/* * Copyright (c) 2013-2014 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * * 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. The rights granted to you under the License * may not be used to create, or enable the creation or redistribution of, * unlawful or unlicensed copies of an Apple operating system, or to * circumvent, violate, or enable the circumvention or violation of, any * terms of an Apple operating system software license agreement. * * 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, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ #include #include #include #include #include #include #include #include #include "hfs.h" #include "hfs_catalog.h" #include "hfs_cnode.h" #include "hfs_endian.h" #include "hfs_btreeio.h" #if CONFIG_PROTECT #include #endif /* Enable/disable debugging code for live volume resizing */ int hfs_resize_debug = 0; static int hfs_file_extent_overlaps(struct hfsmount *hfsmp, u_int32_t allocLimit, struct HFSPlusCatalogFile *filerec); static int hfs_reclaimspace(struct hfsmount *hfsmp, u_int32_t allocLimit, u_int32_t reclaimblks, vfs_context_t context); static int hfs_extend_journal(struct hfsmount *hfsmp, u_int32_t sector_size, u_int64_t sector_count, vfs_context_t context); /* * Extend a file system. */ int hfs_extendfs(struct hfsmount *hfsmp, u_int64_t newsize, vfs_context_t context) { struct proc *p = vfs_context_proc(context); kauth_cred_t cred = vfs_context_ucred(context); struct vnode *vp; struct vnode *devvp; struct buf *bp; struct filefork *fp = NULL; ExtendedVCB *vcb; struct cat_fork forkdata; u_int64_t oldsize; u_int64_t newblkcnt; u_int64_t prev_phys_block_count; u_int32_t addblks; u_int64_t sector_count; u_int32_t sector_size; u_int32_t phys_sector_size; u_int32_t overage_blocks; daddr64_t prev_fs_alt_sector; daddr_t bitmapblks; int lockflags = 0; int error; int64_t oldBitmapSize; Boolean usedExtendFileC = false; int transaction_begun = 0; devvp = hfsmp->hfs_devvp; vcb = HFSTOVCB(hfsmp); /* * - HFS Plus file systems only. * - Journaling must be enabled. * - No embedded volumes. */ if ((vcb->vcbSigWord == kHFSSigWord) || (hfsmp->jnl == NULL) || (vcb->hfsPlusIOPosOffset != 0)) { return (EPERM); } /* * If extending file system by non-root, then verify * ownership and check permissions. */ if (suser(cred, NULL)) { error = hfs_vget(hfsmp, kHFSRootFolderID, &vp, 0, 0); if (error) return (error); error = hfs_owner_rights(hfsmp, VTOC(vp)->c_uid, cred, p, 0); if (error == 0) { error = hfs_write_access(vp, cred, p, false); } hfs_unlock(VTOC(vp)); vnode_put(vp); if (error) return (error); error = vnode_authorize(devvp, NULL, KAUTH_VNODE_READ_DATA | KAUTH_VNODE_WRITE_DATA, context); if (error) return (error); } if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)§or_size, 0, context)) { return (ENXIO); } if (sector_size != hfsmp->hfs_logical_block_size) { return (ENXIO); } if (VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)§or_count, 0, context)) { return (ENXIO); } /* Check if partition size is correct for new file system size */ if ((sector_size * sector_count) < newsize) { printf("hfs_extendfs: not enough space on device (vol=%s)\n", hfsmp->vcbVN); return (ENOSPC); } error = VNOP_IOCTL(devvp, DKIOCGETPHYSICALBLOCKSIZE, (caddr_t)&phys_sector_size, 0, context); if (error) { if ((error != ENOTSUP) && (error != ENOTTY)) { return (ENXIO); } /* If ioctl is not supported, force physical and logical sector size to be same */ phys_sector_size = sector_size; } oldsize = (u_int64_t)hfsmp->totalBlocks * (u_int64_t)hfsmp->blockSize; /* * Validate new size. */ if ((newsize <= oldsize) || (newsize % sector_size) || (newsize % phys_sector_size)) { printf("hfs_extendfs: invalid size (newsize=%qu, oldsize=%qu)\n", newsize, oldsize); return (EINVAL); } newblkcnt = newsize / vcb->blockSize; if (newblkcnt > (u_int64_t)0xFFFFFFFF) { printf ("hfs_extendfs: current blockSize=%u too small for newsize=%qu\n", hfsmp->blockSize, newsize); return (EOVERFLOW); } addblks = newblkcnt - vcb->totalBlocks; if (hfs_resize_debug) { printf ("hfs_extendfs: old: size=%qu, blkcnt=%u\n", oldsize, hfsmp->totalBlocks); printf ("hfs_extendfs: new: size=%qu, blkcnt=%u, addblks=%u\n", newsize, (u_int32_t)newblkcnt, addblks); } printf("hfs_extendfs: will extend \"%s\" by %d blocks\n", vcb->vcbVN, addblks); hfs_lock_mount (hfsmp); if (hfsmp->hfs_flags & HFS_RESIZE_IN_PROGRESS) { hfs_unlock_mount(hfsmp); error = EALREADY; goto out; } hfsmp->hfs_flags |= HFS_RESIZE_IN_PROGRESS; hfs_unlock_mount (hfsmp); /* Start with a clean journal. */ hfs_journal_flush(hfsmp, TRUE); /* * Enclose changes inside a transaction. */ if (hfs_start_transaction(hfsmp) != 0) { error = EINVAL; goto out; } transaction_begun = 1; /* Update the hfsmp fields for the physical information about the device */ prev_phys_block_count = hfsmp->hfs_logical_block_count; prev_fs_alt_sector = hfsmp->hfs_fs_avh_sector; hfsmp->hfs_logical_block_count = sector_count; hfsmp->hfs_logical_bytes = (uint64_t) sector_count * (uint64_t) sector_size; /* * It is possible that the new file system is smaller than the partition size. * Therefore, update offsets for AVH accordingly. */ if (hfs_resize_debug) { printf ("hfs_extendfs: old: partition_avh_sector=%qu, fs_avh_sector=%qu\n", hfsmp->hfs_partition_avh_sector, hfsmp->hfs_fs_avh_sector); } hfsmp->hfs_partition_avh_sector = (hfsmp->hfsPlusIOPosOffset / sector_size) + HFS_ALT_SECTOR(sector_size, hfsmp->hfs_logical_block_count); hfsmp->hfs_fs_avh_sector = (hfsmp->hfsPlusIOPosOffset / sector_size) + HFS_ALT_SECTOR(sector_size, (newsize/hfsmp->hfs_logical_block_size)); if (hfs_resize_debug) { printf ("hfs_extendfs: new: partition_avh_sector=%qu, fs_avh_sector=%qu\n", hfsmp->hfs_partition_avh_sector, hfsmp->hfs_fs_avh_sector); } /* * Note: we take the attributes lock in case we have an attribute data vnode * which needs to change size. */ lockflags = hfs_systemfile_lock(hfsmp, SFL_ATTRIBUTE | SFL_EXTENTS | SFL_BITMAP, HFS_EXCLUSIVE_LOCK); vp = vcb->allocationsRefNum; fp = VTOF(vp); bcopy(&fp->ff_data, &forkdata, sizeof(forkdata)); /* * Calculate additional space required (if any) by allocation bitmap. */ oldBitmapSize = fp->ff_size; bitmapblks = roundup((newblkcnt+7) / 8, vcb->vcbVBMIOSize) / vcb->blockSize; if (bitmapblks > (daddr_t)fp->ff_blocks) bitmapblks -= fp->ff_blocks; else bitmapblks = 0; /* * The allocation bitmap can contain unused bits that are beyond end of * current volume's allocation blocks. Usually they are supposed to be * zero'ed out but there can be cases where they might be marked as used. * After extending the file system, those bits can represent valid * allocation blocks, so we mark all the bits from the end of current * volume to end of allocation bitmap as "free". * * Figure out the number of overage blocks before proceeding though, * so we don't add more bytes to our I/O than necessary. * First figure out the total number of blocks representable by the * end of the bitmap file vs. the total number of blocks in the new FS. * Then subtract away the number of blocks in the current FS. This is how much * we can mark as free right now without having to grow the bitmap file. */ overage_blocks = fp->ff_blocks * vcb->blockSize * 8; overage_blocks = MIN (overage_blocks, newblkcnt); overage_blocks -= vcb->totalBlocks; BlockMarkFreeUnused(vcb, vcb->totalBlocks, overage_blocks); if (bitmapblks > 0) { daddr64_t blkno; daddr_t blkcnt; off_t bytesAdded; /* * Get the bitmap's current size (in allocation blocks) so we know * where to start zero filling once the new space is added. We've * got to do this before the bitmap is grown. */ blkno = (daddr64_t)fp->ff_blocks; /* * Try to grow the allocation file in the normal way, using allocation * blocks already existing in the file system. This way, we might be * able to grow the bitmap contiguously, or at least in the metadata * zone. */ error = ExtendFileC(vcb, fp, bitmapblks * vcb->blockSize, 0, kEFAllMask | kEFNoClumpMask | kEFReserveMask | kEFMetadataMask | kEFContigMask, &bytesAdded); if (error == 0) { usedExtendFileC = true; } else { /* * If the above allocation failed, fall back to allocating the new * extent of the bitmap from the space we're going to add. Since those * blocks don't yet belong to the file system, we have to update the * extent list directly, and manually adjust the file size. */ bytesAdded = 0; error = AddFileExtent(vcb, fp, vcb->totalBlocks, bitmapblks); if (error) { printf("hfs_extendfs: error %d adding extents\n", error); goto out; } fp->ff_blocks += bitmapblks; VTOC(vp)->c_blocks = fp->ff_blocks; VTOC(vp)->c_flag |= C_MODIFIED; } /* * Update the allocation file's size to include the newly allocated * blocks. Note that ExtendFileC doesn't do this, which is why this * statement is outside the above "if" statement. */ fp->ff_size += (u_int64_t)bitmapblks * (u_int64_t)vcb->blockSize; /* * Zero out the new bitmap blocks. */ { bp = NULL; blkcnt = bitmapblks; while (blkcnt > 0) { error = (int)buf_meta_bread(vp, blkno, vcb->blockSize, NOCRED, &bp); if (error) { if (bp) { buf_brelse(bp); } break; } bzero((char *)buf_dataptr(bp), vcb->blockSize); buf_markaged(bp); error = (int)buf_bwrite(bp); if (error) break; --blkcnt; ++blkno; } } if (error) { printf("hfs_extendfs: error %d clearing blocks\n", error); goto out; } /* * Mark the new bitmap space as allocated. * * Note that ExtendFileC will have marked any blocks it allocated, so * this is only needed if we used AddFileExtent. Also note that this * has to come *after* the zero filling of new blocks in the case where * we used AddFileExtent (since the part of the bitmap we're touching * is in those newly allocated blocks). */ if (!usedExtendFileC) { error = BlockMarkAllocated(vcb, vcb->totalBlocks, bitmapblks); if (error) { printf("hfs_extendfs: error %d setting bitmap\n", error); goto out; } vcb->freeBlocks -= bitmapblks; } } /* * Mark the new alternate VH as allocated. */ if (vcb->blockSize == 512) error = BlockMarkAllocated(vcb, vcb->totalBlocks + addblks - 2, 2); else error = BlockMarkAllocated(vcb, vcb->totalBlocks + addblks - 1, 1); if (error) { printf("hfs_extendfs: error %d setting bitmap (VH)\n", error); goto out; } /* * Mark the old alternate VH as free. */ if (vcb->blockSize == 512) (void) BlockMarkFree(vcb, vcb->totalBlocks - 2, 2); else (void) BlockMarkFree(vcb, vcb->totalBlocks - 1, 1); /* * Adjust file system variables for new space. */ vcb->totalBlocks += addblks; vcb->freeBlocks += addblks; MarkVCBDirty(vcb); error = hfs_flushvolumeheader(hfsmp, MNT_WAIT, HFS_ALTFLUSH); if (error) { printf("hfs_extendfs: couldn't flush volume headers (%d)", error); /* * Restore to old state. */ if (usedExtendFileC) { (void) TruncateFileC(vcb, fp, oldBitmapSize, 0, FORK_IS_RSRC(fp), FTOC(fp)->c_fileid, false); } else { fp->ff_blocks -= bitmapblks; fp->ff_size -= (u_int64_t)bitmapblks * (u_int64_t)vcb->blockSize; /* * No need to mark the excess blocks free since those bitmap blocks * are no longer part of the bitmap. But we do need to undo the * effect of the "vcb->freeBlocks -= bitmapblks" above. */ vcb->freeBlocks += bitmapblks; } vcb->totalBlocks -= addblks; vcb->freeBlocks -= addblks; hfsmp->hfs_logical_block_count = prev_phys_block_count; hfsmp->hfs_fs_avh_sector = prev_fs_alt_sector; /* Do not revert hfs_partition_avh_sector because the * partition size is larger than file system size */ MarkVCBDirty(vcb); if (vcb->blockSize == 512) { if (BlockMarkAllocated(vcb, vcb->totalBlocks - 2, 2)) { hfs_mark_inconsistent(hfsmp, HFS_ROLLBACK_FAILED); } } else { if (BlockMarkAllocated(vcb, vcb->totalBlocks - 1, 1)) { hfs_mark_inconsistent(hfsmp, HFS_ROLLBACK_FAILED); } } goto out; } /* * Invalidate the old alternate volume header. We are growing the filesystem so * this sector must be returned to the FS as free space. */ bp = NULL; if (prev_fs_alt_sector) { if (buf_meta_bread(hfsmp->hfs_devvp, HFS_PHYSBLK_ROUNDDOWN(prev_fs_alt_sector, hfsmp->hfs_log_per_phys), hfsmp->hfs_physical_block_size, NOCRED, &bp) == 0) { journal_modify_block_start(hfsmp->jnl, bp); bzero((char *)buf_dataptr(bp) + HFS_ALT_OFFSET(hfsmp->hfs_physical_block_size), kMDBSize); journal_modify_block_end(hfsmp->jnl, bp, NULL, NULL); } else if (bp) { buf_brelse(bp); } } /* * Update the metadata zone size based on current volume size */ hfs_metadatazone_init(hfsmp, false); /* * Adjust the size of hfsmp->hfs_attrdata_vp */ if (hfsmp->hfs_attrdata_vp) { struct cnode *attr_cp; struct filefork *attr_fp; if (vnode_get(hfsmp->hfs_attrdata_vp) == 0) { attr_cp = VTOC(hfsmp->hfs_attrdata_vp); attr_fp = VTOF(hfsmp->hfs_attrdata_vp); attr_cp->c_blocks = newblkcnt; attr_fp->ff_blocks = newblkcnt; attr_fp->ff_extents[0].blockCount = newblkcnt; attr_fp->ff_size = (off_t) newblkcnt * hfsmp->blockSize; ubc_setsize(hfsmp->hfs_attrdata_vp, attr_fp->ff_size); vnode_put(hfsmp->hfs_attrdata_vp); } } /* * We only update hfsmp->allocLimit if totalBlocks actually increased. */ if (error == 0) { UpdateAllocLimit(hfsmp, hfsmp->totalBlocks); } /* Release all locks and sync up journal content before * checking and extending, if required, the journal */ if (lockflags) { hfs_systemfile_unlock(hfsmp, lockflags); lockflags = 0; } if (transaction_begun) { hfs_end_transaction(hfsmp); hfs_journal_flush(hfsmp, TRUE); transaction_begun = 0; } /* Increase the journal size, if required. */ error = hfs_extend_journal(hfsmp, sector_size, sector_count, context); if (error) { printf ("hfs_extendfs: Could not extend journal size\n"); goto out_noalloc; } /* Log successful extending */ printf("hfs_extendfs: extended \"%s\" to %d blocks (was %d blocks)\n", hfsmp->vcbVN, hfsmp->totalBlocks, (u_int32_t)(oldsize/hfsmp->blockSize)); out: if (error && fp) { /* Restore allocation fork. */ bcopy(&forkdata, &fp->ff_data, sizeof(forkdata)); VTOC(vp)->c_blocks = fp->ff_blocks; } out_noalloc: hfs_lock_mount (hfsmp); hfsmp->hfs_flags &= ~HFS_RESIZE_IN_PROGRESS; hfs_unlock_mount (hfsmp); if (lockflags) { hfs_systemfile_unlock(hfsmp, lockflags); } if (transaction_begun) { hfs_end_transaction(hfsmp); hfs_journal_flush(hfsmp, FALSE); /* Just to be sure, sync all data to the disk */ (void) VNOP_IOCTL(hfsmp->hfs_devvp, DKIOCSYNCHRONIZECACHE, NULL, FWRITE, context); } if (error) { printf ("hfs_extentfs: failed error=%d on vol=%s\n", MacToVFSError(error), hfsmp->vcbVN); } return MacToVFSError(error); } #define HFS_MIN_SIZE (32LL * 1024LL * 1024LL) /* * Truncate a file system (while still mounted). */ int hfs_truncatefs(struct hfsmount *hfsmp, u_int64_t newsize, vfs_context_t context) { u_int64_t oldsize; u_int32_t newblkcnt; u_int32_t reclaimblks = 0; int lockflags = 0; int transaction_begun = 0; Boolean updateFreeBlocks = false; Boolean disable_sparse = false; int error = 0; hfs_lock_mount (hfsmp); if (hfsmp->hfs_flags & HFS_RESIZE_IN_PROGRESS) { hfs_unlock_mount (hfsmp); return (EALREADY); } hfsmp->hfs_flags |= HFS_RESIZE_IN_PROGRESS; hfsmp->hfs_resize_blocksmoved = 0; hfsmp->hfs_resize_totalblocks = 0; hfsmp->hfs_resize_progress = 0; hfs_unlock_mount (hfsmp); /* * - Journaled HFS Plus volumes only. * - No embedded volumes. */ if ((hfsmp->jnl == NULL) || (hfsmp->hfsPlusIOPosOffset != 0)) { error = EPERM; goto out; } oldsize = (u_int64_t)hfsmp->totalBlocks * (u_int64_t)hfsmp->blockSize; newblkcnt = newsize / hfsmp->blockSize; reclaimblks = hfsmp->totalBlocks - newblkcnt; if (hfs_resize_debug) { printf ("hfs_truncatefs: old: size=%qu, blkcnt=%u, freeblks=%u\n", oldsize, hfsmp->totalBlocks, hfs_freeblks(hfsmp, 1)); printf ("hfs_truncatefs: new: size=%qu, blkcnt=%u, reclaimblks=%u\n", newsize, newblkcnt, reclaimblks); } /* Make sure new size is valid. */ if ((newsize < HFS_MIN_SIZE) || (newsize >= oldsize) || (newsize % hfsmp->hfs_logical_block_size) || (newsize % hfsmp->hfs_physical_block_size)) { printf ("hfs_truncatefs: invalid size (newsize=%qu, oldsize=%qu)\n", newsize, oldsize); error = EINVAL; goto out; } /* * Make sure that the file system has enough free blocks reclaim. * * Before resize, the disk is divided into four zones - * A. Allocated_Stationary - These are allocated blocks that exist * before the new end of disk. These blocks will not be * relocated or modified during resize. * B. Free_Stationary - These are free blocks that exist before the * new end of disk. These blocks can be used for any new * allocations during resize, including allocation for relocating * data from the area of disk being reclaimed. * C. Allocated_To-Reclaim - These are allocated blocks that exist * beyond the new end of disk. These blocks need to be reclaimed * during resize by allocating equal number of blocks in Free * Stationary zone and copying the data. * D. Free_To-Reclaim - These are free blocks that exist beyond the * new end of disk. Nothing special needs to be done to reclaim * them. * * Total number of blocks on the disk before resize: * ------------------------------------------------ * Total Blocks = Allocated_Stationary + Free_Stationary + * Allocated_To-Reclaim + Free_To-Reclaim * * Total number of blocks that need to be reclaimed: * ------------------------------------------------ * Blocks to Reclaim = Allocated_To-Reclaim + Free_To-Reclaim * * Note that the check below also makes sure that we have enough space * to relocate data from Allocated_To-Reclaim to Free_Stationary. * Therefore we do not need to check total number of blocks to relocate * later in the code. * * The condition below gets converted to: * * Allocated To-Reclaim + Free To-Reclaim >= Free Stationary + Free To-Reclaim * * which is equivalent to: * * Allocated To-Reclaim >= Free Stationary */ if (reclaimblks >= hfs_freeblks(hfsmp, 1)) { printf("hfs_truncatefs: insufficient space (need %u blocks; have %u free blocks)\n", reclaimblks, hfs_freeblks(hfsmp, 1)); error = ENOSPC; goto out; } /* Start with a clean journal. */ hfs_journal_flush(hfsmp, TRUE); if (hfs_start_transaction(hfsmp) != 0) { error = EINVAL; goto out; } transaction_begun = 1; /* Take the bitmap lock to update the alloc limit field */ lockflags = hfs_systemfile_lock(hfsmp, SFL_BITMAP, HFS_EXCLUSIVE_LOCK); /* * Prevent new allocations from using the part we're trying to truncate. * * NOTE: allocLimit is set to the allocation block number where the new * alternate volume header will be. That way there will be no files to * interfere with allocating the new alternate volume header, and no files * in the allocation blocks beyond (i.e. the blocks we're trying to * truncate away. */ if (hfsmp->blockSize == 512) { error = UpdateAllocLimit (hfsmp, newblkcnt - 2); } else { error = UpdateAllocLimit (hfsmp, newblkcnt - 1); } /* Sparse devices use first fit allocation which is not ideal * for volume resize which requires best fit allocation. If a * sparse device is being truncated, disable the sparse device * property temporarily for the duration of resize. Also reset * the free extent cache so that it is rebuilt as sorted by * totalBlocks instead of startBlock. * * Note that this will affect all allocations on the volume and * ideal fix would be just to modify resize-related allocations, * but it will result in complexity like handling of two free * extent caches sorted differently, etc. So we stick to this * solution for now. */ hfs_lock_mount (hfsmp); if (hfsmp->hfs_flags & HFS_HAS_SPARSE_DEVICE) { hfsmp->hfs_flags &= ~HFS_HAS_SPARSE_DEVICE; ResetVCBFreeExtCache(hfsmp); disable_sparse = true; } /* * Update the volume free block count to reflect the total number * of free blocks that will exist after a successful resize. * Relocation of extents will result in no net change in the total * free space on the disk. Therefore the code that allocates * space for new extent and deallocates the old extent explicitly * prevents updating the volume free block count. It will also * prevent false disk full error when the number of blocks in * an extent being relocated is more than the free blocks that * will exist after the volume is resized. */ hfsmp->freeBlocks -= reclaimblks; updateFreeBlocks = true; hfs_unlock_mount(hfsmp); if (lockflags) { hfs_systemfile_unlock(hfsmp, lockflags); lockflags = 0; } /* * Update the metadata zone size to match the new volume size, * and if it too less, metadata zone might be disabled. */ hfs_metadatazone_init(hfsmp, false); /* * If some files have blocks at or beyond the location of the * new alternate volume header, recalculate free blocks and * reclaim blocks. Otherwise just update free blocks count. * * The current allocLimit is set to the location of new alternate * volume header, and reclaimblks are the total number of blocks * that need to be reclaimed. So the check below is really * ignoring the blocks allocated for old alternate volume header. */ if (hfs_isallocated(hfsmp, hfsmp->allocLimit, reclaimblks)) { /* * hfs_reclaimspace will use separate transactions when * relocating files (so we don't overwhelm the journal). */ hfs_end_transaction(hfsmp); transaction_begun = 0; /* Attempt to reclaim some space. */ error = hfs_reclaimspace(hfsmp, hfsmp->allocLimit, reclaimblks, context); if (error != 0) { printf("hfs_truncatefs: couldn't reclaim space on %s (error=%d)\n", hfsmp->vcbVN, error); error = ENOSPC; goto out; } if (hfs_start_transaction(hfsmp) != 0) { error = EINVAL; goto out; } transaction_begun = 1; /* Check if we're clear now. */ error = hfs_isallocated(hfsmp, hfsmp->allocLimit, reclaimblks); if (error != 0) { printf("hfs_truncatefs: didn't reclaim enough space on %s (error=%d)\n", hfsmp->vcbVN, error); error = EAGAIN; /* tell client to try again */ goto out; } } /* * Note: we take the attributes lock in case we have an attribute data vnode * which needs to change size. */ lockflags = hfs_systemfile_lock(hfsmp, SFL_ATTRIBUTE | SFL_EXTENTS | SFL_BITMAP, HFS_EXCLUSIVE_LOCK); /* * Allocate last 1KB for alternate volume header. */ error = BlockMarkAllocated(hfsmp, hfsmp->allocLimit, (hfsmp->blockSize == 512) ? 2 : 1); if (error) { printf("hfs_truncatefs: Error %d allocating new alternate volume header\n", error); goto out; } /* * Mark the old alternate volume header as free. * We don't bother shrinking allocation bitmap file. */ if (hfsmp->blockSize == 512) (void) BlockMarkFree(hfsmp, hfsmp->totalBlocks - 2, 2); else (void) BlockMarkFree(hfsmp, hfsmp->totalBlocks - 1, 1); /* Don't invalidate the old AltVH yet. It is still valid until the partition size is updated ! */ /* Log successful shrinking. */ printf("hfs_truncatefs: shrank \"%s\" to %d blocks (was %d blocks)\n", hfsmp->vcbVN, newblkcnt, hfsmp->totalBlocks); /* * Adjust file system variables and flush them to disk. * * Note that although the logical block size is updated here, it is only * done for the benefit/convenience of the partition management software. The * logical block count change has not yet actually been propagated to * the disk device yet (and we won't get any notification when it does). */ hfsmp->totalBlocks = newblkcnt; hfsmp->hfs_logical_block_count = newsize / hfsmp->hfs_logical_block_size; hfsmp->hfs_logical_bytes = (uint64_t) hfsmp->hfs_logical_block_count * (uint64_t) hfsmp->hfs_logical_block_size; /* * At this point, a smaller HFS file system exists in a larger volume. * As per volume format, the alternate volume header is located 1024 bytes * before end of the partition. So, until the partition is also resized, * a valid alternate volume header will need to be updated at 1024 bytes * before end of the volume. Under normal circumstances, a file system * resize is always followed by a volume resize, so we also need to * write a copy of the new alternate volume header at 1024 bytes before * end of the new file system. */ if (hfs_resize_debug) { printf ("hfs_truncatefs: old: partition_avh_sector=%qu, fs_avh_sector=%qu\n", hfsmp->hfs_partition_avh_sector, hfsmp->hfs_fs_avh_sector); } hfsmp->hfs_fs_avh_sector = HFS_ALT_SECTOR(hfsmp->hfs_logical_block_size, hfsmp->hfs_logical_block_count); /* Note hfs_partition_avh_sector stays unchanged! partition size has not yet been modified */ if (hfs_resize_debug) { printf ("hfs_truncatefs: new: partition_avh_sector=%qu, fs_avh_sector=%qu\n", hfsmp->hfs_partition_avh_sector, hfsmp->hfs_fs_avh_sector); } MarkVCBDirty(hfsmp); error = hfs_flushvolumeheader(hfsmp, MNT_WAIT, HFS_ALTFLUSH); if (error) { panic("hfs_truncatefs: unexpected error flushing volume header (%d)\n", error); } /* * Adjust the size of hfsmp->hfs_attrdata_vp */ if (hfsmp->hfs_attrdata_vp) { struct cnode *cp; struct filefork *fp; if (vnode_get(hfsmp->hfs_attrdata_vp) == 0) { cp = VTOC(hfsmp->hfs_attrdata_vp); fp = VTOF(hfsmp->hfs_attrdata_vp); cp->c_blocks = newblkcnt; fp->ff_blocks = newblkcnt; fp->ff_extents[0].blockCount = newblkcnt; fp->ff_size = (off_t) newblkcnt * hfsmp->blockSize; ubc_setsize(hfsmp->hfs_attrdata_vp, fp->ff_size); vnode_put(hfsmp->hfs_attrdata_vp); } } out: /* * Update the allocLimit to acknowledge the last one or two blocks now. * Add it to the tree as well if necessary. */ UpdateAllocLimit (hfsmp, hfsmp->totalBlocks); hfs_lock_mount (hfsmp); if (disable_sparse == true) { /* Now that resize is completed, set the volume to be sparse * device again so that all further allocations will be first * fit instead of best fit. Reset free extent cache so that * it is rebuilt. */ hfsmp->hfs_flags |= HFS_HAS_SPARSE_DEVICE; ResetVCBFreeExtCache(hfsmp); } if (error && (updateFreeBlocks == true)) { hfsmp->freeBlocks += reclaimblks; } if (hfsmp->nextAllocation >= hfsmp->allocLimit) { hfsmp->nextAllocation = hfsmp->hfs_metazone_end + 1; } hfsmp->hfs_flags &= ~HFS_RESIZE_IN_PROGRESS; hfs_unlock_mount (hfsmp); /* On error, reset the metadata zone for original volume size */ if (error && (updateFreeBlocks == true)) { hfs_metadatazone_init(hfsmp, false); } if (lockflags) { hfs_systemfile_unlock(hfsmp, lockflags); } if (transaction_begun) { hfs_end_transaction(hfsmp); hfs_journal_flush(hfsmp, FALSE); /* Just to be sure, sync all data to the disk */ (void) VNOP_IOCTL(hfsmp->hfs_devvp, DKIOCSYNCHRONIZECACHE, NULL, FWRITE, context); } if (error) { printf ("hfs_truncatefs: failed error=%d on vol=%s\n", MacToVFSError(error), hfsmp->vcbVN); } return MacToVFSError(error); } /* * Invalidate the physical block numbers associated with buffer cache blocks * in the given extent of the given vnode. */ struct hfs_inval_blk_no { daddr64_t sectorStart; daddr64_t sectorCount; }; static int hfs_invalidate_block_numbers_callback(buf_t bp, void *args_in) { daddr64_t blkno; struct hfs_inval_blk_no *args; blkno = buf_blkno(bp); args = args_in; if (blkno >= args->sectorStart && blkno < args->sectorStart+args->sectorCount) buf_setblkno(bp, buf_lblkno(bp)); return BUF_RETURNED; } static void hfs_invalidate_sectors(struct vnode *vp, daddr64_t sectorStart, daddr64_t sectorCount) { struct hfs_inval_blk_no args; args.sectorStart = sectorStart; args.sectorCount = sectorCount; buf_iterate(vp, hfs_invalidate_block_numbers_callback, BUF_SCAN_DIRTY|BUF_SCAN_CLEAN, &args); } /* * Copy the contents of an extent to a new location. Also invalidates the * physical block number of any buffer cache block in the copied extent * (so that if the block is written, it will go through VNOP_BLOCKMAP to * determine the new physical block number). * * At this point, for regular files, we hold the truncate lock exclusive * and the cnode lock exclusive. */ static int hfs_copy_extent( struct hfsmount *hfsmp, struct vnode *vp, /* The file whose extent is being copied. */ u_int32_t oldStart, /* The start of the source extent. */ u_int32_t newStart, /* The start of the destination extent. */ u_int32_t blockCount, /* The number of allocation blocks to copy. */ vfs_context_t context) { int err = 0; size_t bufferSize; void *buffer = NULL; struct vfsioattr ioattr; buf_t bp = NULL; off_t resid; size_t ioSize; u_int32_t ioSizeSectors; /* Device sectors in this I/O */ daddr64_t srcSector, destSector; u_int32_t sectorsPerBlock = hfsmp->blockSize / hfsmp->hfs_logical_block_size; #if CONFIG_PROTECT int cpenabled = 0; #endif /* * Sanity check that we have locked the vnode of the file we're copying. * * But since hfs_systemfile_lock() doesn't actually take the lock on * the allocation file if a journal is active, ignore the check if the * file being copied is the allocation file. */ struct cnode *cp = VTOC(vp); if (cp != hfsmp->hfs_allocation_cp && cp->c_lockowner != current_thread()) panic("hfs_copy_extent: vp=%p (cp=%p) not owned?