Linux Audio

Check our new training course

Loading...
v4.17
  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_format.h"
 21#include "xfs_log_format.h"
 22#include "xfs_trans_resv.h"
 
 
 23#include "xfs_mount.h"
 24#include "xfs_inode.h"
 25#include "xfs_trans.h"
 26#include "xfs_inode_item.h"
 27#include "xfs_error.h"
 28#include "xfs_trace.h"
 29#include "xfs_trans_priv.h"
 30#include "xfs_buf_item.h"
 31#include "xfs_log.h"
 32
 33#include <linux/iversion.h>
 34
 35kmem_zone_t	*xfs_ili_zone;		/* inode log item zone */
 36
 37static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
 38{
 39	return container_of(lip, struct xfs_inode_log_item, ili_item);
 40}
 41
 42STATIC void
 43xfs_inode_item_data_fork_size(
 44	struct xfs_inode_log_item *iip,
 45	int			*nvecs,
 46	int			*nbytes)
 47{
 48	struct xfs_inode	*ip = iip->ili_inode;
 49
 50	switch (ip->i_d.di_format) {
 51	case XFS_DINODE_FMT_EXTENTS:
 52		if ((iip->ili_fields & XFS_ILOG_DEXT) &&
 53		    ip->i_d.di_nextents > 0 &&
 54		    ip->i_df.if_bytes > 0) {
 55			/* worst case, doesn't subtract delalloc extents */
 56			*nbytes += XFS_IFORK_DSIZE(ip);
 57			*nvecs += 1;
 58		}
 59		break;
 60	case XFS_DINODE_FMT_BTREE:
 61		if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 62		    ip->i_df.if_broot_bytes > 0) {
 63			*nbytes += ip->i_df.if_broot_bytes;
 64			*nvecs += 1;
 65		}
 66		break;
 67	case XFS_DINODE_FMT_LOCAL:
 68		if ((iip->ili_fields & XFS_ILOG_DDATA) &&
 69		    ip->i_df.if_bytes > 0) {
 70			*nbytes += roundup(ip->i_df.if_bytes, 4);
 71			*nvecs += 1;
 72		}
 73		break;
 74
 75	case XFS_DINODE_FMT_DEV:
 
 76		break;
 77	default:
 78		ASSERT(0);
 79		break;
 80	}
 81}
 82
 83STATIC void
 84xfs_inode_item_attr_fork_size(
 85	struct xfs_inode_log_item *iip,
 86	int			*nvecs,
 87	int			*nbytes)
 88{
 89	struct xfs_inode	*ip = iip->ili_inode;
 90
 91	switch (ip->i_d.di_aformat) {
 92	case XFS_DINODE_FMT_EXTENTS:
 93		if ((iip->ili_fields & XFS_ILOG_AEXT) &&
 94		    ip->i_d.di_anextents > 0 &&
 95		    ip->i_afp->if_bytes > 0) {
 96			/* worst case, doesn't subtract unused space */
 97			*nbytes += XFS_IFORK_ASIZE(ip);
 98			*nvecs += 1;
 99		}
100		break;
101	case XFS_DINODE_FMT_BTREE:
102		if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
103		    ip->i_afp->if_broot_bytes > 0) {
104			*nbytes += ip->i_afp->if_broot_bytes;
105			*nvecs += 1;
106		}
107		break;
108	case XFS_DINODE_FMT_LOCAL:
109		if ((iip->ili_fields & XFS_ILOG_ADATA) &&
110		    ip->i_afp->if_bytes > 0) {
111			*nbytes += roundup(ip->i_afp->if_bytes, 4);
112			*nvecs += 1;
113		}
114		break;
115	default:
116		ASSERT(0);
117		break;
118	}
119}
120
121/*
122 * This returns the number of iovecs needed to log the given inode item.
123 *
124 * We need one iovec for the inode log format structure, one for the
125 * inode core, and possibly one for the inode data/extents/b-tree root
126 * and one for the inode attribute data/extents/b-tree root.
127 */
128STATIC void
129xfs_inode_item_size(
130	struct xfs_log_item	*lip,
131	int			*nvecs,
132	int			*nbytes)
133{
134	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
135	struct xfs_inode	*ip = iip->ili_inode;
136
137	*nvecs += 2;
138	*nbytes += sizeof(struct xfs_inode_log_format) +
139		   xfs_log_dinode_size(ip->i_d.di_version);
140
141	xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
142	if (XFS_IFORK_Q(ip))
143		xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
144}
145
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
146STATIC void
147xfs_inode_item_format_data_fork(
148	struct xfs_inode_log_item *iip,
149	struct xfs_inode_log_format *ilf,
150	struct xfs_log_vec	*lv,
151	struct xfs_log_iovec	**vecp)
152{
153	struct xfs_inode	*ip = iip->ili_inode;
154	size_t			data_bytes;
155
156	switch (ip->i_d.di_format) {
157	case XFS_DINODE_FMT_EXTENTS:
158		iip->ili_fields &=
159			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
 
160
161		if ((iip->ili_fields & XFS_ILOG_DEXT) &&
162		    ip->i_d.di_nextents > 0 &&
163		    ip->i_df.if_bytes > 0) {
164			struct xfs_bmbt_rec *p;
165
166			ASSERT(xfs_iext_count(&ip->i_df) > 0);
 
167
168			p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
169			data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
170			xlog_finish_iovec(lv, *vecp, data_bytes);
171
172			ASSERT(data_bytes <= ip->i_df.if_bytes);
173
174			ilf->ilf_dsize = data_bytes;
175			ilf->ilf_size++;
176		} else {
177			iip->ili_fields &= ~XFS_ILOG_DEXT;
178		}
179		break;
180	case XFS_DINODE_FMT_BTREE:
181		iip->ili_fields &=
182			~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
 
183
184		if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
185		    ip->i_df.if_broot_bytes > 0) {
186			ASSERT(ip->i_df.if_broot != NULL);
187			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
188					ip->i_df.if_broot,
189					ip->i_df.if_broot_bytes);
190			ilf->ilf_dsize = ip->i_df.if_broot_bytes;
191			ilf->ilf_size++;
192		} else {
193			ASSERT(!(iip->ili_fields &
194				 XFS_ILOG_DBROOT));
195			iip->ili_fields &= ~XFS_ILOG_DBROOT;
196		}
197		break;
198	case XFS_DINODE_FMT_LOCAL:
199		iip->ili_fields &=
200			~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
 
201		if ((iip->ili_fields & XFS_ILOG_DDATA) &&
202		    ip->i_df.if_bytes > 0) {
203			/*
204			 * Round i_bytes up to a word boundary.
205			 * The underlying memory is guaranteed to
206			 * to be there by xfs_idata_realloc().
207			 */
208			data_bytes = roundup(ip->i_df.if_bytes, 4);
209			ASSERT(ip->i_df.if_real_bytes == 0 ||
210			       ip->i_df.if_real_bytes >= data_bytes);
211			ASSERT(ip->i_df.if_u1.if_data != NULL);
212			ASSERT(ip->i_d.di_size > 0);
213			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
214					ip->i_df.if_u1.if_data, data_bytes);
215			ilf->ilf_dsize = (unsigned)data_bytes;
216			ilf->ilf_size++;
217		} else {
218			iip->ili_fields &= ~XFS_ILOG_DDATA;
219		}
220		break;
221	case XFS_DINODE_FMT_DEV:
222		iip->ili_fields &=
223			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
 
