1// SPDX-License-Identifier: GPL-2.0-or-later
   2/* -*- mode: c; c-basic-offset: 8; -*-
   3 * vim: noexpandtab sw=8 ts=8 sts=0:
   4 *
   5 * inode.c
   6 *
   7 * vfs' aops, fops, dops and iops
   8 *
   9 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  10 */
  11
  12#include <linux/fs.h>
  13#include <linux/types.h>
  14#include <linux/highmem.h>
  15#include <linux/pagemap.h>
  16#include <linux/quotaops.h>
  17#include <linux/iversion.h>
  18
  19#include <asm/byteorder.h>
  20
  21#include <cluster/masklog.h>
  22
  23#include "ocfs2.h"
  24
  25#include "alloc.h"
  26#include "dir.h"
  27#include "blockcheck.h"
  28#include "dlmglue.h"
  29#include "extent_map.h"
  30#include "file.h"
  31#include "heartbeat.h"
  32#include "inode.h"
  33#include "journal.h"
  34#include "namei.h"
  35#include "suballoc.h"
  36#include "super.h"
  37#include "symlink.h"
  38#include "sysfile.h"
  39#include "uptodate.h"
  40#include "xattr.h"
  41#include "refcounttree.h"
  42#include "ocfs2_trace.h"
  43#include "filecheck.h"
  44
  45#include "buffer_head_io.h"
  46
  47struct ocfs2_find_inode_args
  48{
  49	u64		fi_blkno;
  50	unsigned long	fi_ino;
  51	unsigned int	fi_flags;
  52	unsigned int	fi_sysfile_type;
  53};
  54
  55static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
  56
  57static int ocfs2_read_locked_inode(struct inode *inode,
  58				   struct ocfs2_find_inode_args *args);
  59static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
  60static int ocfs2_find_actor(struct inode *inode, void *opaque);
  61static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
  62				    struct inode *inode,
  63				    struct buffer_head *fe_bh);
  64
  65static int ocfs2_filecheck_read_inode_block_full(struct inode *inode,
  66						 struct buffer_head **bh,
  67						 int flags, int type);
  68static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
  69						struct buffer_head *bh);
  70static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
  71					      struct buffer_head *bh);
  72
  73void ocfs2_set_inode_flags(struct inode *inode)
  74{
  75	unsigned int flags = OCFS2_I(inode)->ip_attr;
  76
  77	inode->i_flags &= ~(S_IMMUTABLE |
  78		S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
  79
  80	if (flags & OCFS2_IMMUTABLE_FL)
  81		inode->i_flags |= S_IMMUTABLE;
  82
  83	if (flags & OCFS2_SYNC_FL)
  84		inode->i_flags |= S_SYNC;
  85	if (flags & OCFS2_APPEND_FL)
  86		inode->i_flags |= S_APPEND;
  87	if (flags & OCFS2_NOATIME_FL)
  88		inode->i_flags |= S_NOATIME;
  89	if (flags & OCFS2_DIRSYNC_FL)
  90		inode->i_flags |= S_DIRSYNC;
  91}
  92
  93/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
  94void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
  95{
  96	unsigned int flags = oi->vfs_inode.i_flags;
  97
  98	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
  99			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
 100	if (flags & S_SYNC)
 101		oi->ip_attr |= OCFS2_SYNC_FL;
 102	if (flags & S_APPEND)
 103		oi->ip_attr |= OCFS2_APPEND_FL;
 104	if (flags & S_IMMUTABLE)
 105		oi->ip_attr |= OCFS2_IMMUTABLE_FL;
 106	if (flags & S_NOATIME)
 107		oi->ip_attr |= OCFS2_NOATIME_FL;
 108	if (flags & S_DIRSYNC)
 109		oi->ip_attr |= OCFS2_DIRSYNC_FL;
 110}
 111
 112struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
 113{
 114	struct ocfs2_find_inode_args args;
 115
 116	args.fi_blkno = blkno;
 117	args.fi_flags = 0;
 118	args.fi_ino = ino_from_blkno(sb, blkno);
 119	args.fi_sysfile_type = 0;
 120
 121	return ilookup5(sb, blkno, ocfs2_find_actor, &args);
 122}
 123struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
 124			 int sysfile_type)
 125{
 126	int rc = -ESTALE;
 127	struct inode *inode = NULL;
 128	struct super_block *sb = osb->sb;
 129	struct ocfs2_find_inode_args args;
 130	journal_t *journal = OCFS2_SB(sb)->journal->j_journal;
 131
 132	trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
 133			       sysfile_type);
 134
 135	/* Ok. By now we've either got the offsets passed to us by the
 136	 * caller, or we just pulled them off the bh. Lets do some
 137	 * sanity checks to make sure they're OK. */
 138	if (blkno == 0) {
 139		inode = ERR_PTR(-EINVAL);
 140		mlog_errno(PTR_ERR(inode));
 141		goto bail;
 142	}
 143
 144	args.fi_blkno = blkno;
 145	args.fi_flags = flags;
 146	args.fi_ino = ino_from_blkno(sb, blkno);
 147	args.fi_sysfile_type = sysfile_type;
 148
 149	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
 150			     ocfs2_init_locked_inode, &args);
 151	/* inode was *not* in the inode cache. 2.6.x requires
 152	 * us to do our own read_inode call and unlock it
 153	 * afterwards. */
 154	if (inode == NULL) {
 155		inode = ERR_PTR(-ENOMEM);
 156		mlog_errno(PTR_ERR(inode));
 157		goto bail;
 158	}
 159	trace_ocfs2_iget5_locked(inode->i_state);
 160	if (inode->i_state & I_NEW) {
 161		rc = ocfs2_read_locked_inode(inode, &args);
 162		unlock_new_inode(inode);
 163	}
 164	if (is_bad_inode(inode)) {
 165		iput(inode);
 166		inode = ERR_PTR(rc);
 167		goto bail;
 168	}
 169
 170	/*
 171	 * Set transaction id's of transactions that have to be committed
 172	 * to finish f[data]sync. We set them to currently running transaction
 173	 * as we cannot be sure that the inode or some of its metadata isn't
 174	 * part of the transaction - the inode could have been reclaimed and
 175	 * now it is reread from disk.
