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}
v3.15
 
   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/aio.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
  37/**
  38 * struct nilfs_iget_args - arguments used during comparison between inodes
  39 * @ino: inode number
  40 * @cno: checkpoint number
  41 * @root: pointer on NILFS root object (mounted checkpoint)
  42 * @for_gc: inode for GC flag
 
 
  43 */
  44struct nilfs_iget_args {
  45	u64 ino;
  46	__u64 cno;
  47	struct nilfs_root *root;
  48	int for_gc;
 
 
  49};
  50
 
 
  51void nilfs_inode_add_blocks(struct inode *inode, int n)
  52{
  53	struct nilfs_root *root = NILFS_I(inode)->i_root;
  54
  55	inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  56	if (root)
  57		atomic64_add(n, &root->blocks_count);
  58}
  59
  60void nilfs_inode_sub_blocks(struct inode *inode, int n)
  61{
  62	struct nilfs_root *root = NILFS_I(inode)->i_root;
  63
  64	inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  65	if (root)
  66		atomic64_sub(n, &root->blocks_count);
  67}
  68
  69/**
  70 * nilfs_get_block() - get a file block on the filesystem (callback function)
  71 * @inode - inode struct of the target file
  72 * @blkoff - file block number
  73 * @bh_result - buffer head to be mapped on
  74 * @create - indicate whether allocating the block or not when it has not
  75 *      been allocated yet.
  76 *
  77 * This function does not issue actual read request of the specified data
  78 * block. It is done by VFS.
  79 */
  80int nilfs_get_block(struct inode *inode, sector_t blkoff,
  81		    struct buffer_head *bh_result, int create)
  82{
  83	struct nilfs_inode_info *ii = NILFS_I(inode);
  84	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  85	__u64 blknum = 0;
  86	int err = 0, ret;
  87	unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  88
  89	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  90	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  91	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  92	if (ret >= 0) {	/* found */
  93		map_bh(bh_result, inode->i_sb, blknum);
  94		if (ret > 0)
  95			bh_result->b_size = (ret << inode->i_blkbits);
  96		goto out;
  97	}
  98	/* data block was not found */
  99	if (ret == -ENOENT && create) {
 100		struct nilfs_transaction_info ti;
 101
 102		bh_result->b_blocknr = 0;
 103		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 104		if (unlikely(err))
 105			goto out;
 106		err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 107					(unsigned long)bh_result);
 108		if (unlikely(err != 0)) {
 109			if (err == -EEXIST) {
 110				/*
 111				 * The get_block() function could be called
 112				 * from multiple callers for an inode.
 113				 * However, the page having this block must
 114				 * be locked in this case.
 115				 */
 116				printk(KERN_WARNING
 117				       "nilfs_get_block: a race condition "
 118				       "while inserting a data block. "
 119				       "(inode number=%lu, file block "
 120				       "offset=%llu)\n",
 121				       inode->i_ino,
 122				       (unsigned long long)blkoff);
 123				err = 0;
 124			}
 125			nilfs_transaction_abort(inode->i_sb);
 126			goto out;
 127		}
 128		nilfs_mark_inode_dirty(inode);
 129		nilfs_transaction_commit(inode->i_sb); /* never fails */
 130		/* Error handling should be detailed */
 131		set_buffer_new(bh_result);
 132		set_buffer_delay(bh_result);
 133		map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 134						      to proper value */
 
 135	} else if (ret == -ENOENT) {
 136		/* not found is not error (e.g. hole); must return without
 137		   the mapped state flag. */
 
