Linux Audio

Check our new training course

Loading...
v6.2
   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 "nilfs.h"
  19#include "btnode.h"
  20#include "segment.h"
  21#include "page.h"
  22#include "mdt.h"
  23#include "cpfile.h"
  24#include "ifile.h"
  25
  26/**
  27 * struct nilfs_iget_args - arguments used during comparison between inodes
  28 * @ino: inode number
  29 * @cno: checkpoint number
  30 * @root: pointer on NILFS root object (mounted checkpoint)
  31 * @for_gc: inode for GC flag
  32 * @for_btnc: inode for B-tree node cache flag
  33 * @for_shadow: inode for shadowed page cache flag
  34 */
  35struct nilfs_iget_args {
  36	u64 ino;
  37	__u64 cno;
  38	struct nilfs_root *root;
  39	bool for_gc;
  40	bool for_btnc;
  41	bool for_shadow;
  42};
  43
  44static int nilfs_iget_test(struct inode *inode, void *opaque);
  45
  46void nilfs_inode_add_blocks(struct inode *inode, int n)
  47{
  48	struct nilfs_root *root = NILFS_I(inode)->i_root;
  49
  50	inode_add_bytes(inode, i_blocksize(inode) * n);
  51	if (root)
  52		atomic64_add(n, &root->blocks_count);
  53}
  54
  55void nilfs_inode_sub_blocks(struct inode *inode, int n)
  56{
  57	struct nilfs_root *root = NILFS_I(inode)->i_root;
  58
  59	inode_sub_bytes(inode, i_blocksize(inode) * n);
  60	if (root)
  61		atomic64_sub(n, &root->blocks_count);
  62}
  63
  64/**
  65 * nilfs_get_block() - get a file block on the filesystem (callback function)
  66 * @inode: inode struct of the target file
  67 * @blkoff: file block number
  68 * @bh_result: buffer head to be mapped on
  69 * @create: indicate whether allocating the block or not when it has not
  70 *      been allocated yet.
  71 *
  72 * This function does not issue actual read request of the specified data
  73 * block. It is done by VFS.
  74 */
  75int nilfs_get_block(struct inode *inode, sector_t blkoff,
  76		    struct buffer_head *bh_result, int create)
  77{
  78	struct nilfs_inode_info *ii = NILFS_I(inode);
  79	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  80	__u64 blknum = 0;
  81	int err = 0, ret;
  82	unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
  83
  84	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  85	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  86	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  87	if (ret >= 0) {	/* found */
  88		map_bh(bh_result, inode->i_sb, blknum);
  89		if (ret > 0)
  90			bh_result->b_size = (ret << inode->i_blkbits);
  91		goto out;
  92	}
  93	/* data block was not found */
  94	if (ret == -ENOENT && create) {
  95		struct nilfs_transaction_info ti;
  96
  97		bh_result->b_blocknr = 0;
  98		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  99		if (unlikely(err))
 100			goto out;
 101		err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 102					(unsigned long)bh_result);
 103		if (unlikely(err != 0)) {
 104			if (err == -EEXIST) {
 105				/*
 106				 * The get_block() function could be called
 107				 * from multiple callers for an inode.
 108				 * However, the page having this block must
 109				 * be locked in this case.
 110				 */
 111				nilfs_warn(inode->i_sb,
 112					   "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
 113					   __func__, inode->i_ino,
 114					   (unsigned long long)blkoff);
 
 
 
 115				err = 0;
 116			}
 117			nilfs_transaction_abort(inode->i_sb);
 118			goto out;
 119		}
 120		nilfs_mark_inode_dirty_sync(inode);
 121		nilfs_transaction_commit(inode->i_sb); /* never fails */
 122		/* Error handling should be detailed */
 123		set_buffer_new(bh_result);
 124		set_buffer_delay(bh_result);
 125		map_bh(bh_result, inode->i_sb, 0);
 126		/* Disk block number must be changed to proper value */
 127
 128	} else if (ret == -ENOENT) {
 129		/*
 130		 * not found is not error (e.g. hole); must return without
 131		 * the mapped state flag.
 132		 */
 133		;
 134	} else {
 135		err = ret;
 136	}
 137
 138 out:
 139	return err;
 140}
 141
 142/**
 143 * nilfs_read_folio() - implement read_folio() method of nilfs_aops {}
 144 * address_space_operations.
 145 * @file: file struct of the file to be read
 146 * @folio: the folio to be read
 147 */
 148static int nilfs_read_folio(struct file *file, struct folio *folio)
 149{
 150	return mpage_read_folio(folio, nilfs_get_block);
 151}
 152
 153static void nilfs_readahead(struct readahead_control *rac)
 
 
 
 
 
 
 
 
 
 154{
 155	mpage_readahead(rac, nilfs_get_block);
 156}
 157
 158static int nilfs_writepages(struct address_space *mapping,
 159			    struct writeback_control *wbc)
 160{
 161	struct inode *inode = mapping->host;
 162	int err = 0;
 163
 164	if (sb_rdonly(inode->i_sb)) {
 165		nilfs_clear_dirty_pages(mapping, false);
 166		return -EROFS;
 167	}
 168
 169	if (wbc->sync_mode == WB_SYNC_ALL)
 170		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 171						    wbc->range_start,
 172						    wbc->range_end);
 173	return err;
 174}
 175
 176static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 177{
 178	struct inode *inode = page->mapping->host;
 179	int err;
 180
 181	if (sb_rdonly(inode->i_sb)) {
 182		/*
 183		 * It means that filesystem was remounted in read-only
 184		 * mode because of error or metadata corruption. But we
 185		 * have dirty pages that try to be flushed in background.
 186		 * So, here we simply discard this dirty page.
 187		 */
 188		nilfs_clear_dirty_page(page, false);
 189		unlock_page(page);
 190		return -EROFS;
 191	}
 192
 193	redirty_page_for_writepage(wbc, page);
 194	unlock_page(page);
 195
 196	if (wbc->sync_mode == WB_SYNC_ALL) {
 197		err = nilfs_construct_segment(inode->i_sb);
 198		if (unlikely(err))
 199			return err;
 200	} else if (wbc->for_reclaim)
 201		nilfs_flush_segment(inode->i_sb, inode->i_ino);
 202
 203	return 0;
 204}
 205
 206static bool nilfs_dirty_folio(struct address_space *mapping,
 207		struct folio *folio)
 208{
 209	struct inode *inode = mapping->host;
 210	struct buffer_head *head;
 211	unsigned int nr_dirty = 0;
 212	bool ret = filemap_dirty_folio(mapping, folio);
 213
 214	/*
 215	 * The page may not be locked, eg if called from try_to_unmap_one()
 216	 */
 217	spin_lock(&mapping->private_lock);
 218	head = folio_buffers(folio);
 219	if (head) {
 220		struct buffer_head *bh = head;
 221
 222		do {
 223			/* Do not mark hole blocks dirty */
 224			if (buffer_dirty(bh) || !buffer_mapped(bh))
 225				continue;
 226
 227			set_buffer_dirty(bh);
 228			nr_dirty++;
 229		} while (bh = bh->b_this_page, bh != head);
 230	} else if (ret) {
 231		nr_dirty = 1 << (folio_shift(folio) - inode->i_blkbits);
 232	}
 233	spin_unlock(&mapping->private_lock);
 234
 235	if (nr_dirty)
 236		nilfs_set_file_dirty(inode, nr_dirty);
 237	return ret;
 238}
 239
 240void nilfs_write_failed(struct address_space *mapping, loff_t to)
 241{
 242	struct inode *inode = mapping->host;
 243
 244	if (to > inode->i_size) {
 245		truncate_pagecache(inode, inode->i_size);
 246		nilfs_truncate(inode);
 247	}
 
 248}
 249
 250static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 251			     loff_t pos, unsigned len,
 252			     struct page **pagep, void **fsdata)
 253
 254{
 255	struct inode *inode = mapping->host;
 256	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 257
 258	if (unlikely(err))
 259		return err;
 260
 261	err = block_write_begin(mapping, pos, len, pagep, nilfs_get_block);
 
 262	if (unlikely(err)) {
 263		nilfs_write_failed(mapping, pos + len);
 
 
 
 264		nilfs_transaction_abort(inode->i_sb);
 265	}
 266	return err;
 267}
 268
 269static int nilfs_write_end(struct file *file, struct address_space *mapping,
 270			   loff_t pos, unsigned len, unsigned copied,
 271			   struct page *page, void *fsdata)
 272{
 273	struct inode *inode = mapping->host;
 274	unsigned int start = pos & (PAGE_SIZE - 1);
 275	unsigned int nr_dirty;
 276	int err;
 277
 278	nr_dirty = nilfs_page_count_clean_buffers(page, start,
 279						  start + copied);
 280	copied = generic_write_end(file, mapping, pos, len, copied, page,
 281				   fsdata);
 282	nilfs_set_file_dirty(inode, nr_dirty);
 283	err = nilfs_transaction_commit(inode->i_sb);
 284	return err ? : copied;
 285}
 286
 287static ssize_t
 288nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 
 289{
 290	struct inode *inode = file_inode(iocb->ki_filp);
 
