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"
27 #include "xfs_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_trans_priv.h"
30 #include "xfs_extfree_item.h"
33 kmem_zone_t *xfs_efi_zone;
34 kmem_zone_t *xfs_efd_zone;
36 STATIC void xfs_efi_item_unlock(xfs_efi_log_item_t *);
39 xfs_efi_item_free(xfs_efi_log_item_t *efip)
41 int nexts = efip->efi_format.efi_nextents;
43 if (nexts > XFS_EFI_MAX_FAST_EXTENTS) {
46 kmem_zone_free(xfs_efi_zone, efip);
51 * This returns the number of iovecs needed to log the given efi item.
52 * We only need 1 iovec for an efi item. It just logs the efi_log_format
57 xfs_efi_item_size(xfs_efi_log_item_t *efip)
63 * This is called to fill in the vector of log iovecs for the
64 * given efi log item. We use only 1 iovec, and we point that
65 * at the efi_log_format structure embedded in the efi item.
66 * It is at this point that we assert that all of the extent
67 * slots in the efi item have been filled.
70 xfs_efi_item_format(xfs_efi_log_item_t *efip,
71 xfs_log_iovec_t *log_vector)
75 ASSERT(efip->efi_next_extent == efip->efi_format.efi_nextents);
77 efip->efi_format.efi_type = XFS_LI_EFI;
79 size = sizeof(xfs_efi_log_format_t);
80 size += (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
81 efip->efi_format.efi_size = 1;
83 log_vector->i_addr = (xfs_caddr_t)&(efip->efi_format);
84 log_vector->i_len = size;
85 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_EFI_FORMAT);
86 ASSERT(size >= sizeof(xfs_efi_log_format_t));
91 * Pinning has no meaning for an efi item, so just return.
95 xfs_efi_item_pin(xfs_efi_log_item_t *efip)
102 * While EFIs cannot really be pinned, the unpin operation is the
103 * last place at which the EFI is manipulated during a transaction.
104 * Here we coordinate with xfs_efi_cancel() to determine who gets to
109 xfs_efi_item_unpin(xfs_efi_log_item_t *efip, int stale)
112 struct xfs_ail *ailp;
114 mp = efip->efi_item.li_mountp;
115 ailp = efip->efi_item.li_ailp;
116 spin_lock(&ailp->xa_lock);
117 if (efip->efi_flags & XFS_EFI_CANCELED) {
119 * xfs_trans_delete_ail() drops the AIL lock.
121 xfs_trans_delete_ail(mp, (xfs_log_item_t *)efip);
122 xfs_efi_item_free(efip);
124 efip->efi_flags |= XFS_EFI_COMMITTED;
125 spin_unlock(&ailp->xa_lock);
130 * like unpin only we have to also clear the xaction descriptor
131 * pointing the log item if we free the item. This routine duplicates
132 * unpin because efi_flags is protected by the AIL lock. Freeing
133 * the descriptor and then calling unpin would force us to drop the AIL
134 * lock which would open up a race condition.
137 xfs_efi_item_unpin_remove(xfs_efi_log_item_t *efip, xfs_trans_t *tp)
140 struct xfs_ail *ailp;
141 xfs_log_item_desc_t *lidp;
143 mp = efip->efi_item.li_mountp;
144 ailp = efip->efi_item.li_ailp;
145 spin_lock(&ailp->xa_lock);
146 if (efip->efi_flags & XFS_EFI_CANCELED) {
148 * free the xaction descriptor pointing to this item
150 lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) efip);
151 xfs_trans_free_item(tp, lidp);
153 * pull the item off the AIL.
154 * xfs_trans_delete_ail() drops the AIL lock.
156 xfs_trans_delete_ail(mp, (xfs_log_item_t *)efip);
157 xfs_efi_item_free(efip);
159 efip->efi_flags |= XFS_EFI_COMMITTED;
160 spin_unlock(&ailp->xa_lock);
165 * Efi items have no locking or pushing. However, since EFIs are
166 * pulled from the AIL when their corresponding EFDs are committed
167 * to disk, their situation is very similar to being pinned. Return
168 * XFS_ITEM_PINNED so that the caller will eventually flush the log.