\n", vp, cp); #if CONFIG_PROTECT /* * Prepare the CP blob and get it ready for use, if necessary. * * Note that we specifically *exclude* system vnodes (catalog, bitmap, extents, EAs), * because they are implicitly protected via the media key on iOS. As such, they * must not be relocated except with the media key. So it is OK to not pass down * a special cpentry to the IOMedia/LwVM code for handling. */ if (!vnode_issystem (vp) && vnode_isreg(vp) && cp_fs_protected (hfsmp->hfs_mp)) { int cp_err = 0; /* * Ideally, the file whose extents we are about to manipulate is using the * newer offset-based IVs so that we can manipulate it regardless of the * current lock state. However, we must maintain support for older-style * EAs. * * For the older EA case, the IV was tied to the device LBA for file content. * This means that encrypted data cannot be moved from one location to another * in the filesystem without garbling the IV data. As a result, we need to * access the file's plaintext because we cannot do our AES-symmetry trick * here. This requires that we attempt a key-unwrap here (via cp_handle_relocate) * to make forward progress. If the keys are unavailable then we will * simply stop the resize in its tracks here since we cannot move * this extent at this time. */ if ((cp->c_cpentry->cp_flags & CP_OFF_IV_ENABLED) == 0) { cp_err = cp_handle_relocate(cp, hfsmp); } if (cp_err) { printf ("hfs_copy_extent: cp_handle_relocate failed (%d) \n", cp_err); return cp_err; } cpenabled = 1; } #endif /* * Determine the I/O size to use * * NOTE: Many external drives will result in an ioSize of 128KB. * TODO: Should we use a larger buffer, doing several consecutive * reads, then several consecutive writes? */ vfs_ioattr(hfsmp->hfs_mp, &ioattr); bufferSize = MIN(ioattr.io_maxreadcnt, ioattr.io_maxwritecnt); if (kmem_alloc(kernel_map, (vm_offset_t*) &buffer, bufferSize)) return ENOMEM; /* Get a buffer for doing the I/O */ bp = buf_alloc(hfsmp->hfs_devvp); buf_setdataptr(bp, (uintptr_t)buffer); resid = (off_t) blockCount * (off_t) hfsmp->blockSize; srcSector = (daddr64_t) oldStart * hfsmp->blockSize / hfsmp->hfs_logical_block_size; destSector = (daddr64_t) newStart * hfsmp->blockSize / hfsmp->hfs_logical_block_size; while (resid > 0) { ioSize = MIN(bufferSize, (size_t) resid); ioSizeSectors = ioSize / hfsmp->hfs_logical_block_size; /* Prepare the buffer for reading */ buf_reset(bp, B_READ); buf_setsize(bp, ioSize); buf_setcount(bp, ioSize); buf_setblkno(bp, srcSector); buf_setlblkno(bp, srcSector); /* * Note that because this is an I/O to the device vp * it is correct to have lblkno and blkno both point to the * start sector being read from. If it were being issued against the * underlying file then that would be different. */ /* Attach the new CP blob to the buffer if needed */ #if CONFIG_PROTECT if (cpenabled) { if (cp->c_cpentry->cp_flags & CP_OFF_IV_ENABLED) { /* attach the RELOCATION_INFLIGHT flag for the underlying call to VNOP_STRATEGY */ cp->c_cpentry->cp_flags |= CP_RELOCATION_INFLIGHT; buf_setcpaddr(bp, hfsmp->hfs_resize_cpentry); } else { /* * Use the cnode's cp key. This file is tied to the * LBAs of the physical blocks that it occupies. */ buf_setcpaddr (bp, cp->c_cpentry); } /* Initialize the content protection file offset to start at 0 */ buf_setcpoff (bp, 0); } #endif /* Do the read */ err = VNOP_STRATEGY(bp); if (!err) err = buf_biowait(bp); if (err) { #if CONFIG_PROTECT /* Turn the flag off in error cases. */ if (cpenabled) { cp->c_cpentry->cp_flags &= ~CP_RELOCATION_INFLIGHT; } #endif printf("hfs_copy_extent: Error %d from VNOP_STRATEGY (read)\n", err); break; } /* Prepare the buffer for writing */ buf_reset(bp, B_WRITE); buf_setsize(bp, ioSize); buf_setcount(bp, ioSize); buf_setblkno(bp, destSector); buf_setlblkno(bp, destSector); if (vnode_issystem(vp) && journal_uses_fua(hfsmp->jnl)) buf_markfua(bp); #if CONFIG_PROTECT /* Attach the CP to the buffer if needed */ if (cpenabled) { if (cp->c_cpentry->cp_flags & CP_OFF_IV_ENABLED) { buf_setcpaddr(bp, hfsmp->hfs_resize_cpentry); } else { /* * Use the cnode's CP key. This file is still tied * to the LBAs of the physical blocks that it occupies. */ buf_setcpaddr (bp, cp->c_cpentry); } /* * The last STRATEGY call may have updated the cp file offset behind our * back, so we cannot trust it. Re-initialize the content protection * file offset back to 0 before initiating the write portion of this I/O. */ buf_setcpoff (bp, 0); } #endif /* Do the write */ vnode_startwrite(hfsmp->hfs_devvp); err = VNOP_STRATEGY(bp); if (!err) { err = buf_biowait(bp); } #if CONFIG_PROTECT /* Turn the flag off regardless once the strategy call finishes. */ if (cpenabled) { cp->c_cpentry->cp_flags &= ~CP_RELOCATION_INFLIGHT; } #endif if (err) { printf("hfs_copy_extent: Error %d from VNOP_STRATEGY (write)\n", err); break; } resid -= ioSize; srcSector += ioSizeSectors; destSector += ioSizeSectors; } if (bp) buf_free(bp); if (buffer) kmem_free(kernel_map, (vm_offset_t)buffer, bufferSize); /* Make sure all writes have been flushed to disk. */ if (vnode_issystem(vp) && !journal_uses_fua(hfsmp->jnl)) { err = VNOP_IOCTL(hfsmp->hfs_devvp, DKIOCSYNCHRONIZECACHE, NULL, FWRITE, context); if (err) { printf("hfs_copy_extent: DKIOCSYNCHRONIZECACHE failed (%d)\n", err); err = 0; /* Don't fail the copy. */ } } if (!err) hfs_invalidate_sectors(vp, (daddr64_t)oldStart*sectorsPerBlock, (daddr64_t)blockCount*sectorsPerBlock); return err; } /* Structure to store state of reclaiming extents from a * given file. hfs_reclaim_file()/hfs_reclaim_xattr() * initializes the values in this structure which are then * used by code that reclaims and splits the extents. */ struct hfs_reclaim_extent_info { struct vnode *vp; u_int32_t fileID; u_int8_t forkType; u_int8_t is_dirlink; /* Extent belongs to directory hard link */ u_int8_t is_sysfile; /* Extent belongs to system file */ u_int8_t is_xattr; /* Extent belongs to extent-based xattr */ u_int8_t extent_index; int lockflags; /* Locks that reclaim and split code should grab before modifying the extent record */ u_int32_t blocks_relocated; /* Total blocks relocated for this file till now */ u_int32_t recStartBlock; /* File allocation block number (FABN) for current extent record */ u_int32_t cur_blockCount; /* Number of allocation blocks that have been checked for reclaim */ struct filefork *catalog_fp; /* If non-NULL, extent is from catalog record */ union record { HFSPlusExtentRecord overflow;/* Extent record from overflow extents btree */ HFSPlusAttrRecord xattr; /* Attribute record for large EAs */ } record; HFSPlusExtentDescriptor *extents; /* Pointer to current extent record being processed. * For catalog extent record, points to the correct * extent information in filefork. For overflow extent * record, or xattr record, points to extent record * in the structure above */ struct cat_desc *dirlink_desc; struct cat_attr *dirlink_attr; struct filefork *dirlink_fork; /* For directory hard links, fp points actually to this */ struct BTreeIterator *iterator; /* Shared read/write iterator, hfs_reclaim_file/xattr() * use it for reading and hfs_reclaim_extent()/hfs_split_extent() * use it for writing updated extent record */ struct FSBufferDescriptor btdata; /* Shared btdata for reading/writing extent record, same as iterator above */ u_int16_t recordlen; int overflow_count; /* For debugging, counter for overflow extent record */ FCB *fcb; /* Pointer to the current btree being traversed */ }; /* * Split the current extent into two extents, with first extent * to contain given number of allocation blocks. Splitting of * extent creates one new extent entry which can result in * shifting of many entries through all the extent records of a * file, and/or creating a new extent record in the overflow * extent btree. * * Example: * The diagram below represents two consecutive extent records, * for simplicity, lets call them record X and X+1 respectively. * Interesting extent entries have been denoted by letters. * If the letter is unchanged before and after split, it means * that the extent entry was not modified during the split. * A '.' means that the entry remains unchanged after the split * and is not relevant for our example. A '0' means that the * extent entry is empty. * * If there isn't sufficient contiguous free space to relocate * an extent (extent "C" below), we will have to break the one * extent into multiple smaller extents, and relocate each of * the smaller extents individually. The way we do this is by * finding the largest contiguous free space that is currently * available (N allocation blocks), and then convert extent "C" * into two extents, C1 and C2, that occupy exactly the same * allocation blocks as extent C. Extent C1 is the first * N allocation blocks of extent C, and extent C2 is the remainder * of extent C. Then we can relocate extent C1 since we know * we have enough contiguous free space to relocate it in its * entirety. We then repeat the process starting with extent C2. * * In record X, only the entries following entry C are shifted, and * the original entry C is replaced with two entries C1 and C2 which * are actually two extent entries for contiguous allocation blocks. * * Note that the entry E from record X is shifted into record X+1 as * the new first entry. Since the first entry of record X+1 is updated, * the FABN will also get updated with the blockCount of entry E. * This also results in shifting of all extent entries in record X+1. * Note that the number of empty entries after the split has been * changed from 3 to 2. * * Before: * record X record X+1 * ---------------------===--------- --------------------------------- * | A | . | . | . | B | C | D | E | | F | . | . | . | G | 0 | 0 | 0 | * ---------------------===--------- --------------------------------- * * After: * ---------------------=======----- --------------------------------- * | A | . | . | . | B | C1| C2| D | | E | F | . | . | . | G | 0 | 0 | * ---------------------=======----- --------------------------------- * * C1.startBlock = C.startBlock * C1.blockCount = N * * C2.startBlock = C.startBlock + N * C2.blockCount = C.blockCount - N * * FABN = old FABN - E.blockCount * * Inputs: * extent_info - This is the structure that contains state about * the current file, extent, and extent record that * is being relocated. This structure is shared * among code that traverses through all the extents * of the file, code that relocates extents, and * code that splits the extent. * newBlockCount - The blockCount of the extent to be split after * successfully split operation. * Output: * Zero on success, non-zero on failure. */ static int hfs_split_extent(struct hfs_reclaim_extent_info *extent_info, uint32_t newBlockCount) { int error = 0; int index = extent_info->extent_index; int i; HFSPlusExtentDescriptor shift_extent; /* Extent entry that should be shifted into next extent record */ HFSPlusExtentDescriptor last_extent; HFSPlusExtentDescriptor *extents; /* Pointer to current extent record being manipulated */ HFSPlusExtentRecord *extents_rec = NULL; HFSPlusExtentKey *extents_key = NULL; HFSPlusAttrRecord *xattr_rec = NULL; HFSPlusAttrKey *xattr_key = NULL; struct BTreeIterator iterator; struct FSBufferDescriptor btdata; uint16_t reclen; uint32_t read_recStartBlock; /* Starting allocation block number to read old extent record */ uint32_t write_recStartBlock; /* Starting allocation block number to insert newly updated extent record */ Boolean create_record = false; Boolean is_xattr; struct cnode *cp; is_xattr = extent_info->is_xattr; extents = extent_info->extents; cp = VTOC(extent_info->vp); if (newBlockCount == 0) { if (hfs_resize_debug) { printf ("hfs_split_extent: No splitting required for newBlockCount=0\n"); } return error; } if (hfs_resize_debug) { printf ("hfs_split_extent: Split record:%u recStartBlock=%u %u:(%u,%u) for %u blocks\n", extent_info->overflow_count, extent_info->recStartBlock, index, extents[index].startBlock, extents[index].blockCount, newBlockCount); } /* Extents overflow btree can not have more than 8 extents. * No split allowed if the 8th extent is already used. */ if ((extent_info->fileID == kHFSExtentsFileID) && (extents[kHFSPlusExtentDensity - 1].blockCount != 0)) { printf ("hfs_split_extent: Maximum 8 extents allowed for extents overflow btree, cannot split further.