 176	 */
 177	if (journal) {
 178		transaction_t *transaction;
 179		tid_t tid;
 180		struct ocfs2_inode_info *oi = OCFS2_I(inode);
 181
 182		read_lock(&journal->j_state_lock);
 183		if (journal->j_running_transaction)
 184			transaction = journal->j_running_transaction;
 185		else
 186			transaction = journal->j_committing_transaction;
 187		if (transaction)
 188			tid = transaction->t_tid;
 189		else
 190			tid = journal->j_commit_sequence;
 191		read_unlock(&journal->j_state_lock);
 192		oi->i_sync_tid = tid;
 193		oi->i_datasync_tid = tid;
 194	}
 195
 196bail:
 197	if (!IS_ERR(inode)) {
 198		trace_ocfs2_iget_end(inode, 
 199			(unsigned long long)OCFS2_I(inode)->ip_blkno);
 200	}
 201
 202	return inode;
 203}
 204
 205
 206/*
 207 * here's how inodes get read from disk:
 208 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
 209 * found? : return the in-memory inode
 210 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
 211 */
 212
 213static int ocfs2_find_actor(struct inode *inode, void *opaque)
 214{
 215	struct ocfs2_find_inode_args *args = NULL;
 216	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 217	int ret = 0;
 218
 219	args = opaque;
 220
 221	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
 222
 223	trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
 224
 225	if (oi->ip_blkno != args->fi_blkno)
 226		goto bail;
 227
 228	ret = 1;
 229bail:
 230	return ret;
 231}
 232
 233/*
 234 * initialize the new inode, but don't do anything that would cause
 235 * us to sleep.
 236 * return 0 on success, 1 on failure
 237 */
 238static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
 239{
 240	struct ocfs2_find_inode_args *args = opaque;
 241	static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
 242				     ocfs2_file_ip_alloc_sem_key;
 243
 244	inode->i_ino = args->fi_ino;
 245	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
 246	if (args->fi_sysfile_type != 0)
 247		lockdep_set_class(&inode->i_rwsem,
 248			&ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
 249	if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
 250	    args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
 251	    args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
 252	    args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
 253		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
 254				  &ocfs2_quota_ip_alloc_sem_key);
 255	else
 256		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
 257				  &ocfs2_file_ip_alloc_sem_key);
 258
 259	return 0;
 260}
 261
 262void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 263			  int create_ino)
 264{
 265	struct super_block *sb;
 266	struct ocfs2_super *osb;
 267	int use_plocks = 1;
 268
 269	sb = inode->i_sb;
 270	osb = OCFS2_SB(sb);
 271
 272	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
 273	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
 274		use_plocks = 0;
 275
 276	/*
 277	 * These have all been checked by ocfs2_read_inode_block() or set
 278	 * by ocfs2_mknod_locked(), so a failure is a code bug.
 279	 */
 280	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
 281						cannot create a superblock
 282						inode today.  change if
 283						that is needed. */
 284	BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
 285	BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
 286
 287
 288	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 289	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
 290	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
 291
 292	inode_set_iversion(inode, 1);
 293	inode->i_generation = le32_to_cpu(fe->i_generation);
 294	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 295	inode->i_mode = le16_to_cpu(fe->i_mode);
 296	i_uid_write(inode, le32_to_cpu(fe->i_uid));
 297	i_gid_write(inode, le32_to_cpu(fe->i_gid));
 298
 299	/* Fast symlinks will have i_size but no allocated clusters. */
 300	if (S_ISLNK(inode->i_mode) && !fe->i_clusters) {
 301		inode->i_blocks = 0;
 302		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
 303	} else {
 304		inode->i_blocks = ocfs2_inode_sector_count(inode);
 305		inode->i_mapping->a_ops = &ocfs2_aops;
 306	}
 307	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
 308	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
 309	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
 310	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
 311	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
 312	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
 313
 314	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
 315		mlog(ML_ERROR,
 316		     "ip_blkno %llu != i_blkno %llu!\n",
 317		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
 318		     (unsigned long long)le64_to_cpu(fe->i_blkno));
 319
 320	set_nlink(inode, ocfs2_read_links_count(fe));
 321
 322	trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
 323				   le32_to_cpu(fe->i_flags));
 324	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
 325		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
 326		inode->i_flags |= S_NOQUOTA;
 327	}
 328  
 329	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
 330		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
 331	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
 332		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
 333	} else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
 334		inode->i_flags |= S_NOQUOTA;
 335	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
 336		/* we can't actually hit this as read_inode can't
 337		 * handle superblocks today ;-) */
 338		BUG();
 339	}
 340
 341	switch (inode->i_mode & S_IFMT) {
 342	    case S_IFREG:
 343		    if (use_plocks)
 344			    inode->i_fop = &ocfs2_fops;
 345		    else
 346			    inode->i_fop = &ocfs2_fops_no_plocks;
 347		    inode->i_op = &ocfs2_file_iops;
 348		    i_size_write(inode, le64_to_cpu(fe->i_size));
 349		    break;
 350	    case S_IFDIR:
 351		    inode->i_op = &ocfs2_dir_iops;
 352		    if (use_plocks)
 353			    inode->i_fop = &ocfs2_dops;
 354		    else
 355			    inode->i_fop = &ocfs2_dops_no_plocks;
 356		    i_size_write(inode, le64_to_cpu(fe->i_size));
 357		    OCFS2_I(inode)->ip_dir_lock_gen = 1;
 358		    break;
 359	    case S_IFLNK:
 360		    inode->i_op = &ocfs2_symlink_inode_operations;
 361		    inode_nohighmem(inode);
 362		    i_size_write(inode, le64_to_cpu(fe->i_size));
 363		    break;
 364	    default:
 365		    inode->i_op = &ocfs2_special_file_iops;
 366		    init_special_inode(inode, inode->i_mode,
 367				       inode->i_rdev);
 368		    break;
 369	}
 370
 371	if (create_ino) {
 372		inode->i_ino = ino_from_blkno(inode->i_sb,
 373			       le64_to_cpu(fe->i_blkno));
 374
 375		/*
 376		 * If we ever want to create system files from kernel,
 377		 * the generation argument to
 378		 * ocfs2_inode_lock_res_init() will have to change.
 379		 */
 380		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
 381
 382		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
 383					  OCFS2_LOCK_TYPE_META, 0, inode);
 384
 385		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
 386					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
 387	}
 388
 389	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
 390				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
 391				  inode);
 392
 393	ocfs2_set_inode_flags(inode);
 394
 395	OCFS2_I(inode)->ip_last_used_slot = 0;
 396	OCFS2_I(inode)->ip_last_used_group = 0;
 397
 398	if (S_ISDIR(inode->i_mode))
 399		ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
 400				    OCFS2_RESV_FLAG_DIR);
 401}
 402
 403static int ocfs2_read_locked_inode(struct inode *inode,
 404				   struct ocfs2_find_inode_args *args)
 405{
 406	struct super_block *sb;
 407	struct ocfs2_super *osb;
 408	struct ocfs2_dinode *fe;
 409	struct buffer_head *bh = NULL;
 410	int status, can_lock, lock_level = 0;
 411	u32 generation = 0;
 412
 413	status = -EINVAL;
 414	sb = inode->i_sb;
 415	osb = OCFS2_SB(sb);
 416
 417	/*
 418	 * To improve performance of cold-cache inode stats, we take
 419	 * the cluster lock here if possible.