 
 291
 292	if (iov_iter_rw(iter) == WRITE)
 293		return 0;
 294
 295	/* Needs synchronization with the cleaner */
 296	return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 297}
 298
 299const struct address_space_operations nilfs_aops = {
 300	.writepage		= nilfs_writepage,
 301	.read_folio		= nilfs_read_folio,
 302	.writepages		= nilfs_writepages,
 303	.dirty_folio		= nilfs_dirty_folio,
 304	.readahead		= nilfs_readahead,
 305	.write_begin		= nilfs_write_begin,
 306	.write_end		= nilfs_write_end,
 307	.invalidate_folio	= block_invalidate_folio,
 
 308	.direct_IO		= nilfs_direct_IO,
 309	.is_partially_uptodate  = block_is_partially_uptodate,
 310};
 311
 312static int nilfs_insert_inode_locked(struct inode *inode,
 313				     struct nilfs_root *root,
 314				     unsigned long ino)
 315{
 316	struct nilfs_iget_args args = {
 317		.ino = ino, .root = root, .cno = 0, .for_gc = false,
 318		.for_btnc = false, .for_shadow = false
 319	};
 320
 321	return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 322}
 323
 324struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 325{
 326	struct super_block *sb = dir->i_sb;
 327	struct the_nilfs *nilfs = sb->s_fs_info;
 328	struct inode *inode;
 329	struct nilfs_inode_info *ii;
 330	struct nilfs_root *root;
 331	struct buffer_head *bh;
 332	int err = -ENOMEM;
 333	ino_t ino;
 334
 335	inode = new_inode(sb);
 336	if (unlikely(!inode))
 337		goto failed;
 338
 339	mapping_set_gfp_mask(inode->i_mapping,
 340			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 341
 342	root = NILFS_I(dir)->i_root;
 343	ii = NILFS_I(inode);
 344	ii->i_state = BIT(NILFS_I_NEW);
 345	ii->i_root = root;
 346
 347	err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 348	if (unlikely(err))
 349		goto failed_ifile_create_inode;
 350	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
 351
 352	if (unlikely(ino < NILFS_USER_INO)) {
 353		nilfs_warn(sb,
 354			   "inode bitmap is inconsistent for reserved inodes");
 355		do {
 356			brelse(bh);
 357			err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
 358			if (unlikely(err))
 359				goto failed_ifile_create_inode;
 360		} while (ino < NILFS_USER_INO);
 361
 362		nilfs_info(sb, "repaired inode bitmap for reserved inodes");
 363	}
 364	ii->i_bh = bh;
 365
 366	atomic64_inc(&root->inodes_count);
 367	inode_init_owner(&init_user_ns, inode, dir, mode);
 368	inode->i_ino = ino;
 369	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
 370
 371	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 372		err = nilfs_bmap_read(ii->i_bmap, NULL);
 373		if (err < 0)
 374			goto failed_after_creation;
 375
 376		set_bit(NILFS_I_BMAP, &ii->i_state);
 377		/* No lock is needed; iget() ensures it. */
 378	}
 379
 380	ii->i_flags = nilfs_mask_flags(
 381		mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 382
 383	/* ii->i_file_acl = 0; */
 384	/* ii->i_dir_acl = 0; */
 385	ii->i_dir_start_lookup = 0;
 386	nilfs_set_inode_flags(inode);
 387	spin_lock(&nilfs->ns_next_gen_lock);
 388	inode->i_generation = nilfs->ns_next_generation++;
 389	spin_unlock(&nilfs->ns_next_gen_lock);
 390	if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 391		err = -EIO;
 392		goto failed_after_creation;
 393	}
 394
 395	err = nilfs_init_acl(inode, dir);
 396	if (unlikely(err))
 397		/*
 398		 * Never occur.  When supporting nilfs_init_acl(),
 399		 * proper cancellation of above jobs should be considered.
 400		 */
 401		goto failed_after_creation;
 402
 403	return inode;
 404
 405 failed_after_creation:
 406	clear_nlink(inode);
 407	if (inode->i_state & I_NEW)
 408		unlock_new_inode(inode);
 409	iput(inode);  /*
 410		       * raw_inode will be deleted through
 411		       * nilfs_evict_inode().
 412		       */
 413	goto failed;
 414
 415 failed_ifile_create_inode:
 416	make_bad_inode(inode);
 417	iput(inode);
 
 418 failed:
 419	return ERR_PTR(err);
 420}
 421
 422void nilfs_set_inode_flags(struct inode *inode)
 423{
 424	unsigned int flags = NILFS_I(inode)->i_flags;
 425	unsigned int new_fl = 0;
 426
 
