inode.c

   1// SPDX-License-Identifier: GPL-2.0+
   2/*
   3 * NILFS inode operations.
   4 *
   5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6 *
   7 * Written by Ryusuke Konishi.
   8 *
   9 */
  10
  11#include <linux/buffer_head.h>
  12#include <linux/gfp.h>
  13#include <linux/mpage.h>
  14#include <linux/pagemap.h>
  15#include <linux/writeback.h>
  16#include <linux/uio.h>
  17#include <linux/fiemap.h>
  18#include <linux/random.h>
  19#include "nilfs.h"
  20#include "btnode.h"
  21#include "segment.h"
  22#include "page.h"
  23#include "mdt.h"
  24#include "cpfile.h"
  25#include "ifile.h"
  26
  27/**
  28 * struct nilfs_iget_args - arguments used during comparison between inodes
  29 * @ino: inode number
  30 * @cno: checkpoint number
  31 * @root: pointer on NILFS root object (mounted checkpoint)
  32 * @type: inode type
  33 */
  34struct nilfs_iget_args {
  35	u64 ino;
  36	__u64 cno;
  37	struct nilfs_root *root;
  38	unsigned int type;
  39};
  40
  41static int nilfs_iget_test(struct inode *inode, void *opaque);
  42
  43void nilfs_inode_add_blocks(struct inode *inode, int n)
  44{
  45	struct nilfs_root *root = NILFS_I(inode)->i_root;
  46
  47	inode_add_bytes(inode, i_blocksize(inode) * n);
  48	if (root)
  49		atomic64_add(n, &root->blocks_count);
  50}
  51
  52void nilfs_inode_sub_blocks(struct inode *inode, int n)
  53{
  54	struct nilfs_root *root = NILFS_I(inode)->i_root;
  55
  56	inode_sub_bytes(inode, i_blocksize(inode) * n);
  57	if (root)
  58		atomic64_sub(n, &root->blocks_count);
  59}
  60
  61/**
  62 * nilfs_get_block() - get a file block on the filesystem (callback function)
  63 * @inode: inode struct of the target file
  64 * @blkoff: file block number
  65 * @bh_result: buffer head to be mapped on
  66 * @create: indicate whether allocating the block or not when it has not
  67 *      been allocated yet.
  68 *
  69 * This function does not issue actual read request of the specified data
  70 * block. It is done by VFS.
  71 *
  72 * Return: 0 on success, or a negative error code on failure.
  73 */
  74int nilfs_get_block(struct inode *inode, sector_t blkoff,
  75		    struct buffer_head *bh_result, int create)
  76{
  77	struct nilfs_inode_info *ii = NILFS_I(inode);
  78	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  79	__u64 blknum = 0;
  80	int err = 0, ret;
  81	unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
  82
  83	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  84	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  85	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  86	if (ret >= 0) {	/* found */
  87		map_bh(bh_result, inode->i_sb, blknum);
  88		if (ret > 0)
  89			bh_result->b_size = (ret << inode->i_blkbits);
  90		goto out;
  91	}
  92	/* data block was not found */
  93	if (ret == -ENOENT && create) {
  94		struct nilfs_transaction_info ti;
  95
  96		bh_result->b_blocknr = 0;
  97		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  98		if (unlikely(err))
  99			goto out;
 100		err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 101					(unsigned long)bh_result);
 102		if (unlikely(err != 0)) {
 103			if (err == -EEXIST) {
 104				/*
 105				 * The get_block() function could be called
 106				 * from multiple callers for an inode.
 107				 * However, the page having this block must
 108				 * be locked in this case.
 109				 */
 110				nilfs_warn(inode->i_sb,
 111					   "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
 112					   __func__, inode->i_ino,
 113					   (unsigned long long)blkoff);
 114				err = -EAGAIN;
 115			}
 116			nilfs_transaction_abort(inode->i_sb);
 117			goto out;
 118		}
 119		nilfs_mark_inode_dirty_sync(inode);
 120		nilfs_transaction_commit(inode->i_sb); /* never fails */
 121		/* Error handling should be detailed */
 122		set_buffer_new(bh_result);
 123		set_buffer_delay(bh_result);
 124		map_bh(bh_result, inode->i_sb, 0);
 125		/* Disk block number must be changed to proper value */
 126
 127	} else if (ret == -ENOENT) {
 128		/*
 129		 * not found is not error (e.g. hole); must return without
 130		 * the mapped state flag.
 131		 */
 132		;
 133	} else {
 134		err = ret;
 135	}
 136
 137 out:
 138	return err;
 139}
 140
 141/**
 142 * nilfs_read_folio() - implement read_folio() method of nilfs_aops {}
 143 * address_space_operations.
 144 * @file: file struct of the file to be read
 145 * @folio: the folio to be read
 146 *
 147 * Return: 0 on success, or a negative error code on failure.