 420	 *
 421	 * Generally, OCFS2 never trusts the contents of an inode
 422	 * unless it's holding a cluster lock, so taking it here isn't
 423	 * a correctness issue as much as it is a performance
 424	 * improvement.
 425	 *
 426	 * There are three times when taking the lock is not a good idea:
 427	 *
 428	 * 1) During startup, before we have initialized the DLM.
 429	 *
 430	 * 2) If we are reading certain system files which never get
 431	 *    cluster locks (local alloc, truncate log).
 432	 *
 433	 * 3) If the process doing the iget() is responsible for
 434	 *    orphan dir recovery. We're holding the orphan dir lock and
 435	 *    can get into a deadlock with another process on another
 436	 *    node in ->delete_inode().
 437	 *
 438	 * #1 and #2 can be simply solved by never taking the lock
 439	 * here for system files (which are the only type we read
 440	 * during mount). It's a heavier approach, but our main
 441	 * concern is user-accessible files anyway.
 442	 *
 443	 * #3 works itself out because we'll eventually take the
 444	 * cluster lock before trusting anything anyway.
 445	 */
 446	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
 447		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
 448		&& !ocfs2_mount_local(osb);
 449
 450	trace_ocfs2_read_locked_inode(
 451		(unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
 452
 453	/*
 454	 * To maintain backwards compatibility with older versions of
 455	 * ocfs2-tools, we still store the generation value for system
 456	 * files. The only ones that actually matter to userspace are
 457	 * the journals, but it's easier and inexpensive to just flag
 458	 * all system files similarly.
 459	 */
 460	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
 461		generation = osb->fs_generation;
 462
 463	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
 464				  OCFS2_LOCK_TYPE_META,
 465				  generation, inode);
 466
 467	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
 468				  OCFS2_LOCK_TYPE_OPEN,
 469				  0, inode);
 470
 471	if (can_lock) {
 472		status = ocfs2_open_lock(inode);
 473		if (status) {
 474			make_bad_inode(inode);
 475			mlog_errno(status);
 476			return status;
 477		}
 478		status = ocfs2_inode_lock(inode, NULL, lock_level);
 479		if (status) {
 480			make_bad_inode(inode);
 481			mlog_errno(status);
 482			return status;
 483		}
 484	}
 485
 486	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
 487		status = ocfs2_try_open_lock(inode, 0);
 488		if (status) {
 489			make_bad_inode(inode);
 490			return status;
 491		}
 492	}
 493
 494	if (can_lock) {
 495		if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
 496			status = ocfs2_filecheck_read_inode_block_full(inode,
 497						&bh, OCFS2_BH_IGNORE_CACHE, 0);
 498		else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
 499			status = ocfs2_filecheck_read_inode_block_full(inode,
 500						&bh, OCFS2_BH_IGNORE_CACHE, 1);
 501		else
 502			status = ocfs2_read_inode_block_full(inode,
 503						&bh, OCFS2_BH_IGNORE_CACHE);
 504	} else {
 505		status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
 506		/*
 507		 * If buffer is in jbd, then its checksum may not have been
 508		 * computed as yet.
 509		 */
 510		if (!status && !buffer_jbd(bh)) {
 511			if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
 512				status = ocfs2_filecheck_validate_inode_block(
 513								osb->sb, bh);
 514			else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
 515				status = ocfs2_filecheck_repair_inode_block(
 516								osb->sb, bh);
 517			else
 518				status = ocfs2_validate_inode_block(
 519								osb->sb, bh);
 520		}
 521	}
 522	if (status < 0) {
 523		mlog_errno(status);
 524		goto bail;
 525	}
 526
 527	status = -EINVAL;
 528	fe = (struct ocfs2_dinode *) bh->b_data;
 529
 530	/*
 531	 * This is a code bug. Right now the caller needs to
 532	 * understand whether it is asking for a system file inode or
 533	 * not so the proper lock names can be built.
 534	 */
 535	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
 536			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
 537			"Inode %llu: system file state is ambiguous\n",
 538			(unsigned long long)args->fi_blkno);
 539
 540	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
 541	    S_ISBLK(le16_to_cpu(fe->i_mode)))
 542		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 543
 544	ocfs2_populate_inode(inode, fe, 0);
 545
 546	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
 547
 548	if (buffer_dirty(bh) && !buffer_jbd(bh)) {
 549		if (can_lock) {
 550			ocfs2_inode_unlock(inode, lock_level);
 551			lock_level = 1;
 552			ocfs2_inode_lock(inode, NULL, lock_level);
 553		}
 554		status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
 555		if (status < 0) {
 556			mlog_errno(status);
 557			goto bail;
 558		}
 559	}
 560
 561	status = 0;
 562
 563bail:
 564	if (can_lock)
 565		ocfs2_inode_unlock(inode, lock_level);
 566
 567	if (status < 0)
 568		make_bad_inode(inode);
 569
 570	brelse(bh);
 571
 572	return status;
 573}
 574
 575void ocfs2_sync_blockdev(struct super_block *sb)
 576{
 577	sync_blockdev(sb->s_bdev);
 578}
 579
 580static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
 581				     struct inode *inode,
 582				     struct buffer_head *fe_bh)
 583{
 584	int status = 0;
 585	struct ocfs2_dinode *fe;
 586	handle_t *handle = NULL;
 587
 588	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 589
 590	/*
 591	 * This check will also skip truncate of inodes with inline
 592	 * data and fast symlinks.
