summaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
authorDavid Chinner <david@fromorbit.com>2008-10-30 17:06:08 +1100
committerLachlan McIlroy <lachlan@sgi.com>2008-10-30 17:06:08 +1100
commitfe4fa4b8e463fa5848ef9e86ed75d27501d0da1e (patch)
tree8b3ec6b29f7bb2c604aebb7d9e99b2670ec5e87e /fs
parent34519daae6778d129d56688f75ade27f6011bac9 (diff)
downloadlinux-fe4fa4b8e463fa5848ef9e86ed75d27501d0da1e.tar.bz2
[XFS] move sync code to its own file
The sync code in XFS is spread around several files. While it used to make sense to have such a distribution, the code is about to be cleaned up and so centralising it in one spot as the first step makes sense. SGI-PV: 988139 SGI-Modid: xfs-linux-melb:xfs-kern:32282a Signed-off-by: David Chinner <david@fromorbit.com> Signed-off-by: Lachlan McIlroy <lachlan@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org>
Diffstat (limited to 'fs')
-rw-r--r--fs/xfs/Makefile1
-rw-r--r--fs/xfs/linux-2.6/xfs_super.c1
-rw-r--r--fs/xfs/linux-2.6/xfs_sync.c605
-rw-r--r--fs/xfs/linux-2.6/xfs_sync.h7
-rw-r--r--fs/xfs/xfs_vfsops.c560
-rw-r--r--fs/xfs/xfs_vfsops.h1
6 files changed, 615 insertions, 560 deletions
diff --git a/fs/xfs/Makefile b/fs/xfs/Makefile
index 75b2be72c39f..51b87de97f87 100644
--- a/fs/xfs/Makefile
+++ b/fs/xfs/Makefile
@@ -107,6 +107,7 @@ xfs-y += $(addprefix $(XFS_LINUX)/, \
xfs_iops.o \
xfs_lrw.o \
xfs_super.o \
+ xfs_sync.o \
xfs_vnode.o \
xfs_xattr.o)
diff --git a/fs/xfs/linux-2.6/xfs_super.c b/fs/xfs/linux-2.6/xfs_super.c
index 0c71c2158535..727d0e47e80f 100644
--- a/fs/xfs/linux-2.6/xfs_super.c
+++ b/fs/xfs/linux-2.6/xfs_super.c
@@ -59,6 +59,7 @@
#include "xfs_extfree_item.h"
#include "xfs_mru_cache.h"
#include "xfs_inode_item.h"
+#include "xfs_sync.h"
#include <linux/namei.h>
#include <linux/init.h>
diff --git a/fs/xfs/linux-2.6/xfs_sync.c b/fs/xfs/linux-2.6/xfs_sync.c
new file mode 100644
index 000000000000..c765eb2a8dca
--- /dev/null
+++ b/fs/xfs/linux-2.6/xfs_sync.c
@@ -0,0 +1,605 @@
+/*
+ * Copyright (c) 2000-2005 Silicon Graphics, Inc.
+ * All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_types.h"
+#include "xfs_bit.h"
+#include "xfs_log.h"
+#include "xfs_inum.h"
+#include "xfs_trans.h"
+#include "xfs_sb.h"
+#include "xfs_ag.h"
+#include "xfs_dir2.h"
+#include "xfs_dmapi.h"
+#include "xfs_mount.h"
+#include "xfs_bmap_btree.h"
+#include "xfs_alloc_btree.h"
+#include "xfs_ialloc_btree.h"
+#include "xfs_btree.h"
+#include "xfs_dir2_sf.h"
+#include "xfs_attr_sf.h"
+#include "xfs_inode.h"
+#include "xfs_dinode.h"
+#include "xfs_error.h"
+#include "xfs_mru_cache.h"
+#include "xfs_filestream.h"
+#include "xfs_vnodeops.h"
+#include "xfs_utils.h"
+#include "xfs_buf_item.h"
+#include "xfs_inode_item.h"
+#include "xfs_rw.h"
+
+/*
+ * xfs_sync flushes any pending I/O to file system vfsp.
+ *
+ * This routine is called by vfs_sync() to make sure that things make it
+ * out to disk eventually, on sync() system calls to flush out everything,
+ * and when the file system is unmounted. For the vfs_sync() case, all
+ * we really need to do is sync out the log to make all of our meta-data
+ * updates permanent (except for timestamps). For calls from pflushd(),
+ * dirty pages are kept moving by calling pdflush() on the inodes
+ * containing them. We also flush the inodes that we can lock without
+ * sleeping and the superblock if we can lock it without sleeping from
+ * vfs_sync() so that items at the tail of the log are always moving out.
+ *
+ * Flags:
+ * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
+ * to sleep if we can help it. All we really need
+ * to do is ensure that the log is synced at least
+ * periodically. We also push the inodes and
+ * superblock if we can lock them without sleeping
+ * and they are not pinned.
+ * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
+ * set, then we really want to lock each inode and flush
+ * it.
