pagelist.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/fs/nfs/pagelist.c
   4 *
   5 * A set of helper functions for managing NFS read and write requests.
   6 * The main purpose of these routines is to provide support for the
   7 * coalescing of several requests into a single RPC call.
   8 *
   9 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
  10 *
  11 */
  12
  13#include <linux/slab.h>
  14#include <linux/file.h>
  15#include <linux/sched.h>
  16#include <linux/sunrpc/clnt.h>
  17#include <linux/nfs.h>
  18#include <linux/nfs3.h>
  19#include <linux/nfs4.h>
  20#include <linux/nfs_fs.h>
  21#include <linux/nfs_page.h>
  22#include <linux/nfs_mount.h>
  23#include <linux/export.h>
  24#include <linux/filelock.h>
  25
  26#include "internal.h"
  27#include "pnfs.h"
  28#include "nfstrace.h"
  29#include "fscache.h"
  30
  31#define NFSDBG_FACILITY		NFSDBG_PAGECACHE
  32
  33static struct kmem_cache *nfs_page_cachep;
  34static const struct rpc_call_ops nfs_pgio_common_ops;
  35
  36struct nfs_page_iter_page {
  37	const struct nfs_page *req;
  38	size_t count;
  39};
  40
  41static void nfs_page_iter_page_init(struct nfs_page_iter_page *i,
  42				    const struct nfs_page *req)
  43{
  44	i->req = req;
  45	i->count = 0;
  46}
  47
  48static void nfs_page_iter_page_advance(struct nfs_page_iter_page *i, size_t sz)
  49{
  50	const struct nfs_page *req = i->req;
  51	size_t tmp = i->count + sz;
  52
  53	i->count = (tmp < req->wb_bytes) ? tmp : req->wb_bytes;
  54}
  55
  56static struct page *nfs_page_iter_page_get(struct nfs_page_iter_page *i)
  57{
  58	const struct nfs_page *req = i->req;
  59	struct page *page;
  60
  61	if (i->count != req->wb_bytes) {
  62		size_t base = i->count + req->wb_pgbase;
  63		size_t len = PAGE_SIZE - offset_in_page(base);
  64
  65		page = nfs_page_to_page(req, base);
  66		nfs_page_iter_page_advance(i, len);
  67		return page;
  68	}
  69	return NULL;
  70}
  71
  72static struct nfs_pgio_mirror *
  73nfs_pgio_get_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
  74{
  75	if (desc->pg_ops->pg_get_mirror)
  76		return desc->pg_ops->pg_get_mirror(desc, idx);
  77	return &desc->pg_mirrors[0];
  78}
  79
  80struct nfs_pgio_mirror *
  81nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
  82{
  83	return nfs_pgio_get_mirror(desc, desc->pg_mirror_idx);
  84}
  85EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
  86
  87static u32
  88nfs_pgio_set_current_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
  89{
  90	if (desc->pg_ops->pg_set_mirror)
  91		return desc->pg_ops->pg_set_mirror(desc, idx);
  92	return desc->pg_mirror_idx;
  93}
  94
  95void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
  96		       struct nfs_pgio_header *hdr,
  97		       void (*release)(struct nfs_pgio_header *hdr))
  98{
  99	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 100
 101
 102	hdr->req = nfs_list_entry(mirror->pg_list.next);
 103	hdr->inode = desc->pg_inode;
 104	hdr->cred = nfs_req_openctx(hdr->req)->cred;
 105	hdr->io_start = req_offset(hdr->req);
 106	hdr->good_bytes = mirror->pg_count;
 107	hdr->io_completion = desc->pg_io_completion;
 108	hdr->dreq = desc->pg_dreq;
 109	nfs_netfs_set_pgio_header(hdr, desc);
 110	hdr->release = release;
 111	hdr->completion_ops = desc->pg_completion_ops;
 112	if (hdr->completion_ops->init_hdr)
 113		hdr->completion_ops->init_hdr(hdr);
 114
 115	hdr->pgio_mirror_idx = desc->pg_mirror_idx;
 116}
 117EXPORT_SYMBOL_GPL(nfs_pgheader_init);
 118
 119void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
 120{
 121	unsigned int new = pos - hdr->io_start;
 122
 123	trace_nfs_pgio_error(hdr, error, pos);
 124	if (hdr->good_bytes > new) {
 125		hdr->good_bytes = new;
 126		clear_bit(NFS_IOHDR_EOF, &hdr->flags);
 127		if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
 128			hdr->error = error;
 129	}
 130}
 131
 132static inline struct nfs_page *nfs_page_alloc(void)
 133{
 134	struct nfs_page *p =
 135		kmem_cache_zalloc(nfs_page_cachep, nfs_io_gfp_mask());
 136	if (p)
 137		INIT_LIST_HEAD(&p->wb_list);
 138	return p;
 139}
 140
 141static inline void
 142nfs_page_free(struct nfs_page *p)
 143{
 144	kmem_cache_free(nfs_page_cachep, p);
 145}
 146
 147/**
 148 * nfs_iocounter_wait - wait for i/o to complete
 149 * @l_ctx: nfs_lock_context with io_counter to use
 150 *
 151 * returns -ERESTARTSYS if interrupted by a fatal signal.
 152 * Otherwise returns 0 once the io_count hits 0.
 153 */
 154int
 155nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
 156{
 157	return wait_var_event_killable(&l_ctx->io_count,
 158				       !atomic_read(&l_ctx->io_count));
 159}
 160
 161/**
 162 * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
 163 * to complete
 164 * @task: the rpc_task that should wait
 165 * @l_ctx: nfs_lock_context with io_counter to check
 166 *
 167 * Returns true if there is outstanding I/O to wait on and the
 168 * task has been put to sleep.