 
 138		;
 139	} else {
 140		err = ret;
 141	}
 142
 143 out:
 144	return err;
 145}
 146
 147/**
 148 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 149 * address_space_operations.
 150 * @file - file struct of the file to be read
 151 * @page - the page to be read
 152 */
 153static int nilfs_readpage(struct file *file, struct page *page)
 154{
 155	return mpage_readpage(page, nilfs_get_block);
 156}
 157
 158/**
 159 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 160 * address_space_operations.
 161 * @file - file struct of the file to be read
 162 * @mapping - address_space struct used for reading multiple pages
 163 * @pages - the pages to be read
 164 * @nr_pages - number of pages to be read
 165 */
 166static int nilfs_readpages(struct file *file, struct address_space *mapping,
 167			   struct list_head *pages, unsigned nr_pages)
 168{
 169	return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 170}
 171
 172static int nilfs_writepages(struct address_space *mapping,
 173			    struct writeback_control *wbc)
 174{
 175	struct inode *inode = mapping->host;
 176	int err = 0;
 177
 178	if (inode->i_sb->s_flags & MS_RDONLY) {
 179		nilfs_clear_dirty_pages(mapping, false);
 180		return -EROFS;
 181	}
 182
 183	if (wbc->sync_mode == WB_SYNC_ALL)
 184		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 185						    wbc->range_start,
 186						    wbc->range_end);
 187	return err;
 188}
 189
 190static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 191{
 192	struct inode *inode = page->mapping->host;
 193	int err;
 194
 195	if (inode->i_sb->s_flags & MS_RDONLY) {
 196		/*
 197		 * It means that filesystem was remounted in read-only
 198		 * mode because of error or metadata corruption. But we
 199		 * have dirty pages that try to be flushed in background.
 200		 * So, here we simply discard this dirty page.
 201		 */
 202		nilfs_clear_dirty_page(page, false);
 203		unlock_page(page);
 204		return -EROFS;
 205	}
 206
 207	redirty_page_for_writepage(wbc, page);
 208	unlock_page(page);
 209
 210	if (wbc->sync_mode == WB_SYNC_ALL) {
 211		err = nilfs_construct_segment(inode->i_sb);
 212		if (unlikely(err))
 213			return err;
 214	} else if (wbc->for_reclaim)
 215		nilfs_flush_segment(inode->i_sb, inode->i_ino);
 216
 217	return 0;
 218}
 219
 220static int nilfs_set_page_dirty(struct page *page)
 
 221{
 222	int ret = __set_page_dirty_nobuffers(page);
 
 
 
 223
 224	if (page_has_buffers(page)) {
 225		struct inode *inode = page->mapping->host;
 226		unsigned nr_dirty = 0;
 227		struct buffer_head *bh, *head;
 
 
 
 228
 229		/*
 230		 * This page is locked by callers, and no other thread
 231		 * concurrently marks its buffers dirty since they are
 232		 * only dirtied through routines in fs/buffer.c in
 233		 * which call sites of mark_buffer_dirty are protected
 234		 * by page lock.
 235		 */
 236		bh = head = page_buffers(page);
 237		do {
 238			/* Do not mark hole blocks dirty */
 239			if (buffer_dirty(bh) || !buffer_mapped(bh))
 240				continue;
 241
 242			set_buffer_dirty(bh);
 243			nr_dirty++;
 244		} while (bh = bh->b_this_page, bh != head);
 
 
 