+ * SYNC_WAIT - All the flushes that take place in this call should
+ * be synchronous.
+ * SYNC_DELWRI - This tells us to push dirty pages associated with
+ * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
+ * determine if they should be flushed sync, async, or
+ * delwri.
+ * SYNC_CLOSE - This flag is passed when the system is being
+ * unmounted. We should sync and invalidate everything.
+ * SYNC_FSDATA - This indicates that the caller would like to make
+ * sure the superblock is safe on disk. We can ensure
+ * this by simply making sure the log gets flushed
+ * if SYNC_BDFLUSH is set, and by actually writing it
+ * out otherwise.
+ * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
+ * before we return (including direct I/O). Forms the drain
+ * side of the write barrier needed to safely quiesce the
+ * filesystem.
+ *
+ */
+int
+xfs_sync(
+ xfs_mount_t *mp,
+ int flags)
+{
+ int error;
+
+ /*
+ * Get the Quota Manager to flush the dquots.
+ *
+ * If XFS quota support is not enabled or this filesystem
+ * instance does not use quotas XFS_QM_DQSYNC will always
+ * return zero.
+ */
+ error = XFS_QM_DQSYNC(mp, flags);
+ if (error) {
+ /*
+ * If we got an IO error, we will be shutting down.
+ * So, there's nothing more for us to do here.
+ */
+ ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp));
+ if (XFS_FORCED_SHUTDOWN(mp))
+ return XFS_ERROR(error);
+ }
+
+ if (flags & SYNC_IOWAIT)
+ xfs_filestream_flush(mp);
+
+ return xfs_syncsub(mp, flags, NULL);
+}
+
+/*
+ * xfs sync routine for internal use
+ *
+ * This routine supports all of the flags defined for the generic vfs_sync
+ * interface as explained above under xfs_sync.
+ *
+ */
+int
+xfs_sync_inodes(
+ xfs_mount_t *mp,
+ int flags,
+ int *bypassed)
+{
+ xfs_inode_t *ip = NULL;
+ struct inode *vp = NULL;
+ int error;
+ int last_error;
+ uint64_t fflag;
+ uint lock_flags;
+ uint base_lock_flags;
+ boolean_t mount_locked;
+ boolean_t vnode_refed;
+ int preempt;
+ xfs_iptr_t *ipointer;
+#ifdef DEBUG
+ boolean_t ipointer_in = B_FALSE;
+
+#define IPOINTER_SET ipointer_in = B_TRUE
+#define IPOINTER_CLR ipointer_in = B_FALSE
+#else
+#define IPOINTER_SET
+#define IPOINTER_CLR
+#endif
+
+
+/* Insert a marker record into the inode list after inode ip. The list
+ * must be locked when this is called. After the call the list will no
+ * longer be locked.
+ */
+#define IPOINTER_INSERT(ip, mp) { \
+ ASSERT(ipointer_in == B_FALSE); \
+ ipointer->ip_mnext = ip->i_mnext; \
+ ipointer->ip_mprev = ip; \
+ ip->i_mnext = (xfs_inode_t *)ipointer; \
+ ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
+ preempt = 0; \
+ XFS_MOUNT_IUNLOCK(mp); \
+ mount_locked = B_FALSE; \
+ IPOINTER_SET; \
+ }
+
+/* Remove the marker from the inode list. If the marker was the only item
+ * in the list then there are no remaining inodes and we should zero out
+ * the whole list. If we are the current head of the list then move the head
+ * past us.
+ */
+#define IPOINTER_REMOVE(ip, mp) { \
+ ASSERT(ipointer_in == B_TRUE); \
+ if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
+ ip = ipointer->ip_mnext; \
+ ip->i_mprev = ipointer->ip_mprev; \
+ ipointer->ip_mprev->i_mnext = ip; \
+ if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
+ mp->m_inodes = ip; \
+ } \
+ } else { \
+ ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
+ mp->m_inodes = NULL; \
+ ip = NULL; \
+ } \
+ IPOINTER_CLR; \
+ }
+
+#define XFS_PREEMPT_MASK 0x7f
+
+ ASSERT(!(flags & SYNC_BDFLUSH));
+
+ if (bypassed)
+ *bypassed = 0;
+ if (mp->m_flags & XFS_MOUNT_RDONLY)
+ return 0;
+ error = 0;
+ last_error = 0;
+ preempt = 0;
+
+ /* Allocate a reference marker */
+ ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
+
+ fflag = XFS_B_ASYNC; /* default is don't wait */
+ if (flags & SYNC_DELWRI)
+ fflag = XFS_B_DELWRI;
+ if (flags & SYNC_WAIT)
+ fflag = 0; /* synchronous overrides all */
+
+ base_lock_flags = XFS_ILOCK_SHARED;
+ if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
+ /*
+ * We need the I/O lock if we're going to call any of
+ * the flush/inval routines.
