File Index
Files and directories
Sign in
FILE

inode.c

File preview with syntax highlighting.

linux-6.19.14/fs/ocfs2/inode.c C
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * inode.c
   4 *
   5 * vfs' aops, fops, dops and iops
   6 *
   7 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8 */
   9
  10#include <linux/fs.h>
  11#include <linux/types.h>
  12#include <linux/highmem.h>
  13#include <linux/pagemap.h>
  14#include <linux/quotaops.h>
  15#include <linux/iversion.h>
  16
  17#include <asm/byteorder.h>
  18
  19#include <cluster/masklog.h>
  20
  21#include "ocfs2.h"
  22
  23#include "alloc.h"
  24#include "dir.h"
  25#include "blockcheck.h"
  26#include "dlmglue.h"
  27#include "extent_map.h"
  28#include "file.h"
  29#include "heartbeat.h"
  30#include "inode.h"
  31#include "journal.h"
  32#include "namei.h"
  33#include "suballoc.h"
  34#include "super.h"
  35#include "symlink.h"
  36#include "sysfile.h"
  37#include "uptodate.h"
  38#include "xattr.h"
  39#include "refcounttree.h"
  40#include "ocfs2_trace.h"
  41#include "filecheck.h"
  42
  43#include "buffer_head_io.h"
  44
  45struct ocfs2_find_inode_args
  46{
  47	u64		fi_blkno;
  48	unsigned long	fi_ino;
  49	unsigned int	fi_flags;
  50	unsigned int	fi_sysfile_type;
  51};
  52
  53static int ocfs2_read_locked_inode(struct inode *inode,
  54				   struct ocfs2_find_inode_args *args);
  55static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
  56static int ocfs2_find_actor(struct inode *inode, void *opaque);
  57static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
  58				    struct inode *inode,
  59				    struct buffer_head *fe_bh);
  60
  61static int ocfs2_filecheck_read_inode_block_full(struct inode *inode,
  62						 struct buffer_head **bh,
  63						 int flags, int type);
  64static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
  65						struct buffer_head *bh);
  66static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
  67					      struct buffer_head *bh);
  68
  69void ocfs2_set_inode_flags(struct inode *inode)
  70{
  71	unsigned int flags = OCFS2_I(inode)->ip_attr;
  72
  73	inode->i_flags &= ~(S_IMMUTABLE |
  74		S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
  75
  76	if (flags & OCFS2_IMMUTABLE_FL)
  77		inode->i_flags |= S_IMMUTABLE;
  78
  79	if (flags & OCFS2_SYNC_FL)
  80		inode->i_flags |= S_SYNC;
  81	if (flags & OCFS2_APPEND_FL)
  82		inode->i_flags |= S_APPEND;
  83	if (flags & OCFS2_NOATIME_FL)
  84		inode->i_flags |= S_NOATIME;
  85	if (flags & OCFS2_DIRSYNC_FL)
  86		inode->i_flags |= S_DIRSYNC;
  87}
  88
  89/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
  90void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
  91{
  92	unsigned int flags = oi->vfs_inode.i_flags;
  93
  94	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
  95			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
  96	if (flags & S_SYNC)
  97		oi->ip_attr |= OCFS2_SYNC_FL;
  98	if (flags & S_APPEND)
  99		oi->ip_attr |= OCFS2_APPEND_FL;
 100	if (flags & S_IMMUTABLE)
 101		oi->ip_attr |= OCFS2_IMMUTABLE_FL;
 102	if (flags & S_NOATIME)
 103		oi->ip_attr |= OCFS2_NOATIME_FL;
 104	if (flags & S_DIRSYNC)
 105		oi->ip_attr |= OCFS2_DIRSYNC_FL;
 106}
 107
 108struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
 109{
 110	struct ocfs2_find_inode_args args;
 111
 112	args.fi_blkno = blkno;
 113	args.fi_flags = 0;
 114	args.fi_ino = ino_from_blkno(sb, blkno);
 115	args.fi_sysfile_type = 0;
 116
 117	return ilookup5(sb, blkno, ocfs2_find_actor, &args);
 118}
 119struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
 120			 int sysfile_type)
 121{
 122	int rc = -ESTALE;
 123	struct inode *inode = NULL;
 124	struct super_block *sb = osb->sb;
 125	struct ocfs2_find_inode_args args;
 126	journal_t *journal = osb->journal->j_journal;
 127
 128	trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
 129			       sysfile_type);
 130
 131	/* Ok. By now we've either got the offsets passed to us by the
 132	 * caller, or we just pulled them off the bh. Lets do some
 133	 * sanity checks to make sure they're OK. */
 134	if (blkno == 0) {
 135		inode = ERR_PTR(-EINVAL);
 136		mlog_errno(PTR_ERR(inode));
 137		goto bail;
 138	}
 139
 140	args.fi_blkno = blkno;
 141	args.fi_flags = flags;
 142	args.fi_ino = ino_from_blkno(sb, blkno);
 143	args.fi_sysfile_type = sysfile_type;
 144
 145	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
 146			     ocfs2_init_locked_inode, &args);
 147	/* inode was *not* in the inode cache. 2.6.x requires
 148	 * us to do our own read_inode call and unlock it
 149	 * afterwards. */
 150	if (inode == NULL) {
 151		inode = ERR_PTR(-ENOMEM);
 152		mlog_errno(PTR_ERR(inode));
 153		goto bail;
 154	}
 155	trace_ocfs2_iget5_locked(inode_state_read_once(inode));
 156	if (inode_state_read_once(inode) & I_NEW) {
 157		rc = ocfs2_read_locked_inode(inode, &args);
 158		unlock_new_inode(inode);
 159	}
 160	if (is_bad_inode(inode)) {
 161		iput(inode);
 162		inode = ERR_PTR(rc);
 163		goto bail;
 164	}
 165
 166	/*
 167	 * Set transaction id's of transactions that have to be committed
 168	 * to finish f[data]sync. We set them to currently running transaction
 169	 * as we cannot be sure that the inode or some of its metadata isn't
 170	 * part of the transaction - the inode could have been reclaimed and
 171	 * now it is reread from disk.
 172	 */
 173	if (journal) {
 174		transaction_t *transaction;
 175		tid_t tid;
 176		struct ocfs2_inode_info *oi = OCFS2_I(inode);
 177
 178		read_lock(&journal->j_state_lock);
 179		if (journal->j_running_transaction)
 180			transaction = journal->j_running_transaction;
 181		else
 182			transaction = journal->j_committing_transaction;
 183		if (transaction)
 184			tid = transaction->t_tid;
 185		else
 186			tid = journal->j_commit_sequence;
 187		read_unlock(&journal->j_state_lock);
 188		oi->i_sync_tid = tid;
 189		oi->i_datasync_tid = tid;
 190	}
 191
 192bail:
 193	if (!IS_ERR(inode)) {
 194		trace_ocfs2_iget_end(inode, 
 195			(unsigned long long)OCFS2_I(inode)->ip_blkno);
 196	}
 197
 198	return inode;
 199}
 200
 201static int ocfs2_dinode_has_extents(struct ocfs2_dinode *di)
 202{
 203	/* inodes flagged with other stuff in id2 */
 204	if (le32_to_cpu(di->i_flags) &
 205	    (OCFS2_SUPER_BLOCK_FL | OCFS2_LOCAL_ALLOC_FL | OCFS2_CHAIN_FL |
 206	     OCFS2_DEALLOC_FL))
 207		return 0;
 208	/* i_flags doesn't indicate when id2 is a fast symlink */
 209	if (S_ISLNK(le16_to_cpu(di->i_mode)) && le64_to_cpu(di->i_size) &&
 210	    !le32_to_cpu(di->i_clusters))
 211		return 0;
 212	if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)
 213		return 0;
 214
 215	return 1;
 216}
 217
 218/*
 219 * here's how inodes get read from disk:
 220 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
 221 * found? : return the in-memory inode
 222 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
 223 */
 224
 225static int ocfs2_find_actor(struct inode *inode, void *opaque)
 226{
 227	struct ocfs2_find_inode_args *args = NULL;
 228	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 229	int ret = 0;
 230
 231	args = opaque;
 232
 233	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
 234
 235	trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
 236
 237	if (oi->ip_blkno != args->fi_blkno)
 238		goto bail;
 239
 240	ret = 1;
 241bail:
 242	return ret;
 243}
 244
 245/*
 246 * initialize the new inode, but don't do anything that would cause
 247 * us to sleep.
