/*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2002 Apple Computer, Inc. All rights reserved.
*
* @APPLE_LICENSE_HEADER_START@
*
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License"). You may not use this file except in compliance with the
- * License. Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved.
*
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this
+ * file.
+ *
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
*
* @APPLE_LICENSE_HEADER_END@
*/
-/* @(#)hfs_btreeio.c
-*
-* (c) 1998, 2000 Apple Computer, Inc. All Rights Reserved
-*
-* hfs_btreeio.c -- I/O Routines for the HFS B-tree files.
-*
-* HISTORY
-* 15-Feb-2000 Don Brady Added ClearBTNodes.
-* 16-Jul-1998 Don Brady In ExtendBtreeFile force all b-tree nodes to be contiguous on disk.
-* 4-Jun-1998 Pat Dirks Changed to do all B*-Tree writes synchronously (FORCESYNCBTREEWRITES = 1)
-* 18-apr-1998 Don Brady Call brelse on bread failure.
-* 17-Apr-1998 Pat Dirks Fixed ReleaseBTreeBlock to not call brelse when bwrite or bdwrite is called.
-* 13-apr-1998 Don Brady Add ExtendBTreeFile routine (from BTreeWrapper.c).
-* 26-mar-1998 Don Brady SetBTreeBlockSize was incorrectly excluding 512 byte blockSize.
-* 18-feb-1998 Don Brady Initially created file.
-*
-*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/buf.h>
+#include <sys/kernel.h>
#include <sys/mount.h>
#include <sys/vnode.h>
#include "hfs.h"
+#include "hfs_cnode.h"
#include "hfs_dbg.h"
#include "hfs_endian.h"
#define FORCESYNCBTREEWRITES 0
-static OSStatus FlushAlternate( ExtendedVCB *vcb );
static int ClearBTNodes(struct vnode *vp, long blksize, off_t offset, off_t amount);
+__private_extern__
OSStatus SetBTreeBlockSize(FileReference vp, ByteCount blockSize, ItemCount minBlockCount)
{
BTreeControlBlockPtr bTreePtr;
DBG_ASSERT(vp != NULL);
- DBG_ASSERT(VTOFCB(vp) != NULL);
- DBG_ASSERT(VTOFCB(vp)->fcbBTCBPtr != NULL);
DBG_ASSERT(blockSize >= kMinNodeSize);
if (blockSize > MAXBSIZE )
return (fsBTBadNodeSize);
- DBG_TREE(("SetBlockSizeProc: blockSize=%ld for file %ld\n", blockSize, H_FILEID(VTOH(vp))));
-
- bTreePtr = (BTreeControlBlockPtr)(VTOH(vp)->fcbBTCBPtr);
+ bTreePtr = (BTreeControlBlockPtr)VTOF(vp)->fcbBTCBPtr;
bTreePtr->nodeSize = blockSize;
return (E_NONE);
}
+__private_extern__
OSStatus GetBTreeBlock(FileReference vp, UInt32 blockNum, GetBlockOptions options, BlockDescriptor *block)
{
OSStatus retval = E_NONE;
struct buf *bp = NULL;
- if (options & kGetEmptyBlock)
- bp = getblk (vp,
- IOBLKNOFORBLK(blockNum, VTOHFS(vp)->hfs_phys_block_size),
- IOBYTECCNTFORBLK(blockNum, block->blockSize, VTOHFS(vp)->hfs_phys_block_size),
- 0,
- 0,
- BLK_META);
- else
- retval = meta_bread(vp,
- IOBLKNOFORBLK(blockNum, VTOHFS(vp)->hfs_phys_block_size),
- IOBYTECCNTFORBLK(blockNum, block->blockSize, VTOHFS(vp)->hfs_phys_block_size),
- NOCRED,
- &bp);
+ if (options & kGetEmptyBlock)
+ bp = getblk(vp, blockNum, block->blockSize, 0, 0, BLK_META);
+ else
+ retval = meta_bread(vp, blockNum, block->blockSize, NOCRED, &bp);
DBG_ASSERT(bp != NULL);
DBG_ASSERT(bp->b_data != NULL);
if (retval == E_NONE) {
block->blockHeader = bp;