+ */
+ base_lock_flags |= XFS_IOLOCK_SHARED;
+ }
+
+ XFS_MOUNT_ILOCK(mp);
+
+ ip = mp->m_inodes;
+
+ mount_locked = B_TRUE;
+ vnode_refed = B_FALSE;
+
+ IPOINTER_CLR;
+
+ do {
+ ASSERT(ipointer_in == B_FALSE);
+ ASSERT(vnode_refed == B_FALSE);
+
+ lock_flags = base_lock_flags;
+
+ /*
+ * There were no inodes in the list, just break out
+ * of the loop.
+ */
+ if (ip == NULL) {
+ break;
+ }
+
+ /*
+ * We found another sync thread marker - skip it
+ */
+ if (ip->i_mount == NULL) {
+ ip = ip->i_mnext;
+ continue;
+ }
+
+ vp = VFS_I(ip);
+
+ /*
+ * If the vnode is gone then this is being torn down,
+ * call reclaim if it is flushed, else let regular flush
+ * code deal with it later in the loop.
+ */
+
+ if (vp == NULL) {
+ /* Skip ones already in reclaim */
+ if (ip->i_flags & XFS_IRECLAIM) {
+ ip = ip->i_mnext;
+ continue;
+ }
+ if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
+ ip = ip->i_mnext;
+ } else if ((xfs_ipincount(ip) == 0) &&
+ xfs_iflock_nowait(ip)) {
+ IPOINTER_INSERT(ip, mp);
+
+ xfs_finish_reclaim(ip, 1,
+ XFS_IFLUSH_DELWRI_ELSE_ASYNC);
+
+ XFS_MOUNT_ILOCK(mp);
+ mount_locked = B_TRUE;
+ IPOINTER_REMOVE(ip, mp);
+ } else {
+ xfs_iunlock(ip, XFS_ILOCK_EXCL);
+ ip = ip->i_mnext;
+ }
+ continue;
+ }
+
+ if (VN_BAD(vp)) {
+ ip = ip->i_mnext;
+ continue;
+ }
+
+ if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
+ XFS_MOUNT_IUNLOCK(mp);
+ kmem_free(ipointer);
+ return 0;
+ }
+
+ /*
+ * Try to lock without sleeping. We're out of order with
+ * the inode list lock here, so if we fail we need to drop
+ * the mount lock and try again. If we're called from
+ * bdflush() here, then don't bother.
+ *
+ * The inode lock here actually coordinates with the
+ * almost spurious inode lock in xfs_ireclaim() to prevent
+ * the vnode we handle here without a reference from
+ * being freed while we reference it. If we lock the inode
+ * while it's on the mount list here, then the spurious inode
+ * lock in xfs_ireclaim() after the inode is pulled from
+ * the mount list will sleep until we release it here.
+ * This keeps the vnode from being freed while we reference
+ * it.
+ */
+ if (xfs_ilock_nowait(ip, lock_flags) == 0) {
+ if (vp == NULL) {
+ ip = ip->i_mnext;
+ continue;
+ }
+
+ vp = vn_grab(vp);
+ if (vp == NULL) {
+ ip = ip->i_mnext;
+ continue;
+ }
+
+ IPOINTER_INSERT(ip, mp);
+ xfs_ilock(ip, lock_flags);
+
+ ASSERT(vp == VFS_I(ip));
+ ASSERT(ip->i_mount == mp);
+
+ vnode_refed = B_TRUE;
+ }
+
+ /* From here on in the loop we may have a marker record
+ * in the inode list.
+ */
+
+ /*
+ * If we have to flush data or wait for I/O completion
+ * we need to drop the ilock that we currently hold.
+ * If we need to drop the lock, insert a marker if we
+ * have not already done so.
+ */
+ if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
+ ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
+ if (mount_locked) {
+ IPOINTER_INSERT(ip, mp);
+ }
+ xfs_iunlock(ip, XFS_ILOCK_SHARED);
+
+ if (flags & SYNC_CLOSE) {
+ /* Shutdown case. Flush and invalidate. */
+ if (XFS_FORCED_SHUTDOWN(mp))
+ xfs_tosspages(ip, 0, -1,
+ FI_REMAPF);
+ else
+ error = xfs_flushinval_pages(ip,
+ 0, -1, FI_REMAPF);
+ } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
+ error = xfs_flush_pages(ip, 0,
+ -1, fflag, FI_NONE);
+ }
+
+ /*
+ * When freezing, we need to wait ensure all I/O (including direct
+ * I/O) is complete to ensure no further data modification can take
+ * place after this point
+ */
+ if (flags & SYNC_IOWAIT)
+ vn_iowait(ip);
+
+ xfs_ilock(ip, XFS_ILOCK_SHARED);
+ }
+
+ if ((flags & SYNC_ATTR) &&
+ (ip->i_update_core ||
+ (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) {
+ if (mount_locked)
+ IPOINTER_INSERT(ip, mp);
+
+ if (flags & SYNC_WAIT) {
+ xfs_iflock(ip);
+ error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
+
+ /*
+ * If we can't acquire the flush lock, then the inode
+ * is already being flushed so don't bother waiting.