169 * This should help in getting the EFI out of the AIL.
173 xfs_efi_item_trylock(xfs_efi_log_item_t *efip)
175 return XFS_ITEM_PINNED;
179 * Efi items have no locking, so just return.
183 xfs_efi_item_unlock(xfs_efi_log_item_t *efip)
185 if (efip->efi_item.li_flags & XFS_LI_ABORTED)
186 xfs_efi_item_free(efip);
191 * The EFI is logged only once and cannot be moved in the log, so
192 * simply return the lsn at which it's been logged. The canceled
193 * flag is not paid any attention here. Checking for that is delayed
194 * until the EFI is unpinned.
198 xfs_efi_item_committed(xfs_efi_log_item_t *efip, xfs_lsn_t lsn)
204 * There isn't much you can do to push on an efi item. It is simply
205 * stuck waiting for all of its corresponding efd items to be
210 xfs_efi_item_push(xfs_efi_log_item_t *efip)
216 * The EFI dependency tracking op doesn't do squat. It can't because
217 * it doesn't know where the free extent is coming from. The dependency
218 * tracking has to be handled by the "enclosing" metadata object. For
219 * example, for inodes, the inode is locked throughout the extent freeing
220 * so the dependency should be recorded there.
224 xfs_efi_item_committing(xfs_efi_log_item_t *efip, xfs_lsn_t lsn)
230 * This is the ops vector shared by all efi log items.
232 static struct xfs_item_ops xfs_efi_item_ops = {
233 .iop_size = (uint(*)(xfs_log_item_t*))xfs_efi_item_size,
234 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
236 .iop_pin = (void(*)(xfs_log_item_t*))xfs_efi_item_pin,
237 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_efi_item_unpin,
238 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
239 xfs_efi_item_unpin_remove,
240 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_efi_item_trylock,
241 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_efi_item_unlock,
242 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
243 xfs_efi_item_committed,
244 .iop_push = (void(*)(xfs_log_item_t*))xfs_efi_item_push,
246 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
247 xfs_efi_item_committing
252 * Allocate and initialize an efi item with the given number of extents.
255 xfs_efi_init(xfs_mount_t *mp,
259 xfs_efi_log_item_t *efip;
262 ASSERT(nextents > 0);
263 if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
264 size = (uint)(sizeof(xfs_efi_log_item_t) +
265 ((nextents - 1) * sizeof(xfs_extent_t)));
266 efip = (xfs_efi_log_item_t*)kmem_zalloc(size, KM_SLEEP);
268 efip = (xfs_efi_log_item_t*)kmem_zone_zalloc(xfs_efi_zone,
272 efip->efi_item.li_type = XFS_LI_EFI;
273 efip->efi_item.li_ops = &xfs_efi_item_ops;
274 efip->efi_item.li_mountp = mp;
275 efip->efi_item.li_ailp = mp->m_ail;
276 efip->efi_format.efi_nextents = nextents;
277 efip->efi_format.efi_id = (__psint_t)(void*)efip;
283 * Copy an EFI format buffer from the given buf, and into the destination
284 * EFI format structure.
285 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
286 * one of which will be the native format for this kernel.
287 * It will handle the conversion of formats if necessary.