 
 427	if (flags & FS_SYNC_FL)
 428		new_fl |= S_SYNC;
 429	if (flags & FS_APPEND_FL)
 430		new_fl |= S_APPEND;
 431	if (flags & FS_IMMUTABLE_FL)
 432		new_fl |= S_IMMUTABLE;
 433	if (flags & FS_NOATIME_FL)
 434		new_fl |= S_NOATIME;
 435	if (flags & FS_DIRSYNC_FL)
 436		new_fl |= S_DIRSYNC;
 437	inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 438			S_NOATIME | S_DIRSYNC);
 439}
 440
 441int nilfs_read_inode_common(struct inode *inode,
 442			    struct nilfs_inode *raw_inode)
 443{
 444	struct nilfs_inode_info *ii = NILFS_I(inode);
 445	int err;
 446
 447	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 448	i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 449	i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 450	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 451	inode->i_size = le64_to_cpu(raw_inode->i_size);
 452	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 453	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 454	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 455	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 456	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 457	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 458	if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
 459		return -EIO; /* this inode is for metadata and corrupted */
 460	if (inode->i_nlink == 0)
 461		return -ESTALE; /* this inode is deleted */
 462
 463	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 464	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 465#if 0
 466	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 467	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 468		0 : le32_to_cpu(raw_inode->i_dir_acl);
 469#endif
 470	ii->i_dir_start_lookup = 0;
 471	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 472
 473	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 474	    S_ISLNK(inode->i_mode)) {
 475		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 476		if (err < 0)
 477			return err;
 478		set_bit(NILFS_I_BMAP, &ii->i_state);
 479		/* No lock is needed; iget() ensures it. */
 480	}
 481	return 0;
 482}
 483
 484static int __nilfs_read_inode(struct super_block *sb,
 485			      struct nilfs_root *root, unsigned long ino,
 486			      struct inode *inode)
 487{
 488	struct the_nilfs *nilfs = sb->s_fs_info;
 489	struct buffer_head *bh;
 490	struct nilfs_inode *raw_inode;
 491	int err;
 492
 493	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 494	err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 495	if (unlikely(err))
 496		goto bad_inode;
 497
 498	raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 499
 500	err = nilfs_read_inode_common(inode, raw_inode);
 501	if (err)
 502		goto failed_unmap;
 503
 504	if (S_ISREG(inode->i_mode)) {
 505		inode->i_op = &nilfs_file_inode_operations;
 506		inode->i_fop = &nilfs_file_operations;
 507		inode->i_mapping->a_ops = &nilfs_aops;
 508	} else if (S_ISDIR(inode->i_mode)) {
 509		inode->i_op = &nilfs_dir_inode_operations;
 510		inode->i_fop = &nilfs_dir_operations;
 511		inode->i_mapping->a_ops = &nilfs_aops;
 512	} else if (S_ISLNK(inode->i_mode)) {
 513		inode->i_op = &nilfs_symlink_inode_operations;
 514		inode_nohighmem(inode);
 515		inode->i_mapping->a_ops = &nilfs_aops;
 516	} else {
 517		inode->i_op = &nilfs_special_inode_operations;
 518		init_special_inode(
 519			inode, inode->i_mode,
 520			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 521	}
 522	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 523	brelse(bh);
 524	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 525	nilfs_set_inode_flags(inode);
 526	mapping_set_gfp_mask(inode->i_mapping,
 527			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 528	return 0;
 529
 530 failed_unmap:
 531	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 532	brelse(bh);
 533
 534 bad_inode:
 535	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 536	return err;
 537}
 538
 539static int nilfs_iget_test(struct inode *inode, void *opaque)
 540{
 541	struct nilfs_iget_args *args = opaque;
 542	struct nilfs_inode_info *ii;
 543
 544	if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 545		return 0;
 546
 547	ii = NILFS_I(inode);
 548	if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
 549		if (!args->for_btnc)
 550			return 0;
 551	} else if (args->for_btnc) {
 552		return 0;
 553	}
 554	if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
 555		if (!args->for_shadow)
 556			return 0;
 557	} else if (args->for_shadow) {
 558		return 0;
 559	}
 560
 561	if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 562		return !args->for_gc;
 563
 564	return args->for_gc && args->cno == ii->i_cno;
 565}
 566
 567static int nilfs_iget_set(struct inode *inode, void *opaque)
 568{
 569	struct nilfs_iget_args *args = opaque;
 570
 571	inode->i_ino = args->ino;
 572	NILFS_I(inode)->i_cno = args->cno;
 573	NILFS_I(inode)->i_root = args->root;
 574	if (args->root && args->ino == NILFS_ROOT_INO)
 575		nilfs_get_root(args->root);
 576
 577	if (args->for_gc)
 578		NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
 579	if (args->for_btnc)
 580		NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
 581	if (args->for_shadow)
 582		NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
 583	return 0;
 584}
 585
 586struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 587			    unsigned long ino)
 588{
 589	struct nilfs_iget_args args = {
 590		.ino = ino, .root = root, .cno = 0, .for_gc = false,
 591		.for_btnc = false, .for_shadow = false
 592	};
 593
 594	return ilookup5(sb, ino, nilfs_iget_test, &args);
 595}
 596
 597struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 598				unsigned long ino)
 599{
 600	struct nilfs_iget_args args = {
 601		.ino = ino, .root = root, .cno = 0, .for_gc = false,
 602		.for_btnc = false, .for_shadow = false
 603	};
 604
 605	return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 606}
 607
 608struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 609			 unsigned long ino)
 610{
 611	struct inode *inode;
 612	int err;
 613
 614	inode = nilfs_iget_locked(sb, root, ino);
 615	if (unlikely(!inode))
 616		return ERR_PTR(-ENOMEM);
 617	if (!(inode->i_state & I_NEW))
 618		return inode;
 619
 620	err = __nilfs_read_inode(sb, root, ino, inode);
 621	if (unlikely(err)) {
 622		iget_failed(inode);
 623		return ERR_PTR(err);
 624	}
 625	unlock_new_inode(inode);
 626	return inode;
 627}
 628
 629struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 630				__u64 cno)
 631{
 632	struct nilfs_iget_args args = {
 633		.ino = ino, .root = NULL, .cno = cno, .for_gc = true,
 634		.for_btnc = false, .for_shadow = false
 635	};
 636	struct inode *inode;
 637	int err;
 638
 639	inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 640	if (unlikely(!inode))
 641		return ERR_PTR(-ENOMEM);
 642	if (!(inode->i_state & I_NEW))
 643		return inode;
 644
 645	err = nilfs_init_gcinode(inode);
 646	if (unlikely(err)) {
 647		iget_failed(inode);
 648		return ERR_PTR(err);
 649	}
 650	unlock_new_inode(inode);
 651	return inode;
 652}
 653
 654/**
 655 * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
 656 * @inode: inode object
 657 *
 658 * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
 659 * or does nothing if the inode already has it.  This function allocates
 660 * an additional inode to maintain page cache of B-tree nodes one-on-one.
 661 *
 662 * Return Value: On success, 0 is returned. On errors, one of the following
 663 * negative error code is returned.
 664 *
 665 * %-ENOMEM - Insufficient memory available.
 666 */
 667int nilfs_attach_btree_node_cache(struct inode *inode)
 668{
 669	struct nilfs_inode_info *ii = NILFS_I(inode);
 670	struct inode *btnc_inode;
 671	struct nilfs_iget_args args;
 672
 673	if (ii->i_assoc_inode)
 674		return 0;
 675
 676	args.ino = inode->i_ino;
 677	args.root = ii->i_root;
 678	args.cno = ii->i_cno;
 679	args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
 680	args.for_btnc = true;
 681	args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
 682
 683	btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 684				  nilfs_iget_set, &args);
 685	if (unlikely(!btnc_inode))
 686		return -ENOMEM;
 687	if (btnc_inode->i_state & I_NEW) {
 688		nilfs_init_btnc_inode(btnc_inode);
 689		unlock_new_inode(btnc_inode);
 690	}
 691	NILFS_I(btnc_inode)->i_assoc_inode = inode;
 692	NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
 693	ii->i_assoc_inode = btnc_inode;
 694
 695	return 0;
 696}
 697
 698/**
 699 * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
 700 * @inode: inode object
 701 *
 702 * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
 703 * holder inode bound to @inode, or does nothing if @inode doesn't have it.
 704 */
 705void nilfs_detach_btree_node_cache(struct inode *inode)
 706{
 707	struct nilfs_inode_info *ii = NILFS_I(inode);
 708	struct inode *btnc_inode = ii->i_assoc_inode;
 709
 710	if (btnc_inode) {
 711		NILFS_I(btnc_inode)->i_assoc_inode = NULL;
 712		ii->i_assoc_inode = NULL;
 713		iput(btnc_inode);
 714	}
 715}
 716
 717/**
 718 * nilfs_iget_for_shadow - obtain inode for shadow mapping
 719 * @inode: inode object that uses shadow mapping
 720 *
 721 * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
 722 * caches for shadow mapping.  The page cache for data pages is set up
 723 * in one inode and the one for b-tree node pages is set up in the
 724 * other inode, which is attached to the former inode.
 725 *
 726 * Return Value: On success, a pointer to the inode for data pages is
 727 * returned. On errors, one of the following negative error code is returned
 728 * in a pointer type.
 729 *
 730 * %-ENOMEM - Insufficient memory available.
 731 */
 732struct inode *nilfs_iget_for_shadow(struct inode *inode)
 733{
 734	struct nilfs_iget_args args = {
 735		.ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
 736		.for_btnc = false, .for_shadow = true
 737	};
 738	struct inode *s_inode;
 739	int err;
 740
 741	s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
 742			       nilfs_iget_set, &args);
 743	if (unlikely(!s_inode))
 744		return ERR_PTR(-ENOMEM);
 745	if (!(s_inode->i_state & I_NEW))
 746		return inode;
 747
 748	NILFS_I(s_inode)->i_flags = 0;
 749	memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
 750	mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
 751
 752	err = nilfs_attach_btree_node_cache(s_inode);
 753	if (unlikely(err)) {
 754		iget_failed(s_inode);
 755		return ERR_PTR(err);
 756	}
 757	unlock_new_inode(s_inode);
 758	return s_inode;
 759}
 760
 761void nilfs_write_inode_common(struct inode *inode,
 762			      struct nilfs_inode *raw_inode, int has_bmap)
 763{
 764	struct nilfs_inode_info *ii = NILFS_I(inode);
 765
 766	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 767	raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 768	raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 769	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 770	raw_inode->i_size = cpu_to_le64(inode->i_size);
 771	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 772	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 773	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 774	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 775	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 776
 777	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 778	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 779
 780	if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 781		struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 782
 783		/* zero-fill unused portion in the case of super root block */
 784		raw_inode->i_xattr = 0;
 785		raw_inode->i_pad = 0;
 786		memset((void *)raw_inode + sizeof(*raw_inode), 0,
 787		       nilfs->ns_inode_size - sizeof(*raw_inode));
 788	}
 789
 790	if (has_bmap)
 791		nilfs_bmap_write(ii->i_bmap, raw_inode);
 792	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 793		raw_inode->i_device_code =
 794			cpu_to_le64(huge_encode_dev(inode->i_rdev));
 795	/*
 796	 * When extending inode, nilfs->ns_inode_size should be checked
 797	 * for substitutions of appended fields.
 798	 */
 799}
 800
 801void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 802{
 803	ino_t ino = inode->i_ino;
 804	struct nilfs_inode_info *ii = NILFS_I(inode);
 805	struct inode *ifile = ii->i_root->ifile;
 806	struct nilfs_inode *raw_inode;
 807
 808	raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 809
 810	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 811		memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 812	if (flags & I_DIRTY_DATASYNC)
 813		set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 814
 815	nilfs_write_inode_common(inode, raw_inode, 0);
 816		/*
 817		 * XXX: call with has_bmap = 0 is a workaround to avoid
 818		 * deadlock of bmap.  This delays update of i_bmap to just
 819		 * before writing.
 820		 */
 821
 822	nilfs_ifile_unmap_inode(ifile, ino, ibh);
 823}
 824
 825#define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
 826
 827static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 828				unsigned long from)
 829{
 830	__u64 b;
 831	int ret;
 832
 833	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 834		return;
 835repeat:
 836	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 837	if (ret == -ENOENT)
 838		return;
 839	else if (ret < 0)
 840		goto failed;
 841
 842	if (b < from)
 843		return;
 844
 845	b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 846	ret = nilfs_bmap_truncate(ii->i_bmap, b);
 847	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 848	if (!ret || (ret == -ENOMEM &&
 849		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 850		goto repeat;
 851
 852failed:
 853	nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
 854		   ret, ii->vfs_inode.i_ino);
 