+ *
+ * If we can lock it then do a delwri flush so we can
+ * combine multiple inode flushes in each disk write.
+ */
+ } else if (xfs_iflock_nowait(ip)) {
+ error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
+ } else if (bypassed) {
+ (*bypassed)++;
+ }
+ }
+
+ if (lock_flags != 0) {
+ xfs_iunlock(ip, lock_flags);
+ }
+
+ if (vnode_refed) {
+ /*
+ * If we had to take a reference on the vnode
+ * above, then wait until after we've unlocked
+ * the inode to release the reference. This is
+ * because we can be already holding the inode
+ * lock when IRELE() calls xfs_inactive().
+ *
+ * Make sure to drop the mount lock before calling
+ * IRELE() so that we don't trip over ourselves if
+ * we have to go for the mount lock again in the
+ * inactive code.
+ */
+ if (mount_locked) {
+ IPOINTER_INSERT(ip, mp);
+ }
+
+ IRELE(ip);
+
+ vnode_refed = B_FALSE;
+ }
+
+ if (error) {
+ last_error = error;
+ }
+
+ /*
+ * bail out if the filesystem is corrupted.
+ */
+ if (error == EFSCORRUPTED) {
+ if (!mount_locked) {
+ XFS_MOUNT_ILOCK(mp);
+ IPOINTER_REMOVE(ip, mp);
+ }
+ XFS_MOUNT_IUNLOCK(mp);
+ ASSERT(ipointer_in == B_FALSE);
+ kmem_free(ipointer);
+ return XFS_ERROR(error);
+ }
+
+ /* Let other threads have a chance at the mount lock
+ * if we have looped many times without dropping the
+ * lock.
+ */
+ if ((++preempt & XFS_PREEMPT_MASK) == 0) {
+ if (mount_locked) {
+ IPOINTER_INSERT(ip, mp);
+ }
+ }
+
+ if (mount_locked == B_FALSE) {
+ XFS_MOUNT_ILOCK(mp);
+ mount_locked = B_TRUE;
+ IPOINTER_REMOVE(ip, mp);
+ continue;
+ }
+
+ ASSERT(ipointer_in == B_FALSE);
+ ip = ip->i_mnext;
+
+ } while (ip != mp->m_inodes);
+
+ XFS_MOUNT_IUNLOCK(mp);
+
+ ASSERT(ipointer_in == B_FALSE);
+
+ kmem_free(ipointer);
+ return XFS_ERROR(last_error);
+}
+
+/*
+ * xfs sync routine for internal use
+ *
+ * This routine supports all of the flags defined for the generic vfs_sync
+ * interface as explained above under xfs_sync.
+ *
+ */
+int
+xfs_syncsub(
+ xfs_mount_t *mp,
+ int flags,
+ int *bypassed)
+{
+ int error = 0;
+ int last_error = 0;
+ uint log_flags = XFS_LOG_FORCE;
+ xfs_buf_t *bp;
+ xfs_buf_log_item_t *bip;
+
+ /*
+ * Sync out the log. This ensures that the log is periodically
+ * flushed even if there is not enough activity to fill it up.
+ */
+ if (flags & SYNC_WAIT)
+ log_flags |= XFS_LOG_SYNC;
+
+ xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
+
+ if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
+ if (flags & SYNC_BDFLUSH)
+ xfs_finish_reclaim_all(mp, 1);
+ else
+ error = xfs_sync_inodes(mp, flags, bypassed);
+ }
+
+ /*
+ * Flushing out dirty data above probably generated more
+ * log activity, so if this isn't vfs_sync() then flush
+ * the log again.
+ */
+ if (flags & SYNC_DELWRI) {
+ xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
+ }
+
+ if (flags & SYNC_FSDATA) {
+ /*
+ * If this is vfs_sync() then only sync the superblock
+ * if we can lock it without sleeping and it is not pinned.
+ */
+ if (flags & SYNC_BDFLUSH) {
+ bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
+ if (bp != NULL) {
+ bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
+ if ((bip != NULL) &&
+ xfs_buf_item_dirty(bip)) {
+ if (!(XFS_BUF_ISPINNED(bp))) {
+ XFS_BUF_ASYNC(bp);
+ error = xfs_bwrite(mp, bp);
+ } else {
+ xfs_buf_relse(bp);
+ }
+ } else {
+ xfs_buf_relse(bp);
+ }
+ }
+ } else {
+ bp = xfs_getsb(mp, 0);
+ /*
+ * If the buffer is pinned then push on the log so
+ * we won't get stuck waiting in the write for
+ * someone, maybe ourselves, to flush the log.
+ * Even though we just pushed the log above, we
+ * did not have the superblock buffer locked at
+ * that point so it can become pinned in between
+ * there and here.
