xref: /linux/fs/xfs/xfs_extfree_item.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
1 /*
2  * Copyright (c) 2000-2001,2005 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_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_buf_item.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_extfree_item.h"
30 
31 
32 kmem_zone_t	*xfs_efi_zone;
33 kmem_zone_t	*xfs_efd_zone;
34 
35 static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
36 {
37 	return container_of(lip, struct xfs_efi_log_item, efi_item);
38 }
39 
40 void
41 xfs_efi_item_free(
42 	struct xfs_efi_log_item	*efip)
43 {
44 	if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
45 		kmem_free(efip);
46 	else
47 		kmem_zone_free(xfs_efi_zone, efip);
48 }
49 
50 /*
51  * Freeing the efi requires that we remove it from the AIL if it has already
52  * been placed there. However, the EFI may not yet have been placed in the AIL
53  * when called by xfs_efi_release() from EFD processing due to the ordering of
54  * committed vs unpin operations in bulk insert operations. Hence the
55  * test_and_clear_bit(XFS_EFI_COMMITTED) to ensure only the last caller frees
56  * the EFI.
57  */
58 STATIC void
59 __xfs_efi_release(
60 	struct xfs_efi_log_item	*efip)
61 {
62 	struct xfs_ail		*ailp = efip->efi_item.li_ailp;
63 
64 	if (!test_and_clear_bit(XFS_EFI_COMMITTED, &efip->efi_flags)) {
65 		spin_lock(&ailp->xa_lock);
66 		/* xfs_trans_ail_delete() drops the AIL lock. */
67 		xfs_trans_ail_delete(ailp, &efip->efi_item);
68 		xfs_efi_item_free(efip);
69 	}
70 }
71 
72 /*
73  * This returns the number of iovecs needed to log the given efi item.
74  * We only need 1 iovec for an efi item.  It just logs the efi_log_format
75  * structure.
76  */
77 STATIC uint
78 xfs_efi_item_size(
79 	struct xfs_log_item	*lip)
80 {
81 	return 1;
82 }
83 
84 /*
85  * This is called to fill in the vector of log iovecs for the
86  * given efi log item. We use only 1 iovec, and we point that
87  * at the efi_log_format structure embedded in the efi item.
88  * It is at this point that we assert that all of the extent
89  * slots in the efi item have been filled.
90  */
91 STATIC void
92 xfs_efi_item_format(
93 	struct xfs_log_item	*lip,
94 	struct xfs_log_iovec	*log_vector)
95 {
96 	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
97 	uint			size;
98 
99 	ASSERT(atomic_read(&efip->efi_next_extent) ==
100 				efip->efi_format.efi_nextents);
101 
102 	efip->efi_format.efi_type = XFS_LI_EFI;
103 
104 	size = sizeof(xfs_efi_log_format_t);
105 	size += (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
106 	efip->efi_format.efi_size = 1;
107 
108 	log_vector->i_addr = &efip->efi_format;
109 	log_vector->i_len = size;
110 	log_vector->i_type = XLOG_REG_TYPE_EFI_FORMAT;
111 	ASSERT(size >= sizeof(xfs_efi_log_format_t));
112 }
113 
114 
115 /*
116  * Pinning has no meaning for an efi item, so just return.
117  */
118 STATIC void
119 xfs_efi_item_pin(
120 	struct xfs_log_item	*lip)
121 {
122 }
123 
124 /*
125  * While EFIs cannot really be pinned, the unpin operation is the last place at
126  * which the EFI is manipulated during a transaction.  If we are being asked to
127  * remove the EFI it's because the transaction has been cancelled and by
128  * definition that means the EFI cannot be in the AIL so remove it from the
129  * transaction and free it.  Otherwise coordinate with xfs_efi_release() (via
130  * XFS_EFI_COMMITTED) to determine who gets to free the EFI.
131  */
132 STATIC void
133 xfs_efi_item_unpin(
134 	struct xfs_log_item	*lip,
135 	int			remove)
136 {
137 	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
138 
139 	if (remove) {
140 		ASSERT(!(lip->li_flags & XFS_LI_IN_AIL));
141 		if (lip->li_desc)
142 			xfs_trans_del_item(lip);
143 		xfs_efi_item_free(efip);
144 		return;
145 	}
146 	__xfs_efi_release(efip);
147 }
148 
149 /*
150  * Efi items have no locking or pushing.  However, since EFIs are
151  * pulled from the AIL when their corresponding EFDs are committed
152  * to disk, their situation is very similar to being pinned.  Return
153  * XFS_ITEM_PINNED so that the caller will eventually flush the log.
