1/*
   2 * fs/f2fs/data.c
   3 *
   4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5 *             http://www.samsung.com/
   6 *
   7 * This program is free software; you can redistribute it and/or modify
   8 * it under the terms of the GNU General Public License version 2 as
   9 * published by the Free Software Foundation.
  10 */
  11#include <linux/fs.h>
  12#include <linux/f2fs_fs.h>
  13#include <linux/buffer_head.h>
  14#include <linux/mpage.h>
  15#include <linux/writeback.h>
  16#include <linux/backing-dev.h>
  17#include <linux/pagevec.h>
  18#include <linux/blkdev.h>
  19#include <linux/bio.h>
  20#include <linux/prefetch.h>
  21#include <linux/uio.h>
  22#include <linux/mm.h>
  23#include <linux/memcontrol.h>
  24#include <linux/cleancache.h>
  25#include <linux/sched/signal.h>
  26
  27#include "f2fs.h"
  28#include "node.h"
  29#include "segment.h"
  30#include "trace.h"
  31#include <trace/events/f2fs.h>
  32
  33static bool __is_cp_guaranteed(struct page *page)
  34{
  35	struct address_space *mapping = page->mapping;
  36	struct inode *inode;
  37	struct f2fs_sb_info *sbi;
  38
  39	if (!mapping)
  40		return false;
  41
  42	inode = mapping->host;
  43	sbi = F2FS_I_SB(inode);
  44
  45	if (inode->i_ino == F2FS_META_INO(sbi) ||
  46			inode->i_ino ==  F2FS_NODE_INO(sbi) ||
  47			S_ISDIR(inode->i_mode) ||
  48			is_cold_data(page))
  49		return true;
  50	return false;
  51}
  52
  53static void f2fs_read_end_io(struct bio *bio)
  54{
  55	struct bio_vec *bvec;
  56	int i;
  57
  58#ifdef CONFIG_F2FS_FAULT_INJECTION
  59	if (time_to_inject(F2FS_P_SB(bio_first_page_all(bio)), FAULT_IO)) {
  60		f2fs_show_injection_info(FAULT_IO);
  61		bio->bi_status = BLK_STS_IOERR;
  62	}
  63#endif
  64
  65	if (f2fs_bio_encrypted(bio)) {
  66		if (bio->bi_status) {
  67			fscrypt_release_ctx(bio->bi_private);
  68		} else {
  69			fscrypt_decrypt_bio_pages(bio->bi_private, bio);
  70			return;
  71		}
  72	}
  73
  74	bio_for_each_segment_all(bvec, bio, i) {
  75		struct page *page = bvec->bv_page;
  76
  77		if (!bio->bi_status) {
  78			if (!PageUptodate(page))
  79				SetPageUptodate(page);
  80		} else {
  81			ClearPageUptodate(page);
  82			SetPageError(page);
  83		}
  84		unlock_page(page);
  85	}
  86	bio_put(bio);
  87}
  88
  89static void f2fs_write_end_io(struct bio *bio)
  90{
  91	struct f2fs_sb_info *sbi = bio->bi_private;
  92	struct bio_vec *bvec;
  93	int i;
  94
  95	bio_for_each_segment_all(bvec, bio, i) {
  96		struct page *page = bvec->bv_page;
  97		enum count_type type = WB_DATA_TYPE(page);
  98
  99		if (IS_DUMMY_WRITTEN_PAGE(page)) {
 100			set_page_private(page, (unsigned long)NULL);
 101			ClearPagePrivate(page);
 102			unlock_page(page);
 103			mempool_free(page, sbi->write_io_dummy);
 104
 105			if (unlikely(bio->bi_status))
 106				f2fs_stop_checkpoint(sbi, true);
 107			continue;
 108		}
 109
 110		fscrypt_pullback_bio_page(&page, true);
 111
 112		if (unlikely(bio->bi_status)) {
 113			mapping_set_error(page->mapping, -EIO);
 114			if (type == F2FS_WB_CP_DATA)
 115				f2fs_stop_checkpoint(sbi, true);
 116		}
 117
 118		f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 119					page->index != nid_of_node(page));
 120
 121		dec_page_count(sbi, type);
 122		clear_cold_data(page);
 123		end_page_writeback(page);
 
 124	}
 125	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 126				wq_has_sleeper(&sbi->cp_wait))
 127		wake_up(&sbi->cp_wait);
 128
 129	bio_put(bio);
 130}
 131
 132/*
 133 * Return true, if pre_bio's bdev is same as its target device.
 134 */
 135struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 136				block_t blk_addr, struct bio *bio)
 137{
 138	struct block_device *bdev = sbi->sb->s_bdev;
 139	int i;
 140
 141	for (i = 0; i < sbi->s_ndevs; i++) {
 142		if (FDEV(i).start_blk <= blk_addr &&
 143					FDEV(i).end_blk >= blk_addr) {
 144			blk_addr -= FDEV(i).start_blk;
 145			bdev = FDEV(i).bdev;
 146			break;
 147		}
 148	}
 149	if (bio) {
 150		bio_set_dev(bio, bdev);
 151		bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
 152	}
 153	return bdev;
 154}
 155
 156int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 157{
 158	int i;
 159
 160	for (i = 0; i < sbi->s_ndevs; i++)
 161		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 162			return i;
 163	return 0;
 164}
 165
 166static bool __same_bdev(struct f2fs_sb_info *sbi,
 167				block_t blk_addr, struct bio *bio)
 168{
 169	struct block_device *b = f2fs_target_device(sbi, blk_addr, NULL);
 170	return bio->bi_disk == b->bd_disk && bio->bi_partno == b->bd_partno;
 171}
 172
 173/*
 174 * Low-level block read/write IO operations.
 175 */
 176static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
 177				struct writeback_control *wbc,
 178				int npages, bool is_read,
 179				enum page_type type, enum temp_type temp)
 180{
 181	struct bio *bio;
 182
 183	bio = f2fs_bio_alloc(sbi, npages, true);
 184
 185	f2fs_target_device(sbi, blk_addr, bio);
 186	if (is_read) {
 187		bio->bi_end_io = f2fs_read_end_io;
 188		bio->bi_private = NULL;
 189	} else {
 190		bio->bi_end_io = f2fs_write_end_io;
 191		bio->bi_private = sbi;
 192		bio->bi_write_hint = io_type_to_rw_hint(sbi, type, temp);
 193	}
 194	if (wbc)
 195		wbc_init_bio(wbc, bio);
 196
 197	return bio;
 198}
 199
 200static inline void __submit_bio(struct f2fs_sb_info *sbi,
 201				struct bio *bio, enum page_type type)
 202{
 203	if (!is_read_io(bio_op(bio))) {
 204		unsigned int start;
 205
 206		if (type != DATA && type != NODE)
 207			goto submit_io;
 208
 209		if (f2fs_sb_has_blkzoned(sbi->sb) && current->plug)
 210			blk_finish_plug(current->plug);
 211
 212		start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
 213		start %= F2FS_IO_SIZE(sbi);
 214
 215		if (start == 0)
 216			goto submit_io;
 217
 218		/* fill dummy pages */
 219		for (; start < F2FS_IO_SIZE(sbi); start++) {
 220			struct page *page =
 221				mempool_alloc(sbi->write_io_dummy,
 222					GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
 223			f2fs_bug_on(sbi, !page);
 224
 225			SetPagePrivate(page);
 226			set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
 227			lock_page(page);
 228			if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 229				f2fs_bug_on(sbi, 1);
 230		}
 231		/*
 232		 * In the NODE case, we lose next block address chain. So, we
 233		 * need to do checkpoint in f2fs_sync_file.
 234		 */
 235		if (type == NODE)
 236			set_sbi_flag(sbi, SBI_NEED_CP);
 237	}
 238submit_io:
 239	if (is_read_io(bio_op(bio)))
 240		trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 241	else
 242		trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 243	submit_bio(bio);
 244}
 245
 246static void __submit_merged_bio(struct f2fs_bio_info *io)
 247{
 248	struct f2fs_io_info *fio = &io->fio;
 249
 250	if (!io->bio)
 251		return;
 252
 253	bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
 254
 255	if (is_read_io(fio->op))
 256		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 257	else
 258		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 259
 260	__submit_bio(io->sbi, io->bio, fio->type);
 261	io->bio = NULL;
 262}
 263
 264static bool __has_merged_page(struct f2fs_bio_info *io,
 265				struct inode *inode, nid_t ino, pgoff_t idx)
 266{
 267	struct bio_vec *bvec;
 268	struct page *target;
 269	int i;
 270
 271	if (!io->bio)
 272		return false;
 273
 274	if (!inode && !ino)
 275		return true;
 276
 277	bio_for_each_segment_all(bvec, io->bio, i) {
 278
 279		if (bvec->bv_page->mapping)
 280			target = bvec->bv_page;
 281		else
 282			target = fscrypt_control_page(bvec->bv_page);
 283
 284		if (idx != target->index)
 285			continue;
 286
 287		if (inode && inode == target->mapping->host)
 288			return true;
 
 
 289		if (ino && ino == ino_of_node(target))
 290			return true;
 291	}
 292
 293	return false;
 294}
 295
 296static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
 297				nid_t ino, pgoff_t idx, enum page_type type)
 
 298{
 299	enum page_type btype = PAGE_TYPE_OF_BIO(type);
 300	enum temp_type temp;
 301	struct f2fs_bio_info *io;
 302	bool ret = false;
 303
 304	for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 305		io = sbi->write_io[btype] + temp;
 306
 307		down_read(&io->io_rwsem);
 308		ret = __has_merged_page(io, inode, ino, idx);
 309		up_read(&io->io_rwsem);
 310
 311		/* TODO: use HOT temp only for meta pages now. */
 312		if (ret || btype == META)
 313			break;
 314	}
 315	return ret;
 316}
 317
 318static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 319				enum page_type type, enum temp_type temp)
 
 320{
 321	enum page_type btype = PAGE_TYPE_OF_BIO(type);
 322	struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 
 
 323
 324	down_write(&io->io_rwsem);
 325
 
 
 