 148 */
 149static int nilfs_read_folio(struct file *file, struct folio *folio)
 150{
 151	return mpage_read_folio(folio, nilfs_get_block);
 152}
 153
 154static void nilfs_readahead(struct readahead_control *rac)
 155{
 156	mpage_readahead(rac, nilfs_get_block);
 157}
 158
 159static int nilfs_writepages(struct address_space *mapping,
 160			    struct writeback_control *wbc)
 161{
 162	struct inode *inode = mapping->host;
 163	int err = 0;
 164
 165	if (sb_rdonly(inode->i_sb)) {
 166		nilfs_clear_dirty_pages(mapping);
 167		return -EROFS;
 168	}
 169
 170	if (wbc->sync_mode == WB_SYNC_ALL)
 171		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 172						    wbc->range_start,
 173						    wbc->range_end);
 174	return err;
 175}
 176
 177static bool nilfs_dirty_folio(struct address_space *mapping,
 178		struct folio *folio)
 179{
 180	struct inode *inode = mapping->host;
 181	struct buffer_head *head;
 182	unsigned int nr_dirty = 0;
 183	bool ret = filemap_dirty_folio(mapping, folio);
 184
 185	/*
 186	 * The page may not be locked, eg if called from try_to_unmap_one()
 187	 */
 188	spin_lock(&mapping->i_private_lock);
 189	head = folio_buffers(folio);
 190	if (head) {
 191		struct buffer_head *bh = head;
 192
 193		do {
 194			/* Do not mark hole blocks dirty */
 195			if (buffer_dirty(bh) || !buffer_mapped(bh))
 196				continue;
 197
 198			set_buffer_dirty(bh);
 199			nr_dirty++;
 200		} while (bh = bh->b_this_page, bh != head);
 201	} else if (ret) {
 202		nr_dirty = 1 << (folio_shift(folio) - inode->i_blkbits);
 203	}
 204	spin_unlock(&mapping->i_private_lock);
 205
 206	if (nr_dirty)
 207		nilfs_set_file_dirty(inode, nr_dirty);
 208	return ret;
 209}
 210
 211void nilfs_write_failed(struct address_space *mapping, loff_t to)
 212{
 213	struct inode *inode = mapping->host;
 214
 215	if (to > inode->i_size) {
 216		truncate_pagecache(inode, inode->i_size);
 217		nilfs_truncate(inode);
 218	}
 219}
 220
 221static int nilfs_write_begin(const struct kiocb *iocb,
 222			     struct address_space *mapping,
 223			     loff_t pos, unsigned len,
 224			     struct folio **foliop, void **fsdata)
 225
 226{
 227	struct inode *inode = mapping->host;
 228	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 229
 230	if (unlikely(err))
 231		return err;
 232
 233	err = block_write_begin(mapping, pos, len, foliop, nilfs_get_block);
 234	if (unlikely(err)) {
 235		nilfs_write_failed(mapping, pos + len);
 236		nilfs_transaction_abort(inode->i_sb);
 237	}
 238	return err;
 239}
 240
 241static int nilfs_write_end(const struct kiocb *iocb,
 242			   struct address_space *mapping,
 243			   loff_t pos, unsigned len, unsigned copied,
 244			   struct folio *folio, void *fsdata)
 245{
 246	struct inode *inode = mapping->host;
 247	unsigned int start = pos & (PAGE_SIZE - 1);
 248	unsigned int nr_dirty;
 249	int err;
 250
 251	nr_dirty = nilfs_page_count_clean_buffers(folio, start,
 252						  start + copied);
 253	copied = generic_write_end(iocb, mapping, pos, len, copied, folio,
 254				   fsdata);
 255	nilfs_set_file_dirty(inode, nr_dirty);
 256	err = nilfs_transaction_commit(inode->i_sb);
 257	return err ? : copied;
 258}
 259
 260static ssize_t
 261nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 262{
 263	struct inode *inode = file_inode(iocb->ki_filp);
 264
 265	if (iov_iter_rw(iter) == WRITE)
 266		return 0;
 267
 268	/* Needs synchronization with the cleaner */
 269	return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 270}
 271
 272const struct address_space_operations nilfs_aops = {
 273	.read_folio		= nilfs_read_folio,
 274	.writepages		= nilfs_writepages,
 275	.dirty_folio		= nilfs_dirty_folio,
 276	.readahead		= nilfs_readahead,
 277	.write_begin		= nilfs_write_begin,
 278	.write_end		= nilfs_write_end,
 279	.invalidate_folio	= block_invalidate_folio,
 280	.direct_IO		= nilfs_direct_IO,
 281	.migrate_folio		= buffer_migrate_folio_norefs,
 282	.is_partially_uptodate  = block_is_partially_uptodate,
 283};
 284
 285const struct address_space_operations nilfs_buffer_cache_aops = {
 286	.invalidate_folio	= block_invalidate_folio,
 287};
 288
 289static int nilfs_insert_inode_locked(struct inode *inode,
 290				     struct nilfs_root *root,
 291				     unsigned long ino)
 292{
 293	struct nilfs_iget_args args = {
 294		.ino = ino, .root = root, .cno = 0, .type = NILFS_I_TYPE_NORMAL
 295	};
 296
 297	return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 298}
 299
 300struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 301{
 302	struct super_block *sb = dir->i_sb;
 303	struct inode *inode;
 304	struct nilfs_inode_info *ii;
 305	struct nilfs_root *root;
 306	struct buffer_head *bh;