154  * This should help in getting the EFI out of the AIL.
155  */
156 STATIC uint
157 xfs_efi_item_trylock(
158 	struct xfs_log_item	*lip)
159 {
160 	return XFS_ITEM_PINNED;
161 }
162 
163 /*
164  * Efi items have no locking, so just return.
165  */
166 STATIC void
167 xfs_efi_item_unlock(
168 	struct xfs_log_item	*lip)
169 {
170 	if (lip->li_flags & XFS_LI_ABORTED)
171 		xfs_efi_item_free(EFI_ITEM(lip));
172 }
173 
174 /*
175  * The EFI is logged only once and cannot be moved in the log, so simply return
176  * the lsn at which it's been logged.  For bulk transaction committed
177  * processing, the EFI may be processed but not yet unpinned prior to the EFD
178  * being processed. Set the XFS_EFI_COMMITTED flag so this case can be detected
179  * when processing the EFD.
180  */
181 STATIC xfs_lsn_t
182 xfs_efi_item_committed(
183 	struct xfs_log_item	*lip,
184 	xfs_lsn_t		lsn)
185 {
186 	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
187 
188 	set_bit(XFS_EFI_COMMITTED, &efip->efi_flags);
189 	return lsn;
190 }
191 
192 /*
193  * There isn't much you can do to push on an efi item.  It is simply
194  * stuck waiting for all of its corresponding efd items to be
195  * committed to disk.
196  */
197 STATIC void
198 xfs_efi_item_push(
199 	struct xfs_log_item	*lip)
200 {
201 }
202 
203 /*
204  * The EFI dependency tracking op doesn't do squat.  It can't because
205  * it doesn't know where the free extent is coming from.  The dependency
206  * tracking has to be handled by the "enclosing" metadata object.  For
207  * example, for inodes, the inode is locked throughout the extent freeing
208  * so the dependency should be recorded there.
209  */
210 STATIC void
211 xfs_efi_item_committing(
212 	struct xfs_log_item	*lip,
213 	xfs_lsn_t		lsn)
214 {
215 }
216 
217 /*
218  * This is the ops vector shared by all efi log items.
219  */
220 static struct xfs_item_ops xfs_efi_item_ops = {
221 	.iop_size	= xfs_efi_item_size,
222 	.iop_format	= xfs_efi_item_format,
223 	.iop_pin	= xfs_efi_item_pin,
224 	.iop_unpin	= xfs_efi_item_unpin,
225 	.iop_trylock	= xfs_efi_item_trylock,
226 	.iop_unlock	= xfs_efi_item_unlock,
227 	.iop_committed	= xfs_efi_item_committed,
228 	.iop_push	= xfs_efi_item_push,
229 	.iop_committing = xfs_efi_item_committing
230 };
231 
232 
233 /*
234  * Allocate and initialize an efi item with the given number of extents.
235  */
236 struct xfs_efi_log_item *
237 xfs_efi_init(
238 	struct xfs_mount	*mp,
239 	uint			nextents)
240 
241 {
242 	struct xfs_efi_log_item	*efip;
243 	uint			size;
244 
245 	ASSERT(nextents > 0);
246 	if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
247 		size = (uint)(sizeof(xfs_efi_log_item_t) +
248 			((nextents - 1) * sizeof(xfs_extent_t)));
249 		efip = kmem_zalloc(size, KM_SLEEP);
250 	} else {
251 		efip = kmem_zone_zalloc(xfs_efi_zone, KM_SLEEP);
252 	}
253 
254 	xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
255 	efip->efi_format.efi_nextents = nextents;
256 	efip->efi_format.efi_id = (__psint_t)(void*)efip;
257 	atomic_set(&efip->efi_next_extent, 0);
258 
259 	return efip;
260 }
261 
262 /*
263  * Copy an EFI format buffer from the given buf, and into the destination
264  * EFI format structure.
265  * The given buffer can be in 32 bit or 64 bit form (which has different padding),
266  * one of which will be the native format for this kernel.
