Linux Audio

Check our new training course

Loading...
v3.5.6
 
  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_log.h"
 22#include "xfs_trans.h"
 23#include "xfs_sb.h"
 24#include "xfs_ag.h"
 25#include "xfs_mount.h"
 26#include "xfs_bmap_btree.h"
 27#include "xfs_alloc_btree.h"
 28#include "xfs_ialloc_btree.h"
 29#include "xfs_dinode.h"
 30#include "xfs_inode.h"
 31#include "xfs_buf_item.h"
 32#include "xfs_trans_priv.h"
 33#include "xfs_error.h"
 34#include "xfs_trace.h"
 35
 36/*
 37 * Check to see if a buffer matching the given parameters is already
 38 * a part of the given transaction.
 39 */
 40STATIC struct xfs_buf *
 41xfs_trans_buf_item_match(
 42	struct xfs_trans	*tp,
 43	struct xfs_buftarg	*target,
 44	xfs_daddr_t		blkno,
 45	int			len)
 46{
 47	struct xfs_log_item_desc *lidp;
 48	struct xfs_buf_log_item	*blip;
 
 
 
 
 
 49
 50	len = BBTOB(len);
 51	list_for_each_entry(lidp, &tp->t_items, lid_trans) {
 52		blip = (struct xfs_buf_log_item *)lidp->lid_item;
 53		if (blip->bli_item.li_type == XFS_LI_BUF &&
 54		    blip->bli_buf->b_target == target &&
 55		    XFS_BUF_ADDR(blip->bli_buf) == blkno &&
 56		    BBTOB(blip->bli_buf->b_length) == len)
 
 57			return blip->bli_buf;
 
 58	}
 59
 60	return NULL;
 61}
 62
 63/*
 64 * Add the locked buffer to the transaction.
 65 *
 66 * The buffer must be locked, and it cannot be associated with any
 67 * transaction.
 68 *
 69 * If the buffer does not yet have a buf log item associated with it,
 70 * then allocate one for it.  Then add the buf item to the transaction.
 71 */
 72STATIC void
 73_xfs_trans_bjoin(
 74	struct xfs_trans	*tp,
 75	struct xfs_buf		*bp,
 76	int			reset_recur)
 77{
 78	struct xfs_buf_log_item	*bip;
 79
 80	ASSERT(bp->b_transp == NULL);
 81
 82	/*
 83	 * The xfs_buf_log_item pointer is stored in b_fsprivate.  If
 84	 * it doesn't have one yet, then allocate one and initialize it.
 85	 * The checks to see if one is there are in xfs_buf_item_init().
 86	 */
 87	xfs_buf_item_init(bp, tp->t_mountp);
 88	bip = bp->b_fspriv;
 89	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 90	ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
 91	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
 92	if (reset_recur)
 93		bip->bli_recur = 0;
 94
 95	/*
 96	 * Take a reference for this transaction on the buf item.
 97	 */
 98	atomic_inc(&bip->bli_refcount);
 99
100	/*
101	 * Get a log_item_desc to point at the new item.
 
102	 */
103	xfs_trans_add_item(tp, &bip->bli_item);
104
105	/*
106	 * Initialize b_fsprivate2 so we can find it with incore_match()
107	 * in xfs_trans_get_buf() and friends above.
108	 */
109	bp->b_transp = tp;
110
111}
112
113void
114xfs_trans_bjoin(
115	struct xfs_trans	*tp,
116	struct xfs_buf		*bp)
117{
118	_xfs_trans_bjoin(tp, bp, 0);
119	trace_xfs_trans_bjoin(bp->b_fspriv);
120}
121
122/*
123 * Get and lock the buffer for the caller if it is not already
124 * locked within the given transaction.  If it is already locked
125 * within the transaction, just increment its lock recursion count
126 * and return a pointer to it.
127 *
128 * If the transaction pointer is NULL, make this just a normal
129 * get_buf() call.
130 */
131xfs_buf_t *
132xfs_trans_get_buf(xfs_trans_t	*tp,
133		  xfs_buftarg_t	*target_dev,
134		  xfs_daddr_t	blkno,
135		  int		len,
136		  uint		flags)
 
 
137{
138	xfs_buf_t		*bp;
139	xfs_buf_log_item_t	*bip;
 
140
141	/*
142	 * Default to a normal get_buf() call if the tp is NULL.
143	 */
144	if (tp == NULL)
145		return xfs_buf_get(target_dev, blkno, len, flags);
146
147	/*
148	 * If we find the buffer in the cache with this transaction
149	 * pointer in its b_fsprivate2 field, then we know we already
150	 * have it locked.  In this case we just increment the lock
151	 * recursion count and return the buffer to the caller.
152	 */
153	bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
154	if (bp != NULL) {
155		ASSERT(xfs_buf_islocked(bp));
156		if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
157			xfs_buf_stale(bp);
158			XFS_BUF_DONE(bp);
159		}
160
161		ASSERT(bp->b_transp == tp);
162		bip = bp->b_fspriv;
163		ASSERT(bip != NULL);
164		ASSERT(atomic_read(&bip->bli_refcount) > 0);
165		bip->bli_recur++;
166		trace_xfs_trans_get_buf_recur(bip);
167		return (bp);
 
168	}
169
170	bp = xfs_buf_get(target_dev, blkno, len, flags);
171	if (bp == NULL) {
172		return NULL;
173	}
174
175	ASSERT(!bp->b_error);
176
177	_xfs_trans_bjoin(tp, bp, 1);
178	trace_xfs_trans_get_buf(bp->b_fspriv);
179	return (bp);
 
180}
181
182/*
183 * Get and lock the superblock buffer of this file system for the
184 * given transaction.
185 *
186 * We don't need to use incore_match() here, because the superblock
187 * buffer is a private buffer which we keep a pointer to in the
188 * mount structure.
189 */
190xfs_buf_t *
191xfs_trans_getsb(xfs_trans_t	*tp,
192		struct xfs_mount *mp,
193		int		flags)
194{
195	xfs_buf_t		*bp;
196	xfs_buf_log_item_t	*bip;
197
198	/*
199	 * Default to just trying to lock the superblock buffer
200	 * if tp is NULL.
201	 */
202	if (tp == NULL) {
203		return (xfs_getsb(mp, flags));
204	}
205
206	/*
207	 * If the superblock buffer already has this transaction
208	 * pointer in its b_fsprivate2 field, then we know we already
209	 * have it locked.  In this case we just increment the lock
210	 * recursion count and return the buffer to the caller.
211	 */
212	bp = mp->m_sb_bp;
213	if (bp->b_transp == tp) {
214		bip = bp->b_fspriv;
 