 855}
 856
 857void nilfs_truncate(struct inode *inode)
 858{
 859	unsigned long blkoff;
 860	unsigned int blocksize;
 861	struct nilfs_transaction_info ti;
 862	struct super_block *sb = inode->i_sb;
 863	struct nilfs_inode_info *ii = NILFS_I(inode);
 864
 865	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 866		return;
 867	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 868		return;
 869
 870	blocksize = sb->s_blocksize;
 871	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 872	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 873
 874	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 875
 876	nilfs_truncate_bmap(ii, blkoff);
 877
 878	inode->i_mtime = inode->i_ctime = current_time(inode);
 879	if (IS_SYNC(inode))
 880		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 881
 882	nilfs_mark_inode_dirty(inode);
 883	nilfs_set_file_dirty(inode, 0);
 884	nilfs_transaction_commit(sb);
 885	/*
 886	 * May construct a logical segment and may fail in sync mode.
 887	 * But truncate has no return value.
 888	 */
 889}
 890
 891static void nilfs_clear_inode(struct inode *inode)
 892{
 893	struct nilfs_inode_info *ii = NILFS_I(inode);
 
 894
 895	/*
 896	 * Free resources allocated in nilfs_read_inode(), here.
 897	 */
 898	BUG_ON(!list_empty(&ii->i_dirty));
 899	brelse(ii->i_bh);
 900	ii->i_bh = NULL;
 901
 902	if (nilfs_is_metadata_file_inode(inode))
 903		nilfs_mdt_clear(inode);
 904
 905	if (test_bit(NILFS_I_BMAP, &ii->i_state))
 906		nilfs_bmap_clear(ii->i_bmap);
 907
 908	if (!test_bit(NILFS_I_BTNC, &ii->i_state))
 909		nilfs_detach_btree_node_cache(inode);
 910
 911	if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 912		nilfs_put_root(ii->i_root);
 913}
 914
 915void nilfs_evict_inode(struct inode *inode)
 916{
 917	struct nilfs_transaction_info ti;
 918	struct super_block *sb = inode->i_sb;
 919	struct nilfs_inode_info *ii = NILFS_I(inode);
 920	int ret;
 921
 922	if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 923		truncate_inode_pages_final(&inode->i_data);
 924		clear_inode(inode);
 
 925		nilfs_clear_inode(inode);
 926		return;
 927	}
 928	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 929
 930	truncate_inode_pages_final(&inode->i_data);
 
 931
 932	/* TODO: some of the following operations may fail.  */
 933	nilfs_truncate_bmap(ii, 0);
 934	nilfs_mark_inode_dirty(inode);
 935	clear_inode(inode);
 936
 937	ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 938	if (!ret)
 939		atomic64_dec(&ii->i_root->inodes_count);
 940
 941	nilfs_clear_inode(inode);
 942
 943	if (IS_SYNC(inode))
 944		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 945	nilfs_transaction_commit(sb);
 946	/*
 947	 * May construct a logical segment and may fail in sync mode.
 948	 * But delete_inode has no return value.
 949	 */
 950}
 951
 952int nilfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 953		  struct iattr *iattr)
 954{
 955	struct nilfs_transaction_info ti;
 956	struct inode *inode = d_inode(dentry);
 957	struct super_block *sb = inode->i_sb;
 958	int err;
 959
 960	err = setattr_prepare(&init_user_ns, dentry, iattr);
 961	if (err)
 962		return err;
 963
 964	err = nilfs_transaction_begin(sb, &ti, 0);
 965	if (unlikely(err))
 966		return err;
 967
 968	if ((iattr->ia_valid & ATTR_SIZE) &&
 969	    iattr->ia_size != i_size_read(inode)) {
 970		inode_dio_wait(inode);
 971		truncate_setsize(inode, iattr->ia_size);
 972		nilfs_truncate(inode);
 
 
 973	}
 974
 975	setattr_copy(&init_user_ns, inode, iattr);
 976	mark_inode_dirty(inode);
 977
 978	if (iattr->ia_valid & ATTR_MODE) {
 979		err = nilfs_acl_chmod(inode);
 980		if (unlikely(err))
 981			goto out_err;
 982	}
 983
 984	return nilfs_transaction_commit(sb);
 985
 986out_err:
 987	nilfs_transaction_abort(sb);
 988	return err;
 989}
 990
 991int nilfs_permission(struct user_namespace *mnt_userns, struct inode *inode,
 992		     int mask)
 993{
 994	struct nilfs_root *root = NILFS_I(inode)->i_root;
 995
 996	if ((mask & MAY_WRITE) && root &&
 997	    root->cno != NILFS_CPTREE_CURRENT_CNO)
 998		return -EROFS; /* snapshot is not writable */
 999
1000	return generic_permission(&init_user_ns, inode, mask);
1001}
1002
1003int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1004{
1005	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1006	struct nilfs_inode_info *ii = NILFS_I(inode);
1007	int err;
1008
1009	spin_lock(&nilfs->ns_inode_lock);
1010	if (ii->i_bh == NULL) {
1011		spin_unlock(&nilfs->ns_inode_lock);
1012		err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1013						  inode->i_ino, pbh);
1014		if (unlikely(err))
1015			return err;
1016		spin_lock(&nilfs->ns_inode_lock);
1017		if (ii->i_bh == NULL)
1018			ii->i_bh = *pbh;
1019		else {
1020			brelse(*pbh);
1021			*pbh = ii->i_bh;
1022		}
1023	} else
1024		*pbh = ii->i_bh;
1025
1026	get_bh(*pbh);
1027	spin_unlock(&nilfs->ns_inode_lock);
1028	return 0;
1029}
1030
1031int nilfs_inode_dirty(struct inode *inode)
1032{
1033	struct nilfs_inode_info *ii = NILFS_I(inode);
1034	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1035	int ret = 0;
1036
1037	if (!list_empty(&ii->i_dirty)) {
1038		spin_lock(&nilfs->ns_inode_lock);
1039		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1040			test_bit(NILFS_I_BUSY, &ii->i_state);
1041		spin_unlock(&nilfs->ns_inode_lock);
1042	}
1043	return ret;
1044}
1045
1046int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1047{
1048	struct nilfs_inode_info *ii = NILFS_I(inode);
1049	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1050
1051	atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1052
1053	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1054		return 0;
1055
1056	spin_lock(&nilfs->ns_inode_lock);
1057	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1058	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1059		/*
1060		 * Because this routine may race with nilfs_dispose_list(),
1061		 * we have to check NILFS_I_QUEUED here, too.
1062		 */
1063		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1064			/*
1065			 * This will happen when somebody is freeing
1066			 * this inode.
1067			 */
1068			nilfs_warn(inode->i_sb,
1069				   "cannot set file dirty (ino=%lu): the file is being freed",
1070				   inode->i_ino);
1071			spin_unlock(&nilfs->ns_inode_lock);
1072			return -EINVAL; /*
1073					 * NILFS_I_DIRTY may remain for
1074					 * freeing inode.
1075					 */
1076		}
1077		list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1078		set_bit(NILFS_I_QUEUED, &ii->i_state);
1079	}
1080	spin_unlock(&nilfs->ns_inode_lock);
1081	return 0;
1082}
1083
1084int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1085{
1086	struct buffer_head *ibh;
1087	int err;
1088
1089	err = nilfs_load_inode_block(inode, &ibh);
1090	if (unlikely(err)) {
1091		nilfs_warn(inode->i_sb,
1092			   "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1093			   inode->i_ino, err);
1094		return err;
1095	}
1096	nilfs_update_inode(inode, ibh, flags);
1097	mark_buffer_dirty(ibh);
1098	nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1099	brelse(ibh);
1100	return 0;
1101}
1102
1103/**
1104 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1105 * @inode: inode of the file to be registered.
1106 * @flags: flags to determine the dirty state of the inode
1107 *
1108 * nilfs_dirty_inode() loads a inode block containing the specified
1109 * @inode and copies data from a nilfs_inode to a corresponding inode
1110 * entry in the inode block. This operation is excluded from the segment
1111 * construction. This function can be called both as a single operation
1112 * and as a part of indivisible file operations.
1113 */
1114void nilfs_dirty_inode(struct inode *inode, int flags)
1115{
1116	struct nilfs_transaction_info ti;
1117	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1118
1119	if (is_bad_inode(inode)) {
1120		nilfs_warn(inode->i_sb,
1121			   "tried to mark bad_inode dirty. ignored.");
1122		dump_stack();
1123		return;
1124	}
1125	if (mdi) {
1126		nilfs_mdt_mark_dirty(inode);
1127		return;
1128	}
1129	nilfs_transaction_begin(inode->i_sb, &ti, 0);
1130	__nilfs_mark_inode_dirty(inode, flags);
1131	nilfs_transaction_commit(inode->i_sb); /* never fails */
1132}
1133
1134int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1135		 __u64 start, __u64 len)
1136{
1137	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1138	__u64 logical = 0, phys = 0, size = 0;
1139	__u32 flags = 0;
1140	loff_t isize;
1141	sector_t blkoff, end_blkoff;
1142	sector_t delalloc_blkoff;
1143	unsigned long delalloc_blklen;
1144	unsigned int blkbits = inode->i_blkbits;
1145	int ret, n;
1146
1147	ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1148	if (ret)
1149		return ret;
1150
1151	inode_lock(inode);
1152
1153	isize = i_size_read(inode);
1154
1155	blkoff = start >> blkbits;
1156	end_blkoff = (start + len - 1) >> blkbits;
1157
1158	delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1159							&delalloc_blkoff);
1160
1161	do {
1162		__u64 blkphy;
1163		unsigned int maxblocks;
1164
1165		if (delalloc_blklen && blkoff == delalloc_blkoff) {
1166			if (size) {
1167				/* End of the current extent */
1168				ret = fiemap_fill_next_extent(
1169					fieinfo, logical, phys, size, flags);
1170				if (ret)
1171					break;
1172			}
1173			if (blkoff > end_blkoff)
1174				break;
1175
1176			flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1177			logical = blkoff << blkbits;
1178			phys = 0;
1179			size = delalloc_blklen << blkbits;
1180
1181			blkoff = delalloc_blkoff + delalloc_blklen;
1182			delalloc_blklen = nilfs_find_uncommitted_extent(
1183				inode, blkoff, &delalloc_blkoff);
1184			continue;
1185		}
1186
1187		/*
1188		 * Limit the number of blocks that we look up so as
1189		 * not to get into the next delayed allocation extent.
1190		 */
1191		maxblocks = INT_MAX;
1192		if (delalloc_blklen)
1193			maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1194					  maxblocks);
1195		blkphy = 0;
1196
1197		down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1198		n = nilfs_bmap_lookup_contig(
1199			NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1200		up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1201
1202		if (n < 0) {
1203			int past_eof;
1204
1205			if (unlikely(n != -ENOENT))
1206				break; /* error */
1207
1208			/* HOLE */
1209			blkoff++;
1210			past_eof = ((blkoff << blkbits) >= isize);
1211
1212			if (size) {
1213				/* End of the current extent */
1214
1215				if (past_eof)
1216					flags |= FIEMAP_EXTENT_LAST;
1217
1218				ret = fiemap_fill_next_extent(
1219					fieinfo, logical, phys, size, flags);
1220				if (ret)
1221					break;
1222				size = 0;
1223			}
1224			if (blkoff > end_blkoff || past_eof)
1225				break;
1226		} else {
1227			if (size) {
1228				if (phys && blkphy << blkbits == phys + size) {
1229					/* The current extent goes on */
1230					size += n << blkbits;
1231				} else {
1232					/* Terminate the current extent */
1233					ret = fiemap_fill_next_extent(
1234						fieinfo, logical, phys, size,
1235						flags);
1236					if (ret || blkoff > end_blkoff)
1237						break;
1238
1239					/* Start another extent */
1240					flags = FIEMAP_EXTENT_MERGED;
1241					logical = blkoff << blkbits;
1242					phys = blkphy << blkbits;
1243					size = n << blkbits;
1244				}
1245			} else {
1246				/* Start a new extent */
1247				flags = FIEMAP_EXTENT_MERGED;
1248				logical = blkoff << blkbits;
1249				phys = blkphy << blkbits;
1250				size = n << blkbits;
1251			}
1252			blkoff += n;
1253		}
1254		cond_resched();
1255	} while (true);
1256
1257	/* If ret is 1 then we just hit the end of the extent array */
1258	if (ret == 1)
1259		ret = 0;
1260
1261	inode_unlock(inode);
1262	return ret;
1263}
v3.1
 