\n"); error = ENOSPC; goto out; } /* Determine the starting allocation block number for the following * overflow extent record, if any, before the current record * gets modified. */ read_recStartBlock = extent_info->recStartBlock; for (i = 0; i < kHFSPlusExtentDensity; i++) { if (extents[i].blockCount == 0) { break; } read_recStartBlock += extents[i].blockCount; } /* Shift and split */ if (index == kHFSPlusExtentDensity-1) { /* The new extent created after split will go into following overflow extent record */ shift_extent.startBlock = extents[index].startBlock + newBlockCount; shift_extent.blockCount = extents[index].blockCount - newBlockCount; /* Last extent in the record will be split, so nothing to shift */ } else { /* Splitting of extents can result in at most of one * extent entry to be shifted into following overflow extent * record. So, store the last extent entry for later. */ shift_extent = extents[kHFSPlusExtentDensity-1]; if ((hfs_resize_debug) && (shift_extent.blockCount != 0)) { printf ("hfs_split_extent: Save 7:(%u,%u) to shift into overflow record\n", shift_extent.startBlock, shift_extent.blockCount); } /* Start shifting extent information from the end of the extent * record to the index where we want to insert the new extent. * Note that kHFSPlusExtentDensity-1 is already saved above, and * does not need to be shifted. The extent entry that is being * split does not get shifted. */ for (i = kHFSPlusExtentDensity-2; i > index; i--) { if (hfs_resize_debug) { if (extents[i].blockCount) { printf ("hfs_split_extent: Shift %u:(%u,%u) to %u:(%u,%u)\n", i, extents[i].startBlock, extents[i].blockCount, i+1, extents[i].startBlock, extents[i].blockCount); } } extents[i+1] = extents[i]; } } if (index == kHFSPlusExtentDensity-1) { /* The second half of the extent being split will be the overflow * entry that will go into following overflow extent record. The * value has been stored in 'shift_extent' above, so there is * nothing to be done here. */ } else { /* Update the values in the second half of the extent being split * before updating the first half of the split. Note that the * extent to split or first half of the split is at index 'index' * and a new extent or second half of the split will be inserted at * 'index+1' or into following overflow extent record. */ extents[index+1].startBlock = extents[index].startBlock + newBlockCount; extents[index+1].blockCount = extents[index].blockCount - newBlockCount; } /* Update the extent being split, only the block count will change */ extents[index].blockCount = newBlockCount; if (hfs_resize_debug) { printf ("hfs_split_extent: Split %u:(%u,%u) and ", index, extents[index].startBlock, extents[index].blockCount); if (index != kHFSPlusExtentDensity-1) { printf ("%u:(%u,%u)\n", index+1, extents[index+1].startBlock, extents[index+1].blockCount); } else { printf ("overflow:(%u,%u)\n", shift_extent.startBlock, shift_extent.blockCount); } } /* Write out information about the newly split extent to the disk */ if (extent_info->catalog_fp) { /* (extent_info->catalog_fp != NULL) means the newly split * extent exists in the catalog record. This means that * the cnode was updated. Therefore, to write out the changes, * mark the cnode as modified. We cannot call hfs_update() * in this function because the caller hfs_reclaim_extent() * is holding the catalog lock currently. */ cp->c_flag |= C_MODIFIED; } else { /* The newly split extent is for large EAs or is in overflow * extent record, so update it directly in the btree using the * iterator information from the shared extent_info structure */ error = BTReplaceRecord(extent_info->fcb, extent_info->iterator, &(extent_info->btdata), extent_info->recordlen); if (error) { printf ("hfs_split_extent: fileID=%u BTReplaceRecord returned error=%d\n", extent_info->fileID, error); goto out; } } /* No extent entry to be shifted into another extent overflow record */ if (shift_extent.blockCount == 0) { if (hfs_resize_debug) { printf ("hfs_split_extent: No extent entry to be shifted into overflow records\n"); } error = 0; goto out; } /* The overflow extent entry has to be shifted into an extent * overflow record. This means that we might have to shift * extent entries from all subsequent overflow records by one. * We start iteration from the first record to the last record, * and shift the extent entry from one record to another. * We might have to create a new extent record for the last * extent entry for the file. */ /* Initialize iterator to search the next record */ bzero(&iterator, sizeof(iterator)); if (is_xattr) { /* Copy the key from the iterator that was used to update the modified attribute record. */ xattr_key = (HFSPlusAttrKey *)&(iterator.key); bcopy((HFSPlusAttrKey *)&(extent_info->iterator->key), xattr_key, sizeof(HFSPlusAttrKey)); /* Note: xattr_key->startBlock will be initialized later in the iteration loop */ MALLOC(xattr_rec, HFSPlusAttrRecord *, sizeof(HFSPlusAttrRecord), M_TEMP, M_WAITOK); if (xattr_rec == NULL) { error = ENOMEM; goto out; } btdata.bufferAddress = xattr_rec; btdata.itemSize = sizeof(HFSPlusAttrRecord); btdata.itemCount = 1; extents = xattr_rec->overflowExtents.extents; } else { /* Initialize the extent key for the current file */ extents_key = (HFSPlusExtentKey *) &(iterator.key); extents_key->keyLength = kHFSPlusExtentKeyMaximumLength; extents_key->forkType = extent_info->forkType; extents_key->fileID = extent_info->fileID; /* Note: extents_key->startBlock will be initialized later in the iteration loop */ MALLOC(extents_rec, HFSPlusExtentRecord *, sizeof(HFSPlusExtentRecord), M_TEMP, M_WAITOK); if (extents_rec == NULL) { error = ENOMEM; goto out; } btdata.bufferAddress = extents_rec; btdata.itemSize = sizeof(HFSPlusExtentRecord); btdata.itemCount = 1; extents = extents_rec[0]; } /* The overflow extent entry has to be shifted into an extent * overflow record. This means that we might have to shift * extent entries from all subsequent overflow records by one. * We start iteration from the first record to the last record, * examine one extent record in each iteration and shift one * extent entry from one record to another. We might have to * create a new extent record for the last extent entry for the * file. * * If shift_extent.blockCount is non-zero, it means that there is * an extent entry that needs to be shifted into the next * overflow extent record. We keep on going till there are no such * entries left to be shifted. This will also change the starting * allocation block number of the extent record which is part of * the key for the extent record in each iteration. Note that * because the extent record key is changing while we are searching, * the record can not be updated directly, instead it has to be * deleted and inserted again. */ while (shift_extent.blockCount) { if (hfs_resize_debug) { printf ("hfs_split_extent: Will shift (%u,%u) into overflow record with startBlock=%u\n", shift_extent.startBlock, shift_extent.blockCount, read_recStartBlock); } /* Search if there is any existing overflow extent record * that matches the current file and the logical start block * number. * * For this, the logical start block number in the key is * the value calculated based on the logical start block * number of the current extent record and the total number * of blocks existing in the current extent record. */ if (is_xattr) { xattr_key->startBlock = read_recStartBlock; } else { extents_key->startBlock = read_recStartBlock; } error = BTSearchRecord(extent_info->fcb, &iterator, &btdata, &reclen, &iterator); if (error) { if (error != btNotFound) { printf ("hfs_split_extent: fileID=%u startBlock=%u BTSearchRecord error=%d\n", extent_info->fileID, read_recStartBlock, error); goto out; } /* No matching record was found, so create a new extent record. * Note: Since no record was found, we can't rely on the * btree key in the iterator any longer. This will be initialized * later before we insert the record. */ create_record = true; } /* The extra extent entry from the previous record is being inserted * as the first entry in the current extent record. This will change * the file allocation block number (FABN) of the current extent * record, which is the startBlock value from the extent record key. * Since one extra entry is being inserted in the record, the new * FABN for the record will less than old FABN by the number of blocks * in the new extent entry being inserted at the start. We have to * do this before we update read_recStartBlock to point at the * startBlock of the following record. */ write_recStartBlock = read_recStartBlock - shift_extent.blockCount; if (hfs_resize_debug) { if (create_record) { printf ("hfs_split_extent: No records found for startBlock=%u, will create new with startBlock=%u\n", read_recStartBlock, write_recStartBlock); } } /* Now update the read_recStartBlock to account for total number * of blocks in this extent record. It will now point to the * starting allocation block number for the next extent record. */ for (i = 0; i < kHFSPlusExtentDensity; i++) { if (extents[i].blockCount == 0) { break; } read_recStartBlock += extents[i].blockCount; } if (create_record == true) { /* Initialize new record content with only one extent entry */ bzero(extents, sizeof(HFSPlusExtentRecord)); /* The new record will contain only one extent entry */ extents[0] = shift_extent; /* There are no more overflow extents to be shifted */ shift_extent.startBlock = shift_extent.blockCount = 0; if (is_xattr) { /* BTSearchRecord above returned btNotFound, * but since the attribute btree is never empty * if we are trying to insert new overflow * record for the xattrs, the extents_key will * contain correct data. So we don't need to * re-initialize it again like below. */ /* Initialize the new xattr record */ xattr_rec->recordType = kHFSPlusAttrExtents; xattr_rec->overflowExtents.reserved = 0; reclen = sizeof(HFSPlusAttrExtents); } else { /* BTSearchRecord above returned btNotFound, * which means that extents_key content might * not correspond to the record that we are * trying to create, especially when the extents * overflow btree is empty. So we reinitialize * the extents_key again always. */ extents_key->keyLength = kHFSPlusExtentKeyMaximumLength; extents_key->forkType = extent_info->forkType; extents_key->fileID = extent_info->fileID; /* Initialize the new extent record */ reclen = sizeof(HFSPlusExtentRecord); } } else { /* The overflow extent entry from previous record will be * the first entry in this extent record. If the last * extent entry in this record is valid, it will be shifted * into the following extent record as its first entry. So * save the last entry before shifting entries in current * record. */ last_extent = extents[kHFSPlusExtentDensity-1]; /* Shift all entries by one index towards the end */ for (i = kHFSPlusExtentDensity-2; i >= 0; i--) { extents[i+1] = extents[i]; } /* Overflow extent entry saved from previous record * is now the first entry in the current record. */ extents[0] = shift_extent; if (hfs_resize_debug) { printf ("hfs_split_extent: Shift overflow=(%u,%u) to record with updated startBlock=%u\n", shift_extent.startBlock, shift_extent.blockCount, write_recStartBlock); } /* The last entry from current record will be the * overflow entry which will be the first entry for * the following extent record. */ shift_extent = last_extent; /* Since the key->startBlock is being changed for this record, * it should be deleted and inserted with the new key. */ error = BTDeleteRecord(extent_info->fcb, &iterator); if (error) { printf ("hfs_split_extent: fileID=%u startBlock=%u BTDeleteRecord error=%d\n", extent_info->fileID, read_recStartBlock, error); goto out; } if (hfs_resize_debug) { printf ("hfs_split_extent: Deleted extent record with startBlock=%u\n", (is_xattr ? xattr_key->startBlock : extents_key->startBlock)); } } /* Insert the newly created or modified extent record */ bzero(&iterator.hint, sizeof(iterator.hint)); if (is_xattr) { xattr_key->startBlock = write_recStartBlock; } else { extents_key->startBlock = write_recStartBlock; } error = BTInsertRecord(extent_info->fcb, &iterator, &btdata, reclen); if (error) { printf ("hfs_split_extent: fileID=%u, startBlock=%u BTInsertRecord error=%d\n", extent_info->fileID, write_recStartBlock, error); goto out; } if (hfs_resize_debug) { printf ("hfs_split_extent: Inserted extent record with startBlock=%u\n", write_recStartBlock); } } out: /* * Extents overflow btree or attributes btree headers might have * been modified during the split/shift operation, so flush the * changes to the disk while we are inside journal transaction. * We should only be able to generate I/O that modifies the B-Tree * header nodes while we're in the middle of a journal transaction. * Otherwise it might result in panic during unmount. */ BTFlushPath(extent_info->fcb); if (extents_rec) { FREE (extents_rec, M_TEMP); } if (xattr_rec) { FREE (xattr_rec, M_TEMP); } return error; } /* * Relocate an extent if it lies beyond the expected end of volume. * * This function is called for every extent of the file being relocated. * It allocates space for relocation, copies the data, deallocates * the old extent, and update corresponding on-disk extent. If the function * does not find contiguous space to relocate an extent, it splits the * extent in smaller size to be able to relocate it out of the area of * disk being reclaimed. As an optimization, if an extent lies partially * in the area of the disk being reclaimed, it is split so that we only * have to relocate the area that was overlapping with the area of disk * being reclaimed. * * Note that every extent is relocated in its own transaction so that * they do not overwhelm the journal. This function handles the extent * record that exists in the catalog record, extent record from overflow * extents btree, and extents for large EAs. * * Inputs: * extent_info - This is the structure that contains state about * the current file, extent, and extent record that * is being relocated. This structure is shared * among code that traverses through all the extents * of the file, code that relocates extents, and * code that splits the extent. */ static int hfs_reclaim_extent(struct hfsmount *hfsmp, const u_long allocLimit, struct hfs_reclaim_extent_info *extent_info, vfs_context_t context) { int error = 0; int index; struct cnode *cp; u_int32_t oldStartBlock; u_int32_t oldBlockCount; u_int32_t newStartBlock; u_int32_t newBlockCount; u_int32_t roundedBlockCount; uint16_t node_size; uint32_t remainder_blocks; u_int32_t alloc_flags; int blocks_allocated = false; index = extent_info->extent_index; cp = VTOC(extent_info->vp); oldStartBlock = extent_info->extents[index].startBlock; oldBlockCount = extent_info->extents[index].blockCount; if (0 && hfs_resize_debug) { printf ("hfs_reclaim_extent: Examine record:%u recStartBlock=%u, %u:(%u,%u)\n", extent_info->overflow_count, extent_info->recStartBlock, index, oldStartBlock, oldBlockCount); } /* If the current extent lies completely within allocLimit, * it does not require any relocation. */ if ((oldStartBlock + oldBlockCount) <= allocLimit) { extent_info->cur_blockCount += oldBlockCount; return error; } /* Every extent should be relocated in its own transaction * to make sure that we don't overflow the journal buffer. */ error = hfs_start_transaction(hfsmp); if (error) { return error; } extent_info->lockflags = hfs_systemfile_lock(hfsmp, extent_info->lockflags, HFS_EXCLUSIVE_LOCK); /* Check if the extent lies partially in the area to reclaim, * i.e. it starts before allocLimit and ends beyond allocLimit. * We have already skipped extents that lie completely within * allocLimit in the check above, so we only check for the * startBlock. If it lies partially, split it so that we * only relocate part of the extent. */ if (oldStartBlock < allocLimit) { newBlockCount = allocLimit - oldStartBlock; if (hfs_resize_debug) { int idx = extent_info->extent_index; printf ("hfs_reclaim_extent: Split straddling extent %u:(%u,%u) for %u blocks\n", idx, extent_info->extents[idx].startBlock, extent_info->extents[idx].blockCount, newBlockCount); } /* If the extent belongs