224		if (iip->ili_fields & XFS_ILOG_DEV)
225			ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
 
 
 
 
 
 
 
226		break;
227	default:
228		ASSERT(0);
229		break;
230	}
231}
232
233STATIC void
234xfs_inode_item_format_attr_fork(
235	struct xfs_inode_log_item *iip,
236	struct xfs_inode_log_format *ilf,
237	struct xfs_log_vec	*lv,
238	struct xfs_log_iovec	**vecp)
239{
240	struct xfs_inode	*ip = iip->ili_inode;
241	size_t			data_bytes;
242
243	switch (ip->i_d.di_aformat) {
244	case XFS_DINODE_FMT_EXTENTS:
245		iip->ili_fields &=
246			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
247
248		if ((iip->ili_fields & XFS_ILOG_AEXT) &&
249		    ip->i_d.di_anextents > 0 &&
250		    ip->i_afp->if_bytes > 0) {
251			struct xfs_bmbt_rec *p;
252
253			ASSERT(xfs_iext_count(ip->i_afp) ==
254				ip->i_d.di_anextents);
 
255
256			p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
257			data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
258			xlog_finish_iovec(lv, *vecp, data_bytes);
259
260			ilf->ilf_asize = data_bytes;
261			ilf->ilf_size++;
262		} else {
263			iip->ili_fields &= ~XFS_ILOG_AEXT;
264		}
265		break;
266	case XFS_DINODE_FMT_BTREE:
267		iip->ili_fields &=
268			~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
269
270		if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
271		    ip->i_afp->if_broot_bytes > 0) {
272			ASSERT(ip->i_afp->if_broot != NULL);
273
274			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
275					ip->i_afp->if_broot,
276					ip->i_afp->if_broot_bytes);
277			ilf->ilf_asize = ip->i_afp->if_broot_bytes;
278			ilf->ilf_size++;
279		} else {
280			iip->ili_fields &= ~XFS_ILOG_ABROOT;
281		}
282		break;
283	case XFS_DINODE_FMT_LOCAL:
284		iip->ili_fields &=
285			~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
286
287		if ((iip->ili_fields & XFS_ILOG_ADATA) &&
288		    ip->i_afp->if_bytes > 0) {
289			/*
290			 * Round i_bytes up to a word boundary.
291			 * The underlying memory is guaranteed to
292			 * to be there by xfs_idata_realloc().
293			 */
294			data_bytes = roundup(ip->i_afp->if_bytes, 4);
295			ASSERT(ip->i_afp->if_real_bytes == 0 ||
296			       ip->i_afp->if_real_bytes >= data_bytes);
297			ASSERT(ip->i_afp->if_u1.if_data != NULL);
298			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
299					ip->i_afp->if_u1.if_data,
300					data_bytes);
301			ilf->ilf_asize = (unsigned)data_bytes;
302			ilf->ilf_size++;
303		} else {
304			iip->ili_fields &= ~XFS_ILOG_ADATA;
305		}
306		break;
307	default:
308		ASSERT(0);
309		break;
310	}
311}
312
313static void
314xfs_inode_to_log_dinode(
315	struct xfs_inode	*ip,
316	struct xfs_log_dinode	*to,
317	xfs_lsn_t		lsn)
318{
319	struct xfs_icdinode	*from = &ip->i_d;
320	struct inode		*inode = VFS_I(ip);
321
322	to->di_magic = XFS_DINODE_MAGIC;
323
324	to->di_version = from->di_version;
325	to->di_format = from->di_format;
326	to->di_uid = from->di_uid;
327	to->di_gid = from->di_gid;
328	to->di_projid_lo = from->di_projid_lo;
329	to->di_projid_hi = from->di_projid_hi;
330
331	memset(to->di_pad, 0, sizeof(to->di_pad));
332	memset(to->di_pad3, 0, sizeof(to->di_pad3));
333	to->di_atime.t_sec = inode->i_atime.tv_sec;
334	to->di_atime.t_nsec = inode->i_atime.tv_nsec;
335	to->di_mtime.t_sec = inode->i_mtime.tv_sec;
336	to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
337	to->di_ctime.t_sec = inode->i_ctime.tv_sec;
338	to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
339	to->di_nlink = inode->i_nlink;
340	to->di_gen = inode->i_generation;
341	to->di_mode = inode->i_mode;
342
343	to->di_size = from->di_size;
344	to->di_nblocks = from->di_nblocks;
345	to->di_extsize = from->di_extsize;
346	to->di_nextents = from->di_nextents;
347	to->di_anextents = from->di_anextents;
348	to->di_forkoff = from->di_forkoff;
349	to->di_aformat = from->di_aformat;
350	to->di_dmevmask = from->di_dmevmask;
351	to->di_dmstate = from->di_dmstate;
352	to->di_flags = from->di_flags;
353
354	/* log a dummy value to ensure log structure is fully initialised */
355	to->di_next_unlinked = NULLAGINO;
356
357	if (from->di_version == 3) {
358		to->di_changecount = inode_peek_iversion(inode);
359		to->di_crtime.t_sec = from->di_crtime.t_sec;
360		to->di_crtime.t_nsec = from->di_crtime.t_nsec;
361		to->di_flags2 = from->di_flags2;
362		to->di_cowextsize = from->di_cowextsize;
363		to->di_ino = ip->i_ino;
364		to->di_lsn = lsn;
365		memset(to->di_pad2, 0, sizeof(to->di_pad2));
366		uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
367		to->di_flushiter = 0;
368	} else {
369		to->di_flushiter = from->di_flushiter;
370	}
371}
372
373/*
374 * Format the inode core. Current timestamp data is only in the VFS inode
375 * fields, so we need to grab them from there. Hence rather than just copying
376 * the XFS inode core structure, format the fields directly into the iovec.
377 */
378static void
379xfs_inode_item_format_core(
380	struct xfs_inode	*ip,
381	struct xfs_log_vec	*lv,
382	struct xfs_log_iovec	**vecp)
383{
384	struct xfs_log_dinode	*dic;
385
386	dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
387	xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
388	xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_d.di_version));
389}
390
391/*
392 * This is called to fill in the vector of log iovecs for the given inode
393 * log item.  It fills the first item with an inode log format structure,
394 * the second with the on-disk inode structure, and a possible third and/or
395 * fourth with the inode data/extents/b-tree root and inode attributes
396 * data/extents/b-tree root.
397 *
398 * Note: Always use the 64 bit inode log format structure so we don't
399 * leave an uninitialised hole in the format item on 64 bit systems. Log
400 * recovery on 32 bit systems handles this just fine, so there's no reason
401 * for not using an initialising the properly padded structure all the time.
402 */
403STATIC void
404xfs_inode_item_format(
405	struct xfs_log_item	*lip,
406	struct xfs_log_vec	*lv)
407{
408	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
409	struct xfs_inode	*ip = iip->ili_inode;
410	struct xfs_log_iovec	*vecp = NULL;
411	struct xfs_inode_log_format *ilf;
412
413	ASSERT(ip->i_d.di_version > 1);
414
415	ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
416	ilf->ilf_type = XFS_LI_INODE;
417	ilf->ilf_ino = ip->i_ino;
418	ilf->ilf_blkno = ip->i_imap.im_blkno;
419	ilf->ilf_len = ip->i_imap.im_len;
420	ilf->ilf_boffset = ip->i_imap.im_boffset;
421	ilf->ilf_fields = XFS_ILOG_CORE;
422	ilf->ilf_size = 2; /* format + core */
 