 169 */
 170bool
 171nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
 172{
 173	struct inode *inode = d_inode(l_ctx->open_context->dentry);
 174	bool ret = false;
 175
 176	if (atomic_read(&l_ctx->io_count) > 0) {
 177		rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
 178		ret = true;
 179	}
 180
 181	if (atomic_read(&l_ctx->io_count) == 0) {
 182		rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
 183		ret = false;
 184	}
 185
 186	return ret;
 187}
 188EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
 189
 190/*
 191 * nfs_page_set_headlock - set the request PG_HEADLOCK
 192 * @req: request that is to be locked
 193 *
 194 * this lock must be held when modifying req->wb_head
 195 *
 196 * return 0 on success, < 0 on error
 197 */
 198int
 199nfs_page_set_headlock(struct nfs_page *req)
 200{
 201	if (!test_and_set_bit(PG_HEADLOCK, &req->wb_flags))
 202		return 0;
 203
 204	set_bit(PG_CONTENDED1, &req->wb_flags);
 205	smp_mb__after_atomic();
 206	return wait_on_bit_lock(&req->wb_flags, PG_HEADLOCK,
 207				TASK_UNINTERRUPTIBLE);
 208}
 209
 210/*
 211 * nfs_page_clear_headlock - clear the request PG_HEADLOCK
 212 * @req: request that is to be locked
 213 */
 214void
 215nfs_page_clear_headlock(struct nfs_page *req)
 216{
 217	clear_bit_unlock(PG_HEADLOCK, &req->wb_flags);
 218	smp_mb__after_atomic();
 219	if (!test_bit(PG_CONTENDED1, &req->wb_flags))
 220		return;
 221	wake_up_bit(&req->wb_flags, PG_HEADLOCK);
 222}
 223
 224/*
 225 * nfs_page_group_lock - lock the head of the page group
 226 * @req: request in group that is to be locked
 227 *
 228 * this lock must be held when traversing or modifying the page
 229 * group list
 230 *
 231 * return 0 on success, < 0 on error
 232 */
 233int
 234nfs_page_group_lock(struct nfs_page *req)
 235{
 236	int ret;
 237
 238	ret = nfs_page_set_headlock(req);
 239	if (ret || req->wb_head == req)
 240		return ret;
 241	return nfs_page_set_headlock(req->wb_head);
 242}
 243
 244/*
 245 * nfs_page_group_unlock - unlock the head of the page group
 246 * @req: request in group that is to be unlocked
 247 */
 248void
 249nfs_page_group_unlock(struct nfs_page *req)
 250{
 251	if (req != req->wb_head)
 252		nfs_page_clear_headlock(req->wb_head);
 253	nfs_page_clear_headlock(req);
 254}
 255
 256/**
 257 * nfs_page_group_sync_on_bit_locked - Test if all requests have @bit set
 258 * @req: request in page group
 259 * @bit: PG_* bit that is used to sync page group
 260 *
 261 * must be called with page group lock held
 262 */
 263bool nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
 264{
 265	struct nfs_page *head = req->wb_head;
 266	struct nfs_page *tmp;
 267
 268	WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
 269	WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
 270
 271	tmp = req->wb_this_page;
 272	while (tmp != req) {
 273		if (!test_bit(bit, &tmp->wb_flags))
 274			return false;
 275		tmp = tmp->wb_this_page;
 276	}
 277
 278	/* true! reset all bits */
 279	tmp = req;
 280	do {
 281		clear_bit(bit, &tmp->wb_flags);
 282		tmp = tmp->wb_this_page;
 283	} while (tmp != req);
 284
 285	return true;
 286}
 287
 288/*
 289 * nfs_page_group_sync_on_bit - set bit on current request, but only
 290 *   return true if the bit is set for all requests in page group
 291 * @req - request in page group
 292 * @bit - PG_* bit that is used to sync page group
 293 */
 294bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
 295{
 296	bool ret;
 297
 298	nfs_page_group_lock(req);
 299	ret = nfs_page_group_sync_on_bit_locked(req, bit);
 300	nfs_page_group_unlock(req);
 301
 302	return ret;
 303}
 304
 305/*
 306 * nfs_page_group_init - Initialize the page group linkage for @req
 307 * @req - a new nfs request
 308 * @prev - the previous request in page group, or NULL if @req is the first
 309 *         or only request in the group (the head).
 310 */
 311static inline void
 312nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
 313{
 314	struct inode *inode;
 315	WARN_ON_ONCE(prev == req);
 316
 317	if (!prev) {
 318		/* a head request */
 319		req->wb_head = req;
 320		req->wb_this_page = req;
 321	} else {
 322		/* a subrequest */
 323		WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
 324		WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
 325		req->wb_head = prev->wb_head;
 326		req->wb_this_page = prev->wb_this_page;
 327		prev->wb_this_page = req;
 328
 329		/* All subrequests take a ref on the head request until
 330		 * nfs_page_group_destroy is called */
 331		kref_get(&req->wb_head->wb_kref);
 332
 333		/* grab extra ref and bump the request count if head request
 334		 * has extra ref from the write/commit path to handle handoff
 335		 * between write and commit lists. */
 336		if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
 337			inode = nfs_page_to_inode(req);
 338			set_bit(PG_INODE_REF, &req->wb_flags);
 339			kref_get(&req->wb_kref);
 340			atomic_long_inc(&NFS_I(inode)->nrequests);
 341		}
 342	}
 343}
 344
 345/*
 346 * nfs_page_group_destroy - sync the destruction of page groups
 347 * @req - request that no longer needs the page group
 348 *
 349 * releases the page group reference from each member once all
 350 * members have called this function.