290 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
292 xfs_efi_log_format_t *src_efi_fmt = (xfs_efi_log_format_t *)buf->i_addr;
294 uint len = sizeof(xfs_efi_log_format_t) +
295 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
296 uint len32 = sizeof(xfs_efi_log_format_32_t) +
297 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
298 uint len64 = sizeof(xfs_efi_log_format_64_t) +
299 (src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
301 if (buf->i_len == len) {
302 memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
304 } else if (buf->i_len == len32) {
305 xfs_efi_log_format_32_t *src_efi_fmt_32 =
306 (xfs_efi_log_format_32_t *)buf->i_addr;
308 dst_efi_fmt->efi_type = src_efi_fmt_32->efi_type;
309 dst_efi_fmt->efi_size = src_efi_fmt_32->efi_size;
310 dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
311 dst_efi_fmt->efi_id = src_efi_fmt_32->efi_id;
312 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
313 dst_efi_fmt->efi_extents[i].ext_start =
314 src_efi_fmt_32->efi_extents[i].ext_start;
315 dst_efi_fmt->efi_extents[i].ext_len =
316 src_efi_fmt_32->efi_extents[i].ext_len;
319 } else if (buf->i_len == len64) {
320 xfs_efi_log_format_64_t *src_efi_fmt_64 =
321 (xfs_efi_log_format_64_t *)buf->i_addr;
323 dst_efi_fmt->efi_type = src_efi_fmt_64->efi_type;
324 dst_efi_fmt->efi_size = src_efi_fmt_64->efi_size;
325 dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
326 dst_efi_fmt->efi_id = src_efi_fmt_64->efi_id;
327 for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
328 dst_efi_fmt->efi_extents[i].ext_start =
329 src_efi_fmt_64->efi_extents[i].ext_start;
330 dst_efi_fmt->efi_extents[i].ext_len =
331 src_efi_fmt_64->efi_extents[i].ext_len;
339 * This is called by the efd item code below to release references to
340 * the given efi item. Each efd calls this with the number of
341 * extents that it has logged, and when the sum of these reaches
342 * the total number of extents logged by this efi item we can free
345 * Freeing the efi item requires that we remove it from the AIL.
346 * We'll use the AIL lock to protect our counters as well as
347 * the removal from the AIL.
350 xfs_efi_release(xfs_efi_log_item_t *efip,
354 struct xfs_ail *ailp;
357 mp = efip->efi_item.li_mountp;
358 ailp = efip->efi_item.li_ailp;
359 ASSERT(efip->efi_next_extent > 0);
360 ASSERT(efip->efi_flags & XFS_EFI_COMMITTED);
362 spin_lock(&ailp->xa_lock);
363 ASSERT(efip->efi_next_extent >= nextents);
364 efip->efi_next_extent -= nextents;
365 extents_left = efip->efi_next_extent;
366 if (extents_left == 0) {
368 * xfs_trans_delete_ail() drops the AIL lock.
370 xfs_trans_delete_ail(mp, (xfs_log_item_t *)efip);
371 xfs_efi_item_free(efip);
373 spin_unlock(&ailp->xa_lock);
378 xfs_efd_item_free(xfs_efd_log_item_t *efdp)
380 int nexts = efdp->efd_format.efd_nextents;
382 if (nexts > XFS_EFD_MAX_FAST_EXTENTS) {
385 kmem_zone_free(xfs_efd_zone, efdp);
390 * This returns the number of iovecs needed to log the given efd item.
391 * We only need 1 iovec for an efd item. It just logs the efd_log_format
396 xfs_efd_item_size(xfs_efd_log_item_t *efdp)
402 * This is called to fill in the vector of log iovecs for the
403 * given efd log item. We use only 1 iovec, and we point that
404 * at the efd_log_format structure embedded in the efd item.
405 * It is at this point that we assert that all of the extent
406 * slots in the efd item have been filled.
409 xfs_efd_item_format(xfs_efd_log_item_t *efdp,
410 xfs_log_iovec_t *log_vector)
414 ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
416 efdp->efd_format.efd_type = XFS_LI_EFD;
418 size = sizeof(xfs_efd_log_format_t);
419 size += (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
420 efdp->efd_format.efd_size = 1;
422 log_vector->i_addr = (xfs_caddr_t)&(efdp->efd_format);
423 log_vector->i_len = size;
424 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_EFD_FORMAT);
425 ASSERT(size >= sizeof(xfs_efd_log_format_t));
430 * Pinning has no meaning for an efd item, so just return.