- block->buffer = bp->b_data + IOBYTEOFFSETFORBLK(bp->b_blkno, VTOHFS(vp)->hfs_phys_block_size);
+ block->buffer = bp->b_data;
block->blockReadFromDisk = (bp->b_flags & B_CACHE) == 0; /* not found in cache ==> came from disk */
+ // XXXdbg
+ block->isModified = 0;
+
#if BYTE_ORDER == LITTLE_ENDIAN
/* Endian swap B-Tree node (only if it's a valid block) */
if (!(options & kGetEmptyBlock)) {
(SWAP_BE16 (((BTHeaderRec *)((char *)block->buffer + 14))->nodeSize) != bp->b_bcount)) {
/* Don't swap the descriptors at all, we don't care (this block will be invalidated) */
- SWAP_BT_NODE (block, ISHFSPLUS(VTOVCB(vp)), H_FILEID(VTOH(vp)), 3);
+ SWAP_BT_NODE (block, ISHFSPLUS(VTOVCB(vp)), VTOC(vp)->c_fileid, 3);
/* The node needs swapping */
} else if (*((UInt16 *)((char *)block->buffer + (block->blockSize - sizeof (UInt16)))) == 0x0e00) {
- SWAP_BT_NODE (block, ISHFSPLUS(VTOVCB(vp)), H_FILEID(VTOH(vp)), 0);
+ SWAP_BT_NODE (block, ISHFSPLUS(VTOVCB(vp)), VTOC(vp)->c_fileid, 0);
#if 0
/* The node is not already in native byte order, hence corrupt */
} else if (*((UInt16 *)((char *)block->buffer + (block->blockSize - sizeof (UInt16)))) != 0x000e) {
}
+__private_extern__
+void ModifyBlockStart(FileReference vp, BlockDescPtr blockPtr)
+{
+ struct hfsmount *hfsmp = VTOHFS(vp);
+ struct buf *bp = NULL;
+
+ if (hfsmp->jnl == NULL) {
+ return;
+ }
+
+ bp = (struct buf *) blockPtr->blockHeader;
+ if (bp == NULL) {
+ panic("ModifyBlockStart: null bp for blockdescptr 0x%x?!?\n", blockPtr);
+ return;
+ }
+
+ journal_modify_block_start(hfsmp->jnl, bp);
+ blockPtr->isModified = 1;
+}
+
+
+__private_extern__
OSStatus ReleaseBTreeBlock(FileReference vp, BlockDescPtr blockPtr, ReleaseBlockOptions options)
{
+ struct hfsmount *hfsmp = VTOHFS(vp);
+ extern int bdwrite_internal(struct buf *, int);
OSStatus retval = E_NONE;
struct buf *bp = NULL;
bp = (struct buf *) blockPtr->blockHeader;
if (bp == NULL) {
- DBG_TREE(("ReleaseBlockProc: blockHeader is zero!\n"));
retval = -1;
goto exit;
}
if (options & kTrashBlock) {
- bp->b_flags |= B_INVAL;
- brelse(bp); /* note: B-tree code will clear blockPtr->blockHeader and blockPtr->buffer */
+ bp->b_flags |= B_INVAL;
+ if (hfsmp->jnl && (bp->b_flags & B_LOCKED)) {
+ journal_kill_block(hfsmp->jnl, bp);
+ } else {
+ brelse(bp); /* note: B-tree code will clear blockPtr->blockHeader and blockPtr->buffer */
+ }
} else {
if (options & kForceWriteBlock) {
- bp->b_flags |= B_DIRTY;
- retval = VOP_BWRITE(bp);
+ if (hfsmp->jnl) {
+ if (blockPtr->isModified == 0) {
+ panic("hfs: releaseblock: modified is 0 but forcewrite set! bp 0x%x\n", bp);
+ }
+ retval = journal_modify_block_end(hfsmp->jnl, bp);
+ blockPtr->isModified = 0;
+ } else {
+ retval = VOP_BWRITE(bp);
+ }
} else if (options & kMarkBlockDirty) {
- bp->b_flags |= B_DIRTY;
-#if FORCESYNCBTREEWRITES
- VOP_BWRITE(bp);
-#else
- if (options & kLockTransaction) {
-
- /*
- *
- * Set the B_LOCKED flag and unlock the buffer, causing brelse to move
- * the buffer onto the LOCKED free list. This is necessary, otherwise
- * getnewbuf() would try to reclaim the buffers using bawrite, which
- * isn't going to work.