215		ASSERT(bip != NULL);
216		ASSERT(atomic_read(&bip->bli_refcount) > 0);
217		bip->bli_recur++;
 
218		trace_xfs_trans_getsb_recur(bip);
219		return (bp);
 
 
 
 
 
220	}
221
222	bp = xfs_getsb(mp, flags);
223	if (bp == NULL)
224		return NULL;
225
226	_xfs_trans_bjoin(tp, bp, 1);
227	trace_xfs_trans_getsb(bp->b_fspriv);
228	return (bp);
 
 
229}
230
231#ifdef DEBUG
232xfs_buftarg_t *xfs_error_target;
233int	xfs_do_error;
234int	xfs_req_num;
235int	xfs_error_mod = 33;
236#endif
 
 
237
238/*
239 * Get and lock the buffer for the caller if it is not already
240 * locked within the given transaction.  If it has not yet been
241 * read in, read it from disk. If it is already locked
242 * within the transaction and already read in, just increment its
243 * lock recursion count and return a pointer to it.
244 *
245 * If the transaction pointer is NULL, make this just a normal
246 * read_buf() call.
247 */
248int
249xfs_trans_read_buf(
250	xfs_mount_t	*mp,
251	xfs_trans_t	*tp,
252	xfs_buftarg_t	*target,
253	xfs_daddr_t	blkno,
254	int		len,
255	uint		flags,
256	xfs_buf_t	**bpp)
 
257{
258	xfs_buf_t		*bp;
259	xfs_buf_log_item_t	*bip;
260	int			error;
261
262	*bpp = NULL;
263
264	/*
265	 * Default to a normal get_buf() call if the tp is NULL.
266	 */
267	if (tp == NULL) {
268		bp = xfs_buf_read(target, blkno, len, flags);
269		if (!bp)
270			return (flags & XBF_TRYLOCK) ?
271					EAGAIN : XFS_ERROR(ENOMEM);
272
273		if (bp->b_error) {
274			error = bp->b_error;
275			xfs_buf_ioerror_alert(bp, __func__);
276			XFS_BUF_UNDONE(bp);
277			xfs_buf_stale(bp);
278			xfs_buf_relse(bp);
279			return error;
280		}
281#ifdef DEBUG
282		if (xfs_do_error) {
283			if (xfs_error_target == target) {
284				if (((xfs_req_num++) % xfs_error_mod) == 0) {
285					xfs_buf_relse(bp);
286					xfs_debug(mp, "Returning error!");
287					return XFS_ERROR(EIO);
288				}
289			}
290		}
291#endif
292		if (XFS_FORCED_SHUTDOWN(mp))
293			goto shutdown_abort;
294		*bpp = bp;
295		return 0;
296	}
297
298	/*
299	 * If we find the buffer in the cache with this transaction
300	 * pointer in its b_fsprivate2 field, then we know we already
301	 * have it locked.  If it is already read in we just increment
302	 * the lock recursion count and return the buffer to the caller.
303	 * If the buffer is not yet read in, then we read it in, increment
304	 * the lock recursion count, and return it to the caller.
305	 */
306	bp = xfs_trans_buf_item_match(tp, target, blkno, len);
307	if (bp != NULL) {
 
308		ASSERT(xfs_buf_islocked(bp));
309		ASSERT(bp->b_transp == tp);
310		ASSERT(bp->b_fspriv != NULL);
311		ASSERT(!bp->b_error);
312		if (!(XFS_BUF_ISDONE(bp))) {
313			trace_xfs_trans_read_buf_io(bp, _RET_IP_);
314			ASSERT(!XFS_BUF_ISASYNC(bp));
315			XFS_BUF_READ(bp);
316			xfsbdstrat(tp->t_mountp, bp);
317			error = xfs_buf_iowait(bp);
318			if (error) {
319				xfs_buf_ioerror_alert(bp, __func__);
320				xfs_buf_relse(bp);
321				/*
322				 * We can gracefully recover from most read
323				 * errors. Ones we can't are those that happen
324				 * after the transaction's already dirty.
325				 */
326				if (tp->t_flags & XFS_TRANS_DIRTY)
327					xfs_force_shutdown(tp->t_mountp,
328							SHUTDOWN_META_IO_ERROR);
329				return error;
330			}
331		}
332		/*
333		 * We never locked this buf ourselves, so we shouldn't
334		 * brelse it either. Just get out.
335		 */
336		if (XFS_FORCED_SHUTDOWN(mp)) {
337			trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
338			*bpp = NULL;
339			return XFS_ERROR(EIO);
340		}
341
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
342
343		bip = bp->b_fspriv;
344		bip->bli_recur++;
345
346		ASSERT(atomic_read(&bip->bli_refcount) > 0);
347		trace_xfs_trans_read_buf_recur(bip);
 
348		*bpp = bp;
349		return 0;
350	}
351
352	bp = xfs_buf_read(target, blkno, len, flags);
353	if (bp == NULL) {
354		*bpp = NULL;
355		return (flags & XBF_TRYLOCK) ?
356					0 : XFS_ERROR(ENOMEM);
357	}
358	if (bp->b_error) {
359		error = bp->b_error;
360		xfs_buf_stale(bp);
361		XFS_BUF_DONE(bp);
362		xfs_buf_ioerror_alert(bp, __func__);
363		if (tp->t_flags & XFS_TRANS_DIRTY)
364			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
365		xfs_buf_relse(bp);
 
 
366		return error;
367	}
368#ifdef DEBUG
369	if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
370		if (xfs_error_target == target) {
371			if (((xfs_req_num++) % xfs_error_mod) == 0) {
372				xfs_force_shutdown(tp->t_mountp,
373						   SHUTDOWN_META_IO_ERROR);
374				xfs_buf_relse(bp);
375				xfs_debug(mp, "Returning trans error!");
376				return XFS_ERROR(EIO);
377			}
378		}
379	}
380#endif
381	if (XFS_FORCED_SHUTDOWN(mp))
382		goto shutdown_abort;
383
384	_xfs_trans_bjoin(tp, bp, 1);
385	trace_xfs_trans_read_buf(bp->b_fspriv);
 
 
 
386
 
 
 
 
 
387	*bpp = bp;
388	return 0;
389
390shutdown_abort:
391	trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
392	xfs_buf_relse(bp);
393	*bpp = NULL;
394	return XFS_ERROR(EIO);
395}
396
 
 
 
 
 
 
 
 
 
 
 
 
397
398/*
399 * Release the buffer bp which was previously acquired with one of the
400 * xfs_trans_... buffer allocation routines if the buffer has not
401 * been modified within this transaction.  If the buffer is modified
402 * within this transaction, do decrement the recursion count but do
403 * not release the buffer even if the count goes to 0.  If the buffer is not
404 * modified within the transaction, decrement the recursion count and
405 * release the buffer if the recursion count goes to 0.
406 *
407 * If the buffer is to be released and it was not modified before
408 * this transaction began, then free the buf_log_item associated with it.
409 *
410 * If the transaction pointer is NULL, make this just a normal
411 * brelse() call.
412 */
413void
414xfs_trans_brelse(xfs_trans_t	*tp,
415		 xfs_buf_t	*bp)
 