+ */
+ if (XFS_BUF_ISPINNED(bp))
+ xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
+ if (flags & SYNC_WAIT)
+ XFS_BUF_UNASYNC(bp);
+ else
+ XFS_BUF_ASYNC(bp);
+ error = xfs_bwrite(mp, bp);
+ }
+ if (error) {
+ last_error = error;
+ }
+ }
+
+ /*
+ * Now check to see if the log needs a "dummy" transaction.
+ */
+ if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
+ xfs_trans_t *tp;
+ xfs_inode_t *ip;
+
+ /*
+ * Put a dummy transaction in the log to tell
+ * recovery that all others are OK.
+ */
+ tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
+ if ((error = xfs_trans_reserve(tp, 0,
+ XFS_ICHANGE_LOG_RES(mp),
+ 0, 0, 0))) {
+ xfs_trans_cancel(tp, 0);
+ return error;
+ }
+
+ ip = mp->m_rootip;
+ xfs_ilock(ip, XFS_ILOCK_EXCL);
+
+ xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
+ xfs_trans_ihold(tp, ip);
+ xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
+ error = xfs_trans_commit(tp, 0);
+ xfs_iunlock(ip, XFS_ILOCK_EXCL);
+ xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
+ }
+
+ /*
+ * When shutting down, we need to insure that the AIL is pushed
+ * to disk or the filesystem can appear corrupt from the PROM.
+ */
+ if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
+ XFS_bflush(mp->m_ddev_targp);
+ if (mp->m_rtdev_targp) {
+ XFS_bflush(mp->m_rtdev_targp);
+ }
+ }
+
+ return XFS_ERROR(last_error);
+}
diff --git a/fs/xfs/linux-2.6/xfs_sync.h b/fs/xfs/linux-2.6/xfs_sync.h
new file mode 100644
index 000000000000..f4c3b1ea64c0
--- /dev/null
+++ b/fs/xfs/linux-2.6/xfs_sync.h
@@ -0,0 +1,7 @@
+#ifndef XFS_SYNC_H
+#define XFS_SYNC_H 1
+
+int xfs_sync(struct xfs_mount *mp, int flags);
+int xfs_syncsub(struct xfs_mount *mp, int flags, int *bypassed);
+
+#endif
diff --git a/fs/xfs/xfs_vfsops.c b/fs/xfs/xfs_vfsops.c
index 439dd3939dda..01e274b902c0 100644
--- a/fs/xfs/xfs_vfsops.c
+++ b/fs/xfs/xfs_vfsops.c
@@ -56,6 +56,7 @@
#include "xfs_vnodeops.h"
#include "xfs_vfsops.h"
#include "xfs_utils.h"
+#include "xfs_sync.h"
STATIC void
@@ -196,562 +197,3 @@ fscorrupt_out2:
return XFS_ERROR(EFSCORRUPTED);
}
-/*
- * xfs_sync flushes any pending I/O to file system vfsp.
- *
- * This routine is called by vfs_sync() to make sure that things make it
- * out to disk eventually, on sync() system calls to flush out everything,
- * and when the file system is unmounted. For the vfs_sync() case, all
- * we really need to do is sync out the log to make all of our meta-data
- * updates permanent (except for timestamps). For calls from pflushd(),
- * dirty pages are kept moving by calling pdflush() on the inodes
- * containing them. We also flush the inodes that we can lock without
- * sleeping and the superblock if we can lock it without sleeping from
- * vfs_sync() so that items at the tail of the log are always moving out.
- *
- * Flags:
- * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
- * to sleep if we can help it. All we really need
- * to do is ensure that the log is synced at least
- * periodically. We also push the inodes and
- * superblock if we can lock them without sleeping
- * and they are not pinned.
- * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
- * set, then we really want to lock each inode and flush
- * it.
- * SYNC_WAIT - All the flushes that take place in this call should
- * be synchronous.
- * SYNC_DELWRI - This tells us to push dirty pages associated with
- * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
- * determine if they should be flushed sync, async, or
- * delwri.
- * SYNC_CLOSE - This flag is passed when the system is being
- * unmounted. We should sync and invalidate everything.
- * SYNC_FSDATA - This indicates that the caller would like to make
- * sure the superblock is safe on disk. We can ensure
- * this by simply making sure the log gets flushed
- * if SYNC_BDFLUSH is set, and by actually writing it
- * out otherwise.
- * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
- * before we return (including direct I/O). Forms the drain
- * side of the write barrier needed to safely quiesce the
- * filesystem.
- *
- */
-int
-xfs_sync(
- xfs_mount_t *mp,
- int flags)
-{
- int error;
-
- /*
- * Get the Quota Manager to flush the dquots.
- *
- * If XFS quota support is not enabled or this filesystem
- * instance does not use quotas XFS_QM_DQSYNC will always
- * return zero.
- */
- error = XFS_QM_DQSYNC(mp, flags);
- if (error) {
- /*
- * If we got an IO error, we will be shutting down.