 326	/* change META to META_FLUSH in the checkpoint procedure */
 327	if (type >= META_FLUSH) {
 328		io->fio.type = META_FLUSH;
 329		io->fio.op = REQ_OP_WRITE;
 330		io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
 331		if (!test_opt(sbi, NOBARRIER))
 332			io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
 333	}
 334	__submit_merged_bio(io);
 
 335	up_write(&io->io_rwsem);
 336}
 337
 338static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 339				struct inode *inode, nid_t ino, pgoff_t idx,
 340				enum page_type type, bool force)
 341{
 342	enum temp_type temp;
 343
 344	if (!force && !has_merged_page(sbi, inode, ino, idx, type))
 345		return;
 346
 347	for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 348
 349		__f2fs_submit_merged_write(sbi, type, temp);
 350
 351		/* TODO: use HOT temp only for meta pages now. */
 352		if (type >= META)
 353			break;
 354	}
 355}
 356
 357void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 358{
 359	__submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
 360}
 361
 362void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 363				struct inode *inode, nid_t ino, pgoff_t idx,
 364				enum page_type type)
 365{
 366	__submit_merged_write_cond(sbi, inode, ino, idx, type, false);
 
 367}
 368
 369void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 370{
 371	f2fs_submit_merged_write(sbi, DATA);
 372	f2fs_submit_merged_write(sbi, NODE);
 373	f2fs_submit_merged_write(sbi, META);
 374}
 375
 376/*
 377 * Fill the locked page with data located in the block address.
 378 * A caller needs to unlock the page on failure.
 379 */
 380int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 381{
 382	struct bio *bio;
 383	struct page *page = fio->encrypted_page ?
 384			fio->encrypted_page : fio->page;
 385
 386	verify_block_addr(fio, fio->new_blkaddr);
 387	trace_f2fs_submit_page_bio(page, fio);
 388	f2fs_trace_ios(fio, 0);
 389
 390	/* Allocate a new bio */
 391	bio = __bio_alloc(fio->sbi, fio->new_blkaddr, fio->io_wbc,
 392				1, is_read_io(fio->op), fio->type, fio->temp);
 393
 394	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 395		bio_put(bio);
 396		return -EFAULT;
 397	}
 398	bio_set_op_attrs(bio, fio->op, fio->op_flags);
 399
 400	__submit_bio(fio->sbi, bio, fio->type);
 401
 402	if (!is_read_io(fio->op))
 403		inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
 404	return 0;
 405}
 406
 407int f2fs_submit_page_write(struct f2fs_io_info *fio)
 408{
 409	struct f2fs_sb_info *sbi = fio->sbi;
 410	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
 411	struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
 
 412	struct page *bio_page;
 413	int err = 0;
 414
 415	f2fs_bug_on(sbi, is_read_io(fio->op));
 416
 417	down_write(&io->io_rwsem);
 418next:
 419	if (fio->in_list) {
 420		spin_lock(&io->io_lock);
 421		if (list_empty(&io->io_list)) {
 422			spin_unlock(&io->io_lock);
 423			goto out_fail;
 424		}
 425		fio = list_first_entry(&io->io_list,
 426						struct f2fs_io_info, list);
 427		list_del(&fio->list);
 428		spin_unlock(&io->io_lock);
 429	}
 430
 431	if (fio->old_blkaddr != NEW_ADDR)
 432		verify_block_addr(fio, fio->old_blkaddr);
 433	verify_block_addr(fio, fio->new_blkaddr);
 434
 435	bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
 436
 437	/* set submitted = true as a return value */
 438	fio->submitted = true;
 439
 440	inc_page_count(sbi, WB_DATA_TYPE(bio_page));
 
 441
 442	if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
 443	    (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
 444			!__same_bdev(sbi, fio->new_blkaddr, io->bio)))
 445		__submit_merged_bio(io);
 446alloc_new:
 447	if (io->bio == NULL) {
 448		if ((fio->type == DATA || fio->type == NODE) &&
 449				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
 450			err = -EAGAIN;
 451			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
 452			goto out_fail;
 453		}
 454		io->bio = __bio_alloc(sbi, fio->new_blkaddr, fio->io_wbc,
 455						BIO_MAX_PAGES, false,
 456						fio->type, fio->temp);
 457		io->fio = *fio;
 458	}
 459
 460	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
 
 
 
 461		__submit_merged_bio(io);
 462		goto alloc_new;
 463	}
 464
 465	if (fio->io_wbc)
 466		wbc_account_io(fio->io_wbc, bio_page, PAGE_SIZE);
 467
 468	io->last_block_in_bio = fio->new_blkaddr;
 469	f2fs_trace_ios(fio, 0);
 470
 471	trace_f2fs_submit_page_write(fio->page, fio);
 472
 473	if (fio->in_list)
 474		goto next;
 475out_fail:
 476	up_write(&io->io_rwsem);
 477	return err;
 478}
 479
 480static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 481							 unsigned nr_pages)
 482{
 483	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 484	struct fscrypt_ctx *ctx = NULL;
 485	struct bio *bio;
 486
 487	if (f2fs_encrypted_file(inode)) {
 488		ctx = fscrypt_get_ctx(inode, GFP_NOFS);
 489		if (IS_ERR(ctx))
 490			return ERR_CAST(ctx);
 491
 492		/* wait the page to be moved by cleaning */
 493		f2fs_wait_on_block_writeback(sbi, blkaddr);
 494	}
 495
 496	bio = f2fs_bio_alloc(sbi, min_t(int, nr_pages, BIO_MAX_PAGES), false);
 497	if (!bio) {
 498		if (ctx)
 499			fscrypt_release_ctx(ctx);
 500		return ERR_PTR(-ENOMEM);
 501	}
 502	f2fs_target_device(sbi, blkaddr, bio);
 503	bio->bi_end_io = f2fs_read_end_io;
 504	bio->bi_private = ctx;
 505	bio_set_op_attrs(bio, REQ_OP_READ, 0);
 506
 507	return bio;
 508}
 509
 510/* This can handle encryption stuffs */
 511static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 512							block_t blkaddr)
 513{
 514	struct bio *bio = f2fs_grab_read_bio(inode, blkaddr, 1);
 515
 516	if (IS_ERR(bio))
 517		return PTR_ERR(bio);
 518
 519	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 520		bio_put(bio);
 521		return -EFAULT;
 522	}
 523	__submit_bio(F2FS_I_SB(inode), bio, DATA);
 524	return 0;
 525}
 526
 527static void __set_data_blkaddr(struct dnode_of_data *dn)
 528{
 529	struct f2fs_node *rn = F2FS_NODE(dn->node_page);
 530	__le32 *addr_array;
 531	int base = 0;
 532
 533	if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
 534		base = get_extra_isize(dn->inode);
 535
 536	/* Get physical address of data block */
 537	addr_array = blkaddr_in_node(rn);
 538	addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
 539}
 540
 541/*
 542 * Lock ordering for the change of data block address:
 543 * ->data_page
 544 *  ->node_page
 545 *    update block addresses in the node page
 546 */
 547void set_data_blkaddr(struct dnode_of_data *dn)
 548{
 549	f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 550	__set_data_blkaddr(dn);
 551	if (set_page_dirty(dn->node_page))
 
 
 
 
 
 
 
 
 
 
 552		dn->node_changed = true;
 553}
 554
 555void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
 556{
 557	dn->data_blkaddr = blkaddr;
 558	set_data_blkaddr(dn);
 559	f2fs_update_extent_cache(dn);
 560}
 561
 562/* dn->ofs_in_node will be returned with up-to-date last block pointer */
 563int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
 564{
 565	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 566	int err;
 567
 568	if (!count)
 569		return 0;
 570
 571	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 572		return -EPERM;
 573	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 574		return err;
 575
 576	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
 577						dn->ofs_in_node, count);
 578
 579	f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
 580
 581	for (; count > 0; dn->ofs_in_node++) {
 582		block_t blkaddr = datablock_addr(dn->inode,
 583					dn->node_page, dn->ofs_in_node);
 584		if (blkaddr == NULL_ADDR) {
 585			dn->data_blkaddr = NEW_ADDR;
 586			__set_data_blkaddr(dn);
 587			count--;
 588		}
 589	}
 590
 591	if (set_page_dirty(dn->node_page))
 592		dn->node_changed = true;
 593	return 0;
 594}
 595
 596/* Should keep dn->ofs_in_node unchanged */
 597int reserve_new_block(struct dnode_of_data *dn)
 598{
 599	unsigned int ofs_in_node = dn->ofs_in_node;
 600	int ret;
 601
 602	ret = reserve_new_blocks(dn, 1);
 603	dn->ofs_in_node = ofs_in_node;
 604	return ret;
 605}
 606
 607int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
 608{
 609	bool need_put = dn->inode_page ? false : true;
 610	int err;
 611
 612	err = get_dnode_of_data(dn, index, ALLOC_NODE);
 613	if (err)
 614		return err;
 615
 616	if (dn->data_blkaddr == NULL_ADDR)
 617		err = reserve_new_block(dn);
 618	if (err || need_put)
 619		f2fs_put_dnode(dn);
 620	return err;
 621}
 622
 623int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
 624{
 625	struct extent_info ei  = {0,0,0};
 626	struct inode *inode = dn->inode;
 627
 628	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 629		dn->data_blkaddr = ei.blk + index - ei.fofs;
 630		return 0;
 631	}
 632
 633	return f2fs_reserve_block(dn, index);
 634}
 635
 636struct page *get_read_data_page(struct inode *inode, pgoff_t index,
 637						int op_flags, bool for_write)
 638{
 639	struct address_space *mapping = inode->i_mapping;
 640	struct dnode_of_data dn;
 641	struct page *page;
 642	struct extent_info ei = {0,0,0};
 643	int err;
 
 
 
 
 
 
 
 
 
