xref: /openbmc/linux/fs/ocfs2/inode.c (revision 95777591)
1  /* -*- mode: c; c-basic-offset: 8; -*-
2   * vim: noexpandtab sw=8 ts=8 sts=0:
3   *
4   * inode.c
5   *
6   * vfs' aops, fops, dops and iops
7   *
8   * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9   *
10   * This program is free software; you can redistribute it and/or
11   * modify it under the terms of the GNU General Public
12   * License as published by the Free Software Foundation; either
13   * version 2 of the License, or (at your option) any later version.
14   *
15   * This program is distributed in the hope that it will be useful,
16   * but WITHOUT ANY WARRANTY; without even the implied warranty of
17   * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18   * General Public License for more details.
19   *
20   * You should have received a copy of the GNU General Public
21   * License along with this program; if not, write to the
22   * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23   * Boston, MA 021110-1307, USA.
24   */
25  
26  #include <linux/fs.h>
27  #include <linux/types.h>
28  #include <linux/highmem.h>
29  #include <linux/pagemap.h>
30  #include <linux/quotaops.h>
31  #include <linux/iversion.h>
32  
33  #include <asm/byteorder.h>
34  
35  #include <cluster/masklog.h>
36  
37  #include "ocfs2.h"
38  
39  #include "alloc.h"
40  #include "dir.h"
41  #include "blockcheck.h"
42  #include "dlmglue.h"
43  #include "extent_map.h"
44  #include "file.h"
45  #include "heartbeat.h"
46  #include "inode.h"
47  #include "journal.h"
48  #include "namei.h"
49  #include "suballoc.h"
50  #include "super.h"
51  #include "symlink.h"
52  #include "sysfile.h"
53  #include "uptodate.h"
54  #include "xattr.h"
55  #include "refcounttree.h"
56  #include "ocfs2_trace.h"
57  #include "filecheck.h"
58  
59  #include "buffer_head_io.h"
60  
61  struct ocfs2_find_inode_args
62  {
63  	u64		fi_blkno;
64  	unsigned long	fi_ino;
65  	unsigned int	fi_flags;
66  	unsigned int	fi_sysfile_type;
67  };
68  
69  static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
70  
71  static int ocfs2_read_locked_inode(struct inode *inode,
72  				   struct ocfs2_find_inode_args *args);
73  static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
74  static int ocfs2_find_actor(struct inode *inode, void *opaque);
75  static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
76  				    struct inode *inode,
77  				    struct buffer_head *fe_bh);
78  
79  static int ocfs2_filecheck_read_inode_block_full(struct inode *inode,
80  						 struct buffer_head **bh,
81  						 int flags, int type);
82  static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
83  						struct buffer_head *bh);
84  static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
85  					      struct buffer_head *bh);
86  
87  void ocfs2_set_inode_flags(struct inode *inode)
88  {
89  	unsigned int flags = OCFS2_I(inode)->ip_attr;
90  
91  	inode->i_flags &= ~(S_IMMUTABLE |
92  		S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
93  
94  	if (flags & OCFS2_IMMUTABLE_FL)
95  		inode->i_flags |= S_IMMUTABLE;
96  
97  	if (flags & OCFS2_SYNC_FL)
98  		inode->i_flags |= S_SYNC;
99  	if (flags & OCFS2_APPEND_FL)
100  		inode->i_flags |= S_APPEND;
101  	if (flags & OCFS2_NOATIME_FL)
102  		inode->i_flags |= S_NOATIME;
103  	if (flags & OCFS2_DIRSYNC_FL)
104  		inode->i_flags |= S_DIRSYNC;
105  }
106  
107  /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
108  void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
109  {
110  	unsigned int flags = oi->vfs_inode.i_flags;
111  
112  	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
113  			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
114  	if (flags & S_SYNC)
115  		oi->ip_attr |= OCFS2_SYNC_FL;
116  	if (flags & S_APPEND)
117  		oi->ip_attr |= OCFS2_APPEND_FL;
118  	if (flags & S_IMMUTABLE)
119  		oi->ip_attr |= OCFS2_IMMUTABLE_FL;
120  	if (flags & S_NOATIME)
121  		oi->ip_attr |= OCFS2_NOATIME_FL;
122  	if (flags & S_DIRSYNC)
123  		oi->ip_attr |= OCFS2_DIRSYNC_FL;
124  }
125  
126  struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
127  {
128  	struct ocfs2_find_inode_args args;
129  
130  	args.fi_blkno = blkno;
131  	args.fi_flags = 0;
132  	args.fi_ino = ino_from_blkno(sb, blkno);
133  	args.fi_sysfile_type = 0;
134  
135  	return ilookup5(sb, blkno, ocfs2_find_actor, &args);
136  }
137  struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
138  			 int sysfile_type)
139  {
140  	int rc = -ESTALE;
141  	struct inode *inode = NULL;
142  	struct super_block *sb = osb->sb;
143  	struct ocfs2_find_inode_args args;
144  	journal_t *journal = OCFS2_SB(sb)->journal->j_journal;
145  
146  	trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
147  			       sysfile_type);
148  
149  	/* Ok. By now we've either got the offsets passed to us by the
150  	 * caller, or we just pulled them off the bh. Lets do some
151  	 * sanity checks to make sure they're OK. */
152  	if (blkno == 0) {
153  		inode = ERR_PTR(-EINVAL);
154  		mlog_errno(PTR_ERR(inode));
155  		goto bail;
156  	}
157  
158  	args.fi_blkno = blkno;
159  	args.fi_flags = flags;
160  	args.fi_ino = ino_from_blkno(sb, blkno);
161  	args.fi_sysfile_type = sysfile_type;
162  
163  	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
164  			     ocfs2_init_locked_inode, &args);
165  	/* inode was *not* in the inode cache. 2.6.x requires
166  	 * us to do our own read_inode call and unlock it
167  	 * afterwards. */
168  	if (inode == NULL) {
169  		inode = ERR_PTR(-ENOMEM);
170  		mlog_errno(PTR_ERR(inode));
171  		goto bail;
172  	}
173  	trace_ocfs2_iget5_locked(inode->i_state);
174  	if (inode->i_state & I_NEW) {
175  		rc = ocfs2_read_locked_inode(inode, &args);
176  		unlock_new_inode(inode);
177  	}
178  	if (is_bad_inode(inode)) {
179  		iput(inode);
180  		inode = ERR_PTR(rc);
181  		goto bail;
182  	}
183  
184  	/*
185  	 * Set transaction id's of transactions that have to be committed
186  	 * to finish f[data]sync. We set them to currently running transaction
187  	 * as we cannot be sure that the inode or some of its metadata isn't
188  	 * part of the transaction - the inode could have been reclaimed and
189  	 * now it is reread from disk.