416{
417	xfs_buf_log_item_t	*bip;
418
419	/*
420	 * Default to a normal brelse() call if the tp is NULL.
421	 */
422	if (tp == NULL) {
423		ASSERT(bp->b_transp == NULL);
424		xfs_buf_relse(bp);
425		return;
426	}
427
428	ASSERT(bp->b_transp == tp);
429	bip = bp->b_fspriv;
430	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
431	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
432	ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
433	ASSERT(atomic_read(&bip->bli_refcount) > 0);
434
435	trace_xfs_trans_brelse(bip);
436
437	/*
438	 * If the release is just for a recursive lock,
439	 * then decrement the count and return.
440	 */
441	if (bip->bli_recur > 0) {
442		bip->bli_recur--;
443		return;
444	}
445
446	/*
447	 * If the buffer is dirty within this transaction, we can't
448	 * release it until we commit.
449	 */
450	if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
451		return;
452
453	/*
454	 * If the buffer has been invalidated, then we can't release
455	 * it until the transaction commits to disk unless it is re-dirtied
456	 * as part of this transaction.  This prevents us from pulling
457	 * the item from the AIL before we should.
458	 */
459	if (bip->bli_flags & XFS_BLI_STALE)
460		return;
461
462	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
463
464	/*
465	 * Free up the log item descriptor tracking the released item.
 
466	 */
 
467	xfs_trans_del_item(&bip->bli_item);
 
468
469	/*
470	 * Clear the hold flag in the buf log item if it is set.
471	 * We wouldn't want the next user of the buffer to
472	 * get confused.
473	 */
474	if (bip->bli_flags & XFS_BLI_HOLD) {
475		bip->bli_flags &= ~XFS_BLI_HOLD;
476	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
477
478	/*
479	 * Drop our reference to the buf log item.
 
480	 */
481	atomic_dec(&bip->bli_refcount);
482
483	/*
484	 * If the buf item is not tracking data in the log, then
485	 * we must free it before releasing the buffer back to the
486	 * free pool.  Before releasing the buffer to the free pool,
487	 * clear the transaction pointer in b_fsprivate2 to dissolve
488	 * its relation to this transaction.
489	 */
490	if (!xfs_buf_item_dirty(bip)) {
491/***
492		ASSERT(bp->b_pincount == 0);
493***/
494		ASSERT(atomic_read(&bip->bli_refcount) == 0);
495		ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
496		ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
497		xfs_buf_item_relse(bp);
498	}
499
 
 
 
500	bp->b_transp = NULL;
501	xfs_buf_relse(bp);
502}
503
504/*
505 * Mark the buffer as not needing to be unlocked when the buf item's
506 * IOP_UNLOCK() routine is called.  The buffer must already be locked
507 * and associated with the given transaction.
508 */
509/* ARGSUSED */
510void
511xfs_trans_bhold(xfs_trans_t	*tp,
512		xfs_buf_t	*bp)
 
513{
514	xfs_buf_log_item_t	*bip = bp->b_fspriv;
515
516	ASSERT(bp->b_transp == tp);
517	ASSERT(bip != NULL);
518	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
519	ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
520	ASSERT(atomic_read(&bip->bli_refcount) > 0);
521
522	bip->bli_flags |= XFS_BLI_HOLD;
523	trace_xfs_trans_bhold(bip);
524}
525
526/*
527 * Cancel the previous buffer hold request made on this buffer
528 * for this transaction.
529 */
530void
531xfs_trans_bhold_release(xfs_trans_t	*tp,
532			xfs_buf_t	*bp)
 
533{
534	xfs_buf_log_item_t	*bip = bp->b_fspriv;
535
536	ASSERT(bp->b_transp == tp);
537	ASSERT(bip != NULL);
538	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
539	ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
540	ASSERT(atomic_read(&bip->bli_refcount) > 0);
541	ASSERT(bip->bli_flags & XFS_BLI_HOLD);
542
543	bip->bli_flags &= ~XFS_BLI_HOLD;
544	trace_xfs_trans_bhold_release(bip);
545}
546
547/*
548 * This is called to mark bytes first through last inclusive of the given
549 * buffer as needing to be logged when the transaction is committed.
550 * The buffer must already be associated with the given transaction.
551 *
552 * First and last are numbers relative to the beginning of this buffer,
553 * so the first byte in the buffer is numbered 0 regardless of the
554 * value of b_blkno.
555 */
556void
557xfs_trans_log_buf(xfs_trans_t	*tp,
558		  xfs_buf_t	*bp,
559		  uint		first,
560		  uint		last)
561{
562	xfs_buf_log_item_t	*bip = bp->b_fspriv;
563
564	ASSERT(bp->b_transp == tp);
565	ASSERT(bip != NULL);
566	ASSERT(first <= last && last < BBTOB(bp->b_length));
567	ASSERT(bp->b_iodone == NULL ||
568	       bp->b_iodone == xfs_buf_iodone_callbacks);
569
570	/*
571	 * Mark the buffer as needing to be written out eventually,
572	 * and set its iodone function to remove the buffer's buf log
573	 * item from the AIL and free it when the buffer is flushed
574	 * to disk.  See xfs_buf_attach_iodone() for more details
575	 * on li_cb and xfs_buf_iodone_callbacks().
576	 * If we end up aborting this transaction, we trap this buffer
577	 * inside the b_bdstrat callback so that this won't get written to
578	 * disk.
579	 */
580	XFS_BUF_DONE(bp);
581
582	ASSERT(atomic_read(&bip->bli_refcount) > 0);
583	bp->b_iodone = xfs_buf_iodone_callbacks;
584	bip->bli_item.li_cb = xfs_buf_iodone;
585
586	trace_xfs_trans_log_buf(bip);
587
588	/*
589	 * If we invalidated the buffer within this transaction, then
590	 * cancel the invalidation now that we're dirtying the buffer
591	 * again.  There are no races with the code in xfs_buf_item_unpin(),
592	 * because we have a reference to the buffer this entire time.
593	 */
594	if (bip->bli_flags & XFS_BLI_STALE) {
595		bip->bli_flags &= ~XFS_BLI_STALE;
596		ASSERT(XFS_BUF_ISSTALE(bp));
597		XFS_BUF_UNSTALE(bp);
598		bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
599	}
 