 351 */
 352static void
 353nfs_page_group_destroy(struct kref *kref)
 354{
 355	struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 356	struct nfs_page *head = req->wb_head;
 357	struct nfs_page *tmp, *next;
 358
 359	if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
 360		goto out;
 361
 362	tmp = req;
 363	do {
 364		next = tmp->wb_this_page;
 365		/* unlink and free */
 366		tmp->wb_this_page = tmp;
 367		tmp->wb_head = tmp;
 368		nfs_free_request(tmp);
 369		tmp = next;
 370	} while (tmp != req);
 371out:
 372	/* subrequests must release the ref on the head request */
 373	if (head != req)
 374		nfs_release_request(head);
 375}
 376
 377static struct nfs_page *nfs_page_create(struct nfs_lock_context *l_ctx,
 378					unsigned int pgbase, pgoff_t index,
 379					unsigned int offset, unsigned int count)
 380{
 381	struct nfs_page		*req;
 382	struct nfs_open_context *ctx = l_ctx->open_context;
 383
 384	if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
 385		return ERR_PTR(-EBADF);
 386	/* try to allocate the request struct */
 387	req = nfs_page_alloc();
 388	if (req == NULL)
 389		return ERR_PTR(-ENOMEM);
 390
 391	req->wb_lock_context = l_ctx;
 392	refcount_inc(&l_ctx->count);
 393	atomic_inc(&l_ctx->io_count);
 394
 395	/* Initialize the request struct. Initially, we assume a
 396	 * long write-back delay. This will be adjusted in
 397	 * update_nfs_request below if the region is not locked. */
 398	req->wb_pgbase = pgbase;
 399	req->wb_index = index;
 400	req->wb_offset = offset;
 401	req->wb_bytes = count;
 402	kref_init(&req->wb_kref);
 403	req->wb_nio = 0;
 404	return req;
 405}
 406
 407static void nfs_page_assign_folio(struct nfs_page *req, struct folio *folio)
 408{
 409	if (folio != NULL) {
 410		req->wb_folio = folio;
 411		folio_get(folio);
 412		set_bit(PG_FOLIO, &req->wb_flags);
 413	}
 414}
 415
 416static void nfs_page_assign_page(struct nfs_page *req, struct page *page)
 417{
 418	if (page != NULL) {
 419		req->wb_page = page;
 420		get_page(page);
 421	}
 422}
 423
 424/**
 425 * nfs_page_create_from_page - Create an NFS read/write request.
 426 * @ctx: open context to use
 427 * @page: page to write
 428 * @pgbase: starting offset within the page for the write
 429 * @offset: file offset for the write
 430 * @count: number of bytes to read/write
 431 *
 432 * The page must be locked by the caller. This makes sure we never
 433 * create two different requests for the same page.
 434 * User should ensure it is safe to sleep in this function.
 435 */
 436struct nfs_page *nfs_page_create_from_page(struct nfs_open_context *ctx,
 437					   struct page *page,
 438					   unsigned int pgbase, loff_t offset,
 439					   unsigned int count)
 440{
 441	struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
 442	struct nfs_page *ret;
 443
 444	if (IS_ERR(l_ctx))
 445		return ERR_CAST(l_ctx);
 446	ret = nfs_page_create(l_ctx, pgbase, offset >> PAGE_SHIFT,
 447			      offset_in_page(offset), count);
 448	if (!IS_ERR(ret)) {
 449		nfs_page_assign_page(ret, page);
 450		nfs_page_group_init(ret, NULL);
 451	}
 452	nfs_put_lock_context(l_ctx);
 453	return ret;
 454}
 455
 456/**
 457 * nfs_page_create_from_folio - Create an NFS read/write request.
 458 * @ctx: open context to use
 459 * @folio: folio to write
 460 * @offset: starting offset within the folio for the write
 461 * @count: number of bytes to read/write
 462 *
 463 * The page must be locked by the caller. This makes sure we never
 464 * create two different requests for the same page.
 465 * User should ensure it is safe to sleep in this function.
 466 */
 467struct nfs_page *nfs_page_create_from_folio(struct nfs_open_context *ctx,
 468					    struct folio *folio,
 469					    unsigned int offset,
 470					    unsigned int count)
 471{
 472	struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
 473	struct nfs_page *ret;
 474
 475	if (IS_ERR(l_ctx))
 476		return ERR_CAST(l_ctx);
 477	ret = nfs_page_create(l_ctx, offset, folio->index, offset, count);
 478	if (!IS_ERR(ret)) {
 479		nfs_page_assign_folio(ret, folio);
 480		nfs_page_group_init(ret, NULL);
 481	}
 482	nfs_put_lock_context(l_ctx);
 483	return ret;
 484}
 485
 486static struct nfs_page *
 487nfs_create_subreq(struct nfs_page *req,
 488		  unsigned int pgbase,
 489		  unsigned int offset,
 490		  unsigned int count)
 491{
 492	struct nfs_page *last;
 493	struct nfs_page *ret;
 494	struct folio *folio = nfs_page_to_folio(req);
 495	struct page *page = nfs_page_to_page(req, pgbase);
 496
 497	ret = nfs_page_create(req->wb_lock_context, pgbase, req->wb_index,
 498			      offset, count);
 499	if (!IS_ERR(ret)) {
 500		if (folio)
 501			nfs_page_assign_folio(ret, folio);
 502		else
 503			nfs_page_assign_page(ret, page);
 504		/* find the last request */
 505		for (last = req->wb_head;
 506		     last->wb_this_page != req->wb_head;
 507		     last = last->wb_this_page)
 508			;
 509
 510		nfs_lock_request(ret);
 511		nfs_page_group_init(ret, last);
 512		ret->wb_nio = req->wb_nio;
 513	}
 514	return ret;
 515}
 516
 517/**
 518 * nfs_unlock_request - Unlock request and wake up sleepers.
 519 * @req: pointer to request
 520 */
 521void nfs_unlock_request(struct nfs_page *req)
 522{
 523	clear_bit_unlock(PG_BUSY, &req->wb_flags);
 524	smp_mb__after_atomic();
 525	if (!test_bit(PG_CONTENDED2, &req->wb_flags))
 526		return;
 527	wake_up_bit(&req->wb_flags, PG_BUSY);
 528}
 529
 530/**
 531 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 532 * @req: pointer to request
 533 */
 534void nfs_unlock_and_release_request(struct nfs_page *req)
 535{
 536	nfs_unlock_request(req);
 537	nfs_release_request(req);
 538}
 539
 540/*
 541 * nfs_clear_request - Free up all resources allocated to the request
 542 * @req:
 543 *
 544 * Release page and open context resources associated with a read/write
 545 * request after it has completed.