190  	 */
191  	if (journal) {
192  		transaction_t *transaction;
193  		tid_t tid;
194  		struct ocfs2_inode_info *oi = OCFS2_I(inode);
195  
196  		read_lock(&journal->j_state_lock);
197  		if (journal->j_running_transaction)
198  			transaction = journal->j_running_transaction;
199  		else
200  			transaction = journal->j_committing_transaction;
201  		if (transaction)
202  			tid = transaction->t_tid;
203  		else
204  			tid = journal->j_commit_sequence;
205  		read_unlock(&journal->j_state_lock);
206  		oi->i_sync_tid = tid;
207  		oi->i_datasync_tid = tid;
208  	}
209  
210  bail:
211  	if (!IS_ERR(inode)) {
212  		trace_ocfs2_iget_end(inode,
213  			(unsigned long long)OCFS2_I(inode)->ip_blkno);
214  	}
215  
216  	return inode;
217  }
218  
219  
220  /*
221   * here's how inodes get read from disk:
222   * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
223   * found? : return the in-memory inode
224   * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
225   */
226  
227  static int ocfs2_find_actor(struct inode *inode, void *opaque)
228  {
229  	struct ocfs2_find_inode_args *args = NULL;
230  	struct ocfs2_inode_info *oi = OCFS2_I(inode);
231  	int ret = 0;
232  
233  	args = opaque;
234  
235  	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
236  
237  	trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
238  
239  	if (oi->ip_blkno != args->fi_blkno)
240  		goto bail;
241  
242  	ret = 1;
243  bail:
244  	return ret;
245  }
246  
247  /*
248   * initialize the new inode, but don't do anything that would cause
249   * us to sleep.
250   * return 0 on success, 1 on failure
251   */
252  static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
253  {
254  	struct ocfs2_find_inode_args *args = opaque;
255  	static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
256  				     ocfs2_file_ip_alloc_sem_key;
257  
258  	inode->i_ino = args->fi_ino;
259  	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
260  	if (args->fi_sysfile_type != 0)
261  		lockdep_set_class(&inode->i_rwsem,
262  			&ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
263  	if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
264  	    args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
265  	    args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
266  	    args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
267  		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
268  				  &ocfs2_quota_ip_alloc_sem_key);
269  	else
270  		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
271  				  &ocfs2_file_ip_alloc_sem_key);
272  
273  	return 0;
274  }
275  
276  void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
277  			  int create_ino)
278  {
279  	struct super_block *sb;
280  	struct ocfs2_super *osb;
281  	int use_plocks = 1;
282  
283  	sb = inode->i_sb;
284  	osb = OCFS2_SB(sb);
285  
286  	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
287  	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
288  		use_plocks = 0;
289  
290  	/*
291  	 * These have all been checked by ocfs2_read_inode_block() or set
292  	 * by ocfs2_mknod_locked(), so a failure is a code bug.
293  	 */
294  	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
295  						cannot create a superblock
296  						inode today.  change if
297  						that is needed. */
298  	BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
299  	BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
300  
301  
302  	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
303  	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
304  	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
305  
306  	inode_set_iversion(inode, 1);
307  	inode->i_generation = le32_to_cpu(fe->i_generation);
308  	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
309  	inode->i_mode = le16_to_cpu(fe->i_mode);
310  	i_uid_write(inode, le32_to_cpu(fe->i_uid));
311  	i_gid_write(inode, le32_to_cpu(fe->i_gid));
312  
313  	/* Fast symlinks will have i_size but no allocated clusters. */
314  	if (S_ISLNK(inode->i_mode) && !fe->i_clusters) {
315  		inode->i_blocks = 0;
316  		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
317  	} else {
318  		inode->i_blocks = ocfs2_inode_sector_count(inode);
319  		inode->i_mapping->a_ops = &ocfs2_aops;
320  	}
321  	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
322  	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
323  	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
324  	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
325  	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
326  	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
327  
328  	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
329  		mlog(ML_ERROR,
330  		     "ip_blkno %llu != i_blkno %llu!\n",
331  		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
332  		     (unsigned long long)le64_to_cpu(fe->i_blkno));
333  
334  	set_nlink(inode, ocfs2_read_links_count(fe));
335  
336  	trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
337  				   le32_to_cpu(fe->i_flags));
338  	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
339  		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
340  		inode->i_flags |= S_NOQUOTA;
341  	}
342  
343  	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
344  		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
345  	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
346  		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
347  	} else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
348  		inode->i_flags |= S_NOQUOTA;
349  	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
350  		/* we can't actually hit this as read_inode can't
351  		 * handle superblocks today ;-) */
352  		BUG();
353  	}
354  
355  	switch (inode->i_mode & S_IFMT) {
356  	    case S_IFREG:
357  		    if (use_plocks)
358  			    inode->i_fop = &ocfs2_fops;
359  		    else
360  			    inode->i_fop = &ocfs2_fops_no_plocks;
361  		    inode->i_op = &ocfs2_file_iops;
362  		    i_size_write(inode, le64_to_cpu(fe->i_size));
363  		    break;
364  	    case S_IFDIR:
365  		    inode->i_op = &ocfs2_dir_iops;
366  		    if (use_plocks)
367  			    inode->i_fop = &ocfs2_dops;
368  		    else
369  			    inode->i_fop = &ocfs2_dops_no_plocks;
370  		    i_size_write(inode, le64_to_cpu(fe->i_size));
371  		    OCFS2_I(inode)->ip_dir_lock_gen = 1;
372  		    break;
373  	    case S_IFLNK:
374  		    inode->i_op = &ocfs2_symlink_inode_operations;
375  		    inode_nohighmem(inode);
376  		    i_size_write(inode, le64_to_cpu(fe->i_size));
377  		    break;
378  	    default:
379  		    inode->i_op = &ocfs2_special_file_iops;
380  		    init_special_inode(inode, inode->i_mode,
381  				       inode->i_rdev);
382  		    break;
383  	}
384  
385  	if (create_ino) {
386  		inode->i_ino = ino_from_blkno(inode->i_sb,
387  			       le64_to_cpu(fe->i_blkno));
388  
389  		/*
390  		 * If we ever want to create system files from kernel,
391  		 * the generation argument to
392  		 * ocfs2_inode_lock_res_init() will have to change.
393  		 */
394  		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
395  
396  		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
397  					  OCFS2_LOCK_TYPE_META, 0, inode);
398  
399  		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
400  					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
401  	}
402  
403  	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
404  				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
405  				  inode);
406  
407  	ocfs2_set_inode_flags(inode);
408  
409  	OCFS2_I(inode)->ip_last_used_slot = 0;
410  	OCFS2_I(inode)->ip_last_used_group = 0;
411  
412  	if (S_ISDIR(inode->i_mode))
413  		ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
414  				    OCFS2_RESV_FLAG_DIR);
415  }
416  
417  static int ocfs2_read_locked_inode(struct inode *inode,
418  				   struct ocfs2_find_inode_args *args)
419  {
420  	struct super_block *sb;
421  	struct ocfs2_super *osb;
422  	struct ocfs2_dinode *fe;
423  	struct buffer_head *bh = NULL;
424  	int status, can_lock, lock_level = 0;
425  	u32 generation = 0;
426  
427  	status = -EINVAL;
428  	sb = inode->i_sb;
429  	osb = OCFS2_SB(sb);
430  
431  	/*
432  	 * To improve performance of cold-cache inode stats, we take
433  	 * the cluster lock here if possible.