 644
 645	page = f2fs_grab_cache_page(mapping, index, for_write);
 646	if (!page)
 647		return ERR_PTR(-ENOMEM);
 648
 649	if (f2fs_lookup_extent_cache(inode, index, &ei)) {
 650		dn.data_blkaddr = ei.blk + index - ei.fofs;
 651		goto got_it;
 652	}
 653
 654	set_new_dnode(&dn, inode, NULL, NULL, 0);
 655	err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 656	if (err)
 657		goto put_err;
 658	f2fs_put_dnode(&dn);
 659
 660	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
 661		err = -ENOENT;
 662		goto put_err;
 663	}
 664got_it:
 665	if (PageUptodate(page)) {
 666		unlock_page(page);
 667		return page;
 668	}
 669
 670	/*
 671	 * A new dentry page is allocated but not able to be written, since its
 672	 * new inode page couldn't be allocated due to -ENOSPC.
 673	 * In such the case, its blkaddr can be remained as NEW_ADDR.
 674	 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
 675	 */
 676	if (dn.data_blkaddr == NEW_ADDR) {
 677		zero_user_segment(page, 0, PAGE_SIZE);
 678		if (!PageUptodate(page))
 679			SetPageUptodate(page);
 680		unlock_page(page);
 681		return page;
 682	}
 683
 684	err = f2fs_submit_page_read(inode, page, dn.data_blkaddr);
 
 
 685	if (err)
 686		goto put_err;
 687	return page;
 688
 689put_err:
 690	f2fs_put_page(page, 1);
 691	return ERR_PTR(err);
 692}
 693
 694struct page *find_data_page(struct inode *inode, pgoff_t index)
 695{
 696	struct address_space *mapping = inode->i_mapping;
 697	struct page *page;
 698
 699	page = find_get_page(mapping, index);
 700	if (page && PageUptodate(page))
 701		return page;
 702	f2fs_put_page(page, 0);
 703
 704	page = get_read_data_page(inode, index, 0, false);
 705	if (IS_ERR(page))
 706		return page;
 707
 708	if (PageUptodate(page))
 709		return page;
 710
 711	wait_on_page_locked(page);
 712	if (unlikely(!PageUptodate(page))) {
 713		f2fs_put_page(page, 0);
 714		return ERR_PTR(-EIO);
 715	}
 716	return page;
 717}
 718
 719/*
 720 * If it tries to access a hole, return an error.
 721 * Because, the callers, functions in dir.c and GC, should be able to know
 722 * whether this page exists or not.
 723 */
 724struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
 725							bool for_write)
 726{
 727	struct address_space *mapping = inode->i_mapping;
 728	struct page *page;
 729repeat:
 730	page = get_read_data_page(inode, index, 0, for_write);
 731	if (IS_ERR(page))
 732		return page;
 733
 734	/* wait for read completion */
 735	lock_page(page);
 736	if (unlikely(page->mapping != mapping)) {
 737		f2fs_put_page(page, 1);
 738		goto repeat;
 739	}
 740	if (unlikely(!PageUptodate(page))) {
 741		f2fs_put_page(page, 1);
 742		return ERR_PTR(-EIO);
 743	}
 
 
 
 
 744	return page;
 745}
 746
 747/*
 748 * Caller ensures that this data page is never allocated.
 749 * A new zero-filled data page is allocated in the page cache.
 750 *
 751 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
 752 * f2fs_unlock_op().
 753 * Note that, ipage is set only by make_empty_dir, and if any error occur,
 754 * ipage should be released by this function.
 755 */
 756struct page *get_new_data_page(struct inode *inode,
 757		struct page *ipage, pgoff_t index, bool new_i_size)
 758{
 759	struct address_space *mapping = inode->i_mapping;
 760	struct page *page;
 761	struct dnode_of_data dn;
 762	int err;
 763
 764	page = f2fs_grab_cache_page(mapping, index, true);
 765	if (!page) {
 766		/*
 767		 * before exiting, we should make sure ipage will be released
 768		 * if any error occur.
 769		 */
 770		f2fs_put_page(ipage, 1);
 771		return ERR_PTR(-ENOMEM);
 772	}
 773
 774	set_new_dnode(&dn, inode, ipage, NULL, 0);
 775	err = f2fs_reserve_block(&dn, index);
 776	if (err) {
 777		f2fs_put_page(page, 1);
 778		return ERR_PTR(err);
 779	}
 780	if (!ipage)
 781		f2fs_put_dnode(&dn);
 782
 783	if (PageUptodate(page))
 784		goto got_it;
 785
 786	if (dn.data_blkaddr == NEW_ADDR) {
 787		zero_user_segment(page, 0, PAGE_SIZE);
 788		if (!PageUptodate(page))
 789			SetPageUptodate(page);
 790	} else {
 791		f2fs_put_page(page, 1);
 792
 793		/* if ipage exists, blkaddr should be NEW_ADDR */
 794		f2fs_bug_on(F2FS_I_SB(inode), ipage);
 795		page = get_lock_data_page(inode, index, true);
 796		if (IS_ERR(page))
 797			return page;
 798	}
 799got_it:
 800	if (new_i_size && i_size_read(inode) <
 801				((loff_t)(index + 1) << PAGE_SHIFT))
 802		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
 
 
 
 803	return page;
 804}
 805
 806static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
 807{
 808	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 809	struct f2fs_summary sum;
 810	struct node_info ni;
 
 811	pgoff_t fofs;
 812	blkcnt_t count = 1;
 813	int err;
 814
 815	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
 816		return -EPERM;
 817
 818	dn->data_blkaddr = datablock_addr(dn->inode,
 819				dn->node_page, dn->ofs_in_node);
 820	if (dn->data_blkaddr == NEW_ADDR)
 821		goto alloc;
 822
 823	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
 824		return err;
 825
 826alloc:
 827	get_node_info(sbi, dn->nid, &ni);
 828	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
 829
 
 
 
 830	allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
 831					&sum, seg_type, NULL, false);
 832	set_data_blkaddr(dn);
 833
 834	/* update i_size */
 835	fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
 836							dn->ofs_in_node;
 837	if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
 838		f2fs_i_size_write(dn->inode,
 839				((loff_t)(fofs + 1) << PAGE_SHIFT));
 840	return 0;
 841}
 842
 843int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
 844{
 845	struct inode *inode = file_inode(iocb->ki_filp);
 846	struct f2fs_map_blocks map;
 847	int flag;
 848	int err = 0;
 849	bool direct_io = iocb->ki_flags & IOCB_DIRECT;
 850
 851	/* convert inline data for Direct I/O*/
 852	if (direct_io) {
 853		err = f2fs_convert_inline_inode(inode);
 854		if (err)
 855			return err;
 856	}
 857
 858	if (is_inode_flag_set(inode, FI_NO_PREALLOC))
 859		return 0;
 860
 861	map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
 862	map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
 863	if (map.m_len > map.m_lblk)
 864		map.m_len -= map.m_lblk;
 865	else
 866		map.m_len = 0;
 867
 
 
 868	map.m_next_pgofs = NULL;
 869	map.m_next_extent = NULL;
 870	map.m_seg_type = NO_CHECK_TYPE;
 871
 872	if (direct_io) {
 873		map.m_seg_type = rw_hint_to_seg_type(iocb->ki_hint);
 874		flag = f2fs_force_buffered_io(inode, WRITE) ?
 875					F2FS_GET_BLOCK_PRE_AIO :
 876					F2FS_GET_BLOCK_PRE_DIO;
 877		goto map_blocks;
 878	}
 879	if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
 880		err = f2fs_convert_inline_inode(inode);
 881		if (err)
 882			return err;
 883	}
 884	if (f2fs_has_inline_data(inode))
 885		return err;
 886
 887	flag = F2FS_GET_BLOCK_PRE_AIO;
 888
 889map_blocks:
 890	err = f2fs_map_blocks(inode, &map, 1, flag);
 891	if (map.m_len > 0 && err == -ENOSPC) {
 892		if (!direct_io)
 893			set_inode_flag(inode, FI_NO_PREALLOC);
 894		err = 0;
 895	}
 896	return err;
 897}
 898
 899static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
 900{
 901	if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 902		if (lock)
 903			down_read(&sbi->node_change);
 904		else
 905			up_read(&sbi->node_change);
 906	} else {
 907		if (lock)
 908			f2fs_lock_op(sbi);
 909		else
 910			f2fs_unlock_op(sbi);
 911	}
 
 
 
 912}
 913
 914/*
 915 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
 916 * f2fs_map_blocks structure.
 917 * If original data blocks are allocated, then give them to blockdev.
 918 * Otherwise,
 919 *     a. preallocate requested block addresses
 920 *     b. do not use extent cache for better performance
 921 *     c. give the block addresses to blockdev
 922 */
 923int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
 924						int create, int flag)
 925{
 926	unsigned int maxblocks = map->m_len;
 927	struct dnode_of_data dn;
 928	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 929	int mode = create ? ALLOC_NODE : LOOKUP_NODE;
 930	pgoff_t pgofs, end_offset, end;
 931	int err = 0, ofs = 1;
 932	unsigned int ofs_in_node, last_ofs_in_node;
 933	blkcnt_t prealloc;
 934	struct extent_info ei = {0,0,0};
 935	block_t blkaddr;
 936	unsigned int start_pgofs;
 937
 938	if (!maxblocks)
 939		return 0;
 940
 941	map->m_len = 0;
 942	map->m_flags = 0;
 943
 944	/* it only supports block size == page size */
 945	pgofs =	(pgoff_t)map->m_lblk;
 946	end = pgofs + maxblocks;
 947
 948	if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
 949		map->m_pblk = ei.blk + pgofs - ei.fofs;
 950		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
 951		map->m_flags = F2FS_MAP_MAPPED;
 952		if (map->m_next_extent)
 953			*map->m_next_extent = pgofs + map->m_len;
 954		goto out;
 955	}
 956
 957next_dnode:
 958	if (create)
 959		__do_map_lock(sbi, flag, true);
 960
 961	/* When reading holes, we need its node page */
 962	set_new_dnode(&dn, inode, NULL, NULL, 0);
 963	err = get_dnode_of_data(&dn, pgofs, mode);
 964	if (err) {
 965		if (flag == F2FS_GET_BLOCK_BMAP)
 966			map->m_pblk = 0;
 967		if (err == -ENOENT) {
 968			err = 0;
 969			if (map->m_next_pgofs)
 970				*map->m_next_pgofs =
 971					get_next_page_offset(&dn, pgofs);
 972			if (map->m_next_extent)
 973				*map->m_next_extent =
 974					get_next_page_offset(&dn, pgofs);
 975		}
 976		goto unlock_out;
 977	}
 978
 979	start_pgofs = pgofs;
 980	prealloc = 0;
 981	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
 982	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
 983
 984next_block:
 985	blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
 986
 987	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
 988		if (create) {
 989			if (unlikely(f2fs_cp_error(sbi))) {
 990				err = -EIO;
 991				goto sync_out;
 992			}
 993			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
 994				if (blkaddr == NULL_ADDR) {
 995					prealloc++;
 996					last_ofs_in_node = dn.ofs_in_node;
 997				}
 998			} else {
 999				err = __allocate_data_block(&dn,
1000							map->m_seg_type);
1001				if (!err)
1002					set_inode_flag(inode, FI_APPEND_WRITE);
1003			}
1004			if (err)
1005				goto sync_out;
1006			map->m_flags |= F2FS_MAP_NEW;
 