 
 245
 246		if (nr_dirty)
 247			nilfs_set_file_dirty(inode, nr_dirty);
 248	}
 249	return ret;
 250}
 251
 252void nilfs_write_failed(struct address_space *mapping, loff_t to)
 253{
 254	struct inode *inode = mapping->host;
 255
 256	if (to > inode->i_size) {
 257		truncate_pagecache(inode, inode->i_size);
 258		nilfs_truncate(inode);
 259	}
 260}
 261
 262static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 263			     loff_t pos, unsigned len, unsigned flags,
 264			     struct page **pagep, void **fsdata)
 265
 266{
 267	struct inode *inode = mapping->host;
 268	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 269
 270	if (unlikely(err))
 271		return err;
 272
 273	err = block_write_begin(mapping, pos, len, flags, pagep,
 274				nilfs_get_block);
 275	if (unlikely(err)) {
 276		nilfs_write_failed(mapping, pos + len);
 277		nilfs_transaction_abort(inode->i_sb);
 278	}
 279	return err;
 280}
 281
 282static int nilfs_write_end(struct file *file, struct address_space *mapping,
 283			   loff_t pos, unsigned len, unsigned copied,
 284			   struct page *page, void *fsdata)
 285{
 286	struct inode *inode = mapping->host;
 287	unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 288	unsigned nr_dirty;
 289	int err;
 290
 291	nr_dirty = nilfs_page_count_clean_buffers(page, start,
 292						  start + copied);
 293	copied = generic_write_end(file, mapping, pos, len, copied, page,
 294				   fsdata);
 295	nilfs_set_file_dirty(inode, nr_dirty);
 296	err = nilfs_transaction_commit(inode->i_sb);
 297	return err ? : copied;
 298}
 299
 300static ssize_t
 301nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 302		loff_t offset, unsigned long nr_segs)
 303{
 304	struct file *file = iocb->ki_filp;
 305	struct address_space *mapping = file->f_mapping;
 306	struct inode *inode = file->f_mapping->host;
 307	ssize_t size;
 308
 309	if (rw == WRITE)
 310		return 0;
 311
 312	/* Needs synchronization with the cleaner */
 313	size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 314				  nilfs_get_block);
 315
 316	/*
 317	 * In case of error extending write may have instantiated a few
 318	 * blocks outside i_size. Trim these off again.
 319	 */
 320	if (unlikely((rw & WRITE) && size < 0)) {
 321		loff_t isize = i_size_read(inode);
 322		loff_t end = offset + iov_length(iov, nr_segs);
 323
 324		if (end > isize)
 325			nilfs_write_failed(mapping, end);
 326	}
 327
 328	return size;
 329}
 330
 331const struct address_space_operations nilfs_aops = {
 332	.writepage		= nilfs_writepage,
 333	.readpage		= nilfs_readpage,
 334	.writepages		= nilfs_writepages,
 335	.set_page_dirty		= nilfs_set_page_dirty,
 336	.readpages		= nilfs_readpages,
 337	.write_begin		= nilfs_write_begin,
 338	.write_end		= nilfs_write_end,
 339	/* .releasepage		= nilfs_releasepage, */
 340	.invalidatepage		= block_invalidatepage,
 341	.direct_IO		= nilfs_direct_IO,
 342	.is_partially_uptodate  = block_is_partially_uptodate,
 343};
 344
 
 
 
 
 
 
 
 
 
 
 
 
 345struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 346{
 347	struct super_block *sb = dir->i_sb;
 348	struct the_nilfs *nilfs = sb->s_fs_info;
 349	struct inode *inode;
 350	struct nilfs_inode_info *ii;
 351	struct nilfs_root *root;
 
 352	int err = -ENOMEM;
 353	ino_t ino;
 354
 355	inode = new_inode(sb);
 356	if (unlikely(!inode))
 357		goto failed;
 358
 359	mapping_set_gfp_mask(inode->i_mapping,
 360			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 361
 362	root = NILFS_I(dir)->i_root;
 363	ii = NILFS_I(inode);
 364	ii->i_state = 1 << NILFS_I_NEW;
 365	ii->i_root = root;
 366
 367	err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 368	if (unlikely(err))
 369		goto failed_ifile_create_inode;
 370	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
 371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 372	atomic64_inc(&root->inodes_count);
 373	inode_init_owner(inode, dir, mode);
 374	inode->i_ino = ino;
 375	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 376
 377	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 378		err = nilfs_bmap_read(ii->i_bmap, NULL);
 379		if (err < 0)
 380			goto failed_bmap;
 381
 382		set_bit(NILFS_I_BMAP, &ii->i_state);
 383		/* No lock is needed; iget() ensures it. */
 384	}
 385
 386	ii->i_flags = nilfs_mask_flags(
 387		mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 388
 389	/* ii->i_file_acl = 0; */
 390	/* ii->i_dir_acl = 0; */
 391	ii->i_dir_start_lookup = 0;
 392	nilfs_set_inode_flags(inode);
 393	spin_lock(&nilfs->ns_next_gen_lock);
 394	inode->i_generation = nilfs->ns_next_generation++;
 395	spin_unlock(&nilfs->ns_next_gen_lock);
 396	insert_inode_hash(inode);
 
 
 
 397
 398	err = nilfs_init_acl(inode, dir);
 399	if (unlikely(err))
 400		goto failed_acl; /* never occur. When supporting
 401				    nilfs_init_acl(), proper cancellation of
 402				    above jobs should be considered */
 
 
 403
 404	return inode;
 405
 406 failed_acl:
 407 failed_bmap:
 408	clear_nlink(inode);
 409	iput(inode);  /* raw_inode will be deleted through
 410			 generic_delete_inode() */
 
 
 