 248 * return 0 on success, 1 on failure
 249 */
 250static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
 251{
 252	struct ocfs2_find_inode_args *args = opaque;
 253#ifdef CONFIG_LOCKDEP
 254	static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
 255	static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
 256				     ocfs2_file_ip_alloc_sem_key;
 257#endif
 258
 259	inode->i_ino = args->fi_ino;
 260	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
 261#ifdef CONFIG_LOCKDEP
 262	switch (args->fi_sysfile_type) {
 263	case BAD_BLOCK_SYSTEM_INODE:
 264		break;
 265	case GLOBAL_INODE_ALLOC_SYSTEM_INODE:
 266		lockdep_set_class(&inode->i_rwsem,
 267				  &ocfs2_sysfile_lock_key[GLOBAL_INODE_ALLOC_SYSTEM_INODE]);
 268		break;
 269	case SLOT_MAP_SYSTEM_INODE:
 270		lockdep_set_class(&inode->i_rwsem,
 271				  &ocfs2_sysfile_lock_key[SLOT_MAP_SYSTEM_INODE]);
 272		break;
 273	case HEARTBEAT_SYSTEM_INODE:
 274		lockdep_set_class(&inode->i_rwsem,
 275				  &ocfs2_sysfile_lock_key[HEARTBEAT_SYSTEM_INODE]);
 276		break;
 277	case GLOBAL_BITMAP_SYSTEM_INODE:
 278		lockdep_set_class(&inode->i_rwsem,
 279				  &ocfs2_sysfile_lock_key[GLOBAL_BITMAP_SYSTEM_INODE]);
 280		break;
 281	case USER_QUOTA_SYSTEM_INODE:
 282		lockdep_set_class(&inode->i_rwsem,
 283				  &ocfs2_sysfile_lock_key[USER_QUOTA_SYSTEM_INODE]);
 284		break;
 285	case GROUP_QUOTA_SYSTEM_INODE:
 286		lockdep_set_class(&inode->i_rwsem,
 287				  &ocfs2_sysfile_lock_key[GROUP_QUOTA_SYSTEM_INODE]);
 288		break;
 289	case ORPHAN_DIR_SYSTEM_INODE:
 290		lockdep_set_class(&inode->i_rwsem,
 291				  &ocfs2_sysfile_lock_key[ORPHAN_DIR_SYSTEM_INODE]);
 292		break;
 293	case EXTENT_ALLOC_SYSTEM_INODE:
 294		lockdep_set_class(&inode->i_rwsem,
 295				  &ocfs2_sysfile_lock_key[EXTENT_ALLOC_SYSTEM_INODE]);
 296		break;
 297	case INODE_ALLOC_SYSTEM_INODE:
 298		lockdep_set_class(&inode->i_rwsem,
 299				  &ocfs2_sysfile_lock_key[INODE_ALLOC_SYSTEM_INODE]);
 300		break;
 301	case JOURNAL_SYSTEM_INODE:
 302		lockdep_set_class(&inode->i_rwsem,
 303				  &ocfs2_sysfile_lock_key[JOURNAL_SYSTEM_INODE]);
 304		break;
 305	case LOCAL_ALLOC_SYSTEM_INODE:
 306		lockdep_set_class(&inode->i_rwsem,
 307				  &ocfs2_sysfile_lock_key[LOCAL_ALLOC_SYSTEM_INODE]);
 308		break;
 309	case TRUNCATE_LOG_SYSTEM_INODE:
 310		lockdep_set_class(&inode->i_rwsem,
 311				  &ocfs2_sysfile_lock_key[TRUNCATE_LOG_SYSTEM_INODE]);
 312		break;
 313	case LOCAL_USER_QUOTA_SYSTEM_INODE:
 314		lockdep_set_class(&inode->i_rwsem,
 315				  &ocfs2_sysfile_lock_key[LOCAL_USER_QUOTA_SYSTEM_INODE]);
 316		break;
 317	case LOCAL_GROUP_QUOTA_SYSTEM_INODE:
 318		lockdep_set_class(&inode->i_rwsem,
 319				  &ocfs2_sysfile_lock_key[LOCAL_GROUP_QUOTA_SYSTEM_INODE]);
 320		break;
 321	default:
 322		WARN_ONCE(1, "Unknown sysfile type %d\n", args->fi_sysfile_type);
 323	}
 324	if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
 325	    args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
 326	    args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
 327	    args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
 328		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
 329				  &ocfs2_quota_ip_alloc_sem_key);
 330	else
 331		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
 332				  &ocfs2_file_ip_alloc_sem_key);
 333#endif
 334
 335	return 0;
 336}
 337
 338void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 339			  int create_ino)
 340{
 341	struct super_block *sb;
 342	struct ocfs2_super *osb;
 343	int use_plocks = 1;
 344
 345	sb = inode->i_sb;
 346	osb = OCFS2_SB(sb);
 347
 348	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
 349	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
 350		use_plocks = 0;
 351
 352	/*
 353	 * These have all been checked by ocfs2_read_inode_block() or set
 354	 * by ocfs2_mknod_locked(), so a failure is a code bug.
 355	 */
 356	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
 357						cannot create a superblock
 358						inode today.  change if
 359						that is needed. */
 360	BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
 361	BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
 362
 363
 364	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 365	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
 366	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
 367
 368	inode_set_iversion(inode, 1);
 369	inode->i_generation = le32_to_cpu(fe->i_generation);
 370	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 371	inode->i_mode = le16_to_cpu(fe->i_mode);
 372	i_uid_write(inode, le32_to_cpu(fe->i_uid));
 373	i_gid_write(inode, le32_to_cpu(fe->i_gid));
 374
 375	/* Fast symlinks will have i_size but no allocated clusters. */
 376	if (S_ISLNK(inode->i_mode) && !fe->i_clusters) {
 377		inode->i_blocks = 0;
 378		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
 379	} else {
 380		inode->i_blocks = ocfs2_inode_sector_count(inode);
 381		inode->i_mapping->a_ops = &ocfs2_aops;
 382	}
 383	inode_set_atime(inode, le64_to_cpu(fe->i_atime),
 384		        le32_to_cpu(fe->i_atime_nsec));
 385	inode_set_mtime(inode, le64_to_cpu(fe->i_mtime),
 386		        le32_to_cpu(fe->i_mtime_nsec));
 387	inode_set_ctime(inode, le64_to_cpu(fe->i_ctime),
 388		        le32_to_cpu(fe->i_ctime_nsec));
 389
 390	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
 391		mlog(ML_ERROR,
 392		     "ip_blkno %llu != i_blkno %llu!\n",
 393		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
 394		     (unsigned long long)le64_to_cpu(fe->i_blkno));
 395
 396	set_nlink(inode, ocfs2_read_links_count(fe));
 397
 398	trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
 399				   le32_to_cpu(fe->i_flags));
 400	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
 401		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
 402		inode->i_flags |= S_NOQUOTA;
 403	}
 404  
 405	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
 406		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
 407	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
 408		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
 409	} else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
 410		inode->i_flags |= S_NOQUOTA;
 411	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
 412		/* we can't actually hit this as read_inode can't
 413		 * handle superblocks today ;-) */
 414		BUG();
 415	}
 416
 417	switch (inode->i_mode & S_IFMT) {
 418	    case S_IFREG:
 419		    if (use_plocks)
 420			    inode->i_fop = &ocfs2_fops;
 421		    else
 422			    inode->i_fop = &ocfs2_fops_no_plocks;
 423		    inode->i_op = &ocfs2_file_iops;
 424		    i_size_write(inode, le64_to_cpu(fe->i_size));
 425		    break;
 426	    case S_IFDIR:
 427		    inode->i_op = &ocfs2_dir_iops;
 428		    if (use_plocks)
 429			    inode->i_fop = &ocfs2_dops;
 430		    else
 431			    inode->i_fop = &ocfs2_dops_no_plocks;
 432		    i_size_write(inode, le64_to_cpu(fe->i_size));
 433		    OCFS2_I(inode)->ip_dir_lock_gen = 1;
 434		    break;
 435	    case S_IFLNK:
 436		    inode->i_op = &ocfs2_symlink_inode_operations;
 437		    inode_nohighmem(inode);
 438		    i_size_write(inode, le64_to_cpu(fe->i_size));
 439		    break;
 440	    default:
 441		    inode->i_op = &ocfs2_special_file_iops;
 442		    init_special_inode(inode, inode->i_mode,
 443				       inode->i_rdev);
 444		    break;
 445	}
 446
 447	if (create_ino) {
 448		inode->i_ino = ino_from_blkno(inode->i_sb,
 449			       le64_to_cpu(fe->i_blkno));
 450
 451		/*
 452		 * If we ever want to create system files from kernel,
 453		 * the generation argument to
 454		 * ocfs2_inode_lock_res_init() will have to change.