1007			blkaddr = dn.data_blkaddr;
1008		} else {
1009			if (flag == F2FS_GET_BLOCK_BMAP) {
1010				map->m_pblk = 0;
1011				goto sync_out;
1012			}
1013			if (flag == F2FS_GET_BLOCK_PRECACHE)
1014				goto sync_out;
1015			if (flag == F2FS_GET_BLOCK_FIEMAP &&
1016						blkaddr == NULL_ADDR) {
1017				if (map->m_next_pgofs)
1018					*map->m_next_pgofs = pgofs + 1;
1019				goto sync_out;
1020			}
1021			if (flag != F2FS_GET_BLOCK_FIEMAP) {
1022				/* for defragment case */
1023				if (map->m_next_pgofs)
1024					*map->m_next_pgofs = pgofs + 1;
1025				goto sync_out;
1026			}
1027		}
1028	}
1029
1030	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1031		goto skip;
1032
1033	if (map->m_len == 0) {
1034		/* preallocated unwritten block should be mapped for fiemap. */
1035		if (blkaddr == NEW_ADDR)
1036			map->m_flags |= F2FS_MAP_UNWRITTEN;
1037		map->m_flags |= F2FS_MAP_MAPPED;
1038
1039		map->m_pblk = blkaddr;
1040		map->m_len = 1;
1041	} else if ((map->m_pblk != NEW_ADDR &&
1042			blkaddr == (map->m_pblk + ofs)) ||
1043			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1044			flag == F2FS_GET_BLOCK_PRE_DIO) {
 
1045		ofs++;
1046		map->m_len++;
1047	} else {
1048		goto sync_out;
1049	}
1050
1051skip:
1052	dn.ofs_in_node++;
1053	pgofs++;
1054
1055	/* preallocate blocks in batch for one dnode page */
1056	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1057			(pgofs == end || dn.ofs_in_node == end_offset)) {
1058
1059		dn.ofs_in_node = ofs_in_node;
1060		err = reserve_new_blocks(&dn, prealloc);
1061		if (err)
1062			goto sync_out;
1063
1064		map->m_len += dn.ofs_in_node - ofs_in_node;
1065		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1066			err = -ENOSPC;
1067			goto sync_out;
1068		}
1069		dn.ofs_in_node = end_offset;
1070	}
1071
1072	if (pgofs >= end)
1073		goto sync_out;
1074	else if (dn.ofs_in_node < end_offset)
1075		goto next_block;
1076
1077	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1078		if (map->m_flags & F2FS_MAP_MAPPED) {
1079			unsigned int ofs = start_pgofs - map->m_lblk;
1080
1081			f2fs_update_extent_cache_range(&dn,
1082				start_pgofs, map->m_pblk + ofs,
1083				map->m_len - ofs);
1084		}
 
 
1085	}
1086
1087	f2fs_put_dnode(&dn);
1088
1089	if (create) {
1090		__do_map_lock(sbi, flag, false);
1091		f2fs_balance_fs(sbi, dn.node_changed);
1092	}
1093	goto next_dnode;
1094
1095sync_out:
1096	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1097		if (map->m_flags & F2FS_MAP_MAPPED) {
1098			unsigned int ofs = start_pgofs - map->m_lblk;
1099
1100			f2fs_update_extent_cache_range(&dn,
1101				start_pgofs, map->m_pblk + ofs,
1102				map->m_len - ofs);
1103		}
1104		if (map->m_next_extent)
1105			*map->m_next_extent = pgofs + 1;
1106	}
1107	f2fs_put_dnode(&dn);
1108unlock_out:
1109	if (create) {
1110		__do_map_lock(sbi, flag, false);
1111		f2fs_balance_fs(sbi, dn.node_changed);
1112	}
1113out:
1114	trace_f2fs_map_blocks(inode, map, err);
1115	return err;
1116}
1117
1118bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1119{
1120	struct f2fs_map_blocks map;
1121	block_t last_lblk;
1122	int err;
1123
1124	if (pos + len > i_size_read(inode))
1125		return false;
1126
1127	map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1128	map.m_next_pgofs = NULL;
1129	map.m_next_extent = NULL;
1130	map.m_seg_type = NO_CHECK_TYPE;
1131	last_lblk = F2FS_BLK_ALIGN(pos + len);
1132
1133	while (map.m_lblk < last_lblk) {
1134		map.m_len = last_lblk - map.m_lblk;
1135		err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_DEFAULT);
1136		if (err || map.m_len == 0)
1137			return false;
1138		map.m_lblk += map.m_len;
1139	}
1140	return true;
1141}
1142
1143static int __get_data_block(struct inode *inode, sector_t iblock,
1144			struct buffer_head *bh, int create, int flag,
1145			pgoff_t *next_pgofs, int seg_type)
1146{
1147	struct f2fs_map_blocks map;
1148	int err;
1149
1150	map.m_lblk = iblock;
1151	map.m_len = bh->b_size >> inode->i_blkbits;
1152	map.m_next_pgofs = next_pgofs;
1153	map.m_next_extent = NULL;
1154	map.m_seg_type = seg_type;
1155
1156	err = f2fs_map_blocks(inode, &map, create, flag);
1157	if (!err) {
1158		map_bh(bh, inode->i_sb, map.m_pblk);
1159		bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1160		bh->b_size = (u64)map.m_len << inode->i_blkbits;
1161	}
1162	return err;
1163}
1164
1165static int get_data_block(struct inode *inode, sector_t iblock,
1166			struct buffer_head *bh_result, int create, int flag,
1167			pgoff_t *next_pgofs)
1168{
1169	return __get_data_block(inode, iblock, bh_result, create,
1170							flag, next_pgofs,
1171							NO_CHECK_TYPE);
1172}
1173
1174static int get_data_block_dio(struct inode *inode, sector_t iblock,
1175			struct buffer_head *bh_result, int create)
1176{
1177	return __get_data_block(inode, iblock, bh_result, create,
1178						F2FS_GET_BLOCK_DEFAULT, NULL,
1179						rw_hint_to_seg_type(
1180							inode->i_write_hint));
1181}
1182
1183static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1184			struct buffer_head *bh_result, int create)
1185{
1186	/* Block number less than F2FS MAX BLOCKS */
1187	if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1188		return -EFBIG;
1189
1190	return __get_data_block(inode, iblock, bh_result, create,
1191						F2FS_GET_BLOCK_BMAP, NULL,
1192						NO_CHECK_TYPE);
1193}
1194
1195static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1196{
1197	return (offset >> inode->i_blkbits);
1198}
1199
1200static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1201{
1202	return (blk << inode->i_blkbits);
1203}
1204
1205static int f2fs_xattr_fiemap(struct inode *inode,
1206				struct fiemap_extent_info *fieinfo)
1207{
1208	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1209	struct page *page;
1210	struct node_info ni;
1211	__u64 phys = 0, len;
1212	__u32 flags;
1213	nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1214	int err = 0;
1215
1216	if (f2fs_has_inline_xattr(inode)) {
1217		int offset;
1218
1219		page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1220						inode->i_ino, false);
1221		if (!page)
1222			return -ENOMEM;
1223
1224		get_node_info(sbi, inode->i_ino, &ni);
1225
1226		phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1227		offset = offsetof(struct f2fs_inode, i_addr) +
1228					sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1229					get_inline_xattr_addrs(inode));
1230
1231		phys += offset;
1232		len = inline_xattr_size(inode);
1233
1234		f2fs_put_page(page, 1);
1235
1236		flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1237
1238		if (!xnid)
1239			flags |= FIEMAP_EXTENT_LAST;
1240
1241		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1242		if (err || err == 1)
1243			return err;
1244	}
1245
1246	if (xnid) {
1247		page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1248		if (!page)
1249			return -ENOMEM;
1250
1251		get_node_info(sbi, xnid, &ni);
1252
1253		phys = (__u64)blk_to_logical(inode, ni.blk_addr);
1254		len = inode->i_sb->s_blocksize;
1255
1256		f2fs_put_page(page, 1);
1257
1258		flags = FIEMAP_EXTENT_LAST;
1259	}
1260
1261	if (phys)
1262		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1263
1264	return (err < 0 ? err : 0);
1265}
1266
1267int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1268		u64 start, u64 len)
1269{
1270	struct buffer_head map_bh;
1271	sector_t start_blk, last_blk;
1272	pgoff_t next_pgofs;
 
1273	u64 logical = 0, phys = 0, size = 0;
1274	u32 flags = 0;
1275	int ret = 0;
1276
1277	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1278		ret = f2fs_precache_extents(inode);
1279		if (ret)
1280			return ret;
1281	}
1282
1283	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR);
1284	if (ret)
1285		return ret;
1286
1287	inode_lock(inode);
1288
1289	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1290		ret = f2fs_xattr_fiemap(inode, fieinfo);
1291		goto out;
1292	}
1293
1294	if (f2fs_has_inline_data(inode)) {
1295		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1296		if (ret != -EAGAIN)
1297			goto out;
1298	}
1299
 
 
 
 
 
 
 
 
 