   1/*
   2 * inode.c - NILFS inode operations.
   3 *
   4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5 *
   6 * This program is free software; you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License as published by
   8 * the Free Software Foundation; either version 2 of the License, or
   9 * (at your option) any later version.
  10 *
  11 * This program is distributed in the hope that it will be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with this program; if not, write to the Free Software
  18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19 *
  20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21 *
  22 */
  23
  24#include <linux/buffer_head.h>
  25#include <linux/gfp.h>
  26#include <linux/mpage.h>
 
  27#include <linux/writeback.h>
  28#include <linux/uio.h>
 
  29#include "nilfs.h"
  30#include "btnode.h"
  31#include "segment.h"
  32#include "page.h"
  33#include "mdt.h"
  34#include "cpfile.h"
  35#include "ifile.h"
  36
 
 
 
 
 
 
 
 
 
  37struct nilfs_iget_args {
  38	u64 ino;
  39	__u64 cno;
  40	struct nilfs_root *root;
  41	int for_gc;
 
 
  42};
  43
 
 
  44void nilfs_inode_add_blocks(struct inode *inode, int n)
  45{
  46	struct nilfs_root *root = NILFS_I(inode)->i_root;
  47
  48	inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  49	if (root)
  50		atomic_add(n, &root->blocks_count);
  51}
  52
  53void nilfs_inode_sub_blocks(struct inode *inode, int n)
  54{
  55	struct nilfs_root *root = NILFS_I(inode)->i_root;
  56
  57	inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  58	if (root)
  59		atomic_sub(n, &root->blocks_count);
  60}
  61
  62/**
  63 * nilfs_get_block() - get a file block on the filesystem (callback function)
  64 * @inode - inode struct of the target file
  65 * @blkoff - file block number
  66 * @bh_result - buffer head to be mapped on
  67 * @create - indicate whether allocating the block or not when it has not
  68 *      been allocated yet.
  69 *
  70 * This function does not issue actual read request of the specified data
  71 * block. It is done by VFS.
  72 */
  73int nilfs_get_block(struct inode *inode, sector_t blkoff,
  74		    struct buffer_head *bh_result, int create)
  75{
  76	struct nilfs_inode_info *ii = NILFS_I(inode);
  77	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  78	__u64 blknum = 0;
  79	int err = 0, ret;
  80	unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  81
  82	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  83	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  84	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  85	if (ret >= 0) {	/* found */
  86		map_bh(bh_result, inode->i_sb, blknum);
  87		if (ret > 0)
  88			bh_result->b_size = (ret << inode->i_blkbits);
  89		goto out;
  90	}
  91	/* data block was not found */
  92	if (ret == -ENOENT && create) {
  93		struct nilfs_transaction_info ti;
  94
  95		bh_result->b_blocknr = 0;
  96		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  97		if (unlikely(err))
  98			goto out;
  99		err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 100					(unsigned long)bh_result);
 101		if (unlikely(err != 0)) {
 102			if (err == -EEXIST) {
 103				/*
 104				 * The get_block() function could be called
 105				 * from multiple callers for an inode.
 106				 * However, the page having this block must
 107				 * be locked in this case.
 108				 */
 109				printk(KERN_WARNING
 110				       "nilfs_get_block: a race condition "
 111				       "while inserting a data block. "
 112				       "(inode number=%lu, file block "
 113				       "offset=%llu)\n",
 114				       inode->i_ino,
 115				       (unsigned long long)blkoff);
 116				err = 0;
 117			}
 118			nilfs_transaction_abort(inode->i_sb);
 119			goto out;
 120		}
 121		nilfs_mark_inode_dirty(inode);
 122		nilfs_transaction_commit(inode->i_sb); /* never fails */
 123		/* Error handling should be detailed */
 124		set_buffer_new(bh_result);
 125		set_buffer_delay(bh_result);
 126		map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 127						      to proper value */
 
 128	} else if (ret == -ENOENT) {
 129		/* not found is not error (e.g. hole); must return without
 130		   the mapped state flag. */
 
 
 131		;
 132	} else {
 133		err = ret;
 134	}
 135
 136 out:
 137	return err;
 138}
 139
 140/**
 141 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 142 * address_space_operations.
 143 * @file - file struct of the file to be read
 144 * @page - the page to be read
 145 */
 146static int nilfs_readpage(struct file *file, struct page *page)
 147{
 148	return mpage_readpage(page, nilfs_get_block);
 149}
 150
 151/**
 152 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 153 * address_space_operations.
 154 * @file - file struct of the file to be read
 155 * @mapping - address_space struct used for reading multiple pages
 156 * @pages - the pages to be read
 157 * @nr_pages - number of pages to be read
 158 */
 159static int nilfs_readpages(struct file *file, struct address_space *mapping,
 160			   struct list_head *pages, unsigned nr_pages)
 161{
 162	return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 163}
 164
 165static int nilfs_writepages(struct address_space *mapping,
 166			    struct writeback_control *wbc)
 167{
 168	struct inode *inode = mapping->host;
 169	int err = 0;
 170
 
 
 
 
 
 171	if (wbc->sync_mode == WB_SYNC_ALL)
 172		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 173						    wbc->range_start,
 174						    wbc->range_end);
 175	return err;
 176}
 177
 178static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 179{
 180	struct inode *inode = page->mapping->host;
 181	int err;
 182
 
 
 
 
 
 
 
 
 
 
 
 
 183	redirty_page_for_writepage(wbc, page);
 184	unlock_page(page);
 185
 186	if (wbc->sync_mode == WB_SYNC_ALL) {
 187		err = nilfs_construct_segment(inode->i_sb);
 188		if (unlikely(err))
 189			return err;
 190	} else if (wbc->for_reclaim)
 191		nilfs_flush_segment(inode->i_sb, inode->i_ino);
 192
 193	return 0;
 194}
 195
 196static int nilfs_set_page_dirty(struct page *page)
 
 197{
 198	int ret = __set_page_dirty_buffers(page);
 