to a btree, check and trim * it to be multiple of the node size. */ if (extent_info->is_sysfile) { node_size = get_btree_nodesize(extent_info->vp); /* If the btree node size is less than the block size, * splitting this extent will not split a node across * different extents. So we only check and trim if * node size is more than the allocation block size. */ if (node_size > hfsmp->blockSize) { remainder_blocks = newBlockCount % (node_size / hfsmp->blockSize); if (remainder_blocks) { newBlockCount -= remainder_blocks; if (hfs_resize_debug) { printf ("hfs_reclaim_extent: Round-down newBlockCount to be multiple of nodeSize, node_allocblks=%u, old=%u, new=%u\n", node_size/hfsmp->blockSize, newBlockCount + remainder_blocks, newBlockCount); } } } /* The newBlockCount is zero because of rounding-down so that * btree nodes are not split across extents. Therefore this * straddling extent across resize-boundary does not require * splitting. Skip over to relocating of complete extent. */ if (newBlockCount == 0) { if (hfs_resize_debug) { printf ("hfs_reclaim_extent: After round-down newBlockCount=0, skip split, relocate full extent\n"); } goto relocate_full_extent; } } /* Split the extents into two parts --- the first extent lies * completely within allocLimit and therefore does not require * relocation. The second extent will require relocation which * will be handled when the caller calls this function again * for the next extent. */ error = hfs_split_extent(extent_info, newBlockCount); if (error == 0) { /* Split success, no relocation required */ goto out; } /* Split failed, so try to relocate entire extent */ if (hfs_resize_debug) { int idx = extent_info->extent_index; printf ("hfs_reclaim_extent: Split straddling extent %u:(%u,%u) for %u blocks failed, relocate full extent\n", idx, extent_info->extents[idx].startBlock, extent_info->extents[idx].blockCount, newBlockCount); } } relocate_full_extent: /* At this point, the current extent requires relocation. * We will try to allocate space equal to the size of the extent * being relocated first to try to relocate it without splitting. * If the allocation fails, we will try to allocate contiguous * blocks out of metadata zone. If that allocation also fails, * then we will take a whatever contiguous block run is returned * by the allocation, split the extent into two parts, and then * relocate the first splitted extent. */ alloc_flags = HFS_ALLOC_FORCECONTIG | HFS_ALLOC_SKIPFREEBLKS; if (extent_info->is_sysfile) { alloc_flags |= HFS_ALLOC_METAZONE; } error = BlockAllocate(hfsmp, 1, oldBlockCount, oldBlockCount, alloc_flags, &newStartBlock, &newBlockCount); if ((extent_info->is_sysfile == false) && ((error == dskFulErr) || (error == ENOSPC))) { /* For non-system files, try reallocating space in metadata zone */ alloc_flags |= HFS_ALLOC_METAZONE; error = BlockAllocate(hfsmp, 1, oldBlockCount, oldBlockCount, alloc_flags, &newStartBlock, &newBlockCount); } if ((error == dskFulErr) || (error == ENOSPC)) { /* * We did not find desired contiguous space for this * extent, when we asked for it, including the metazone allocations. * At this point we are not worrying about getting contiguity anymore. * * HOWEVER, if we now allow blocks to be used which were recently * de-allocated, we may find a contiguous range (though this seems * unlikely). As a result, assume that we will have to split the * current extent into two pieces, but if we are able to satisfy * the request with a single extent, detect that as well. */ alloc_flags &= ~HFS_ALLOC_FORCECONTIG; alloc_flags |= HFS_ALLOC_FLUSHTXN; error = BlockAllocate(hfsmp, 1, oldBlockCount, oldBlockCount, alloc_flags, &newStartBlock, &newBlockCount); if (error) { printf ("hfs_reclaim_extent: fileID=%u start=%u, %u:(%u,%u) BlockAllocate error=%d\n", extent_info->fileID, extent_info->recStartBlock, index, oldStartBlock, oldBlockCount, error); goto out; } /* * Allowing recently deleted extents may now allow us to find * a single contiguous extent in the amount & size desired. If so, * do NOT split this extent into two pieces. This is technically a * check for "< oldBlockCount", but we use != to highlight the point * that the special case is when they're equal. The allocator should * never vend back more blocks than were requested. */ if (newBlockCount != oldBlockCount) { blocks_allocated = true; /* The number of blocks allocated is less than the requested * number of blocks. For btree extents, check and trim the * extent to be multiple of the node size. */ if (extent_info->is_sysfile) { node_size = get_btree_nodesize(extent_info->vp); if (node_size > hfsmp->blockSize) { remainder_blocks = newBlockCount % (node_size / hfsmp->blockSize); if (remainder_blocks) { roundedBlockCount = newBlockCount - remainder_blocks; /* Free tail-end blocks of the newly allocated extent */ BlockDeallocate(hfsmp, newStartBlock + roundedBlockCount, newBlockCount - roundedBlockCount, HFS_ALLOC_SKIPFREEBLKS); newBlockCount = roundedBlockCount; if (hfs_resize_debug) { printf ("hfs_reclaim_extent: Fixing extent block count, node_blks=%u, old=%u, new=%u\n", node_size/hfsmp->blockSize, newBlockCount + remainder_blocks, newBlockCount); } if (newBlockCount == 0) { printf ("hfs_reclaim_extent: Not enough contiguous blocks available to relocate fileID=%d\n", extent_info->fileID); error = ENOSPC; goto out; } } } } /* The number of blocks allocated is less than the number of * blocks requested, so split this extent --- the first extent * will be relocated as part of this function call and the caller * will handle relocating the second extent by calling this * function again for the second extent. */ error = hfs_split_extent(extent_info, newBlockCount); if (error) { printf ("hfs_reclaim_extent: fileID=%u start=%u, %u:(%u,%u) split error=%d\n", extent_info->fileID, extent_info->recStartBlock, index, oldStartBlock, oldBlockCount, error); goto out; } oldBlockCount = newBlockCount; } /* end oldBlockCount != newBlockCount */ } /* end allocation request for any available free space */ if (error) { printf ("hfs_reclaim_extent: fileID=%u start=%u, %u:(%u,%u) contig BlockAllocate error=%d\n", extent_info->fileID, extent_info->recStartBlock, index, oldStartBlock, oldBlockCount, error); goto out; } blocks_allocated = true; /* Copy data from old location to new location */ error = hfs_copy_extent(hfsmp, extent_info->vp, oldStartBlock, newStartBlock, newBlockCount, context); if (error) { printf ("hfs_reclaim_extent: fileID=%u start=%u, %u:(%u,%u)=>(%u,%u) hfs_copy_extent error=%d\n", extent_info->fileID, extent_info->recStartBlock, index, oldStartBlock, oldBlockCount, newStartBlock, newBlockCount, error); goto out; } /* Update the extent record with the new start block information */ extent_info->extents[index].startBlock = newStartBlock; /* Sync the content back to the disk */ if (extent_info->catalog_fp) { /* Update the extents in catalog record */ if (extent_info->is_dirlink) { error = cat_update_dirlink(hfsmp, extent_info->forkType, extent_info->dirlink_desc, extent_info->dirlink_attr, &(extent_info->dirlink_fork->ff_data)); } else { cp->c_flag |= C_MODIFIED; /* If this is a system file, sync volume headers on disk */ if (extent_info->is_sysfile) { error = hfs_flushvolumeheader(hfsmp, MNT_WAIT, HFS_ALTFLUSH); } } } else { /* Replace record for extents overflow or extents-based xattrs */ error = BTReplaceRecord(extent_info->fcb, extent_info->iterator, &(extent_info->btdata), extent_info->recordlen); } if (error) { printf ("hfs_reclaim_extent: fileID=%u, update record error=%u\n", extent_info->fileID, error); goto out; } /* Deallocate the old extent */ error = BlockDeallocate(hfsmp, oldStartBlock, oldBlockCount, HFS_ALLOC_SKIPFREEBLKS); if (error) { printf ("hfs_reclaim_extent: fileID=%u start=%u, %u:(%u,%u) BlockDeallocate error=%d\n", extent_info->fileID, extent_info->recStartBlock, index, oldStartBlock, oldBlockCount, error); goto out; } extent_info->blocks_relocated += newBlockCount; if (hfs_resize_debug) { printf ("hfs_reclaim_extent: Relocated record:%u %u:(%u,%u) to (%u,%u)\n", extent_info->overflow_count, index, oldStartBlock, oldBlockCount, newStartBlock, newBlockCount); } out: if (error != 0) { if (blocks_allocated == true) { BlockDeallocate(hfsmp, newStartBlock, newBlockCount, HFS_ALLOC_SKIPFREEBLKS); } } else { /* On success, increment the total allocation blocks processed */ extent_info->cur_blockCount += newBlockCount; } hfs_systemfile_unlock(hfsmp, extent_info->lockflags); /* For a non-system file, if an extent entry from catalog record * was modified, sync the in-memory changes to the catalog record * on disk before ending the transaction. */ if ((extent_info->catalog_fp) && (extent_info->is_sysfile == false)) { (void) hfs_update(extent_info->vp, MNT_WAIT); } hfs_end_transaction(hfsmp); return error; } /* Report intermediate progress during volume resize */ static void hfs_truncatefs_progress(struct hfsmount *hfsmp) { u_int32_t cur_progress = 0; hfs_resize_progress(hfsmp, &cur_progress); if (cur_progress > (hfsmp->hfs_resize_progress + 9)) { printf("hfs_truncatefs: %d%% done...\n", cur_progress); hfsmp->hfs_resize_progress = cur_progress; } return; } /* * Reclaim space at the end of a volume for given file and forktype. * * This routine attempts to move any extent which contains allocation blocks * at or after "allocLimit." A separate transaction is used for every extent * that needs to be moved. If there is not contiguous space available for * moving an extent, it can be split into smaller extents. The contents of * any moved extents are read and written via the volume's device vnode -- * NOT via "vp." During the move, moved blocks which are part of a transaction * have their physical block numbers invalidated so they will eventually be * written to their new locations. * * This function is also called for directory hard links. Directory hard links * are regular files with no data fork and resource fork that contains alias * information for backward compatibility with pre-Leopard systems. However * non-Mac OS X implementation can add/modify data fork or resource fork * information to directory hard links, so we check, and if required, relocate * both data fork and resource fork. * * Inputs: * hfsmp The volume being resized. * vp The vnode for the system file. * fileID ID of the catalog record that needs to be relocated * forktype The type of fork that needs relocated, * kHFSResourceForkType for resource fork, * kHFSDataForkType for data fork * allocLimit Allocation limit for the new volume size, * do not use this block or beyond. All extents * that use this block or any blocks beyond this limit * will be relocated. * * Side Effects: * hfsmp->hfs_resize_blocksmoved is incremented by the number of allocation * blocks that were relocated. */ static int hfs_reclaim_file(struct hfsmount *hfsmp, struct vnode *vp, u_int32_t fileID, u_int8_t forktype, u_long allocLimit, vfs_context_t context) { int error = 0; struct hfs_reclaim_extent_info *extent_info; int i; int lockflags = 0; struct cnode *cp; struct filefork *fp; int took_truncate_lock = false; int release_desc = false; HFSPlusExtentKey *key; /* If there is no vnode for this file, then there's nothing to do. */ if (vp == NULL) { return 0; } cp = VTOC(vp); if (hfs_resize_debug) { const char *filename = (const char *) cp->c_desc.cd_nameptr; int namelen = cp->c_desc.cd_namelen; if (filename == NULL) { filename = ""; namelen = 0; } printf("hfs_reclaim_file: reclaiming '%.*s'\n", namelen, filename); } MALLOC(extent_info, struct hfs_reclaim_extent_info *, sizeof(struct hfs_reclaim_extent_info), M_TEMP, M_WAITOK); if (extent_info == NULL) { return ENOMEM; } bzero(extent_info, sizeof(struct hfs_reclaim_extent_info)); extent_info->vp = vp; extent_info->fileID = fileID; extent_info->forkType = forktype; extent_info->is_sysfile = vnode_issystem(vp); if (vnode_isdir(vp) && (cp->c_flag & C_HARDLINK)) { extent_info->is_dirlink = true; } /* We always need allocation bitmap and extent btree lock */ lockflags = SFL_BITMAP | SFL_EXTENTS; if ((fileID == kHFSCatalogFileID) || (extent_info->is_dirlink == true)) { lockflags |= SFL_CATALOG; } else if (fileID == kHFSAttributesFileID) { lockflags |= SFL_ATTRIBUTE; } else if (fileID == kHFSStartupFileID) { lockflags |= SFL_STARTUP; } extent_info->lockflags = lockflags; extent_info->fcb = VTOF(hfsmp->hfs_extents_vp); /* Flush data associated with current file on disk. * * If the current vnode is directory hard link, no flushing of * journal or vnode is required. The current kernel does not * modify data/resource fork of directory hard links, so nothing * will be in the cache. If a directory hard link is newly created, * the resource fork data is written directly using devvp and * the code that actually relocates data (hfs_copy_extent()) also * uses devvp for its I/O --- so they will see a consistent copy. */ if (extent_info->is_sysfile) { /* If the current vnode is system vnode, flush journal * to make sure that all data is written to the disk. */ error = hfs_journal_flush(hfsmp, TRUE); if (error) { printf ("hfs_reclaim_file: journal_flush returned %d\n", error); goto out; } } else if (extent_info->is_dirlink == false) { /* Flush all blocks associated with this regular file vnode. * Normally there should not be buffer cache blocks for regular * files, but for objects like symlinks, we can have buffer cache * blocks associated with the vnode. Therefore we call * buf_flushdirtyblks() also. */ buf_flushdirtyblks(vp, 0, BUF_SKIP_LOCKED, "hfs_reclaim_file"); hfs_unlock(cp); hfs_lock_truncate(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT); took_truncate_lock = true; (void) cluster_push(vp, 0); error = hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS); if (error) { goto out; } /* If the file no longer exists, nothing left to do */ if (cp->c_flag & C_NOEXISTS) { error = 0; goto out; } /* Wait for any in-progress writes to this vnode to complete, so that we'll * be copying consistent bits. (Otherwise, it's possible that an async * write will complete to the old extent after we read from it. That * could lead to corruption.) */ error = vnode_waitforwrites(vp, 0, 0, 0, "hfs_reclaim_file"); if (error) { goto