423
424	/*
425	 * make sure we don't leak uninitialised data into the log in the case
426	 * when we don't log every field in the inode.
427	 */
428	ilf->ilf_dsize = 0;
429	ilf->ilf_asize = 0;
430	ilf->ilf_pad = 0;
431	memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
432
433	xlog_finish_iovec(lv, vecp, sizeof(*ilf));
434
435	xfs_inode_item_format_core(ip, lv, &vecp);
436	xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
437	if (XFS_IFORK_Q(ip)) {
438		xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
439	} else {
440		iip->ili_fields &=
441			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
442	}
443
444	/* update the format with the exact fields we actually logged */
445	ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
446}
447
448/*
449 * This is called to pin the inode associated with the inode log
450 * item in memory so it cannot be written out.
451 */
452STATIC void
453xfs_inode_item_pin(
454	struct xfs_log_item	*lip)
455{
456	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
457
458	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
459
460	trace_xfs_inode_pin(ip, _RET_IP_);
461	atomic_inc(&ip->i_pincount);
462}
463
464
465/*
466 * This is called to unpin the inode associated with the inode log
467 * item which was previously pinned with a call to xfs_inode_item_pin().
468 *
469 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
470 */
471STATIC void
472xfs_inode_item_unpin(
473	struct xfs_log_item	*lip,
474	int			remove)
475{
476	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
477
478	trace_xfs_inode_unpin(ip, _RET_IP_);
479	ASSERT(atomic_read(&ip->i_pincount) > 0);
480	if (atomic_dec_and_test(&ip->i_pincount))
481		wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
482}
483
484/*
485 * Callback used to mark a buffer with XFS_LI_FAILED when items in the buffer
486 * have been failed during writeback
487 *
488 * This informs the AIL that the inode is already flush locked on the next push,
489 * and acquires a hold on the buffer to ensure that it isn't reclaimed before
490 * dirty data makes it to disk.
491 */
492STATIC void
493xfs_inode_item_error(
494	struct xfs_log_item	*lip,
495	struct xfs_buf		*bp)
496{
497	ASSERT(xfs_isiflocked(INODE_ITEM(lip)->ili_inode));
498	xfs_set_li_failed(lip, bp);
499}
500
501STATIC uint
502xfs_inode_item_push(
503	struct xfs_log_item	*lip,
504	struct list_head	*buffer_list)
505		__releases(&lip->li_ailp->ail_lock)
506		__acquires(&lip->li_ailp->ail_lock)
507{
508	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
509	struct xfs_inode	*ip = iip->ili_inode;
510	struct xfs_buf		*bp = lip->li_buf;
511	uint			rval = XFS_ITEM_SUCCESS;
512	int			error;
513
514	if (xfs_ipincount(ip) > 0)
515		return XFS_ITEM_PINNED;
516
517	/*
518	 * The buffer containing this item failed to be written back
519	 * previously. Resubmit the buffer for IO.
520	 */
521	if (lip->li_flags & XFS_LI_FAILED) {
522		if (!xfs_buf_trylock(bp))
523			return XFS_ITEM_LOCKED;
524
525		if (!xfs_buf_resubmit_failed_buffers(bp, buffer_list))
526			rval = XFS_ITEM_FLUSHING;
527
528		xfs_buf_unlock(bp);
529		return rval;
530	}
531
532	if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
533		return XFS_ITEM_LOCKED;
534
535	/*
536	 * Re-check the pincount now that we stabilized the value by
537	 * taking the ilock.
538	 */
539	if (xfs_ipincount(ip) > 0) {
540		rval = XFS_ITEM_PINNED;
541		goto out_unlock;
542	}
543
544	/*
545	 * Stale inode items should force out the iclog.
546	 */
547	if (ip->i_flags & XFS_ISTALE) {
548		rval = XFS_ITEM_PINNED;
549		goto out_unlock;
550	}
551
552	/*
553	 * Someone else is already flushing the inode.  Nothing we can do
554	 * here but wait for the flush to finish and remove the item from
555	 * the AIL.
556	 */
557	if (!xfs_iflock_nowait(ip)) {
558		rval = XFS_ITEM_FLUSHING;
559		goto out_unlock;
560	}
561
562	ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
563	ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
564
565	spin_unlock(&lip->li_ailp->ail_lock);
566
567	error = xfs_iflush(ip, &bp);
568	if (!error) {
569		if (!xfs_buf_delwri_queue(bp, buffer_list))
570			rval = XFS_ITEM_FLUSHING;
571		xfs_buf_relse(bp);
572	}
573
574	spin_lock(&lip->li_ailp->ail_lock);
575out_unlock:
576	xfs_iunlock(ip, XFS_ILOCK_SHARED);
577	return rval;
578}
579
580/*
581 * Unlock the inode associated with the inode log item.
 
 
 
582 */
583STATIC void
584xfs_inode_item_unlock(
585	struct xfs_log_item	*lip)
586{
587	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
588	struct xfs_inode	*ip = iip->ili_inode;
589	unsigned short		lock_flags;
590
591	ASSERT(ip->i_itemp != NULL);
592	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
593
594	lock_flags = iip->ili_lock_flags;
595	iip->ili_lock_flags = 0;
596	if (lock_flags)
597		xfs_iunlock(ip, lock_flags);
598}
599
600/*
601 * This is called to find out where the oldest active copy of the inode log
602 * item in the on disk log resides now that the last log write of it completed
603 * at the given lsn.  Since we always re-log all dirty data in an inode, the
604 * latest copy in the on disk log is the only one that matters.  Therefore,
605 * simply return the given lsn.
606 *
607 * If the inode has been marked stale because the cluster is being freed, we
608 * don't want to (re-)insert this inode into the AIL. There is a race condition
609 * where the cluster buffer may be unpinned before the inode is inserted into
610 * the AIL during transaction committed processing. If the buffer is unpinned
611 * before the inode item has been committed and inserted, then it is possible
612 * for the buffer to be written and IO completes before the inode is inserted
613 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
614 * AIL which will never get removed. It will, however, get reclaimed which
615 * triggers an assert in xfs_inode_free() complaining about freein an inode
616 * still in the AIL.
617 *
618 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
619 * transaction committed code knows that it does not need to do any further
620 * processing on the item.
621 */
622STATIC xfs_lsn_t
623xfs_inode_item_committed(
624	struct xfs_log_item	*lip,
625	xfs_lsn_t		lsn)
626{
627	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
628	struct xfs_inode	*ip = iip->ili_inode;
629
630	if (xfs_iflags_test(ip, XFS_ISTALE)) {
631		xfs_inode_item_unpin(lip, 0);
632		return -1;
633	}
634	return lsn;
635}
636
 