434  	 *
435  	 * Generally, OCFS2 never trusts the contents of an inode
436  	 * unless it's holding a cluster lock, so taking it here isn't
437  	 * a correctness issue as much as it is a performance
438  	 * improvement.
439  	 *
440  	 * There are three times when taking the lock is not a good idea:
441  	 *
442  	 * 1) During startup, before we have initialized the DLM.
443  	 *
444  	 * 2) If we are reading certain system files which never get
445  	 *    cluster locks (local alloc, truncate log).
446  	 *
447  	 * 3) If the process doing the iget() is responsible for
448  	 *    orphan dir recovery. We're holding the orphan dir lock and
449  	 *    can get into a deadlock with another process on another
450  	 *    node in ->delete_inode().
451  	 *
452  	 * #1 and #2 can be simply solved by never taking the lock
453  	 * here for system files (which are the only type we read
454  	 * during mount). It's a heavier approach, but our main
455  	 * concern is user-accessible files anyway.
456  	 *
457  	 * #3 works itself out because we'll eventually take the
458  	 * cluster lock before trusting anything anyway.
459  	 */
460  	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
461  		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
462  		&& !ocfs2_mount_local(osb);
463  
464  	trace_ocfs2_read_locked_inode(
465  		(unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
466  
467  	/*
468  	 * To maintain backwards compatibility with older versions of
469  	 * ocfs2-tools, we still store the generation value for system
470  	 * files. The only ones that actually matter to userspace are
471  	 * the journals, but it's easier and inexpensive to just flag
472  	 * all system files similarly.
473  	 */
474  	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
475  		generation = osb->fs_generation;
476  
477  	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
478  				  OCFS2_LOCK_TYPE_META,
479  				  generation, inode);
480  
481  	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
482  				  OCFS2_LOCK_TYPE_OPEN,
483  				  0, inode);
484  
485  	if (can_lock) {
486  		status = ocfs2_open_lock(inode);
487  		if (status) {
488  			make_bad_inode(inode);
489  			mlog_errno(status);
490  			return status;
491  		}
492  		status = ocfs2_inode_lock(inode, NULL, lock_level);
493  		if (status) {
494  			make_bad_inode(inode);
495  			mlog_errno(status);
496  			return status;
497  		}
498  	}
499  
500  	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
501  		status = ocfs2_try_open_lock(inode, 0);
502  		if (status) {
503  			make_bad_inode(inode);
504  			return status;
505  		}
506  	}
507  
508  	if (can_lock) {
509  		if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
510  			status = ocfs2_filecheck_read_inode_block_full(inode,
511  						&bh, OCFS2_BH_IGNORE_CACHE, 0);
512  		else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
513  			status = ocfs2_filecheck_read_inode_block_full(inode,
514  						&bh, OCFS2_BH_IGNORE_CACHE, 1);
515  		else
516  			status = ocfs2_read_inode_block_full(inode,
517  						&bh, OCFS2_BH_IGNORE_CACHE);
518  	} else {
519  		status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
520  		/*
521  		 * If buffer is in jbd, then its checksum may not have been
522  		 * computed as yet.
523  		 */
524  		if (!status && !buffer_jbd(bh)) {
525  			if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK)
526  				status = ocfs2_filecheck_validate_inode_block(
527  								osb->sb, bh);
528  			else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX)
529  				status = ocfs2_filecheck_repair_inode_block(
530  								osb->sb, bh);
531  			else
532  				status = ocfs2_validate_inode_block(
533  								osb->sb, bh);
534  		}
535  	}
536  	if (status < 0) {
537  		mlog_errno(status);
538  		goto bail;
539  	}
540  
541  	status = -EINVAL;
542  	fe = (struct ocfs2_dinode *) bh->b_data;
543  
544  	/*
545  	 * This is a code bug. Right now the caller needs to
546  	 * understand whether it is asking for a system file inode or
547  	 * not so the proper lock names can be built.
548  	 */
549  	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
550  			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
551  			"Inode %llu: system file state is ambigous\n",
552  			(unsigned long long)args->fi_blkno);
553  
554  	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
555  	    S_ISBLK(le16_to_cpu(fe->i_mode)))
556  		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
557  
558  	ocfs2_populate_inode(inode, fe, 0);
559  
560  	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
561  
562  	if (buffer_dirty(bh) && !buffer_jbd(bh)) {
563  		if (can_lock) {
564  			ocfs2_inode_unlock(inode, lock_level);
565  			lock_level = 1;
566  			ocfs2_inode_lock(inode, NULL, lock_level);
567  		}
568  		status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
569  		if (status < 0) {
570  			mlog_errno(status);
571  			goto bail;
572  		}
573  	}
574  
575  	status = 0;
576  
577  bail:
578  	if (can_lock)
579  		ocfs2_inode_unlock(inode, lock_level);
580  
581  	if (status < 0)
582  		make_bad_inode(inode);
583  
584  	brelse(bh);
585  
586  	return status;
587  }
588  
589  void ocfs2_sync_blockdev(struct super_block *sb)
590  {
591  	sync_blockdev(sb->s_bdev);
592  }
593  
594  static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
595  				     struct inode *inode,
596  				     struct buffer_head *fe_bh)
597  {
598  	int status = 0;
599  	struct ocfs2_dinode *fe;
600  	handle_t *handle = NULL;
601  
602  	fe = (struct ocfs2_dinode *) fe_bh->b_data;
603  
604  	/*
605  	 * This check will also skip truncate of inodes with inline
606  	 * data and fast symlinks.