- *
- */
- bp->b_flags |= B_LOCKED;
- };
- bdwrite(bp);
+ if ((options & kLockTransaction) && hfsmp->jnl == NULL) {
+ /*
+ *
+ * Set the B_LOCKED flag and unlock the buffer, causing brelse to move
+ * the buffer onto the LOCKED free list. This is necessary, otherwise
+ * getnewbuf() would try to reclaim the buffers using bawrite, which
+ * isn't going to work.
+ *
+ */
+ extern int count_lock_queue __P((void));
+ /* Don't hog all the buffers... */
+ if (count_lock_queue() > kMaxLockedMetaBuffers) {
+ hfs_btsync(vp, HFS_SYNCTRANS);
+ /* Rollback sync time to cause a sync on lock release... */
+ (void) BTSetLastSync(VTOF(vp), time.tv_sec - (kMaxSecsForFsync + 1));
+ }
+
+ bp->b_flags |= B_LOCKED;
+ }
-#endif
+ /*
+ * Delay-write this block.
+ * If the maximum delayed buffers has been exceeded then
+ * free up some buffers and fall back to an asynchronous write.
+ */
+ if (hfsmp->jnl) {
+ if (blockPtr->isModified == 0) {
+ panic("hfs: releaseblock: modified is 0 but markdirty set! bp 0x%x\n", bp);
+ }
+ retval = journal_modify_block_end(hfsmp->jnl, bp);
+ blockPtr->isModified = 0;
+ } else if (bdwrite_internal(bp, 1) != 0) {
+ hfs_btsync(vp, 0);
+ /* Rollback sync time to cause a sync on lock release... */
+ (void) BTSetLastSync(VTOF(vp), time.tv_sec - (kMaxSecsForFsync + 1));
+ bp->b_flags &= ~B_LOCKED;
+ bawrite(bp);
+ }
} else {
- brelse(bp); /* note: B-tree code will clear blockPtr->blockHeader and blockPtr->buffer */
+ // check if we had previously called journal_modify_block_start()
+ // on this block and if so, abort it (which will call brelse()).
+ if (hfsmp->jnl && blockPtr->isModified) {
+ // XXXdbg - I don't want to call modify_block_abort()
+ // because I think it may be screwing up the
+ // journal and blowing away a block that has
+ // valid data in it.
+ //
+ // journal_modify_block_abort(hfsmp->jnl, bp);
+ //panic("hfs: releaseblock called for 0x%x but mod_block_start previously called.\n", bp);
+ journal_modify_block_end(hfsmp->jnl, bp);
+ blockPtr->isModified = 0;
+ } else {
+ brelse(bp); /* note: B-tree code will clear blockPtr->blockHeader and blockPtr->buffer */
+ }
};
};
}
+__private_extern__
OSStatus ExtendBTreeFile(FileReference vp, FSSize minEOF, FSSize maxEOF)
{
#pragma unused (maxEOF)
- OSStatus retval;
- UInt64 actualBytesAdded;
+ OSStatus retval, ret;
+ UInt64 actualBytesAdded, origSize;
UInt64 bytesToAdd;
- UInt32 extendFlags;
+ u_int32_t startAllocation;
+ u_int32_t fileblocks;
BTreeInfoRec btInfo;
ExtendedVCB *vcb;
FCB *filePtr;
struct proc *p = NULL;
-
+ UInt64 trim = 0;
filePtr = GetFileControlBlock(vp);
{
bytesToAdd = minEOF - filePtr->fcbEOF;
- if (bytesToAdd < filePtr->fcbClmpSize)
- bytesToAdd = filePtr->fcbClmpSize; //XXX why not always be a mutiple of clump size?