 546 */
 547static void nfs_clear_request(struct nfs_page *req)
 548{
 549	struct folio *folio = nfs_page_to_folio(req);
 550	struct page *page = req->wb_page;
 551	struct nfs_lock_context *l_ctx = req->wb_lock_context;
 552	struct nfs_open_context *ctx;
 553
 554	if (folio != NULL) {
 555		folio_put(folio);
 556		req->wb_folio = NULL;
 557		clear_bit(PG_FOLIO, &req->wb_flags);
 558	} else if (page != NULL) {
 559		put_page(page);
 560		req->wb_page = NULL;
 561	}
 562	if (l_ctx != NULL) {
 563		if (atomic_dec_and_test(&l_ctx->io_count)) {
 564			wake_up_var(&l_ctx->io_count);
 565			ctx = l_ctx->open_context;
 566			if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
 567				rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
 568		}
 569		nfs_put_lock_context(l_ctx);
 570		req->wb_lock_context = NULL;
 571	}
 572}
 573
 574/**
 575 * nfs_free_request - Release the count on an NFS read/write request
 576 * @req: request to release
 577 *
 578 * Note: Should never be called with the spinlock held!
 579 */
 580void nfs_free_request(struct nfs_page *req)
 581{
 582	WARN_ON_ONCE(req->wb_this_page != req);
 583
 584	/* extra debug: make sure no sync bits are still set */
 585	WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
 586	WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
 587	WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
 588	WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
 589	WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
 590
 591	/* Release struct file and open context */
 592	nfs_clear_request(req);
 593	nfs_page_free(req);
 594}
 595
 596void nfs_release_request(struct nfs_page *req)
 597{
 598	kref_put(&req->wb_kref, nfs_page_group_destroy);
 599}
 600EXPORT_SYMBOL_GPL(nfs_release_request);
 601
 602/*
 603 * nfs_generic_pg_test - determine if requests can be coalesced
 604 * @desc: pointer to descriptor
 605 * @prev: previous request in desc, or NULL
 606 * @req: this request
 607 *
 608 * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
 609 * the size of the request.
 610 */
 611size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
 612			   struct nfs_page *prev, struct nfs_page *req)
 613{
 614	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 615
 616
 617	if (mirror->pg_count > mirror->pg_bsize) {
 618		/* should never happen */
 619		WARN_ON_ONCE(1);
 620		return 0;
 621	}
 622
 623	/*
 624	 * Limit the request size so that we can still allocate a page array
 625	 * for it without upsetting the slab allocator.
 626	 */
 627	if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
 628			sizeof(struct page *) > PAGE_SIZE)
 629		return 0;
 630
 631	return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
 632}
 633EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 634
 635struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
 636{
 637	struct nfs_pgio_header *hdr = ops->rw_alloc_header();
 638
 639	if (hdr) {
 640		INIT_LIST_HEAD(&hdr->pages);
 641		hdr->rw_ops = ops;
 642	}
 643	return hdr;
 644}
 645EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
 646
 647/**
 648 * nfs_pgio_data_destroy - make @hdr suitable for reuse
 649 *
 650 * Frees memory and releases refs from nfs_generic_pgio, so that it may
 651 * be called again.
 652 *
 653 * @hdr: A header that has had nfs_generic_pgio called
 654 */
 655static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
 656{
 657	if (hdr->args.context)
 658		put_nfs_open_context(hdr->args.context);
 659	if (hdr->page_array.pagevec != hdr->page_array.page_array)
 660		kfree(hdr->page_array.pagevec);
 661}
 662
 663/*
 664 * nfs_pgio_header_free - Free a read or write header
 665 * @hdr: The header to free
 666 */
 667void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
 668{
 669	nfs_pgio_data_destroy(hdr);
 670	hdr->rw_ops->rw_free_header(hdr);
 671}
 672EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
 673
 674/**
 675 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
 676 * @hdr: The pageio hdr
 677 * @pgbase: base
 678 * @count: Number of bytes to read
 679 * @how: How to commit data (writes only)
 680 * @cinfo: Commit information for the call (writes only)
 681 */
 682static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr, unsigned int pgbase,
 683			      unsigned int count, int how,
 684			      struct nfs_commit_info *cinfo)
 685{
 686	struct nfs_page *req = hdr->req;
 687
 688	/* Set up the RPC argument and reply structs
 689	 * NB: take care not to mess about with hdr->commit et al. */
 690
 691	hdr->args.fh     = NFS_FH(hdr->inode);
 692	hdr->args.offset = req_offset(req);
 693	/* pnfs_set_layoutcommit needs this */
 694	hdr->mds_offset = hdr->args.offset;
 695	hdr->args.pgbase = pgbase;
 696	hdr->args.pages  = hdr->page_array.pagevec;
 697	hdr->args.count  = count;
 698	hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
 699	hdr->args.lock_context = req->wb_lock_context;
 700	hdr->args.stable  = NFS_UNSTABLE;
 701	switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
 702	case 0:
 703		break;
 704	case FLUSH_COND_STABLE:
 705		if (nfs_reqs_to_commit(cinfo))