 593	 */
 594	if (fe->i_clusters) {
 595		if (ocfs2_should_order_data(inode))
 596			ocfs2_begin_ordered_truncate(inode, 0);
 597
 598		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
 599		if (IS_ERR(handle)) {
 600			status = PTR_ERR(handle);
 601			handle = NULL;
 602			mlog_errno(status);
 603			goto out;
 604		}
 605
 606		status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 607						 fe_bh,
 608						 OCFS2_JOURNAL_ACCESS_WRITE);
 609		if (status < 0) {
 610			mlog_errno(status);
 611			goto out;
 612		}
 613
 614		i_size_write(inode, 0);
 615
 616		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 617		if (status < 0) {
 618			mlog_errno(status);
 619			goto out;
 620		}
 621
 622		ocfs2_commit_trans(osb, handle);
 623		handle = NULL;
 624
 625		status = ocfs2_commit_truncate(osb, inode, fe_bh);
 626		if (status < 0)
 627			mlog_errno(status);
 628	}
 629
 630out:
 631	if (handle)
 632		ocfs2_commit_trans(osb, handle);
 633	return status;
 634}
 635
 636static int ocfs2_remove_inode(struct inode *inode,
 637			      struct buffer_head *di_bh,
 638			      struct inode *orphan_dir_inode,
 639			      struct buffer_head *orphan_dir_bh)
 640{
 641	int status;
 642	struct inode *inode_alloc_inode = NULL;
 643	struct buffer_head *inode_alloc_bh = NULL;
 644	handle_t *handle;
 645	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 646	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
 647
 648	inode_alloc_inode =
 649		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
 650					    le16_to_cpu(di->i_suballoc_slot));
 651	if (!inode_alloc_inode) {
 652		status = -ENOENT;
 653		mlog_errno(status);
 654		goto bail;
 655	}
 656
 657	inode_lock(inode_alloc_inode);
 658	status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
 659	if (status < 0) {
 660		inode_unlock(inode_alloc_inode);
 661
 662		mlog_errno(status);
 663		goto bail;
 664	}
 665
 666	handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
 667				   ocfs2_quota_trans_credits(inode->i_sb));
 668	if (IS_ERR(handle)) {
 669		status = PTR_ERR(handle);
 670		mlog_errno(status);
 671		goto bail_unlock;
 672	}
 673
 674	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
 675		status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
 676					  orphan_dir_bh, false);
 677		if (status < 0) {
 678			mlog_errno(status);
 679			goto bail_commit;
 680		}
 681	}
 682
 683	/* set the inodes dtime */
 684	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 685					 OCFS2_JOURNAL_ACCESS_WRITE);
 686	if (status < 0) {
 687		mlog_errno(status);
 688		goto bail_commit;
 689	}
 690
 691	di->i_dtime = cpu_to_le64(ktime_get_real_seconds());
 692	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
 693	ocfs2_journal_dirty(handle, di_bh);
 694
 695	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
 696	dquot_free_inode(inode);
 697
 698	status = ocfs2_free_dinode(handle, inode_alloc_inode,
 699				   inode_alloc_bh, di);
 700	if (status < 0)
 701		mlog_errno(status);
 702
 703bail_commit:
 704	ocfs2_commit_trans(osb, handle);
 705bail_unlock:
 706	ocfs2_inode_unlock(inode_alloc_inode, 1);
 707	inode_unlock(inode_alloc_inode);
 708	brelse(inode_alloc_bh);
 709bail:
 710	iput(inode_alloc_inode);
 711
 712	return status;
 713}
 714
 715/*
 716 * Serialize with orphan dir recovery. If the process doing
 717 * recovery on this orphan dir does an iget() with the dir
 718 * i_mutex held, we'll deadlock here. Instead we detect this
 719 * and exit early - recovery will wipe this inode for us.
 720 */
 721static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
 722					     int slot)
 723{
 724	int ret = 0;
 725
 726	spin_lock(&osb->osb_lock);
 727	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
 728		ret = -EDEADLK;
 729		goto out;
 730	}
 731	/* This signals to the orphan recovery process that it should
 732	 * wait for us to handle the wipe. */
 733	osb->osb_orphan_wipes[slot]++;
 734out:
 735	spin_unlock(&osb->osb_lock);
 736	trace_ocfs2_check_orphan_recovery_state(slot, ret);
 737	return ret;
 738}
 739
 740static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
 741					 int slot)
 742{
 743	spin_lock(&osb->osb_lock);
 744	osb->osb_orphan_wipes[slot]--;
 745	spin_unlock(&osb->osb_lock);
 746
 747	wake_up(&osb->osb_wipe_event);
 748}
 749
 750static int ocfs2_wipe_inode(struct inode *inode,
 751			    struct buffer_head *di_bh)
 752{
 753	int status, orphaned_slot = -1;
 754	struct inode *orphan_dir_inode = NULL;
 755	struct buffer_head *orphan_dir_bh = NULL;
 756	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 757	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
 758
 759	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
 760		orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
 761
 762		status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
 763		if (status)
 764			return status;
 765
 766		orphan_dir_inode = ocfs2_get_system_file_inode(osb,
 767							       ORPHAN_DIR_SYSTEM_INODE,
 768							       orphaned_slot);
 769		if (!orphan_dir_inode) {
 770			status = -ENOENT;
 771			mlog_errno(status);
 772			goto bail;
 773		}
 774
 775		/* Lock the orphan dir. The lock will be held for the entire
 776		 * delete_inode operation. We do this now to avoid races with
 777		 * recovery completion on other nodes. */
 778		inode_lock(orphan_dir_inode);
 779		status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
 780		if (status < 0) {
 781			inode_unlock(orphan_dir_inode);
 782
 783			mlog_errno(status);
 784			goto bail;
 785		}
 786	}
 787
 788	/* we do this while holding the orphan dir lock because we
 789	 * don't want recovery being run from another node to try an
 790	 * inode delete underneath us -- this will result in two nodes
 791	 * truncating the same file! */
 792	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
 793	if (status < 0) {
 794		mlog_errno(status);
 795		goto bail_unlock_dir;
 796	}
 797
 798	/* Remove any dir index tree */
 799	if (S_ISDIR(inode->i_mode)) {
 800		status = ocfs2_dx_dir_truncate(inode, di_bh);
 801		if (status) {
 802			mlog_errno(status);
 803			goto bail_unlock_dir;
 804		}
 805	}
 806
 807	/*Free extended attribute resources associated with this inode.*/
 808	status = ocfs2_xattr_remove(inode, di_bh);
 809	if (status < 0) {
 810		mlog_errno(status);
 811		goto bail_unlock_dir;
 812	}
 813
 814	status = ocfs2_remove_refcount_tree(inode, di_bh);
 815	if (status < 0) {
 816		mlog_errno(status);
 817		goto bail_unlock_dir;
 818	}
 819
 820	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
 821				    orphan_dir_bh);
 822	if (status < 0)
 823		mlog_errno(status);
 824
 825bail_unlock_dir:
 826	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
 827		return status;
 828
 829	ocfs2_inode_unlock(orphan_dir_inode, 1);
 830	inode_unlock(orphan_dir_inode);
 831	brelse(orphan_dir_bh);
 832bail:
 833	iput(orphan_dir_inode);
 834	ocfs2_signal_wipe_completion(osb, orphaned_slot);
 835
 836	return status;
 837}
 838
 839/* There is a series of simple checks that should be done before a
 840 * trylock is even considered. Encapsulate those in this function. */
 841static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
 842{
 843	int ret = 0;
 844	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 845	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 846
 847	trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
 848					     (unsigned long long)oi->ip_blkno,
 849					     oi->ip_flags);
 850
 851	/* We shouldn't be getting here for the root directory
 852	 * inode.. */
 853	if (inode == osb->root_inode) {
 854		mlog(ML_ERROR, "Skipping delete of root inode.\n");
 855		goto bail;
 856	}
 857
 858	/*
 859	 * If we're coming from downconvert_thread we can't go into our own
 860	 * voting [hello, deadlock city!] so we cannot delete the inode. But
 861	 * since we dropped last inode ref when downconverting dentry lock,
 862	 * we cannot have the file open and thus the node doing unlink will
 863	 * take care of deleting the inode.