600
601	tp->t_flags |= XFS_TRANS_DIRTY;
602	bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
603	bip->bli_flags |= XFS_BLI_LOGGED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
604	xfs_buf_item_log(bip, first, last);
605}
606
607
608/*
609 * Invalidate a buffer that is being used within a transaction.
610 *
611 * Typically this is because the blocks in the buffer are being freed, so we
612 * need to prevent it from being written out when we're done.  Allowing it
613 * to be written again might overwrite data in the free blocks if they are
614 * reallocated to a file.
615 *
616 * We prevent the buffer from being written out by marking it stale.  We can't
617 * get rid of the buf log item at this point because the buffer may still be
618 * pinned by another transaction.  If that is the case, then we'll wait until
619 * the buffer is committed to disk for the last time (we can tell by the ref
620 * count) and free it in xfs_buf_item_unpin().  Until that happens we will
621 * keep the buffer locked so that the buffer and buf log item are not reused.
622 *
623 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
624 * the buf item.  This will be used at recovery time to determine that copies
625 * of the buffer in the log before this should not be replayed.
626 *
627 * We mark the item descriptor and the transaction dirty so that we'll hold
628 * the buffer until after the commit.
629 *
630 * Since we're invalidating the buffer, we also clear the state about which
631 * parts of the buffer have been logged.  We also clear the flag indicating
632 * that this is an inode buffer since the data in the buffer will no longer
633 * be valid.
634 *
635 * We set the stale bit in the buffer as well since we're getting rid of it.
636 */
637void
638xfs_trans_binval(
639	xfs_trans_t	*tp,
640	xfs_buf_t	*bp)
641{
642	xfs_buf_log_item_t	*bip = bp->b_fspriv;
 
643
644	ASSERT(bp->b_transp == tp);
645	ASSERT(bip != NULL);
646	ASSERT(atomic_read(&bip->bli_refcount) > 0);
647
648	trace_xfs_trans_binval(bip);
649
650	if (bip->bli_flags & XFS_BLI_STALE) {
651		/*
652		 * If the buffer is already invalidated, then
653		 * just return.
654		 */
655		ASSERT(XFS_BUF_ISSTALE(bp));
656		ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
657		ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
658		ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
659		ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
 
660		ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
661		return;
662	}
663
664	xfs_buf_stale(bp);
665
666	bip->bli_flags |= XFS_BLI_STALE;
667	bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
668	bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
669	bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
670	memset((char *)(bip->bli_format.blf_data_map), 0,
671	      (bip->bli_format.blf_map_size * sizeof(uint)));
672	bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
 
 
 
673	tp->t_flags |= XFS_TRANS_DIRTY;
674}
675
676/*
677 * This call is used to indicate that the buffer contains on-disk inodes which
678 * must be handled specially during recovery.  They require special handling
679 * because only the di_next_unlinked from the inodes in the buffer should be
680 * recovered.  The rest of the data in the buffer is logged via the inodes
681 * themselves.
682 *
683 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
684 * transferred to the buffer's log format structure so that we'll know what to
685 * do at recovery time.
686 */
687void
688xfs_trans_inode_buf(
689	xfs_trans_t	*tp,
690	xfs_buf_t	*bp)
691{
692	xfs_buf_log_item_t	*bip = bp->b_fspriv;
693
694	ASSERT(bp->b_transp == tp);
695	ASSERT(bip != NULL);
696	ASSERT(atomic_read(&bip->bli_refcount) > 0);
697
698	bip->bli_flags |= XFS_BLI_INODE_BUF;
 
 
699}
700
701/*
702 * This call is used to indicate that the buffer is going to
703 * be staled and was an inode buffer. This means it gets
704 * special processing during unpin - where any inodes 
705 * associated with the buffer should be removed from ail.
706 * There is also special processing during recovery,
707 * any replay of the inodes in the buffer needs to be
708 * prevented as the buffer may have been reused.
709 */
710void
711xfs_trans_stale_inode_buf(
712	xfs_trans_t	*tp,
713	xfs_buf_t	*bp)
714{
715	xfs_buf_log_item_t	*bip = bp->b_fspriv;
716
717	ASSERT(bp->b_transp == tp);
718	ASSERT(bip != NULL);
719	ASSERT(atomic_read(&bip->bli_refcount) > 0);
720
721	bip->bli_flags |= XFS_BLI_STALE_INODE;
722	bip->bli_item.li_cb = xfs_buf_iodone;
 
723}
724
725/*
726 * Mark the buffer as being one which contains newly allocated
727 * inodes.  We need to make sure that even if this buffer is
728 * relogged as an 'inode buf' we still recover all of the inode
729 * images in the face of a crash.  This works in coordination with
730 * xfs_buf_item_committed() to ensure that the buffer remains in the
731 * AIL at its original location even after it has been relogged.
732 */
733/* ARGSUSED */
734void
735xfs_trans_inode_alloc_buf(
736	xfs_trans_t	*tp,
737	xfs_buf_t	*bp)
738{
739	xfs_buf_log_item_t	*bip = bp->b_fspriv;
740
741	ASSERT(bp->b_transp == tp);
742	ASSERT(bip != NULL);
743	ASSERT(atomic_read(&bip->bli_refcount) > 0);
744
745	bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
 
 
746}
747
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
748
749/*
750 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
751 * dquots. However, unlike in inode buffer recovery, dquot buffers get
752 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
753 * The only thing that makes dquot buffers different from regular
754 * buffers is that we must not replay dquot bufs when recovering
755 * if a _corresponding_ quotaoff has happened. We also have to distinguish
756 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
757 * can be turned off independently.
758 */
759/* ARGSUSED */
760void
761xfs_trans_dquot_buf(
762	xfs_trans_t	*tp,
763	xfs_buf_t	*bp,
764	uint		type)
765{
766	xfs_buf_log_item_t	*bip = bp->b_fspriv;
767
768	ASSERT(bp->b_transp == tp);
769	ASSERT(bip != NULL);
770	ASSERT(type == XFS_BLF_UDQUOT_BUF ||
771	       type == XFS_BLF_PDQUOT_BUF ||
772	       type == XFS_BLF_GDQUOT_BUF);
773	ASSERT(atomic_read(&bip->bli_refcount) > 0);
774
775	bip->bli_format.blf_flags |= type;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
776}
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
  4 * All Rights Reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
  5 */
  6#include "xfs.h"
  7#include "xfs_fs.h"
  8#include "xfs_shared.h"
  9#include "xfs_format.h"
 10#include "xfs_log_format.h"
 11#include "xfs_trans_resv.h"
 
 12#include "xfs_mount.h"
 13#include "xfs_trans.h"
 
 
 
 
 14#include "xfs_buf_item.h"
 15#include "xfs_trans_priv.h"
 