434 xfs_efd_item_pin(xfs_efd_log_item_t *efdp)
441 * Since pinning has no meaning for an efd item, unpinning does
446 xfs_efd_item_unpin(xfs_efd_log_item_t *efdp, int stale)
453 xfs_efd_item_unpin_remove(xfs_efd_log_item_t *efdp, xfs_trans_t *tp)
459 * Efd items have no locking, so just return success.
463 xfs_efd_item_trylock(xfs_efd_log_item_t *efdp)
465 return XFS_ITEM_LOCKED;
469 * Efd items have no locking or pushing, so return failure
470 * so that the caller doesn't bother with us.
474 xfs_efd_item_unlock(xfs_efd_log_item_t *efdp)
476 if (efdp->efd_item.li_flags & XFS_LI_ABORTED)
477 xfs_efd_item_free(efdp);
482 * When the efd item is committed to disk, all we need to do
483 * is delete our reference to our partner efi item and then
484 * free ourselves. Since we're freeing ourselves we must
485 * return -1 to keep the transaction code from further referencing
490 xfs_efd_item_committed(xfs_efd_log_item_t *efdp, xfs_lsn_t lsn)
493 * If we got a log I/O error, it's always the case that the LR with the
494 * EFI got unpinned and freed before the EFD got aborted.
496 if ((efdp->efd_item.li_flags & XFS_LI_ABORTED) == 0)
497 xfs_efi_release(efdp->efd_efip, efdp->efd_format.efd_nextents);
499 xfs_efd_item_free(efdp);
500 return (xfs_lsn_t)-1;
504 * There isn't much you can do to push on an efd item. It is simply
505 * stuck waiting for the log to be flushed to disk.
509 xfs_efd_item_push(xfs_efd_log_item_t *efdp)
515 * The EFD dependency tracking op doesn't do squat. It can't because
516 * it doesn't know where the free extent is coming from. The dependency
517 * tracking has to be handled by the "enclosing" metadata object. For
518 * example, for inodes, the inode is locked throughout the extent freeing
519 * so the dependency should be recorded there.
523 xfs_efd_item_committing(xfs_efd_log_item_t *efip, xfs_lsn_t lsn)
529 * This is the ops vector shared by all efd log items.
531 static struct xfs_item_ops xfs_efd_item_ops = {
532 .iop_size = (uint(*)(xfs_log_item_t*))xfs_efd_item_size,
533 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
535 .iop_pin = (void(*)(xfs_log_item_t*))xfs_efd_item_pin,
536 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_efd_item_unpin,
537 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
538 xfs_efd_item_unpin_remove,
539 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_efd_item_trylock,
540 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_efd_item_unlock,
541 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
542 xfs_efd_item_committed,
543 .iop_push = (void(*)(xfs_log_item_t*))xfs_efd_item_push,
545 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
546 xfs_efd_item_committing
551 * Allocate and initialize an efd item with the given number of extents.
554 xfs_efd_init(xfs_mount_t *mp,
555 xfs_efi_log_item_t *efip,
559 xfs_efd_log_item_t *efdp;
562 ASSERT(nextents > 0);
563 if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
564 size = (uint)(sizeof(xfs_efd_log_item_t) +
565 ((nextents - 1) * sizeof(xfs_extent_t)));
566 efdp = (xfs_efd_log_item_t*)kmem_zalloc(size, KM_SLEEP);
568 efdp = (xfs_efd_log_item_t*)kmem_zone_zalloc(xfs_efd_zone,
572 efdp->efd_item.li_type = XFS_LI_EFD;
573 efdp->efd_item.li_ops = &xfs_efd_item_ops;
574 efdp->efd_item.li_mountp = mp;
575 efdp->efd_item.li_ailp = mp->m_ail;
576 efdp->efd_efip = efip;
577 efdp->efd_format.efd_nextents = nextents;
578 efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;