 455		 */
 456		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
 457
 458		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
 459					  OCFS2_LOCK_TYPE_META, 0, inode);
 460
 461		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
 462					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
 463	}
 464
 465	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
 466				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
 467				  inode);
 468
 469	ocfs2_set_inode_flags(inode);
 470
 471	OCFS2_I(inode)->ip_last_used_slot = 0;
 472	OCFS2_I(inode)->ip_last_used_group = 0;
 473
 474	if (S_ISDIR(inode->i_mode))
 475		ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
 476				    OCFS2_RESV_FLAG_DIR);
 477}
 478
 479static int ocfs2_read_locked_inode(struct inode *inode,
 480				   struct ocfs2_find_inode_args *args)
 481{
 482	struct super_block *sb;
 483	struct ocfs2_super *osb;
 484	struct ocfs2_dinode *fe;
 485	struct buffer_head *bh = NULL;
 486	int status, can_lock, lock_level = 0;
 487	u32 generation = 0;
 488
 489	status = -EINVAL;
 490	sb = inode->i_sb;
 491	osb = OCFS2_SB(sb);
 492
 493	/*
 494	 * To improve performance of cold-cache inode stats, we take
 495	 * the cluster lock here if possible.
 496	 *
 497	 * Generally, OCFS2 never trusts the contents of an inode
 498	 * unless it's holding a cluster lock, so taking it here isn't
 499	 * a correctness issue as much as it is a performance
 500	 * improvement.
 501	 *
 502	 * There are three times when taking the lock is not a good idea:
 503	 *
 504	 * 1) During startup, before we have initialized the DLM.
 505	 *
 506	 * 2) If we are reading certain system files which never get
 507	 *    cluster locks (local alloc, truncate log).
 508	 *
 509	 * 3) If the process doing the iget() is responsible for
 510	 *    orphan dir recovery. We're holding the orphan dir lock and
 511	 *    can get into a deadlock with another process on another
 512	 *    node in ->delete_inode().
 513	 *
 514	 * #1 and #2 can be simply solved by never taking the lock
 515	 * here for system files (which are the only type we read
 516	 * during mount). It's a heavier approach, but our main
 517	 * concern is user-accessible files anyway.
 518	 *
 519	 * #3 works itself out because we'll eventually take the
 520	 * cluster lock before trusting anything anyway.
 521	 */
 522	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
 523		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
 524		&& !ocfs2_mount_local(osb);
 525
 526	trace_ocfs2_read_locked_inode(
 527		(unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
 528
 529	/*
 530	 * To maintain backwards compatibility with older versions of
 531	 * ocfs2-tools, we still store the generation value for system
 532	 * files. The only ones that actually matter to userspace are
 533	 * the journals, but it's easier and inexpensive to just flag
 534	 * all system files similarly.
 535	 */
 536	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
 537		generation = osb->fs_generation;
 538
 539	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
 540				  OCFS2_LOCK_TYPE_META,
 541				  generation, inode);
 542
 543	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
 544				  OCFS2_LOCK_TYPE_OPEN,
 545				  0, inode);
 546
 547	if (can_lock) {
 548		status = ocfs2_open_lock(inode);
 549		if (status) {
 550			make_bad_inode(inode);
 551			mlog_errno(status);
 552			return status;
 553		}
 554		status = ocfs2_inode_lock(inode, NULL, lock_level);
 555		if (status) {
 556			make_bad_inode(inode);
 557			mlog_errno(status);
 558			return status;
 559		}
 560	}
 561
 562	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
 563		status = ocfs2_try_open_lock(inode, 0);
 564		if (status) {
 565			make_bad_inode(inode);
 566			return status;
 567		}
 568	}
 569
 570	if (can_lock) {
 571		if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
 572			status = ocfs2_filecheck_read_inode_block_full(inode,
 573						&bh, OCFS2_BH_IGNORE_CACHE, 0);
 574		else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
 575			status = ocfs2_filecheck_read_inode_block_full(inode,
 576						&bh, OCFS2_BH_IGNORE_CACHE, 1);
 577		else
 578			status = ocfs2_read_inode_block_full(inode,
 579						&bh, OCFS2_BH_IGNORE_CACHE);
 580	} else {
 581		status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
 582		/*
 583		 * If buffer is in jbd, then its checksum may not have been
 584		 * computed as yet.
 585		 */
 586		if (!status && !buffer_jbd(bh)) {
 587			if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
 588				status = ocfs2_filecheck_validate_inode_block(
 589								osb->sb, bh);
 590			else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
 591				status = ocfs2_filecheck_repair_inode_block(
 592								osb->sb, bh);
 593			else
 594				status = ocfs2_validate_inode_block(
 595								osb->sb, bh);
 596		}
 597	}
 598	if (status < 0) {
 599		mlog_errno(status);
 600		goto bail;
 601	}
 602
 603	status = -EINVAL;
 604	fe = (struct ocfs2_dinode *) bh->b_data;
 605
 606	/*
 607	 * This is a code bug. Right now the caller needs to
 608	 * understand whether it is asking for a system file inode or
 609	 * not so the proper lock names can be built.
 610	 */
 611	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
 612			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
 613			"Inode %llu: system file state is ambiguous\n",
 614			(unsigned long long)args->fi_blkno);
 615
 616	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
 617	    S_ISBLK(le16_to_cpu(fe->i_mode)))
 618		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 619
 620	ocfs2_populate_inode(inode, fe, 0);
 621
 622	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
 623
 624	if (buffer_dirty(bh) && !buffer_jbd(bh)) {
 625		if (can_lock) {
 626			ocfs2_inode_unlock(inode, lock_level);
 627			lock_level = 1;
 628			ocfs2_inode_lock(inode, NULL, lock_level);
 629		}
 630		status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
 631		if (status < 0) {
 632			mlog_errno(status);
 633			goto bail;
 634		}
 635	}
 636
 637	status = 0;
 638
 639bail:
 640	if (can_lock)
 641		ocfs2_inode_unlock(inode, lock_level);
 642
 643	if (status < 0)
 644		make_bad_inode(inode);
 645
 646	brelse(bh);
 647
 648	return status;
 649}
 650
 651void ocfs2_sync_blockdev(struct super_block *sb)
 652{
 653	sync_blockdev(sb->s_bdev);
 654}
 655
 656static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
 657				     struct inode *inode,
 658				     struct buffer_head *fe_bh)
 659{
 660	int status = 0;
 661	struct ocfs2_dinode *fe;
 662	handle_t *handle = NULL;
 663
 664	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 665
 666	/*
 667	 * This check will also skip truncate of inodes with inline
 668	 * data and fast symlinks.