607  	 */
608  	if (fe->i_clusters) {
609  		if (ocfs2_should_order_data(inode))
610  			ocfs2_begin_ordered_truncate(inode, 0);
611  
612  		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
613  		if (IS_ERR(handle)) {
614  			status = PTR_ERR(handle);
615  			handle = NULL;
616  			mlog_errno(status);
617  			goto out;
618  		}
619  
620  		status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
621  						 fe_bh,
622  						 OCFS2_JOURNAL_ACCESS_WRITE);
623  		if (status < 0) {
624  			mlog_errno(status);
625  			goto out;
626  		}
627  
628  		i_size_write(inode, 0);
629  
630  		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
631  		if (status < 0) {
632  			mlog_errno(status);
633  			goto out;
634  		}
635  
636  		ocfs2_commit_trans(osb, handle);
637  		handle = NULL;
638  
639  		status = ocfs2_commit_truncate(osb, inode, fe_bh);
640  		if (status < 0)
641  			mlog_errno(status);
642  	}
643  
644  out:
645  	if (handle)
646  		ocfs2_commit_trans(osb, handle);
647  	return status;
648  }
649  
650  static int ocfs2_remove_inode(struct inode *inode,
651  			      struct buffer_head *di_bh,
652  			      struct inode *orphan_dir_inode,
653  			      struct buffer_head *orphan_dir_bh)
654  {
655  	int status;
656  	struct inode *inode_alloc_inode = NULL;
657  	struct buffer_head *inode_alloc_bh = NULL;
658  	handle_t *handle;
659  	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
660  	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
661  
662  	inode_alloc_inode =
663  		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
664  					    le16_to_cpu(di->i_suballoc_slot));
665  	if (!inode_alloc_inode) {
666  		status = -ENOENT;
667  		mlog_errno(status);
668  		goto bail;
669  	}
670  
671  	inode_lock(inode_alloc_inode);
672  	status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
673  	if (status < 0) {
674  		inode_unlock(inode_alloc_inode);
675  
676  		mlog_errno(status);
677  		goto bail;
678  	}
679  
680  	handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
681  				   ocfs2_quota_trans_credits(inode->i_sb));
682  	if (IS_ERR(handle)) {
683  		status = PTR_ERR(handle);
684  		mlog_errno(status);
685  		goto bail_unlock;
686  	}
687  
688  	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
689  		status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
690  					  orphan_dir_bh, false);
691  		if (status < 0) {
692  			mlog_errno(status);
693  			goto bail_commit;
694  		}
695  	}
696  
697  	/* set the inodes dtime */
698  	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
699  					 OCFS2_JOURNAL_ACCESS_WRITE);
700  	if (status < 0) {
701  		mlog_errno(status);
702  		goto bail_commit;
703  	}
704  
705  	di->i_dtime = cpu_to_le64(ktime_get_real_seconds());
706  	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
707  	ocfs2_journal_dirty(handle, di_bh);
708  
709  	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
710  	dquot_free_inode(inode);
711  
712  	status = ocfs2_free_dinode(handle, inode_alloc_inode,
713  				   inode_alloc_bh, di);
714  	if (status < 0)
715  		mlog_errno(status);
716  
717  bail_commit:
718  	ocfs2_commit_trans(osb, handle);
719  bail_unlock:
720  	ocfs2_inode_unlock(inode_alloc_inode, 1);
721  	inode_unlock(inode_alloc_inode);
722  	brelse(inode_alloc_bh);
723  bail:
724  	iput(inode_alloc_inode);
725  
726  	return status;
727  }
728  
729  /*
730   * Serialize with orphan dir recovery. If the process doing
731   * recovery on this orphan dir does an iget() with the dir
732   * i_mutex held, we'll deadlock here. Instead we detect this
733   * and exit early - recovery will wipe this inode for us.
734   */
735  static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
736  					     int slot)
737  {
738  	int ret = 0;
739  
740  	spin_lock(&osb->osb_lock);
741  	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
742  		ret = -EDEADLK;
743  		goto out;
744  	}
745  	/* This signals to the orphan recovery process that it should
746  	 * wait for us to handle the wipe. */
747  	osb->osb_orphan_wipes[slot]++;
748  out:
749  	spin_unlock(&osb->osb_lock);
750  	trace_ocfs2_check_orphan_recovery_state(slot, ret);
751  	return ret;
752  }
753  
754  static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
755  					 int slot)
756  {
757  	spin_lock(&osb->osb_lock);
758  	osb->osb_orphan_wipes[slot]--;
759  	spin_unlock(&osb->osb_lock);
760  
761  	wake_up(&osb->osb_wipe_event);
762  }
763  
764  static int ocfs2_wipe_inode(struct inode *inode,
765  			    struct buffer_head *di_bh)
766  {
767  	int status, orphaned_slot = -1;
768  	struct inode *orphan_dir_inode = NULL;
769  	struct buffer_head *orphan_dir_bh = NULL;
770  	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
771  	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
772  
773  	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
774  		orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
775  
776  		status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
777  		if (status)
778  			return status;
779  
780  		orphan_dir_inode = ocfs2_get_system_file_inode(osb,
781  							       ORPHAN_DIR_SYSTEM_INODE,
782  							       orphaned_slot);
783  		if (!orphan_dir_inode) {
784  			status = -ENOENT;
785  			mlog_errno(status);
786  			goto bail;
787  		}
788  
789  		/* Lock the orphan dir. The lock will be held for the entire
790  		 * delete_inode operation. We do this now to avoid races with
791  		 * recovery completion on other nodes. */
792  		inode_lock(orphan_dir_inode);
793  		status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
794  		if (status < 0) {
795  			inode_unlock(orphan_dir_inode);
796  
797  			mlog_errno(status);
798  			goto bail;
799  		}
800  	}
801  
802  	/* we do this while holding the orphan dir lock because we
803  	 * don't want recovery being run from another node to try an
804  	 * inode delete underneath us -- this will result in two nodes
805  	 * truncating the same file! */
806  	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
807  	if (status < 0) {
808  		mlog_errno(status);
809  		goto bail_unlock_dir;
810  	}
811  
812  	/* Remove any dir index tree */
813  	if (S_ISDIR(inode->i_mode)) {
814  		status = ocfs2_dx_dir_truncate(inode, di_bh);
815  		if (status) {
816  			mlog_errno(status);
817  			goto bail_unlock_dir;
818  		}
819  	}
820  
821  	/*Free extended attribute resources associated with this inode.*/
822  	status = ocfs2_xattr_remove(inode, di_bh);
823  	if (status < 0) {
824  		mlog_errno(status);
825  		goto bail_unlock_dir;
826  	}
827  
828  	status = ocfs2_remove_refcount_tree(inode, di_bh);
829  	if (status < 0) {
830  		mlog_errno(status);
831  		goto bail_unlock_dir;
832  	}
833  
834  	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
835  				    orphan_dir_bh);
836  	if (status < 0)
837  		mlog_errno(status);
838  
839  bail_unlock_dir:
840  	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
841  		return status;
842  
843  	ocfs2_inode_unlock(orphan_dir_inode, 1);
844  	inode_unlock(orphan_dir_inode);
845  	brelse(orphan_dir_bh);
846  bail:
847  	iput(orphan_dir_inode);
848  	ocfs2_signal_wipe_completion(osb, orphaned_slot);
849  
850  	return status;
851  }
852  
853  /* There is a series of simple checks that should be done before a
854   * trylock is even considered. Encapsulate those in this function. */
855  static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
856  {
857  	int ret = 0;
858  	struct ocfs2_inode_info *oi = OCFS2_I(inode);
859  	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
860  
861  	trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
862  					     (unsigned long long)oi->ip_blkno,
863  					     oi->ip_flags);
864  
865  	/* We shouldn't be getting here for the root directory
866  	 * inode.. */
867  	if (inode == osb->root_inode) {
868  		mlog(ML_ERROR, "Skipping delete of root inode.\n");
869  		goto bail;
870  	}
871  
872  	/*
873  	 * If we're coming from downconvert_thread we can't go into our own
874  	 * voting [hello, deadlock city!] so we cannot delete the inode. But
875  	 * since we dropped last inode ref when downconverting dentry lock,
876  	 * we cannot have the file open and thus the node doing unlink will
877  	 * take care of deleting the inode.
