2 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
24 #include "xfs_buf_item.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_extfree_item.h"
32 kmem_zone_t *xfs_efi_zone;
33 kmem_zone_t *xfs_efd_zone;
35 STATIC void xfs_efi_item_unlock(xfs_efi_log_item_t *);
36 STATIC void xfs_efi_item_abort(xfs_efi_log_item_t *);
37 STATIC void xfs_efd_item_abort(xfs_efd_log_item_t *);
41 xfs_efi_item_free(xfs_efi_log_item_t *efip)
43 int nexts = efip->efi_format.efi_nextents;
45 if (nexts > XFS_EFI_MAX_FAST_EXTENTS) {
46 kmem_free(efip, sizeof(xfs_efi_log_item_t) +
47 (nexts - 1) * sizeof(xfs_extent_t));
49 kmem_zone_free(xfs_efi_zone, efip);
54 * This returns the number of iovecs needed to log the given efi item.
55 * We only need 1 iovec for an efi item. It just logs the efi_log_format
60 xfs_efi_item_size(xfs_efi_log_item_t *efip)
66 * This is called to fill in the vector of log iovecs for the
67 * given efi log item. We use only 1 iovec, and we point that
68 * at the efi_log_format structure embedded in the efi item.
69 * It is at this point that we assert that all of the extent
70 * slots in the efi item have been filled.
73 xfs_efi_item_format(xfs_efi_log_item_t *efip,
74 xfs_log_iovec_t *log_vector)
78 ASSERT(efip->efi_next_extent == efip->efi_format.efi_nextents);
80 efip->efi_format.efi_type = XFS_LI_EFI;
82 size = sizeof(xfs_efi_log_format_t);
83 size += (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
84 efip->efi_format.efi_size = 1;
86 log_vector->i_addr = (xfs_caddr_t)&(efip->efi_format);
87 log_vector->i_len = size;
88 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_EFI_FORMAT);
89 ASSERT(size >= sizeof(xfs_efi_log_format_t));
94 * Pinning has no meaning for an efi item, so just return.
98 xfs_efi_item_pin(xfs_efi_log_item_t *efip)
105 * While EFIs cannot really be pinned, the unpin operation is the
106 * last place at which the EFI is manipulated during a transaction.
107 * Here we coordinate with xfs_efi_cancel() to determine who gets to
112 xfs_efi_item_unpin(xfs_efi_log_item_t *efip, int stale)
117 mp = efip->efi_item.li_mountp;
119 if (efip->efi_flags & XFS_EFI_CANCELED) {
121 * xfs_trans_delete_ail() drops the AIL lock.
123 xfs_trans_delete_ail(mp, (xfs_log_item_t *)efip, s);
124 xfs_efi_item_free(efip);
126 efip->efi_flags |= XFS_EFI_COMMITTED;
132 * like unpin only we have to also clear the xaction descriptor
133 * pointing the log item if we free the item. This routine duplicates
134 * unpin because efi_flags is protected by the AIL lock. Freeing
135 * the descriptor and then calling unpin would force us to drop the AIL
136 * lock which would open up a race condition.
139 xfs_efi_item_unpin_remove(xfs_efi_log_item_t *efip, xfs_trans_t *tp)
142 xfs_log_item_desc_t *lidp;
145 mp = efip->efi_item.li_mountp;
147 if (efip->efi_flags & XFS_EFI_CANCELED) {
149 * free the xaction descriptor pointing to this item
151 lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) efip);
152 xfs_trans_free_item(tp, lidp);
154 * pull the item off the AIL.
155 * xfs_trans_delete_ail() drops the AIL lock.
157 xfs_trans_delete_ail(mp, (xfs_log_item_t *)efip, s);
158 xfs_efi_item_free(efip);
160 efip->efi_flags |= XFS_EFI_COMMITTED;
166 * Efi items have no locking or pushing. However, since EFIs are
167 * pulled from the AIL when their corresponding EFDs are committed
168 * to disk, their situation is very similar to being pinned. Return
169 * XFS_ITEM_PINNED so that the caller will eventually flush the log.
170 * This should help in getting the EFI out of the AIL.
174 xfs_efi_item_trylock(xfs_efi_log_item_t *efip)
176 return XFS_ITEM_PINNED;
180 * Efi items have no locking, so just return.