+ if (bytesToAdd < filePtr->ff_clumpsize)
+ bytesToAdd = filePtr->ff_clumpsize; //XXX why not always be a mutiple of clump size?
}
else
{
- DBG_TREE((" ExtendBTreeFile: minEOF is smaller than current size!"));
return -1;
}
- vcb = FCBTOVCB(filePtr);
+ vcb = VTOVCB(vp);
/*
* The Extents B-tree can't have overflow extents. ExtendFileC will
* when the resident extents are exhausted.
*/
/* XXX warning - this can leave the volume bitmap unprotected during ExtendFileC call */
- if(H_FILEID(filePtr) != kHFSExtentsFileID)
+ if(VTOC(vp)->c_fileid != kHFSExtentsFileID)
{
p = current_proc();
/* lock extents b-tree (also protects volume bitmap) */
- retval = hfs_metafilelocking(VTOHFS(vp), kHFSExtentsFileID, LK_EXCLUSIVE, p);
+ retval = hfs_metafilelocking(VTOHFS(vp), kHFSExtentsFileID, LK_EXCLUSIVE, p);
if (retval)
return (retval);
}
(void) BTGetInformation(filePtr, 0, &btInfo);
+#if 0 // XXXdbg
/*
- * The b-tree code expects nodes to be contiguous. So when
+ * The b-tree code expects nodes to be contiguous. So when
* the allocation block size is less than the b-tree node
- * size, we need to force disk allocations to be contiguous.
- */
+ * size, we need to force disk allocations to be contiguous.
+ */
if (vcb->blockSize >= btInfo.nodeSize) {
extendFlags = 0;
} else {
/* Ensure that all b-tree nodes are contiguous on disk */
- extendFlags = kEFAllMask | kEFContigMask;
+ extendFlags = kEFContigMask;
}
+#endif
- retval = ExtendFileC(vcb, filePtr, bytesToAdd, extendFlags, &actualBytesAdded );
-
- if(H_FILEID(filePtr) != kHFSExtentsFileID)
- (void) hfs_metafilelocking(VTOHFS(vp), kHFSExtentsFileID, LK_RELEASE, p);
-
- if (retval)
- return (retval);
-
- if (actualBytesAdded < bytesToAdd)
- DBG_TREE((" ExtendBTreeFile: actualBytesAdded < bytesToAdd!"));
-
- filePtr->fcbEOF = filePtr->fcbPLen;
-
- retval = ClearBTNodes(vp, btInfo.nodeSize, filePtr->fcbEOF - actualBytesAdded, actualBytesAdded);
- if (retval)
- return (retval);
+ origSize = filePtr->fcbEOF;
+ fileblocks = filePtr->ff_blocks;
+ startAllocation = vcb->nextAllocation;
+
+ // loop trying to get a contiguous chunk that's an integer multiple
+ // of the btree node size. if we can't get a contiguous chunk that
+ // is at least the node size then we break out of the loop and let
+ // the error propagate back up.
+ do {
+ retval = ExtendFileC(vcb, filePtr, bytesToAdd, 0, kEFContigMask, &actualBytesAdded);
+ if (retval == dskFulErr && actualBytesAdded == 0) {
+
+ if (bytesToAdd == btInfo.nodeSize || bytesToAdd < (minEOF - origSize)) {
+ // if we're here there's nothing else to try, we're out
+ // of space so we break and bail out.