 864	 */
 865	if (current == osb->dc_task)
 866		goto bail;
 867
 868	spin_lock(&oi->ip_lock);
 869	/* OCFS2 *never* deletes system files. This should technically
 870	 * never get here as system file inodes should always have a
 871	 * positive link count. */
 872	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
 873		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
 874		     (unsigned long long)oi->ip_blkno);
 875		goto bail_unlock;
 876	}
 877
 878	ret = 1;
 879bail_unlock:
 880	spin_unlock(&oi->ip_lock);
 881bail:
 882	return ret;
 883}
 884
 885/* Query the cluster to determine whether we should wipe an inode from
 886 * disk or not.
 887 *
 888 * Requires the inode to have the cluster lock. */
 889static int ocfs2_query_inode_wipe(struct inode *inode,
 890				  struct buffer_head *di_bh,
 891				  int *wipe)
 892{
 893	int status = 0, reason = 0;
 894	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 895	struct ocfs2_dinode *di;
 896
 897	*wipe = 0;
 898
 899	trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
 900					   inode->i_nlink);
 901
 902	/* While we were waiting for the cluster lock in
 903	 * ocfs2_delete_inode, another node might have asked to delete
 904	 * the inode. Recheck our flags to catch this. */
 905	if (!ocfs2_inode_is_valid_to_delete(inode)) {
 906		reason = 1;
 907		goto bail;
 908	}
 909
 910	/* Now that we have an up to date inode, we can double check
 911	 * the link count. */
 912	if (inode->i_nlink)
 913		goto bail;
 914
 915	/* Do some basic inode verification... */
 916	di = (struct ocfs2_dinode *) di_bh->b_data;
 917	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
 918	    !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
 919		/*
 920		 * Inodes in the orphan dir must have ORPHANED_FL.  The only
 921		 * inodes that come back out of the orphan dir are reflink
 922		 * targets. A reflink target may be moved out of the orphan
 923		 * dir between the time we scan the directory and the time we
 924		 * process it. This would lead to HAS_REFCOUNT_FL being set but
 925		 * ORPHANED_FL not.
 926		 */
 927		if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
 928			reason = 2;
 929			goto bail;
 930		}
 931
 932		/* for lack of a better error? */
 933		status = -EEXIST;
 934		mlog(ML_ERROR,
 935		     "Inode %llu (on-disk %llu) not orphaned! "
 936		     "Disk flags  0x%x, inode flags 0x%x\n",
 937		     (unsigned long long)oi->ip_blkno,
 938		     (unsigned long long)le64_to_cpu(di->i_blkno),
 939		     le32_to_cpu(di->i_flags), oi->ip_flags);
 940		goto bail;
 941	}
 942
 943	/* has someone already deleted us?! baaad... */
 944	if (di->i_dtime) {
 945		status = -EEXIST;
 946		mlog_errno(status);
 947		goto bail;
 948	}
 949
 950	/*
 951	 * This is how ocfs2 determines whether an inode is still live
 952	 * within the cluster. Every node takes a shared read lock on
 953	 * the inode open lock in ocfs2_read_locked_inode(). When we
 954	 * get to ->delete_inode(), each node tries to convert it's
 955	 * lock to an exclusive. Trylocks are serialized by the inode
 956	 * meta data lock. If the upconvert succeeds, we know the inode
 957	 * is no longer live and can be deleted.
 958	 *
 959	 * Though we call this with the meta data lock held, the
 960	 * trylock keeps us from ABBA deadlock.
 961	 */
 962	status = ocfs2_try_open_lock(inode, 1);
 963	if (status == -EAGAIN) {
 964		status = 0;
 965		reason = 3;
 966		goto bail;
 967	}
 968	if (status < 0) {
 969		mlog_errno(status);
 970		goto bail;
 971	}
 972
 973	*wipe = 1;
 974	trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
 975
 976bail:
 977	trace_ocfs2_query_inode_wipe_end(status, reason);
 978	return status;
 979}
 980
 981/* Support function for ocfs2_delete_inode. Will help us keep the
 982 * inode data in a consistent state for clear_inode. Always truncates
 983 * pages, optionally sync's them first. */
 984static void ocfs2_cleanup_delete_inode(struct inode *inode,
 985				       int sync_data)
 986{
 987	trace_ocfs2_cleanup_delete_inode(
 988		(unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
 989	if (sync_data)
 990		filemap_write_and_wait(inode->i_mapping);
 991	truncate_inode_pages_final(&inode->i_data);
 992}
 993
 994static void ocfs2_delete_inode(struct inode *inode)
 995{
 996	int wipe, status;
 997	sigset_t oldset;
 998	struct buffer_head *di_bh = NULL;
 999	struct ocfs2_dinode *di = NULL;
1000
1001	trace_ocfs2_delete_inode(inode->i_ino,
1002				 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1003				 is_bad_inode(inode));
1004
1005	/* When we fail in read_inode() we mark inode as bad. The second test
1006	 * catches the case when inode allocation fails before allocating
1007	 * a block for inode. */
1008	if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
1009		goto bail;
1010
1011	if (!ocfs2_inode_is_valid_to_delete(inode)) {
1012		/* It's probably not necessary to truncate_inode_pages
1013		 * here but we do it for safety anyway (it will most
1014		 * likely be a no-op anyway) */
1015		ocfs2_cleanup_delete_inode(inode, 0);
1016		goto bail;
1017	}
1018
1019	dquot_initialize(inode);
1020
1021	/* We want to block signals in delete_inode as the lock and
1022	 * messaging paths may return us -ERESTARTSYS. Which would
1023	 * cause us to exit early, resulting in inodes being orphaned
1024	 * forever. */
1025	ocfs2_block_signals(&oldset);
1026
1027	/*
1028	 * Synchronize us against ocfs2_get_dentry. We take this in
1029	 * shared mode so that all nodes can still concurrently
1030	 * process deletes.
1031	 */
1032	status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1033	if (status < 0) {
1034		mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1035		ocfs2_cleanup_delete_inode(inode, 0);
1036		goto bail_unblock;
1037	}
1038	/* Lock down the inode. This gives us an up to date view of
1039	 * it's metadata (for verification), and allows us to
1040	 * serialize delete_inode on multiple nodes.
1041	 *
1042	 * Even though we might be doing a truncate, we don't take the
1043	 * allocation lock here as it won't be needed - nobody will
1044	 * have the file open.