267  * It will handle the conversion of formats if necessary.
268  */
269 int
270 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
271 {
272 	xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
273 	uint i;
274 	uint len = sizeof(xfs_efi_log_format_t) +
275 		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
276 	uint len32 = sizeof(xfs_efi_log_format_32_t) +
277 		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
278 	uint len64 = sizeof(xfs_efi_log_format_64_t) +
279 		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
280 
281 	if (buf->i_len == len) {
282 		memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
283 		return 0;
284 	} else if (buf->i_len == len32) {
285 		xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
286 
287 		dst_efi_fmt->efi_type     = src_efi_fmt_32->efi_type;
288 		dst_efi_fmt->efi_size     = src_efi_fmt_32->efi_size;
289 		dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
290 		dst_efi_fmt->efi_id       = src_efi_fmt_32->efi_id;
291 		for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
292 			dst_efi_fmt->efi_extents[i].ext_start =
293 				src_efi_fmt_32->efi_extents[i].ext_start;
294 			dst_efi_fmt->efi_extents[i].ext_len =
295 				src_efi_fmt_32->efi_extents[i].ext_len;
296 		}
297 		return 0;
298 	} else if (buf->i_len == len64) {
299 		xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
300 
301 		dst_efi_fmt->efi_type     = src_efi_fmt_64->efi_type;
302 		dst_efi_fmt->efi_size     = src_efi_fmt_64->efi_size;
303 		dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
304 		dst_efi_fmt->efi_id       = src_efi_fmt_64->efi_id;
305 		for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
306 			dst_efi_fmt->efi_extents[i].ext_start =
307 				src_efi_fmt_64->efi_extents[i].ext_start;
308 			dst_efi_fmt->efi_extents[i].ext_len =
309 				src_efi_fmt_64->efi_extents[i].ext_len;
310 		}
311 		return 0;
312 	}
313 	return EFSCORRUPTED;
314 }
315 
316 /*
317  * This is called by the efd item code below to release references to the given
318  * efi item.  Each efd calls this with the number of extents that it has
319  * logged, and when the sum of these reaches the total number of extents logged
320  * by this efi item we can free the efi item.
321  */
322 void
323 xfs_efi_release(xfs_efi_log_item_t	*efip,
324 		uint			nextents)
325 {
326 	ASSERT(atomic_read(&efip->efi_next_extent) >= nextents);
327 	if (atomic_sub_and_test(nextents, &efip->efi_next_extent))
328 		__xfs_efi_release(efip);
329 }
330 
331 static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
332 {
333 	return container_of(lip, struct xfs_efd_log_item, efd_item);
334 }
335 
336 STATIC void
337 xfs_efd_item_free(struct xfs_efd_log_item *efdp)
338 {
339 	if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
340 		kmem_free(efdp);
341 	else
342 		kmem_zone_free(xfs_efd_zone, efdp);
343 }
344 
345 /*
346  * This returns the number of iovecs needed to log the given efd item.
347  * We only need 1 iovec for an efd item.  It just logs the efd_log_format
348  * structure.
349  */
350 STATIC uint
351 xfs_efd_item_size(
352 	struct xfs_log_item	*lip)
353 {
354 	return 1;
355 }
356 
357 /*
358  * This is called to fill in the vector of log iovecs for the
359  * given efd log item. We use only 1 iovec, and we point that
360  * at the efd_log_format structure embedded in the efd item.
361  * It is at this point that we assert that all of the extent
362  * slots in the efd item have been filled.
363  */
364 STATIC void
365 xfs_efd_item_format(
366 	struct xfs_log_item	*lip,
367 	struct xfs_log_iovec	*log_vector)
368 {
369 	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
370 	uint			size;
371 
372 	ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
373 
374 	efdp->efd_format.efd_type = XFS_LI_EFD;
375 
376 	size = sizeof(xfs_efd_log_format_t);
377 	size += (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
378 	efdp->efd_format.efd_size = 1;
379 
380 	log_vector->i_addr = &efdp->efd_format;
381 	log_vector->i_len = size;
382 	log_vector->i_type = XLOG_REG_TYPE_EFD_FORMAT;
383 	ASSERT(size >= sizeof(xfs_efd_log_format_t));
384 }
385 
386 /*
387  * Pinning has no meaning for an efd item, so just return.