 706			break;
 707		fallthrough;
 708	default:
 709		hdr->args.stable = NFS_FILE_SYNC;
 710	}
 711
 712	hdr->res.fattr   = &hdr->fattr;
 713	hdr->res.count   = 0;
 714	hdr->res.eof     = 0;
 715	hdr->res.verf    = &hdr->verf;
 716	nfs_fattr_init(&hdr->fattr);
 717}
 718
 719/**
 720 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
 721 * @task: The current task
 722 * @calldata: pageio header to prepare
 723 */
 724static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
 725{
 726	struct nfs_pgio_header *hdr = calldata;
 727	int err;
 728	err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
 729	if (err)
 730		rpc_exit(task, err);
 731}
 732
 733int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
 734		      const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
 735		      const struct rpc_call_ops *call_ops, int how, int flags,
 736		      struct nfsd_file *localio)
 737{
 738	struct rpc_task *task;
 739	struct rpc_message msg = {
 740		.rpc_argp = &hdr->args,
 741		.rpc_resp = &hdr->res,
 742		.rpc_cred = cred,
 743	};
 744	struct rpc_task_setup task_setup_data = {
 745		.rpc_client = clnt,
 746		.task = &hdr->task,
 747		.rpc_message = &msg,
 748		.callback_ops = call_ops,
 749		.callback_data = hdr,
 750		.workqueue = nfsiod_workqueue,
 751		.flags = RPC_TASK_ASYNC | flags,
 752	};
 753
 754	if (nfs_server_capable(hdr->inode, NFS_CAP_MOVEABLE))
 755		task_setup_data.flags |= RPC_TASK_MOVEABLE;
 756
 757	hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
 758
 759	dprintk("NFS: initiated pgio call "
 760		"(req %s/%llu, %u bytes @ offset %llu)\n",
 761		hdr->inode->i_sb->s_id,
 762		(unsigned long long)NFS_FILEID(hdr->inode),
 763		hdr->args.count,
 764		(unsigned long long)hdr->args.offset);
 765
 766	if (localio)
 767		return nfs_local_doio(NFS_SERVER(hdr->inode)->nfs_client,
 768				      localio, hdr, call_ops);
 769
 770	task = rpc_run_task(&task_setup_data);
 771	if (IS_ERR(task))
 772		return PTR_ERR(task);
 773	rpc_put_task(task);
 774	return 0;
 775}
 776EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
 777
 778/**
 779 * nfs_pgio_error - Clean up from a pageio error
 780 * @hdr: pageio header
 781 */
 782static void nfs_pgio_error(struct nfs_pgio_header *hdr)
 783{
 784	set_bit(NFS_IOHDR_REDO, &hdr->flags);
 785	hdr->completion_ops->completion(hdr);
 786}
 787
 788/**
 789 * nfs_pgio_release - Release pageio data
 790 * @calldata: The pageio header to release
 791 */
 792static void nfs_pgio_release(void *calldata)
 793{
 794	struct nfs_pgio_header *hdr = calldata;
 795	hdr->completion_ops->completion(hdr);
 796}
 797
 798static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
 799				   unsigned int bsize)
 800{
 801	INIT_LIST_HEAD(&mirror->pg_list);
 802	mirror->pg_bytes_written = 0;
 803	mirror->pg_count = 0;
 804	mirror->pg_bsize = bsize;
 805	mirror->pg_base = 0;
 806	mirror->pg_recoalesce = 0;
 807}
 808
 809/**
 810 * nfs_pageio_init - initialise a page io descriptor
 811 * @desc: pointer to descriptor
 812 * @inode: pointer to inode
 813 * @pg_ops: pointer to pageio operations
 814 * @compl_ops: pointer to pageio completion operations
 815 * @rw_ops: pointer to nfs read/write operations
 816 * @bsize: io block size
 817 * @io_flags: extra parameters for the io function
 818 */
 819void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 820		     struct inode *inode,
 821		     const struct nfs_pageio_ops *pg_ops,
 822		     const struct nfs_pgio_completion_ops *compl_ops,
 823		     const struct nfs_rw_ops *rw_ops,
 824		     size_t bsize,
 825		     int io_flags)
 826{
 827	desc->pg_moreio = 0;
 828	desc->pg_inode = inode;
 829	desc->pg_ops = pg_ops;
 830	desc->pg_completion_ops = compl_ops;
 831	desc->pg_rw_ops = rw_ops;
 832	desc->pg_ioflags = io_flags;
 833	desc->pg_error = 0;
 834	desc->pg_lseg = NULL;
 835	desc->pg_io_completion = NULL;
 836	desc->pg_dreq = NULL;
 837	nfs_netfs_reset_pageio_descriptor(desc);
 838	desc->pg_bsize = bsize;
 839
 840	desc->pg_mirror_count = 1;
 841	desc->pg_mirror_idx = 0;
 842
 843	desc->pg_mirrors_dynamic = NULL;
 844	desc->pg_mirrors = desc->pg_mirrors_static;
 845	nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
 846	desc->pg_maxretrans = 0;
 847}
 848
 849/**
 850 * nfs_pgio_result - Basic pageio error handling
 851 * @task: The task that ran
 852 * @calldata: Pageio header to check
 853 */
 854static void nfs_pgio_result(struct rpc_task *task, void *calldata)
 855{
 856	struct nfs_pgio_header *hdr = calldata;
 857	struct inode *inode = hdr->inode;
 858
 859	if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
 860		return;
 861	if (task->tk_status < 0)
 862		nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
 863	else
 864		hdr->rw_ops->rw_result(task, hdr);
 865}
 866
 867/*
 868 * Create an RPC task for the given read or write request and kick it.
 869 * The page must have been locked by the caller.
 870 *
 871 * It may happen that the page we're passed is not marked dirty.
 872 * This is the case if nfs_updatepage detects a conflicting request
 873 * that has been written but not committed.