 16#include "xfs_trace.h"
 17
 18/*
 19 * Check to see if a buffer matching the given parameters is already
 20 * a part of the given transaction.
 21 */
 22STATIC struct xfs_buf *
 23xfs_trans_buf_item_match(
 24	struct xfs_trans	*tp,
 25	struct xfs_buftarg	*target,
 26	struct xfs_buf_map	*map,
 27	int			nmaps)
 28{
 29	struct xfs_log_item	*lip;
 30	struct xfs_buf_log_item	*blip;
 31	int			len = 0;
 32	int			i;
 33
 34	for (i = 0; i < nmaps; i++)
 35		len += map[i].bm_len;
 36
 37	list_for_each_entry(lip, &tp->t_items, li_trans) {
 38		blip = (struct xfs_buf_log_item *)lip;
 
 39		if (blip->bli_item.li_type == XFS_LI_BUF &&
 40		    blip->bli_buf->b_target == target &&
 41		    xfs_buf_daddr(blip->bli_buf) == map[0].bm_bn &&
 42		    blip->bli_buf->b_length == len) {
 43			ASSERT(blip->bli_buf->b_map_count == nmaps);
 44			return blip->bli_buf;
 45		}
 46	}
 47
 48	return NULL;
 49}
 50
 51/*
 52 * Add the locked buffer to the transaction.
 53 *
 54 * The buffer must be locked, and it cannot be associated with any
 55 * transaction.
 56 *
 57 * If the buffer does not yet have a buf log item associated with it,
 58 * then allocate one for it.  Then add the buf item to the transaction.
 59 */
 60STATIC void
 61_xfs_trans_bjoin(
 62	struct xfs_trans	*tp,
 63	struct xfs_buf		*bp,
 64	int			reset_recur)
 65{
 66	struct xfs_buf_log_item	*bip;
 67
 68	ASSERT(bp->b_transp == NULL);
 69
 70	/*
 71	 * The xfs_buf_log_item pointer is stored in b_log_item.  If
 72	 * it doesn't have one yet, then allocate one and initialize it.
 73	 * The checks to see if one is there are in xfs_buf_item_init().
 74	 */
 75	xfs_buf_item_init(bp, tp->t_mountp);
 76	bip = bp->b_log_item;
 77	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 78	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
 79	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
 80	if (reset_recur)
 81		bip->bli_recur = 0;
 82
 83	/*
 84	 * Take a reference for this transaction on the buf item.
 85	 */
 86	atomic_inc(&bip->bli_refcount);
 87
 88	/*
 89	 * Attach the item to the transaction so we can find it in
 90	 * xfs_trans_get_buf() and friends.
 91	 */
 92	xfs_trans_add_item(tp, &bip->bli_item);
 
 
 
 
 
 93	bp->b_transp = tp;
 94
 95}
 96
 97void
 98xfs_trans_bjoin(
 99	struct xfs_trans	*tp,
100	struct xfs_buf		*bp)
101{
102	_xfs_trans_bjoin(tp, bp, 0);
103	trace_xfs_trans_bjoin(bp->b_log_item);
104}
105
106/*
107 * Get and lock the buffer for the caller if it is not already
108 * locked within the given transaction.  If it is already locked
109 * within the transaction, just increment its lock recursion count
110 * and return a pointer to it.
111 *
112 * If the transaction pointer is NULL, make this just a normal
113 * get_buf() call.
114 */
115int
116xfs_trans_get_buf_map(
117	struct xfs_trans	*tp,
118	struct xfs_buftarg	*target,
119	struct xfs_buf_map	*map,
120	int			nmaps,
121	xfs_buf_flags_t		flags,
122	struct xfs_buf		**bpp)
123{
124	struct xfs_buf		*bp;
125	struct xfs_buf_log_item	*bip;
126	int			error;
127
128	*bpp = NULL;
129	if (!tp)
130		return xfs_buf_get_map(target, map, nmaps, flags, bpp);
 
 
131
132	/*
133	 * If we find the buffer in the cache with this transaction
134	 * pointer in its b_fsprivate2 field, then we know we already
135	 * have it locked.  In this case we just increment the lock
136	 * recursion count and return the buffer to the caller.
137	 */
138	bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
139	if (bp != NULL) {
140		ASSERT(xfs_buf_islocked(bp));
141		if (xfs_is_shutdown(tp->t_mountp)) {
142			xfs_buf_stale(bp);
143			bp->b_flags |= XBF_DONE;
144		}
145
146		ASSERT(bp->b_transp == tp);
147		bip = bp->b_log_item;
148		ASSERT(bip != NULL);
149		ASSERT(atomic_read(&bip->bli_refcount) > 0);
150		bip->bli_recur++;
151		trace_xfs_trans_get_buf_recur(bip);
152		*bpp = bp;
153		return 0;
154	}
155
156	error = xfs_buf_get_map(target, map, nmaps, flags, &bp);
157	if (error)
158		return error;
 
159
160	ASSERT(!bp->b_error);
161
162	_xfs_trans_bjoin(tp, bp, 1);
163	trace_xfs_trans_get_buf(bp->b_log_item);
164	*bpp = bp;
165	return 0;
166}
167
168/*
169 * Get and lock the superblock buffer for the given transaction.
 
 
 
 
 
170 */
171static struct xfs_buf *
172__xfs_trans_getsb(
173	struct xfs_trans	*tp,
174	struct xfs_buf		*bp)
175{
 
 
 
176	/*
177	 * Just increment the lock recursion count if the buffer is already
178	 * attached to this transaction.
179	 */
 
 
 
 
 
 
 
 
 
 
 
180	if (bp->b_transp == tp) {
181		struct xfs_buf_log_item	*bip = bp->b_log_item;
182
183		ASSERT(bip != NULL);
184		ASSERT(atomic_read(&bip->bli_refcount) > 0);
185		bip->bli_recur++;
186
187		trace_xfs_trans_getsb_recur(bip);
188	} else {
189		xfs_buf_lock(bp);
190		xfs_buf_hold(bp);
191		_xfs_trans_bjoin(tp, bp, 1);
192
193		trace_xfs_trans_getsb(bp->b_log_item);
194	}
195
196	return bp;
197}
 