878  	 */
879  	if (current == osb->dc_task)
880  		goto bail;
881  
882  	spin_lock(&oi->ip_lock);
883  	/* OCFS2 *never* deletes system files. This should technically
884  	 * never get here as system file inodes should always have a
885  	 * positive link count. */
886  	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
887  		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
888  		     (unsigned long long)oi->ip_blkno);
889  		goto bail_unlock;
890  	}
891  
892  	ret = 1;
893  bail_unlock:
894  	spin_unlock(&oi->ip_lock);
895  bail:
896  	return ret;
897  }
898  
899  /* Query the cluster to determine whether we should wipe an inode from
900   * disk or not.
901   *
902   * Requires the inode to have the cluster lock. */
903  static int ocfs2_query_inode_wipe(struct inode *inode,
904  				  struct buffer_head *di_bh,
905  				  int *wipe)
906  {
907  	int status = 0, reason = 0;
908  	struct ocfs2_inode_info *oi = OCFS2_I(inode);
909  	struct ocfs2_dinode *di;
910  
911  	*wipe = 0;
912  
913  	trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
914  					   inode->i_nlink);
915  
916  	/* While we were waiting for the cluster lock in
917  	 * ocfs2_delete_inode, another node might have asked to delete
918  	 * the inode. Recheck our flags to catch this. */
919  	if (!ocfs2_inode_is_valid_to_delete(inode)) {
920  		reason = 1;
921  		goto bail;
922  	}
923  
924  	/* Now that we have an up to date inode, we can double check
925  	 * the link count. */
926  	if (inode->i_nlink)
927  		goto bail;
928  
929  	/* Do some basic inode verification... */
930  	di = (struct ocfs2_dinode *) di_bh->b_data;
931  	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
932  	    !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
933  		/*
934  		 * Inodes in the orphan dir must have ORPHANED_FL.  The only
935  		 * inodes that come back out of the orphan dir are reflink
936  		 * targets. A reflink target may be moved out of the orphan
937  		 * dir between the time we scan the directory and the time we
938  		 * process it. This would lead to HAS_REFCOUNT_FL being set but
939  		 * ORPHANED_FL not.
940  		 */
941  		if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
942  			reason = 2;
943  			goto bail;
944  		}
945  
946  		/* for lack of a better error? */
947  		status = -EEXIST;
948  		mlog(ML_ERROR,
949  		     "Inode %llu (on-disk %llu) not orphaned! "
950  		     "Disk flags  0x%x, inode flags 0x%x\n",
951  		     (unsigned long long)oi->ip_blkno,
952  		     (unsigned long long)le64_to_cpu(di->i_blkno),
953  		     le32_to_cpu(di->i_flags), oi->ip_flags);
954  		goto bail;
955  	}
956  
957  	/* has someone already deleted us?! baaad... */
958  	if (di->i_dtime) {
959  		status = -EEXIST;
960  		mlog_errno(status);
961  		goto bail;
962  	}
963  
964  	/*
965  	 * This is how ocfs2 determines whether an inode is still live
966  	 * within the cluster. Every node takes a shared read lock on
967  	 * the inode open lock in ocfs2_read_locked_inode(). When we
968  	 * get to ->delete_inode(), each node tries to convert it's
969  	 * lock to an exclusive. Trylocks are serialized by the inode
970  	 * meta data lock. If the upconvert succeeds, we know the inode
971  	 * is no longer live and can be deleted.
972  	 *
973  	 * Though we call this with the meta data lock held, the
974  	 * trylock keeps us from ABBA deadlock.
975  	 */
976  	status = ocfs2_try_open_lock(inode, 1);
977  	if (status == -EAGAIN) {
978  		status = 0;
979  		reason = 3;
980  		goto bail;
981  	}
982  	if (status < 0) {
983  		mlog_errno(status);
984  		goto bail;
985  	}
986  
987  	*wipe = 1;
988  	trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
989  
990  bail:
991  	trace_ocfs2_query_inode_wipe_end(status, reason);
992  	return status;
993  }
994  
995  /* Support function for ocfs2_delete_inode. Will help us keep the
996   * inode data in a consistent state for clear_inode. Always truncates
997   * pages, optionally sync's them first. */
998  static void ocfs2_cleanup_delete_inode(struct inode *inode,
999  				       int sync_data)
1000  {
1001  	trace_ocfs2_cleanup_delete_inode(
1002  		(unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
1003  	if (sync_data)
1004  		filemap_write_and_wait(inode->i_mapping);
1005  	truncate_inode_pages_final(&inode->i_data);
1006  }
1007  
1008  static void ocfs2_delete_inode(struct inode *inode)
1009  {
1010  	int wipe, status;
1011  	sigset_t oldset;
1012  	struct buffer_head *di_bh = NULL;
1013  	struct ocfs2_dinode *di = NULL;
1014  
1015  	trace_ocfs2_delete_inode(inode->i_ino,
1016  				 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1017  				 is_bad_inode(inode));
1018  
1019  	/* When we fail in read_inode() we mark inode as bad. The second test
1020  	 * catches the case when inode allocation fails before allocating
1021  	 * a block for inode. */
1022  	if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
1023  		goto bail;
1024  
1025  	if (!ocfs2_inode_is_valid_to_delete(inode)) {
1026  		/* It's probably not necessary to truncate_inode_pages
1027  		 * here but we do it for safety anyway (it will most
1028  		 * likely be a no-op anyway) */
1029  		ocfs2_cleanup_delete_inode(inode, 0);
1030  		goto bail;
1031  	}
1032  
1033  	dquot_initialize(inode);
1034  
1035  	/* We want to block signals in delete_inode as the lock and
1036  	 * messaging paths may return us -ERESTARTSYS. Which would
1037  	 * cause us to exit early, resulting in inodes being orphaned
1038  	 * forever. */
1039  	ocfs2_block_signals(&oldset);
1040  
1041  	/*
1042  	 * Synchronize us against ocfs2_get_dentry. We take this in
1043  	 * shared mode so that all nodes can still concurrently
1044  	 * process deletes.
1045  	 */
1046  	status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1047  	if (status < 0) {
1048  		mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1049  		ocfs2_cleanup_delete_inode(inode, 0);
1050  		goto bail_unblock;
1051  	}
1052  	/* Lock down the inode. This gives us an up to date view of
1053  	 * it's metadata (for verification), and allows us to
1054  	 * serialize delete_inode on multiple nodes.
1055  	 *
1056  	 * Even though we might be doing a truncate, we don't take the
1057  	 * allocation lock here as it won't be needed - nobody will
1058  	 * have the file open.