184 xfs_efi_item_unlock(xfs_efi_log_item_t *efip)
186 if (efip->efi_item.li_flags & XFS_LI_ABORTED)
187 xfs_efi_item_abort(efip);
192 * The EFI is logged only once and cannot be moved in the log, so
193 * simply return the lsn at which it's been logged. The canceled
194 * flag is not paid any attention here. Checking for that is delayed
195 * until the EFI is unpinned.
199 xfs_efi_item_committed(xfs_efi_log_item_t *efip, xfs_lsn_t lsn)
205 * This is called when the transaction logging the EFI is aborted.
206 * Free up the EFI and return. No need to clean up the slot for
207 * the item in the transaction. That was done by the unpin code
208 * which is called prior to this routine in the abort/fs-shutdown path.
211 xfs_efi_item_abort(xfs_efi_log_item_t *efip)
213 xfs_efi_item_free(efip);
217 * There isn't much you can do to push on an efi item. It is simply
218 * stuck waiting for all of its corresponding efd items to be
223 xfs_efi_item_push(xfs_efi_log_item_t *efip)
229 * The EFI dependency tracking op doesn't do squat. It can't because
230 * it doesn't know where the free extent is coming from. The dependency
231 * tracking has to be handled by the "enclosing" metadata object. For
232 * example, for inodes, the inode is locked throughout the extent freeing
233 * so the dependency should be recorded there.
237 xfs_efi_item_committing(xfs_efi_log_item_t *efip, xfs_lsn_t lsn)
243 * This is the ops vector shared by all efi log items.
245 STATIC struct xfs_item_ops xfs_efi_item_ops = {
246 .iop_size = (uint(*)(xfs_log_item_t*))xfs_efi_item_size,
247 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
249 .iop_pin = (void(*)(xfs_log_item_t*))xfs_efi_item_pin,
250 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_efi_item_unpin,
251 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
252 xfs_efi_item_unpin_remove,
253 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_efi_item_trylock,
254 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_efi_item_unlock,
255 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
256 xfs_efi_item_committed,
257 .iop_push = (void(*)(xfs_log_item_t*))xfs_efi_item_push,
258 .iop_abort = (void(*)(xfs_log_item_t*))xfs_efi_item_abort,
260 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
261 xfs_efi_item_committing
266 * Allocate and initialize an efi item with the given number of extents.
269 xfs_efi_init(xfs_mount_t *mp,
273 xfs_efi_log_item_t *efip;
276 ASSERT(nextents > 0);
277 if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
278 size = (uint)(sizeof(xfs_efi_log_item_t) +
279 ((nextents - 1) * sizeof(xfs_extent_t)));
280 efip = (xfs_efi_log_item_t*)kmem_zalloc(size, KM_SLEEP);
282 efip = (xfs_efi_log_item_t*)kmem_zone_zalloc(xfs_efi_zone,
286 efip->efi_item.li_type = XFS_LI_EFI;
287 efip->efi_item.li_ops = &xfs_efi_item_ops;
288 efip->efi_item.li_mountp = mp;
289 efip->efi_format.efi_nextents = nextents;
290 efip->efi_format.efi_id = (__psint_t)(void*)efip;
296 * Copy an EFI format buffer from the given buf, and into the destination
297 * EFI format structure.
298 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
299 * one of which will be the native format for this kernel.
300 * It will handle the conversion of formats if necessary.