1045	 */
1046	status = ocfs2_inode_lock(inode, &di_bh, 1);
1047	if (status < 0) {
1048		if (status != -ENOENT)
1049			mlog_errno(status);
1050		ocfs2_cleanup_delete_inode(inode, 0);
1051		goto bail_unlock_nfs_sync;
1052	}
1053
1054	di = (struct ocfs2_dinode *)di_bh->b_data;
1055	/* Skip inode deletion and wait for dio orphan entry recovered
1056	 * first */
1057	if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
1058		ocfs2_cleanup_delete_inode(inode, 0);
1059		goto bail_unlock_inode;
1060	}
1061
1062	/* Query the cluster. This will be the final decision made
1063	 * before we go ahead and wipe the inode. */
1064	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1065	if (!wipe || status < 0) {
1066		/* Error and remote inode busy both mean we won't be
1067		 * removing the inode, so they take almost the same
1068		 * path. */
1069		if (status < 0)
1070			mlog_errno(status);
1071
1072		/* Someone in the cluster has disallowed a wipe of
1073		 * this inode, or it was never completely
1074		 * orphaned. Write out the pages and exit now. */
1075		ocfs2_cleanup_delete_inode(inode, 1);
1076		goto bail_unlock_inode;
1077	}
1078
1079	ocfs2_cleanup_delete_inode(inode, 0);
1080
1081	status = ocfs2_wipe_inode(inode, di_bh);
1082	if (status < 0) {
1083		if (status != -EDEADLK)
1084			mlog_errno(status);
1085		goto bail_unlock_inode;
1086	}
1087
1088	/*
1089	 * Mark the inode as successfully deleted.
1090	 *
1091	 * This is important for ocfs2_clear_inode() as it will check
1092	 * this flag and skip any checkpointing work
1093	 *
1094	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1095	 * the LVB for other nodes.
1096	 */
1097	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1098
1099bail_unlock_inode:
1100	ocfs2_inode_unlock(inode, 1);
1101	brelse(di_bh);
1102
1103bail_unlock_nfs_sync:
1104	ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1105
1106bail_unblock:
1107	ocfs2_unblock_signals(&oldset);
1108bail:
1109	return;
1110}
1111
1112static void ocfs2_clear_inode(struct inode *inode)
1113{
1114	int status;
1115	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1116	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1117
1118	clear_inode(inode);
1119	trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1120				inode->i_nlink);
1121
1122	mlog_bug_on_msg(osb == NULL,
1123			"Inode=%lu\n", inode->i_ino);
1124
1125	dquot_drop(inode);
1126
1127	/* To preven remote deletes we hold open lock before, now it
1128	 * is time to unlock PR and EX open locks. */
1129	ocfs2_open_unlock(inode);
1130
1131	/* Do these before all the other work so that we don't bounce
1132	 * the downconvert thread while waiting to destroy the locks. */
1133	ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres);
1134	ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres);
1135	ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres);
1136
1137	ocfs2_resv_discard(&osb->osb_la_resmap,
1138			   &oi->ip_la_data_resv);
1139	ocfs2_resv_init_once(&oi->ip_la_data_resv);
1140
1141	/* We very well may get a clear_inode before all an inodes
1142	 * metadata has hit disk. Of course, we can't drop any cluster
1143	 * locks until the journal has finished with it. The only
1144	 * exception here are successfully wiped inodes - their
1145	 * metadata can now be considered to be part of the system
1146	 * inodes from which it came. */
1147	if (!(oi->ip_flags & OCFS2_INODE_DELETED))
1148		ocfs2_checkpoint_inode(inode);
1149
1150	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1151			"Clear inode of %llu, inode has io markers\n",
1152			(unsigned long long)oi->ip_blkno);
1153	mlog_bug_on_msg(!list_empty(&oi->ip_unwritten_list),
1154			"Clear inode of %llu, inode has unwritten extents\n",
1155			(unsigned long long)oi->ip_blkno);
1156
1157	ocfs2_extent_map_trunc(inode, 0);
1158
1159	status = ocfs2_drop_inode_locks(inode);
1160	if (status < 0)
1161		mlog_errno(status);
1162
1163	ocfs2_lock_res_free(&oi->ip_rw_lockres);
1164	ocfs2_lock_res_free(&oi->ip_inode_lockres);
1165	ocfs2_lock_res_free(&oi->ip_open_lockres);
1166
1167	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1168
1169	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1170			"Clear inode of %llu, inode has %u cache items\n",
1171			(unsigned long long)oi->ip_blkno,
1172			INODE_CACHE(inode)->ci_num_cached);
1173
1174	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1175			"Clear inode of %llu, inode has a bad flag\n",
1176			(unsigned long long)oi->ip_blkno);
1177
1178	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1179			"Clear inode of %llu, inode is locked\n",
1180			(unsigned long long)oi->ip_blkno);
1181
1182	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1183			"Clear inode of %llu, io_mutex is locked\n",
1184			(unsigned long long)oi->ip_blkno);
1185	mutex_unlock(&oi->ip_io_mutex);
1186
1187	/*
1188	 * down_trylock() returns 0, down_write_trylock() returns 1
1189	 * kernel 1, world 0
1190	 */
1191	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1192			"Clear inode of %llu, alloc_sem is locked\n",
1193			(unsigned long long)oi->ip_blkno);
1194	up_write(&oi->ip_alloc_sem);
1195
1196	mlog_bug_on_msg(oi->ip_open_count,
1197			"Clear inode of %llu has open count %d\n",
1198			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
1199
1200	/* Clear all other flags. */
1201	oi->ip_flags = 0;
1202	oi->ip_dir_start_lookup = 0;
1203	oi->ip_blkno = 0ULL;
1204
1205	/*
1206	 * ip_jinode is used to track txns against this inode. We ensure that
1207	 * the journal is flushed before journal shutdown. Thus it is safe to
1208	 * have inodes get cleaned up after journal shutdown.
1209	 */
1210	jbd2_journal_release_jbd_inode(osb->journal->j_journal,
1211				       &oi->ip_jinode);
1212}
1213
1214void ocfs2_evict_inode(struct inode *inode)
1215{
1216	if (!inode->i_nlink ||
1217	    (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1218		ocfs2_delete_inode(inode);
1219	} else {
1220		truncate_inode_pages_final(&inode->i_data);
1221	}
1222	ocfs2_clear_inode(inode);
1223}
1224
1225/* Called under inode_lock, with no more references on the
1226 * struct inode, so it's safe here to check the flags field
1227 * and to manipulate i_nlink without any other locks. */
1228int ocfs2_drop_inode(struct inode *inode)
1229{
1230	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1231
1232	trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
1233				inode->i_nlink, oi->ip_flags);
1234
1235	assert_spin_locked(&inode->i_lock);
1236	inode->i_state |= I_WILL_FREE;
1237	spin_unlock(&inode->i_lock);
1238	write_inode_now(inode, 1);
1239	spin_lock(&inode->i_lock);
1240	WARN_ON(inode->i_state & I_NEW);
1241	inode->i_state &= ~I_WILL_FREE;
1242
1243	return 1;
1244}
1245
1246/*
1247 * This is called from our getattr.