 874 */
 875int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
 876		     struct nfs_pgio_header *hdr)
 877{
 878	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 879
 880	struct nfs_page		*req;
 881	struct page		**pages,
 882				*last_page;
 883	struct list_head *head = &mirror->pg_list;
 884	struct nfs_commit_info cinfo;
 885	struct nfs_page_array *pg_array = &hdr->page_array;
 886	unsigned int pagecount, pageused;
 887	unsigned int pg_base = offset_in_page(mirror->pg_base);
 888	gfp_t gfp_flags = nfs_io_gfp_mask();
 889
 890	pagecount = nfs_page_array_len(pg_base, mirror->pg_count);
 891	pg_array->npages = pagecount;
 892
 893	if (pagecount <= ARRAY_SIZE(pg_array->page_array))
 894		pg_array->pagevec = pg_array->page_array;
 895	else {
 896		pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
 897		if (!pg_array->pagevec) {
 898			pg_array->npages = 0;
 899			nfs_pgio_error(hdr);
 900			desc->pg_error = -ENOMEM;
 901			return desc->pg_error;
 902		}
 903	}
 904
 905	nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
 906	pages = hdr->page_array.pagevec;
 907	last_page = NULL;
 908	pageused = 0;
 909	while (!list_empty(head)) {
 910		struct nfs_page_iter_page i;
 911		struct page *page;
 912
 913		req = nfs_list_entry(head->next);
 914		nfs_list_move_request(req, &hdr->pages);
 915
 916		if (req->wb_pgbase == 0)
 917			last_page = NULL;
 918
 919		nfs_page_iter_page_init(&i, req);
 920		while ((page = nfs_page_iter_page_get(&i)) != NULL) {
 921			if (last_page != page) {
 922				pageused++;
 923				if (pageused > pagecount)
 924					goto full;
 925				*pages++ = last_page = page;
 926			}
 927		}
 928	}
 929full:
 930	if (WARN_ON_ONCE(pageused != pagecount)) {
 931		nfs_pgio_error(hdr);
 932		desc->pg_error = -EINVAL;
 933		return desc->pg_error;
 934	}
 935
 936	if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
 937	    (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
 938		desc->pg_ioflags &= ~FLUSH_COND_STABLE;
 939
 940	/* Set up the argument struct */
 941	nfs_pgio_rpcsetup(hdr, pg_base, mirror->pg_count, desc->pg_ioflags,
 942			  &cinfo);
 943	desc->pg_rpc_callops = &nfs_pgio_common_ops;
 944	return 0;
 945}
 946EXPORT_SYMBOL_GPL(nfs_generic_pgio);
 947
 948static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 949{
 950	struct nfs_pgio_header *hdr;
 951	int ret;
 952	unsigned short task_flags = 0;
 953
 954	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
 955	if (!hdr) {
 956		desc->pg_error = -ENOMEM;
 957		return desc->pg_error;
 958	}
 959	nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
 960	ret = nfs_generic_pgio(desc, hdr);
 961	if (ret == 0) {
 962		struct nfs_client *clp = NFS_SERVER(hdr->inode)->nfs_client;
 963
 964		struct nfsd_file *localio =
 965			nfs_local_open_fh(clp, hdr->cred, hdr->args.fh,
 966					  &hdr->args.context->nfl,
 967					  hdr->args.context->mode);
 968
 969		if (NFS_SERVER(hdr->inode)->nfs_client->cl_minorversion)
 970			task_flags = RPC_TASK_MOVEABLE;
 971		ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
 972					hdr,
 973					hdr->cred,
 974					NFS_PROTO(hdr->inode),
 975					desc->pg_rpc_callops,
 976					desc->pg_ioflags,
 977					RPC_TASK_CRED_NOREF | task_flags,
 978					localio);
 979	}
 980	return ret;
 981}
 982
 983static struct nfs_pgio_mirror *
 984nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
 985		unsigned int mirror_count)
 986{
 987	struct nfs_pgio_mirror *ret;
 988	unsigned int i;
 989
 990	kfree(desc->pg_mirrors_dynamic);
 991	desc->pg_mirrors_dynamic = NULL;
 992	if (mirror_count == 1)
 993		return desc->pg_mirrors_static;
 994	ret = kmalloc_array(mirror_count, sizeof(*ret), nfs_io_gfp_mask());
 995	if (ret != NULL) {
 996		for (i = 0; i < mirror_count; i++)
 997			nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
 998		desc->pg_mirrors_dynamic = ret;
 999	}
1000	return ret;
1001}
1002
1003/*
1004 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
1005 *				by calling the pg_get_mirror_count op
1006 */
1007static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
1008				       struct nfs_page *req)
1009{
1010	unsigned int mirror_count = 1;
1011
1012	if (pgio->pg_ops->pg_get_mirror_count)
1013		mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
1014	if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
1015		return;
1016
1017	if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
1018		pgio->pg_error = -EINVAL;
1019		return;
1020	}
1021
1022	pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
1023	if (pgio->pg_mirrors == NULL) {
1024		pgio->pg_error = -ENOMEM;
1025		pgio->pg_mirrors = pgio->pg_mirrors_static;
1026		mirror_count = 1;
1027	}
1028	pgio->pg_mirror_count = mirror_count;
1029}
1030
1031static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
1032{
1033	pgio->pg_mirror_count = 1;
1034	pgio->pg_mirror_idx = 0;
1035	pgio->pg_mirrors = pgio->pg_mirrors_static;
1036	kfree(pgio->pg_mirrors_dynamic);
1037	pgio->pg_mirrors_dynamic = NULL;
1038}
1039
1040static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
1041		const struct nfs_lock_context *l2)
1042{
1043	return l1->lockowner == l2->lockowner;
1044}
1045
1046static bool nfs_page_is_contiguous(const struct nfs_page *prev,
1047				   const struct nfs_page *req)
1048{
1049	size_t prev_end = prev->wb_pgbase + prev->wb_bytes;
1050
1051	if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
1052		return false;
1053	if (req->wb_pgbase == 0)
1054		return prev_end == nfs_page_max_length(prev);
1055	if (req->wb_pgbase == prev_end) {
1056		struct folio *folio = nfs_page_to_folio(req);
1057		if (folio)
1058			return folio == nfs_page_to_folio(prev);
1059		return req->wb_page == prev->wb_page;
1060	}
1061	return false;
1062}
1063
1064/**
1065 * nfs_coalesce_size - test two requests for compatibility
1066 * @prev: pointer to nfs_page
1067 * @req: pointer to nfs_page
1068 * @pgio: pointer to nfs_pagio_descriptor
1069 *
1070 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
1071 * page data area they describe is contiguous, and that their RPC
1072 * credentials, NFSv4 open state, and lockowners are the same.
1073 *
1074 * Returns size of the request that can be coalesced
1075 */
1076static unsigned int nfs_coalesce_size(struct nfs_page *prev,
1077				      struct nfs_page *req,
1078				      struct nfs_pageio_descriptor *pgio)
1079{
1080	struct file_lock_context *flctx;
1081
1082	if (prev) {
1083		if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
1084			return 0;
1085		flctx = locks_inode_context(d_inode(nfs_req_openctx(req)->dentry));
1086		if (flctx != NULL &&
1087		    !(list_empty_careful(&flctx->flc_posix) &&
1088		      list_empty_careful(&flctx->flc_flock)) &&
1089		    !nfs_match_lock_context(req->wb_lock_context,
1090					    prev->wb_lock_context))
1091			return 0;
1092		if (!nfs_page_is_contiguous(prev, req))
1093			return 0;
1094	}
1095	return pgio->pg_ops->pg_test(pgio, prev, req);
1096}
1097
1098/**
1099 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
1100 * @desc: destination io descriptor
1101 * @req: request
1102 *
1103 * If the request 'req' was successfully coalesced into the existing list
1104 * of pages 'desc', it returns the size of req.