 669	 */
 670	if (fe->i_clusters) {
 671		if (ocfs2_should_order_data(inode))
 672			ocfs2_begin_ordered_truncate(inode, 0);
 673
 674		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 675		if (IS_ERR(handle)) {
 676			status = PTR_ERR(handle);
 677			handle = NULL;
 678			mlog_errno(status);
 679			goto out;
 680		}
 681
 682		status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 683						 fe_bh,
 684						 OCFS2_JOURNAL_ACCESS_WRITE);
 685		if (status < 0) {
 686			mlog_errno(status);
 687			goto out;
 688		}
 689
 690		i_size_write(inode, 0);
 691
 692		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 693		if (status < 0) {
 694			mlog_errno(status);
 695			goto out;
 696		}
 697
 698		ocfs2_commit_trans(osb, handle);
 699		handle = NULL;
 700
 701		status = ocfs2_commit_truncate(osb, inode, fe_bh);
 702		if (status < 0)
 703			mlog_errno(status);
 704	}
 705
 706out:
 707	if (handle)
 708		ocfs2_commit_trans(osb, handle);
 709	return status;
 710}
 711
 712static int ocfs2_remove_inode(struct inode *inode,
 713			      struct buffer_head *di_bh,
 714			      struct inode *orphan_dir_inode,
 715			      struct buffer_head *orphan_dir_bh)
 716{
 717	int status;
 718	struct inode *inode_alloc_inode = NULL;
 719	struct buffer_head *inode_alloc_bh = NULL;
 720	handle_t *handle;
 721	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 722	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
 723
 724	inode_alloc_inode =
 725		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
 726					    le16_to_cpu(di->i_suballoc_slot));
 727	if (!inode_alloc_inode) {
 728		status = -ENOENT;
 729		mlog_errno(status);
 730		goto bail;
 731	}
 732
 733	inode_lock(inode_alloc_inode);
 734	status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
 735	if (status < 0) {
 736		inode_unlock(inode_alloc_inode);
 737
 738		mlog_errno(status);
 739		goto bail;
 740	}
 741
 742	handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
 743				   ocfs2_quota_trans_credits(inode->i_sb));
 744	if (IS_ERR(handle)) {
 745		status = PTR_ERR(handle);
 746		mlog_errno(status);
 747		goto bail_unlock;
 748	}
 749
 750	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
 751		status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
 752					  orphan_dir_bh, false);
 753		if (status < 0) {
 754			mlog_errno(status);
 755			goto bail_commit;
 756		}
 757	}
 758
 759	/* set the inodes dtime */
 760	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 761					 OCFS2_JOURNAL_ACCESS_WRITE);
 762	if (status < 0) {
 763		mlog_errno(status);
 764		goto bail_commit;
 765	}
 766
 767	di->i_dtime = cpu_to_le64(ktime_get_real_seconds());
 768	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
 769	ocfs2_journal_dirty(handle, di_bh);
 770
 771	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
 772	dquot_free_inode(inode);
 773
 774	status = ocfs2_free_dinode(handle, inode_alloc_inode,
 775				   inode_alloc_bh, di);
 776	if (status < 0)
 777		mlog_errno(status);
 778
 779bail_commit:
 780	ocfs2_commit_trans(osb, handle);
 781bail_unlock:
 782	ocfs2_inode_unlock(inode_alloc_inode, 1);
 783	inode_unlock(inode_alloc_inode);
 784	brelse(inode_alloc_bh);
 785bail:
 786	iput(inode_alloc_inode);
 787
 788	return status;
 789}
 790
 791/*
 792 * Serialize with orphan dir recovery. If the process doing
 793 * recovery on this orphan dir does an iget() with the dir
 794 * i_rwsem held, we'll deadlock here. Instead we detect this
 795 * and exit early - recovery will wipe this inode for us.
 796 */
 797static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
 798					     int slot)
 799{
 800	int ret = 0;
 801
 802	spin_lock(&osb->osb_lock);
 803	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
 804		ret = -EDEADLK;
 805		goto out;
 806	}
 807	/* This signals to the orphan recovery process that it should
 808	 * wait for us to handle the wipe. */
 809	osb->osb_orphan_wipes[slot]++;
 810out:
 811	spin_unlock(&osb->osb_lock);
 812	trace_ocfs2_check_orphan_recovery_state(slot, ret);
 813	return ret;
 814}
 815
 816static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
 817					 int slot)
 818{
 819	spin_lock(&osb->osb_lock);
 820	osb->osb_orphan_wipes[slot]--;
 821	spin_unlock(&osb->osb_lock);
 822
 823	wake_up(&osb->osb_wipe_event);
 824}
 825
 826static int ocfs2_wipe_inode(struct inode *inode,
 827			    struct buffer_head *di_bh)
 828{
 829	int status, orphaned_slot = -1;
 830	struct inode *orphan_dir_inode = NULL;
 831	struct buffer_head *orphan_dir_bh = NULL;
 832	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 833	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
 834
 835	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
 836		orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
 837
 838		status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
 839		if (status)
 840			return status;
 841
 842		orphan_dir_inode = ocfs2_get_system_file_inode(osb,
 843							       ORPHAN_DIR_SYSTEM_INODE,
 844							       orphaned_slot);
 845		if (!orphan_dir_inode) {
 846			status = -ENOENT;
 847			mlog_errno(status);
 848			goto bail;
 849		}
 850
 851		/* Lock the orphan dir. The lock will be held for the entire
 852		 * delete_inode operation. We do this now to avoid races with
 853		 * recovery completion on other nodes. */
 854		inode_lock(orphan_dir_inode);
 855		status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
 856		if (status < 0) {
 857			inode_unlock(orphan_dir_inode);
 858
 859			mlog_errno(status);
 860			goto bail;
 861		}
 862	}
 863
 864	/* we do this while holding the orphan dir lock because we
 865	 * don't want recovery being run from another node to try an
 866	 * inode delete underneath us -- this will result in two nodes
 867	 * truncating the same file! */
 868	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
 869	if (status < 0) {
 870		mlog_errno(status);
 871		goto bail_unlock_dir;
 872	}
 873
 874	/* Remove any dir index tree */
 875	if (S_ISDIR(inode->i_mode)) {
 876		status = ocfs2_dx_dir_truncate(inode, di_bh);
 877		if (status) {
 878			mlog_errno(status);
 879			goto bail_unlock_dir;
 880		}
 881	}
 882
 883	/*Free extended attribute resources associated with this inode.*/
 884	status = ocfs2_xattr_remove(inode, di_bh);
 885	if (status < 0) {
 886		mlog_errno(status);
 887		goto bail_unlock_dir;
 888	}
 889
 890	status = ocfs2_remove_refcount_tree(inode, di_bh);
 891	if (status < 0) {
 892		mlog_errno(status);
 893		goto bail_unlock_dir;
 894	}
 895
 896	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
 897				    orphan_dir_bh);
 898	if (status < 0)
 899		mlog_errno(status);
 900
 901bail_unlock_dir:
 902	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
 903		return status;
 904
 905	ocfs2_inode_unlock(orphan_dir_inode, 1);
 906	inode_unlock(orphan_dir_inode);
 907	brelse(orphan_dir_bh);
 908bail:
 909	iput(orphan_dir_inode);
 910	ocfs2_signal_wipe_completion(osb, orphaned_slot);
 911
 912	return status;
 913}
 914
 915/* There is a series of simple checks that should be done before a
 916 * trylock is even considered. Encapsulate those in this function. */
 917static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
 918{
 919	int ret = 0;
 920	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 921	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 922
 923	trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
 924					     (unsigned long long)oi->ip_blkno,
 925					     oi->ip_flags);
 926
 927	/* We shouldn't be getting here for the root directory
 928	 * inode.. */
 929	if (inode == osb->root_inode) {
 930		mlog(ML_ERROR, "Skipping delete of root inode.\n");
 931		goto bail;
 932	}
 933
 934	/*
 935	 * If we're coming from downconvert_thread we can't go into our own
 936	 * voting [hello, deadlock city!] so we cannot delete the inode. But
 937	 * since we dropped last inode ref when downconverting dentry lock,
 938	 * we cannot have the file open and thus the node doing unlink will
 939	 * take care of deleting the inode.