- * So, there's nothing more for us to do here.
- */
- ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp));
- if (XFS_FORCED_SHUTDOWN(mp))
- return XFS_ERROR(error);
- }
-
- if (flags & SYNC_IOWAIT)
- xfs_filestream_flush(mp);
-
- return xfs_syncsub(mp, flags, NULL);
-}
-
-/*
- * xfs sync routine for internal use
- *
- * This routine supports all of the flags defined for the generic vfs_sync
- * interface as explained above under xfs_sync.
- *
- */
-int
-xfs_sync_inodes(
- xfs_mount_t *mp,
- int flags,
- int *bypassed)
-{
- xfs_inode_t *ip = NULL;
- struct inode *vp = NULL;
- int error;
- int last_error;
- uint64_t fflag;
- uint lock_flags;
- uint base_lock_flags;
- boolean_t mount_locked;
- boolean_t vnode_refed;
- int preempt;
- xfs_iptr_t *ipointer;
-#ifdef DEBUG
- boolean_t ipointer_in = B_FALSE;
-
-#define IPOINTER_SET ipointer_in = B_TRUE
-#define IPOINTER_CLR ipointer_in = B_FALSE
-#else
-#define IPOINTER_SET
-#define IPOINTER_CLR
-#endif
-
-
-/* Insert a marker record into the inode list after inode ip. The list
- * must be locked when this is called. After the call the list will no
- * longer be locked.
- */
-#define IPOINTER_INSERT(ip, mp) { \
- ASSERT(ipointer_in == B_FALSE); \
- ipointer->ip_mnext = ip->i_mnext; \
- ipointer->ip_mprev = ip; \
- ip->i_mnext = (xfs_inode_t *)ipointer; \
- ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
- preempt = 0; \
- XFS_MOUNT_IUNLOCK(mp); \
- mount_locked = B_FALSE; \
- IPOINTER_SET; \
- }
-
-/* Remove the marker from the inode list. If the marker was the only item
- * in the list then there are no remaining inodes and we should zero out
- * the whole list. If we are the current head of the list then move the head
- * past us.
- */
-#define IPOINTER_REMOVE(ip, mp) { \
- ASSERT(ipointer_in == B_TRUE); \
- if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
- ip = ipointer->ip_mnext; \
- ip->i_mprev = ipointer->ip_mprev; \
- ipointer->ip_mprev->i_mnext = ip; \
- if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
- mp->m_inodes = ip; \
- } \
- } else { \
- ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
- mp->m_inodes = NULL; \
- ip = NULL; \
- } \
- IPOINTER_CLR; \
- }
-
-#define XFS_PREEMPT_MASK 0x7f
-
- ASSERT(!(flags & SYNC_BDFLUSH));
-
- if (bypassed)
- *bypassed = 0;
- if (mp->m_flags & XFS_MOUNT_RDONLY)
- return 0;
- error = 0;
- last_error = 0;
- preempt = 0;
-
- /* Allocate a reference marker */
- ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
-
- fflag = XFS_B_ASYNC; /* default is don't wait */
- if (flags & SYNC_DELWRI)
- fflag = XFS_B_DELWRI;
- if (flags & SYNC_WAIT)
- fflag = 0; /* synchronous overrides all */
-
- base_lock_flags = XFS_ILOCK_SHARED;
- if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
- /*
- * We need the I/O lock if we're going to call any of
- * the flush/inval routines.
- */
- base_lock_flags |= XFS_IOLOCK_SHARED;
- }
-
- XFS_MOUNT_ILOCK(mp);
-
- ip = mp->m_inodes;
-
- mount_locked = B_TRUE;
- vnode_refed = B_FALSE;
-
- IPOINTER_CLR;
-
- do {
- ASSERT(ipointer_in == B_FALSE);
- ASSERT(vnode_refed == B_FALSE);
-
- lock_flags = base_lock_flags;
-
- /*
- * There were no inodes in the list, just break out
- * of the loop.
- */
- if (ip == NULL) {
- break;
- }
-
- /*
- * We found another sync thread marker - skip it
- */
- if (ip->i_mount == NULL) {
- ip = ip->i_mnext;
- continue;
- }
-
- vp = VFS_I(ip);
-
- /*
- * If the vnode is gone then this is being torn down,
- * call reclaim if it is flushed, else let regular flush
- * code deal with it later in the loop.