 
 411	goto failed;
 412
 413 failed_ifile_create_inode:
 414	make_bad_inode(inode);
 415	iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 416			 called */
 417 failed:
 418	return ERR_PTR(err);
 419}
 420
 421void nilfs_set_inode_flags(struct inode *inode)
 422{
 423	unsigned int flags = NILFS_I(inode)->i_flags;
 
 424
 425	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 426			    S_DIRSYNC);
 427	if (flags & FS_SYNC_FL)
 428		inode->i_flags |= S_SYNC;
 429	if (flags & FS_APPEND_FL)
 430		inode->i_flags |= S_APPEND;
 431	if (flags & FS_IMMUTABLE_FL)
 432		inode->i_flags |= S_IMMUTABLE;
 433	if (flags & FS_NOATIME_FL)
 434		inode->i_flags |= S_NOATIME;
 435	if (flags & FS_DIRSYNC_FL)
 436		inode->i_flags |= S_DIRSYNC;
 437	mapping_set_gfp_mask(inode->i_mapping,
 438			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 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 (inode->i_nlink == 0 && inode->i_mode == 0)
 459		return -EINVAL; /* this inode is deleted */
 
 
 460
 461	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 462	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 463#if 0
 464	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 465	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 466		0 : le32_to_cpu(raw_inode->i_dir_acl);
 467#endif
 468	ii->i_dir_start_lookup = 0;
 469	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 470
 471	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 472	    S_ISLNK(inode->i_mode)) {
 473		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 474		if (err < 0)
 475			return err;
 476		set_bit(NILFS_I_BMAP, &ii->i_state);
 477		/* No lock is needed; iget() ensures it. */
 478	}
 479	return 0;
 480}
 481
 482static int __nilfs_read_inode(struct super_block *sb,
 483			      struct nilfs_root *root, unsigned long ino,
 484			      struct inode *inode)
 485{
 486	struct the_nilfs *nilfs = sb->s_fs_info;
 487	struct buffer_head *bh;
 488	struct nilfs_inode *raw_inode;
 489	int err;
 490
 491	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 492	err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 493	if (unlikely(err))
 494		goto bad_inode;
 495
 496	raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 497
 498	err = nilfs_read_inode_common(inode, raw_inode);
 499	if (err)
 500		goto failed_unmap;
 501
 502	if (S_ISREG(inode->i_mode)) {
 503		inode->i_op = &nilfs_file_inode_operations;
 504		inode->i_fop = &nilfs_file_operations;
 505		inode->i_mapping->a_ops = &nilfs_aops;
 506	} else if (S_ISDIR(inode->i_mode)) {
 507		inode->i_op = &nilfs_dir_inode_operations;
 508		inode->i_fop = &nilfs_dir_operations;
 509		inode->i_mapping->a_ops = &nilfs_aops;
 510	} else if (S_ISLNK(inode->i_mode)) {
 511		inode->i_op = &nilfs_symlink_inode_operations;
 
 512		inode->i_mapping->a_ops = &nilfs_aops;
 513	} else {
 514		inode->i_op = &nilfs_special_inode_operations;
 515		init_special_inode(
 516			inode, inode->i_mode,
 517			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 518	}
 519	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 520	brelse(bh);
 521	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 522	nilfs_set_inode_flags(inode);
 
 
 523	return 0;
 524
 525 failed_unmap:
 526	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 527	brelse(bh);
 528
 529 bad_inode:
 530	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 531	return err;
 532}
 533
 534static int nilfs_iget_test(struct inode *inode, void *opaque)
 535{
 536	struct nilfs_iget_args *args = opaque;
 537	struct nilfs_inode_info *ii;
 538
 539	if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 540		return 0;
 541
 542	ii = NILFS_I(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 543	if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 544		return !args->for_gc;
 545
 546	return args->for_gc && args->cno == ii->i_cno;
 547}
 548
 549static int nilfs_iget_set(struct inode *inode, void *opaque)
 550{
 551	struct nilfs_iget_args *args = opaque;
 552
 553	inode->i_ino = args->ino;
 554	if (args->for_gc) {
 555		NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 556		NILFS_I(inode)->i_cno = args->cno;
 557		NILFS_I(inode)->i_root = NULL;
 558	} else {
 559		if (args->root && args->ino == NILFS_ROOT_INO)
 560			nilfs_get_root(args->root);
 561		NILFS_I(inode)->i_root = args->root;
 562	}
 