198
199struct xfs_buf *
200xfs_trans_getsb(
201	struct xfs_trans	*tp)
202{
203	return __xfs_trans_getsb(tp, tp->t_mountp->m_sb_bp);
204}
205
206struct xfs_buf *
207xfs_trans_getrtsb(
208	struct xfs_trans	*tp)
209{
210	if (!tp->t_mountp->m_rtsb_bp)
211		return NULL;
212	return __xfs_trans_getsb(tp, tp->t_mountp->m_rtsb_bp);
213}
214
215/*
216 * Get and lock the buffer for the caller if it is not already
217 * locked within the given transaction.  If it has not yet been
218 * read in, read it from disk. If it is already locked
219 * within the transaction and already read in, just increment its
220 * lock recursion count and return a pointer to it.
221 *
222 * If the transaction pointer is NULL, make this just a normal
223 * read_buf() call.
224 */
225int
226xfs_trans_read_buf_map(
227	struct xfs_mount	*mp,
228	struct xfs_trans	*tp,
229	struct xfs_buftarg	*target,
230	struct xfs_buf_map	*map,
231	int			nmaps,
232	xfs_buf_flags_t		flags,
233	struct xfs_buf		**bpp,
234	const struct xfs_buf_ops *ops)
235{
236	struct xfs_buf		*bp = NULL;
237	struct xfs_buf_log_item	*bip;
238	int			error;
239
240	*bpp = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
241	/*
242	 * If we find the buffer in the cache with this transaction
243	 * pointer in its b_fsprivate2 field, then we know we already
244	 * have it locked.  If it is already read in we just increment
245	 * the lock recursion count and return the buffer to the caller.
246	 * If the buffer is not yet read in, then we read it in, increment
247	 * the lock recursion count, and return it to the caller.
248	 */
249	if (tp)
250		bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
251	if (bp) {
252		ASSERT(xfs_buf_islocked(bp));
253		ASSERT(bp->b_transp == tp);
254		ASSERT(bp->b_log_item != NULL);
255		ASSERT(!bp->b_error);
256		ASSERT(bp->b_flags & XBF_DONE);
257
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
258		/*
259		 * We never locked this buf ourselves, so we shouldn't
260		 * brelse it either. Just get out.
261		 */
262		if (xfs_is_shutdown(mp)) {
263			trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
264			return -EIO;
 
265		}
266
267		/*
268		 * Check if the caller is trying to read a buffer that is
269		 * already attached to the transaction yet has no buffer ops
270		 * assigned.  Ops are usually attached when the buffer is
271		 * attached to the transaction, or by the read caller if
272		 * special circumstances.  That didn't happen, which is not
273		 * how this is supposed to go.
274		 *
275		 * If the buffer passes verification we'll let this go, but if
276		 * not we have to shut down.  Let the transaction cleanup code
277		 * release this buffer when it kills the tranaction.
278		 */
279		ASSERT(bp->b_ops != NULL);
280		error = xfs_buf_reverify(bp, ops);
281		if (error) {
282			xfs_buf_ioerror_alert(bp, __return_address);
283
284			if (tp->t_flags & XFS_TRANS_DIRTY)
285				xfs_force_shutdown(tp->t_mountp,
286						SHUTDOWN_META_IO_ERROR);
287
288			/* bad CRC means corrupted metadata */
289			if (error == -EFSBADCRC)
290				error = -EFSCORRUPTED;
291			return error;
292		}
293
294		bip = bp->b_log_item;
295		bip->bli_recur++;
296
297		ASSERT(atomic_read(&bip->bli_refcount) > 0);
298		trace_xfs_trans_read_buf_recur(bip);
299		ASSERT(bp->b_ops != NULL || ops == NULL);
300		*bpp = bp;
301		return 0;
302	}
303
304	error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops,
305			__return_address);
306	switch (error) {
307	case 0:
308		break;
309	default:
310		if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
 
 
 
 
 
311			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
312		fallthrough;
313	case -ENOMEM:
314	case -EAGAIN:
315		return error;
316	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
317
318	if (xfs_is_shutdown(mp)) {
319		xfs_buf_relse(bp);
320		trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
321		return -EIO;
322	}
323
324	if (tp) {
325		_xfs_trans_bjoin(tp, bp, 1);
326		trace_xfs_trans_read_buf(bp->b_log_item);
327	}
328	ASSERT(bp->b_ops != NULL || ops == NULL);
329	*bpp = bp;
330	return 0;
331
 
 
 
 
 
332}
333
334/* Has this buffer been dirtied by anyone? */
335bool
336xfs_trans_buf_is_dirty(
337	struct xfs_buf		*bp)
338{
339	struct xfs_buf_log_item	*bip = bp->b_log_item;
340
341	if (!bip)
342		return false;
343	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
344	return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
345}
346
347/*
348 * Release a buffer previously joined to the transaction. If the buffer is
349 * modified within this transaction, decrement the recursion count but do not
350 * release the buffer even if the count goes to 0. If the buffer is not modified
351 * within the transaction, decrement the recursion count and release the buffer
352 * if the recursion count goes to 0.
 
 
353 *
354 * If the buffer is to be released and it was not already dirty before this
355 * transaction began, then also free the buf_log_item associated with it.
356 *
357 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
 
358 */
359void
360xfs_trans_brelse(
361	struct xfs_trans	*tp,
362	struct xfs_buf		*bp)
363{
364	struct xfs_buf_log_item	*bip = bp->b_log_item;
365
366	ASSERT(bp->b_transp == tp);
367
368	if (!tp) {
 
 
369		xfs_buf_relse(bp);
370		return;
371	}
372
373	trace_xfs_trans_brelse(bip);
 
374	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
 
 
375	ASSERT(atomic_read(&bip->bli_refcount) > 0);
376
 
 
377	/*
378	 * If the release is for a recursive lookup, then decrement the count
379	 * and return.
380	 */
381	if (bip->bli_recur > 0) {
382		bip->bli_recur--;
383		return;
384	}
385
386	/*
387	 * If the buffer is invalidated or dirty in this transaction, we can't
388	 * release it until we commit.
389	 */
390	if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
391		return;
 
 
 
 
 
 
 
392	if (bip->bli_flags & XFS_BLI_STALE)
393		return;
394
 
 
395	/*
396	 * Unlink the log item from the transaction and clear the hold flag, if
397	 * set. We wouldn't want the next user of the buffer to get confused.
398	 */
399	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
400	xfs_trans_del_item(&bip->bli_item);
401	bip->bli_flags &= ~XFS_BLI_HOLD;
402
403	/* drop the reference to the bli */
404	xfs_buf_item_put(bip);
405
406	bp->b_transp = NULL;
407	xfs_buf_relse(bp);
408}
409
410/*
411 * Forcibly detach a buffer previously joined to the transaction.  The caller
412 * will retain its locked reference to the buffer after this function returns.
413 * The buffer must be completely clean and must not be held to the transaction.
414 */
415void
416xfs_trans_bdetach(
417	struct xfs_trans	*tp,
418	struct xfs_buf		*bp)
419{
420	struct xfs_buf_log_item	*bip = bp->b_log_item;
421
422	ASSERT(tp != NULL);
423	ASSERT(bp->b_transp == tp);
424	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
425	ASSERT(atomic_read(&bip->bli_refcount) > 0);
426
427	trace_xfs_trans_bdetach(bip);
428
429	/*
430	 * Erase all recursion count, since we're removing this buffer from the
431	 * transaction.
432	 */
433	bip->bli_recur = 0;
434
435	/*
436	 * The buffer must be completely clean.  Specifically, it had better
437	 * not be dirty, stale, logged, ordered, or held to the transaction.
 