 940	 */
 941	if (current == osb->dc_task)
 942		goto bail;
 943
 944	spin_lock(&oi->ip_lock);
 945	/* OCFS2 *never* deletes system files. This should technically
 946	 * never get here as system file inodes should always have a
 947	 * positive link count. */
 948	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
 949		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
 950		     (unsigned long long)oi->ip_blkno);
 951		goto bail_unlock;
 952	}
 953
 954	ret = 1;
 955bail_unlock:
 956	spin_unlock(&oi->ip_lock);
 957bail:
 958	return ret;
 959}
 960
 961/* Query the cluster to determine whether we should wipe an inode from
 962 * disk or not.
 963 *
 964 * Requires the inode to have the cluster lock. */
 965static int ocfs2_query_inode_wipe(struct inode *inode,
 966				  struct buffer_head *di_bh,
 967				  int *wipe)
 968{
 969	int status = 0, reason = 0;
 970	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 971	struct ocfs2_dinode *di;
 972
 973	*wipe = 0;
 974
 975	trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
 976					   inode->i_nlink);
 977
 978	/* While we were waiting for the cluster lock in
 979	 * ocfs2_delete_inode, another node might have asked to delete
 980	 * the inode. Recheck our flags to catch this. */
 981	if (!ocfs2_inode_is_valid_to_delete(inode)) {
 982		reason = 1;
 983		goto bail;
 984	}
 985
 986	/* Now that we have an up to date inode, we can double check
 987	 * the link count. */
 988	if (inode->i_nlink)
 989		goto bail;
 990
 991	/* Do some basic inode verification... */
 992	di = (struct ocfs2_dinode *) di_bh->b_data;
 993	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
 994	    !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
 995		/*
 996		 * Inodes in the orphan dir must have ORPHANED_FL.  The only
 997		 * inodes that come back out of the orphan dir are reflink
 998		 * targets. A reflink target may be moved out of the orphan
 999		 * dir between the time we scan the directory and the time we
1000		 * process it. This would lead to HAS_REFCOUNT_FL being set but
1001		 * ORPHANED_FL not.
1002		 */
1003		if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
1004			reason = 2;
1005			goto bail;
1006		}
1007
1008		/* for lack of a better error? */
1009		status = -EEXIST;
1010		mlog(ML_ERROR,
1011		     "Inode %llu (on-disk %llu) not orphaned! "
1012		     "Disk flags  0x%x, inode flags 0x%x\n",
1013		     (unsigned long long)oi->ip_blkno,
1014		     (unsigned long long)le64_to_cpu(di->i_blkno),
1015		     le32_to_cpu(di->i_flags), oi->ip_flags);
1016		goto bail;
1017	}
1018
1019	/* has someone already deleted us?! baaad... */
1020	if (di->i_dtime) {
1021		status = -EEXIST;
1022		mlog_errno(status);
1023		goto bail;
1024	}
1025
1026	/*
1027	 * This is how ocfs2 determines whether an inode is still live
1028	 * within the cluster. Every node takes a shared read lock on
1029	 * the inode open lock in ocfs2_read_locked_inode(). When we
1030	 * get to ->delete_inode(), each node tries to convert it's
1031	 * lock to an exclusive. Trylocks are serialized by the inode
1032	 * meta data lock. If the upconvert succeeds, we know the inode
1033	 * is no longer live and can be deleted.
1034	 *
1035	 * Though we call this with the meta data lock held, the
1036	 * trylock keeps us from ABBA deadlock.
1037	 */
1038	status = ocfs2_try_open_lock(inode, 1);
1039	if (status == -EAGAIN) {
1040		status = 0;
1041		reason = 3;
1042		goto bail;
1043	}
1044	if (status < 0) {
1045		mlog_errno(status);
1046		goto bail;
1047	}
1048
1049	*wipe = 1;
1050	trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
1051
1052bail:
1053	trace_ocfs2_query_inode_wipe_end(status, reason);
1054	return status;
1055}
1056
1057/* Support function for ocfs2_delete_inode. Will help us keep the
1058 * inode data in a consistent state for clear_inode. Always truncates
1059 * pages, optionally sync's them first. */
1060static void ocfs2_cleanup_delete_inode(struct inode *inode,
1061				       int sync_data)
1062{
1063	trace_ocfs2_cleanup_delete_inode(
1064		(unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
1065	if (sync_data)
1066		filemap_write_and_wait(inode->i_mapping);
1067	truncate_inode_pages_final(&inode->i_data);
1068}
1069
1070static void ocfs2_delete_inode(struct inode *inode)
1071{
1072	int wipe, status;
1073	sigset_t oldset;
1074	struct buffer_head *di_bh = NULL;
1075	struct ocfs2_dinode *di = NULL;
1076
1077	trace_ocfs2_delete_inode(inode->i_ino,
1078				 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1079				 is_bad_inode(inode));
1080
1081	/* When we fail in read_inode() we mark inode as bad. The second test
1082	 * catches the case when inode allocation fails before allocating
1083	 * a block for inode. */
1084	if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
1085		goto bail;
1086
1087	if (!ocfs2_inode_is_valid_to_delete(inode)) {
1088		/* It's probably not necessary to truncate_inode_pages
1089		 * here but we do it for safety anyway (it will most
1090		 * likely be a no-op anyway) */
1091		ocfs2_cleanup_delete_inode(inode, 0);
1092		goto bail;
1093	}
1094
1095	dquot_initialize(inode);
1096
1097	/* We want to block signals in delete_inode as the lock and
1098	 * messaging paths may return us -ERESTARTSYS. Which would
1099	 * cause us to exit early, resulting in inodes being orphaned
1100	 * forever. */
1101	ocfs2_block_signals(&oldset);
1102
1103	/*
1104	 * Synchronize us against ocfs2_get_dentry. We take this in
1105	 * shared mode so that all nodes can still concurrently
1106	 * process deletes.
1107	 */
1108	status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1109	if (status < 0) {
1110		mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1111		ocfs2_cleanup_delete_inode(inode, 0);
1112		goto bail_unblock;
1113	}
1114	/* Lock down the inode. This gives us an up to date view of
1115	 * it's metadata (for verification), and allows us to
1116	 * serialize delete_inode on multiple nodes.
1117	 *
1118	 * Even though we might be doing a truncate, we don't take the
1119	 * allocation lock here as it won't be needed - nobody will
1120	 * have the file open.
1121	 */
1122	status = ocfs2_inode_lock(inode, &di_bh, 1);
1123	if (status < 0) {
1124		if (status != -ENOENT)
1125			mlog_errno(status);
1126		ocfs2_cleanup_delete_inode(inode, 0);
1127		goto bail_unlock_nfs_sync;
1128	}
1129
1130	di = (struct ocfs2_dinode *)di_bh->b_data;
1131	/* Skip inode deletion and wait for dio orphan entry recovered
1132	 * first */
1133	if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
1134		ocfs2_cleanup_delete_inode(inode, 0);
1135		goto bail_unlock_inode;
1136	}
1137
1138	/* Query the cluster. This will be the final decision made
1139	 * before we go ahead and wipe the inode. */
1140	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1141	if (!wipe || status < 0) {
1142		/* Error and remote inode busy both mean we won't be
1143		 * removing the inode, so they take almost the same
1144		 * path. */
1145		if (status < 0)
1146			mlog_errno(status);
1147
1148		/* Someone in the cluster has disallowed a wipe of
1149		 * this inode, or it was never completely
1150		 * orphaned. Write out the pages and exit now. */
1151		ocfs2_cleanup_delete_inode(inode, 1);
1152		goto bail_unlock_inode;
1153	}
1154
1155	ocfs2_cleanup_delete_inode(inode, 0);
1156
1157	status = ocfs2_wipe_inode(inode, di_bh);
1158	if (status < 0) {
1159		if (status != -EDEADLK)
1160			mlog_errno(status);
1161		goto bail_unlock_inode;
1162	}
1163
1164	/*
1165	 * Mark the inode as successfully deleted.