+ break;
+ } else {
+ bytesToAdd >>= 1;
+ if (bytesToAdd < btInfo.nodeSize) {
+ bytesToAdd = btInfo.nodeSize;
+ } else if ((bytesToAdd % btInfo.nodeSize) != 0) {
+ // make sure it's an integer multiple of the nodeSize
+ bytesToAdd -= (bytesToAdd % btInfo.nodeSize);
+ }
+ }
+ }
+ } while (retval == dskFulErr && actualBytesAdded == 0);
/*
- * Update the Alternate MDB or Alternate VolumeHeader
+ * If a new extent was added then move the roving allocator
+ * reference forward by the current b-tree file size so
+ * there's plenty of room to grow.
*/
- if ((H_FILEID(filePtr) == kHFSExtentsFileID) ||
- (H_FILEID(filePtr) == kHFSCatalogFileID) ||
- (H_FILEID(filePtr) == kHFSAttributesFileID)
- ) {
- MarkVCBDirty( vcb );
- if (vcb->vcbSigWord == kHFSPlusSigWord) {
- retval = hfs_flushvolumeheader(VCBTOHFS(vcb), 0);
+ if ((retval == 0) &&
+ (vcb->nextAllocation > startAllocation) &&
+ ((vcb->nextAllocation + fileblocks) < vcb->totalBlocks)) {
+ vcb->nextAllocation += fileblocks;
+ }
+
+ filePtr->fcbEOF = (u_int64_t)filePtr->ff_blocks * (u_int64_t)vcb->blockSize;
+
+ // XXXdbg ExtendFileC() could have returned an error even though
+ // it grew the file to be big enough for our needs. If this is
+ // the case, we don't care about retval so we blow it away.
+ //
+ if (filePtr->fcbEOF >= minEOF && retval != 0) {
+ retval = 0;
+ }
+
+ // XXXdbg if the file grew but isn't large enough or isn't an
+ // even multiple of the nodeSize then trim things back. if
+ // the file isn't large enough we trim back to the original
+ // size. otherwise we trim back to be an even multiple of the
+ // btree node size.
+ //
+ if ((filePtr->fcbEOF < minEOF) || (actualBytesAdded % btInfo.nodeSize) != 0) {
+
+ if (filePtr->fcbEOF < minEOF) {
+ retval = dskFulErr;
+
+ if (filePtr->fcbEOF < origSize) {
+ panic("hfs: btree file eof %lld less than orig size %lld!\n",
+ filePtr->fcbEOF, origSize);
+ }
+
+ trim = filePtr->fcbEOF - origSize;
+ if (trim != actualBytesAdded) {
+ panic("hfs: trim == %lld but actualBytesAdded == %lld\n",
+ trim, actualBytesAdded);
+ }
} else {
- retval = hfs_flushMDB(VCBTOHFS(vcb), 0);
+ trim = (actualBytesAdded % btInfo.nodeSize);
}
- if (retval == 0) {
- retval = FlushAlternate(vcb);
+
+ ret = TruncateFileC(vcb, filePtr, filePtr->fcbEOF - trim, 0);
+ filePtr->fcbEOF = (u_int64_t)filePtr->ff_blocks * (u_int64_t)vcb->blockSize;
+
+ // XXXdbg - panic if the file didn't get trimmed back properly
+ if ((filePtr->fcbEOF % btInfo.nodeSize) != 0) {
+ panic("hfs: truncate file didn't! fcbEOF %lld nsize %d fcb 0x%x\n",
+ filePtr->fcbEOF, btInfo.nodeSize, filePtr);
}
+
+ if (ret) {
+ // XXXdbg - this probably doesn't need to be a panic()
+ panic("hfs: error truncating btree files (sz 0x%llx, trim %lld, ret %d)\n",
+ filePtr->fcbEOF, trim, ret);
+ return ret;
+ }
+ actualBytesAdded -= trim;
}
-
- return retval;
-}
+ if(VTOC(vp)->c_fileid != kHFSExtentsFileID) {
+ /*
+ * Get any extents overflow b-tree changes to disk ASAP!