 
 
 
637STATIC void
638xfs_inode_item_committing(
639	struct xfs_log_item	*lip,
640	xfs_lsn_t		lsn)
641{
642	INODE_ITEM(lip)->ili_last_lsn = lsn;
643}
644
645/*
646 * This is the ops vector shared by all buf log items.
647 */
648static const struct xfs_item_ops xfs_inode_item_ops = {
649	.iop_size	= xfs_inode_item_size,
650	.iop_format	= xfs_inode_item_format,
651	.iop_pin	= xfs_inode_item_pin,
652	.iop_unpin	= xfs_inode_item_unpin,
653	.iop_unlock	= xfs_inode_item_unlock,
654	.iop_committed	= xfs_inode_item_committed,
655	.iop_push	= xfs_inode_item_push,
656	.iop_committing = xfs_inode_item_committing,
657	.iop_error	= xfs_inode_item_error
658};
659
660
661/*
662 * Initialize the inode log item for a newly allocated (in-core) inode.
663 */
664void
665xfs_inode_item_init(
666	struct xfs_inode	*ip,
667	struct xfs_mount	*mp)
668{
669	struct xfs_inode_log_item *iip;
670
671	ASSERT(ip->i_itemp == NULL);
672	iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
673
674	iip->ili_inode = ip;
675	xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
676						&xfs_inode_item_ops);
677}
678
679/*
680 * Free the inode log item and any memory hanging off of it.
681 */
682void
683xfs_inode_item_destroy(
684	xfs_inode_t	*ip)
685{
686	kmem_free(ip->i_itemp->ili_item.li_lv_shadow);
687	kmem_zone_free(xfs_ili_zone, ip->i_itemp);
688}
689
690
691/*
692 * This is the inode flushing I/O completion routine.  It is called
693 * from interrupt level when the buffer containing the inode is
694 * flushed to disk.  It is responsible for removing the inode item
695 * from the AIL if it has not been re-logged, and unlocking the inode's
696 * flush lock.
697 *
698 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
699 * list for other inodes that will run this function. We remove them from the
700 * buffer list so we can process all the inode IO completions in one AIL lock
701 * traversal.
702 */
703void
704xfs_iflush_done(
705	struct xfs_buf		*bp,
706	struct xfs_log_item	*lip)
707{
708	struct xfs_inode_log_item *iip;
709	struct xfs_log_item	*blip, *n;
 
 
710	struct xfs_ail		*ailp = lip->li_ailp;
711	int			need_ail = 0;
712	LIST_HEAD(tmp);
713
714	/*
715	 * Scan the buffer IO completions for other inodes being completed and
716	 * attach them to the current inode log item.
717	 */
 
 
 
 
 
 
 
 
718
719	list_add_tail(&lip->li_bio_list, &tmp);
 
 
 
 
 
 
720
721	list_for_each_entry_safe(blip, n, &bp->b_li_list, li_bio_list) {
722		if (lip->li_cb != xfs_iflush_done)
723			continue;
724
725		list_move_tail(&blip->li_bio_list, &tmp);
726		/*
727		 * while we have the item, do the unlocked check for needing
728		 * the AIL lock.
729		 */
730		iip = INODE_ITEM(blip);
731		if ((iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn) ||
732		    (blip->li_flags & XFS_LI_FAILED))
733			need_ail++;
 
 
734	}
735
736	/* make sure we capture the state of the initial inode. */
737	iip = INODE_ITEM(lip);
738	if ((iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn) ||
739	    lip->li_flags & XFS_LI_FAILED)
740		need_ail++;
741
742	/*
743	 * We only want to pull the item from the AIL if it is
744	 * actually there and its location in the log has not
745	 * changed since we started the flush.  Thus, we only bother
746	 * if the ili_logged flag is set and the inode's lsn has not
747	 * changed.  First we check the lsn outside
748	 * the lock since it's cheaper, and then we recheck while
749	 * holding the lock before removing the inode from the AIL.
750	 */
751	if (need_ail) {
752		bool			mlip_changed = false;
753
754		/* this is an opencoded batch version of xfs_trans_ail_delete */
755		spin_lock(&ailp->ail_lock);
756		list_for_each_entry(blip, &tmp, li_bio_list) {
757			if (INODE_ITEM(blip)->ili_logged &&
758			    blip->li_lsn == INODE_ITEM(blip)->ili_flush_lsn)
759				mlip_changed |= xfs_ail_delete_one(ailp, blip);
760			else {
761				xfs_clear_li_failed(blip);
762			}
 
763		}
764
765		if (mlip_changed) {
766			if (!XFS_FORCED_SHUTDOWN(ailp->ail_mount))
767				xlog_assign_tail_lsn_locked(ailp->ail_mount);
768			if (list_empty(&ailp->ail_head))
769				wake_up_all(&ailp->ail_empty);
770		}
771		spin_unlock(&ailp->ail_lock);
772
773		if (mlip_changed)
774			xfs_log_space_wake(ailp->ail_mount);
775	}
776
 
777	/*
778	 * clean up and unlock the flush lock now we are done. We can clear the
779	 * ili_last_fields bits now that we know that the data corresponding to
780	 * them is safely on disk.
781	 */
782	list_for_each_entry_safe(blip, n, &tmp, li_bio_list) {
783		list_del_init(&blip->li_bio_list);
 
 
784		iip = INODE_ITEM(blip);
785		iip->ili_logged = 0;
786		iip->ili_last_fields = 0;
787		xfs_ifunlock(iip->ili_inode);
788	}
789	list_del(&tmp);
790}
791
792/*
793 * This is the inode flushing abort routine.  It is called from xfs_iflush when
794 * the filesystem is shutting down to clean up the inode state.  It is
795 * responsible for removing the inode item from the AIL if it has not been
796 * re-logged, and unlocking the inode's flush lock.
797 */
798void
799xfs_iflush_abort(
800	xfs_inode_t		*ip,
801	bool			stale)
802{
803	xfs_inode_log_item_t	*iip = ip->i_itemp;
804
805	if (iip) {
 
806		if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
807			xfs_trans_ail_remove(&iip->ili_item,
808					     stale ? SHUTDOWN_LOG_IO_ERROR :
 
 
 
 
809						     SHUTDOWN_CORRUPT_INCORE);
 
 
810		}
811		iip->ili_logged = 0;
812		/*
813		 * Clear the ili_last_fields bits now that we know that the
814		 * data corresponding to them is safely on disk.
815		 */
816		iip->ili_last_fields = 0;
817		/*
818		 * Clear the inode logging fields so no more flushes are
819		 * attempted.
820		 */
821		iip->ili_fields = 0;
822		iip->ili_fsync_fields = 0;
823	}
824	/*
825	 * Release the inode's flush lock since we're done with it.
826	 */
827	xfs_ifunlock(ip);
828}
829
830void
831xfs_istale_done(
832	struct xfs_buf		*bp,
833	struct xfs_log_item	*lip)
834{
835	xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
836}
837
838/*
839 * convert an xfs_inode_log_format struct from the old 32 bit version
840 * (which can have different field alignments) to the native 64 bit version
841 */
842int
843xfs_inode_item_format_convert(
844	struct xfs_log_iovec		*buf,
845	struct xfs_inode_log_format	*in_f)
846{
847	struct xfs_inode_log_format_32	*in_f32 = buf->i_addr;
 