388  */
389 STATIC void
390 xfs_efd_item_pin(
391 	struct xfs_log_item	*lip)
392 {
393 }
394 
395 /*
396  * Since pinning has no meaning for an efd item, unpinning does
397  * not either.
398  */
399 STATIC void
400 xfs_efd_item_unpin(
401 	struct xfs_log_item	*lip,
402 	int			remove)
403 {
404 }
405 
406 /*
407  * Efd items have no locking, so just return success.
408  */
409 STATIC uint
410 xfs_efd_item_trylock(
411 	struct xfs_log_item	*lip)
412 {
413 	return XFS_ITEM_LOCKED;
414 }
415 
416 /*
417  * Efd items have no locking or pushing, so return failure
418  * so that the caller doesn't bother with us.
419  */
420 STATIC void
421 xfs_efd_item_unlock(
422 	struct xfs_log_item	*lip)
423 {
424 	if (lip->li_flags & XFS_LI_ABORTED)
425 		xfs_efd_item_free(EFD_ITEM(lip));
426 }
427 
428 /*
429  * When the efd item is committed to disk, all we need to do
430  * is delete our reference to our partner efi item and then
431  * free ourselves.  Since we're freeing ourselves we must
432  * return -1 to keep the transaction code from further referencing
433  * this item.
434  */
435 STATIC xfs_lsn_t
436 xfs_efd_item_committed(
437 	struct xfs_log_item	*lip,
438 	xfs_lsn_t		lsn)
439 {
440 	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
441 
442 	/*
443 	 * If we got a log I/O error, it's always the case that the LR with the
444 	 * EFI got unpinned and freed before the EFD got aborted.
445 	 */
446 	if (!(lip->li_flags & XFS_LI_ABORTED))
447 		xfs_efi_release(efdp->efd_efip, efdp->efd_format.efd_nextents);
448 
449 	xfs_efd_item_free(efdp);
450 	return (xfs_lsn_t)-1;
451 }
452 
453 /*
454  * There isn't much you can do to push on an efd item.  It is simply
455  * stuck waiting for the log to be flushed to disk.
456  */
457 STATIC void
458 xfs_efd_item_push(
459 	struct xfs_log_item	*lip)
460 {
461 }
462 
463 /*
464  * The EFD dependency tracking op doesn't do squat.  It can't because
465  * it doesn't know where the free extent is coming from.  The dependency
466  * tracking has to be handled by the "enclosing" metadata object.  For
467  * example, for inodes, the inode is locked throughout the extent freeing
468  * so the dependency should be recorded there.
469  */
470 STATIC void
471 xfs_efd_item_committing(
472 	struct xfs_log_item	*lip,
473 	xfs_lsn_t		lsn)
474 {
475 }
476 
477 /*
478  * This is the ops vector shared by all efd log items.
479  */
480 static struct xfs_item_ops xfs_efd_item_ops = {
481 	.iop_size	= xfs_efd_item_size,
482 	.iop_format	= xfs_efd_item_format,
483 	.iop_pin	= xfs_efd_item_pin,
484 	.iop_unpin	= xfs_efd_item_unpin,
485 	.iop_trylock	= xfs_efd_item_trylock,
486 	.iop_unlock	= xfs_efd_item_unlock,
487 	.iop_committed	= xfs_efd_item_committed,
488 	.iop_push	= xfs_efd_item_push,
489 	.iop_committing = xfs_efd_item_committing
490 };
491 
492 /*
493  * Allocate and initialize an efd item with the given number of extents.
494  */
495 struct xfs_efd_log_item *
496 xfs_efd_init(
497 	struct xfs_mount	*mp,
498 	struct xfs_efi_log_item	*efip,
499 	uint			nextents)
500 
501 {
502 	struct xfs_efd_log_item	*efdp;
503 	uint			size;
504 
505 	ASSERT(nextents > 0);
506 	if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
507 		size = (uint)(sizeof(xfs_efd_log_item_t) +
508 			((nextents - 1) * sizeof(xfs_extent_t)));
509 		efdp = kmem_zalloc(size, KM_SLEEP);
510 	} else {
511 		efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
512 	}
513 
514 	xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
515 	efdp->efd_efip = efip;
516 	efdp->efd_format.efd_nextents = nextents;
517 	efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
518 
519 	return efdp;
520 }
521