- */
-
- if (vp == NULL) {
- /* Skip ones already in reclaim */
- if (ip->i_flags & XFS_IRECLAIM) {
- ip = ip->i_mnext;
- continue;
- }
- if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
- ip = ip->i_mnext;
- } else if ((xfs_ipincount(ip) == 0) &&
- xfs_iflock_nowait(ip)) {
- IPOINTER_INSERT(ip, mp);
-
- xfs_finish_reclaim(ip, 1,
- XFS_IFLUSH_DELWRI_ELSE_ASYNC);
-
- XFS_MOUNT_ILOCK(mp);
- mount_locked = B_TRUE;
- IPOINTER_REMOVE(ip, mp);
- } else {
- xfs_iunlock(ip, XFS_ILOCK_EXCL);
- ip = ip->i_mnext;
- }
- continue;
- }
-
- if (VN_BAD(vp)) {
- ip = ip->i_mnext;
- continue;
- }
-
- if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
- XFS_MOUNT_IUNLOCK(mp);
- kmem_free(ipointer);
- return 0;
- }
-
- /*
- * Try to lock without sleeping. We're out of order with
- * the inode list lock here, so if we fail we need to drop
- * the mount lock and try again. If we're called from
- * bdflush() here, then don't bother.
- *
- * The inode lock here actually coordinates with the
- * almost spurious inode lock in xfs_ireclaim() to prevent
- * the vnode we handle here without a reference from
- * being freed while we reference it. If we lock the inode
- * while it's on the mount list here, then the spurious inode
- * lock in xfs_ireclaim() after the inode is pulled from
- * the mount list will sleep until we release it here.
- * This keeps the vnode from being freed while we reference
- * it.
- */
- if (xfs_ilock_nowait(ip, lock_flags) == 0) {
- if (vp == NULL) {
- ip = ip->i_mnext;
- continue;
- }
-
- vp = vn_grab(vp);
- if (vp == NULL) {
- ip = ip->i_mnext;
- continue;
- }
-
- IPOINTER_INSERT(ip, mp);
- xfs_ilock(ip, lock_flags);
-
- ASSERT(vp == VFS_I(ip));
- ASSERT(ip->i_mount == mp);
-
- vnode_refed = B_TRUE;
- }
-
- /* From here on in the loop we may have a marker record
- * in the inode list.
- */
-
- /*
- * If we have to flush data or wait for I/O completion
- * we need to drop the ilock that we currently hold.
- * If we need to drop the lock, insert a marker if we
- * have not already done so.
- */
- if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
- ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
- if (mount_locked) {
- IPOINTER_INSERT(ip, mp);
- }
- xfs_iunlock(ip, XFS_ILOCK_SHARED);
-
- if (flags & SYNC_CLOSE) {
- /* Shutdown case. Flush and invalidate. */
- if (XFS_FORCED_SHUTDOWN(mp))
- xfs_tosspages(ip, 0, -1,
- FI_REMAPF);
- else
- error = xfs_flushinval_pages(ip,
- 0, -1, FI_REMAPF);
- } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
- error = xfs_flush_pages(ip, 0,
- -1, fflag, FI_NONE);
- }
-
- /*
- * When freezing, we need to wait ensure all I/O (including direct
- * I/O) is complete to ensure no further data modification can take
- * place after this point
- */
- if (flags & SYNC_IOWAIT)
- vn_iowait(ip);
-
- xfs_ilock(ip, XFS_ILOCK_SHARED);
- }
-
- if ((flags & SYNC_ATTR) &&
- (ip->i_update_core ||
- (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) {
- if (mount_locked)
- IPOINTER_INSERT(ip, mp);
-
- if (flags & SYNC_WAIT) {
- xfs_iflock(ip);
- error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
-
- /*
- * If we can't acquire the flush lock, then the inode
- * is already being flushed so don't bother waiting.
- *
- * If we can lock it then do a delwri flush so we can
- * combine multiple inode flushes in each disk write.
- */
- } else if (xfs_iflock_nowait(ip)) {
- error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
- } else if (bypassed) {
- (*bypassed)++;
- }
- }
-
- if (lock_flags != 0) {
- xfs_iunlock(ip, lock_flags);
- }
-
- if (vnode_refed) {
- /*
- * If we had to take a reference on the vnode
- * above, then wait until after we've unlocked
- * the inode to release the reference. This is
- * because we can be already holding the inode
- * lock when IRELE() calls xfs_inactive().
- *
- * Make sure to drop the mount lock before calling
- * IRELE() so that we don't trip over ourselves if
- * we have to go for the mount lock again in the
- * inactive code.
- */
- if (mount_locked) {
- IPOINTER_INSERT(ip, mp);
- }
-
- IRELE(ip);
-
- vnode_refed = B_FALSE;
- }
-
- if (error) {
- last_error = error;
- }
-
- /*
- * bail out if the filesystem is corrupted.
- */
- if (error == EFSCORRUPTED) {
- if (!mount_locked) {
- XFS_MOUNT_ILOCK(mp);
- IPOINTER_REMOVE(ip, mp);
- }
- XFS_MOUNT_IUNLOCK(mp);
- ASSERT(ipointer_in == B_FALSE);
- kmem_free(ipointer);
- return XFS_ERROR(error);
- }
-
- /* Let other threads have a chance at the mount lock
- * if we have looped many times without dropping the
- * lock.
