}
1045
1046MODULE_ALIAS("nfs-layouttype4-3");
1047MODULE_ALIAS("nfs-layouttype4-5");
1048
1049module_init(nfs4blocklayout_init);
1050module_exit(nfs4blocklayout_exit);