1105 */
1106static unsigned int
1107nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
1108		struct nfs_page *req)
1109{
1110	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1111	struct nfs_page *prev = NULL;
1112	unsigned int size;
1113
1114	if (list_empty(&mirror->pg_list)) {
1115		if (desc->pg_ops->pg_init)
1116			desc->pg_ops->pg_init(desc, req);
1117		if (desc->pg_error < 0)
1118			return 0;
1119		mirror->pg_base = req->wb_pgbase;
1120		mirror->pg_count = 0;
1121		mirror->pg_recoalesce = 0;
1122	} else
1123		prev = nfs_list_entry(mirror->pg_list.prev);
1124
1125	if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
1126		if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
1127			desc->pg_error = -ETIMEDOUT;
1128		else
1129			desc->pg_error = -EIO;
1130		return 0;
1131	}
1132
1133	size = nfs_coalesce_size(prev, req, desc);
1134	if (size < req->wb_bytes)
1135		return size;
1136	nfs_list_move_request(req, &mirror->pg_list);
1137	mirror->pg_count += req->wb_bytes;
1138	return req->wb_bytes;
1139}
1140
1141/*
1142 * Helper for nfs_pageio_add_request and nfs_pageio_complete
1143 */
1144static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1145{
1146	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1147
1148	if (!list_empty(&mirror->pg_list)) {
1149		int error = desc->pg_ops->pg_doio(desc);
1150		if (error < 0)
1151			desc->pg_error = error;
1152		if (list_empty(&mirror->pg_list))
1153			mirror->pg_bytes_written += mirror->pg_count;
1154	}
1155}
1156
1157static void
1158nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1159		struct nfs_page *req)
1160{
1161	LIST_HEAD(head);
1162
1163	nfs_list_move_request(req, &head);
1164	desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1165}
1166
1167/**
1168 * __nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1169 * @desc: destination io descriptor
1170 * @req: request
1171 *
1172 * This may split a request into subrequests which are all part of the
1173 * same page group. If so, it will submit @req as the last one, to ensure
1174 * the pointer to @req is still valid in case of failure.
1175 *
1176 * Returns true if the request 'req' was successfully coalesced into the
1177 * existing list of pages 'desc'.
1178 */
1179static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1180			   struct nfs_page *req)
1181{
1182	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1183	struct nfs_page *subreq;
1184	unsigned int size, subreq_size;
1185
1186	nfs_page_group_lock(req);
1187
1188	subreq = req;
1189	subreq_size = subreq->wb_bytes;
1190	for(;;) {
1191		size = nfs_pageio_do_add_request(desc, subreq);
1192		if (size == subreq_size) {
1193			/* We successfully submitted a request */
1194			if (subreq == req)
1195				break;
1196			req->wb_pgbase += size;
1197			req->wb_bytes -= size;
1198			req->wb_offset += size;
1199			subreq_size = req->wb_bytes;
1200			subreq = req;
1201			continue;
1202		}
1203		if (WARN_ON_ONCE(subreq != req)) {
1204			nfs_page_group_unlock(req);
1205			nfs_pageio_cleanup_request(desc, subreq);
1206			subreq = req;
1207			subreq_size = req->wb_bytes;
1208			nfs_page_group_lock(req);
1209		}
1210		if (!size) {
1211			/* Can't coalesce any more, so do I/O */
1212			nfs_page_group_unlock(req);
1213			desc->pg_moreio = 1;
1214			nfs_pageio_doio(desc);
1215			if (desc->pg_error < 0 || mirror->pg_recoalesce)
1216				return 0;
1217			/* retry add_request for this subreq */
1218			nfs_page_group_lock(req);
1219			continue;
1220		}
1221		subreq = nfs_create_subreq(req, req->wb_pgbase,
1222				req->wb_offset, size);
1223		if (IS_ERR(subreq))
1224			goto err_ptr;
1225		subreq_size = size;
1226	}
1227
1228	nfs_page_group_unlock(req);
1229	return 1;
1230err_ptr:
1231	desc->pg_error = PTR_ERR(subreq);
1232	nfs_page_group_unlock(req);
1233	return 0;
1234}
1235
1236static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1237{
1238	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1239	LIST_HEAD(head);
1240
1241	do {
1242		list_splice_init(&mirror->pg_list, &head);
1243		mirror->pg_recoalesce = 0;
1244
1245		while (!list_empty(&head)) {
1246			struct nfs_page *req;
1247
1248			req = list_first_entry(&head, struct nfs_page, wb_list);
1249			if (__nfs_pageio_add_request(desc, req))
1250				continue;
1251			if (desc->pg_error < 0) {
1252				list_splice_tail(&head, &mirror->pg_list);
1253				mirror->pg_recoalesce = 1;
1254				return 0;
1255			}
1256			break;
1257		}
1258	} while (mirror->pg_recoalesce);
1259	return 1;
1260}
1261
1262static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1263		struct nfs_page *req)
1264{
1265	int ret;
1266
1267	do {
1268		ret = __nfs_pageio_add_request(desc, req);
1269		if (ret)
1270			break;
1271		if (desc->pg_error < 0)
1272			break;
1273		ret = nfs_do_recoalesce(desc);
1274	} while (ret);
1275
1276	return ret;
1277}
1278
1279static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1280{
1281	u32 midx;
1282	struct nfs_pgio_mirror *mirror;
1283
1284	if (!desc->pg_error)
1285		return;
1286
1287	for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1288		mirror = nfs_pgio_get_mirror(desc, midx);
1289		desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1290				desc->pg_error);
1291	}
1292}
1293
1294int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1295			   struct nfs_page *req)
1296{
1297	u32 midx;
1298	unsigned int pgbase, offset, bytes;
1299	struct nfs_page *dupreq;
1300
1301	pgbase = req->wb_pgbase;
1302	offset = req->wb_offset;
1303	bytes = req->wb_bytes;