1166	 *
1167	 * This is important for ocfs2_clear_inode() as it will check
1168	 * this flag and skip any checkpointing work
1169	 *
1170	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1171	 * the LVB for other nodes.
1172	 */
1173	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1174
1175bail_unlock_inode:
1176	ocfs2_inode_unlock(inode, 1);
1177	brelse(di_bh);
1178
1179bail_unlock_nfs_sync:
1180	ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1181
1182bail_unblock:
1183	ocfs2_unblock_signals(&oldset);
1184bail:
1185	return;
1186}
1187
1188static void ocfs2_clear_inode(struct inode *inode)
1189{
1190	int status;
1191	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1192	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1193
1194	clear_inode(inode);
1195	trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1196				inode->i_nlink);
1197
1198	mlog_bug_on_msg(osb == NULL,
1199			"Inode=%lu\n", inode->i_ino);
1200
1201	dquot_drop(inode);
1202
1203	/* To prevent remote deletes we hold open lock before, now it
1204	 * is time to unlock PR and EX open locks. */
1205	ocfs2_open_unlock(inode);
1206
1207	/* Do these before all the other work so that we don't bounce
1208	 * the downconvert thread while waiting to destroy the locks. */
1209	ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres);
1210	ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres);
1211	ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres);
1212
1213	ocfs2_resv_discard(&osb->osb_la_resmap,
1214			   &oi->ip_la_data_resv);
1215	ocfs2_resv_init_once(&oi->ip_la_data_resv);
1216
1217	/* We very well may get a clear_inode before all an inodes
1218	 * metadata has hit disk. Of course, we can't drop any cluster
1219	 * locks until the journal has finished with it. The only
1220	 * exception here are successfully wiped inodes - their
1221	 * metadata can now be considered to be part of the system
1222	 * inodes from which it came. */
1223	if (!(oi->ip_flags & OCFS2_INODE_DELETED))
1224		ocfs2_checkpoint_inode(inode);
1225
1226	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1227			"Clear inode of %llu, inode has io markers\n",
1228			(unsigned long long)oi->ip_blkno);
1229	mlog_bug_on_msg(!list_empty(&oi->ip_unwritten_list),
1230			"Clear inode of %llu, inode has unwritten extents\n",
1231			(unsigned long long)oi->ip_blkno);
1232
1233	ocfs2_extent_map_trunc(inode, 0);
1234
1235	status = ocfs2_drop_inode_locks(inode);
1236	if (status < 0)
1237		mlog_errno(status);
1238
1239	ocfs2_lock_res_free(&oi->ip_rw_lockres);
1240	ocfs2_lock_res_free(&oi->ip_inode_lockres);
1241	ocfs2_lock_res_free(&oi->ip_open_lockres);
1242
1243	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1244
1245	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1246			"Clear inode of %llu, inode has %u cache items\n",
1247			(unsigned long long)oi->ip_blkno,
1248			INODE_CACHE(inode)->ci_num_cached);
1249
1250	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1251			"Clear inode of %llu, inode has a bad flag\n",
1252			(unsigned long long)oi->ip_blkno);
1253
1254	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1255			"Clear inode of %llu, inode is locked\n",
1256			(unsigned long long)oi->ip_blkno);
1257
1258	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1259			"Clear inode of %llu, io_mutex is locked\n",
1260			(unsigned long long)oi->ip_blkno);
1261	mutex_unlock(&oi->ip_io_mutex);
1262
1263	/*
1264	 * down_trylock() returns 0, down_write_trylock() returns 1
1265	 * kernel 1, world 0
1266	 */
1267	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1268			"Clear inode of %llu, alloc_sem is locked\n",
1269			(unsigned long long)oi->ip_blkno);
1270	up_write(&oi->ip_alloc_sem);
1271
1272	mlog_bug_on_msg(oi->ip_open_count,
1273			"Clear inode of %llu has open count %d\n",
1274			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
1275
1276	/* Clear all other flags. */
1277	oi->ip_flags = 0;
1278	oi->ip_dir_start_lookup = 0;
1279	oi->ip_blkno = 0ULL;
1280
1281	/*
1282	 * ip_jinode is used to track txns against this inode. We ensure that
1283	 * the journal is flushed before journal shutdown. Thus it is safe to
1284	 * have inodes get cleaned up after journal shutdown.
1285	 */
1286	if (!osb->journal)
1287		return;
1288
1289	jbd2_journal_release_jbd_inode(osb->journal->j_journal,
1290				       &oi->ip_jinode);
1291}
1292
1293void ocfs2_evict_inode(struct inode *inode)
1294{
1295	write_inode_now(inode, 1);
1296
1297	if (!inode->i_nlink ||
1298	    (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1299		ocfs2_delete_inode(inode);
1300	} else {
1301		truncate_inode_pages_final(&inode->i_data);
1302	}
1303	ocfs2_clear_inode(inode);
1304}
1305
1306/*
1307 * This is called from our getattr.
1308 */
1309int ocfs2_inode_revalidate(struct dentry *dentry)
1310{
1311	struct inode *inode = d_inode(dentry);
1312	int status = 0;
1313
1314	trace_ocfs2_inode_revalidate(inode,
1315		inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1316		inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
1317
1318	if (!inode) {
1319		status = -ENOENT;
1320		goto bail;
1321	}
1322
1323	spin_lock(&OCFS2_I(inode)->ip_lock);
1324	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1325		spin_unlock(&OCFS2_I(inode)->ip_lock);
1326		status = -ENOENT;
1327		goto bail;
1328	}
1329	spin_unlock(&OCFS2_I(inode)->ip_lock);
1330
1331	/* Let ocfs2_inode_lock do the work of updating our struct
1332	 * inode for us. */
1333	status = ocfs2_inode_lock(inode, NULL, 0);
1334	if (status < 0) {
1335		if (status != -ENOENT)
1336			mlog_errno(status);
1337		goto bail;
1338	}
1339	ocfs2_inode_unlock(inode, 0);
1340bail:
1341	return status;
1342}
1343
1344/*
1345 * Updates a disk inode from a
1346 * struct inode.
1347 * Only takes ip_lock.
1348 */
1349int ocfs2_mark_inode_dirty(handle_t *handle,
1350			   struct inode *inode,
1351			   struct buffer_head *bh)
1352{
1353	int status;
1354	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1355
1356	trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
1357
1358	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1359					 OCFS2_JOURNAL_ACCESS_WRITE);
1360	if (status < 0) {
1361		mlog_errno(status);
1362		goto leave;
1363	}
1364
1365	spin_lock(&OCFS2_I(inode)->ip_lock);
1366	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1367	ocfs2_get_inode_flags(OCFS2_I(inode));
1368	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1369	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1370	spin_unlock(&OCFS2_I(inode)->ip_lock);
1371
1372	fe->i_size = cpu_to_le64(i_size_read(inode));
1373	ocfs2_set_links_count(fe, inode->i_nlink);
1374	fe->i_uid = cpu_to_le32(i_uid_read(inode));
1375	fe->i_gid = cpu_to_le32(i_gid_read(inode));
1376	fe->i_mode = cpu_to_le16(inode->i_mode);
1377	fe->i_atime = cpu_to_le64(inode_get_atime_sec(inode));
1378	fe->i_atime_nsec = cpu_to_le32(inode_get_atime_nsec(inode));
1379	fe->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
1380	fe->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
1381	fe->i_mtime = cpu_to_le64(inode_get_mtime_sec(inode));
1382	fe->i_mtime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
1383
1384	ocfs2_journal_dirty(handle, bh);
1385	ocfs2_update_inode_fsync_trans(handle, inode, 1);
1386leave:
1387	return status;
1388}
1389
1390/*
1391 *
1392 * Updates a struct inode from a disk inode.