 307	int err = -ENOMEM;
 308	ino_t ino;
 309
 310	inode = new_inode(sb);
 311	if (unlikely(!inode))
 312		goto failed;
 313
 314	mapping_set_gfp_mask(inode->i_mapping,
 315			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 316
 317	root = NILFS_I(dir)->i_root;
 318	ii = NILFS_I(inode);
 319	ii->i_state = BIT(NILFS_I_NEW);
 320	ii->i_type = NILFS_I_TYPE_NORMAL;
 321	ii->i_root = root;
 322
 323	err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 324	if (unlikely(err))
 325		goto failed_ifile_create_inode;
 326	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
 327	ii->i_bh = bh;
 328
 329	atomic64_inc(&root->inodes_count);
 330	inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
 331	inode->i_ino = ino;
 332	simple_inode_init_ts(inode);
 333
 334	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 335		err = nilfs_bmap_read(ii->i_bmap, NULL);
 336		if (err < 0)
 337			goto failed_after_creation;
 338
 339		set_bit(NILFS_I_BMAP, &ii->i_state);
 340		/* No lock is needed; iget() ensures it. */
 341	}
 342
 343	ii->i_flags = nilfs_mask_flags(
 344		mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 345
 346	/* ii->i_file_acl = 0; */
 347	/* ii->i_dir_acl = 0; */
 348	ii->i_dir_start_lookup = 0;
 349	nilfs_set_inode_flags(inode);
 350	inode->i_generation = get_random_u32();
 351	if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 352		err = -EIO;
 353		goto failed_after_creation;
 354	}
 355
 356	err = nilfs_init_acl(inode, dir);
 357	if (unlikely(err))
 358		/*
 359		 * Never occur.  When supporting nilfs_init_acl(),
 360		 * proper cancellation of above jobs should be considered.
 361		 */
 362		goto failed_after_creation;
 363
 364	return inode;
 365
 366 failed_after_creation:
 367	clear_nlink(inode);
 368	if (inode_state_read_once(inode) & I_NEW)
 369		unlock_new_inode(inode);
 370	iput(inode);  /*
 371		       * raw_inode will be deleted through
 372		       * nilfs_evict_inode().
 373		       */
 374	goto failed;
 375
 376 failed_ifile_create_inode:
 377	make_bad_inode(inode);
 378	iput(inode);
 379 failed:
 380	return ERR_PTR(err);
 381}
 382
 383void nilfs_set_inode_flags(struct inode *inode)
 384{
 385	unsigned int flags = NILFS_I(inode)->i_flags;
 386	unsigned int new_fl = 0;
 387
 388	if (flags & FS_SYNC_FL)
 389		new_fl |= S_SYNC;
 390	if (flags & FS_APPEND_FL)
 391		new_fl |= S_APPEND;
 392	if (flags & FS_IMMUTABLE_FL)
 393		new_fl |= S_IMMUTABLE;
 394	if (flags & FS_NOATIME_FL)
 395		new_fl |= S_NOATIME;
 396	if (flags & FS_DIRSYNC_FL)
 397		new_fl |= S_DIRSYNC;
 398	inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 399			S_NOATIME | S_DIRSYNC);
 400}
 401
 402int nilfs_read_inode_common(struct inode *inode,
 403			    struct nilfs_inode *raw_inode)
 404{
 405	struct nilfs_inode_info *ii = NILFS_I(inode);
 406	int err;
 407
 408	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 409	i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 410	i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 411	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 412	inode->i_size = le64_to_cpu(raw_inode->i_size);
 413	inode_set_atime(inode, le64_to_cpu(raw_inode->i_mtime),
 414			le32_to_cpu(raw_inode->i_mtime_nsec));
 415	inode_set_ctime(inode, le64_to_cpu(raw_inode->i_ctime),
 416			le32_to_cpu(raw_inode->i_ctime_nsec));
 417	inode_set_mtime(inode, le64_to_cpu(raw_inode->i_mtime),
 418			le32_to_cpu(raw_inode->i_mtime_nsec));
 419	if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
 420		return -EIO; /* this inode is for metadata and corrupted */
 421	if (inode->i_nlink == 0)
 422		return -ESTALE; /* this inode is deleted */
 423
 424	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 425	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 426#if 0
 427	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 428	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 429		0 : le32_to_cpu(raw_inode->i_dir_acl);
 430#endif
 431	ii->i_dir_start_lookup = 0;
 432	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 433
 434	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 435	    S_ISLNK(inode->i_mode)) {
 436		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 437		if (err < 0)
 438			return err;
 439		set_bit(NILFS_I_BMAP, &ii->i_state);
 440		/* No lock is needed; iget() ensures it. */
 441	}
 442	return 0;
 443}
 444
 445static int __nilfs_read_inode(struct super_block *sb,
 446			      struct nilfs_root *root, unsigned long ino,
 447			      struct inode *inode)
 448{
 449	struct the_nilfs *nilfs = sb->s_fs_info;
 450	struct buffer_head *bh;
 451	struct nilfs_inode *raw_inode;
 452	int err;
 453