1300	if (logical_to_blk(inode, len) == 0)
1301		len = blk_to_logical(inode, 1);
1302
1303	start_blk = logical_to_blk(inode, start);
1304	last_blk = logical_to_blk(inode, start + len - 1);
1305
1306next:
1307	memset(&map_bh, 0, sizeof(struct buffer_head));
1308	map_bh.b_size = len;
1309
1310	ret = get_data_block(inode, start_blk, &map_bh, 0,
1311					F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1312	if (ret)
1313		goto out;
1314
1315	/* HOLE */
1316	if (!buffer_mapped(&map_bh)) {
1317		start_blk = next_pgofs;
1318
1319		if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1320					F2FS_I_SB(inode)->max_file_blocks))
1321			goto prep_next;
1322
 
 
 
1323		flags |= FIEMAP_EXTENT_LAST;
1324	}
1325
1326	if (size) {
1327		if (f2fs_encrypted_inode(inode))
1328			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1329
1330		ret = fiemap_fill_next_extent(fieinfo, logical,
1331				phys, size, flags);
1332	}
1333
1334	if (start_blk > last_blk || ret)
1335		goto out;
1336
1337	logical = blk_to_logical(inode, start_blk);
1338	phys = blk_to_logical(inode, map_bh.b_blocknr);
1339	size = map_bh.b_size;
1340	flags = 0;
1341	if (buffer_unwritten(&map_bh))
1342		flags = FIEMAP_EXTENT_UNWRITTEN;
1343
1344	start_blk += logical_to_blk(inode, size);
1345
1346prep_next:
1347	cond_resched();
1348	if (fatal_signal_pending(current))
1349		ret = -EINTR;
1350	else
1351		goto next;
1352out:
1353	if (ret == 1)
1354		ret = 0;
1355
1356	inode_unlock(inode);
1357	return ret;
1358}
1359
1360/*
1361 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1362 * Major change was from block_size == page_size in f2fs by default.
1363 */
1364static int f2fs_mpage_readpages(struct address_space *mapping,
1365			struct list_head *pages, struct page *page,
1366			unsigned nr_pages)
1367{
1368	struct bio *bio = NULL;
 
1369	sector_t last_block_in_bio = 0;
1370	struct inode *inode = mapping->host;
1371	const unsigned blkbits = inode->i_blkbits;
1372	const unsigned blocksize = 1 << blkbits;
1373	sector_t block_in_file;
1374	sector_t last_block;
1375	sector_t last_block_in_file;
1376	sector_t block_nr;
 
1377	struct f2fs_map_blocks map;
1378
1379	map.m_pblk = 0;
1380	map.m_lblk = 0;
1381	map.m_len = 0;
1382	map.m_flags = 0;
1383	map.m_next_pgofs = NULL;
1384	map.m_next_extent = NULL;
1385	map.m_seg_type = NO_CHECK_TYPE;
1386
1387	for (; nr_pages; nr_pages--) {
1388		if (pages) {
1389			page = list_last_entry(pages, struct page, lru);
1390
1391			prefetchw(&page->flags);
 
 
1392			list_del(&page->lru);
1393			if (add_to_page_cache_lru(page, mapping,
1394						  page->index,
1395						  readahead_gfp_mask(mapping)))
1396				goto next_page;
1397		}
1398
1399		block_in_file = (sector_t)page->index;
1400		last_block = block_in_file + nr_pages;
1401		last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1402								blkbits;
1403		if (last_block > last_block_in_file)
1404			last_block = last_block_in_file;
1405
1406		/*
1407		 * Map blocks using the previous result first.
1408		 */
1409		if ((map.m_flags & F2FS_MAP_MAPPED) &&
1410				block_in_file > map.m_lblk &&
1411				block_in_file < (map.m_lblk + map.m_len))
1412			goto got_it;
1413
1414		/*
1415		 * Then do more f2fs_map_blocks() calls until we are
1416		 * done with this page.
1417		 */
1418		map.m_flags = 0;
1419
1420		if (block_in_file < last_block) {
1421			map.m_lblk = block_in_file;
1422			map.m_len = last_block - block_in_file;
1423
1424			if (f2fs_map_blocks(inode, &map, 0,
1425						F2FS_GET_BLOCK_DEFAULT))
1426				goto set_error_page;
1427		}
1428got_it:
1429		if ((map.m_flags & F2FS_MAP_MAPPED)) {
1430			block_nr = map.m_pblk + block_in_file - map.m_lblk;
1431			SetPageMappedToDisk(page);
1432
1433			if (!PageUptodate(page) && !cleancache_get_page(page)) {
1434				SetPageUptodate(page);
1435				goto confused;
1436			}
1437		} else {
1438			zero_user_segment(page, 0, PAGE_SIZE);
1439			if (!PageUptodate(page))
1440				SetPageUptodate(page);
1441			unlock_page(page);
1442			goto next_page;
1443		}
1444
1445		/*
1446		 * This page will go to BIO.  Do we need to send this
1447		 * BIO off first?
1448		 */
1449		if (bio && (last_block_in_bio != block_nr - 1 ||
1450			!__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1451submit_and_realloc:
1452			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1453			bio = NULL;
1454		}
1455		if (bio == NULL) {
1456			bio = f2fs_grab_read_bio(inode, block_nr, nr_pages);
1457			if (IS_ERR(bio)) {
1458				bio = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1459				goto set_error_page;
1460			}
 
 
 
 
1461		}
1462
1463		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1464			goto submit_and_realloc;
1465
1466		last_block_in_bio = block_nr;
1467		goto next_page;
1468set_error_page:
1469		SetPageError(page);
1470		zero_user_segment(page, 0, PAGE_SIZE);
1471		unlock_page(page);
1472		goto next_page;
1473confused:
1474		if (bio) {
1475			__submit_bio(F2FS_I_SB(inode), bio, DATA);
1476			bio = NULL;
1477		}
1478		unlock_page(page);
1479next_page:
1480		if (pages)
1481			put_page(page);
1482	}
1483	BUG_ON(pages && !list_empty(pages));
1484	if (bio)
1485		__submit_bio(F2FS_I_SB(inode), bio, DATA);
1486	return 0;
1487}
1488
1489static int f2fs_read_data_page(struct file *file, struct page *page)
1490{
1491	struct inode *inode = page->mapping->host;
1492	int ret = -EAGAIN;
1493
1494	trace_f2fs_readpage(page, DATA);
1495
1496	/* If the file has inline data, try to read it directly */
1497	if (f2fs_has_inline_data(inode))
1498		ret = f2fs_read_inline_data(inode, page);
1499	if (ret == -EAGAIN)
1500		ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1501	return ret;
1502}
1503
1504static int f2fs_read_data_pages(struct file *file,
1505			struct address_space *mapping,
1506			struct list_head *pages, unsigned nr_pages)
1507{
1508	struct inode *inode = mapping->host;
1509	struct page *page = list_last_entry(pages, struct page, lru);
1510
1511	trace_f2fs_readpages(inode, page, nr_pages);
1512
1513	/* If the file has inline data, skip readpages */
1514	if (f2fs_has_inline_data(inode))
1515		return 0;
1516
1517	return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1518}
1519
1520static int encrypt_one_page(struct f2fs_io_info *fio)
1521{
1522	struct inode *inode = fio->page->mapping->host;
1523	gfp_t gfp_flags = GFP_NOFS;
1524
1525	if (!f2fs_encrypted_file(inode))
1526		return 0;
1527
1528	/* wait for GCed encrypted page writeback */
1529	f2fs_wait_on_block_writeback(fio->sbi, fio->old_blkaddr);
1530
1531retry_encrypt:
1532	fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1533			PAGE_SIZE, 0, fio->page->index, gfp_flags);
1534	if (!IS_ERR(fio->encrypted_page))
1535		return 0;
1536
1537	/* flush pending IOs and wait for a while in the ENOMEM case */
1538	if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1539		f2fs_flush_merged_writes(fio->sbi);
1540		congestion_wait(BLK_RW_ASYNC, HZ/50);
1541		gfp_flags |= __GFP_NOFAIL;
1542		goto retry_encrypt;
1543	}
1544	return PTR_ERR(fio->encrypted_page);
1545}
1546
1547static inline bool check_inplace_update_policy(struct inode *inode,
1548				struct f2fs_io_info *fio)
1549{
1550	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1551	unsigned int policy = SM_I(sbi)->ipu_policy;
1552
1553	if (policy & (0x1 << F2FS_IPU_FORCE))
1554		return true;
1555	if (policy & (0x1 << F2FS_IPU_SSR) && need_SSR(sbi))
1556		return true;
1557	if (policy & (0x1 << F2FS_IPU_UTIL) &&
1558			utilization(sbi) > SM_I(sbi)->min_ipu_util)
1559		return true;
1560	if (policy & (0x1 << F2FS_IPU_SSR_UTIL) && need_SSR(sbi) &&
1561			utilization(sbi) > SM_I(sbi)->min_ipu_util)
1562		return true;
1563
1564	/*
1565	 * IPU for rewrite async pages
1566	 */
1567	if (policy & (0x1 << F2FS_IPU_ASYNC) &&
1568			fio && fio->op == REQ_OP_WRITE &&
1569			!(fio->op_flags & REQ_SYNC) &&
1570			!f2fs_encrypted_inode(inode))
1571		return true;
1572
1573	/* this is only set during fdatasync */
1574	if (policy & (0x1 << F2FS_IPU_FSYNC) &&
1575			is_inode_flag_set(inode, FI_NEED_IPU))
1576		return true;
1577
1578	return false;
1579}
1580
1581bool should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
1582{
1583	if (f2fs_is_pinned_file(inode))
1584		return true;
1585
1586	/* if this is cold file, we should overwrite to avoid fragmentation */
1587	if (file_is_cold(inode))
1588		return true;
1589
1590	return check_inplace_update_policy(inode, fio);
1591}
1592
1593bool should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
1594{
1595	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1596
1597	if (test_opt(sbi, LFS))
1598		return true;
1599	if (S_ISDIR(inode->i_mode))
1600		return true;
1601	if (f2fs_is_atomic_file(inode))
1602		return true;
1603	if (fio) {
1604		if (is_cold_data(fio->page))
1605			return true;
1606		if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1607			return true;
1608	}
1609	return false;
1610}
1611
1612static inline bool need_inplace_update(struct f2fs_io_info *fio)
1613{
1614	struct inode *inode = fio->page->mapping->host;
1615
1616	if (should_update_outplace(inode, fio))
1617		return false;
1618
1619	return should_update_inplace(inode, fio);
1620}
1621
1622static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1623{
1624	if (fio->old_blkaddr == NEW_ADDR)
1625		return false;
1626	if (fio->old_blkaddr == NULL_ADDR)
1627		return false;
1628	return true;
1629}
1630
1631int do_write_data_page(struct f2fs_io_info *fio)
1632{
1633	struct page *page = fio->page;
1634	struct inode *inode = page->mapping->host;
1635	struct dnode_of_data dn;
1636	struct extent_info ei = {0,0,0};
1637	bool ipu_force = false;
1638	int err = 0;
1639
1640	set_new_dnode(&dn, inode, NULL, NULL, 0);
1641	if (need_inplace_update(fio) &&
1642			f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1643		fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1644
1645		if (valid_ipu_blkaddr(fio)) {
1646			ipu_force = true;
1647			fio->need_lock = LOCK_DONE;
1648			goto got_it;
1649		}
1650	}
1651
1652	/* Deadlock due to between page->lock and f2fs_lock_op */
1653	if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1654		return -EAGAIN;
1655
1656	err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1657	if (err)
1658		goto out;
1659
1660	fio->old_blkaddr = dn.data_blkaddr;
1661
1662	/* This page is already truncated */
1663	if (fio->old_blkaddr == NULL_ADDR) {
1664		ClearPageUptodate(page);
1665		goto out_writepage;
1666	}
1667got_it:
1668	/*
1669	 * If current allocation needs SSR,
1670	 * it had better in-place writes for updated data.
1671	 */
1672	if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1673		err = encrypt_one_page(fio);
1674		if (err)
1675			goto out_writepage;
1676
1677		set_page_writeback(page);
1678		f2fs_put_dnode(&dn);
1679		if (fio->need_lock == LOCK_REQ)
1680			f2fs_unlock_op(fio->sbi);
1681		err = rewrite_data_page(fio);
1682		trace_f2fs_do_write_data_page(fio->page, IPU);
1683		set_inode_flag(inode, FI_UPDATE_WRITE);
1684		return err;
1685	}
1686
1687	if (fio->need_lock == LOCK_RETRY) {
1688		if (!f2fs_trylock_op(fio->sbi)) {
1689			err = -EAGAIN;
 