 
 563	return 0;
 564}
 565
 566struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 567			    unsigned long ino)
 568{
 569	struct nilfs_iget_args args = {
 570		.ino = ino, .root = root, .cno = 0, .for_gc = 0
 
 571	};
 572
 573	return ilookup5(sb, ino, nilfs_iget_test, &args);
 574}
 575
 576struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 577				unsigned long ino)
 578{
 579	struct nilfs_iget_args args = {
 580		.ino = ino, .root = root, .cno = 0, .for_gc = 0
 
 581	};
 582
 583	return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 584}
 585
 586struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 587			 unsigned long ino)
 588{
 589	struct inode *inode;
 590	int err;
 591
 592	inode = nilfs_iget_locked(sb, root, ino);
 593	if (unlikely(!inode))
 594		return ERR_PTR(-ENOMEM);
 595	if (!(inode->i_state & I_NEW))
 596		return inode;
 597
 598	err = __nilfs_read_inode(sb, root, ino, inode);
 599	if (unlikely(err)) {
 600		iget_failed(inode);
 601		return ERR_PTR(err);
 602	}
 603	unlock_new_inode(inode);
 604	return inode;
 605}
 606
 607struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 608				__u64 cno)
 609{
 610	struct nilfs_iget_args args = {
 611		.ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 
 612	};
 613	struct inode *inode;
 614	int err;
 615
 616	inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 617	if (unlikely(!inode))
 618		return ERR_PTR(-ENOMEM);
 619	if (!(inode->i_state & I_NEW))
 620		return inode;
 621
 622	err = nilfs_init_gcinode(inode);
 623	if (unlikely(err)) {
 624		iget_failed(inode);
 625		return ERR_PTR(err);
 626	}
 627	unlock_new_inode(inode);
 628	return inode;
 629}
 630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 631void nilfs_write_inode_common(struct inode *inode,
 632			      struct nilfs_inode *raw_inode, int has_bmap)
 633{
 634	struct nilfs_inode_info *ii = NILFS_I(inode);
 635
 636	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 637	raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 638	raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 639	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 640	raw_inode->i_size = cpu_to_le64(inode->i_size);
 641	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 642	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 643	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 644	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 645	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 646
 647	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 648	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 649
 650	if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 651		struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 652
 653		/* zero-fill unused portion in the case of super root block */
 654		raw_inode->i_xattr = 0;
 655		raw_inode->i_pad = 0;
 656		memset((void *)raw_inode + sizeof(*raw_inode), 0,
 657		       nilfs->ns_inode_size - sizeof(*raw_inode));
 658	}
 659
 660	if (has_bmap)
 661		nilfs_bmap_write(ii->i_bmap, raw_inode);
 662	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 663		raw_inode->i_device_code =
 664			cpu_to_le64(huge_encode_dev(inode->i_rdev));
 665	/* When extending inode, nilfs->ns_inode_size should be checked
 666	   for substitutions of appended fields */
 
 
 667}
 668
 669void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 670{
 671	ino_t ino = inode->i_ino;
 672	struct nilfs_inode_info *ii = NILFS_I(inode);
 673	struct inode *ifile = ii->i_root->ifile;
 674	struct nilfs_inode *raw_inode;
 675
 676	raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 677
 678	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 679		memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 680	set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 
 681
 682	nilfs_write_inode_common(inode, raw_inode, 0);
 683		/* XXX: call with has_bmap = 0 is a workaround to avoid
 684		   deadlock of bmap. This delays update of i_bmap to just
 685		   before writing */
 
 
 