 454	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 455	err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 456	if (unlikely(err))
 457		goto bad_inode;
 458
 459	raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 460
 461	err = nilfs_read_inode_common(inode, raw_inode);
 462	if (err)
 463		goto failed_unmap;
 464
 465	if (S_ISREG(inode->i_mode)) {
 466		inode->i_op = &nilfs_file_inode_operations;
 467		inode->i_fop = &nilfs_file_operations;
 468		inode->i_mapping->a_ops = &nilfs_aops;
 469	} else if (S_ISDIR(inode->i_mode)) {
 470		inode->i_op = &nilfs_dir_inode_operations;
 471		inode->i_fop = &nilfs_dir_operations;
 472		inode->i_mapping->a_ops = &nilfs_aops;
 473	} else if (S_ISLNK(inode->i_mode)) {
 474		inode->i_op = &nilfs_symlink_inode_operations;
 475		inode_nohighmem(inode);
 476		inode->i_mapping->a_ops = &nilfs_aops;
 477	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 478		   S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 479		inode->i_op = &nilfs_special_inode_operations;
 480		init_special_inode(
 481			inode, inode->i_mode,
 482			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 483	} else {
 484		nilfs_error(sb,
 485			    "invalid file type bits in mode 0%o for inode %lu",
 486			    inode->i_mode, ino);
 487		err = -EIO;
 488		goto failed_unmap;
 489	}
 490	nilfs_ifile_unmap_inode(raw_inode);
 491	brelse(bh);
 492	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 493	nilfs_set_inode_flags(inode);
 494	mapping_set_gfp_mask(inode->i_mapping,
 495			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 496	return 0;
 497
 498 failed_unmap:
 499	nilfs_ifile_unmap_inode(raw_inode);
 500	brelse(bh);
 501
 502 bad_inode:
 503	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 504	return err;
 505}
 506
 507static int nilfs_iget_test(struct inode *inode, void *opaque)
 508{
 509	struct nilfs_iget_args *args = opaque;
 510	struct nilfs_inode_info *ii;
 511
 512	if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 513		return 0;
 514
 515	ii = NILFS_I(inode);
 516	if (ii->i_type != args->type)
 517		return 0;
 518
 519	return !(args->type & NILFS_I_TYPE_GC) || args->cno == ii->i_cno;
 520}
 521
 522static int nilfs_iget_set(struct inode *inode, void *opaque)
 523{
 524	struct nilfs_iget_args *args = opaque;
 525
 526	inode->i_ino = args->ino;
 527	NILFS_I(inode)->i_cno = args->cno;
 528	NILFS_I(inode)->i_root = args->root;
 529	NILFS_I(inode)->i_type = args->type;
 530	if (args->root && args->ino == NILFS_ROOT_INO)
 531		nilfs_get_root(args->root);
 532	return 0;
 533}
 534
 535struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 536			    unsigned long ino)
 537{
 538	struct nilfs_iget_args args = {
 539		.ino = ino, .root = root, .cno = 0, .type = NILFS_I_TYPE_NORMAL
 540	};
 541
 542	return ilookup5(sb, ino, nilfs_iget_test, &args);
 543}
 544
 545struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 546				unsigned long ino)
 547{
 548	struct nilfs_iget_args args = {
 549		.ino = ino, .root = root, .cno = 0, .type = NILFS_I_TYPE_NORMAL
 550	};
 551
 552	return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 553}
 554
 555struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 556			 unsigned long ino)
 557{
 558	struct inode *inode;
 559	int err;
 560
 561	inode = nilfs_iget_locked(sb, root, ino);
 562	if (unlikely(!inode))
 563		return ERR_PTR(-ENOMEM);
 564
 565	if (!(inode_state_read_once(inode) & I_NEW)) {
 566		if (!inode->i_nlink) {
 567			iput(inode);
 568			return ERR_PTR(-ESTALE);
 569		}
 570		return inode;
 571	}
 572
 573	err = __nilfs_read_inode(sb, root, ino, inode);
 574	if (unlikely(err)) {
 575		iget_failed(inode);
 576		return ERR_PTR(err);
 577	}
 578	unlock_new_inode(inode);
 579	return inode;
 580}
 581
 582struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 583				__u64 cno)
 584{
 585	struct nilfs_iget_args args = {
 586		.ino = ino, .root = NULL, .cno = cno, .type = NILFS_I_TYPE_GC
 587	};
 588	struct inode *inode;
 589	int err;
 590
 591	inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 592	if (unlikely(!inode))
 593		return ERR_PTR(-ENOMEM);
 594	if (!(inode_state_read_once(inode) & I_NEW))
 595		return inode;
 596
 597	err = nilfs_init_gcinode(inode);
 598	if (unlikely(err)) {
 599		iget_failed(inode);
 600		return ERR_PTR(err);
 601	}
 602	unlock_new_inode(inode);
 603	return inode;
 604}
 605
 606/**
 607 * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 608 * @inode: inode object
 609 *
 610 * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 611 * or does nothing if the inode already has it.  This function allocates
 612 * an additional inode to maintain page cache of B-tree nodes one-on-one.