1059  	 */
1060  	status = ocfs2_inode_lock(inode, &di_bh, 1);
1061  	if (status < 0) {
1062  		if (status != -ENOENT)
1063  			mlog_errno(status);
1064  		ocfs2_cleanup_delete_inode(inode, 0);
1065  		goto bail_unlock_nfs_sync;
1066  	}
1067  
1068  	di = (struct ocfs2_dinode *)di_bh->b_data;
1069  	/* Skip inode deletion and wait for dio orphan entry recovered
1070  	 * first */
1071  	if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
1072  		ocfs2_cleanup_delete_inode(inode, 0);
1073  		goto bail_unlock_inode;
1074  	}
1075  
1076  	/* Query the cluster. This will be the final decision made
1077  	 * before we go ahead and wipe the inode. */
1078  	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1079  	if (!wipe || status < 0) {
1080  		/* Error and remote inode busy both mean we won't be
1081  		 * removing the inode, so they take almost the same
1082  		 * path. */
1083  		if (status < 0)
1084  			mlog_errno(status);
1085  
1086  		/* Someone in the cluster has disallowed a wipe of
1087  		 * this inode, or it was never completely
1088  		 * orphaned. Write out the pages and exit now. */
1089  		ocfs2_cleanup_delete_inode(inode, 1);
1090  		goto bail_unlock_inode;
1091  	}
1092  
1093  	ocfs2_cleanup_delete_inode(inode, 0);
1094  
1095  	status = ocfs2_wipe_inode(inode, di_bh);
1096  	if (status < 0) {
1097  		if (status != -EDEADLK)
1098  			mlog_errno(status);
1099  		goto bail_unlock_inode;
1100  	}
1101  
1102  	/*
1103  	 * Mark the inode as successfully deleted.
1104  	 *
1105  	 * This is important for ocfs2_clear_inode() as it will check
1106  	 * this flag and skip any checkpointing work
1107  	 *
1108  	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1109  	 * the LVB for other nodes.
1110  	 */
1111  	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1112  
1113  bail_unlock_inode:
1114  	ocfs2_inode_unlock(inode, 1);
1115  	brelse(di_bh);
1116  
1117  bail_unlock_nfs_sync:
1118  	ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1119  
1120  bail_unblock:
1121  	ocfs2_unblock_signals(&oldset);
1122  bail:
1123  	return;
1124  }
1125  
1126  static void ocfs2_clear_inode(struct inode *inode)
1127  {
1128  	int status;
1129  	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1130  	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1131  
1132  	clear_inode(inode);
1133  	trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1134  				inode->i_nlink);
1135  
1136  	mlog_bug_on_msg(osb == NULL,
1137  			"Inode=%lu\n", inode->i_ino);
1138  
1139  	dquot_drop(inode);
1140  
1141  	/* To preven remote deletes we hold open lock before, now it
1142  	 * is time to unlock PR and EX open locks. */
1143  	ocfs2_open_unlock(inode);
1144  
1145  	/* Do these before all the other work so that we don't bounce
1146  	 * the downconvert thread while waiting to destroy the locks. */
1147  	ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres);
1148  	ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres);
1149  	ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres);
1150  
1151  	ocfs2_resv_discard(&osb->osb_la_resmap,
1152  			   &oi->ip_la_data_resv);
1153  	ocfs2_resv_init_once(&oi->ip_la_data_resv);
1154  
1155  	/* We very well may get a clear_inode before all an inodes
1156  	 * metadata has hit disk. Of course, we can't drop any cluster
1157  	 * locks until the journal has finished with it. The only
1158  	 * exception here are successfully wiped inodes - their
1159  	 * metadata can now be considered to be part of the system
1160  	 * inodes from which it came. */
1161  	if (!(oi->ip_flags & OCFS2_INODE_DELETED))
1162  		ocfs2_checkpoint_inode(inode);
1163  
1164  	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1165  			"Clear inode of %llu, inode has io markers\n",
1166  			(unsigned long long)oi->ip_blkno);
1167  	mlog_bug_on_msg(!list_empty(&oi->ip_unwritten_list),
1168  			"Clear inode of %llu, inode has unwritten extents\n",
1169  			(unsigned long long)oi->ip_blkno);
1170  
1171  	ocfs2_extent_map_trunc(inode, 0);
1172  
1173  	status = ocfs2_drop_inode_locks(inode);
1174  	if (status < 0)
1175  		mlog_errno(status);
1176  
1177  	ocfs2_lock_res_free(&oi->ip_rw_lockres);
1178  	ocfs2_lock_res_free(&oi->ip_inode_lockres);
1179  	ocfs2_lock_res_free(&oi->ip_open_lockres);
1180  
1181  	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1182  
1183  	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1184  			"Clear inode of %llu, inode has %u cache items\n",
1185  			(unsigned long long)oi->ip_blkno,
1186  			INODE_CACHE(inode)->ci_num_cached);
1187  
1188  	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1189  			"Clear inode of %llu, inode has a bad flag\n",
1190  			(unsigned long long)oi->ip_blkno);
1191  
1192  	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1193  			"Clear inode of %llu, inode is locked\n",
1194  			(unsigned long long)oi->ip_blkno);
1195  
1196  	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1197  			"Clear inode of %llu, io_mutex is locked\n",
1198  			(unsigned long long)oi->ip_blkno);
1199  	mutex_unlock(&oi->ip_io_mutex);
1200  
1201  	/*
1202  	 * down_trylock() returns 0, down_write_trylock() returns 1
1203  	 * kernel 1, world 0
1204  	 */
1205  	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1206  			"Clear inode of %llu, alloc_sem is locked\n",
1207  			(unsigned long long)oi->ip_blkno);
1208  	up_write(&oi->ip_alloc_sem);
1209  
1210  	mlog_bug_on_msg(oi->ip_open_count,
1211  			"Clear inode of %llu has open count %d\n",
1212  			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
1213  
1214  	/* Clear all other flags. */
1215  	oi->ip_flags = 0;
1216  	oi->ip_dir_start_lookup = 0;
1217  	oi->ip_blkno = 0ULL;
1218  
1219  	/*
1220  	 * ip_jinode is used to track txns against this inode. We ensure that
1221  	 * the journal is flushed before journal shutdown. Thus it is safe to
1222  	 * have inodes get cleaned up after journal shutdown.
1223  	 */
1224  	jbd2_journal_release_jbd_inode(osb->journal->j_journal,
1225  				       &oi->ip_jinode);
1226  }
1227  
1228  void ocfs2_evict_inode(struct inode *inode)
1229  {
1230  	if (!inode->i_nlink ||
1231  	    (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1232  		ocfs2_delete_inode(inode);
1233  	} else {
1234  		truncate_inode_pages_final(&inode->i_data);
1235  	}
1236  	ocfs2_clear_inode(inode);
1237  }
1238  
1239  /* Called under inode_lock, with no more references on the
1240   * struct inode, so it's safe here to check the flags field
1241   * and to manipulate i_nlink without any other locks. */
1242  int ocfs2_drop_inode(struct inode *inode)
1243  {
1244  	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1245  
1246  	trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
1247  				inode->i_nlink, oi->ip_flags);
1248  
1249  	assert_spin_locked(&inode->i_lock);
1250  	inode->i_state |= I_WILL_FREE;
1251  	spin_unlock(&inode->i_lock);
1252  	write_inode_now(inode, 1);
1253  	spin_lock(&inode->i_lock);
1254  	WARN_ON(inode->i_state & I_NEW);
1255  	inode->i_state &= ~I_WILL_FREE;
1256  
1257  	return 1;
1258  }
1259  
1260  /*
1261   * This is called from our getattr.