1393 * does no i/o, only takes ip_lock.
1394 */
1395void ocfs2_refresh_inode(struct inode *inode,
1396			 struct ocfs2_dinode *fe)
1397{
1398	spin_lock(&OCFS2_I(inode)->ip_lock);
1399
1400	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1401	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1402	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1403	ocfs2_set_inode_flags(inode);
1404	i_size_write(inode, le64_to_cpu(fe->i_size));
1405	set_nlink(inode, ocfs2_read_links_count(fe));
1406	i_uid_write(inode, le32_to_cpu(fe->i_uid));
1407	i_gid_write(inode, le32_to_cpu(fe->i_gid));
1408	inode->i_mode = le16_to_cpu(fe->i_mode);
1409	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1410		inode->i_blocks = 0;
1411	else
1412		inode->i_blocks = ocfs2_inode_sector_count(inode);
1413	inode_set_atime(inode, le64_to_cpu(fe->i_atime),
1414			le32_to_cpu(fe->i_atime_nsec));
1415	inode_set_mtime(inode, le64_to_cpu(fe->i_mtime),
1416			le32_to_cpu(fe->i_mtime_nsec));
1417	inode_set_ctime(inode, le64_to_cpu(fe->i_ctime),
1418			le32_to_cpu(fe->i_ctime_nsec));
1419
1420	spin_unlock(&OCFS2_I(inode)->ip_lock);
1421}
1422
1423int ocfs2_validate_inode_block(struct super_block *sb,
1424			       struct buffer_head *bh)
1425{
1426	int rc;
1427	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1428
1429	trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
1430
1431	BUG_ON(!buffer_uptodate(bh));
1432
1433	/*
1434	 * If the ecc fails, we return the error but otherwise
1435	 * leave the filesystem running.  We know any error is
1436	 * local to this block.
1437	 */
1438	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1439	if (rc) {
1440		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1441		     (unsigned long long)bh->b_blocknr);
1442		goto bail;
1443	}
1444
1445	if ((!di->i_links_count && !di->i_links_count_hi) || !di->i_mode) {
1446		mlog(ML_ERROR, "Invalid dinode #%llu: "
1447			"Corrupt state (nlink = %u or mode = %u) detected!\n",
1448		        (unsigned long long)bh->b_blocknr,
1449			ocfs2_read_links_count(di), le16_to_cpu(di->i_mode));
1450		rc = -EFSCORRUPTED;
1451		goto bail;
1452	}
1453	/*
1454	 * Errors after here are fatal.
1455	 */
1456
1457	rc = -EINVAL;
1458
1459	if (!OCFS2_IS_VALID_DINODE(di)) {
1460		rc = ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1461				 (unsigned long long)bh->b_blocknr, 7,
1462				 di->i_signature);
1463		goto bail;
1464	}
1465
1466	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1467		rc = ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1468				 (unsigned long long)bh->b_blocknr,
1469				 (unsigned long long)le64_to_cpu(di->i_blkno));
1470		goto bail;
1471	}
1472
1473	if (!(le32_to_cpu(di->i_flags) & OCFS2_VALID_FL)) {
1474		rc = ocfs2_error(sb,
1475				 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1476				 (unsigned long long)bh->b_blocknr);
1477		goto bail;
1478	}
1479
1480	if (le32_to_cpu(di->i_fs_generation) !=
1481	    OCFS2_SB(sb)->fs_generation) {
1482		rc = ocfs2_error(sb,
1483				 "Invalid dinode #%llu: fs_generation is %u\n",
1484				 (unsigned long long)bh->b_blocknr,
1485				 le32_to_cpu(di->i_fs_generation));
1486		goto bail;
1487	}
1488
1489	if (le16_to_cpu(di->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
1490	    (u32)le16_to_cpu(di->i_suballoc_slot) > OCFS2_SB(sb)->max_slots - 1) {
1491		rc = ocfs2_error(sb, "Invalid dinode %llu: suballoc slot %u\n",
1492				 (unsigned long long)bh->b_blocknr,
1493				 le16_to_cpu(di->i_suballoc_slot));
1494		goto bail;
1495	}
1496
1497	if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) {
1498		struct ocfs2_inline_data *data = &di->id2.i_data;
1499
1500		if (le32_to_cpu(di->i_clusters)) {
1501			rc = ocfs2_error(sb,
1502					 "Invalid dinode %llu: %u clusters\n",
1503					 (unsigned long long)bh->b_blocknr,
1504					 le32_to_cpu(di->i_clusters));
1505			goto bail;
1506		}
1507
1508		if (le16_to_cpu(data->id_count) >
1509		    ocfs2_max_inline_data_with_xattr(sb, di)) {
1510			rc = ocfs2_error(sb,
1511					 "Invalid dinode #%llu: inline data id_count %u exceeds max %d\n",
1512					 (unsigned long long)bh->b_blocknr,
1513					 le16_to_cpu(data->id_count),
1514					 ocfs2_max_inline_data_with_xattr(sb, di));
1515			goto bail;
1516		}
1517
1518		if (le64_to_cpu(di->i_size) > le16_to_cpu(data->id_count)) {
1519			rc = ocfs2_error(sb,
1520					 "Invalid dinode #%llu: inline data i_size %llu exceeds id_count %u\n",
1521					 (unsigned long long)bh->b_blocknr,
1522					 (unsigned long long)le64_to_cpu(di->i_size),
1523					 le16_to_cpu(data->id_count));
1524			goto bail;
1525		}
1526	}
1527
1528	if (le32_to_cpu(di->i_flags) & OCFS2_CHAIN_FL) {
1529		struct ocfs2_chain_list *cl = &di->id2.i_chain;
1530		u16 bpc = 1 << (OCFS2_SB(sb)->s_clustersize_bits -
1531				sb->s_blocksize_bits);
1532
1533		if (le16_to_cpu(cl->cl_count) != ocfs2_chain_recs_per_inode(sb)) {
1534			rc = ocfs2_error(sb, "Invalid dinode %llu: chain list count %u\n",
1535					 (unsigned long long)bh->b_blocknr,
1536					 le16_to_cpu(cl->cl_count));
1537			goto bail;
1538		}
1539		if (le16_to_cpu(cl->cl_next_free_rec) > le16_to_cpu(cl->cl_count)) {
1540			rc = ocfs2_error(sb, "Invalid dinode %llu: chain list index %u\n",
1541					 (unsigned long long)bh->b_blocknr,
1542					 le16_to_cpu(cl->cl_next_free_rec));
1543			goto bail;
1544		}
1545		if (OCFS2_SB(sb)->bitmap_blkno &&
1546		    OCFS2_SB(sb)->bitmap_blkno != le64_to_cpu(di->i_blkno) &&
1547		    le16_to_cpu(cl->cl_bpc) != bpc) {
1548			rc = ocfs2_error(sb, "Invalid dinode %llu: bits per cluster %u\n",
1549					 (unsigned long long)bh->b_blocknr,
1550					 le16_to_cpu(cl->cl_bpc));
1551			goto bail;
1552		}
1553	}
1554
1555	rc = 0;
1556
1557bail:
1558	return rc;
1559}
1560
1561static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
1562						struct buffer_head *bh)
1563{
1564	int rc = 0;
1565	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1566
1567	trace_ocfs2_filecheck_validate_inode_block(
1568		(unsigned long long)bh->b_blocknr);
1569
1570	BUG_ON(!buffer_uptodate(bh));
1571
1572	/*
1573	 * Call ocfs2_validate_meta_ecc() first since it has ecc repair
1574	 * function, but we should not return error immediately when ecc
1575	 * validation fails, because the reason is quite likely the invalid
1576	 * inode number inputted.