 613 *
 614 * Return: 0 on success, or %-ENOMEM if memory is insufficient.
 615 */
 616int nilfs_attach_btree_node_cache(struct inode *inode)
 617{
 618	struct nilfs_inode_info *ii = NILFS_I(inode);
 619	struct inode *btnc_inode;
 620	struct nilfs_iget_args args;
 621
 622	if (ii->i_assoc_inode)
 623		return 0;
 624
 625	args.ino = inode->i_ino;
 626	args.root = ii->i_root;
 627	args.cno = ii->i_cno;
 628	args.type = ii->i_type | NILFS_I_TYPE_BTNC;
 629
 630	btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 631				  nilfs_iget_set, &args);
 632	if (unlikely(!btnc_inode))
 633		return -ENOMEM;
 634	if (inode_state_read_once(btnc_inode) & I_NEW) {
 635		nilfs_init_btnc_inode(btnc_inode);
 636		unlock_new_inode(btnc_inode);
 637	}
 638	NILFS_I(btnc_inode)->i_assoc_inode = inode;
 639	NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 640	ii->i_assoc_inode = btnc_inode;
 641
 642	return 0;
 643}
 644
 645/**
 646 * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 647 * @inode: inode object
 648 *
 649 * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 650 * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 651 */
 652void nilfs_detach_btree_node_cache(struct inode *inode)
 653{
 654	struct nilfs_inode_info *ii = NILFS_I(inode);
 655	struct inode *btnc_inode = ii->i_assoc_inode;
 656
 657	if (btnc_inode) {
 658		NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 659		ii->i_assoc_inode = NULL;
 660		iput(btnc_inode);
 661	}
 662}
 663
 664/**
 665 * nilfs_iget_for_shadow - obtain inode for shadow mapping
 666 * @inode: inode object that uses shadow mapping
 667 *
 668 * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 669 * caches for shadow mapping.  The page cache for data pages is set up
 670 * in one inode and the one for b-tree node pages is set up in the
 671 * other inode, which is attached to the former inode.
 672 *
 673 * Return: a pointer to the inode for data pages on success, or %-ENOMEM
 674 * if memory is insufficient.
 675 */
 676struct inode *nilfs_iget_for_shadow(struct inode *inode)
 677{
 678	struct nilfs_iget_args args = {
 679		.ino = inode->i_ino, .root = NULL, .cno = 0,
 680		.type = NILFS_I_TYPE_SHADOW
 681	};
 682	struct inode *s_inode;
 683	int err;
 684
 685	s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 686			       nilfs_iget_set, &args);
 687	if (unlikely(!s_inode))
 688		return ERR_PTR(-ENOMEM);
 689	if (!(inode_state_read_once(s_inode) & I_NEW))
 690		return inode;
 691
 692	NILFS_I(s_inode)->i_flags = 0;
 693	memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 694	mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 695	s_inode->i_mapping->a_ops = &nilfs_buffer_cache_aops;
 696
 697	err = nilfs_attach_btree_node_cache(s_inode);
 698	if (unlikely(err)) {
 699		iget_failed(s_inode);
 700		return ERR_PTR(err);
 701	}
 702	unlock_new_inode(s_inode);
 703	return s_inode;
 704}
 705
 706/**
 707 * nilfs_write_inode_common - export common inode information to on-disk inode
 708 * @inode:     inode object
 709 * @raw_inode: on-disk inode
 710 *
 711 * This function writes standard information from the on-memory inode @inode
 712 * to @raw_inode on ifile, cpfile or a super root block.  Since inode bmap
 713 * data is not exported, nilfs_bmap_write() must be called separately during
 714 * log writing.
 715 */
 716void nilfs_write_inode_common(struct inode *inode,
 717			      struct nilfs_inode *raw_inode)
 718{
 719	struct nilfs_inode_info *ii = NILFS_I(inode);
 720
 721	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 722	raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 723	raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 724	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 725	raw_inode->i_size = cpu_to_le64(inode->i_size);
 726	raw_inode->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
 727	raw_inode->i_mtime = cpu_to_le64(inode_get_mtime_sec(inode));
 728	raw_inode->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
 729	raw_inode->i_mtime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
 730	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 731
 732	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 733	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 734
 735	/*
 736	 * When extending inode, nilfs->ns_inode_size should be checked
 737	 * for substitutions of appended fields.