848
849	if (buf->i_len != sizeof(*in_f32))
850		return -EFSCORRUPTED;
851
852	in_f->ilf_type = in_f32->ilf_type;
853	in_f->ilf_size = in_f32->ilf_size;
854	in_f->ilf_fields = in_f32->ilf_fields;
855	in_f->ilf_asize = in_f32->ilf_asize;
856	in_f->ilf_dsize = in_f32->ilf_dsize;
857	in_f->ilf_ino = in_f32->ilf_ino;
858	memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
859	in_f->ilf_blkno = in_f32->ilf_blkno;
860	in_f->ilf_len = in_f32->ilf_len;
861	in_f->ilf_boffset = in_f32->ilf_boffset;
862	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
863}
v3.15
  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_format.h"
 21#include "xfs_log_format.h"
 22#include "xfs_trans_resv.h"
 23#include "xfs_sb.h"
 24#include "xfs_ag.h"
 25#include "xfs_mount.h"
 26#include "xfs_inode.h"
 27#include "xfs_trans.h"
 28#include "xfs_inode_item.h"
 29#include "xfs_error.h"
 30#include "xfs_trace.h"
 31#include "xfs_trans_priv.h"
 32#include "xfs_dinode.h"
 33#include "xfs_log.h"
 34
 
 35
 36kmem_zone_t	*xfs_ili_zone;		/* inode log item zone */
 37
 38static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
 39{
 40	return container_of(lip, struct xfs_inode_log_item, ili_item);
 41}
 42
 43STATIC void
 44xfs_inode_item_data_fork_size(
 45	struct xfs_inode_log_item *iip,
 46	int			*nvecs,
 47	int			*nbytes)
 48{
 49	struct xfs_inode	*ip = iip->ili_inode;
 50
 51	switch (ip->i_d.di_format) {
 52	case XFS_DINODE_FMT_EXTENTS:
 53		if ((iip->ili_fields & XFS_ILOG_DEXT) &&
 54		    ip->i_d.di_nextents > 0 &&
 55		    ip->i_df.if_bytes > 0) {
 56			/* worst case, doesn't subtract delalloc extents */
 57			*nbytes += XFS_IFORK_DSIZE(ip);
 58			*nvecs += 1;
 59		}
 60		break;
 61	case XFS_DINODE_FMT_BTREE:
 62		if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 63		    ip->i_df.if_broot_bytes > 0) {
 64			*nbytes += ip->i_df.if_broot_bytes;
 65			*nvecs += 1;
 66		}
 67		break;
 68	case XFS_DINODE_FMT_LOCAL:
 69		if ((iip->ili_fields & XFS_ILOG_DDATA) &&
 70		    ip->i_df.if_bytes > 0) {
 71			*nbytes += roundup(ip->i_df.if_bytes, 4);
 72			*nvecs += 1;
 73		}
 74		break;
 75
 76	case XFS_DINODE_FMT_DEV:
 77	case XFS_DINODE_FMT_UUID:
 78		break;
 79	default:
 80		ASSERT(0);
 81		break;
 82	}
 83}
 84
 85STATIC void
 86xfs_inode_item_attr_fork_size(
 87	struct xfs_inode_log_item *iip,
 88	int			*nvecs,
 89	int			*nbytes)
 90{
 91	struct xfs_inode	*ip = iip->ili_inode;
 92
 93	switch (ip->i_d.di_aformat) {
 94	case XFS_DINODE_FMT_EXTENTS:
 95		if ((iip->ili_fields & XFS_ILOG_AEXT) &&
 96		    ip->i_d.di_anextents > 0 &&
 97		    ip->i_afp->if_bytes > 0) {
 98			/* worst case, doesn't subtract unused space */
 99			*nbytes += XFS_IFORK_ASIZE(ip);
100			*nvecs += 1;
101		}
102		break;
103	case XFS_DINODE_FMT_BTREE:
104		if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
105		    ip->i_afp->if_broot_bytes > 0) {
106			*nbytes += ip->i_afp->if_broot_bytes;
107			*nvecs += 1;
108		}
109		break;
110	case XFS_DINODE_FMT_LOCAL:
111		if ((iip->ili_fields & XFS_ILOG_ADATA) &&
112		    ip->i_afp->if_bytes > 0) {
113			*nbytes += roundup(ip->i_afp->if_bytes, 4);
114			*nvecs += 1;
115		}
116		break;
117	default:
118		ASSERT(0);
119		break;
120	}
121}
122
123/*
124 * This returns the number of iovecs needed to log the given inode item.
125 *
126 * We need one iovec for the inode log format structure, one for the
127 * inode core, and possibly one for the inode data/extents/b-tree root
128 * and one for the inode attribute data/extents/b-tree root.
129 */
130STATIC void
131xfs_inode_item_size(
132	struct xfs_log_item	*lip,
133	int			*nvecs,
134	int			*nbytes)
135{
136	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
137	struct xfs_inode	*ip = iip->ili_inode;
138
139	*nvecs += 2;
140	*nbytes += sizeof(struct xfs_inode_log_format) +
141		   xfs_icdinode_size(ip->i_d.di_version);
142
143	xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
144	if (XFS_IFORK_Q(ip))
145		xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
146}
147
148/*
149 * If this is a v1 format inode, then we need to log it as such.  This means
150 * that we have to copy the link count from the new field to the old.  We
151 * don't have to worry about the new fields, because nothing trusts them as
152 * long as the old inode version number is there.
153 */
154STATIC void
155xfs_inode_item_format_v1_inode(
156	struct xfs_inode	*ip)
157{
158	if (!xfs_sb_version_hasnlink(&ip->i_mount->m_sb)) {
159		/*
160		 * Convert it back.
161		 */
162		ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
163		ip->i_d.di_onlink = ip->i_d.di_nlink;
164	} else {
165		/*
166		 * The superblock version has already been bumped,
167		 * so just make the conversion to the new inode
168		 * format permanent.
169		 */
170		ip->i_d.di_version = 2;
171		ip->i_d.di_onlink = 0;
172		memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
173	}
174}
175
176STATIC void
177xfs_inode_item_format_data_fork(
178	struct xfs_inode_log_item *iip,
179	struct xfs_inode_log_format *ilf,
180	struct xfs_log_vec	*lv,
181	struct xfs_log_iovec	**vecp)
182{
183	struct xfs_inode	*ip = iip->ili_inode;
184	size_t			data_bytes;
185
186	switch (ip->i_d.di_format) {
187	case XFS_DINODE_FMT_EXTENTS:
188		iip->ili_fields &=
189			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
190			  XFS_ILOG_DEV | XFS_ILOG_UUID);
191
192		if ((iip->ili_fields & XFS_ILOG_DEXT) &&
193		    ip->i_d.di_nextents > 0 &&
194		    ip->i_df.if_bytes > 0) {