 
 
 
 
 
 
 
 
 
 
 
 
1690			goto out_writepage;
1691		}
1692		fio->need_lock = LOCK_REQ;
1693	}
1694
1695	err = encrypt_one_page(fio);
1696	if (err)
1697		goto out_writepage;
1698
1699	set_page_writeback(page);
1700
1701	/* LFS mode write path */
1702	write_data_page(&dn, fio);
1703	trace_f2fs_do_write_data_page(page, OPU);
1704	set_inode_flag(inode, FI_APPEND_WRITE);
1705	if (page->index == 0)
1706		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 
 
 
 
 
 
 
 
 
 
 
 
1707out_writepage:
1708	f2fs_put_dnode(&dn);
1709out:
1710	if (fio->need_lock == LOCK_REQ)
1711		f2fs_unlock_op(fio->sbi);
1712	return err;
1713}
1714
1715static int __write_data_page(struct page *page, bool *submitted,
1716				struct writeback_control *wbc,
1717				enum iostat_type io_type)
1718{
1719	struct inode *inode = page->mapping->host;
1720	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1721	loff_t i_size = i_size_read(inode);
1722	const pgoff_t end_index = ((unsigned long long) i_size)
1723							>> PAGE_SHIFT;
1724	loff_t psize = (page->index + 1) << PAGE_SHIFT;
1725	unsigned offset = 0;
1726	bool need_balance_fs = false;
1727	int err = 0;
1728	struct f2fs_io_info fio = {
1729		.sbi = sbi,
1730		.ino = inode->i_ino,
1731		.type = DATA,
1732		.op = REQ_OP_WRITE,
1733		.op_flags = wbc_to_write_flags(wbc),
1734		.old_blkaddr = NULL_ADDR,
1735		.page = page,
1736		.encrypted_page = NULL,
1737		.submitted = false,
1738		.need_lock = LOCK_RETRY,
1739		.io_type = io_type,
1740		.io_wbc = wbc,
1741	};
1742
1743	trace_f2fs_writepage(page, DATA);
1744
1745	/* we should bypass data pages to proceed the kworkder jobs */
1746	if (unlikely(f2fs_cp_error(sbi))) {
1747		mapping_set_error(page->mapping, -EIO);
1748		goto out;
1749	}
1750
1751	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1752		goto redirty_out;
1753
1754	if (page->index < end_index)
1755		goto write;
1756
1757	/*
1758	 * If the offset is out-of-range of file size,
1759	 * this page does not have to be written to disk.
1760	 */
1761	offset = i_size & (PAGE_SIZE - 1);
1762	if ((page->index >= end_index + 1) || !offset)
1763		goto out;
1764
1765	zero_user_segment(page, offset, PAGE_SIZE);
1766write:
 
 
1767	if (f2fs_is_drop_cache(inode))
1768		goto out;
1769	/* we should not write 0'th page having journal header */
1770	if (f2fs_is_volatile_file(inode) && (!page->index ||
1771			(!wbc->for_reclaim &&
1772			available_free_memory(sbi, BASE_CHECK))))
1773		goto redirty_out;
1774
1775	/* Dentry blocks are controlled by checkpoint */
1776	if (S_ISDIR(inode->i_mode)) {
1777		fio.need_lock = LOCK_DONE;
 
1778		err = do_write_data_page(&fio);
1779		goto done;
1780	}
1781
 
 
 
 
 
 
1782	if (!wbc->for_reclaim)
1783		need_balance_fs = true;
1784	else if (has_not_enough_free_secs(sbi, 0, 0))
1785		goto redirty_out;
1786	else
1787		set_inode_flag(inode, FI_HOT_DATA);
1788
1789	err = -EAGAIN;
1790	if (f2fs_has_inline_data(inode)) {
 
1791		err = f2fs_write_inline_data(inode, page);
1792		if (!err)
1793			goto out;
1794	}
1795
1796	if (err == -EAGAIN) {
1797		err = do_write_data_page(&fio);
1798		if (err == -EAGAIN) {
1799			fio.need_lock = LOCK_REQ;
1800			err = do_write_data_page(&fio);
1801		}
1802	}
1803
1804	if (err) {
1805		file_set_keep_isize(inode);
1806	} else {
1807		down_write(&F2FS_I(inode)->i_sem);
1808		if (F2FS_I(inode)->last_disk_size < psize)
1809			F2FS_I(inode)->last_disk_size = psize;
1810		up_write(&F2FS_I(inode)->i_sem);
1811	}
1812
1813done:
1814	if (err && err != -ENOENT)
1815		goto redirty_out;
1816
 
1817out:
1818	inode_dec_dirty_pages(inode);
1819	if (err)
1820		ClearPageUptodate(page);
1821
1822	if (wbc->for_reclaim) {
1823		f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1824		clear_inode_flag(inode, FI_HOT_DATA);
1825		remove_dirty_inode(inode);
1826		submitted = NULL;
1827	}
1828
1829	unlock_page(page);
1830	if (!S_ISDIR(inode->i_mode))
1831		f2fs_balance_fs(sbi, need_balance_fs);
1832
1833	if (unlikely(f2fs_cp_error(sbi))) {
1834		f2fs_submit_merged_write(sbi, DATA);
1835		submitted = NULL;
1836	}
1837
1838	if (submitted)
1839		*submitted = fio.submitted;
1840
1841	return 0;
1842
1843redirty_out:
1844	redirty_page_for_writepage(wbc, page);
1845	if (!err)
1846		return AOP_WRITEPAGE_ACTIVATE;
1847	unlock_page(page);
1848	return err;
1849}
1850
1851static int f2fs_write_data_page(struct page *page,
1852					struct writeback_control *wbc)
1853{
1854	return __write_data_page(page, NULL, wbc, FS_DATA_IO);
 
 
 
1855}
1856
1857/*
1858 * This function was copied from write_cche_pages from mm/page-writeback.c.
1859 * The major change is making write step of cold data page separately from
1860 * warm/hot data page.
1861 */
1862static int f2fs_write_cache_pages(struct address_space *mapping,
1863					struct writeback_control *wbc,
1864					enum iostat_type io_type)
1865{
1866	int ret = 0;
1867	int done = 0;
1868	struct pagevec pvec;
1869	int nr_pages;
1870	pgoff_t uninitialized_var(writeback_index);
1871	pgoff_t index;
1872	pgoff_t end;		/* Inclusive */
1873	pgoff_t done_index;
1874	pgoff_t last_idx = ULONG_MAX;
1875	int cycled;
1876	int range_whole = 0;
1877	int tag;
 
1878
1879	pagevec_init(&pvec);
1880
1881	if (get_dirty_pages(mapping->host) <=
1882				SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1883		set_inode_flag(mapping->host, FI_HOT_DATA);
1884	else
1885		clear_inode_flag(mapping->host, FI_HOT_DATA);
1886
1887	if (wbc->range_cyclic) {
1888		writeback_index = mapping->writeback_index; /* prev offset */
1889		index = writeback_index;
1890		if (index == 0)
1891			cycled = 1;
1892		else
1893			cycled = 0;
1894		end = -1;
1895	} else {
1896		index = wbc->range_start >> PAGE_SHIFT;
1897		end = wbc->range_end >> PAGE_SHIFT;
1898		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1899			range_whole = 1;
1900		cycled = 1; /* ignore range_cyclic tests */
1901	}
1902	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1903		tag = PAGECACHE_TAG_TOWRITE;
1904	else
1905		tag = PAGECACHE_TAG_DIRTY;
1906retry:
1907	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1908		tag_pages_for_writeback(mapping, index, end);
1909	done_index = index;
1910	while (!done && (index <= end)) {
1911		int i;
1912
1913		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
1914				tag);
1915		if (nr_pages == 0)
1916			break;
1917
1918		for (i = 0; i < nr_pages; i++) {
1919			struct page *page = pvec.pages[i];
1920			bool submitted = false;
 
 
 