 738	 */
 739}
 740
 741void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 742{
 743	ino_t ino = inode->i_ino;
 744	struct nilfs_inode_info *ii = NILFS_I(inode);
 745	struct inode *ifile = ii->i_root->ifile;
 746	struct nilfs_inode *raw_inode;
 747
 748	raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 749
 750	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 751		memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 752	if (flags & I_DIRTY_DATASYNC)
 753		set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 754
 755	nilfs_write_inode_common(inode, raw_inode);
 756
 757	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 758		raw_inode->i_device_code =
 759			cpu_to_le64(huge_encode_dev(inode->i_rdev));
 760
 761	nilfs_ifile_unmap_inode(raw_inode);
 762}
 763
 764#define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
 765
 766static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 767				unsigned long from)
 768{
 769	__u64 b;
 770	int ret;
 771
 772	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 773		return;
 774repeat:
 775	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 776	if (ret == -ENOENT)
 777		return;
 778	else if (ret < 0)
 779		goto failed;
 780
 781	if (b < from)
 782		return;
 783
 784	b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 785	ret = nilfs_bmap_truncate(ii->i_bmap, b);
 786	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 787	if (!ret || (ret == -ENOMEM &&
 788		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 789		goto repeat;
 790
 791failed:
 792	nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
 793		   ret, ii->vfs_inode.i_ino);
 794}
 795
 796void nilfs_truncate(struct inode *inode)
 797{
 798	unsigned long blkoff;
 799	unsigned int blocksize;
 800	struct nilfs_transaction_info ti;
 801	struct super_block *sb = inode->i_sb;
 802	struct nilfs_inode_info *ii = NILFS_I(inode);
 803
 804	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 805		return;
 806	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 807		return;
 808
 809	blocksize = sb->s_blocksize;
 810	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 811	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 812
 813	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 814
 815	nilfs_truncate_bmap(ii, blkoff);
 816
 817	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
 818	if (IS_SYNC(inode))
 819		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 820
 821	nilfs_mark_inode_dirty(inode);
 822	nilfs_set_file_dirty(inode, 0);
 823	nilfs_transaction_commit(sb);
 824	/*
 825	 * May construct a logical segment and may fail in sync mode.
 826	 * But truncate has no return value.
 827	 */
 828}
 829
 830static void nilfs_clear_inode(struct inode *inode)
 831{
 832	struct nilfs_inode_info *ii = NILFS_I(inode);
 833
 834	/*
 835	 * Free resources allocated in nilfs_read_inode(), here.
 836	 */
 837	BUG_ON(!list_empty(&ii->i_dirty));
 838	brelse(ii->i_bh);
 839	ii->i_bh = NULL;
 840
 841	if (nilfs_is_metadata_file_inode(inode))
 842		nilfs_mdt_clear(inode);
 843
 844	if (test_bit(NILFS_I_BMAP, &ii->i_state))
 845		nilfs_bmap_clear(ii->i_bmap);
 846
 847	if (!(ii->i_type & NILFS_I_TYPE_BTNC))
 848		nilfs_detach_btree_node_cache(inode);
 849
 850	if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 851		nilfs_put_root(ii->i_root);
 852}
 853
 854void nilfs_evict_inode(struct inode *inode)
 855{
 856	struct nilfs_transaction_info ti;
 857	struct super_block *sb = inode->i_sb;
 858	struct nilfs_inode_info *ii = NILFS_I(inode);
 859	struct the_nilfs *nilfs;
 860	int ret;
 861
 862	if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 863		truncate_inode_pages_final(&inode->i_data);
 864		clear_inode(inode);
 865		nilfs_clear_inode(inode);
 866		return;
 867	}
 868	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 869
 870	truncate_inode_pages_final(&inode->i_data);
 871
 872	nilfs = sb->s_fs_info;
 873	if (unlikely(sb_rdonly(sb) || !nilfs->ns_writer)) {
 874		/*
 875		 * If this inode is about to be disposed after the file system
 876		 * has been degraded to read-only due to file system corruption
 877		 * or after the writer has been detached, do not make any
 878		 * changes that cause writes, just clear it.
 879		 * Do this check after read-locking ns_segctor_sem by
 880		 * nilfs_transaction_begin() in order to avoid a race with
 881		 * the writer detach operation.
 882		 */
 883		clear_inode(inode);
 884		nilfs_clear_inode(inode);
 885		nilfs_transaction_abort(sb);
 886		return;
 887	}
 888
 889	/* TODO: some of the following operations may fail.  */
 890	nilfs_truncate_bmap(ii, 0);
 891	nilfs_mark_inode_dirty(inode);
 892	clear_inode(inode);
 893
 894	ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 895	if (!ret)
 896		atomic64_dec(&ii->i_root->inodes_count);
 897
 898	nilfs_clear_inode(inode);
 899
 900	if (IS_SYNC(inode))
 901		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 902	nilfs_transaction_commit(sb);
 903	/*
 904	 * May construct a logical segment and may fail in sync mode.
 905	 * But delete_inode has no return value.