 
 
 199
 200	if (ret) {
 201		struct inode *inode = page->mapping->host;
 202		unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 203
 
 204		nilfs_set_file_dirty(inode, nr_dirty);
 
 
 
 
 
 
 
 
 
 
 205	}
 206	return ret;
 207}
 208
 209static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 210			     loff_t pos, unsigned len, unsigned flags,
 211			     struct page **pagep, void **fsdata)
 212
 213{
 214	struct inode *inode = mapping->host;
 215	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 216
 217	if (unlikely(err))
 218		return err;
 219
 220	err = block_write_begin(mapping, pos, len, flags, pagep,
 221				nilfs_get_block);
 222	if (unlikely(err)) {
 223		loff_t isize = mapping->host->i_size;
 224		if (pos + len > isize)
 225			vmtruncate(mapping->host, isize);
 226
 227		nilfs_transaction_abort(inode->i_sb);
 228	}
 229	return err;
 230}
 231
 232static int nilfs_write_end(struct file *file, struct address_space *mapping,
 233			   loff_t pos, unsigned len, unsigned copied,
 234			   struct page *page, void *fsdata)
 235{
 236	struct inode *inode = mapping->host;
 237	unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 238	unsigned nr_dirty;
 239	int err;
 240
 241	nr_dirty = nilfs_page_count_clean_buffers(page, start,
 242						  start + copied);
 243	copied = generic_write_end(file, mapping, pos, len, copied, page,
 244				   fsdata);
 245	nilfs_set_file_dirty(inode, nr_dirty);
 246	err = nilfs_transaction_commit(inode->i_sb);
 247	return err ? : copied;
 248}
 249
 250static ssize_t
 251nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 252		loff_t offset, unsigned long nr_segs)
 253{
 254	struct file *file = iocb->ki_filp;
 255	struct inode *inode = file->f_mapping->host;
 256	ssize_t size;
 257
 258	if (rw == WRITE)
 259		return 0;
 260
 261	/* Needs synchronization with the cleaner */
 262	size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 263				  nilfs_get_block);
 264
 265	/*
 266	 * In case of error extending write may have instantiated a few
 267	 * blocks outside i_size. Trim these off again.
 268	 */
 269	if (unlikely((rw & WRITE) && size < 0)) {
 270		loff_t isize = i_size_read(inode);
 271		loff_t end = offset + iov_length(iov, nr_segs);
 272
 273		if (end > isize)
 274			vmtruncate(inode, isize);
 275	}
 276
 277	return size;
 278}
 279
 280const struct address_space_operations nilfs_aops = {
 281	.writepage		= nilfs_writepage,
 282	.readpage		= nilfs_readpage,
 283	.writepages		= nilfs_writepages,
 284	.set_page_dirty		= nilfs_set_page_dirty,
 285	.readpages		= nilfs_readpages,
 286	.write_begin		= nilfs_write_begin,
 287	.write_end		= nilfs_write_end,
 288	/* .releasepage		= nilfs_releasepage, */
 289	.invalidatepage		= block_invalidatepage,
 290	.direct_IO		= nilfs_direct_IO,
 291	.is_partially_uptodate  = block_is_partially_uptodate,
 292};
 293
 294struct inode *nilfs_new_inode(struct inode *dir, int mode)
 
 
 
 
 
 
 
 
 
 
 
 
 295{
 296	struct super_block *sb = dir->i_sb;
 297	struct the_nilfs *nilfs = sb->s_fs_info;
 298	struct inode *inode;
 299	struct nilfs_inode_info *ii;
 300	struct nilfs_root *root;
 
 301	int err = -ENOMEM;
 302	ino_t ino;
 303
 304	inode = new_inode(sb);
 305	if (unlikely(!inode))
 306		goto failed;
 307
 308	mapping_set_gfp_mask(inode->i_mapping,
 309			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 310
 311	root = NILFS_I(dir)->i_root;
 312	ii = NILFS_I(inode);
 313	ii->i_state = 1 << NILFS_I_NEW;
 314	ii->i_root = root;
 315
 316	err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 317	if (unlikely(err))
 318		goto failed_ifile_create_inode;
 319	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
 320
 321	atomic_inc(&root->inodes_count);
 322	inode_init_owner(inode, dir, mode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 323	inode->i_ino = ino;
 324	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 325
 326	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 327		err = nilfs_bmap_read(ii->i_bmap, NULL);
 328		if (err < 0)
 329			goto failed_bmap;
 330
 331		set_bit(NILFS_I_BMAP, &ii->i_state);
 332		/* No lock is needed; iget() ensures it. */
 333	}
 334
 335	ii->i_flags = nilfs_mask_flags(
 336		mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 337
 338	/* ii->i_file_acl = 0; */
 339	/* ii->i_dir_acl = 0; */
 340	ii->i_dir_start_lookup = 0;
 341	nilfs_set_inode_flags(inode);
 342	spin_lock(&nilfs->ns_next_gen_lock);
 343	inode->i_generation = nilfs->ns_next_generation++;
 344	spin_unlock(&nilfs->ns_next_gen_lock);
 345	insert_inode_hash(inode);
 
 
 
 346
 347	err = nilfs_init_acl(inode, dir);
 348	if (unlikely(err))
 349		goto failed_acl; /* never occur. When supporting
 350				    nilfs_init_acl(), proper cancellation of
 351				    above jobs should be considered */
 
 
 352
 353	return inode;
 354
 355 failed_acl:
 356 failed_bmap:
 357	inode->i_nlink = 0;
 358	iput(inode);  /* raw_inode will be deleted through
 359			 generic_delete_inode() */
 
 
 
 360	goto failed;
 361
 362 failed_ifile_create_inode:
 363	make_bad_inode(inode);
 364	iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 365			 called */
 366 failed:
 367	return ERR_PTR(err);
 368}
 369
 370void nilfs_set_inode_flags(struct inode *inode)
 371{
 372	unsigned int flags = NILFS_I(inode)->i_flags;
 
 373
 374	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 375			    S_DIRSYNC);
 376	if (flags & FS_SYNC_FL)
 377		inode->i_flags |= S_SYNC;
 378	if (flags & FS_APPEND_FL)
 379		inode->i_flags |= S_APPEND;
 380	if (flags & FS_IMMUTABLE_FL)
 381		inode->i_flags |= S_IMMUTABLE;
 382	if (flags & FS_NOATIME_FL)
 383		inode->i_flags |= S_NOATIME;
 384	if (flags & FS_DIRSYNC_FL)
 385		inode->i_flags |= S_DIRSYNC;
 386	mapping_set_gfp_mask(inode->i_mapping,
 387			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 388}
 389
 390int nilfs_read_inode_common(struct inode *inode,
 391			    struct nilfs_inode *raw_inode)
 392{
 393	struct nilfs_inode_info *ii = NILFS_I(inode);
 394	int err;
 395
 396	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 397	inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 398	inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 399	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 400	inode->i_size = le64_to_cpu(raw_inode->i_size);
 401	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 402	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 403	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 404	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 405	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 406	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 407	if (inode->i_nlink == 0 && inode->i_mode == 0)
 408		return -EINVAL; /* this inode is deleted */
 
 
 409
 410	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 411	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 412#if 0
 413	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 414	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 415		0 : le32_to_cpu(raw_inode->i_dir_acl);
 416#endif
 417	ii->i_dir_start_lookup = 0;
 418	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 419
 420	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 421	    S_ISLNK(inode->i_mode)) {
 422		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 423		if (err < 0)
 424			return err;
 425		set_bit(NILFS_I_BMAP, &ii->i_state);
 426		/* No lock is needed; iget() ensures it. */
 427	}
 428	return 0;
 429}
 430
 431static int __nilfs_read_inode(struct super_block *sb,
 432			      struct nilfs_root *root, unsigned long ino,
 433			      struct inode *inode)
 434{
 435	struct the_nilfs *nilfs = sb->s_fs_info;
 436	struct buffer_head *bh;
 437	struct nilfs_inode *raw_inode;
 438	int err;
 439
 440	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 441	err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 442	if (unlikely(err))
 443		goto bad_inode;
 444
 445	raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 446
 447	err = nilfs_read_inode_common(inode, raw_inode);
 448	if (err)
 449		goto failed_unmap;
 450
 451	if (S_ISREG(inode->i_mode)) {
 452		inode->i_op = &nilfs_file_inode_operations;
 453		inode->i_fop = &nilfs_file_operations;
 454		inode->i_mapping->a_ops = &nilfs_aops;
 455	} else if (S_ISDIR(inode->i_mode)) {
 456		inode->i_op = &nilfs_dir_inode_operations;
 457		inode->i_fop = &nilfs_dir_operations;
 458		inode->i_mapping->a_ops = &nilfs_aops;
 459	} else if (S_ISLNK(inode->i_mode)) {
 460		inode->i_op = &nilfs_symlink_inode_operations;
 