1304
1305	nfs_pageio_setup_mirroring(desc, req);
1306	if (desc->pg_error < 0)
1307		goto out_failed;
1308
1309	/* Create the mirror instances first, and fire them off */
1310	for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1311		nfs_page_group_lock(req);
1312
1313		dupreq = nfs_create_subreq(req,
1314				pgbase, offset, bytes);
1315
1316		nfs_page_group_unlock(req);
1317		if (IS_ERR(dupreq)) {
1318			desc->pg_error = PTR_ERR(dupreq);
1319			goto out_failed;
1320		}
1321
1322		nfs_pgio_set_current_mirror(desc, midx);
1323		if (!nfs_pageio_add_request_mirror(desc, dupreq))
1324			goto out_cleanup_subreq;
1325	}
1326
1327	nfs_pgio_set_current_mirror(desc, 0);
1328	if (!nfs_pageio_add_request_mirror(desc, req))
1329		goto out_failed;
1330
1331	return 1;
1332
1333out_cleanup_subreq:
1334	nfs_pageio_cleanup_request(desc, dupreq);
1335out_failed:
1336	nfs_pageio_error_cleanup(desc);
1337	return 0;
1338}
1339
1340/*
1341 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1342 *				nfs_pageio_descriptor
1343 * @desc: pointer to io descriptor
1344 * @mirror_idx: pointer to mirror index
1345 */
1346static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1347				       u32 mirror_idx)
1348{
1349	struct nfs_pgio_mirror *mirror;
1350	u32 restore_idx;
1351
1352	restore_idx = nfs_pgio_set_current_mirror(desc, mirror_idx);
1353	mirror = nfs_pgio_current_mirror(desc);
1354
1355	for (;;) {
1356		nfs_pageio_doio(desc);
1357		if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1358			break;
1359		if (!nfs_do_recoalesce(desc))
1360			break;
1361	}
1362	nfs_pgio_set_current_mirror(desc, restore_idx);
1363}
1364
1365/*
1366 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1367 * @hdr - the pgio header to move request from
1368 * @desc - the pageio descriptor to add requests to
1369 *
1370 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1371 * to send them.
1372 *
1373 * Returns 0 on success and < 0 on error.
1374 */
1375int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1376		      struct nfs_pgio_header *hdr)
1377{
1378	LIST_HEAD(pages);
1379
1380	desc->pg_io_completion = hdr->io_completion;
1381	desc->pg_dreq = hdr->dreq;
1382	nfs_netfs_set_pageio_descriptor(desc, hdr);
1383	list_splice_init(&hdr->pages, &pages);
1384	while (!list_empty(&pages)) {
1385		struct nfs_page *req = nfs_list_entry(pages.next);
1386
1387		if (!nfs_pageio_add_request(desc, req))
1388			break;
1389	}
1390	nfs_pageio_complete(desc);
1391	if (!list_empty(&pages)) {
1392		int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1393		hdr->completion_ops->error_cleanup(&pages, err);
1394		nfs_set_pgio_error(hdr, err, hdr->io_start);
1395		return err;
1396	}
1397	return 0;
1398}
1399EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1400
1401/**
1402 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1403 * @desc: pointer to io descriptor
1404 */
1405void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1406{
1407	u32 midx;
1408
1409	for (midx = 0; midx < desc->pg_mirror_count; midx++)
1410		nfs_pageio_complete_mirror(desc, midx);
1411
1412	if (desc->pg_error < 0)
1413		nfs_pageio_error_cleanup(desc);
1414	if (desc->pg_ops->pg_cleanup)
1415		desc->pg_ops->pg_cleanup(desc);
1416	nfs_pageio_cleanup_mirroring(desc);
1417}
1418
1419/**
1420 * nfs_pageio_cond_complete - Conditional I/O completion
1421 * @desc: pointer to io descriptor
1422 * @index: page index
1423 *
1424 * It is important to ensure that processes don't try to take locks
1425 * on non-contiguous ranges of pages as that might deadlock. This
1426 * function should be called before attempting to wait on a locked
1427 * nfs_page. It will complete the I/O if the page index 'index'
1428 * is not contiguous with the existing list of pages in 'desc'.
1429 */
1430void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1431{
1432	struct nfs_pgio_mirror *mirror;
1433	struct nfs_page *prev;
1434	struct folio *folio;
1435	u32 midx;
1436
1437	for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1438		mirror = nfs_pgio_get_mirror(desc, midx);
1439		if (!list_empty(&mirror->pg_list)) {
1440			prev = nfs_list_entry(mirror->pg_list.prev);
1441			folio = nfs_page_to_folio(prev);
1442			if (folio) {
1443				if (index == folio_next_index(folio))
1444					continue;
1445			} else if (index == prev->wb_index + 1)
1446				continue;
1447			/*
1448			 * We will submit more requests after these. Indicate
1449			 * this to the underlying layers.
1450			 */
1451			desc->pg_moreio = 1;
1452			nfs_pageio_complete(desc);
1453			break;
1454		}
1455	}
1456}
1457
1458/*
1459 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1460 */
1461void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1462{
1463	nfs_pageio_complete(pgio);
1464}
1465
1466int __init nfs_init_nfspagecache(void)
1467{
1468	nfs_page_cachep = kmem_cache_create("nfs_page",
1469					    sizeof(struct nfs_page),
1470					    0, SLAB_HWCACHE_ALIGN,
1471					    NULL);
1472	if (nfs_page_cachep == NULL)
1473		return -ENOMEM;
1474
1475	return 0;
1476}
1477
1478void nfs_destroy_nfspagecache(void)
1479{
1480	kmem_cache_destroy(nfs_page_cachep);
1481}
1482
1483static const struct rpc_call_ops nfs_pgio_common_ops = {
1484	.rpc_call_prepare = nfs_pgio_prepare,
1485	.rpc_call_done = nfs_pgio_result,
1486	.rpc_release = nfs_pgio_release,
1487};
1488
1489const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1490	.pg_test = nfs_generic_pg_test,
1491	.pg_doio = nfs_generic_pg_pgios,
1492};