 906	 */
 907}
 908
 909int nilfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 910		  struct iattr *iattr)
 911{
 912	struct nilfs_transaction_info ti;
 913	struct inode *inode = d_inode(dentry);
 914	struct super_block *sb = inode->i_sb;
 915	int err;
 916
 917	err = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
 918	if (err)
 919		return err;
 920
 921	err = nilfs_transaction_begin(sb, &ti, 0);
 922	if (unlikely(err))
 923		return err;
 924
 925	if ((iattr->ia_valid & ATTR_SIZE) &&
 926	    iattr->ia_size != i_size_read(inode)) {
 927		inode_dio_wait(inode);
 928		truncate_setsize(inode, iattr->ia_size);
 929		nilfs_truncate(inode);
 930	}
 931
 932	setattr_copy(&nop_mnt_idmap, inode, iattr);
 933	mark_inode_dirty(inode);
 934
 935	if (iattr->ia_valid & ATTR_MODE) {
 936		err = nilfs_acl_chmod(inode);
 937		if (unlikely(err))
 938			goto out_err;
 939	}
 940
 941	return nilfs_transaction_commit(sb);
 942
 943out_err:
 944	nilfs_transaction_abort(sb);
 945	return err;
 946}
 947
 948int nilfs_permission(struct mnt_idmap *idmap, struct inode *inode,
 949		     int mask)
 950{
 951	struct nilfs_root *root = NILFS_I(inode)->i_root;
 952
 953	if ((mask & MAY_WRITE) && root &&
 954	    root->cno != NILFS_CPTREE_CURRENT_CNO)
 955		return -EROFS; /* snapshot is not writable */
 956
 957	return generic_permission(&nop_mnt_idmap, inode, mask);
 958}
 959
 960int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 961{
 962	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 963	struct nilfs_inode_info *ii = NILFS_I(inode);
 964	int err;
 965
 966	spin_lock(&nilfs->ns_inode_lock);
 967	if (ii->i_bh == NULL || unlikely(!buffer_uptodate(ii->i_bh))) {
 968		spin_unlock(&nilfs->ns_inode_lock);
 969		err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 970						  inode->i_ino, pbh);
 971		if (unlikely(err))
 972			return err;
 973		spin_lock(&nilfs->ns_inode_lock);
 974		if (ii->i_bh == NULL)
 975			ii->i_bh = *pbh;
 976		else if (unlikely(!buffer_uptodate(ii->i_bh))) {
 977			__brelse(ii->i_bh);
 978			ii->i_bh = *pbh;
 979		} else {
 980			brelse(*pbh);
 981			*pbh = ii->i_bh;
 982		}
 983	} else
 984		*pbh = ii->i_bh;
 985
 986	get_bh(*pbh);
 987	spin_unlock(&nilfs->ns_inode_lock);
 988	return 0;
 989}
 990
 991int nilfs_inode_dirty(struct inode *inode)
 992{
 993	struct nilfs_inode_info *ii = NILFS_I(inode);
 994	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 995	int ret = 0;
 996
 997	if (!list_empty(&ii->i_dirty)) {
 998		spin_lock(&nilfs->ns_inode_lock);
 999		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1000			test_bit(NILFS_I_BUSY, &ii->i_state);
1001		spin_unlock(&nilfs->ns_inode_lock);
1002	}
1003	return ret;
1004}
1005
1006int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1007{
1008	struct nilfs_inode_info *ii = NILFS_I(inode);
1009	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1010
1011	atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1012
1013	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1014		return 0;
1015
1016	spin_lock(&nilfs->ns_inode_lock);
1017	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1018	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1019		/*
1020		 * Because this routine may race with nilfs_dispose_list(),
1021		 * we have to check NILFS_I_QUEUED here, too.
1022		 */
1023		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1024			/*
1025			 * This will happen when somebody is freeing
1026			 * this inode.
1027			 */
1028			nilfs_warn(inode->i_sb,
1029				   "cannot set file dirty (ino=%lu): the file is being freed",
1030				   inode->i_ino);
1031			spin_unlock(&nilfs->ns_inode_lock);
1032			return -EINVAL; /*
1033					 * NILFS_I_DIRTY may remain for
1034					 * freeing inode.
1035					 */
1036		}
1037		list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1038		set_bit(NILFS_I_QUEUED, &ii->i_state);
1039	}
1040	spin_unlock(&nilfs->ns_inode_lock);
1041	return 0;
1042}
1043
1044int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1045{
1046	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1047	struct buffer_head *ibh;
1048	int err;
1049
1050	/*
1051	 * Do not dirty inodes after the log writer has been detached
1052	 * and its nilfs_root struct has been freed.
1053	 */
1054	if (unlikely(nilfs_purging(nilfs)))
1055		return 0;
1056
1057	err = nilfs_load_inode_block(inode, &ibh);
1058	if (unlikely(err)) {
1059		nilfs_warn(inode->i_sb,
1060			   "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1061			   inode->i_ino, err);
1062		return err;
1063	}
1064	nilfs_update_inode(inode, ibh, flags);
1065	mark_buffer_dirty(ibh);
1066	nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1067	brelse(ibh);
1068	return 0;
1069}
1070
1071/**
1072 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1073 * @inode: inode of the file to be registered.