+ */
+ (void) BTFlushPath(VTOF(vcb->extentsRefNum));
+ (void) VOP_FSYNC(vcb->extentsRefNum, NOCRED, MNT_WAIT, p);
-static OSStatus
-FlushAlternate( ExtendedVCB *vcb )
-{
- void *maindata;
- void *altdata;
- int result;
-
- /* Get the main MDB/VolumeHeader block */
- result = GetBlock_glue(gbDefault,
- (vcb->hfsPlusIOPosOffset / kHFSBlockSize) + kMasterDirectoryBlock,
- (Ptr *)&maindata, kNoFileReference, vcb);
- if (result) return (result);
-
- /* Get the alternate MDB/VolumeHeader block */
- result = GetBlock_glue( gbDefault, vcb->altIDSector,
- (Ptr *)&altdata, kNoFileReference, vcb );
+ (void) hfs_metafilelocking(VTOHFS(vp), kHFSExtentsFileID, LK_RELEASE, p);
+ }
- if (result == 0) {
- bcopy(maindata, altdata, kMDBSize);
+ if ((filePtr->fcbEOF % btInfo.nodeSize) != 0) {
+ panic("hfs: extendbtree: fcb 0x%x has eof 0x%llx not a multiple of 0x%x (trim %llx)\n",
+ filePtr, filePtr->fcbEOF, btInfo.nodeSize, trim);
+ }
- result = RelBlock_glue( (Ptr)altdata, rbWriteMask );
+ /*
+ * Update the Alternate MDB or Alternate VolumeHeader
+ */
+ if ((VTOC(vp)->c_fileid == kHFSExtentsFileID) ||
+ (VTOC(vp)->c_fileid == kHFSCatalogFileID) ||
+ (VTOC(vp)->c_fileid == kHFSAttributesFileID)
+ ) {
+ MarkVCBDirty( vcb );
+ ret = hfs_flushvolumeheader(VCBTOHFS(vcb), MNT_WAIT, HFS_ALTFLUSH);
}
- (void) RelBlock_glue( (Ptr)maindata, rbFreeMask );
+ ret = ClearBTNodes(vp, btInfo.nodeSize, filePtr->fcbEOF - actualBytesAdded, actualBytesAdded);
+ if (ret)
+ return (ret);
- return (result);
+ return retval;
}
static int
ClearBTNodes(struct vnode *vp, long blksize, off_t offset, off_t amount)
{
+ struct hfsmount *hfsmp = VTOHFS(vp);
struct buf *bp = NULL;
daddr_t blk;
daddr_t blkcnt;
bp = getblk(vp, blk, blksize, 0, 0, BLK_META);
if (bp == NULL)
continue;
- bzero((char *)bp->b_data, blksize);
- bp->b_flags |= (B_DIRTY | B_AGE);
- /* wait/yield every 32 blocks so we don't hog all the buffers */
- if ((blk % 32) == 0)
- VOP_BWRITE(bp);
- else
- bawrite(bp);
+ // XXXdbg
+ if (hfsmp->jnl) {
+ // XXXdbg -- skipping this for now since it makes a transaction
+ // become *way* too large
+ //journal_modify_block_start(hfsmp->jnl, bp);
+ }
+
+ bzero((char *)bp->b_data, blksize);
+ bp->b_flags |= B_AGE;
+
+ // XXXdbg
+ if (hfsmp->jnl) {
+ // XXXdbg -- skipping this for now since it makes a transaction
+ // become *way* too large
+ //journal_modify_block_end(hfsmp->jnl, bp);
+
+ // XXXdbg - remove this once we decide what to do with the
+ // writes to the journal
+ if ((blk % 32) == 0)
+ VOP_BWRITE(bp);
+ else
+ bawrite(bp);
+ } else {
+ /* wait/yield every 32 blocks so we don't hog all the buffers */
+ if ((blk % 32) == 0)
+ VOP_BWRITE(bp);
+ else
+ bawrite(bp);
+ }
--blkcnt;
++blk;
}