 
1921
1922			done_index = page->index;
1923retry_write:
1924			lock_page(page);
1925
1926			if (unlikely(page->mapping != mapping)) {
1927continue_unlock:
1928				unlock_page(page);
1929				continue;
1930			}
1931
1932			if (!PageDirty(page)) {
1933				/* someone wrote it for us */
1934				goto continue_unlock;
1935			}
1936
 
 
 
1937			if (PageWriteback(page)) {
1938				if (wbc->sync_mode != WB_SYNC_NONE)
1939					f2fs_wait_on_page_writeback(page,
1940								DATA, true);
1941				else
1942					goto continue_unlock;
1943			}
1944
1945			BUG_ON(PageWriteback(page));
1946			if (!clear_page_dirty_for_io(page))
1947				goto continue_unlock;
1948
1949			ret = __write_data_page(page, &submitted, wbc, io_type);
1950			if (unlikely(ret)) {
1951				/*
1952				 * keep nr_to_write, since vfs uses this to
1953				 * get # of written pages.
1954				 */
1955				if (ret == AOP_WRITEPAGE_ACTIVATE) {
1956					unlock_page(page);
1957					ret = 0;
1958					continue;
1959				} else if (ret == -EAGAIN) {
1960					ret = 0;
1961					if (wbc->sync_mode == WB_SYNC_ALL) {
1962						cond_resched();
1963						congestion_wait(BLK_RW_ASYNC,
1964									HZ/50);
1965						goto retry_write;
1966					}
1967					continue;
1968				}
1969				done_index = page->index + 1;
1970				done = 1;
1971				break;
1972			} else if (submitted) {
1973				last_idx = page->index;
1974			}
1975
1976			/* give a priority to WB_SYNC threads */
1977			if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1978					--wbc->nr_to_write <= 0) &&
1979					wbc->sync_mode == WB_SYNC_NONE) {
1980				done = 1;
1981				break;
1982			}
1983		}
1984		pagevec_release(&pvec);
1985		cond_resched();
1986	}
1987
 
 
 
 
 
1988	if (!cycled && !done) {
1989		cycled = 1;
1990		index = 0;
1991		end = writeback_index - 1;
1992		goto retry;
1993	}
1994	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1995		mapping->writeback_index = done_index;
1996
1997	if (last_idx != ULONG_MAX)
1998		f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
1999						0, last_idx, DATA);
2000
2001	return ret;
2002}
2003
2004int __f2fs_write_data_pages(struct address_space *mapping,
2005						struct writeback_control *wbc,
2006						enum iostat_type io_type)
2007{
2008	struct inode *inode = mapping->host;
2009	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2010	struct blk_plug plug;
2011	int ret;
 
2012
2013	/* deal with chardevs and other special file */
2014	if (!mapping->a_ops->writepage)
2015		return 0;
2016
2017	/* skip writing if there is no dirty page in this inode */
2018	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
2019		return 0;
2020
2021	/* during POR, we don't need to trigger writepage at all. */
2022	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2023		goto skip_write;
2024
2025	if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
2026			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
2027			available_free_memory(sbi, DIRTY_DENTS))
2028		goto skip_write;
2029
2030	/* skip writing during file defragment */
2031	if (is_inode_flag_set(inode, FI_DO_DEFRAG))
2032		goto skip_write;
2033
2034	trace_f2fs_writepages(mapping->host, wbc, DATA);
2035
2036	/* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
2037	if (wbc->sync_mode == WB_SYNC_ALL)
2038		atomic_inc(&sbi->wb_sync_req);
2039	else if (atomic_read(&sbi->wb_sync_req))
2040		goto skip_write;
2041
2042	blk_start_plug(&plug);
2043	ret = f2fs_write_cache_pages(mapping, wbc, io_type);
2044	blk_finish_plug(&plug);
2045
2046	if (wbc->sync_mode == WB_SYNC_ALL)
2047		atomic_dec(&sbi->wb_sync_req);
2048	/*
2049	 * if some pages were truncated, we cannot guarantee its mapping->host
2050	 * to detect pending bios.
2051	 */
 
 
 
 
2052
2053	remove_dirty_inode(inode);
 
 
2054	return ret;
2055
2056skip_write:
2057	wbc->pages_skipped += get_dirty_pages(inode);
2058	trace_f2fs_writepages(mapping->host, wbc, DATA);
2059	return 0;
2060}
2061
2062static int f2fs_write_data_pages(struct address_space *mapping,
2063			    struct writeback_control *wbc)
2064{
2065	struct inode *inode = mapping->host;
2066
2067	return __f2fs_write_data_pages(mapping, wbc,
2068			F2FS_I(inode)->cp_task == current ?
2069			FS_CP_DATA_IO : FS_DATA_IO);
2070}
2071
2072static void f2fs_write_failed(struct address_space *mapping, loff_t to)
2073{
2074	struct inode *inode = mapping->host;
2075	loff_t i_size = i_size_read(inode);
2076
2077	if (to > i_size) {
2078		down_write(&F2FS_I(inode)->i_mmap_sem);
2079		truncate_pagecache(inode, i_size);
2080		truncate_blocks(inode, i_size, true);
2081		up_write(&F2FS_I(inode)->i_mmap_sem);
2082	}
2083}
2084
2085static int prepare_write_begin(struct f2fs_sb_info *sbi,
2086			struct page *page, loff_t pos, unsigned len,
2087			block_t *blk_addr, bool *node_changed)
2088{
2089	struct inode *inode = page->mapping->host;
2090	pgoff_t index = page->index;
2091	struct dnode_of_data dn;
2092	struct page *ipage;
2093	bool locked = false;
2094	struct extent_info ei = {0,0,0};
2095	int err = 0;
2096
2097	/*
2098	 * we already allocated all the blocks, so we don't need to get
2099	 * the block addresses when there is no need to fill the page.
2100	 */
2101	if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
2102			!is_inode_flag_set(inode, FI_NO_PREALLOC))
2103		return 0;
2104
2105	if (f2fs_has_inline_data(inode) ||
2106			(pos & PAGE_MASK) >= i_size_read(inode)) {
2107		__do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
2108		locked = true;
2109	}
2110restart:
2111	/* check inline_data */
2112	ipage = get_node_page(sbi, inode->i_ino);
2113	if (IS_ERR(ipage)) {
2114		err = PTR_ERR(ipage);
2115		goto unlock_out;
2116	}
2117
2118	set_new_dnode(&dn, inode, ipage, ipage, 0);
2119
2120	if (f2fs_has_inline_data(inode)) {
2121		if (pos + len <= MAX_INLINE_DATA(inode)) {
2122			read_inline_data(page, ipage);
2123			set_inode_flag(inode, FI_DATA_EXIST);
2124			if (inode->i_nlink)
2125				set_inline_node(ipage);
2126		} else {
2127			err = f2fs_convert_inline_page(&dn, page);
2128			if (err)
2129				goto out;
2130			if (dn.data_blkaddr == NULL_ADDR)
2131				err = f2fs_get_block(&dn, index);
2132		}
2133	} else if (locked) {
2134		err = f2fs_get_block(&dn, index);
2135	} else {
2136		if (f2fs_lookup_extent_cache(inode, index, &ei)) {
2137			dn.data_blkaddr = ei.blk + index - ei.fofs;
2138		} else {
2139			/* hole case */
2140			err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
2141			if (err || dn.data_blkaddr == NULL_ADDR) {
2142				f2fs_put_dnode(&dn);
2143				__do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
2144								true);
2145				locked = true;
2146				goto restart;
2147			}
2148		}
2149	}
2150
2151	/* convert_inline_page can make node_changed */
2152	*blk_addr = dn.data_blkaddr;
2153	*node_changed = dn.node_changed;
2154out:
2155	f2fs_put_dnode(&dn);
2156unlock_out:
2157	if (locked)
2158		__do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
2159	return err;
2160}
2161
2162static int f2fs_write_begin(struct file *file, struct address_space *mapping,
2163		loff_t pos, unsigned len, unsigned flags,
2164		struct page **pagep, void **fsdata)
2165{
2166	struct inode *inode = mapping->host;
2167	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2168	struct page *page = NULL;
2169	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
2170	bool need_balance = false, drop_atomic = false;
2171	block_t blkaddr = NULL_ADDR;
2172	int err = 0;
2173
2174	trace_f2fs_write_begin(inode, pos, len, flags);
2175
2176	if (f2fs_is_atomic_file(inode) &&
2177			!available_free_memory(sbi, INMEM_PAGES)) {
2178		err = -ENOMEM;
2179		drop_atomic = true;
2180		goto fail;
2181	}
2182
2183	/*
2184	 * We should check this at this moment to avoid deadlock on inode page
2185	 * and #0 page. The locking rule for inline_data conversion should be:
2186	 * lock_page(page #0) -> lock_page(inode_page)
2187	 */
2188	if (index != 0) {
2189		err = f2fs_convert_inline_inode(inode);
2190		if (err)
2191			goto fail;
2192	}
2193repeat:
2194	/*
2195	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
2196	 * wait_for_stable_page. Will wait that below with our IO control.
2197	 */
2198	page = f2fs_pagecache_get_page(mapping, index,
2199				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
2200	if (!page) {
2201		err = -ENOMEM;
2202		goto fail;
2203	}
2204
2205	*pagep = page;
2206
2207	err = prepare_write_begin(sbi, page, pos, len,
2208					&blkaddr, &need_balance);
2209	if (err)
2210		goto fail;
2211
2212	if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
2213		unlock_page(page);
2214		f2fs_balance_fs(sbi, true);
2215		lock_page(page);
2216		if (page->mapping != mapping) {
2217			/* The page got truncated from under us */
2218			f2fs_put_page(page, 1);
2219			goto repeat;
2220		}
2221	}
2222
2223	f2fs_wait_on_page_writeback(page, DATA, false);
2224
2225	/* wait for GCed encrypted page writeback */
2226	if (f2fs_encrypted_file(inode))
2227		f2fs_wait_on_block_writeback(sbi, blkaddr);
2228
2229	if (len == PAGE_SIZE || PageUptodate(page))
2230		return 0;
 
 
2231
2232	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
2233		zero_user_segment(page, len, PAGE_SIZE);
2234		return 0;
 
 
 