 
 
438	 */
439	ASSERT(!test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
440	ASSERT(!(bip->bli_flags & XFS_BLI_DIRTY));
441	ASSERT(!(bip->bli_flags & XFS_BLI_HOLD));
442	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
443	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
444	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 
 
 
445
446	/* Unlink the log item from the transaction and drop the log item. */
447	xfs_trans_del_item(&bip->bli_item);
448	xfs_buf_item_put(bip);
449	bp->b_transp = NULL;
 
450}
451
452/*
453 * Mark the buffer as not needing to be unlocked when the buf item's
454 * iop_committing() routine is called.  The buffer must already be locked
455 * and associated with the given transaction.
456 */
457/* ARGSUSED */
458void
459xfs_trans_bhold(
460	xfs_trans_t		*tp,
461	struct xfs_buf		*bp)
462{
463	struct xfs_buf_log_item	*bip = bp->b_log_item;
464
465	ASSERT(bp->b_transp == tp);
466	ASSERT(bip != NULL);
467	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
468	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
469	ASSERT(atomic_read(&bip->bli_refcount) > 0);
470
471	bip->bli_flags |= XFS_BLI_HOLD;
472	trace_xfs_trans_bhold(bip);
473}
474
475/*
476 * Cancel the previous buffer hold request made on this buffer
477 * for this transaction.
478 */
479void
480xfs_trans_bhold_release(
481	xfs_trans_t		*tp,
482	struct xfs_buf		*bp)
483{
484	struct xfs_buf_log_item	*bip = bp->b_log_item;
485
486	ASSERT(bp->b_transp == tp);
487	ASSERT(bip != NULL);
488	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
489	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
490	ASSERT(atomic_read(&bip->bli_refcount) > 0);
491	ASSERT(bip->bli_flags & XFS_BLI_HOLD);
492
493	bip->bli_flags &= ~XFS_BLI_HOLD;
494	trace_xfs_trans_bhold_release(bip);
495}
496
497/*
498 * Mark a buffer dirty in the transaction.
 
 
 
 
 
 
499 */
500void
501xfs_trans_dirty_buf(
502	struct xfs_trans	*tp,
503	struct xfs_buf		*bp)
 
504{
505	struct xfs_buf_log_item	*bip = bp->b_log_item;
506
507	ASSERT(bp->b_transp == tp);
508	ASSERT(bip != NULL);
 
 
 
509
510	/*
511	 * Mark the buffer as needing to be written out eventually,
512	 * and set its iodone function to remove the buffer's buf log
513	 * item from the AIL and free it when the buffer is flushed
514	 * to disk.
 
 
 
 
515	 */
516	bp->b_flags |= XBF_DONE;
517
518	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 
 
 
 