out; } } if (hfs_resize_debug) { printf("hfs_reclaim_file: === Start reclaiming %sfork for %sid=%u ===\n", (forktype ? "rsrc" : "data"), (extent_info->is_dirlink ? "dirlink" : "file"), fileID); } if (extent_info->is_dirlink) { MALLOC(extent_info->dirlink_desc, struct cat_desc *, sizeof(struct cat_desc), M_TEMP, M_WAITOK); MALLOC(extent_info->dirlink_attr, struct cat_attr *, sizeof(struct cat_attr), M_TEMP, M_WAITOK); MALLOC(extent_info->dirlink_fork, struct filefork *, sizeof(struct filefork), M_TEMP, M_WAITOK); if ((extent_info->dirlink_desc == NULL) || (extent_info->dirlink_attr == NULL) || (extent_info->dirlink_fork == NULL)) { error = ENOMEM; goto out; } /* Lookup catalog record for directory hard link and * create a fake filefork for the value looked up from * the disk. */ fp = extent_info->dirlink_fork; bzero(extent_info->dirlink_fork, sizeof(struct filefork)); extent_info->dirlink_fork->ff_cp = cp; lockflags = hfs_systemfile_lock(hfsmp, lockflags, HFS_EXCLUSIVE_LOCK); error = cat_lookup_dirlink(hfsmp, fileID, forktype, extent_info->dirlink_desc, extent_info->dirlink_attr, &(extent_info->dirlink_fork->ff_data)); hfs_systemfile_unlock(hfsmp, lockflags); if (error) { printf ("hfs_reclaim_file: cat_lookup_dirlink for fileID=%u returned error=%u\n", fileID, error); goto out; } release_desc = true; } else { fp = VTOF(vp); } extent_info->catalog_fp = fp; extent_info->recStartBlock = 0; extent_info->extents = extent_info->catalog_fp->ff_extents; /* Relocate extents from the catalog record */ for (i = 0; i < kHFSPlusExtentDensity; ++i) { if (fp->ff_extents[i].blockCount == 0) { break; } extent_info->extent_index = i; error = hfs_reclaim_extent(hfsmp, allocLimit, extent_info, context); if (error) { printf ("hfs_reclaim_file: fileID=%u #%d %u:(%u,%u) hfs_reclaim_extent error=%d\n", fileID, extent_info->overflow_count, i, fp->ff_extents[i].startBlock, fp->ff_extents[i].blockCount, error); goto out; } } /* If the number of allocation blocks processed for reclaiming * are less than total number of blocks for the file, continuing * working on overflow extents record. */ if (fp->ff_blocks <= extent_info->cur_blockCount) { if (0 && hfs_resize_debug) { printf ("hfs_reclaim_file: Nothing more to relocate, offset=%d, ff_blocks=%u, cur_blockCount=%u\n", i, fp->ff_blocks, extent_info->cur_blockCount); } goto out; } if (hfs_resize_debug) { printf ("hfs_reclaim_file: Will check overflow records, offset=%d, ff_blocks=%u, cur_blockCount=%u\n", i, fp->ff_blocks, extent_info->cur_blockCount); } MALLOC(extent_info->iterator, struct BTreeIterator *, sizeof(struct BTreeIterator), M_TEMP, M_WAITOK); if (extent_info->iterator == NULL) { error = ENOMEM; goto out; } bzero(extent_info->iterator, sizeof(struct BTreeIterator)); key = (HFSPlusExtentKey *) &(extent_info->iterator->key); key->keyLength = kHFSPlusExtentKeyMaximumLength; key->forkType = forktype; key->fileID = fileID; key->startBlock = extent_info->cur_blockCount; extent_info->btdata.bufferAddress = extent_info->record.overflow; extent_info->btdata.itemSize = sizeof(HFSPlusExtentRecord); extent_info->btdata.itemCount = 1; extent_info->catalog_fp = NULL; /* Search the first overflow extent with expected startBlock as 'cur_blockCount' */ lockflags = hfs_systemfile_lock(hfsmp, lockflags, HFS_EXCLUSIVE_LOCK); error = BTSearchRecord(extent_info->fcb, extent_info->iterator, &(extent_info->btdata), &(extent_info->recordlen), extent_info->iterator); hfs_systemfile_unlock(hfsmp, lockflags); while (error == 0) { extent_info->overflow_count++; extent_info->recStartBlock = key->startBlock; extent_info->extents = extent_info->record.overflow; for (i = 0; i < kHFSPlusExtentDensity; i++) { if (extent_info->record.overflow[i].blockCount == 0) { goto out; } extent_info->extent_index = i; error = hfs_reclaim_extent(hfsmp, allocLimit, extent_info, context); if (error) { printf ("hfs_reclaim_file: fileID=%u #%d %u:(%u,%u) hfs_reclaim_extent error=%d\n", fileID, extent_info->overflow_count, i, extent_info->record.overflow[i].startBlock, extent_info->record.overflow[i].blockCount, error); goto out; } } /* Look for more overflow records */ lockflags = hfs_systemfile_lock(hfsmp, lockflags, HFS_EXCLUSIVE_LOCK); error = BTIterateRecord(extent_info->fcb, kBTreeNextRecord, extent_info->iterator, &(extent_info->btdata), &(extent_info->recordlen)); hfs_systemfile_unlock(hfsmp, lockflags); if (error) { break; } /* Stop when we encounter a different file or fork. */ if ((key->fileID != fileID) || (key->forkType != forktype)) { break; } } if (error == fsBTRecordNotFoundErr || error == fsBTEndOfIterationErr) { error = 0; } out: /* If any blocks were relocated, account them and report progress */ if (extent_info->blocks_relocated) { hfsmp->hfs_resize_blocksmoved += extent_info->blocks_relocated; hfs_truncatefs_progress(hfsmp); if (fileID < kHFSFirstUserCatalogNodeID) { printf ("hfs_reclaim_file: Relocated %u blocks from fileID=%u on \"%s\"\n", extent_info->blocks_relocated, fileID, hfsmp->vcbVN); } } if (extent_info->iterator) { FREE(extent_info->iterator, M_TEMP); } if (release_desc == true) { cat_releasedesc(extent_info->dirlink_desc); } if (extent_info->dirlink_desc) { FREE(extent_info->dirlink_desc, M_TEMP); } if (extent_info->dirlink_attr) { FREE(extent_info->dirlink_attr, M_TEMP); } if (extent_info->dirlink_fork) { FREE(extent_info->dirlink_fork, M_TEMP); } if ((extent_info->blocks_relocated != 0) && (extent_info->is_sysfile == false)) { (void) hfs_update(vp, MNT_WAIT); } if (took_truncate_lock) { hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT); } if (extent_info) { FREE(extent_info, M_TEMP); } if (hfs_resize_debug) { printf("hfs_reclaim_file: === Finished relocating %sfork for fileid=%u (error=%d) ===\n", (forktype ? "rsrc" : "data"), fileID, error); } return error; } /* * This journal_relocate callback updates the journal info block to point * at the new journal location. This write must NOT be done using the * transaction. We must write the block immediately. We must also force * it to get to the media so that the new journal location will be seen by * the replay code before we can safely let journaled blocks be written * to their normal locations. * * The tests for journal_uses_fua below are mildly hacky. Since the journal * and the file system are both on the same device, I'm leveraging what * the journal has decided about FUA. */ struct hfs_journal_relocate_args { struct hfsmount *hfsmp; vfs_context_t context; u_int32_t newStartBlock; u_int32_t newBlockCount; }; static errno_t hfs_journal_relocate_callback(void *_args) { int error; struct hfs_journal_relocate_args *args = _args; struct hfsmount *hfsmp = args->hfsmp; buf_t bp; JournalInfoBlock *jibp; error = buf_meta_bread(hfsmp->hfs_devvp, hfsmp->vcbJinfoBlock * (hfsmp->blockSize/hfsmp->hfs_logical_block_size), hfsmp->blockSize, vfs_context_ucred(args->context), &bp); if (error) { printf("hfs_journal_relocate_callback: failed to read JIB (%d)\n", error); if (bp) { buf_brelse(bp); } return error; } jibp = (JournalInfoBlock*) buf_dataptr(bp); jibp->offset = SWAP_BE64((u_int64_t)args->newStartBlock * hfsmp->blockSize); jibp->size = SWAP_BE64((u_int64_t)args->newBlockCount * hfsmp->blockSize); if (journal_uses_fua(hfsmp->jnl)) buf_markfua(bp); error = buf_bwrite(bp); if (error) { printf("hfs_journal_relocate_callback: failed to write JIB (%d)\n", error); return error; } if (!journal_uses_fua(hfsmp->jnl)) { error = VNOP_IOCTL(hfsmp->hfs_devvp, DKIOCSYNCHRONIZECACHE, NULL, FWRITE, args->context); if (error) { printf("hfs_journal_relocate_callback: DKIOCSYNCHRONIZECACHE failed (%d)\n", error); error = 0; /* Don't fail the operation. */ } } return error; } /* Type of resize operation in progress */ #define HFS_RESIZE_TRUNCATE 1 #define HFS_RESIZE_EXTEND 2 /* * Core function to relocate the journal file. This function takes the * journal size of the newly relocated journal --- the caller can * provide a new journal size if they want to change the size of * the journal. The function takes care of updating the journal info * block and all other data structures correctly. * * Note: This function starts a transaction and grabs the btree locks. */ static int hfs_relocate_journal_file(struct hfsmount *hfsmp, u_int32_t jnl_size, int resize_type, vfs_context_t context) { int error; int journal_err; int lockflags; u_int32_t oldStartBlock; u_int32_t newStartBlock; u_int32_t oldBlockCount; u_int32_t newBlockCount; u_int32_t jnlBlockCount; u_int32_t alloc_skipfreeblks; struct cat_desc journal_desc; struct cat_attr journal_attr; struct cat_fork journal_fork; struct hfs_journal_relocate_args callback_args; /* Calculate the number of allocation blocks required for the journal */ jnlBlockCount = howmany(jnl_size, hfsmp->blockSize); /* * During truncatefs(), the volume free block count is updated * before relocating data and reflects the total number of free * blocks that will exist on volume after the resize is successful. * This means that the allocation blocks required for relocation * have already been reserved and accounted for in the free block * count. Therefore, block allocation and deallocation routines * can skip the free block check by passing HFS_ALLOC_SKIPFREEBLKS * flag. * * This special handling is not required when the file system * is being extended as we want all the allocated and deallocated * blocks to be accounted for correctly. */ if (resize_type == HFS_RESIZE_TRUNCATE) { alloc_skipfreeblks = HFS_ALLOC_SKIPFREEBLKS; } else { alloc_skipfreeblks = 0; } error = hfs_start_transaction(hfsmp); if (error) { printf("hfs_relocate_journal_file: hfs_start_transaction returned %d\n", error); return error; } lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG | SFL_BITMAP, HFS_EXCLUSIVE_LOCK); error = BlockAllocate(hfsmp, 1, jnlBlockCount, jnlBlockCount, HFS_ALLOC_METAZONE | HFS_ALLOC_FORCECONTIG | HFS_ALLOC_FLUSHTXN | alloc_skipfreeblks, &newStartBlock, &newBlockCount); if (error) { printf("hfs_relocate_journal_file: BlockAllocate returned %d\n", error); goto fail; } if (newBlockCount != jnlBlockCount) { printf("hfs_relocate_journal_file: newBlockCount != jnlBlockCount (%u, %u)\n", newBlockCount, jnlBlockCount); goto free_fail; } error = cat_idlookup(hfsmp, hfsmp->hfs_jnlfileid, 1, 0, &journal_desc, &journal_attr, &journal_fork); if (error) { printf("hfs_relocate_journal_file: cat_idlookup returned %d\n", error); goto free_fail; } oldStartBlock = journal_fork.cf_extents[0].startBlock; oldBlockCount = journal_fork.cf_extents[0].blockCount; error = BlockDeallocate(hfsmp, oldStartBlock, oldBlockCount, alloc_skipfreeblks); if (error) { printf("hfs_relocate_journal_file: BlockDeallocate returned %d\n", error); goto free_fail; } /* Update the catalog record for .journal */ journal_fork.cf_size = newBlockCount * hfsmp->blockSize; journal_fork.cf_extents[0].startBlock = newStartBlock; journal_fork.cf_extents[0].blockCount = newBlockCount; journal_fork.cf_blocks = newBlockCount; error = cat_update(hfsmp, &journal_desc, &journal_attr, &journal_fork, NULL); cat_releasedesc(&journal_desc); /* all done with cat descriptor */ if (error) { printf("hfs_relocate_journal_file: cat_update returned %d\n", error); goto free_fail; } /* * If the journal is part of the file system, then tell the journal * code about the new location. If the journal is on an external * device, then just keep using it as-is. */ if (hfsmp->jvp == hfsmp->hfs_devvp) { callback_args.hfsmp = hfsmp; callback_args.context = context; callback_args.newStartBlock = newStartBlock; callback_args.newBlockCount = newBlockCount; error = journal_relocate(hfsmp->jnl, (off_t)newStartBlock*hfsmp->blockSize, (off_t)newBlockCount*hfsmp->blockSize, 0, hfs_journal_relocate_callback, &callback_args); if (error) { /* NOTE: journal_relocate will mark the journal invalid. */ printf("hfs_relocate_journal_file: journal_relocate returned %d\n", error); goto fail; } if (hfs_resize_debug) { printf ("hfs_relocate_journal_file: Successfully relocated journal from (%u,%u) to (%u,%u)\n", oldStartBlock, oldBlockCount, newStartBlock, newBlockCount); } hfsmp->jnl_start = newStartBlock; hfsmp->jnl_size = (off_t)newBlockCount * hfsmp->blockSize; } hfs_systemfile_unlock(hfsmp, lockflags); error = hfs_end_transaction(hfsmp); if (error) { printf("hfs_relocate_journal_file: hfs_end_transaction returned %d\n", error); } return error; free_fail: journal_err = BlockDeallocate(hfsmp, newStartBlock, newBlockCount, HFS_ALLOC_SKIPFREEBLKS); if (journal_err) { printf("hfs_relocate_journal_file: BlockDeallocate returned %d\n", error); hfs_mark_inconsistent(hfsmp, HFS_ROLLBACK_FAILED); } fail: hfs_systemfile_unlock(hfsmp, lockflags); (void) hfs_end_transaction(hfsmp); if (hfs_resize_debug) { printf ("hfs_relocate_journal_file: Error relocating journal file (error=%d)\n", error); } return error; } /* * Relocate the journal file when the file system is being truncated. * We do not down-size the journal when the file system size is * reduced, so we always provide the current journal size to the * relocate code. */ static int hfs_reclaim_journal_file(struct hfsmount *hfsmp, u_int32_t allocLimit, vfs_context_t context) { int error = 0; u_int32_t startBlock; u_int32_t blockCount = hfsmp->jnl_size / hfsmp->blockSize; /* * Figure out the location of the .journal file. When the journal * is on an external device, we need to look up the .journal file. */ if (hfsmp->jvp == hfsmp->hfs_devvp) { startBlock = hfsmp->jnl_start; blockCount = hfsmp->jnl_size / hfsmp->blockSize; } else { u_int32_t fileid; u_int32_t old_jnlfileid; struct cat_attr attr; struct cat_fork fork; /* * The cat_lookup inside GetFileInfo will fail because hfs_jnlfileid * is set, and it is trying to hide the .journal file. So temporarily * unset the field while calling GetFileInfo. */ old_jnlfileid = hfsmp->hfs_jnlfileid; hfsmp->hfs_jnlfileid = 0; fileid = GetFileInfo(hfsmp, kHFSRootFolderID, ".journal", &attr, &fork); hfsmp->hfs_jnlfileid = old_jnlfileid; if (fileid != old_jnlfileid) { printf("hfs_reclaim_journal_file: cannot find .journal file!