 461		inode->i_mapping->a_ops = &nilfs_aops;
 462	} else {
 463		inode->i_op = &nilfs_special_inode_operations;
 464		init_special_inode(
 465			inode, inode->i_mode,
 466			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 467	}
 468	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 469	brelse(bh);
 470	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 471	nilfs_set_inode_flags(inode);
 
 
 472	return 0;
 473
 474 failed_unmap:
 475	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 476	brelse(bh);
 477
 478 bad_inode:
 479	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 480	return err;
 481}
 482
 483static int nilfs_iget_test(struct inode *inode, void *opaque)
 484{
 485	struct nilfs_iget_args *args = opaque;
 486	struct nilfs_inode_info *ii;
 487
 488	if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 489		return 0;
 490
 491	ii = NILFS_I(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 492	if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 493		return !args->for_gc;
 494
 495	return args->for_gc && args->cno == ii->i_cno;
 496}
 497
 498static int nilfs_iget_set(struct inode *inode, void *opaque)
 499{
 500	struct nilfs_iget_args *args = opaque;
 501
 502	inode->i_ino = args->ino;
 503	if (args->for_gc) {
 504		NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 505		NILFS_I(inode)->i_cno = args->cno;
 506		NILFS_I(inode)->i_root = NULL;
 507	} else {
 508		if (args->root && args->ino == NILFS_ROOT_INO)
 509			nilfs_get_root(args->root);
 510		NILFS_I(inode)->i_root = args->root;
 511	}
 
 
 512	return 0;
 513}
 514
 515struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 516			    unsigned long ino)
 517{
 518	struct nilfs_iget_args args = {
 519		.ino = ino, .root = root, .cno = 0, .for_gc = 0
 
 520	};
 521
 522	return ilookup5(sb, ino, nilfs_iget_test, &args);
 523}
 524
 525struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 526				unsigned long ino)
 527{
 528	struct nilfs_iget_args args = {
 529		.ino = ino, .root = root, .cno = 0, .for_gc = 0
 
 530	};
 531
 532	return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 533}
 534
 535struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 536			 unsigned long ino)
 537{
 538	struct inode *inode;
 539	int err;
 540
 541	inode = nilfs_iget_locked(sb, root, ino);
 542	if (unlikely(!inode))
 543		return ERR_PTR(-ENOMEM);
 544	if (!(inode->i_state & I_NEW))
 545		return inode;
 546
 547	err = __nilfs_read_inode(sb, root, ino, inode);
 548	if (unlikely(err)) {
 549		iget_failed(inode);
 550		return ERR_PTR(err);
 551	}
 552	unlock_new_inode(inode);
 553	return inode;
 554}
 555
 556struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 557				__u64 cno)
 558{
 559	struct nilfs_iget_args args = {
 560		.ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 
 561	};
 562	struct inode *inode;
 563	int err;
 564
 565	inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 566	if (unlikely(!inode))
 567		return ERR_PTR(-ENOMEM);
 568	if (!(inode->i_state & I_NEW))
 569		return inode;
 570
 571	err = nilfs_init_gcinode(inode);
 572	if (unlikely(err)) {
 573		iget_failed(inode);
 574		return ERR_PTR(err);
 575	}
 576	unlock_new_inode(inode);
 577	return inode;
 578}
 579
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 580void nilfs_write_inode_common(struct inode *inode,
 581			      struct nilfs_inode *raw_inode, int has_bmap)
 582{
 583	struct nilfs_inode_info *ii = NILFS_I(inode);
 584
 585	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 586	raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 587	raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 588	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 589	raw_inode->i_size = cpu_to_le64(inode->i_size);
 590	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 591	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 592	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 593	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 594	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 595
 596	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 597	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 598
 599	if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 600		struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 601
 602		/* zero-fill unused portion in the case of super root block */
 603		raw_inode->i_xattr = 0;
 604		raw_inode->i_pad = 0;
 605		memset((void *)raw_inode + sizeof(*raw_inode), 0,
 606		       nilfs->ns_inode_size - sizeof(*raw_inode));
 607	}
 608
 609	if (has_bmap)
 610		nilfs_bmap_write(ii->i_bmap, raw_inode);
 611	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 612		raw_inode->i_device_code =
 613			cpu_to_le64(huge_encode_dev(inode->i_rdev));
 614	/* When extending inode, nilfs->ns_inode_size should be checked
 615	   for substitutions of appended fields */
 
 
 616}
 617
 618void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 619{
 620	ino_t ino = inode->i_ino;
 621	struct nilfs_inode_info *ii = NILFS_I(inode);
 622	struct inode *ifile = ii->i_root->ifile;
 623	struct nilfs_inode *raw_inode;
 624
 625	raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 626
 627	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 628		memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 629	set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 
 630
 631	nilfs_write_inode_common(inode, raw_inode, 0);
 632		/* XXX: call with has_bmap = 0 is a workaround to avoid
 633		   deadlock of bmap. This delays update of i_bmap to just
 634		   before writing */
 
 
 
 635	nilfs_ifile_unmap_inode(ifile, ino, ibh);
 636}
 637
 638#define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
 639
 640static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 641				unsigned long from)
 642{
 643	unsigned long b;
 644	int ret;
 645
 646	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 647		return;
 648repeat:
 649	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 650	if (ret == -ENOENT)
 651		return;
 652	else if (ret < 0)
 653		goto failed;
 654
 655	if (b < from)
 656		return;
 657
 658	b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 659	ret = nilfs_bmap_truncate(ii->i_bmap, b);
 660	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 661	if (!ret || (ret == -ENOMEM &&
 662		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 663		goto repeat;
 664
 665failed:
 666	nilfs_warning(ii->vfs_inode.i_sb, __func__,
 667		      "failed to truncate bmap (ino=%lu, err=%d)",
 668		      ii->vfs_inode.i_ino, ret);
 669}
 670
 671void nilfs_truncate(struct inode *inode)
 672{
 673	unsigned long blkoff;
 674	unsigned int blocksize;
 675	struct nilfs_transaction_info ti;
 676	struct super_block *sb = inode->i_sb;
 677	struct nilfs_inode_info *ii = NILFS_I(inode);
 678
 679	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 680		return;
 681	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 682		return;
 683
 684	blocksize = sb->s_blocksize;
 685	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 686	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 687
 688	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 689
 690	nilfs_truncate_bmap(ii, blkoff);
 691
 692	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 693	if (IS_SYNC(inode))
 694		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 695
 696	nilfs_mark_inode_dirty(inode);
 697	nilfs_set_file_dirty(inode, 0);
 698	nilfs_transaction_commit(sb);
 699	/* May construct a logical segment and may fail in sync mode.
 700	   But truncate has no return value. */
 
 
 701}
 702
 703static void nilfs_clear_inode(struct inode *inode)
 704{
 705	struct nilfs_inode_info *ii = NILFS_I(inode);
 706	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 707
 708	/*
 709	 * Free resources allocated in nilfs_read_inode(), here.
 710	 */
 711	BUG_ON(!list_empty(&ii->i_dirty));
 712	brelse(ii->i_bh);
 713	ii->i_bh = NULL;
 714
 715	if (mdi && mdi->mi_palloc_cache)
 716		nilfs_palloc_destroy_cache(inode);
 717
 718	if (test_bit(NILFS_I_BMAP, &ii->i_state))
 719		nilfs_bmap_clear(ii->i_bmap);
 720
 721	nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 
 722
 723	if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 724		nilfs_put_root(ii->i_root);
 725}
 726
 727void nilfs_evict_inode(struct inode *inode)
 728{
 729	struct nilfs_transaction_info ti;
 730	struct super_block *sb = inode->i_sb;
 731	struct nilfs_inode_info *ii = NILFS_I(inode);
 732	int ret;
 733
 734	if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 735		if (inode->i_data.nrpages)
 736			truncate_inode_pages(&inode->i_data, 0);
 737		end_writeback(inode);
 738		nilfs_clear_inode(inode);
 739		return;
 740	}
 741	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 742
 743	if (inode->i_data.nrpages)
 744		truncate_inode_pages(&inode->i_data, 0);
 745
 746	/* TODO: some of the following operations may fail.  */
 747	nilfs_truncate_bmap(ii, 0);
 748	nilfs_mark_inode_dirty(inode);
 749	end_writeback(inode);
 750
 751	ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 752	if (!ret)
 753		atomic_dec(&ii->i_root->inodes_count);
 754
 755	nilfs_clear_inode(inode);
 756
 757	if (IS_SYNC(inode))
 758		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 759	nilfs_transaction_commit(sb);
 760	/* May construct a logical segment and may fail in sync mode.
 761	   But delete_inode has no return value. */
 