195			struct xfs_bmbt_rec *p;
196
197			ASSERT(ip->i_df.if_u1.if_extents != NULL);
198			ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0);
199
200			p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
201			data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
202			xlog_finish_iovec(lv, *vecp, data_bytes);
203
204			ASSERT(data_bytes <= ip->i_df.if_bytes);
205
206			ilf->ilf_dsize = data_bytes;
207			ilf->ilf_size++;
208		} else {
209			iip->ili_fields &= ~XFS_ILOG_DEXT;
210		}
211		break;
212	case XFS_DINODE_FMT_BTREE:
213		iip->ili_fields &=
214			~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
215			  XFS_ILOG_DEV | XFS_ILOG_UUID);
216
217		if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
218		    ip->i_df.if_broot_bytes > 0) {
219			ASSERT(ip->i_df.if_broot != NULL);
220			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
221					ip->i_df.if_broot,
222					ip->i_df.if_broot_bytes);
223			ilf->ilf_dsize = ip->i_df.if_broot_bytes;
224			ilf->ilf_size++;
225		} else {
226			ASSERT(!(iip->ili_fields &
227				 XFS_ILOG_DBROOT));
228			iip->ili_fields &= ~XFS_ILOG_DBROOT;
229		}
230		break;
231	case XFS_DINODE_FMT_LOCAL:
232		iip->ili_fields &=
233			~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
234			  XFS_ILOG_DEV | XFS_ILOG_UUID);
235		if ((iip->ili_fields & XFS_ILOG_DDATA) &&
236		    ip->i_df.if_bytes > 0) {
237			/*
238			 * Round i_bytes up to a word boundary.
239			 * The underlying memory is guaranteed to
240			 * to be there by xfs_idata_realloc().
241			 */
242			data_bytes = roundup(ip->i_df.if_bytes, 4);
243			ASSERT(ip->i_df.if_real_bytes == 0 ||
244			       ip->i_df.if_real_bytes == data_bytes);
245			ASSERT(ip->i_df.if_u1.if_data != NULL);
246			ASSERT(ip->i_d.di_size > 0);
247			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
248					ip->i_df.if_u1.if_data, data_bytes);
249			ilf->ilf_dsize = (unsigned)data_bytes;
250			ilf->ilf_size++;
251		} else {
252			iip->ili_fields &= ~XFS_ILOG_DDATA;
253		}
254		break;
255	case XFS_DINODE_FMT_DEV:
256		iip->ili_fields &=
257			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
258			  XFS_ILOG_DEXT | XFS_ILOG_UUID);
259		if (iip->ili_fields & XFS_ILOG_DEV)
260			ilf->ilf_u.ilfu_rdev = ip->i_df.if_u2.if_rdev;
261		break;
262	case XFS_DINODE_FMT_UUID:
263		iip->ili_fields &=
264			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
265			  XFS_ILOG_DEXT | XFS_ILOG_DEV);
266		if (iip->ili_fields & XFS_ILOG_UUID)
267			ilf->ilf_u.ilfu_uuid = ip->i_df.if_u2.if_uuid;
268		break;
269	default:
270		ASSERT(0);
271		break;
272	}
273}
274
275STATIC void
276xfs_inode_item_format_attr_fork(
277	struct xfs_inode_log_item *iip,
278	struct xfs_inode_log_format *ilf,
279	struct xfs_log_vec	*lv,
280	struct xfs_log_iovec	**vecp)
281{
282	struct xfs_inode	*ip = iip->ili_inode;
283	size_t			data_bytes;
284
285	switch (ip->i_d.di_aformat) {
286	case XFS_DINODE_FMT_EXTENTS:
287		iip->ili_fields &=
288			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
289
290		if ((iip->ili_fields & XFS_ILOG_AEXT) &&
291		    ip->i_d.di_anextents > 0 &&
292		    ip->i_afp->if_bytes > 0) {
293			struct xfs_bmbt_rec *p;
294
295			ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) ==
296				ip->i_d.di_anextents);
297			ASSERT(ip->i_afp->if_u1.if_extents != NULL);
298
299			p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
300			data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
301			xlog_finish_iovec(lv, *vecp, data_bytes);
302
303			ilf->ilf_asize = data_bytes;
304			ilf->ilf_size++;
305		} else {
306			iip->ili_fields &= ~XFS_ILOG_AEXT;
307		}
308		break;
309	case XFS_DINODE_FMT_BTREE:
310		iip->ili_fields &=
311			~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
312
313		if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
314		    ip->i_afp->if_broot_bytes > 0) {
315			ASSERT(ip->i_afp->if_broot != NULL);
316
317			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
318					ip->i_afp->if_broot,
319					ip->i_afp->if_broot_bytes);
320			ilf->ilf_asize = ip->i_afp->if_broot_bytes;
321			ilf->ilf_size++;
322		} else {
323			iip->ili_fields &= ~XFS_ILOG_ABROOT;
324		}
325		break;
326	case XFS_DINODE_FMT_LOCAL:
327		iip->ili_fields &=
328			~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
329
330		if ((iip->ili_fields & XFS_ILOG_ADATA) &&
331		    ip->i_afp->if_bytes > 0) {
332			/*
333			 * Round i_bytes up to a word boundary.
334			 * The underlying memory is guaranteed to
335			 * to be there by xfs_idata_realloc().
336			 */
337			data_bytes = roundup(ip->i_afp->if_bytes, 4);
338			ASSERT(ip->i_afp->if_real_bytes == 0 ||
339			       ip->i_afp->if_real_bytes == data_bytes);
340			ASSERT(ip->i_afp->if_u1.if_data != NULL);
341			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
342					ip->i_afp->if_u1.if_data,
343					data_bytes);
344			ilf->ilf_asize = (unsigned)data_bytes;
345			ilf->ilf_size++;
346		} else {
347			iip->ili_fields &= ~XFS_ILOG_ADATA;
348		}
349		break;
350	default:
351		ASSERT(0);
352		break;
353	}
354}
355
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
356/*
357 * This is called to fill in the vector of log iovecs for the given inode
358 * log item.  It fills the first item with an inode log format structure,
359 * the second with the on-disk inode structure, and a possible third and/or
360 * fourth with the inode data/extents/b-tree root and inode attributes
361 * data/extents/b-tree root.
 