1074 * @flags: flags to determine the dirty state of the inode
1075 *
1076 * nilfs_dirty_inode() loads a inode block containing the specified
1077 * @inode and copies data from a nilfs_inode to a corresponding inode
1078 * entry in the inode block. This operation is excluded from the segment
1079 * construction. This function can be called both as a single operation
1080 * and as a part of indivisible file operations.
1081 */
1082void nilfs_dirty_inode(struct inode *inode, int flags)
1083{
1084	struct nilfs_transaction_info ti;
1085	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1086
1087	if (is_bad_inode(inode)) {
1088		nilfs_warn(inode->i_sb,
1089			   "tried to mark bad_inode dirty. ignored.");
1090		dump_stack();
1091		return;
1092	}
1093	if (mdi) {
1094		nilfs_mdt_mark_dirty(inode);
1095		return;
1096	}
1097	nilfs_transaction_begin(inode->i_sb, &ti, 0);
1098	__nilfs_mark_inode_dirty(inode, flags);
1099	nilfs_transaction_commit(inode->i_sb); /* never fails */
1100}
1101
1102int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1103		 __u64 start, __u64 len)
1104{
1105	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1106	__u64 logical = 0, phys = 0, size = 0;
1107	__u32 flags = 0;
1108	loff_t isize;
1109	sector_t blkoff, end_blkoff;
1110	sector_t delalloc_blkoff;
1111	unsigned long delalloc_blklen;
1112	unsigned int blkbits = inode->i_blkbits;
1113	int ret, n;
1114
1115	ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1116	if (ret)
1117		return ret;
1118
1119	inode_lock(inode);
1120
1121	isize = i_size_read(inode);
1122
1123	blkoff = start >> blkbits;
1124	end_blkoff = (start + len - 1) >> blkbits;
1125
1126	delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1127							&delalloc_blkoff);
1128
1129	do {
1130		__u64 blkphy;
1131		unsigned int maxblocks;
1132
1133		if (delalloc_blklen && blkoff == delalloc_blkoff) {
1134			if (size) {
1135				/* End of the current extent */
1136				ret = fiemap_fill_next_extent(
1137					fieinfo, logical, phys, size, flags);
1138				if (ret)
1139					break;
1140			}
1141			if (blkoff > end_blkoff)
1142				break;
1143
1144			flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1145			logical = blkoff << blkbits;
1146			phys = 0;
1147			size = delalloc_blklen << blkbits;
1148
1149			blkoff = delalloc_blkoff + delalloc_blklen;
1150			delalloc_blklen = nilfs_find_uncommitted_extent(
1151				inode, blkoff, &delalloc_blkoff);
1152			continue;
1153		}
1154
1155		/*
1156		 * Limit the number of blocks that we look up so as
1157		 * not to get into the next delayed allocation extent.
1158		 */
1159		maxblocks = INT_MAX;
1160		if (delalloc_blklen)
1161			maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1162					  maxblocks);
1163		blkphy = 0;
1164
1165		down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1166		n = nilfs_bmap_lookup_contig(
1167			NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1168		up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1169
1170		if (n < 0) {
1171			int past_eof;
1172
1173			if (unlikely(n != -ENOENT))
1174				break; /* error */
1175
1176			/* HOLE */
1177			blkoff++;
1178			past_eof = ((blkoff << blkbits) >= isize);
1179
1180			if (size) {
1181				/* End of the current extent */
1182
1183				if (past_eof)
1184					flags |= FIEMAP_EXTENT_LAST;
1185
1186				ret = fiemap_fill_next_extent(
1187					fieinfo, logical, phys, size, flags);
1188				if (ret)
1189					break;
1190				size = 0;
1191			}
1192			if (blkoff > end_blkoff || past_eof)
1193				break;
1194		} else {
1195			if (size) {
1196				if (phys && blkphy << blkbits == phys + size) {
1197					/* The current extent goes on */
1198					size += (u64)n << blkbits;
1199				} else {
1200					/* Terminate the current extent */
1201					ret = fiemap_fill_next_extent(
1202						fieinfo, logical, phys, size,
1203						flags);
1204					if (ret || blkoff > end_blkoff)
1205						break;
1206
1207					/* Start another extent */
1208					flags = FIEMAP_EXTENT_MERGED;
1209					logical = blkoff << blkbits;
1210					phys = blkphy << blkbits;
1211					size = (u64)n << blkbits;
1212				}
1213			} else {
1214				/* Start a new extent */
1215				flags = FIEMAP_EXTENT_MERGED;
1216				logical = blkoff << blkbits;
1217				phys = blkphy << blkbits;
1218				size = (u64)n << blkbits;
1219			}
1220			blkoff += n;
1221		}
1222		cond_resched();
1223	} while (true);
1224
1225	/* If ret is 1 then we just hit the end of the extent array */
1226	if (ret == 1)
1227		ret = 0;
1228
1229	inode_unlock(inode);
1230	return ret;
1231}