1577	 */
1578	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1579	if (rc) {
1580		mlog(ML_ERROR,
1581		     "Filecheck: checksum failed for dinode %llu\n",
1582		     (unsigned long long)bh->b_blocknr);
1583		rc = -OCFS2_FILECHECK_ERR_BLOCKECC;
1584	}
1585
1586	if (!OCFS2_IS_VALID_DINODE(di)) {
1587		mlog(ML_ERROR,
1588		     "Filecheck: invalid dinode #%llu: signature = %.*s\n",
1589		     (unsigned long long)bh->b_blocknr, 7, di->i_signature);
1590		rc = -OCFS2_FILECHECK_ERR_INVALIDINO;
1591		goto bail;
1592	} else if (rc)
1593		goto bail;
1594
1595	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1596		mlog(ML_ERROR,
1597		     "Filecheck: invalid dinode #%llu: i_blkno is %llu\n",
1598		     (unsigned long long)bh->b_blocknr,
1599		     (unsigned long long)le64_to_cpu(di->i_blkno));
1600		rc = -OCFS2_FILECHECK_ERR_BLOCKNO;
1601		goto bail;
1602	}
1603
1604	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1605		mlog(ML_ERROR,
1606		     "Filecheck: invalid dinode #%llu: OCFS2_VALID_FL "
1607		     "not set\n",
1608		     (unsigned long long)bh->b_blocknr);
1609		rc = -OCFS2_FILECHECK_ERR_VALIDFLAG;
1610		goto bail;
1611	}
1612
1613	if (le32_to_cpu(di->i_fs_generation) !=
1614	    OCFS2_SB(sb)->fs_generation) {
1615		mlog(ML_ERROR,
1616		     "Filecheck: invalid dinode #%llu: fs_generation is %u\n",
1617		     (unsigned long long)bh->b_blocknr,
1618		     le32_to_cpu(di->i_fs_generation));
1619		rc = -OCFS2_FILECHECK_ERR_GENERATION;
1620	}
1621
1622bail:
1623	return rc;
1624}
1625
1626static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
1627					      struct buffer_head *bh)
1628{
1629	int changed = 0;
1630	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1631
1632	if (!ocfs2_filecheck_validate_inode_block(sb, bh))
1633		return 0;
1634
1635	trace_ocfs2_filecheck_repair_inode_block(
1636		(unsigned long long)bh->b_blocknr);
1637
1638	if (unlikely(ocfs2_emergency_state(OCFS2_SB(sb)))) {
1639		mlog(ML_ERROR,
1640		     "Filecheck: cannot repair dinode #%llu "
1641		     "on readonly filesystem\n",
1642		     (unsigned long long)bh->b_blocknr);
1643		return -OCFS2_FILECHECK_ERR_READONLY;
1644	}
1645
1646	if (buffer_jbd(bh)) {
1647		mlog(ML_ERROR,
1648		     "Filecheck: cannot repair dinode #%llu, "
1649		     "its buffer is in jbd\n",
1650		     (unsigned long long)bh->b_blocknr);
1651		return -OCFS2_FILECHECK_ERR_INJBD;
1652	}
1653
1654	if (!OCFS2_IS_VALID_DINODE(di)) {
1655		/* Cannot fix invalid inode block */
1656		return -OCFS2_FILECHECK_ERR_INVALIDINO;
1657	}
1658
1659	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1660		/* Cannot just add VALID_FL flag back as a fix,
1661		 * need more things to check here.
1662		 */
1663		return -OCFS2_FILECHECK_ERR_VALIDFLAG;
1664	}
1665
1666	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1667		di->i_blkno = cpu_to_le64(bh->b_blocknr);
1668		changed = 1;
1669		mlog(ML_ERROR,
1670		     "Filecheck: reset dinode #%llu: i_blkno to %llu\n",
1671		     (unsigned long long)bh->b_blocknr,
1672		     (unsigned long long)le64_to_cpu(di->i_blkno));
1673	}
1674
1675	if (le32_to_cpu(di->i_fs_generation) !=
1676	    OCFS2_SB(sb)->fs_generation) {
1677		di->i_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1678		changed = 1;
1679		mlog(ML_ERROR,
1680		     "Filecheck: reset dinode #%llu: fs_generation to %u\n",
1681		     (unsigned long long)bh->b_blocknr,
1682		     le32_to_cpu(di->i_fs_generation));
1683	}
1684
1685	if (ocfs2_dinode_has_extents(di) &&
1686	    le16_to_cpu(di->id2.i_list.l_next_free_rec) > le16_to_cpu(di->id2.i_list.l_count)) {
1687		di->id2.i_list.l_next_free_rec = di->id2.i_list.l_count;
1688		changed = 1;
1689		mlog(ML_ERROR,
1690		     "Filecheck: reset dinode #%llu: l_next_free_rec to %u\n",
1691		     (unsigned long long)bh->b_blocknr,
1692		     le16_to_cpu(di->id2.i_list.l_next_free_rec));
1693	}
1694
1695	if (changed || ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check)) {
1696		ocfs2_compute_meta_ecc(sb, bh->b_data, &di->i_check);
1697		mark_buffer_dirty(bh);
1698		mlog(ML_ERROR,
1699		     "Filecheck: reset dinode #%llu: compute meta ecc\n",
1700		     (unsigned long long)bh->b_blocknr);
1701	}
1702
1703	return 0;
1704}
1705
1706static int
1707ocfs2_filecheck_read_inode_block_full(struct inode *inode,
1708				      struct buffer_head **bh,
1709				      int flags, int type)
1710{
1711	int rc;
1712	struct buffer_head *tmp = *bh;
1713
1714	if (!type) /* Check inode block */
1715		rc = ocfs2_read_blocks(INODE_CACHE(inode),
1716				OCFS2_I(inode)->ip_blkno,
1717				1, &tmp, flags,
1718				ocfs2_filecheck_validate_inode_block);
1719	else /* Repair inode block */
1720		rc = ocfs2_read_blocks(INODE_CACHE(inode),
1721				OCFS2_I(inode)->ip_blkno,
1722				1, &tmp, flags,
1723				ocfs2_filecheck_repair_inode_block);
1724
1725	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1726	if (!rc && !*bh)
1727		*bh = tmp;
1728
1729	return rc;
1730}
1731
1732int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1733				int flags)
1734{
1735	int rc;
1736	struct buffer_head *tmp = *bh;
1737
1738	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1739			       1, &tmp, flags, ocfs2_validate_inode_block);
1740
1741	if (rc < 0)
1742		make_bad_inode(inode);
1743	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1744	if (!rc && !*bh)
1745		*bh = tmp;
1746
1747	return rc;
1748}
1749
1750int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1751{
1752	return ocfs2_read_inode_block_full(inode, bh, 0);
1753}
1754
1755
1756static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1757{
1758	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1759
1760	return oi->ip_blkno;
1761}
1762
1763static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1764{
1765	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1766
1767	return oi->vfs_inode.i_sb;
1768}
1769
1770static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1771__acquires(&oi->ip_lock)
1772{
1773	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1774
1775	spin_lock(&oi->ip_lock);
1776}
1777
1778static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1779__releases(&oi->ip_lock)
1780{
1781	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1782
1783	spin_unlock(&oi->ip_lock);
1784}
1785
1786static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1787{
1788	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1789
1790	mutex_lock(&oi->ip_io_mutex);
1791}
1792
1793static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1794{
1795	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1796
1797	mutex_unlock(&oi->ip_io_mutex);
1798}
1799
1800const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1801	.co_owner		= ocfs2_inode_cache_owner,
1802	.co_get_super		= ocfs2_inode_cache_get_super,
1803	.co_cache_lock		= ocfs2_inode_cache_lock,
1804	.co_cache_unlock	= ocfs2_inode_cache_unlock,
1805	.co_io_lock		= ocfs2_inode_cache_io_lock,
1806	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
1807};
1808
~/ Read-only file preview.