 686	nilfs_ifile_unmap_inode(ifile, ino, ibh);
 687}
 688
 689#define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
 690
 691static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 692				unsigned long from)
 693{
 694	unsigned long b;
 695	int ret;
 696
 697	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 698		return;
 699repeat:
 700	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 701	if (ret == -ENOENT)
 702		return;
 703	else if (ret < 0)
 704		goto failed;
 705
 706	if (b < from)
 707		return;
 708
 709	b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 710	ret = nilfs_bmap_truncate(ii->i_bmap, b);
 711	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 712	if (!ret || (ret == -ENOMEM &&
 713		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 714		goto repeat;
 715
 716failed:
 717	nilfs_warning(ii->vfs_inode.i_sb, __func__,
 718		      "failed to truncate bmap (ino=%lu, err=%d)",
 719		      ii->vfs_inode.i_ino, ret);
 720}
 721
 722void nilfs_truncate(struct inode *inode)
 723{
 724	unsigned long blkoff;
 725	unsigned int blocksize;
 726	struct nilfs_transaction_info ti;
 727	struct super_block *sb = inode->i_sb;
 728	struct nilfs_inode_info *ii = NILFS_I(inode);
 729
 730	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 731		return;
 732	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 733		return;
 734
 735	blocksize = sb->s_blocksize;
 736	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 737	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 738
 739	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 740
 741	nilfs_truncate_bmap(ii, blkoff);
 742
 743	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 744	if (IS_SYNC(inode))
 745		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 746
 747	nilfs_mark_inode_dirty(inode);
 748	nilfs_set_file_dirty(inode, 0);
 749	nilfs_transaction_commit(sb);
 750	/* May construct a logical segment and may fail in sync mode.
 751	   But truncate has no return value. */
 
 
 752}
 753
 754static void nilfs_clear_inode(struct inode *inode)
 755{
 756	struct nilfs_inode_info *ii = NILFS_I(inode);
 757	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 758
 759	/*
 760	 * Free resources allocated in nilfs_read_inode(), here.
 761	 */
 762	BUG_ON(!list_empty(&ii->i_dirty));
 763	brelse(ii->i_bh);
 764	ii->i_bh = NULL;
 765
 766	if (mdi && mdi->mi_palloc_cache)
 767		nilfs_palloc_destroy_cache(inode);
 768
 769	if (test_bit(NILFS_I_BMAP, &ii->i_state))
 770		nilfs_bmap_clear(ii->i_bmap);
 771
 772	nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 
 773
 774	if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 775		nilfs_put_root(ii->i_root);
 776}
 777
 778void nilfs_evict_inode(struct inode *inode)
 779{
 780	struct nilfs_transaction_info ti;
 781	struct super_block *sb = inode->i_sb;
 782	struct nilfs_inode_info *ii = NILFS_I(inode);
 783	int ret;
 784
 785	if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 786		truncate_inode_pages_final(&inode->i_data);
 787		clear_inode(inode);
 788		nilfs_clear_inode(inode);
 789		return;
 790	}
 791	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 792
 793	truncate_inode_pages_final(&inode->i_data);
 794
 795	/* TODO: some of the following operations may fail.  */
 796	nilfs_truncate_bmap(ii, 0);
 797	nilfs_mark_inode_dirty(inode);
 798	clear_inode(inode);
 799
 800	ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 801	if (!ret)
 802		atomic64_dec(&ii->i_root->inodes_count);
 803
 804	nilfs_clear_inode(inode);
 805
 806	if (IS_SYNC(inode))
 807		nilfs_set_transaction_flag(NILFS_TI_SYNC);
 808	nilfs_transaction_commit(sb);
 809	/* May construct a logical segment and may fail in sync mode.
 810	   But delete_inode has no return value. */
 
 
 811}
 812
 813int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 
 814{
 815	struct nilfs_transaction_info ti;
 816	struct inode *inode = dentry->d_inode;
 817	struct super_block *sb = inode->i_sb;
 818	int err;
 819
 820	err = inode_change_ok(inode, iattr);
 821	if (err)
 822		return err;
 823
 824	err = nilfs_transaction_begin(sb, &ti, 0);
 825	if (unlikely(err))
 826		return err;
 827
 828	if ((iattr->ia_valid & ATTR_SIZE) &&
 829	    iattr->ia_size != i_size_read(inode)) {
 830		inode_dio_wait(inode);
 831		truncate_setsize(inode, iattr->ia_size);
 832		nilfs_truncate(inode);
 833	}
 834
 835	setattr_copy(inode, iattr);
 836	mark_inode_dirty(inode);
 837
 838	if (iattr->ia_valid & ATTR_MODE) {
 839		err = nilfs_acl_chmod(inode);
 840		if (unlikely(err))
 841			goto out_err;
 842	}
 843
 844	return nilfs_transaction_commit(sb);
 845
 846out_err:
 847	nilfs_transaction_abort(sb);
 848	return err;
 849}
 850
 851int nilfs_permission(struct inode *inode, int mask)
 