\n"); return EIO; } startBlock = fork.cf_extents[0].startBlock; blockCount = fork.cf_extents[0].blockCount; } if (startBlock + blockCount <= allocLimit) { /* The journal file does not require relocation */ return 0; } error = hfs_relocate_journal_file(hfsmp, blockCount * hfsmp->blockSize, HFS_RESIZE_TRUNCATE, context); if (error == 0) { hfsmp->hfs_resize_blocksmoved += blockCount; hfs_truncatefs_progress(hfsmp); printf ("hfs_reclaim_journal_file: Relocated %u blocks from journal on \"%s\"\n", blockCount, hfsmp->vcbVN); } return error; } /* * Move the journal info block to a new location. We have to make sure the * new copy of the journal info block gets to the media first, then change * the field in the volume header and the catalog record. */ static int hfs_reclaim_journal_info_block(struct hfsmount *hfsmp, u_int32_t allocLimit, vfs_context_t context) { int error; int journal_err; int lockflags; u_int32_t oldBlock; u_int32_t newBlock; u_int32_t blockCount; struct cat_desc jib_desc; struct cat_attr jib_attr; struct cat_fork jib_fork; buf_t old_bp, new_bp; if (hfsmp->vcbJinfoBlock <= allocLimit) { /* The journal info block does not require relocation */ return 0; } error = hfs_start_transaction(hfsmp); if (error) { printf("hfs_reclaim_journal_info_block: hfs_start_transaction returned %d\n", error); return error; } lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG | SFL_BITMAP, HFS_EXCLUSIVE_LOCK); error = BlockAllocate(hfsmp, 1, 1, 1, HFS_ALLOC_METAZONE | HFS_ALLOC_FORCECONTIG | HFS_ALLOC_SKIPFREEBLKS | HFS_ALLOC_FLUSHTXN, &newBlock, &blockCount); if (error) { printf("hfs_reclaim_journal_info_block: BlockAllocate returned %d\n", error); goto fail; } if (blockCount != 1) { printf("hfs_reclaim_journal_info_block: blockCount != 1 (%u)\n", blockCount); goto free_fail; } /* Copy the old journal info block content to the new location */ error = buf_meta_bread(hfsmp->hfs_devvp, hfsmp->vcbJinfoBlock * (hfsmp->blockSize/hfsmp->hfs_logical_block_size), hfsmp->blockSize, vfs_context_ucred(context), &old_bp); if (error) { printf("hfs_reclaim_journal_info_block: failed to read JIB (%d)\n", error); if (old_bp) { buf_brelse(old_bp); } goto free_fail; } new_bp = buf_getblk(hfsmp->hfs_devvp, newBlock * (hfsmp->blockSize/hfsmp->hfs_logical_block_size), hfsmp->blockSize, 0, 0, BLK_META); bcopy((char*)buf_dataptr(old_bp), (char*)buf_dataptr(new_bp), hfsmp->blockSize); buf_brelse(old_bp); if (journal_uses_fua(hfsmp->jnl)) buf_markfua(new_bp); error = buf_bwrite(new_bp); if (error) { printf("hfs_reclaim_journal_info_block: failed to write new JIB (%d)\n", error); goto free_fail; } if (!journal_uses_fua(hfsmp->jnl)) { error = VNOP_IOCTL(hfsmp->hfs_devvp, DKIOCSYNCHRONIZECACHE, NULL, FWRITE, context); if (error) { printf("hfs_reclaim_journal_info_block: DKIOCSYNCHRONIZECACHE failed (%d)\n", error); /* Don't fail the operation. */ } } /* Deallocate the old block once the new one has the new valid content */ error = BlockDeallocate(hfsmp, hfsmp->vcbJinfoBlock, 1, HFS_ALLOC_SKIPFREEBLKS); if (error) { printf("hfs_reclaim_journal_info_block: BlockDeallocate returned %d\n", error); goto free_fail; } /* Update the catalog record for .journal_info_block */ error = cat_idlookup(hfsmp, hfsmp->hfs_jnlinfoblkid, 1, 0, &jib_desc, &jib_attr, &jib_fork); if (error) { printf("hfs_reclaim_journal_info_block: cat_idlookup returned %d\n", error); goto fail; } oldBlock = jib_fork.cf_extents[0].startBlock; jib_fork.cf_size = hfsmp->blockSize; jib_fork.cf_extents[0].startBlock = newBlock; jib_fork.cf_extents[0].blockCount = 1; jib_fork.cf_blocks = 1; error = cat_update(hfsmp, &jib_desc, &jib_attr, &jib_fork, NULL); cat_releasedesc(&jib_desc); /* all done with cat descriptor */ if (error) { printf("hfs_reclaim_journal_info_block: cat_update returned %d\n", error); goto fail; } /* Update the pointer to the journal info block in the volume header. */ hfsmp->vcbJinfoBlock = newBlock; error = hfs_flushvolumeheader(hfsmp, MNT_WAIT, HFS_ALTFLUSH); if (error) { printf("hfs_reclaim_journal_info_block: hfs_flushvolumeheader returned %d\n", error); goto fail; } hfs_systemfile_unlock(hfsmp, lockflags); error = hfs_end_transaction(hfsmp); if (error) { printf("hfs_reclaim_journal_info_block: hfs_end_transaction returned %d\n", error); } error = hfs_journal_flush(hfsmp, FALSE); if (error) { printf("hfs_reclaim_journal_info_block: journal_flush returned %d\n", error); } /* Account for the block relocated and print progress */ hfsmp->hfs_resize_blocksmoved += 1; hfs_truncatefs_progress(hfsmp); if (!error) { printf ("hfs_reclaim_journal_info: Relocated 1 block from journal info on \"%s\"\n", hfsmp->vcbVN); if (hfs_resize_debug) { printf ("hfs_reclaim_journal_info_block: Successfully relocated journal info block from (%u,%u) to (%u,%u)\n", oldBlock, blockCount, newBlock, blockCount); } } return error; free_fail: journal_err = BlockDeallocate(hfsmp, newBlock, blockCount, HFS_ALLOC_SKIPFREEBLKS); if (journal_err) { printf("hfs_reclaim_journal_info_block: BlockDeallocate returned %d\n", error); hfs_mark_inconsistent(hfsmp, HFS_ROLLBACK_FAILED); } fail: hfs_systemfile_unlock(hfsmp, lockflags); (void) hfs_end_transaction(hfsmp); if (hfs_resize_debug) { printf ("hfs_reclaim_journal_info_block: Error relocating journal info block (error=%d)\n", error); } return error; } static u_int64_t calculate_journal_size(struct hfsmount *hfsmp, u_int32_t sector_size, u_int64_t sector_count) { u_int64_t journal_size; u_int32_t journal_scale; #define DEFAULT_JOURNAL_SIZE (8*1024*1024) #define MAX_JOURNAL_SIZE (512*1024*1024) /* Calculate the journal size for this volume. We want * at least 8 MB of journal for each 100 GB of disk space. * We cap the size at 512 MB, unless the allocation block * size is larger, in which case, we use one allocation * block. */ journal_scale = (sector_size * sector_count) / ((u_int64_t)100 * 1024 * 1024 * 1024); journal_size = DEFAULT_JOURNAL_SIZE * (journal_scale + 1); if (journal_size > MAX_JOURNAL_SIZE) { journal_size = MAX_JOURNAL_SIZE; } if (journal_size < hfsmp->blockSize) { journal_size = hfsmp->blockSize; } return journal_size; } /* * Calculate the expected journal size based on current partition size. * If the size of the current journal is less than the calculated size, * force journal relocation with the new journal size. */ static int hfs_extend_journal(struct hfsmount *hfsmp, u_int32_t sector_size, u_int64_t sector_count, vfs_context_t context) { int error = 0; u_int64_t calc_journal_size; if (hfsmp->jvp != hfsmp->hfs_devvp) { if (hfs_resize_debug) { printf("hfs_extend_journal: not resizing the journal because it is on an external device.\n"); } return 0; } calc_journal_size = calculate_journal_size(hfsmp, sector_size, sector_count); if (calc_journal_size <= hfsmp->jnl_size) { /* The journal size requires no modification */ goto out; } if (hfs_resize_debug) { printf ("hfs_extend_journal: journal old=%u, new=%qd\n", hfsmp->jnl_size, calc_journal_size); } /* Extend the journal to the new calculated size */ error = hfs_relocate_journal_file(hfsmp, calc_journal_size, HFS_RESIZE_EXTEND, context); if (error == 0) { printf ("hfs_extend_journal: Extended journal size to %u bytes on \"%s\"\n", hfsmp->jnl_size, hfsmp->vcbVN); } out: return error; } /* * This function traverses through all extended attribute records for a given * fileID, and calls function that reclaims data blocks that exist in the * area of the disk being reclaimed which in turn is responsible for allocating * new space, copying extent data, deallocating new space, and if required, * splitting the extent. * * Note: The caller has already acquired the cnode lock on the file. Therefore * we are assured that no other thread would be creating/deleting/modifying * extended attributes for this file. * * Side Effects: * hfsmp->hfs_resize_blocksmoved is incremented by the number of allocation * blocks that were relocated. * * Returns: * 0 on success, non-zero on failure. */ static int hfs_reclaim_xattr(struct hfsmount *hfsmp, struct vnode *vp, u_int32_t fileID, u_int32_t allocLimit, vfs_context_t context) { int error = 0; struct hfs_reclaim_extent_info *extent_info; int i; HFSPlusAttrKey *key; int *lockflags; if (hfs_resize_debug) { printf("hfs_reclaim_xattr: === Start reclaiming xattr for id=%u ===\n", fileID); } MALLOC(extent_info, struct hfs_reclaim_extent_info *, sizeof(struct hfs_reclaim_extent_info), M_TEMP, M_WAITOK); if (extent_info == NULL) { return ENOMEM; } bzero(extent_info, sizeof(struct hfs_reclaim_extent_info)); extent_info->vp = vp; extent_info->fileID = fileID; extent_info->is_xattr = true; extent_info->is_sysfile = vnode_issystem(vp); extent_info->fcb = VTOF(hfsmp->hfs_attribute_vp); lockflags = &(extent_info->lockflags); *lockflags = SFL_ATTRIBUTE | SFL_BITMAP; /* Initialize iterator from the extent_info structure */ MALLOC(extent_info->iterator, struct BTreeIterator *, sizeof(struct BTreeIterator), M_TEMP, M_WAITOK); if (extent_info->iterator == NULL) { error = ENOMEM; goto out; } bzero(extent_info->iterator, sizeof(struct BTreeIterator)); /* Build attribute key */ key = (HFSPlusAttrKey *)&(extent_info->iterator->key); error = hfs_buildattrkey(fileID, NULL, key); if (error) { goto out; } /* Initialize btdata from extent_info structure. Note that the * buffer pointer actually points to the xattr record from the * extent_info structure itself. */ extent_info->btdata.bufferAddress = &(extent_info->record.xattr); extent_info->btdata.itemSize = sizeof(HFSPlusAttrRecord); extent_info->btdata.itemCount = 1; /* * Sync all extent-based attribute data to the disk. * * All extent-based attribute data I/O is performed via cluster * I/O using a virtual file that spans across entire file system * space. */ hfs_lock_truncate(VTOC(hfsmp->hfs_attrdata_vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT); (void)cluster_push(hfsmp->hfs_attrdata_vp, 0); error = vnode_waitforwrites(hfsmp->hfs_attrdata_vp, 0, 0, 0, "hfs_reclaim_xattr"); hfs_unlock_truncate(VTOC(hfsmp->hfs_attrdata_vp), HFS_LOCK_DEFAULT); if (error) { goto out; } /* Search for extended attribute for current file. This * will place the iterator before the first matching record. */ *lockflags = hfs_systemfile_lock(hfsmp, *lockflags, HFS_EXCLUSIVE_LOCK); error = BTSearchRecord(extent_info->fcb, extent_info->iterator, &(extent_info->btdata), &(extent_info->recordlen), extent_info->iterator); hfs_systemfile_unlock(hfsmp, *lockflags); if (error) { if (error != btNotFound) { goto out; } /* btNotFound is expected here, so just mask it */ error = 0; } while (1) { /* Iterate to the next record */ *lockflags = hfs_systemfile_lock(hfsmp, *lockflags, HFS_EXCLUSIVE_LOCK); error = BTIterateRecord(extent_info->fcb, kBTreeNextRecord, extent_info->iterator, &(extent_info->btdata), &(extent_info->recordlen)); hfs_systemfile_unlock(hfsmp, *lockflags); /* Stop the iteration if we encounter end of btree or xattr with different fileID */ if (error || key->fileID != fileID) { if (error == fsBTRecordNotFoundErr || error == fsBTEndOfIterationErr) { error = 0; } break; } /* We only care about extent-based EAs */ if ((extent_info->record.xattr.recordType != kHFSPlusAttrForkData) && (extent_info->record.xattr.recordType != kHFSPlusAttrExtents)) { continue; } if (extent_info->record.xattr.recordType == kHFSPlusAttrForkData) { extent_info->overflow_count = 0; extent_info->extents = extent_info->record.xattr.forkData.theFork.extents; } else if (extent_info->record.xattr.recordType == kHFSPlusAttrExtents) { extent_info->overflow_count++; extent_info->extents = extent_info->record.xattr.overflowExtents.extents; } extent_info->recStartBlock = key->startBlock; for (i = 0; i < kHFSPlusExtentDensity; i++) { if (extent_info->extents[i].blockCount == 0) { break; } extent_info->extent_index = i; error = hfs_reclaim_extent(hfsmp, allocLimit, extent_info, context); if (error) { printf ("hfs_reclaim_xattr: fileID=%u hfs_reclaim_extent error=%d\n", fileID, error); goto out; } } } out: /* If any blocks were relocated, account them and report progress */ if (extent_info->blocks_relocated) { hfsmp->hfs_resize_blocksmoved += extent_info->blocks_relocated; hfs_truncatefs_progress(hfsmp); } if (extent_info->iterator) { FREE(extent_info->iterator, M_TEMP); } if (extent_info) { FREE(extent_info, M_TEMP); } if (hfs_resize_debug) { printf("hfs_reclaim_xattr: === Finished relocating xattr for fileid=%u (error=%d) ===\n", fileID, error); } return error; } /* * Reclaim any extent-based extended attributes allocation blocks from * the area of the disk that is being truncated. * * The function traverses the attribute btree to find out the fileIDs * of the extended attributes that need to be relocated. For every * file whose large EA requires relocation, it looks up the cnode and * calls hfs_reclaim_xattr() to do all the work for allocating * new space, copying data, deallocating old space, and if required, * splitting the extents. * * Inputs: * allocLimit - starting block of the area being reclaimed * * Returns: * returns 0 on success, non-zero on failure. */ static int hfs_reclaim_xattrspace(struct hfsmount *hfsmp, u_int32_t allocLimit, vfs_context_t context) { int error = 0; FCB *fcb; struct BTreeIterator *iterator = NULL; struct FSBufferDescriptor btdata; HFSPlusAttrKey *key; HFSPlusAttrRecord rec; int lockflags = 0; cnid_t prev_fileid = 0; struct vnode *vp; int need_relocate; int btree_operation; u_int32_t files_moved = 0; u_int32_t prev_blocksmoved; int i; fcb = VTOF(hfsmp->hfs_attribute_vp); /* Store the value to print total blocks moved by this function in end */ prev_blocksmoved = hfsmp->hfs_resize_blocksmoved; if (kmem_alloc(kernel_map, (vm_offset_t *)&iterator, sizeof(*iterator))) { return ENOMEM; } bzero(iterator, sizeof(*iterator)); key = (HFSPlusAttrKey *)&iterator->key; btdata.bufferAddress = &rec; btdata.itemSize = sizeof(rec); btdata.itemCount = 1; need_relocate = false; btree_operation = kBTreeFirstRecord; /* Traverse the attribute btree to find extent-based EAs to reclaim */ while (1) { lockflags = hfs_systemfile_lock(hfsmp, SFL_ATTRIBUTE, HFS_SHARED_LOCK); error = BTIterateRecord(fcb, btree_operation, iterator, &btdata, NULL); hfs_systemfile_unlock(hfsmp, lockflags); if (error) { if (error == fsBTRecordNotFoundErr || error == fsBTEndOfIterationErr) { error = 0; } break; } btree_operation = kBTreeNextRecord; /* If the extents of current fileID were