1262   */
1263  int ocfs2_inode_revalidate(struct dentry *dentry)
1264  {
1265  	struct inode *inode = d_inode(dentry);
1266  	int status = 0;
1267  
1268  	trace_ocfs2_inode_revalidate(inode,
1269  		inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1270  		inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
1271  
1272  	if (!inode) {
1273  		status = -ENOENT;
1274  		goto bail;
1275  	}
1276  
1277  	spin_lock(&OCFS2_I(inode)->ip_lock);
1278  	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1279  		spin_unlock(&OCFS2_I(inode)->ip_lock);
1280  		status = -ENOENT;
1281  		goto bail;
1282  	}
1283  	spin_unlock(&OCFS2_I(inode)->ip_lock);
1284  
1285  	/* Let ocfs2_inode_lock do the work of updating our struct
1286  	 * inode for us. */
1287  	status = ocfs2_inode_lock(inode, NULL, 0);
1288  	if (status < 0) {
1289  		if (status != -ENOENT)
1290  			mlog_errno(status);
1291  		goto bail;
1292  	}
1293  	ocfs2_inode_unlock(inode, 0);
1294  bail:
1295  	return status;
1296  }
1297  
1298  /*
1299   * Updates a disk inode from a
1300   * struct inode.
1301   * Only takes ip_lock.
1302   */
1303  int ocfs2_mark_inode_dirty(handle_t *handle,
1304  			   struct inode *inode,
1305  			   struct buffer_head *bh)
1306  {
1307  	int status;
1308  	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1309  
1310  	trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
1311  
1312  	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1313  					 OCFS2_JOURNAL_ACCESS_WRITE);
1314  	if (status < 0) {
1315  		mlog_errno(status);
1316  		goto leave;
1317  	}
1318  
1319  	spin_lock(&OCFS2_I(inode)->ip_lock);
1320  	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1321  	ocfs2_get_inode_flags(OCFS2_I(inode));
1322  	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1323  	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1324  	spin_unlock(&OCFS2_I(inode)->ip_lock);
1325  
1326  	fe->i_size = cpu_to_le64(i_size_read(inode));
1327  	ocfs2_set_links_count(fe, inode->i_nlink);
1328  	fe->i_uid = cpu_to_le32(i_uid_read(inode));
1329  	fe->i_gid = cpu_to_le32(i_gid_read(inode));
1330  	fe->i_mode = cpu_to_le16(inode->i_mode);
1331  	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1332  	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1333  	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1334  	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1335  	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1336  	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1337  
1338  	ocfs2_journal_dirty(handle, bh);
1339  	ocfs2_update_inode_fsync_trans(handle, inode, 1);
1340  leave:
1341  	return status;
1342  }
1343  
1344  /*
1345   *
1346   * Updates a struct inode from a disk inode.
1347   * does no i/o, only takes ip_lock.
1348   */
1349  void ocfs2_refresh_inode(struct inode *inode,
1350  			 struct ocfs2_dinode *fe)
1351  {
1352  	spin_lock(&OCFS2_I(inode)->ip_lock);
1353  
1354  	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1355  	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1356  	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1357  	ocfs2_set_inode_flags(inode);
1358  	i_size_write(inode, le64_to_cpu(fe->i_size));
1359  	set_nlink(inode, ocfs2_read_links_count(fe));
1360  	i_uid_write(inode, le32_to_cpu(fe->i_uid));
1361  	i_gid_write(inode, le32_to_cpu(fe->i_gid));
1362  	inode->i_mode = le16_to_cpu(fe->i_mode);
1363  	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1364  		inode->i_blocks = 0;
1365  	else
1366  		inode->i_blocks = ocfs2_inode_sector_count(inode);
1367  	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1368  	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1369  	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1370  	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1371  	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1372  	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1373  
1374  	spin_unlock(&OCFS2_I(inode)->ip_lock);
1375  }
1376  
1377  int ocfs2_validate_inode_block(struct super_block *sb,
1378  			       struct buffer_head *bh)
1379  {
1380  	int rc;
1381  	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1382  
1383  	trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
1384  
1385  	BUG_ON(!buffer_uptodate(bh));
1386  
1387  	/*
1388  	 * If the ecc fails, we return the error but otherwise
1389  	 * leave the filesystem running.  We know any error is
1390  	 * local to this block.
1391  	 */
1392  	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1393  	if (rc) {
1394  		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1395  		     (unsigned long long)bh->b_blocknr);
1396  		goto bail;
1397  	}
1398  
1399  	/*
1400  	 * Errors after here are fatal.
1401  	 */
1402  
1403  	rc = -EINVAL;
1404  
1405  	if (!OCFS2_IS_VALID_DINODE(di)) {
1406  		rc = ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1407  				 (unsigned long long)bh->b_blocknr, 7,
1408  				 di->i_signature);
1409  		goto bail;
1410  	}
1411  
1412  	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1413  		rc = ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1414  				 (unsigned long long)bh->b_blocknr,
1415  				 (unsigned long long)le64_to_cpu(di->i_blkno));
1416  		goto bail;
1417  	}
1418  
1419  	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1420  		rc = ocfs2_error(sb,
1421  				 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1422  				 (unsigned long long)bh->b_blocknr);
1423  		goto bail;
1424  	}
1425  
1426  	if (le32_to_cpu(di->i_fs_generation) !=
1427  	    OCFS2_SB(sb)->fs_generation) {
1428  		rc = ocfs2_error(sb,
1429  				 "Invalid dinode #%llu: fs_generation is %u\n",
1430  				 (unsigned long long)bh->b_blocknr,
1431  				 le32_to_cpu(di->i_fs_generation));
1432  		goto bail;
1433  	}
1434  
1435  	rc = 0;
1436  
1437  bail:
1438  	return rc;
1439  }
1440  
1441  static int ocfs2_filecheck_validate_inode_block(struct super_block *sb,
1442  						struct buffer_head *bh)
1443  {
1444  	int rc = 0;
1445  	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1446  
1447  	trace_ocfs2_filecheck_validate_inode_block(
1448  		(unsigned long long)bh->b_blocknr);
1449  
1450  	BUG_ON(!buffer_uptodate(bh));
1451  
1452  	/*
1453  	 * Call ocfs2_validate_meta_ecc() first since it has ecc repair
1454  	 * function, but we should not return error immediately when ecc
1455  	 * validation fails, because the reason is quite likely the invalid
1456  	 * inode number inputed.
1457  	 */
1458  	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1459  	if (rc) {
1460  		mlog(ML_ERROR,
1461  		     "Filecheck: checksum failed for dinode %llu\n",
1462  		     (unsigned long long)bh->b_blocknr);
1463  		rc = -OCFS2_FILECHECK_ERR_BLOCKECC;
1464  	}
1465  
1466  	if (!OCFS2_IS_VALID_DINODE(di)) {
1467  		mlog(ML_ERROR,
1468  		     "Filecheck: invalid dinode #%llu: signature = %.*s\n",
1469  		     (unsigned long long)bh->b_blocknr, 7, di->i_signature);
1470  		rc = -OCFS2_FILECHECK_ERR_INVALIDINO;
1471  		goto bail;
1472  	} else if (rc)
1473  		goto bail;
1474  
1475  	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1476  		mlog(ML_ERROR,
1477  		     "Filecheck: invalid dinode #%llu: i_blkno is %llu\n",
1478  		     (unsigned long long)bh->b_blocknr,
1479  		     (unsigned long long)le64_to_cpu(di->i_blkno));
1480  		rc = -OCFS2_FILECHECK_ERR_BLOCKNO;
1481  		goto bail;
1482  	}
1483  
1484  	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1485  		mlog(ML_ERROR,
1486  		     "Filecheck: invalid dinode #%llu: OCFS2_VALID_FL "
1487  		     "not set\n",
1488  		     (unsigned long long)bh->b_blocknr);
1489  		rc = -OCFS2_FILECHECK_ERR_VALIDFLAG;
1490  		goto bail;
1491  	}
1492  
1493  	if (le32_to_cpu(di->i_fs_generation) !=
1494  	    OCFS2_SB(sb)->fs_generation) {
1495  		mlog(ML_ERROR,
1496  		     "Filecheck: invalid dinode #%llu: fs_generation is %u\n",
1497  		     (unsigned long long)bh->b_blocknr,
1498  		     le32_to_cpu(di->i_fs_generation));
1499  		rc = -OCFS2_FILECHECK_ERR_GENERATION;
1500  	}
1501  
1502  bail:
1503  	return rc;
1504  }
1505  
1506  static int ocfs2_filecheck_repair_inode_block(struct super_block *sb,
1507  					      struct buffer_head *bh)
1508  {
1509  	int changed = 0;
1510  	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1511  
1512  	if (!ocfs2_filecheck_validate_inode_block(sb, bh))
1513  		return 0;
1514  
1515  	trace_ocfs2_filecheck_repair_inode_block(
1516  		(unsigned long long)bh->b_blocknr);
1517  
1518  	if (ocfs2_is_hard_readonly(OCFS2_SB(sb)) ||
1519  	    ocfs2_is_soft_readonly(OCFS2_SB(sb))) {
1520  		mlog(ML_ERROR,
1521  		     "Filecheck: cannot repair dinode #%llu "
1522  		     "on readonly filesystem\n",
1523  		     (unsigned long long)bh->b_blocknr);
1524  		return -OCFS2_FILECHECK_ERR_READONLY;
1525  	}
1526  
1527  	if (buffer_jbd(bh)) {
1528  		mlog(ML_ERROR,
1529  		     "Filecheck: cannot repair dinode #%llu, "
1530  		     "its buffer is in jbd\n",
1531  		     (unsigned long long)bh->b_blocknr);
1532  		return -OCFS2_FILECHECK_ERR_INJBD;
1533  	}
1534  
1535  	if (!OCFS2_IS_VALID_DINODE(di)) {
1536  		/* Cannot fix invalid inode block */
1537  		return -OCFS2_FILECHECK_ERR_INVALIDINO;
1538  	}
1539  
1540  	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1541  		/* Cannot just add VALID_FL flag back as a fix,
1542  		 * need more things to check here.
1543  		 */
1544  		return -OCFS2_FILECHECK_ERR_VALIDFLAG;
1545  	}
1546  
1547  	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1548  		di->i_blkno = cpu_to_le64(bh->b_blocknr);
1549  		changed = 1;
1550  		mlog(ML_ERROR,
1551  		     "Filecheck: reset dinode #%llu: i_blkno to %llu\n",
1552  		     (unsigned long long)bh->b_blocknr,
1553  		     (unsigned long long)le64_to_cpu(di->i_blkno));
1554  	}
1555  
1556  	if (le32_to_cpu(di->i_fs_generation) !=
1557  	    OCFS2_SB(sb)->fs_generation) {
1558  		di->i_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1559  		changed = 1;
1560  		mlog(ML_ERROR,
1561  		     "Filecheck: reset dinode #%llu: fs_generation to %u\n",
1562  		     (unsigned long long)bh->b_blocknr,
1563  		     le32_to_cpu(di->i_fs_generation));
1564  	}
1565  
1566  	if (changed || ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check)) {
1567  		ocfs2_compute_meta_ecc(sb, bh->b_data, &di->i_check);
1568  		mark_buffer_dirty(bh);
1569  		mlog(ML_ERROR,
1570  		     "Filecheck: reset dinode #%llu: compute meta ecc\n",
1571  		     (unsigned long long)bh->b_blocknr);
1572  	}
1573  
1574  	return 0;
1575  }
1576  
1577  static int
1578  ocfs2_filecheck_read_inode_block_full(struct inode *inode,
1579  				      struct buffer_head **bh,
1580  				      int flags, int type)
1581  {
1582  	int rc;
1583  	struct buffer_head *tmp = *bh;
1584  
1585  	if (!type) /* Check inode block */
1586  		rc = ocfs2_read_blocks(INODE_CACHE(inode),
1587  				OCFS2_I(inode)->ip_blkno,
1588  				1, &tmp, flags,
1589  				ocfs2_filecheck_validate_inode_block);
1590  	else /* Repair inode block */
1591  		rc = ocfs2_read_blocks(INODE_CACHE(inode),
1592  				OCFS2_I(inode)->ip_blkno,
1593  				1, &tmp, flags,
1594  				ocfs2_filecheck_repair_inode_block);
1595  
1596  	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1597  	if (!rc && !*bh)
1598  		*bh = tmp;
1599  
1600  	return rc;
1601  }
1602  
1603  int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1604  				int flags)
1605  {
1606  	int rc;
1607  	struct buffer_head *tmp = *bh;
1608  
1609  	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1610  			       1, &tmp, flags, ocfs2_validate_inode_block);
1611  
1612  	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
1613  	if (!rc && !*bh)
1614  		*bh = tmp;
1615  
1616  	return rc;
1617  }
1618  
1619  int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1620  {
1621  	return ocfs2_read_inode_block_full(inode, bh, 0);
1622  }
1623  
1624  
1625  static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1626  {
1627  	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1628  
1629  	return oi->ip_blkno;
1630  }
1631  
1632  static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1633  {
1634  	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1635  
1636  	return oi->vfs_inode.i_sb;
1637  }
1638  
1639  static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1640  {
1641  	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1642  
1643  	spin_lock(&oi->ip_lock);
1644  }
1645  
1646  static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1647  {
1648  	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1649  
1650  	spin_unlock(&oi->ip_lock);
1651  }
1652  
1653  static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1654  {
1655  	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1656  
1657  	mutex_lock(&oi->ip_io_mutex);
1658  }
1659  
1660  static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1661  {
1662  	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1663  
1664  	mutex_unlock(&oi->ip_io_mutex);
1665  }
1666  
1667  const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1668  	.co_owner		= ocfs2_inode_cache_owner,
1669  	.co_get_super		= ocfs2_inode_cache_get_super,
1670  	.co_cache_lock		= ocfs2_inode_cache_lock,
1671  	.co_cache_unlock	= ocfs2_inode_cache_unlock,
1672  	.co_io_lock		= ocfs2_inode_cache_io_lock,
1673  	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
1674  };
1675  
1676