 852{
 853	struct nilfs_root *root = NILFS_I(inode)->i_root;
 
 854	if ((mask & MAY_WRITE) && root &&
 855	    root->cno != NILFS_CPTREE_CURRENT_CNO)
 856		return -EROFS; /* snapshot is not writable */
 857
 858	return generic_permission(inode, mask);
 859}
 860
 861int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 862{
 863	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 864	struct nilfs_inode_info *ii = NILFS_I(inode);
 865	int err;
 866
 867	spin_lock(&nilfs->ns_inode_lock);
 868	if (ii->i_bh == NULL) {
 869		spin_unlock(&nilfs->ns_inode_lock);
 870		err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 871						  inode->i_ino, pbh);
 872		if (unlikely(err))
 873			return err;
 874		spin_lock(&nilfs->ns_inode_lock);
 875		if (ii->i_bh == NULL)
 876			ii->i_bh = *pbh;
 877		else {
 878			brelse(*pbh);
 879			*pbh = ii->i_bh;
 880		}
 881	} else
 882		*pbh = ii->i_bh;
 883
 884	get_bh(*pbh);
 885	spin_unlock(&nilfs->ns_inode_lock);
 886	return 0;
 887}
 888
 889int nilfs_inode_dirty(struct inode *inode)
 890{
 891	struct nilfs_inode_info *ii = NILFS_I(inode);
 892	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 893	int ret = 0;
 894
 895	if (!list_empty(&ii->i_dirty)) {
 896		spin_lock(&nilfs->ns_inode_lock);
 897		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 898			test_bit(NILFS_I_BUSY, &ii->i_state);
 899		spin_unlock(&nilfs->ns_inode_lock);
 900	}
 901	return ret;
 902}
 903
 904int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 905{
 906	struct nilfs_inode_info *ii = NILFS_I(inode);
 907	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 908
 909	atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 910
 911	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 912		return 0;
 913
 914	spin_lock(&nilfs->ns_inode_lock);
 915	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 916	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 917		/* Because this routine may race with nilfs_dispose_list(),
 918		   we have to check NILFS_I_QUEUED here, too. */
 
 
 919		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 920			/* This will happen when somebody is freeing
 921			   this inode. */
 922			nilfs_warning(inode->i_sb, __func__,
 923				      "cannot get inode (ino=%lu)\n",
 924				      inode->i_ino);
 
 
 925			spin_unlock(&nilfs->ns_inode_lock);
 926			return -EINVAL; /* NILFS_I_DIRTY may remain for
 927					   freeing inode */
 
 
 928		}
 929		list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 930		set_bit(NILFS_I_QUEUED, &ii->i_state);
 931	}
 932	spin_unlock(&nilfs->ns_inode_lock);
 933	return 0;
 934}
 935
 936int nilfs_mark_inode_dirty(struct inode *inode)
 937{
 938	struct buffer_head *ibh;
 939	int err;
 940
 941	err = nilfs_load_inode_block(inode, &ibh);
 942	if (unlikely(err)) {
 943		nilfs_warning(inode->i_sb, __func__,
 944			      "failed to reget inode block.\n");
 
 945		return err;
 946	}
 947	nilfs_update_inode(inode, ibh);
 948	mark_buffer_dirty(ibh);
 949	nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 950	brelse(ibh);
 951	return 0;
 952}
 953
 954/**
 955 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 956 * @inode: inode of the file to be registered.
 