1248 */
1249int ocfs2_inode_revalidate(struct dentry *dentry)
1250{
1251	struct inode *inode = d_inode(dentry);
1252	int status = 0;
1253
1254	trace_ocfs2_inode_revalidate(inode,
1255		inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1256		inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
1257
1258	if (!inode) {
1259		status = -ENOENT;
1260		goto bail;
1261	}
1262
1263	spin_lock(&OCFS2_I(inode)->ip_lock);
1264	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1265		spin_unlock(&OCFS2_I(inode)->ip_lock);
1266		status = -ENOENT;
1267		goto bail;
1268	}
1269	spin_unlock(&OCFS2_I(inode)->ip_lock);
1270
1271	/* Let ocfs2_inode_lock do the work of updating our struct
1272	 * inode for us. */
1273	status = ocfs2_inode_lock(inode, NULL, 0);
1274	if (status < 0) {
1275		if (status != -ENOENT)
1276			mlog_errno(status);
1277		goto bail;
1278	}
1279	ocfs2_inode_unlock(inode, 0);
1280bail:
1281	return status;
1282}
1283
1284/*
1285 * Updates a disk inode from a
1286 * struct inode.
1287 * Only takes ip_lock.
1288 */
1289int ocfs2_mark_inode_dirty(handle_t *handle,
1290			   struct inode *inode,
1291			   struct buffer_head *bh)
1292{
1293	int status;
1294	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1295
1296	trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
1297
1298	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1299					 OCFS2_JOURNAL_ACCESS_WRITE);
1300	if (status < 0) {
1301		mlog_errno(status);
1302		goto leave;
1303	}
1304
1305	spin_lock(&OCFS2_I(inode)->ip_lock);
1306	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1307	ocfs2_get_inode_flags(OCFS2_I(inode));
1308	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1309	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1310	spin_unlock(&OCFS2_I(inode)->ip_lock);
1311
1312	fe->i_size = cpu_to_le64(i_size_read(inode));
1313	ocfs2_set_links_count(fe, inode->i_nlink);
1314	fe->i_uid = cpu_to_le32(i_uid_read(inode));
1315	fe->i_gid = cpu_to_le32(i_gid_read(inode));
1316	fe->i_mode = cpu_to_le16(inode->i_mode);
1317	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1318	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1319	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1320	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1321	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1322	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1323
1324	ocfs2_journal_dirty(handle, bh);
1325	ocfs2_update_inode_fsync_trans(handle, inode, 1);
1326leave:
1327	return status;
1328}
1329
1330/*
1331 *
1332 * Updates a struct inode from a disk inode.
1333 * does no i/o, only takes ip_lock.
1334 */
1335void ocfs2_refresh_inode(struct inode *inode,
1336			 struct ocfs2_dinode *fe)
1337{
1338	spin_lock(&OCFS2_I(inode)->ip_lock);
1339
1340	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1341	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1342	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1343	ocfs2_set_inode_flags(inode);
1344	i_size_write(inode, le64_to_cpu(fe->i_size));
1345	set_nlink(inode, ocfs2_read_links_count(fe));
1346	i_uid_write(inode, le32_to_cpu(fe->i_uid));
1347	i_gid_write(inode, le32_to_cpu(fe->i_gid));
1348	inode->i_mode = le16_to_cpu(fe->i_mode);
1349	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1350		inode->i_blocks = 0;
1351	else
1352		inode->i_blocks = ocfs2_inode_sector_count(inode);
1353	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1354	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1355	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1356	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1357	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1358	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1359
1360	spin_unlock(&OCFS2_I(inode)->ip_lock);
1361}
1362
1363int ocfs2_validate_inode_block(struct super_block *sb,
1364			       struct buffer_head *bh)
1365{
1366	int rc;
1367	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1368
1369	trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
1370
1371	BUG_ON(!buffer_uptodate(bh));
1372
1373	/*
1374	 * If the ecc fails, we return the error but otherwise
1375	 * leave the filesystem running.  We know any error is
1376	 * local to this block.
1377	 */
1378	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1379	if (rc) {
1380		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1381		     (unsigned long long)bh->b_blocknr);
1382		goto bail;
1383	}
1384
1385	/*
1386	 * Errors after here are fatal.
1387	 */
1388
1389	rc = -EINVAL;
1390
1391	if (!OCFS2_IS_VALID_DINODE(di)) {
1392		rc = ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1393				 (unsigned long long)bh->b_blocknr, 7,
1394				 di->i_signature);
1395		goto bail;
1396	}
1397
1398	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1399		rc = ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1400				 (unsigned long long)bh->b_blocknr,
1401				 (unsigned long long)le64_to_cpu(di->i_blkno));
1402		goto bail;
1403	}
1404
1405	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1406		rc = ocfs2_error(sb,
1407				 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1408				 (unsigned long long)bh->b_blocknr);
1409		goto bail;
1410	}
1411
1412	if (le32_to_cpu(di->i_fs_generation) !=
1413	    OCFS2_SB(sb)->fs_generation) {
1414		rc = ocfs2_error(sb,
1415				 "Invalid dinode #%llu: fs_generation is %u\n",
1416				 (unsigned long long)bh->b_blocknr,
1417				 le32_to_cpu(di->i_fs_generation));
1418		goto bail;
1419	}
1420
1421	rc = 0;
1422
1423bail:
1424	return rc;
1425}
1426
1427static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
1428						struct buffer_head *bh)
1429{
1430	int rc = 0;
1431	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1432
1433	trace_ocfs2_filecheck_validate_inode_block(
1434		(unsigned long long)bh->b_blocknr);
1435
1436	BUG_ON(!buffer_uptodate(bh));
1437
1438	/*
1439	 * Call ocfs2_validate_meta_ecc() first since it has ecc repair
1440	 * function, but we should not return error immediately when ecc
1441	 * validation fails, because the reason is quite likely the invalid
1442	 * inode number inputed.