 
2235	}
2236
2237	if (blkaddr == NEW_ADDR) {
2238		zero_user_segment(page, 0, PAGE_SIZE);
2239		SetPageUptodate(page);
2240	} else {
2241		err = f2fs_submit_page_read(inode, page, blkaddr);
 
 
 
 
 
 
 
 
 
2242		if (err)
2243			goto fail;
2244
2245		lock_page(page);
 
 
 
 
2246		if (unlikely(page->mapping != mapping)) {
2247			f2fs_put_page(page, 1);
2248			goto repeat;
2249		}
2250		if (unlikely(!PageUptodate(page))) {
2251			err = -EIO;
2252			goto fail;
 
 
 
2253		}
2254	}
 
 
 
 
2255	return 0;
2256
2257fail:
2258	f2fs_put_page(page, 1);
2259	f2fs_write_failed(mapping, pos + len);
2260	if (drop_atomic)
2261		drop_inmem_pages_all(sbi);
2262	return err;
2263}
2264
2265static int f2fs_write_end(struct file *file,
2266			struct address_space *mapping,
2267			loff_t pos, unsigned len, unsigned copied,
2268			struct page *page, void *fsdata)
2269{
2270	struct inode *inode = page->mapping->host;
2271
2272	trace_f2fs_write_end(inode, pos, len, copied);
2273
2274	/*
2275	 * This should be come from len == PAGE_SIZE, and we expect copied
2276	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2277	 * let generic_perform_write() try to copy data again through copied=0.
2278	 */
2279	if (!PageUptodate(page)) {
2280		if (unlikely(copied != len))
2281			copied = 0;
2282		else
2283			SetPageUptodate(page);
2284	}
2285	if (!copied)
2286		goto unlock_out;
2287
2288	set_page_dirty(page);
2289
2290	if (pos + copied > i_size_read(inode))
2291		f2fs_i_size_write(inode, pos + copied);
2292unlock_out:
 
 
2293	f2fs_put_page(page, 1);
2294	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2295	return copied;
2296}
2297
2298static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2299			   loff_t offset)
2300{
2301	unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2302
2303	if (offset & blocksize_mask)
2304		return -EINVAL;
2305
2306	if (iov_iter_alignment(iter) & blocksize_mask)
2307		return -EINVAL;
2308
2309	return 0;
2310}
2311
2312static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 
2313{
2314	struct address_space *mapping = iocb->ki_filp->f_mapping;
2315	struct inode *inode = mapping->host;
2316	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2317	size_t count = iov_iter_count(iter);
2318	loff_t offset = iocb->ki_pos;
2319	int rw = iov_iter_rw(iter);
2320	int err;
2321	enum rw_hint hint = iocb->ki_hint;
2322	int whint_mode = F2FS_OPTION(sbi).whint_mode;
2323
2324	err = check_direct_IO(inode, iter, offset);
2325	if (err)
2326		return err;
2327
2328	if (f2fs_force_buffered_io(inode, rw))
2329		return 0;
2330
2331	trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2332
2333	if (rw == WRITE && whint_mode == WHINT_MODE_OFF)
2334		iocb->ki_hint = WRITE_LIFE_NOT_SET;
 
2335
2336	if (!down_read_trylock(&F2FS_I(inode)->dio_rwsem[rw])) {
2337		if (iocb->ki_flags & IOCB_NOWAIT) {
2338			iocb->ki_hint = hint;
2339			err = -EAGAIN;
2340			goto out;
2341		}
2342		down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2343	}
2344
2345	err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2346	up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2347
2348	if (rw == WRITE) {
2349		if (whint_mode == WHINT_MODE_OFF)
2350			iocb->ki_hint = hint;
2351		if (err > 0) {
2352			f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2353									err);
2354			set_inode_flag(inode, FI_UPDATE_WRITE);
2355		} else if (err < 0) {
2356			f2fs_write_failed(mapping, offset + count);
2357		}
2358	}
2359
2360out:
2361	trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2362
2363	return err;
2364}
2365
2366void f2fs_invalidate_page(struct page *page, unsigned int offset,
2367							unsigned int length)
2368{
2369	struct inode *inode = page->mapping->host;
2370	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2371
2372	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2373		(offset % PAGE_SIZE || length != PAGE_SIZE))
2374		return;
2375
2376	if (PageDirty(page)) {
2377		if (inode->i_ino == F2FS_META_INO(sbi)) {
2378			dec_page_count(sbi, F2FS_DIRTY_META);
2379		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2380			dec_page_count(sbi, F2FS_DIRTY_NODES);
2381		} else {
2382			inode_dec_dirty_pages(inode);
2383			remove_dirty_inode(inode);
2384		}
2385	}
2386
2387	/* This is atomic written page, keep Private */
2388	if (IS_ATOMIC_WRITTEN_PAGE(page))
2389		return drop_inmem_page(inode, page);
2390
2391	set_page_private(page, 0);
2392	ClearPagePrivate(page);
2393}
2394
2395int f2fs_release_page(struct page *page, gfp_t wait)
2396{
2397	/* If this is dirty page, keep PagePrivate */
2398	if (PageDirty(page))
2399		return 0;
2400
2401	/* This is atomic written page, keep Private */
2402	if (IS_ATOMIC_WRITTEN_PAGE(page))
2403		return 0;
2404
2405	set_page_private(page, 0);
2406	ClearPagePrivate(page);
2407	return 1;
2408}
2409
2410/*
2411 * This was copied from __set_page_dirty_buffers which gives higher performance
2412 * in very high speed storages. (e.g., pmem)
2413 */
2414void f2fs_set_page_dirty_nobuffers(struct page *page)
2415{
2416	struct address_space *mapping = page->mapping;
2417	unsigned long flags;
2418
2419	if (unlikely(!mapping))
2420		return;
2421
2422	spin_lock(&mapping->private_lock);
2423	lock_page_memcg(page);
2424	SetPageDirty(page);
2425	spin_unlock(&mapping->private_lock);
2426
2427	xa_lock_irqsave(&mapping->i_pages, flags);
2428	WARN_ON_ONCE(!PageUptodate(page));
2429	account_page_dirtied(page, mapping);
2430	radix_tree_tag_set(&mapping->i_pages,
2431			page_index(page), PAGECACHE_TAG_DIRTY);
2432	xa_unlock_irqrestore(&mapping->i_pages, flags);
2433	unlock_page_memcg(page);
2434
2435	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2436	return;
2437}
2438
2439static int f2fs_set_data_page_dirty(struct page *page)
2440{
2441	struct address_space *mapping = page->mapping;
2442	struct inode *inode = mapping->host;
2443
2444	trace_f2fs_set_page_dirty(page, DATA);
2445
2446	if (!PageUptodate(page))
2447		SetPageUptodate(page);
2448
2449	if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2450		if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2451			register_inmem_page(inode, page);
2452			return 1;
2453		}
2454		/*
2455		 * Previously, this page has been registered, we just
2456		 * return here.
2457		 */
2458		return 0;
2459	}
2460
2461	if (!PageDirty(page)) {
2462		f2fs_set_page_dirty_nobuffers(page);
2463		update_dirty_page(inode, page);
2464		return 1;
2465	}
2466	return 0;
2467}
2468
2469static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2470{
2471	struct inode *inode = mapping->host;
2472
2473	if (f2fs_has_inline_data(inode))
2474		return 0;
2475
2476	/* make sure allocating whole blocks */
2477	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2478		filemap_write_and_wait(mapping);
2479
2480	return generic_block_bmap(mapping, block, get_data_block_bmap);
2481}
2482
2483#ifdef CONFIG_MIGRATION
2484#include <linux/migrate.h>
2485
2486int f2fs_migrate_page(struct address_space *mapping,
2487		struct page *newpage, struct page *page, enum migrate_mode mode)
2488{
2489	int rc, extra_count;
2490	struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2491	bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2492
2493	BUG_ON(PageWriteback(page));
2494
2495	/* migrating an atomic written page is safe with the inmem_lock hold */
2496	if (atomic_written) {
2497		if (mode != MIGRATE_SYNC)
2498			return -EBUSY;
2499		if (!mutex_trylock(&fi->inmem_lock))
2500			return -EAGAIN;
2501	}
2502
2503	/*
2504	 * A reference is expected if PagePrivate set when move mapping,
2505	 * however F2FS breaks this for maintaining dirty page counts when
2506	 * truncating pages. So here adjusting the 'extra_count' make it work.
2507	 */
2508	extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2509	rc = migrate_page_move_mapping(mapping, newpage,
2510				page, NULL, mode, extra_count);
2511	if (rc != MIGRATEPAGE_SUCCESS) {
2512		if (atomic_written)
2513			mutex_unlock(&fi->inmem_lock);
2514		return rc;
2515	}
2516
2517	if (atomic_written) {
2518		struct inmem_pages *cur;
2519		list_for_each_entry(cur, &fi->inmem_pages, list)
2520			if (cur->page == page) {
2521				cur->page = newpage;
2522				break;
2523			}
2524		mutex_unlock(&fi->inmem_lock);
2525		put_page(page);
2526		get_page(newpage);
2527	}
2528
2529	if (PagePrivate(page))
2530		SetPagePrivate(newpage);
2531	set_page_private(newpage, page_private(page));
2532
2533	if (mode != MIGRATE_SYNC_NO_COPY)
2534		migrate_page_copy(newpage, page);
2535	else
2536		migrate_page_states(newpage, page);
2537
2538	return MIGRATEPAGE_SUCCESS;
2539}
2540#endif
2541
2542const struct address_space_operations f2fs_dblock_aops = {
2543	.readpage	= f2fs_read_data_page,
2544	.readpages	= f2fs_read_data_pages,
2545	.writepage	= f2fs_write_data_page,
2546	.writepages	= f2fs_write_data_pages,
2547	.write_begin	= f2fs_write_begin,
2548	.write_end	= f2fs_write_end,
2549	.set_page_dirty	= f2fs_set_data_page_dirty,
2550	.invalidatepage	= f2fs_invalidate_page,
2551	.releasepage	= f2fs_release_page,
2552	.direct_IO	= f2fs_direct_IO,
2553	.bmap		= f2fs_bmap,
2554#ifdef CONFIG_MIGRATION
2555	.migratepage    = f2fs_migrate_page,
2556#endif
2557};