- */
- if ((++preempt & XFS_PREEMPT_MASK) == 0) {
- if (mount_locked) {
- IPOINTER_INSERT(ip, mp);
- }
- }
-
- if (mount_locked == B_FALSE) {
- XFS_MOUNT_ILOCK(mp);
- mount_locked = B_TRUE;
- IPOINTER_REMOVE(ip, mp);
- continue;
- }
-
- ASSERT(ipointer_in == B_FALSE);
- ip = ip->i_mnext;
-
- } while (ip != mp->m_inodes);
-
- XFS_MOUNT_IUNLOCK(mp);
-
- ASSERT(ipointer_in == B_FALSE);
-
- kmem_free(ipointer);
- return XFS_ERROR(last_error);
-}
-
-/*
- * xfs sync routine for internal use
- *
- * This routine supports all of the flags defined for the generic vfs_sync
- * interface as explained above under xfs_sync.
- *
- */
-int
-xfs_syncsub(
- xfs_mount_t *mp,
- int flags,
- int *bypassed)
-{
- int error = 0;
- int last_error = 0;
- uint log_flags = XFS_LOG_FORCE;
- xfs_buf_t *bp;
- xfs_buf_log_item_t *bip;
-
- /*
- * Sync out the log. This ensures that the log is periodically
- * flushed even if there is not enough activity to fill it up.
- */
- if (flags & SYNC_WAIT)
- log_flags |= XFS_LOG_SYNC;
-
- xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
-
- if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
- if (flags & SYNC_BDFLUSH)
- xfs_finish_reclaim_all(mp, 1);
- else
- error = xfs_sync_inodes(mp, flags, bypassed);
- }
-
- /*
- * Flushing out dirty data above probably generated more
- * log activity, so if this isn't vfs_sync() then flush
- * the log again.
- */
- if (flags & SYNC_DELWRI) {
- xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
- }
-
- if (flags & SYNC_FSDATA) {
- /*
- * If this is vfs_sync() then only sync the superblock
- * if we can lock it without sleeping and it is not pinned.
- */
- if (flags & SYNC_BDFLUSH) {
- bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
- if (bp != NULL) {
- bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
- if ((bip != NULL) &&
- xfs_buf_item_dirty(bip)) {
- if (!(XFS_BUF_ISPINNED(bp))) {
- XFS_BUF_ASYNC(bp);
- error = xfs_bwrite(mp, bp);
- } else {
- xfs_buf_relse(bp);
- }
- } else {
- xfs_buf_relse(bp);
- }
- }
- } else {
- bp = xfs_getsb(mp, 0);
- /*
- * If the buffer is pinned then push on the log so
- * we won't get stuck waiting in the write for
- * someone, maybe ourselves, to flush the log.
- * Even though we just pushed the log above, we
- * did not have the superblock buffer locked at
- * that point so it can become pinned in between
- * there and here.
- */
- if (XFS_BUF_ISPINNED(bp))
- xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
- if (flags & SYNC_WAIT)
- XFS_BUF_UNASYNC(bp);
- else
- XFS_BUF_ASYNC(bp);
- error = xfs_bwrite(mp, bp);
- }
- if (error) {
- last_error = error;
- }
- }
-
- /*
- * Now check to see if the log needs a "dummy" transaction.
- */
- if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
- xfs_trans_t *tp;
- xfs_inode_t *ip;
-
- /*
- * Put a dummy transaction in the log to tell
- * recovery that all others are OK.
- */
- tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
- if ((error = xfs_trans_reserve(tp, 0,
- XFS_ICHANGE_LOG_RES(mp),
- 0, 0, 0))) {
- xfs_trans_cancel(tp, 0);
- return error;
- }
-
- ip = mp->m_rootip;
- xfs_ilock(ip, XFS_ILOCK_EXCL);
-
- xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
- xfs_trans_ihold(tp, ip);
- xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
- error = xfs_trans_commit(tp, 0);
- xfs_iunlock(ip, XFS_ILOCK_EXCL);
- xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
- }
-
- /*
- * When shutting down, we need to insure that the AIL is pushed
- * to disk or the filesystem can appear corrupt from the PROM.
- */
- if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
- XFS_bflush(mp->m_ddev_targp);
- if (mp->m_rtdev_targp) {
- XFS_bflush(mp->m_rtdev_targp);
- }
- }
-
- return XFS_ERROR(last_error);
-}
diff --git a/fs/xfs/xfs_vfsops.h b/fs/xfs/xfs_vfsops.h
index a74b05087da4..6701d0ed8adc 100644
--- a/fs/xfs/xfs_vfsops.h
+++ b/fs/xfs/xfs_vfsops.h
@@ -8,7 +8,6 @@ struct kstatfs;
struct xfs_mount;
struct xfs_mount_args;
-int xfs_sync(struct xfs_mount *mp, int flags);
void xfs_do_force_shutdown(struct xfs_mount *mp, int flags, char *fname,
int lnnum);
void xfs_attr_quiesce(struct xfs_mount *mp);