303 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
305 xfs_efi_log_format_t *src_efi_fmt = (xfs_efi_log_format_t *)buf->i_addr;
307 uint len = sizeof(xfs_efi_log_format_t) +
308 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
309 uint len32 = sizeof(xfs_efi_log_format_32_t) +
310 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
311 uint len64 = sizeof(xfs_efi_log_format_64_t) +
312 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
314 if (buf->i_len == len) {
315 memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
317 } else if (buf->i_len == len32) {
318 xfs_efi_log_format_32_t *src_efi_fmt_32 =
319 (xfs_efi_log_format_32_t *)buf->i_addr;
321 dst_efi_fmt->efi_type = src_efi_fmt_32->efi_type;
322 dst_efi_fmt->efi_size = src_efi_fmt_32->efi_size;
323 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
324 dst_efi_fmt->efi_id = src_efi_fmt_32->efi_id;
325 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
326 dst_efi_fmt->efi_extents[i].ext_start =
327 src_efi_fmt_32->efi_extents[i].ext_start;
328 dst_efi_fmt->efi_extents[i].ext_len =
329 src_efi_fmt_32->efi_extents[i].ext_len;
332 } else if (buf->i_len == len64) {
333 xfs_efi_log_format_64_t *src_efi_fmt_64 =
334 (xfs_efi_log_format_64_t *)buf->i_addr;
336 dst_efi_fmt->efi_type = src_efi_fmt_64->efi_type;
337 dst_efi_fmt->efi_size = src_efi_fmt_64->efi_size;
338 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
339 dst_efi_fmt->efi_id = src_efi_fmt_64->efi_id;
340 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
341 dst_efi_fmt->efi_extents[i].ext_start =
342 src_efi_fmt_64->efi_extents[i].ext_start;
343 dst_efi_fmt->efi_extents[i].ext_len =
344 src_efi_fmt_64->efi_extents[i].ext_len;
352 * This is called by the efd item code below to release references to
353 * the given efi item. Each efd calls this with the number of
354 * extents that it has logged, and when the sum of these reaches
355 * the total number of extents logged by this efi item we can free
358 * Freeing the efi item requires that we remove it from the AIL.
359 * We'll use the AIL lock to protect our counters as well as
360 * the removal from the AIL.
363 xfs_efi_release(xfs_efi_log_item_t *efip,
370 mp = efip->efi_item.li_mountp;
371 ASSERT(efip->efi_next_extent > 0);
372 ASSERT(efip->efi_flags & XFS_EFI_COMMITTED);
375 ASSERT(efip->efi_next_extent >= nextents);
376 efip->efi_next_extent -= nextents;
377 extents_left = efip->efi_next_extent;
378 if (extents_left == 0) {
380 * xfs_trans_delete_ail() drops the AIL lock.
382 xfs_trans_delete_ail(mp, (xfs_log_item_t *)efip, s);
383 xfs_efi_item_free(efip);
390 * This is called when the transaction that should be committing the
391 * EFD corresponding to the given EFI is aborted. The committed and
392 * canceled flags are used to coordinate the freeing of the EFI and
393 * the references by the transaction that committed it.
397 xfs_efi_log_item_t *efip)
402 mp = efip->efi_item.li_mountp;
404 if (efip->efi_flags & XFS_EFI_COMMITTED) {
406 * xfs_trans_delete_ail() drops the AIL lock.
408 xfs_trans_delete_ail(mp, (xfs_log_item_t *)efip, s);
409 xfs_efi_item_free(efip);
411 efip->efi_flags |= XFS_EFI_CANCELED;
417 xfs_efd_item_free(xfs_efd_log_item_t *efdp)
419 int nexts = efdp->efd_format.efd_nextents;
421 if (nexts > XFS_EFD_MAX_FAST_EXTENTS) {
422 kmem_free(efdp, sizeof(xfs_efd_log_item_t) +
423 (nexts - 1) * sizeof(xfs_extent_t));
425 kmem_zone_free(xfs_efd_zone, efdp);
430 * This returns the number of iovecs needed to log the given efd item.
431 * We only need 1 iovec for an efd item. It just logs the efd_log_format
436 xfs_efd_item_size(xfs_efd_log_item_t *efdp)
442 * This is called to fill in the vector of log iovecs for the
443 * given efd log item. We use only 1 iovec, and we point that
444 * at the efd_log_format structure embedded in the efd item.
445 * It is at this point that we assert that all of the extent
446 * slots in the efd item have been filled.
449 xfs_efd_item_format(xfs_efd_log_item_t *efdp,
450 xfs_log_iovec_t *log_vector)
454 ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
456 efdp->efd_format.efd_type = XFS_LI_EFD;
458 size = sizeof(xfs_efd_log_format_t);
459 size += (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
460 efdp->efd_format.efd_size = 1;
462 log_vector->i_addr = (xfs_caddr_t)&(efdp->efd_format);
463 log_vector->i_len = size;
464 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_EFD_FORMAT);
465 ASSERT(size >= sizeof(xfs_efd_log_format_t));
470 * Pinning has no meaning for an efd item, so just return.
474 xfs_efd_item_pin(xfs_efd_log_item_t *efdp)
481 * Since pinning has no meaning for an efd item, unpinning does
486 xfs_efd_item_unpin(xfs_efd_log_item_t *efdp, int stale)
493 xfs_efd_item_unpin_remove(xfs_efd_log_item_t *efdp, xfs_trans_t *tp)
499 * Efd items have no locking, so just return success.