already relocated, skip it */ if (prev_fileid == key->fileID) { continue; } /* Check if any of the extents in the current record need to be relocated */ need_relocate = false; switch(rec.recordType) { case kHFSPlusAttrForkData: for (i = 0; i < kHFSPlusExtentDensity; i++) { if (rec.forkData.theFork.extents[i].blockCount == 0) { break; } if ((rec.forkData.theFork.extents[i].startBlock + rec.forkData.theFork.extents[i].blockCount) > allocLimit) { need_relocate = true; break; } } break; case kHFSPlusAttrExtents: for (i = 0; i < kHFSPlusExtentDensity; i++) { if (rec.overflowExtents.extents[i].blockCount == 0) { break; } if ((rec.overflowExtents.extents[i].startBlock + rec.overflowExtents.extents[i].blockCount) > allocLimit) { need_relocate = true; break; } } break; }; /* Continue iterating to next attribute record */ if (need_relocate == false) { continue; } /* Look up the vnode for corresponding file. The cnode * will be locked which will ensure that no one modifies * the xattrs when we are relocating them. * * We want to allow open-unlinked files to be moved, * so provide allow_deleted == 1 for hfs_vget(). */ if (hfs_vget(hfsmp, key->fileID, &vp, 0, 1) != 0) { continue; } error = hfs_reclaim_xattr(hfsmp, vp, key->fileID, allocLimit, context); hfs_unlock(VTOC(vp)); vnode_put(vp); if (error) { printf ("hfs_reclaim_xattrspace: Error relocating xattrs for fileid=%u (error=%d)\n", key->fileID, error); break; } prev_fileid = key->fileID; files_moved++; } if (files_moved) { printf("hfs_reclaim_xattrspace: Relocated %u xattr blocks from %u files on \"%s\"\n", (hfsmp->hfs_resize_blocksmoved - prev_blocksmoved), files_moved, hfsmp->vcbVN); } kmem_free(kernel_map, (vm_offset_t)iterator, sizeof(*iterator)); return error; } /* * Reclaim blocks from regular files. * * This function iterates over all the record in catalog btree looking * for files with extents that overlap into the space we're trying to * free up. If a file extent requires relocation, it looks up the vnode * and calls function to relocate the data. * * Returns: * Zero on success, non-zero on failure. */ static int hfs_reclaim_filespace(struct hfsmount *hfsmp, u_int32_t allocLimit, vfs_context_t context) { int error; FCB *fcb; struct BTreeIterator *iterator = NULL; struct FSBufferDescriptor btdata; int btree_operation; int lockflags; struct HFSPlusCatalogFile filerec; struct vnode *vp; struct vnode *rvp; struct filefork *datafork; u_int32_t files_moved = 0; u_int32_t prev_blocksmoved; #if CONFIG_PROTECT int keys_generated = 0; #endif fcb = VTOF(hfsmp->hfs_catalog_vp); /* Store the value to print total blocks moved by this function at the end */ prev_blocksmoved = hfsmp->hfs_resize_blocksmoved; if (kmem_alloc(kernel_map, (vm_offset_t *)&iterator, sizeof(*iterator))) { error = ENOMEM; goto reclaim_filespace_done; } #if CONFIG_PROTECT /* * For content-protected filesystems, we may need to relocate files that * are encrypted. If they use the new-style offset-based IVs, then * we can move them regardless of the lock state. We create a temporary * key here that we use to read/write the data, then we discard it at the * end of the function. */ if (cp_fs_protected (hfsmp->hfs_mp)) { int needs = 0; error = cp_needs_tempkeys(hfsmp, &needs); if ((error == 0) && (needs)) { error = cp_entry_gentempkeys(&hfsmp->hfs_resize_cpentry, hfsmp); if (error == 0) { keys_generated = 1; } } if (error) { printf("hfs_reclaimspace: Error generating temporary keys for resize (%d)\n", error); goto reclaim_filespace_done; } } #endif bzero(iterator, sizeof(*iterator)); btdata.bufferAddress = &filerec; btdata.itemSize = sizeof(filerec); btdata.itemCount = 1; btree_operation = kBTreeFirstRecord; while (1) { lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_SHARED_LOCK); error = BTIterateRecord(fcb, btree_operation, iterator, &btdata, NULL); hfs_systemfile_unlock(hfsmp, lockflags); if (error) { if (error == fsBTRecordNotFoundErr || error == fsBTEndOfIterationErr) { error = 0; } break; } btree_operation = kBTreeNextRecord; if (filerec.recordType != kHFSPlusFileRecord) { continue; } /* Check if any of the extents require relocation */ if (hfs_file_extent_overlaps(hfsmp, allocLimit, &filerec) == false) { continue; } /* We want to allow open-unlinked files to be moved, so allow_deleted == 1 */ if (hfs_vget(hfsmp, filerec.fileID, &vp, 0, 1) != 0) { if (hfs_resize_debug) { printf("hfs_reclaim_filespace: hfs_vget(%u) failed.\n", filerec.fileID); } continue; } /* If data fork exists or item is a directory hard link, relocate blocks */ datafork = VTOF(vp); if ((datafork && datafork->ff_blocks > 0) || vnode_isdir(vp)) { error = hfs_reclaim_file(hfsmp, vp, filerec.fileID, kHFSDataForkType, allocLimit, context); if (error) { printf ("hfs_reclaimspace: Error reclaiming datafork blocks of fileid=%u (error=%d)\n", filerec.fileID, error); hfs_unlock(VTOC(vp)); vnode_put(vp); break; } } /* If resource fork exists or item is a directory hard link, relocate blocks */ if (((VTOC(vp)->c_blocks - (datafork ? datafork->ff_blocks : 0)) > 0) || vnode_isdir(vp)) { if (vnode_isdir(vp)) { /* Resource fork vnode lookup is invalid for directory hard link. * So we fake data fork vnode as resource fork vnode. */ rvp = vp; } else { error = hfs_vgetrsrc(hfsmp, vp, &rvp); if (error) { printf ("hfs_reclaimspace: Error looking up rvp for fileid=%u (error=%d)\n", filerec.fileID, error); hfs_unlock(VTOC(vp)); vnode_put(vp); break; } VTOC(rvp)->c_flag |= C_NEED_RVNODE_PUT; } error = hfs_reclaim_file(hfsmp, rvp, filerec.fileID, kHFSResourceForkType, allocLimit, context); if (error) { printf ("hfs_reclaimspace: Error reclaiming rsrcfork blocks of fileid=%u (error=%d)\n", filerec.fileID, error); hfs_unlock(VTOC(vp)); vnode_put(vp); break; } } /* The file forks were relocated successfully, now drop the * cnode lock and vnode reference, and continue iterating to * next catalog record. */ hfs_unlock(VTOC(vp)); vnode_put(vp); files_moved++; } if (files_moved) { printf("hfs_reclaim_filespace: Relocated %u blocks from %u files on \"%s\"\n", (hfsmp->hfs_resize_blocksmoved - prev_blocksmoved), files_moved, hfsmp->vcbVN); } reclaim_filespace_done: if (iterator) { kmem_free(kernel_map, (vm_offset_t)iterator, sizeof(*iterator)); } #if CONFIG_PROTECT if (keys_generated) { cp_entry_destroy(hfsmp->hfs_resize_cpentry); hfsmp->hfs_resize_cpentry = NULL; } #endif return error; } /* * Reclaim space at the end of a file system. * * Inputs - * allocLimit - start block of the space being reclaimed * reclaimblks - number of allocation blocks to reclaim */ static int hfs_reclaimspace(struct hfsmount *hfsmp, u_int32_t allocLimit, u_int32_t reclaimblks, vfs_context_t context) { int error = 0; /* * Preflight the bitmap to find out total number of blocks that need * relocation. * * Note: Since allocLimit is set to the location of new alternate volume * header, the check below does not account for blocks allocated for old * alternate volume header. */ error = hfs_count_allocated(hfsmp, allocLimit, reclaimblks, &(hfsmp->hfs_resize_totalblocks)); if (error) { printf ("hfs_reclaimspace: Unable to determine total blocks to reclaim error=%d\n", error); return error; } if (hfs_resize_debug) { printf ("hfs_reclaimspace: Total number of blocks to reclaim = %u\n", hfsmp->hfs_resize_totalblocks); } /* Just to be safe, sync the content of the journal to the disk before we proceed */ hfs_journal_flush(hfsmp, TRUE); /* First, relocate journal file blocks if they're in the way. * Doing this first will make sure that journal relocate code * gets access to contiguous blocks on disk first. The journal * file has to be contiguous on the disk, otherwise resize will * fail. */ error = hfs_reclaim_journal_file(hfsmp, allocLimit, context); if (error) { printf("hfs_reclaimspace: hfs_reclaim_journal_file failed (%d)\n", error); return error; } /* Relocate journal info block blocks if they're in the way. */ error = hfs_reclaim_journal_info_block(hfsmp, allocLimit, context); if (error) { printf("hfs_reclaimspace: hfs_reclaim_journal_info_block failed (%d)\n", error); return error; } /* Relocate extents of the Extents B-tree if they're in the way. * Relocating extents btree before other btrees is important as * this will provide access to largest contiguous block range on * the disk for relocating extents btree. Note that extents btree * can only have maximum of 8 extents. */ error = hfs_reclaim_file(hfsmp, hfsmp->hfs_extents_vp, kHFSExtentsFileID, kHFSDataForkType, allocLimit, context); if (error) { printf("hfs_reclaimspace: reclaim extents b-tree returned %d\n", error); return error; } /* Relocate extents of the Allocation file if they're in the way. */ error = hfs_reclaim_file(hfsmp, hfsmp->hfs_allocation_vp, kHFSAllocationFileID, kHFSDataForkType, allocLimit, context); if (error) { printf("hfs_reclaimspace: reclaim allocation file returned %d\n", error); return error; } /* Relocate extents of the Catalog B-tree if they're in the way. */ error = hfs_reclaim_file(hfsmp, hfsmp->hfs_catalog_vp, kHFSCatalogFileID, kHFSDataForkType, allocLimit, context); if (error) { printf("hfs_reclaimspace: reclaim catalog b-tree returned %d\n", error); return error; } /* Relocate extents of the Attributes B-tree if they're in the way. */ error = hfs_reclaim_file(hfsmp, hfsmp->hfs_attribute_vp, kHFSAttributesFileID, kHFSDataForkType, allocLimit, context); if (error) { printf("hfs_reclaimspace: reclaim attribute b-tree returned %d\n", error); return error; } /* Relocate extents of the Startup File if there is one and they're in the way. */ error = hfs_reclaim_file(hfsmp, hfsmp->hfs_startup_vp, kHFSStartupFileID, kHFSDataForkType, allocLimit, context); if (error) { printf("hfs_reclaimspace: reclaim startup file returned %d\n", error); return error; } /* * We need to make sure the alternate volume header gets flushed if we moved * any extents in the volume header. But we need to do that before * shrinking the size of the volume, or else the journal code will panic * with an invalid (too large) block number. * * Note that blks_moved will be set if ANY extent was moved, even * if it was just an overflow extent. In this case, the journal_flush isn't * strictly required, but shouldn't hurt. */ if (hfsmp->hfs_resize_blocksmoved) { hfs_journal_flush(hfsmp, TRUE); } /* Reclaim extents from catalog file records */ error = hfs_reclaim_filespace(hfsmp, allocLimit, context); if (error) { printf ("hfs_reclaimspace: hfs_reclaim_filespace returned error=%d\n", error); return error; } /* Reclaim extents from extent-based extended attributes, if any */ error = hfs_reclaim_xattrspace(hfsmp, allocLimit, context); if (error) { printf ("hfs_reclaimspace: hfs_reclaim_xattrspace returned error=%d\n", error); return error; } return error; } /* * Check if there are any extents (including overflow extents) that overlap * into the disk space that is being reclaimed. * * Output - * true - One of the extents need to be relocated * false - No overflow extents need to be relocated, or there was an error */ static int hfs_file_extent_overlaps(struct hfsmount *hfsmp, u_int32_t allocLimit, struct HFSPlusCatalogFile *filerec) { struct BTreeIterator * iterator = NULL; struct FSBufferDescriptor btdata; HFSPlusExtentRecord extrec; HFSPlusExtentKey *extkeyptr; FCB *fcb; int overlapped = false; int i, j; int error; int lockflags = 0; u_int32_t endblock; /* Check if data fork overlaps the target space */ for (i = 0; i < kHFSPlusExtentDensity; ++i) { if (filerec->dataFork.extents[i].blockCount == 0) { break; } endblock = filerec->dataFork.extents[i].startBlock + filerec->dataFork.extents[i].blockCount; if (endblock > allocLimit) { overlapped = true; goto out; } } /* Check if resource fork overlaps the target space */ for (j = 0; j < kHFSPlusExtentDensity; ++j) { if (filerec->resourceFork.extents[j].blockCount == 0) { break; } endblock = filerec->resourceFork.extents[j].startBlock + filerec->resourceFork.extents[j].blockCount; if (endblock > allocLimit) { overlapped = true; goto out; } } /* Return back if there are no overflow extents for this file */ if ((i < kHFSPlusExtentDensity) && (j < kHFSPlusExtentDensity)) { goto out; } if (kmem_alloc(kernel_map, (vm_offset_t *)&iterator, sizeof(*iterator))) { return 0; } bzero(iterator, sizeof(*iterator)); extkeyptr = (HFSPlusExtentKey *)&iterator->key; extkeyptr->keyLength = kHFSPlusExtentKeyMaximumLength; extkeyptr->forkType = 0; extkeyptr->fileID = filerec->fileID; extkeyptr->startBlock = 0; btdata.bufferAddress = &extrec; btdata.itemSize = sizeof(extrec); btdata.itemCount = 1; fcb = VTOF(hfsmp->hfs_extents_vp); lockflags = hfs_systemfile_lock(hfsmp, SFL_EXTENTS, HFS_SHARED_LOCK); /* This will position the iterator just before the first overflow * extent record for given fileID. It will always return btNotFound, * so we special case the error code. */ error = BTSearchRecord(fcb, iterator, &btdata, NULL, iterator); if (error && (error != btNotFound)) { goto out; } /* BTIterateRecord() might return error if the btree is empty, and * therefore we return that the extent does not overflow to the caller */ error = BTIterateRecord(fcb, kBTreeNextRecord, iterator, &btdata, NULL); while (error == 0) { /* Stop when we encounter a different file. */ if (extkeyptr->fileID != filerec->fileID) { break; } /* Check if any of the forks exist in the target space. */ for (i = 0; i < kHFSPlusExtentDensity; ++i) { if (extrec[i].blockCount == 0) { break; } endblock = extrec[i].startBlock + extrec[i].blockCount; if (endblock > allocLimit) { overlapped = true; goto out; } } /* Look for more records. */ error = BTIterateRecord(fcb, kBTreeNextRecord, iterator, &btdata, NULL); } out: if (lockflags) { hfs_systemfile_unlock(hfsmp, lockflags); } if (iterator) { kmem_free(kernel_map, (vm_offset_t)iterator, sizeof(*iterator)); } return overlapped; } /* * Calculate the progress of a file system resize operation. */ __private_extern__ int hfs_resize_progress(struct hfsmount *hfsmp, u_int32_t *progress) { if ((hfsmp->hfs_flags & HFS_RESIZE_IN_PROGRESS) == 0) { return (ENXIO); } if (hfsmp->hfs_resize_totalblocks > 0) { *progress = (u_int32_t)((hfsmp->hfs_resize_blocksmoved * 100ULL) / hfsmp->hfs_resize_totalblocks); } else { *progress = 0; } return (0); }