 
 
 
 
362 */
363STATIC void
364xfs_inode_item_format(
365	struct xfs_log_item	*lip,
366	struct xfs_log_vec	*lv)
367{
368	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
369	struct xfs_inode	*ip = iip->ili_inode;
 
370	struct xfs_inode_log_format *ilf;
371	struct xfs_log_iovec	*vecp = NULL;
 
372
373	ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
374	ilf->ilf_type = XFS_LI_INODE;
375	ilf->ilf_ino = ip->i_ino;
376	ilf->ilf_blkno = ip->i_imap.im_blkno;
377	ilf->ilf_len = ip->i_imap.im_len;
378	ilf->ilf_boffset = ip->i_imap.im_boffset;
379	ilf->ilf_fields = XFS_ILOG_CORE;
380	ilf->ilf_size = 2; /* format + core */
381	xlog_finish_iovec(lv, vecp, sizeof(struct xfs_inode_log_format));
382
383	if (ip->i_d.di_version == 1)
384		xfs_inode_item_format_v1_inode(ip);
385	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ICORE,
386			&ip->i_d,
387			xfs_icdinode_size(ip->i_d.di_version));
 
 
 
388
 
 
 
389	xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
390	if (XFS_IFORK_Q(ip)) {
391		xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
392	} else {
393		iip->ili_fields &=
394			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
395	}
396
397	/* update the format with the exact fields we actually logged */
398	ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
399}
400
401/*
402 * This is called to pin the inode associated with the inode log
403 * item in memory so it cannot be written out.
404 */
405STATIC void
406xfs_inode_item_pin(
407	struct xfs_log_item	*lip)
408{
409	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
410
411	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
412
413	trace_xfs_inode_pin(ip, _RET_IP_);
414	atomic_inc(&ip->i_pincount);
415}
416
417
418/*
419 * This is called to unpin the inode associated with the inode log
420 * item which was previously pinned with a call to xfs_inode_item_pin().
421 *
422 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
423 */
424STATIC void
425xfs_inode_item_unpin(
426	struct xfs_log_item	*lip,
427	int			remove)
428{
429	struct xfs_inode	*ip = INODE_ITEM(lip)->ili_inode;
430
431	trace_xfs_inode_unpin(ip, _RET_IP_);
432	ASSERT(atomic_read(&ip->i_pincount) > 0);
433	if (atomic_dec_and_test(&ip->i_pincount))
434		wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
435}
436
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
437STATIC uint
438xfs_inode_item_push(
439	struct xfs_log_item	*lip,
440	struct list_head	*buffer_list)
 
 
441{
442	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
443	struct xfs_inode	*ip = iip->ili_inode;
444	struct xfs_buf		*bp = NULL;
445	uint			rval = XFS_ITEM_SUCCESS;
446	int			error;
447
448	if (xfs_ipincount(ip) > 0)
449		return XFS_ITEM_PINNED;
450
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
451	if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
452		return XFS_ITEM_LOCKED;
453
454	/*
455	 * Re-check the pincount now that we stabilized the value by
456	 * taking the ilock.
457	 */
458	if (xfs_ipincount(ip) > 0) {
459		rval = XFS_ITEM_PINNED;
460		goto out_unlock;
461	}
462
463	/*
464	 * Stale inode items should force out the iclog.
465	 */
466	if (ip->i_flags & XFS_ISTALE) {
467		rval = XFS_ITEM_PINNED;
468		goto out_unlock;
469	}
470
471	/*
472	 * Someone else is already flushing the inode.  Nothing we can do
473	 * here but wait for the flush to finish and remove the item from
474	 * the AIL.
475	 */
476	if (!xfs_iflock_nowait(ip)) {
477		rval = XFS_ITEM_FLUSHING;
478		goto out_unlock;
479	}
480
481	ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
482	ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
483
484	spin_unlock(&lip->li_ailp->xa_lock);
485
486	error = xfs_iflush(ip, &bp);
487	if (!error) {
488		if (!xfs_buf_delwri_queue(bp, buffer_list))
489			rval = XFS_ITEM_FLUSHING;
490		xfs_buf_relse(bp);
491	}
492
493	spin_lock(&lip->li_ailp->xa_lock);
494out_unlock:
495	xfs_iunlock(ip, XFS_ILOCK_SHARED);
496	return rval;
497}
498
499/*
500 * Unlock the inode associated with the inode log item.
501 * Clear the fields of the inode and inode log item that
502 * are specific to the current transaction.  If the
503 * hold flags is set, do not unlock the inode.
504 */
505STATIC void
506xfs_inode_item_unlock(
507	struct xfs_log_item	*lip)
508{
509	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
510	struct xfs_inode	*ip = iip->ili_inode;
511	unsigned short		lock_flags;
512
513	ASSERT(ip->i_itemp != NULL);
514	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
515
516	lock_flags = iip->ili_lock_flags;
517	iip->ili_lock_flags = 0;
518	if (lock_flags)
519		xfs_iunlock(ip, lock_flags);
520}
521
522/*
523 * This is called to find out where the oldest active copy of the inode log
524 * item in the on disk log resides now that the last log write of it completed
525 * at the given lsn.  Since we always re-log all dirty data in an inode, the
526 * latest copy in the on disk log is the only one that matters.  Therefore,
527 * simply return the given lsn.
528 *
529 * If the inode has been marked stale because the cluster is being freed, we
530 * don't want to (re-)insert this inode into the AIL. There is a race condition
531 * where the cluster buffer may be unpinned before the inode is inserted into
532 * the AIL during transaction committed processing. If the buffer is unpinned
533 * before the inode item has been committed and inserted, then it is possible
534 * for the buffer to be written and IO completes before the inode is inserted
535 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
536 * AIL which will never get removed. It will, however, get reclaimed which
537 * triggers an assert in xfs_inode_free() complaining about freein an inode
538 * still in the AIL.
539 *
540 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
541 * transaction committed code knows that it does not need to do any further
542 * processing on the item.
543 */
544STATIC xfs_lsn_t
545xfs_inode_item_committed(
546	struct xfs_log_item	*lip,
547	xfs_lsn_t		lsn)
548{
549	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
550	struct xfs_inode	*ip = iip->ili_inode;
551
552	if (xfs_iflags_test(ip, XFS_ISTALE)) {
553		xfs_inode_item_unpin(lip, 0);
554		return -1;
555	}
556	return lsn;
557}
558
559/*
560 * XXX rcc - this one really has to do something.  Probably needs
561 * to stamp in a new field in the incore inode.
562 */
563STATIC void
564xfs_inode_item_committing(
565	struct xfs_log_item	*lip,
566	xfs_lsn_t		lsn)
567{
568	INODE_ITEM(lip)->ili_last_lsn = lsn;
569}
570
571/*
572 * This is the ops vector shared by all buf log items.
573 */
574static const struct xfs_item_ops xfs_inode_item_ops = {
575	.iop_size	= xfs_inode_item_size,
576	.iop_format	= xfs_inode_item_format,
577	.iop_pin	= xfs_inode_item_pin,
578	.iop_unpin	= xfs_inode_item_unpin,
579	.iop_unlock	= xfs_inode_item_unlock,
580	.iop_committed	= xfs_inode_item_committed,
581	.iop_push	= xfs_inode_item_push,
582	.iop_committing = xfs_inode_item_committing
 
583};
584
585
586/*
587 * Initialize the inode log item for a newly allocated (in-core) inode.
588 */
589void
590xfs_inode_item_init(
591	struct xfs_inode	*ip,
592	struct xfs_mount	*mp)
593{
594	struct xfs_inode_log_item *iip;
595
596	ASSERT(ip->i_itemp == NULL);
597	iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
598
599	iip->ili_inode = ip;
600	xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
601						&xfs_inode_item_ops);
602}
603
604/*
605 * Free the inode log item and any memory hanging off of it.
606 */
607void
608xfs_inode_item_destroy(
609	xfs_inode_t	*ip)
610{
 