 957 *
 958 * nilfs_dirty_inode() loads a inode block containing the specified
 959 * @inode and copies data from a nilfs_inode to a corresponding inode
 960 * entry in the inode block. This operation is excluded from the segment
 961 * construction. This function can be called both as a single operation
 962 * and as a part of indivisible file operations.
 963 */
 964void nilfs_dirty_inode(struct inode *inode, int flags)
 965{
 966	struct nilfs_transaction_info ti;
 967	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 968
 969	if (is_bad_inode(inode)) {
 970		nilfs_warning(inode->i_sb, __func__,
 971			      "tried to mark bad_inode dirty. ignored.\n");
 972		dump_stack();
 973		return;
 974	}
 975	if (mdi) {
 976		nilfs_mdt_mark_dirty(inode);
 977		return;
 978	}
 979	nilfs_transaction_begin(inode->i_sb, &ti, 0);
 980	nilfs_mark_inode_dirty(inode);
 981	nilfs_transaction_commit(inode->i_sb); /* never fails */
 982}
 983
 984int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 985		 __u64 start, __u64 len)
 986{
 987	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 988	__u64 logical = 0, phys = 0, size = 0;
 989	__u32 flags = 0;
 990	loff_t isize;
 991	sector_t blkoff, end_blkoff;
 992	sector_t delalloc_blkoff;
 993	unsigned long delalloc_blklen;
 994	unsigned int blkbits = inode->i_blkbits;
 995	int ret, n;
 996
 997	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 998	if (ret)
 999		return ret;
1000
1001	mutex_lock(&inode->i_mutex);
1002
1003	isize = i_size_read(inode);
1004
1005	blkoff = start >> blkbits;
1006	end_blkoff = (start + len - 1) >> blkbits;
1007
1008	delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1009							&delalloc_blkoff);
1010
1011	do {
1012		__u64 blkphy;
1013		unsigned int maxblocks;
1014
1015		if (delalloc_blklen && blkoff == delalloc_blkoff) {
1016			if (size) {
1017				/* End of the current extent */
1018				ret = fiemap_fill_next_extent(
1019					fieinfo, logical, phys, size, flags);
1020				if (ret)
1021					break;
1022			}
1023			if (blkoff > end_blkoff)
1024				break;
1025
1026			flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1027			logical = blkoff << blkbits;
1028			phys = 0;
1029			size = delalloc_blklen << blkbits;
1030
1031			blkoff = delalloc_blkoff + delalloc_blklen;
1032			delalloc_blklen = nilfs_find_uncommitted_extent(
1033				inode, blkoff, &delalloc_blkoff);
1034			continue;
1035		}
1036
1037		/*
1038		 * Limit the number of blocks that we look up so as
1039		 * not to get into the next delayed allocation extent.
1040		 */
1041		maxblocks = INT_MAX;
1042		if (delalloc_blklen)
1043			maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1044					  maxblocks);
1045		blkphy = 0;
1046
1047		down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1048		n = nilfs_bmap_lookup_contig(
1049			NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1050		up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1051
1052		if (n < 0) {
1053			int past_eof;
1054
1055			if (unlikely(n != -ENOENT))
1056				break; /* error */
1057
1058			/* HOLE */
1059			blkoff++;
1060			past_eof = ((blkoff << blkbits) >= isize);
1061
1062			if (size) {
1063				/* End of the current extent */
1064
1065				if (past_eof)
1066					flags |= FIEMAP_EXTENT_LAST;
1067
1068				ret = fiemap_fill_next_extent(
1069					fieinfo, logical, phys, size, flags);
1070				if (ret)
1071					break;
1072				size = 0;
1073			}
1074			if (blkoff > end_blkoff || past_eof)
1075				break;
1076		} else {
1077			if (size) {
1078				if (phys && blkphy << blkbits == phys + size) {
1079					/* The current extent goes on */
1080					size += n << blkbits;
1081				} else {
1082					/* Terminate the current extent */
1083					ret = fiemap_fill_next_extent(
1084						fieinfo, logical, phys, size,
1085						flags);
1086					if (ret || blkoff > end_blkoff)
1087						break;
1088
1089					/* Start another extent */
1090					flags = FIEMAP_EXTENT_MERGED;
1091					logical = blkoff << blkbits;
1092					phys = blkphy << blkbits;
1093					size = n << blkbits;
1094				}
1095			} else {
1096				/* Start a new extent */
1097				flags = FIEMAP_EXTENT_MERGED;
1098				logical = blkoff << blkbits;
1099				phys = blkphy << blkbits;
1100				size = n << blkbits;
1101			}
1102			blkoff += n;
1103		}
1104		cond_resched();
1105	} while (true);
1106
1107	/* If ret is 1 then we just hit the end of the extent array */
1108	if (ret == 1)
1109		ret = 0;
1110
1111	mutex_unlock(&inode->i_mutex);
1112	return ret;
1113}