 
 762}
 763
 764int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 
 765{
 766	struct nilfs_transaction_info ti;
 767	struct inode *inode = dentry->d_inode;
 768	struct super_block *sb = inode->i_sb;
 769	int err;
 770
 771	err = inode_change_ok(inode, iattr);
 772	if (err)
 773		return err;
 774
 775	err = nilfs_transaction_begin(sb, &ti, 0);
 776	if (unlikely(err))
 777		return err;
 778
 779	if ((iattr->ia_valid & ATTR_SIZE) &&
 780	    iattr->ia_size != i_size_read(inode)) {
 781		inode_dio_wait(inode);
 782
 783		err = vmtruncate(inode, iattr->ia_size);
 784		if (unlikely(err))
 785			goto out_err;
 786	}
 787
 788	setattr_copy(inode, iattr);
 789	mark_inode_dirty(inode);
 790
 791	if (iattr->ia_valid & ATTR_MODE) {
 792		err = nilfs_acl_chmod(inode);
 793		if (unlikely(err))
 794			goto out_err;
 795	}
 796
 797	return nilfs_transaction_commit(sb);
 798
 799out_err:
 800	nilfs_transaction_abort(sb);
 801	return err;
 802}
 803
 804int nilfs_permission(struct inode *inode, int mask)
 
 805{
 806	struct nilfs_root *root = NILFS_I(inode)->i_root;
 
 807	if ((mask & MAY_WRITE) && root &&
 808	    root->cno != NILFS_CPTREE_CURRENT_CNO)
 809		return -EROFS; /* snapshot is not writable */
 810
 811	return generic_permission(inode, mask);
 812}
 813
 814int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 815{
 816	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 817	struct nilfs_inode_info *ii = NILFS_I(inode);
 818	int err;
 819
 820	spin_lock(&nilfs->ns_inode_lock);
 821	if (ii->i_bh == NULL) {
 822		spin_unlock(&nilfs->ns_inode_lock);
 823		err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 824						  inode->i_ino, pbh);
 825		if (unlikely(err))
 826			return err;
 827		spin_lock(&nilfs->ns_inode_lock);
 828		if (ii->i_bh == NULL)
 829			ii->i_bh = *pbh;
 830		else {
 831			brelse(*pbh);
 832			*pbh = ii->i_bh;
 833		}
 834	} else
 835		*pbh = ii->i_bh;
 836
 837	get_bh(*pbh);
 838	spin_unlock(&nilfs->ns_inode_lock);
 839	return 0;
 840}
 841
 842int nilfs_inode_dirty(struct inode *inode)
 843{
 844	struct nilfs_inode_info *ii = NILFS_I(inode);
 845	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 846	int ret = 0;
 847
 848	if (!list_empty(&ii->i_dirty)) {
 849		spin_lock(&nilfs->ns_inode_lock);
 850		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 851			test_bit(NILFS_I_BUSY, &ii->i_state);
 852		spin_unlock(&nilfs->ns_inode_lock);
 853	}
 854	return ret;
 855}
 856
 857int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 858{
 859	struct nilfs_inode_info *ii = NILFS_I(inode);
 860	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 861
 862	atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 863
 864	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 865		return 0;
 866
 867	spin_lock(&nilfs->ns_inode_lock);
 868	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 869	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 870		/* Because this routine may race with nilfs_dispose_list(),
 871		   we have to check NILFS_I_QUEUED here, too. */
 
 
 872		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 873			/* This will happen when somebody is freeing
 874			   this inode. */
 875			nilfs_warning(inode->i_sb, __func__,
 876				      "cannot get inode (ino=%lu)\n",
 877				      inode->i_ino);
 
 
 878			spin_unlock(&nilfs->ns_inode_lock);
 879			return -EINVAL; /* NILFS_I_DIRTY may remain for
 880					   freeing inode */
 
 
 881		}
 882		list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 883		set_bit(NILFS_I_QUEUED, &ii->i_state);
 884	}
 885	spin_unlock(&nilfs->ns_inode_lock);
 886	return 0;
 887}
 888
 889int nilfs_mark_inode_dirty(struct inode *inode)
 890{
 891	struct buffer_head *ibh;
 892	int err;
 893
 894	err = nilfs_load_inode_block(inode, &ibh);
 895	if (unlikely(err)) {
 896		nilfs_warning(inode->i_sb, __func__,
 897			      "failed to reget inode block.\n");
 
 898		return err;
 899	}
 900	nilfs_update_inode(inode, ibh);
 901	mark_buffer_dirty(ibh);
 902	nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 903	brelse(ibh);
 904	return 0;
 905}
 906
 907/**
 908 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 909 * @inode: inode of the file to be registered.
 
 910 *
 911 * nilfs_dirty_inode() loads a inode block containing the specified
 912 * @inode and copies data from a nilfs_inode to a corresponding inode
 913 * entry in the inode block. This operation is excluded from the segment
 914 * construction. This function can be called both as a single operation
 915 * and as a part of indivisible file operations.
 916 */
 917void nilfs_dirty_inode(struct inode *inode, int flags)
 918{
 919	struct nilfs_transaction_info ti;
 920	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 921
 922	if (is_bad_inode(inode)) {
 923		nilfs_warning(inode->i_sb, __func__,
 924			      "tried to mark bad_inode dirty. ignored.\n");
 925		dump_stack();
 926		return;
 927	}
 928	if (mdi) {
 929		nilfs_mdt_mark_dirty(inode);
 930		return;
 931	}
 932	nilfs_transaction_begin(inode->i_sb, &ti, 0);
 933	nilfs_mark_inode_dirty(inode);
 934	nilfs_transaction_commit(inode->i_sb); /* never fails */
 935}
 936
 937int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 938		 __u64 start, __u64 len)
 939{
 940	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 941	__u64 logical = 0, phys = 0, size = 0;
 942	__u32 flags = 0;
 943	loff_t isize;
 944	sector_t blkoff, end_blkoff;
 945	sector_t delalloc_blkoff;
 946	unsigned long delalloc_blklen;
 947	unsigned int blkbits = inode->i_blkbits;
 948	int ret, n;
 949
 950	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 951	if (ret)
 952		return ret;
 953
 954	mutex_lock(&inode->i_mutex);
 955
 956	isize = i_size_read(inode);
 957
 958	blkoff = start >> blkbits;
 959	end_blkoff = (start + len - 1) >> blkbits;
 960
 961	delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
 962							&delalloc_blkoff);
 963
 964	do {
 965		__u64 blkphy;
 966		unsigned int maxblocks;
 967
 968		if (delalloc_blklen && blkoff == delalloc_blkoff) {
 969			if (size) {
 970				/* End of the current extent */
 971				ret = fiemap_fill_next_extent(
 972					fieinfo, logical, phys, size, flags);
 973				if (ret)
 974					break;
 975			}
 976			if (blkoff > end_blkoff)
 977				break;
 978
 979			flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
 980			logical = blkoff << blkbits;
 981			phys = 0;
 982			size = delalloc_blklen << blkbits;
 983
 984			blkoff = delalloc_blkoff + delalloc_blklen;
 985			delalloc_blklen = nilfs_find_uncommitted_extent(
 986				inode, blkoff, &delalloc_blkoff);
 987			continue;
 988		}
 989
 990		/*
 991		 * Limit the number of blocks that we look up so as
 992		 * not to get into the next delayed allocation extent.
 993		 */
 994		maxblocks = INT_MAX;
 995		if (delalloc_blklen)
 996			maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
 997					  maxblocks);
 998		blkphy = 0;
 999
1000		down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1001		n = nilfs_bmap_lookup_contig(
1002			NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1003		up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1004
1005		if (n < 0) {
1006			int past_eof;
1007
1008			if (unlikely(n != -ENOENT))
1009				break; /* error */
1010
1011			/* HOLE */
1012			blkoff++;
1013			past_eof = ((blkoff << blkbits) >= isize);
1014
1015			if (size) {
1016				/* End of the current extent */
1017
1018				if (past_eof)
1019					flags |= FIEMAP_EXTENT_LAST;
1020
1021				ret = fiemap_fill_next_extent(
1022					fieinfo, logical, phys, size, flags);
1023				if (ret)
1024					break;
1025				size = 0;
1026			}
1027			if (blkoff > end_blkoff || past_eof)
1028				break;
1029		} else {
1030			if (size) {
1031				if (phys && blkphy << blkbits == phys + size) {
1032					/* The current extent goes on */
1033					size += n << blkbits;
1034				} else {
1035					/* Terminate the current extent */
1036					ret = fiemap_fill_next_extent(
1037						fieinfo, logical, phys, size,
1038						flags);
1039					if (ret || blkoff > end_blkoff)
1040						break;
1041
1042					/* Start another extent */
1043					flags = FIEMAP_EXTENT_MERGED;
1044					logical = blkoff << blkbits;
1045					phys = blkphy << blkbits;
1046					size = n << blkbits;
1047				}
1048			} else {
1049				/* Start a new extent */
1050				flags = FIEMAP_EXTENT_MERGED;
1051				logical = blkoff << blkbits;
1052				phys = blkphy << blkbits;
1053				size = n << blkbits;
1054			}
1055			blkoff += n;
1056		}
1057		cond_resched();
1058	} while (true);
1059
1060	/* If ret is 1 then we just hit the end of the extent array */
1061	if (ret == 1)
1062		ret = 0;
1063
1064	mutex_unlock(&inode->i_mutex);
1065	return ret;
1066}