1443	 */
1444	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1445	if (rc) {
1446		mlog(ML_ERROR,
1447		     "Filecheck: checksum failed for dinode %llu\n",
1448		     (unsigned long long)bh->b_blocknr);
1449		rc = -OCFS2_FILECHECK_ERR_BLOCKECC;
1450	}
1451
1452	if (!OCFS2_IS_VALID_DINODE(di)) {
1453		mlog(ML_ERROR,
1454		     "Filecheck: invalid dinode #%llu: signature = %.*s\n",
1455		     (unsigned long long)bh->b_blocknr, 7, di->i_signature);
1456		rc = -OCFS2_FILECHECK_ERR_INVALIDINO;
1457		goto bail;
1458	} else if (rc)
1459		goto bail;
1460
1461	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1462		mlog(ML_ERROR,
1463		     "Filecheck: invalid dinode #%llu: i_blkno is %llu\n",
1464		     (unsigned long long)bh->b_blocknr,
1465		     (unsigned long long)le64_to_cpu(di->i_blkno));
1466		rc = -OCFS2_FILECHECK_ERR_BLOCKNO;
1467		goto bail;
1468	}
1469
1470	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1471		mlog(ML_ERROR,
1472		     "Filecheck: invalid dinode #%llu: OCFS2_VALID_FL "
1473		     "not set\n",
1474		     (unsigned long long)bh->b_blocknr);
1475		rc = -OCFS2_FILECHECK_ERR_VALIDFLAG;
1476		goto bail;
1477	}
1478
1479	if (le32_to_cpu(di->i_fs_generation) !=
1480	    OCFS2_SB(sb)->fs_generation) {
1481		mlog(ML_ERROR,
1482		     "Filecheck: invalid dinode #%llu: fs_generation is %u\n",
1483		     (unsigned long long)bh->b_blocknr,
1484		     le32_to_cpu(di->i_fs_generation));
1485		rc = -OCFS2_FILECHECK_ERR_GENERATION;
1486	}
1487
1488bail:
1489	return rc;
1490}
1491
1492static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
1493					      struct buffer_head *bh)
1494{
1495	int changed = 0;
1496	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1497
1498	if (!ocfs2_filecheck_validate_inode_block(sb, bh))
1499		return 0;
1500
1501	trace_ocfs2_filecheck_repair_inode_block(
1502		(unsigned long long)bh->b_blocknr);
1503
1504	if (ocfs2_is_hard_readonly(OCFS2_SB(sb)) ||
1505	    ocfs2_is_soft_readonly(OCFS2_SB(sb))) {
1506		mlog(ML_ERROR,
1507		     "Filecheck: cannot repair dinode #%llu "
1508		     "on readonly filesystem\n",
1509		     (unsigned long long)bh->b_blocknr);
1510		return -OCFS2_FILECHECK_ERR_READONLY;
1511	}
1512
1513	if (buffer_jbd(bh)) {
1514		mlog(ML_ERROR,
1515		     "Filecheck: cannot repair dinode #%llu, "
1516		     "its buffer is in jbd\n",
1517		     (unsigned long long)bh->b_blocknr);
1518		return -OCFS2_FILECHECK_ERR_INJBD;
1519	}
1520
1521	if (!OCFS2_IS_VALID_DINODE(di)) {
1522		/* Cannot fix invalid inode block */
1523		return -OCFS2_FILECHECK_ERR_INVALIDINO;
1524	}
1525
1526	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1527		/* Cannot just add VALID_FL flag back as a fix,
1528		 * need more things to check here.
1529		 */
1530		return -OCFS2_FILECHECK_ERR_VALIDFLAG;
1531	}
1532
1533	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1534		di->i_blkno = cpu_to_le64(bh->b_blocknr);
1535		changed = 1;
1536		mlog(ML_ERROR,
1537		     "Filecheck: reset dinode #%llu: i_blkno to %llu\n",
1538		     (unsigned long long)bh->b_blocknr,
1539		     (unsigned long long)le64_to_cpu(di->i_blkno));
1540	}
1541
1542	if (le32_to_cpu(di->i_fs_generation) !=
1543	    OCFS2_SB(sb)->fs_generation) {
1544		di->i_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1545		changed = 1;
1546		mlog(ML_ERROR,
1547		     "Filecheck: reset dinode #%llu: fs_generation to %u\n",
1548		     (unsigned long long)bh->b_blocknr,
1549		     le32_to_cpu(di->i_fs_generation));
1550	}
1551
1552	if (changed || ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check)) {
1553		ocfs2_compute_meta_ecc(sb, bh->b_data, &di->i_check);
1554		mark_buffer_dirty(bh);
1555		mlog(ML_ERROR,
1556		     "Filecheck: reset dinode #%llu: compute meta ecc\n",
1557		     (unsigned long long)bh->b_blocknr);
1558	}
1559
1560	return 0;
1561}
1562
1563static int
1564ocfs2_filecheck_read_inode_block_full(struct inode *inode,
1565				      struct buffer_head **bh,
1566				      int flags, int type)
1567{
1568	int rc;
1569	struct buffer_head *tmp = *bh;
1570
1571	if (!type) /* Check inode block */
1572		rc = ocfs2_read_blocks(INODE_CACHE(inode),
1573				OCFS2_I(inode)->ip_blkno,
1574				1, &tmp, flags,
1575				ocfs2_filecheck_validate_inode_block);
1576	else /* Repair inode block */
1577		rc = ocfs2_read_blocks(INODE_CACHE(inode),
1578				OCFS2_I(inode)->ip_blkno,
1579				1, &tmp, flags,
1580				ocfs2_filecheck_repair_inode_block);
1581
1582	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1583	if (!rc && !*bh)
1584		*bh = tmp;
1585
1586	return rc;
1587}
1588
1589int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1590				int flags)
1591{
1592	int rc;
1593	struct buffer_head *tmp = *bh;
1594
1595	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1596			       1, &tmp, flags, ocfs2_validate_inode_block);
1597
1598	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1599	if (!rc && !*bh)
1600		*bh = tmp;
1601
1602	return rc;
1603}
1604
1605int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1606{
1607	return ocfs2_read_inode_block_full(inode, bh, 0);
1608}
1609
1610
1611static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1612{
1613	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1614
1615	return oi->ip_blkno;
1616}
1617
1618static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1619{
1620	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1621
1622	return oi->vfs_inode.i_sb;
1623}
1624
1625static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1626{
1627	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1628
1629	spin_lock(&oi->ip_lock);
1630}
1631
1632static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1633{
1634	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1635
1636	spin_unlock(&oi->ip_lock);
1637}
1638
1639static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1640{
1641	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1642
1643	mutex_lock(&oi->ip_io_mutex);
1644}
1645
1646static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1647{
1648	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1649
1650	mutex_unlock(&oi->ip_io_mutex);
1651}
1652
1653const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1654	.co_owner		= ocfs2_inode_cache_owner,
1655	.co_get_super		= ocfs2_inode_cache_get_super,
1656	.co_cache_lock		= ocfs2_inode_cache_lock,
1657	.co_cache_unlock	= ocfs2_inode_cache_unlock,
1658	.co_io_lock		= ocfs2_inode_cache_io_lock,
1659	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
1660};
1661