611	kmem_zone_free(xfs_ili_zone, ip->i_itemp);
612}
613
614
615/*
616 * This is the inode flushing I/O completion routine.  It is called
617 * from interrupt level when the buffer containing the inode is
618 * flushed to disk.  It is responsible for removing the inode item
619 * from the AIL if it has not been re-logged, and unlocking the inode's
620 * flush lock.
621 *
622 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
623 * list for other inodes that will run this function. We remove them from the
624 * buffer list so we can process all the inode IO completions in one AIL lock
625 * traversal.
626 */
627void
628xfs_iflush_done(
629	struct xfs_buf		*bp,
630	struct xfs_log_item	*lip)
631{
632	struct xfs_inode_log_item *iip;
633	struct xfs_log_item	*blip;
634	struct xfs_log_item	*next;
635	struct xfs_log_item	*prev;
636	struct xfs_ail		*ailp = lip->li_ailp;
637	int			need_ail = 0;
 
638
639	/*
640	 * Scan the buffer IO completions for other inodes being completed and
641	 * attach them to the current inode log item.
642	 */
643	blip = bp->b_fspriv;
644	prev = NULL;
645	while (blip != NULL) {
646		if (lip->li_cb != xfs_iflush_done) {
647			prev = blip;
648			blip = blip->li_bio_list;
649			continue;
650		}
651
652		/* remove from list */
653		next = blip->li_bio_list;
654		if (!prev) {
655			bp->b_fspriv = next;
656		} else {
657			prev->li_bio_list = next;
658		}
659
660		/* add to current list */
661		blip->li_bio_list = lip->li_bio_list;
662		lip->li_bio_list = blip;
663
 
664		/*
665		 * while we have the item, do the unlocked check for needing
666		 * the AIL lock.
667		 */
668		iip = INODE_ITEM(blip);
669		if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
 
670			need_ail++;
671
672		blip = next;
673	}
674
675	/* make sure we capture the state of the initial inode. */
676	iip = INODE_ITEM(lip);
677	if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
 
678		need_ail++;
679
680	/*
681	 * We only want to pull the item from the AIL if it is
682	 * actually there and its location in the log has not
683	 * changed since we started the flush.  Thus, we only bother
684	 * if the ili_logged flag is set and the inode's lsn has not
685	 * changed.  First we check the lsn outside
686	 * the lock since it's cheaper, and then we recheck while
687	 * holding the lock before removing the inode from the AIL.
688	 */
689	if (need_ail) {
690		struct xfs_log_item *log_items[need_ail];
691		int i = 0;
692		spin_lock(&ailp->xa_lock);
693		for (blip = lip; blip; blip = blip->li_bio_list) {
694			iip = INODE_ITEM(blip);
695			if (iip->ili_logged &&
696			    blip->li_lsn == iip->ili_flush_lsn) {
697				log_items[i++] = blip;
 
 
698			}
699			ASSERT(i <= need_ail);
700		}
701		/* xfs_trans_ail_delete_bulk() drops the AIL lock. */
702		xfs_trans_ail_delete_bulk(ailp, log_items, i,
703					  SHUTDOWN_CORRUPT_INCORE);
 
 
 
 
 
 
 
 
704	}
705
706
707	/*
708	 * clean up and unlock the flush lock now we are done. We can clear the
709	 * ili_last_fields bits now that we know that the data corresponding to
710	 * them is safely on disk.
711	 */
712	for (blip = lip; blip; blip = next) {
713		next = blip->li_bio_list;
714		blip->li_bio_list = NULL;
715
716		iip = INODE_ITEM(blip);
717		iip->ili_logged = 0;
718		iip->ili_last_fields = 0;
719		xfs_ifunlock(iip->ili_inode);
720	}
 
721}
722
723/*
724 * This is the inode flushing abort routine.  It is called from xfs_iflush when
725 * the filesystem is shutting down to clean up the inode state.  It is
726 * responsible for removing the inode item from the AIL if it has not been
727 * re-logged, and unlocking the inode's flush lock.
728 */
729void
730xfs_iflush_abort(
731	xfs_inode_t		*ip,
732	bool			stale)
733{
734	xfs_inode_log_item_t	*iip = ip->i_itemp;
735
736	if (iip) {
737		struct xfs_ail	*ailp = iip->ili_item.li_ailp;
738		if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
739			spin_lock(&ailp->xa_lock);
740			if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
741				/* xfs_trans_ail_delete() drops the AIL lock. */
742				xfs_trans_ail_delete(ailp, &iip->ili_item,
743						stale ?
744						     SHUTDOWN_LOG_IO_ERROR :
745						     SHUTDOWN_CORRUPT_INCORE);
746			} else
747				spin_unlock(&ailp->xa_lock);
748		}
749		iip->ili_logged = 0;
750		/*
751		 * Clear the ili_last_fields bits now that we know that the
752		 * data corresponding to them is safely on disk.
753		 */
754		iip->ili_last_fields = 0;
755		/*
756		 * Clear the inode logging fields so no more flushes are
757		 * attempted.
758		 */
759		iip->ili_fields = 0;
 
760	}
761	/*
762	 * Release the inode's flush lock since we're done with it.
763	 */
764	xfs_ifunlock(ip);
765}
766
767void
768xfs_istale_done(
769	struct xfs_buf		*bp,
770	struct xfs_log_item	*lip)
771{
772	xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
773}
774
775/*
776 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
777 * (which can have different field alignments) to the native version
778 */
779int
780xfs_inode_item_format_convert(
781	xfs_log_iovec_t		*buf,
782	xfs_inode_log_format_t	*in_f)
783{
784	if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
785		xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
786
787		in_f->ilf_type = in_f32->ilf_type;
788		in_f->ilf_size = in_f32->ilf_size;
789		in_f->ilf_fields = in_f32->ilf_fields;
790		in_f->ilf_asize = in_f32->ilf_asize;
791		in_f->ilf_dsize = in_f32->ilf_dsize;
792		in_f->ilf_ino = in_f32->ilf_ino;
793		/* copy biggest field of ilf_u */
794		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
795		       in_f32->ilf_u.ilfu_uuid.__u_bits,
796		       sizeof(uuid_t));
797		in_f->ilf_blkno = in_f32->ilf_blkno;
798		in_f->ilf_len = in_f32->ilf_len;
799		in_f->ilf_boffset = in_f32->ilf_boffset;
800		return 0;
801	} else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
802		xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
803
804		in_f->ilf_type = in_f64->ilf_type;
805		in_f->ilf_size = in_f64->ilf_size;
806		in_f->ilf_fields = in_f64->ilf_fields;
807		in_f->ilf_asize = in_f64->ilf_asize;
808		in_f->ilf_dsize = in_f64->ilf_dsize;
809		in_f->ilf_ino = in_f64->ilf_ino;
810		/* copy biggest field of ilf_u */
811		memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
812		       in_f64->ilf_u.ilfu_uuid.__u_bits,
813		       sizeof(uuid_t));
814		in_f->ilf_blkno = in_f64->ilf_blkno;
815		in_f->ilf_len = in_f64->ilf_len;
816		in_f->ilf_boffset = in_f64->ilf_boffset;
817		return 0;
818	}
819	return EFSCORRUPTED;
820}