519
520	/*
521	 * If we invalidated the buffer within this transaction, then
522	 * cancel the invalidation now that we're dirtying the buffer
523	 * again.  There are no races with the code in xfs_buf_item_unpin(),
524	 * because we have a reference to the buffer this entire time.
525	 */
526	if (bip->bli_flags & XFS_BLI_STALE) {
527		bip->bli_flags &= ~XFS_BLI_STALE;
528		ASSERT(bp->b_flags & XBF_STALE);
529		bp->b_flags &= ~XBF_STALE;
530		bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
531	}
532	bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
533
534	tp->t_flags |= XFS_TRANS_DIRTY;
535	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
536}
537
538/*
539 * This is called to mark bytes first through last inclusive of the given
540 * buffer as needing to be logged when the transaction is committed.
541 * The buffer must already be associated with the given transaction.
542 *
543 * First and last are numbers relative to the beginning of this buffer,
544 * so the first byte in the buffer is numbered 0 regardless of the
545 * value of b_blkno.
546 */
547void
548xfs_trans_log_buf(
549	struct xfs_trans	*tp,
550	struct xfs_buf		*bp,
551	uint			first,
552	uint			last)
553{
554	struct xfs_buf_log_item	*bip = bp->b_log_item;
555
556	ASSERT(first <= last && last < BBTOB(bp->b_length));
557	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
558
559	xfs_trans_dirty_buf(tp, bp);
560
561	trace_xfs_trans_log_buf(bip);
562	xfs_buf_item_log(bip, first, last);
563}
564
565
566/*
567 * Invalidate a buffer that is being used within a transaction.
568 *
569 * Typically this is because the blocks in the buffer are being freed, so we
570 * need to prevent it from being written out when we're done.  Allowing it
571 * to be written again might overwrite data in the free blocks if they are
572 * reallocated to a file.
573 *
574 * We prevent the buffer from being written out by marking it stale.  We can't
575 * get rid of the buf log item at this point because the buffer may still be
576 * pinned by another transaction.  If that is the case, then we'll wait until
577 * the buffer is committed to disk for the last time (we can tell by the ref
578 * count) and free it in xfs_buf_item_unpin().  Until that happens we will
579 * keep the buffer locked so that the buffer and buf log item are not reused.
580 *
581 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
582 * the buf item.  This will be used at recovery time to determine that copies
583 * of the buffer in the log before this should not be replayed.
584 *
585 * We mark the item descriptor and the transaction dirty so that we'll hold
586 * the buffer until after the commit.
587 *
588 * Since we're invalidating the buffer, we also clear the state about which
589 * parts of the buffer have been logged.  We also clear the flag indicating
590 * that this is an inode buffer since the data in the buffer will no longer
591 * be valid.
592 *
593 * We set the stale bit in the buffer as well since we're getting rid of it.
594 */
595void
596xfs_trans_binval(
597	xfs_trans_t		*tp,
598	struct xfs_buf		*bp)
599{
600	struct xfs_buf_log_item	*bip = bp->b_log_item;
601	int			i;
602
603	ASSERT(bp->b_transp == tp);
604	ASSERT(bip != NULL);
605	ASSERT(atomic_read(&bip->bli_refcount) > 0);
606
607	trace_xfs_trans_binval(bip);
608
609	if (bip->bli_flags & XFS_BLI_STALE) {
610		/*
611		 * If the buffer is already invalidated, then
612		 * just return.
613		 */
614		ASSERT(bp->b_flags & XBF_STALE);
615		ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
616		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
617		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
618		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
619		ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
620		ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
621		return;
622	}
623
624	xfs_buf_stale(bp);
625
626	bip->bli_flags |= XFS_BLI_STALE;
627	bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
628	bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
629	bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
630	bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
631	for (i = 0; i < bip->bli_format_count; i++) {
632		memset(bip->bli_formats[i].blf_data_map, 0,
633		       (bip->bli_formats[i].blf_map_size * sizeof(uint)));
634	}
635	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
636	tp->t_flags |= XFS_TRANS_DIRTY;
637}
638
639/*
640 * This call is used to indicate that the buffer contains on-disk inodes which
641 * must be handled specially during recovery.  They require special handling
642 * because only the di_next_unlinked from the inodes in the buffer should be
643 * recovered.  The rest of the data in the buffer is logged via the inodes
644 * themselves.
645 *
646 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
647 * transferred to the buffer's log format structure so that we'll know what to
648 * do at recovery time.
649 */
650void
651xfs_trans_inode_buf(
652	xfs_trans_t		*tp,
653	struct xfs_buf		*bp)
654{
655	struct xfs_buf_log_item	*bip = bp->b_log_item;
656
657	ASSERT(bp->b_transp == tp);
658	ASSERT(bip != NULL);
659	ASSERT(atomic_read(&bip->bli_refcount) > 0);
660
661	bip->bli_flags |= XFS_BLI_INODE_BUF;
662	bp->b_flags |= _XBF_INODES;
663	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
664}
665
666/*
667 * This call is used to indicate that the buffer is going to
668 * be staled and was an inode buffer. This means it gets
669 * special processing during unpin - where any inodes
670 * associated with the buffer should be removed from ail.
671 * There is also special processing during recovery,
672 * any replay of the inodes in the buffer needs to be
673 * prevented as the buffer may have been reused.
674 */
675void
676xfs_trans_stale_inode_buf(
677	xfs_trans_t		*tp,
678	struct xfs_buf		*bp)
679{
680	struct xfs_buf_log_item	*bip = bp->b_log_item;
681
682	ASSERT(bp->b_transp == tp);
683	ASSERT(bip != NULL);
684	ASSERT(atomic_read(&bip->bli_refcount) > 0);
685
686	bip->bli_flags |= XFS_BLI_STALE_INODE;
687	bp->b_flags |= _XBF_INODES;
688	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
689}
690
691/*
692 * Mark the buffer as being one which contains newly allocated
693 * inodes.  We need to make sure that even if this buffer is
694 * relogged as an 'inode buf' we still recover all of the inode
695 * images in the face of a crash.  This works in coordination with
696 * xfs_buf_item_committed() to ensure that the buffer remains in the
697 * AIL at its original location even after it has been relogged.
698 */
699/* ARGSUSED */
700void
701xfs_trans_inode_alloc_buf(
702	xfs_trans_t		*tp,
703	struct xfs_buf		*bp)
704{
705	struct xfs_buf_log_item	*bip = bp->b_log_item;
706
707	ASSERT(bp->b_transp == tp);
708	ASSERT(bip != NULL);
709	ASSERT(atomic_read(&bip->bli_refcount) > 0);
710
711	bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
712	bp->b_flags |= _XBF_INODES;
713	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
714}
715
716/*
717 * Mark the buffer as ordered for this transaction. This means that the contents
718 * of the buffer are not recorded in the transaction but it is tracked in the
719 * AIL as though it was. This allows us to record logical changes in
720 * transactions rather than the physical changes we make to the buffer without
721 * changing writeback ordering constraints of metadata buffers.
722 */
723bool
724xfs_trans_ordered_buf(
725	struct xfs_trans	*tp,
726	struct xfs_buf		*bp)
727{
728	struct xfs_buf_log_item	*bip = bp->b_log_item;
729
730	ASSERT(bp->b_transp == tp);
731	ASSERT(bip != NULL);
732	ASSERT(atomic_read(&bip->bli_refcount) > 0);
733
734	if (xfs_buf_item_dirty_format(bip))
735		return false;
736
737	bip->bli_flags |= XFS_BLI_ORDERED;
738	trace_xfs_buf_item_ordered(bip);
739
740	/*
741	 * We don't log a dirty range of an ordered buffer but it still needs
742	 * to be marked dirty and that it has been logged.
743	 */
744	xfs_trans_dirty_buf(tp, bp);
745	return true;
746}
747
748/*
749 * Set the type of the buffer for log recovery so that it can correctly identify
750 * and hence attach the correct buffer ops to the buffer after replay.
751 */
752void
753xfs_trans_buf_set_type(
754	struct xfs_trans	*tp,
755	struct xfs_buf		*bp,
756	enum xfs_blft		type)
757{
758	struct xfs_buf_log_item	*bip = bp->b_log_item;
759
760	if (!tp)
761		return;
762
763	ASSERT(bp->b_transp == tp);
764	ASSERT(bip != NULL);
765	ASSERT(atomic_read(&bip->bli_refcount) > 0);
766
767	xfs_blft_to_flags(&bip->__bli_format, type);
768}
769
770void
771xfs_trans_buf_copy_type(
772	struct xfs_buf		*dst_bp,
773	struct xfs_buf		*src_bp)
774{
775	struct xfs_buf_log_item	*sbip = src_bp->b_log_item;
776	struct xfs_buf_log_item	*dbip = dst_bp->b_log_item;
777	enum xfs_blft		type;
778
779	type = xfs_blft_from_flags(&sbip->__bli_format);
780	xfs_blft_to_flags(&dbip->__bli_format, type);
781}
782
783/*
784 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
785 * dquots. However, unlike in inode buffer recovery, dquot buffers get
786 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
787 * The only thing that makes dquot buffers different from regular
788 * buffers is that we must not replay dquot bufs when recovering
789 * if a _corresponding_ quotaoff has happened. We also have to distinguish
790 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
791 * can be turned off independently.
792 */
793/* ARGSUSED */
794void
795xfs_trans_dquot_buf(
796	xfs_trans_t		*tp,
797	struct xfs_buf		*bp,
798	uint			type)
799{
800	struct xfs_buf_log_item	*bip = bp->b_log_item;
801
 
 
802	ASSERT(type == XFS_BLF_UDQUOT_BUF ||
803	       type == XFS_BLF_PDQUOT_BUF ||
804	       type == XFS_BLF_GDQUOT_BUF);
 
805
806	bip->__bli_format.blf_flags |= type;
807
808	switch (type) {
809	case XFS_BLF_UDQUOT_BUF:
810		type = XFS_BLFT_UDQUOT_BUF;
811		break;
812	case XFS_BLF_PDQUOT_BUF:
813		type = XFS_BLFT_PDQUOT_BUF;
814		break;
815	case XFS_BLF_GDQUOT_BUF:
816		type = XFS_BLFT_GDQUOT_BUF;
817		break;
818	default:
819		type = XFS_BLFT_UNKNOWN_BUF;
820		break;
821	}
822
823	bp->b_flags |= _XBF_DQUOTS;
824	xfs_trans_buf_set_type(tp, bp, type);
825}