GIT 7dfbad57adcf068d66289f6789556e7970d70e3e master.kernel.org:/pub/scm/linux/kernel/git/aia21/ntfs-2.6-devel.git --- diff --git a/Documentation/filesystems/ntfs.txt b/Documentation/filesystems/ntfs.txt --- a/Documentation/filesystems/ntfs.txt +++ b/Documentation/filesystems/ntfs.txt @@ -50,9 +50,14 @@ userspace utilities, etc. Features ======== -- This is a complete rewrite of the NTFS driver that used to be in the kernel. - This new driver implements NTFS read support and is functionally equivalent - to the old ntfs driver. +- This is a complete rewrite of the NTFS driver that used to be in the 2.4 and + earlier kernels. This new driver implements NTFS read support and is + functionally equivalent to the old ntfs driver and it also implements limited + write support. The biggest limitation at present is that files/directories + cannot be created or deleted. See below for the list of write features that + are so far supported. Another limitation is that writing to compressed files + is not implemented at all. Also, neither read nor write access to encrypted + files is so far implemented. - The new driver has full support for sparse files on NTFS 3.x volumes which the old driver isn't happy with. - The new driver supports execution of binaries due to mmap() now being @@ -78,7 +83,20 @@ Features - The new driver supports fsync(2), fdatasync(2), and msync(2). - The new driver supports readv(2) and writev(2). - The new driver supports access time updates (including mtime and ctime). - +- The new driver supports truncate(2) and open(2) with O_TRUNC. But at present + only very limited support for highly fragmented files, i.e. ones which have + their data attribute split across multiple extents, is included. Another + limitation is that at present truncate(2) will never create sparse files, + since to mark a file sparse we need to modify the directory entry for the + file and we do not implement directory modifications yet. +- The new driver supports write(2) which can both overwrite existing data and + extend the file size so that you can write beyond the existing data. Also, + writing into sparse regions is supported and the holes are filled in with + clusters. But at present only limited support for highly fragmented files, + i.e. ones which have their data attribute split across multiple extents, is + included. Another limitation is that write(2) will never create sparse + files, since to mark a file sparse we need to modify the directory entry for + the file and we do not implement directory modifications yet. Supported mount options ======================= @@ -439,6 +457,22 @@ ChangeLog Note, a technical ChangeLog aimed at kernel hackers is in fs/ntfs/ChangeLog. +2.1.25: + - Write support is now extended with write(2) being able to both + overwrite existing file data and to extend files. Also, if a write + to a sparse region occurs, write(2) will fill in the hole. Note, + mmap(2) based writes still do not support writing into holes or + writing beyond the initialized size. + - Write support has a new feature and that is that truncate(2) and + open(2) with O_TRUNC are now implemented thus files can be both made + smaller and larger. + - Note: Both write(2) and truncate(2)/open(2) with O_TRUNC still have + limitations in that they + - only provide limited support for highly fragmented files. + - only work on regular, i.e. uncompressed and unencrypted files. + - never create sparse files although this will change once directory + operations are implemented. + - Lots of bug fixes and enhancements across the board. 2.1.24: - Support journals ($LogFile) which have been modified by chkdsk. This means users can boot into Windows after we marked the volume dirty. diff --git a/fs/ntfs/ChangeLog b/fs/ntfs/ChangeLog --- a/fs/ntfs/ChangeLog +++ b/fs/ntfs/ChangeLog @@ -1,18 +1,15 @@ ToDo/Notes: - Find and fix bugs. - - In between ntfs_prepare/commit_write, need exclusion between - simultaneous file extensions. This is given to us by holding i_sem - on the inode. The only places in the kernel when a file is resized - are prepare/commit write and truncate for both of which i_sem is - held. