xref: /openbmc/linux/fs/xfs/xfs_inode_item.c (revision 7dd65feb)
1 /*
2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_buf_item.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_trans_priv.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_btree.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_rw.h"
43 #include "xfs_error.h"
44 #include "xfs_trace.h"
45 
46 
47 kmem_zone_t	*xfs_ili_zone;		/* inode log item zone */
48 
49 /*
50  * This returns the number of iovecs needed to log the given inode item.
51  *
52  * We need one iovec for the inode log format structure, one for the
53  * inode core, and possibly one for the inode data/extents/b-tree root
54  * and one for the inode attribute data/extents/b-tree root.
55  */
56 STATIC uint
57 xfs_inode_item_size(
58 	xfs_inode_log_item_t	*iip)
59 {
60 	uint		nvecs;
61 	xfs_inode_t	*ip;
62 
63 	ip = iip->ili_inode;
64 	nvecs = 2;
65 
66 	/*
67 	 * Only log the data/extents/b-tree root if there is something
68 	 * left to log.
69 	 */
70 	iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
71 
72 	switch (ip->i_d.di_format) {
73 	case XFS_DINODE_FMT_EXTENTS:
74 		iip->ili_format.ilf_fields &=
75 			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
76 			  XFS_ILOG_DEV | XFS_ILOG_UUID);
77 		if ((iip->ili_format.ilf_fields & XFS_ILOG_DEXT) &&
78 		    (ip->i_d.di_nextents > 0) &&
79 		    (ip->i_df.if_bytes > 0)) {
80 			ASSERT(ip->i_df.if_u1.if_extents != NULL);
81 			nvecs++;
82 		} else {
83 			iip->ili_format.ilf_fields &= ~XFS_ILOG_DEXT;
84 		}
85 		break;
86 
87 	case XFS_DINODE_FMT_BTREE:
88 		ASSERT(ip->i_df.if_ext_max ==
89 		       XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
90 		iip->ili_format.ilf_fields &=
91 			~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
92 			  XFS_ILOG_DEV | XFS_ILOG_UUID);
93 		if ((iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) &&
94 		    (ip->i_df.if_broot_bytes > 0)) {
95 			ASSERT(ip->i_df.if_broot != NULL);
96 			nvecs++;
97 		} else {
98 			ASSERT(!(iip->ili_format.ilf_fields &
99 				 XFS_ILOG_DBROOT));
100 #ifdef XFS_TRANS_DEBUG
101 			if (iip->ili_root_size > 0) {
102 				ASSERT(iip->ili_root_size ==
103 				       ip->i_df.if_broot_bytes);
104 				ASSERT(memcmp(iip->ili_orig_root,
105 					    ip->i_df.if_broot,
106 					    iip->ili_root_size) == 0);
107 			} else {
108 				ASSERT(ip->i_df.if_broot_bytes == 0);
109 			}
110 #endif
111 			iip->ili_format.ilf_fields &= ~XFS_ILOG_DBROOT;
112 		}
113 		break;
114 
115 	case XFS_DINODE_FMT_LOCAL:
116 		iip->ili_format.ilf_fields &=
117 			~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
118 			  XFS_ILOG_DEV | XFS_ILOG_UUID);
119 		if ((iip->ili_format.ilf_fields & XFS_ILOG_DDATA) &&
120 		    (ip->i_df.if_bytes > 0)) {
121 			ASSERT(ip->i_df.if_u1.if_data != NULL);
122 			ASSERT(ip->i_d.di_size > 0);
123 			nvecs++;
124 		} else {
125 			iip->ili_format.ilf_fields &= ~XFS_ILOG_DDATA;
126 		}
127 		break;
128 
129 	case XFS_DINODE_FMT_DEV:
130 		iip->ili_format.ilf_fields &=
131 			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
132 			  XFS_ILOG_DEXT | XFS_ILOG_UUID);
133 		break;
134 
135 	case XFS_DINODE_FMT_UUID:
136 		iip->ili_format.ilf_fields &=
137 			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
138 			  XFS_ILOG_DEXT | XFS_ILOG_DEV);
139 		break;
140 
141 	default:
142 		ASSERT(0);
143 		break;
144 	}
145 
146 	/*
147 	 * If there are no attributes associated with this file,
148 	 * then there cannot be anything more to log.
149 	 * Clear all attribute-related log flags.
150 	 */
151 	if (!XFS_IFORK_Q(ip)) {
152 		iip->ili_format.ilf_fields &=
153 			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
154 		return nvecs;
155 	}
156 
157 	/*
158 	 * Log any necessary attribute data.
159 	 */
160 	switch (ip->i_d.di_aformat) {
161 	case XFS_DINODE_FMT_EXTENTS:
162 		iip->ili_format.ilf_fields &=
163 			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
164 		if ((iip->ili_format.ilf_fields & XFS_ILOG_AEXT) &&
165 		    (ip->i_d.di_anextents > 0) &&
166 		    (ip->i_afp->if_bytes > 0)) {
167 			ASSERT(ip->i_afp->if_u1.if_extents != NULL);
168 			nvecs++;
169 		} else {
170 			iip->ili_format.ilf_fields &= ~XFS_ILOG_AEXT;
171 		}
172 		break;
173 
174 	case XFS_DINODE_FMT_BTREE:
175 		iip->ili_format.ilf_fields &=
176 			~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
177 		if ((iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) &&
178 		    (ip->i_afp->if_broot_bytes > 0)) {
179 			ASSERT(ip->i_afp->if_broot != NULL);
180 			nvecs++;
181 		} else {
182 			iip->ili_format.ilf_fields &= ~XFS_ILOG_ABROOT;
183 		}
184 		break;
185 
186 	case XFS_DINODE_FMT_LOCAL:
187 		iip->ili_format.ilf_fields &=
188 			~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
189 		if ((iip->ili_format.ilf_fields & XFS_ILOG_ADATA) &&
190 		    (ip->i_afp->if_bytes > 0)) {
191 			ASSERT(ip->i_afp->if_u1.if_data != NULL);
192 			nvecs++;
193 		} else {
194 			iip->ili_format.ilf_fields &= ~XFS_ILOG_ADATA;
195 		}
196 		break;
197 
198 	default:
199 		ASSERT(0);
200 		break;
201 	}
202 
203 	return nvecs;
204 }
205 
206 /*
207  * This is called to fill in the vector of log iovecs for the
208  * given inode log item.  It fills the first item with an inode
209  * log format structure, the second with the on-disk inode structure,
210  * and a possible third and/or fourth with the inode data/extents/b-tree
211  * root and inode attributes data/extents/b-tree root.
212  */
213 STATIC void
214 xfs_inode_item_format(
215 	xfs_inode_log_item_t	*iip,
216 	xfs_log_iovec_t		*log_vector)
217 {
218 	uint			nvecs;
219 	xfs_log_iovec_t		*vecp;
220 	xfs_inode_t		*ip;
221 	size_t			data_bytes;
222 	xfs_bmbt_rec_t		*ext_buffer;
223 	int			nrecs;
224 	xfs_mount_t		*mp;
225 
226 	ip = iip->ili_inode;
227 	vecp = log_vector;
228 
229 	vecp->i_addr = (xfs_caddr_t)&iip->ili_format;
230 	vecp->i_len  = sizeof(xfs_inode_log_format_t);
231 	XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_IFORMAT);
232 	vecp++;
233 	nvecs	     = 1;
234 
235 	/*
236 	 * Make sure the linux inode is dirty. We do this before
237 	 * clearing i_update_core as the VFS will call back into
238 	 * XFS here and set i_update_core, so we need to dirty the
239 	 * inode first so that the ordering of i_update_core and
240 	 * unlogged modifications still works as described below.
241 	 */
242 	xfs_mark_inode_dirty_sync(ip);
243 
244 	/*
245 	 * Clear i_update_core if the timestamps (or any other
246 	 * non-transactional modification) need flushing/logging
247 	 * and we're about to log them with the rest of the core.
248 	 *
249 	 * This is the same logic as xfs_iflush() but this code can't
250 	 * run at the same time as xfs_iflush because we're in commit
251 	 * processing here and so we have the inode lock held in
252 	 * exclusive mode.  Although it doesn't really matter
253 	 * for the timestamps if both routines were to grab the
254 	 * timestamps or not.  That would be ok.
255 	 *
256 	 * We clear i_update_core before copying out the data.
257 	 * This is for coordination with our timestamp updates
258 	 * that don't hold the inode lock. They will always
259 	 * update the timestamps BEFORE setting i_update_core,
260 	 * so if we clear i_update_core after they set it we
261 	 * are guaranteed to see their updates to the timestamps
262 	 * either here.  Likewise, if they set it after we clear it
263 	 * here, we'll see it either on the next commit of this
264 	 * inode or the next time the inode gets flushed via
265 	 * xfs_iflush().  This depends on strongly ordered memory
266 	 * semantics, but we have that.  We use the SYNCHRONIZE
267 	 * macro to make sure that the compiler does not reorder
268 	 * the i_update_core access below the data copy below.
269 	 */
270 	if (ip->i_update_core)  {
271 		ip->i_update_core = 0;
272 		SYNCHRONIZE();
273 	}
274 
275 	/*
276 	 * Make sure to get the latest timestamps from the Linux inode.
277 	 */
278 	xfs_synchronize_times(ip);
279 
280 	vecp->i_addr = (xfs_caddr_t)&ip->i_d;
281 	vecp->i_len  = sizeof(struct xfs_icdinode);
282 	XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_ICORE);
283 	vecp++;
284 	nvecs++;
285 	iip->ili_format.ilf_fields |= XFS_ILOG_CORE;
286 
287 	/*
288 	 * If this is really an old format inode, then we need to
289 	 * log it as such.  This means that we have to copy the link
290 	 * count from the new field to the old.  We don't have to worry
291 	 * about the new fields, because nothing trusts them as long as
292 	 * the old inode version number is there.  If the superblock already
293 	 * has a new version number, then we don't bother converting back.
294 	 */
295 	mp = ip->i_mount;
296 	ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
297 	if (ip->i_d.di_version == 1) {
298 		if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
299 			/*
300 			 * Convert it back.
301 			 */
302 			ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
303 			ip->i_d.di_onlink = ip->i_d.di_nlink;
304 		} else {
305 			/*
306 			 * The superblock version has already been bumped,
307 			 * so just make the conversion to the new inode
308 			 * format permanent.
309 			 */
310 			ip->i_d.di_version = 2;
311 			ip->i_d.di_onlink = 0;
312 			memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
313 		}
314 	}
315 
316 	switch (ip->i_d.di_format) {
317 	case XFS_DINODE_FMT_EXTENTS:
318 		ASSERT(!(iip->ili_format.ilf_fields &
319 			 (XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
320 			  XFS_ILOG_DEV | XFS_ILOG_UUID)));
321 		if (iip->ili_format.ilf_fields & XFS_ILOG_DEXT) {
322 			ASSERT(ip->i_df.if_bytes > 0);
323 			ASSERT(ip->i_df.if_u1.if_extents != NULL);
324 			ASSERT(ip->i_d.di_nextents > 0);
325 			ASSERT(iip->ili_extents_buf == NULL);
326 			nrecs = ip->i_df.if_bytes /
327 				(uint)sizeof(xfs_bmbt_rec_t);
328 			ASSERT(nrecs > 0);
329 #ifdef XFS_NATIVE_HOST
330 			if (nrecs == ip->i_d.di_nextents) {
331 				/*
332 				 * There are no delayed allocation
333 				 * extents, so just point to the
334 				 * real extents array.
335 				 */
336 				vecp->i_addr =
337 					(char *)(ip->i_df.if_u1.if_extents);
338 				vecp->i_len = ip->i_df.if_bytes;
339 				XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_IEXT);
340 			} else
341 #endif
342 			{
343 				/*
344 				 * There are delayed allocation extents
345 				 * in the inode, or we need to convert
346 				 * the extents to on disk format.
347 				 * Use xfs_iextents_copy()
348 				 * to copy only the real extents into
349 				 * a separate buffer.  We'll free the
350 				 * buffer in the unlock routine.
351 				 */
352 				ext_buffer = kmem_alloc(ip->i_df.if_bytes,
353 					KM_SLEEP);
354 				iip->ili_extents_buf = ext_buffer;
355 				vecp->i_addr = (xfs_caddr_t)ext_buffer;
356 				vecp->i_len = xfs_iextents_copy(ip, ext_buffer,
357 						XFS_DATA_FORK);
358 				XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_IEXT);
359 			}
360 			ASSERT(vecp->i_len <= ip->i_df.if_bytes);
361 			iip->ili_format.ilf_dsize = vecp->i_len;
362 			vecp++;
363 			nvecs++;
364 		}
365 		break;
366 
367 	case XFS_DINODE_FMT_BTREE:
368 		ASSERT(!(iip->ili_format.ilf_fields &
369 			 (XFS_ILOG_DDATA | XFS_ILOG_DEXT |
370 			  XFS_ILOG_DEV | XFS_ILOG_UUID)));
371 		if (iip->ili_format.ilf_fields & XFS_ILOG_DBROOT) {
372 			ASSERT(ip->i_df.if_broot_bytes > 0);
373 			ASSERT(ip->i_df.if_broot != NULL);
374 			vecp->i_addr = (xfs_caddr_t)ip->i_df.if_broot;
375 			vecp->i_len = ip->i_df.if_broot_bytes;
376 			XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_IBROOT);
377 			vecp++;
378 			nvecs++;
379 			iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
380 		}
381 		break;
382 
383 	case XFS_DINODE_FMT_LOCAL:
384 		ASSERT(!(iip->ili_format.ilf_fields &
385 			 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
386 			  XFS_ILOG_DEV | XFS_ILOG_UUID)));
387 		if (iip->ili_format.ilf_fields & XFS_ILOG_DDATA) {
388 			ASSERT(ip->i_df.if_bytes > 0);
389 			ASSERT(ip->i_df.if_u1.if_data != NULL);
390 			ASSERT(ip->i_d.di_size > 0);
391 
392 			vecp->i_addr = (xfs_caddr_t)ip->i_df.if_u1.if_data;
393 			/*
394 			 * Round i_bytes up to a word boundary.
395 			 * The underlying memory is guaranteed to
396 			 * to be there by xfs_idata_realloc().
397 			 */
398 			data_bytes = roundup(ip->i_df.if_bytes, 4);
399 			ASSERT((ip->i_df.if_real_bytes == 0) ||
400 			       (ip->i_df.if_real_bytes == data_bytes));
401 			vecp->i_len = (int)data_bytes;
402 			XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_ILOCAL);
403 			vecp++;
404 			nvecs++;
405 			iip->ili_format.ilf_dsize = (unsigned)data_bytes;
406 		}
407 		break;
408 
409 	case XFS_DINODE_FMT_DEV:
410 		ASSERT(!(iip->ili_format.ilf_fields &
411 			 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
412 			  XFS_ILOG_DDATA | XFS_ILOG_UUID)));
413 		if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
414 			iip->ili_format.ilf_u.ilfu_rdev =
415 				ip->i_df.if_u2.if_rdev;
416 		}
417 		break;
418 
419 	case XFS_DINODE_FMT_UUID:
420 		ASSERT(!(iip->ili_format.ilf_fields &
421 			 (XFS_ILOG_DBROOT | XFS_ILOG_DEXT |
422 			  XFS_ILOG_DDATA | XFS_ILOG_DEV)));
423 		if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
424 			iip->ili_format.ilf_u.ilfu_uuid =
425 				ip->i_df.if_u2.if_uuid;
426 		}
427 		break;
428 
429 	default:
430 		ASSERT(0);
431 		break;
432 	}
433 
434 	/*
435 	 * If there are no attributes associated with the file,
436 	 * then we're done.
437 	 * Assert that no attribute-related log flags are set.
438 	 */
439 	if (!XFS_IFORK_Q(ip)) {
440 		ASSERT(nvecs == iip->ili_item.li_desc->lid_size);
441 		iip->ili_format.ilf_size = nvecs;
442 		ASSERT(!(iip->ili_format.ilf_fields &
443 			 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
444 		return;
445 	}
446 
447 	switch (ip->i_d.di_aformat) {
448 	case XFS_DINODE_FMT_EXTENTS:
449 		ASSERT(!(iip->ili_format.ilf_fields &
450 			 (XFS_ILOG_ADATA | XFS_ILOG_ABROOT)));
451 		if (iip->ili_format.ilf_fields & XFS_ILOG_AEXT) {
452 			ASSERT(ip->i_afp->if_bytes > 0);
453 			ASSERT(ip->i_afp->if_u1.if_extents != NULL);
454 			ASSERT(ip->i_d.di_anextents > 0);
455 #ifdef DEBUG
456 			nrecs = ip->i_afp->if_bytes /
457 				(uint)sizeof(xfs_bmbt_rec_t);
458 #endif
459 			ASSERT(nrecs > 0);
460 			ASSERT(nrecs == ip->i_d.di_anextents);
461 #ifdef XFS_NATIVE_HOST
462 			/*
463 			 * There are not delayed allocation extents
464 			 * for attributes, so just point at the array.
465 			 */
466 			vecp->i_addr = (char *)(ip->i_afp->if_u1.if_extents);
467 			vecp->i_len = ip->i_afp->if_bytes;
468 #else
469 			ASSERT(iip->ili_aextents_buf == NULL);
470 			/*
471 			 * Need to endian flip before logging
472 			 */
473 			ext_buffer = kmem_alloc(ip->i_afp->if_bytes,
474 				KM_SLEEP);
475 			iip->ili_aextents_buf = ext_buffer;
476 			vecp->i_addr = (xfs_caddr_t)ext_buffer;
477 			vecp->i_len = xfs_iextents_copy(ip, ext_buffer,
478 					XFS_ATTR_FORK);
479 #endif
480 			XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_IATTR_EXT);
481 			iip->ili_format.ilf_asize = vecp->i_len;
482 			vecp++;
483 			nvecs++;
484 		}
485 		break;
486 
487 	case XFS_DINODE_FMT_BTREE:
488 		ASSERT(!(iip->ili_format.ilf_fields &
489 			 (XFS_ILOG_ADATA | XFS_ILOG_AEXT)));
490 		if (iip->ili_format.ilf_fields & XFS_ILOG_ABROOT) {
491 			ASSERT(ip->i_afp->if_broot_bytes > 0);
492 			ASSERT(ip->i_afp->if_broot != NULL);
493 			vecp->i_addr = (xfs_caddr_t)ip->i_afp->if_broot;
494 			vecp->i_len = ip->i_afp->if_broot_bytes;
495 			XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_IATTR_BROOT);
496 			vecp++;
497 			nvecs++;
498 			iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
499 		}
500 		break;
501 
502 	case XFS_DINODE_FMT_LOCAL:
503 		ASSERT(!(iip->ili_format.ilf_fields &
504 			 (XFS_ILOG_ABROOT | XFS_ILOG_AEXT)));
505 		if (iip->ili_format.ilf_fields & XFS_ILOG_ADATA) {
506 			ASSERT(ip->i_afp->if_bytes > 0);
507 			ASSERT(ip->i_afp->if_u1.if_data != NULL);
508 
509 			vecp->i_addr = (xfs_caddr_t)ip->i_afp->if_u1.if_data;
510 			/*
511 			 * Round i_bytes up to a word boundary.
512 			 * The underlying memory is guaranteed to
513 			 * to be there by xfs_idata_realloc().
514 			 */
515 			data_bytes = roundup(ip->i_afp->if_bytes, 4);
516 			ASSERT((ip->i_afp->if_real_bytes == 0) ||
517 			       (ip->i_afp->if_real_bytes == data_bytes));
518 			vecp->i_len = (int)data_bytes;
519 			XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_IATTR_LOCAL);
520 			vecp++;
521 			nvecs++;
522 			iip->ili_format.ilf_asize = (unsigned)data_bytes;
523 		}
524 		break;
525 
526 	default:
527 		ASSERT(0);
528 		break;
529 	}
530 
531 	ASSERT(nvecs == iip->ili_item.li_desc->lid_size);
532 	iip->ili_format.ilf_size = nvecs;
533 }
534 
535 
536 /*
537  * This is called to pin the inode associated with the inode log
538  * item in memory so it cannot be written out.  Do this by calling
539  * xfs_ipin() to bump the pin count in the inode while holding the
540  * inode pin lock.
541  */
542 STATIC void
543 xfs_inode_item_pin(
544 	xfs_inode_log_item_t	*iip)
545 {
546 	ASSERT(xfs_isilocked(iip->ili_inode, XFS_ILOCK_EXCL));
547 	xfs_ipin(iip->ili_inode);
548 }
549 
550 
551 /*
552  * This is called to unpin the inode associated with the inode log
553  * item which was previously pinned with a call to xfs_inode_item_pin().
554  * Just call xfs_iunpin() on the inode to do this.
555  */
556 /* ARGSUSED */
557 STATIC void
558 xfs_inode_item_unpin(
559 	xfs_inode_log_item_t	*iip,
560 	int			stale)
561 {
562 	xfs_iunpin(iip->ili_inode);
563 }
564 
565 /* ARGSUSED */
566 STATIC void
567 xfs_inode_item_unpin_remove(
568 	xfs_inode_log_item_t	*iip,
569 	xfs_trans_t		*tp)
570 {
571 	xfs_iunpin(iip->ili_inode);
572 }
573 
574 /*
575  * This is called to attempt to lock the inode associated with this
576  * inode log item, in preparation for the push routine which does the actual
577  * iflush.  Don't sleep on the inode lock or the flush lock.
578  *
579  * If the flush lock is already held, indicating that the inode has
580  * been or is in the process of being flushed, then (ideally) we'd like to
581  * see if the inode's buffer is still incore, and if so give it a nudge.
582  * We delay doing so until the pushbuf routine, though, to avoid holding
583  * the AIL lock across a call to the blackhole which is the buffer cache.
584  * Also we don't want to sleep in any device strategy routines, which can happen
585  * if we do the subsequent bawrite in here.
586  */
587 STATIC uint
588 xfs_inode_item_trylock(
589 	xfs_inode_log_item_t	*iip)
590 {
591 	register xfs_inode_t	*ip;
592 
593 	ip = iip->ili_inode;
594 
595 	if (xfs_ipincount(ip) > 0) {
596 		return XFS_ITEM_PINNED;
597 	}
598 
599 	if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
600 		return XFS_ITEM_LOCKED;
601 	}
602 
603 	if (!xfs_iflock_nowait(ip)) {
604 		/*
605 		 * If someone else isn't already trying to push the inode
606 		 * buffer, we get to do it.
607 		 */
608 		if (iip->ili_pushbuf_flag == 0) {
609 			iip->ili_pushbuf_flag = 1;
610 #ifdef DEBUG
611 			iip->ili_push_owner = current_pid();
612 #endif
613 			/*
614 			 * Inode is left locked in shared mode.
615 			 * Pushbuf routine gets to unlock it.
616 			 */
617 			return XFS_ITEM_PUSHBUF;
618 		} else {
619 			/*
620 			 * We hold the AIL lock, so we must specify the
621 			 * NONOTIFY flag so that we won't double trip.
622 			 */
623 			xfs_iunlock(ip, XFS_ILOCK_SHARED|XFS_IUNLOCK_NONOTIFY);
624 			return XFS_ITEM_FLUSHING;
625 		}
626 		/* NOTREACHED */
627 	}
628 
629 	/* Stale items should force out the iclog */
630 	if (ip->i_flags & XFS_ISTALE) {
631 		xfs_ifunlock(ip);
632 		xfs_iunlock(ip, XFS_ILOCK_SHARED|XFS_IUNLOCK_NONOTIFY);
633 		return XFS_ITEM_PINNED;
634 	}
635 
636 #ifdef DEBUG
637 	if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
638 		ASSERT(iip->ili_format.ilf_fields != 0);
639 		ASSERT(iip->ili_logged == 0);
640 		ASSERT(iip->ili_item.li_flags & XFS_LI_IN_AIL);
641 	}
642 #endif
643 	return XFS_ITEM_SUCCESS;
644 }
645 
646 /*
647  * Unlock the inode associated with the inode log item.
648  * Clear the fields of the inode and inode log item that
649  * are specific to the current transaction.  If the
650  * hold flags is set, do not unlock the inode.
651  */
652 STATIC void
653 xfs_inode_item_unlock(
654 	xfs_inode_log_item_t	*iip)
655 {
656 	uint		hold;
657 	uint		iolocked;
658 	uint		lock_flags;
659 	xfs_inode_t	*ip;
660 
661 	ASSERT(iip != NULL);
662 	ASSERT(iip->ili_inode->i_itemp != NULL);
663 	ASSERT(xfs_isilocked(iip->ili_inode, XFS_ILOCK_EXCL));
664 	ASSERT((!(iip->ili_inode->i_itemp->ili_flags &
665 		  XFS_ILI_IOLOCKED_EXCL)) ||
666 	       xfs_isilocked(iip->ili_inode, XFS_IOLOCK_EXCL));
667 	ASSERT((!(iip->ili_inode->i_itemp->ili_flags &
668 		  XFS_ILI_IOLOCKED_SHARED)) ||
669 	       xfs_isilocked(iip->ili_inode, XFS_IOLOCK_SHARED));
670 	/*
671 	 * Clear the transaction pointer in the inode.
672 	 */
673 	ip = iip->ili_inode;
674 	ip->i_transp = NULL;
675 
676 	/*
677 	 * If the inode needed a separate buffer with which to log
678 	 * its extents, then free it now.
679 	 */
680 	if (iip->ili_extents_buf != NULL) {
681 		ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
682 		ASSERT(ip->i_d.di_nextents > 0);
683 		ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_DEXT);
684 		ASSERT(ip->i_df.if_bytes > 0);
685 		kmem_free(iip->ili_extents_buf);
686 		iip->ili_extents_buf = NULL;
687 	}
688 	if (iip->ili_aextents_buf != NULL) {
689 		ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
690 		ASSERT(ip->i_d.di_anextents > 0);
691 		ASSERT(iip->ili_format.ilf_fields & XFS_ILOG_AEXT);
692 		ASSERT(ip->i_afp->if_bytes > 0);
693 		kmem_free(iip->ili_aextents_buf);
694 		iip->ili_aextents_buf = NULL;
695 	}
696 
697 	/*
698 	 * Figure out if we should unlock the inode or not.
699 	 */
700 	hold = iip->ili_flags & XFS_ILI_HOLD;
701 
702 	/*
703 	 * Before clearing out the flags, remember whether we
704 	 * are holding the inode's IO lock.
705 	 */
706 	iolocked = iip->ili_flags & XFS_ILI_IOLOCKED_ANY;
707 
708 	/*
709 	 * Clear out the fields of the inode log item particular
710 	 * to the current transaction.
711 	 */
712 	iip->ili_flags = 0;
713 
714 	/*
715 	 * Unlock the inode if XFS_ILI_HOLD was not set.
716 	 */
717 	if (!hold) {
718 		lock_flags = XFS_ILOCK_EXCL;
719 		if (iolocked & XFS_ILI_IOLOCKED_EXCL) {
720 			lock_flags |= XFS_IOLOCK_EXCL;
721 		} else if (iolocked & XFS_ILI_IOLOCKED_SHARED) {
722 			lock_flags |= XFS_IOLOCK_SHARED;
723 		}
724 		xfs_iput(iip->ili_inode, lock_flags);
725 	}
726 }
727 
728 /*
729  * This is called to find out where the oldest active copy of the
730  * inode log item in the on disk log resides now that the last log
731  * write of it completed at the given lsn.  Since we always re-log
732  * all dirty data in an inode, the latest copy in the on disk log
733  * is the only one that matters.  Therefore, simply return the
734  * given lsn.
735  */
736 /*ARGSUSED*/
737 STATIC xfs_lsn_t
738 xfs_inode_item_committed(
739 	xfs_inode_log_item_t	*iip,
740 	xfs_lsn_t		lsn)
741 {
742 	return (lsn);
743 }
744 
745 /*
746  * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
747  * failed to get the inode flush lock but did get the inode locked SHARED.
748  * Here we're trying to see if the inode buffer is incore, and if so whether it's
749  * marked delayed write. If that's the case, we'll initiate a bawrite on that
750  * buffer to expedite the process.
751  *
752  * We aren't holding the AIL lock (or the flush lock) when this gets called,
753  * so it is inherently race-y.
754  */
755 STATIC void
756 xfs_inode_item_pushbuf(
757 	xfs_inode_log_item_t	*iip)
758 {
759 	xfs_inode_t	*ip;
760 	xfs_mount_t	*mp;
761 	xfs_buf_t	*bp;
762 	uint		dopush;
763 
764 	ip = iip->ili_inode;
765 
766 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
767 
768 	/*
769 	 * The ili_pushbuf_flag keeps others from
770 	 * trying to duplicate our effort.
771 	 */
772 	ASSERT(iip->ili_pushbuf_flag != 0);
773 	ASSERT(iip->ili_push_owner == current_pid());
774 
775 	/*
776 	 * If a flush is not in progress anymore, chances are that the
777 	 * inode was taken off the AIL. So, just get out.
778 	 */
779 	if (completion_done(&ip->i_flush) ||
780 	    ((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0)) {
781 		iip->ili_pushbuf_flag = 0;
782 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
783 		return;
784 	}
785 
786 	mp = ip->i_mount;
787 	bp = xfs_incore(mp->m_ddev_targp, iip->ili_format.ilf_blkno,
788 		    iip->ili_format.ilf_len, XFS_INCORE_TRYLOCK);
789 
790 	if (bp != NULL) {
791 		if (XFS_BUF_ISDELAYWRITE(bp)) {
792 			/*
793 			 * We were racing with iflush because we don't hold
794 			 * the AIL lock or the flush lock. However, at this point,
795 			 * we have the buffer, and we know that it's dirty.
796 			 * So, it's possible that iflush raced with us, and
797 			 * this item is already taken off the AIL.
798 			 * If not, we can flush it async.
799 			 */
800 			dopush = ((iip->ili_item.li_flags & XFS_LI_IN_AIL) &&
801 				  !completion_done(&ip->i_flush));
802 			iip->ili_pushbuf_flag = 0;
803 			xfs_iunlock(ip, XFS_ILOCK_SHARED);
804 
805 			trace_xfs_inode_item_push(bp, _RET_IP_);
806 
807 			if (XFS_BUF_ISPINNED(bp)) {
808 				xfs_log_force(mp, (xfs_lsn_t)0,
809 					      XFS_LOG_FORCE);
810 			}
811 			if (dopush) {
812 				int	error;
813 				error = xfs_bawrite(mp, bp);
814 				if (error)
815 					xfs_fs_cmn_err(CE_WARN, mp,
816 		"xfs_inode_item_pushbuf: pushbuf error %d on iip %p, bp %p",
817 							error, iip, bp);
818 			} else {
819 				xfs_buf_relse(bp);
820 			}
821 		} else {
822 			iip->ili_pushbuf_flag = 0;
823 			xfs_iunlock(ip, XFS_ILOCK_SHARED);
824 			xfs_buf_relse(bp);
825 		}
826 		return;
827 	}
828 	/*
829 	 * We have to be careful about resetting pushbuf flag too early (above).
830 	 * Even though in theory we can do it as soon as we have the buflock,
831 	 * we don't want others to be doing work needlessly. They'll come to
832 	 * this function thinking that pushing the buffer is their
833 	 * responsibility only to find that the buffer is still locked by
834 	 * another doing the same thing
835 	 */
836 	iip->ili_pushbuf_flag = 0;
837 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
838 	return;
839 }
840 
841 
842 /*
843  * This is called to asynchronously write the inode associated with this
844  * inode log item out to disk. The inode will already have been locked by
845  * a successful call to xfs_inode_item_trylock().
846  */
847 STATIC void
848 xfs_inode_item_push(
849 	xfs_inode_log_item_t	*iip)
850 {
851 	xfs_inode_t	*ip;
852 
853 	ip = iip->ili_inode;
854 
855 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
856 	ASSERT(!completion_done(&ip->i_flush));
857 	/*
858 	 * Since we were able to lock the inode's flush lock and
859 	 * we found it on the AIL, the inode must be dirty.  This
860 	 * is because the inode is removed from the AIL while still
861 	 * holding the flush lock in xfs_iflush_done().  Thus, if
862 	 * we found it in the AIL and were able to obtain the flush
863 	 * lock without sleeping, then there must not have been
864 	 * anyone in the process of flushing the inode.
865 	 */
866 	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
867 	       iip->ili_format.ilf_fields != 0);
868 
869 	/*
870 	 * Write out the inode.  The completion routine ('iflush_done') will
871 	 * pull it from the AIL, mark it clean, unlock the flush lock.
872 	 */
873 	(void) xfs_iflush(ip, XFS_IFLUSH_ASYNC);
874 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
875 
876 	return;
877 }
878 
879 /*
880  * XXX rcc - this one really has to do something.  Probably needs
881  * to stamp in a new field in the incore inode.
882  */
883 /* ARGSUSED */
884 STATIC void
885 xfs_inode_item_committing(
886 	xfs_inode_log_item_t	*iip,
887 	xfs_lsn_t		lsn)
888 {
889 	iip->ili_last_lsn = lsn;
890 	return;
891 }
892 
893 /*
894  * This is the ops vector shared by all buf log items.
895  */
896 static struct xfs_item_ops xfs_inode_item_ops = {
897 	.iop_size	= (uint(*)(xfs_log_item_t*))xfs_inode_item_size,
898 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
899 					xfs_inode_item_format,
900 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_inode_item_pin,
901 	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_inode_item_unpin,
902 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
903 					xfs_inode_item_unpin_remove,
904 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_inode_item_trylock,
905 	.iop_unlock	= (void(*)(xfs_log_item_t*))xfs_inode_item_unlock,
906 	.iop_committed	= (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
907 					xfs_inode_item_committed,
908 	.iop_push	= (void(*)(xfs_log_item_t*))xfs_inode_item_push,
909 	.iop_pushbuf	= (void(*)(xfs_log_item_t*))xfs_inode_item_pushbuf,
910 	.iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
911 					xfs_inode_item_committing
912 };
913 
914 
915 /*
916  * Initialize the inode log item for a newly allocated (in-core) inode.
917  */
918 void
919 xfs_inode_item_init(
920 	xfs_inode_t	*ip,
921 	xfs_mount_t	*mp)
922 {
923 	xfs_inode_log_item_t	*iip;
924 
925 	ASSERT(ip->i_itemp == NULL);
926 	iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
927 
928 	iip->ili_item.li_type = XFS_LI_INODE;
929 	iip->ili_item.li_ops = &xfs_inode_item_ops;
930 	iip->ili_item.li_mountp = mp;
931 	iip->ili_item.li_ailp = mp->m_ail;
932 	iip->ili_inode = ip;
933 
934 	/*
935 	   We have zeroed memory. No need ...
936 	   iip->ili_extents_buf = NULL;
937 	   iip->ili_pushbuf_flag = 0;
938 	 */
939 
940 	iip->ili_format.ilf_type = XFS_LI_INODE;
941 	iip->ili_format.ilf_ino = ip->i_ino;
942 	iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
943 	iip->ili_format.ilf_len = ip->i_imap.im_len;
944 	iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
945 }
946 
947 /*
948  * Free the inode log item and any memory hanging off of it.
949  */
950 void
951 xfs_inode_item_destroy(
952 	xfs_inode_t	*ip)
953 {
954 #ifdef XFS_TRANS_DEBUG
955 	if (ip->i_itemp->ili_root_size != 0) {
956 		kmem_free(ip->i_itemp->ili_orig_root);
957 	}
958 #endif
959 	kmem_zone_free(xfs_ili_zone, ip->i_itemp);
960 }
961 
962 
963 /*
964  * This is the inode flushing I/O completion routine.  It is called
965  * from interrupt level when the buffer containing the inode is
966  * flushed to disk.  It is responsible for removing the inode item
967  * from the AIL if it has not been re-logged, and unlocking the inode's
968  * flush lock.
969  */
970 /*ARGSUSED*/
971 void
972 xfs_iflush_done(
973 	xfs_buf_t		*bp,
974 	xfs_inode_log_item_t	*iip)
975 {
976 	xfs_inode_t		*ip = iip->ili_inode;
977 	struct xfs_ail		*ailp = iip->ili_item.li_ailp;
978 
979 	/*
980 	 * We only want to pull the item from the AIL if it is
981 	 * actually there and its location in the log has not
982 	 * changed since we started the flush.  Thus, we only bother
983 	 * if the ili_logged flag is set and the inode's lsn has not
984 	 * changed.  First we check the lsn outside
985 	 * the lock since it's cheaper, and then we recheck while
986 	 * holding the lock before removing the inode from the AIL.
987 	 */
988 	if (iip->ili_logged &&
989 	    (iip->ili_item.li_lsn == iip->ili_flush_lsn)) {
990 		spin_lock(&ailp->xa_lock);
991 		if (iip->ili_item.li_lsn == iip->ili_flush_lsn) {
992 			/* xfs_trans_ail_delete() drops the AIL lock. */
993 			xfs_trans_ail_delete(ailp, (xfs_log_item_t*)iip);
994 		} else {
995 			spin_unlock(&ailp->xa_lock);
996 		}
997 	}
998 
999 	iip->ili_logged = 0;
1000 
1001 	/*
1002 	 * Clear the ili_last_fields bits now that we know that the
1003 	 * data corresponding to them is safely on disk.
1004 	 */
1005 	iip->ili_last_fields = 0;
1006 
1007 	/*
1008 	 * Release the inode's flush lock since we're done with it.
1009 	 */
1010 	xfs_ifunlock(ip);
1011 
1012 	return;
1013 }
1014 
1015 /*
1016  * This is the inode flushing abort routine.  It is called
1017  * from xfs_iflush when the filesystem is shutting down to clean
1018  * up the inode state.
1019  * It is responsible for removing the inode item
1020  * from the AIL if it has not been re-logged, and unlocking the inode's
1021  * flush lock.
1022  */
1023 void
1024 xfs_iflush_abort(
1025 	xfs_inode_t		*ip)
1026 {
1027 	xfs_inode_log_item_t	*iip = ip->i_itemp;
1028 	xfs_mount_t		*mp;
1029 
1030 	iip = ip->i_itemp;
1031 	mp = ip->i_mount;
1032 	if (iip) {
1033 		struct xfs_ail	*ailp = iip->ili_item.li_ailp;
1034 		if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
1035 			spin_lock(&ailp->xa_lock);
1036 			if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
1037 				/* xfs_trans_ail_delete() drops the AIL lock. */
1038 				xfs_trans_ail_delete(ailp, (xfs_log_item_t *)iip);
1039 			} else
1040 				spin_unlock(&ailp->xa_lock);
1041 		}
1042 		iip->ili_logged = 0;
1043 		/*
1044 		 * Clear the ili_last_fields bits now that we know that the
1045 		 * data corresponding to them is safely on disk.
1046 		 */
1047 		iip->ili_last_fields = 0;
1048 		/*
1049 		 * Clear the inode logging fields so no more flushes are
1050 		 * attempted.
1051 		 */
1052 		iip->ili_format.ilf_fields = 0;
1053 	}
1054 	/*
1055 	 * Release the inode's flush lock since we're done with it.
1056 	 */
1057 	xfs_ifunlock(ip);
1058 }
1059 
1060 void
1061 xfs_istale_done(
1062 	xfs_buf_t		*bp,
1063 	xfs_inode_log_item_t	*iip)
1064 {
1065 	xfs_iflush_abort(iip->ili_inode);
1066 }
1067 
1068 /*
1069  * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
1070  * (which can have different field alignments) to the native version
1071  */
1072 int
1073 xfs_inode_item_format_convert(
1074 	xfs_log_iovec_t		*buf,
1075 	xfs_inode_log_format_t	*in_f)
1076 {
1077 	if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
1078 		xfs_inode_log_format_32_t *in_f32;
1079 
1080 		in_f32 = (xfs_inode_log_format_32_t *)buf->i_addr;
1081 		in_f->ilf_type = in_f32->ilf_type;
1082 		in_f->ilf_size = in_f32->ilf_size;
1083 		in_f->ilf_fields = in_f32->ilf_fields;
1084 		in_f->ilf_asize = in_f32->ilf_asize;
1085 		in_f->ilf_dsize = in_f32->ilf_dsize;
1086 		in_f->ilf_ino = in_f32->ilf_ino;
1087 		/* copy biggest field of ilf_u */
1088 		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1089 		       in_f32->ilf_u.ilfu_uuid.__u_bits,
1090 		       sizeof(uuid_t));
1091 		in_f->ilf_blkno = in_f32->ilf_blkno;
1092 		in_f->ilf_len = in_f32->ilf_len;
1093 		in_f->ilf_boffset = in_f32->ilf_boffset;
1094 		return 0;
1095 	} else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
1096 		xfs_inode_log_format_64_t *in_f64;
1097 
1098 		in_f64 = (xfs_inode_log_format_64_t *)buf->i_addr;
1099 		in_f->ilf_type = in_f64->ilf_type;
1100 		in_f->ilf_size = in_f64->ilf_size;
1101 		in_f->ilf_fields = in_f64->ilf_fields;
1102 		in_f->ilf_asize = in_f64->ilf_asize;
1103 		in_f->ilf_dsize = in_f64->ilf_dsize;
1104 		in_f->ilf_ino = in_f64->ilf_ino;
1105 		/* copy biggest field of ilf_u */
1106 		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
1107 		       in_f64->ilf_u.ilfu_uuid.__u_bits,
1108 		       sizeof(uuid_t));
1109 		in_f->ilf_blkno = in_f64->ilf_blkno;
1110 		in_f->ilf_len = in_f64->ilf_len;
1111 		in_f->ilf_boffset = in_f64->ilf_boffset;
1112 		return 0;
1113 	}
1114 	return EFSCORRUPTED;
1115 }
1116