503 xfs_efd_item_trylock(xfs_efd_log_item_t *efdp)
505 return XFS_ITEM_LOCKED;
509 * Efd items have no locking or pushing, so return failure
510 * so that the caller doesn't bother with us.
514 xfs_efd_item_unlock(xfs_efd_log_item_t *efdp)
516 if (efdp->efd_item.li_flags & XFS_LI_ABORTED)
517 xfs_efd_item_abort(efdp);
522 * When the efd item is committed to disk, all we need to do
523 * is delete our reference to our partner efi item and then
524 * free ourselves. Since we're freeing ourselves we must
525 * return -1 to keep the transaction code from further referencing
530 xfs_efd_item_committed(xfs_efd_log_item_t *efdp, xfs_lsn_t lsn)
533 * If we got a log I/O error, it's always the case that the LR with the
534 * EFI got unpinned and freed before the EFD got aborted.
536 if ((efdp->efd_item.li_flags & XFS_LI_ABORTED) == 0)
537 xfs_efi_release(efdp->efd_efip, efdp->efd_format.efd_nextents);
539 xfs_efd_item_free(efdp);
540 return (xfs_lsn_t)-1;
544 * The transaction of which this EFD is a part has been aborted.
545 * Inform its companion EFI of this fact and then clean up after
546 * ourselves. No need to clean up the slot for the item in the
547 * transaction. That was done by the unpin code which is called
548 * prior to this routine in the abort/fs-shutdown path.
551 xfs_efd_item_abort(xfs_efd_log_item_t *efdp)
554 * If we got a log I/O error, it's always the case that the LR with the
555 * EFI got unpinned and freed before the EFD got aborted. So don't
556 * reference the EFI at all in that case.
558 if ((efdp->efd_item.li_flags & XFS_LI_ABORTED) == 0)
559 xfs_efi_cancel(efdp->efd_efip);
561 xfs_efd_item_free(efdp);
565 * There isn't much you can do to push on an efd item. It is simply
566 * stuck waiting for the log to be flushed to disk.
570 xfs_efd_item_push(xfs_efd_log_item_t *efdp)
576 * The EFD dependency tracking op doesn't do squat. It can't because
577 * it doesn't know where the free extent is coming from. The dependency
578 * tracking has to be handled by the "enclosing" metadata object. For
579 * example, for inodes, the inode is locked throughout the extent freeing
580 * so the dependency should be recorded there.
584 xfs_efd_item_committing(xfs_efd_log_item_t *efip, xfs_lsn_t lsn)
590 * This is the ops vector shared by all efd log items.
592 STATIC struct xfs_item_ops xfs_efd_item_ops = {
593 .iop_size = (uint(*)(xfs_log_item_t*))xfs_efd_item_size,
594 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
596 .iop_pin = (void(*)(xfs_log_item_t*))xfs_efd_item_pin,
597 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_efd_item_unpin,
598 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
599 xfs_efd_item_unpin_remove,
600 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_efd_item_trylock,
601 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_efd_item_unlock,
602 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
603 xfs_efd_item_committed,
604 .iop_push = (void(*)(xfs_log_item_t*))xfs_efd_item_push,
605 .iop_abort = (void(*)(xfs_log_item_t*))xfs_efd_item_abort,
607 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
608 xfs_efd_item_committing
613 * Allocate and initialize an efd item with the given number of extents.
616 xfs_efd_init(xfs_mount_t *mp,
617 xfs_efi_log_item_t *efip,
621 xfs_efd_log_item_t *efdp;
624 ASSERT(nextents > 0);
625 if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
626 size = (uint)(sizeof(xfs_efd_log_item_t) +
627 ((nextents - 1) * sizeof(xfs_extent_t)));
628 efdp = (xfs_efd_log_item_t*)kmem_zalloc(size, KM_SLEEP);
630 efdp = (xfs_efd_log_item_t*)kmem_zone_zalloc(xfs_efd_zone,
634 efdp->efd_item.li_type = XFS_LI_EFD;
635 efdp->efd_item.li_ops = &xfs_efd_item_ops;
636 efdp->efd_item.li_mountp = mp;
637 efdp->efd_efip = efip;
638 efdp->efd_format.efd_nextents = nextents;
639 efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;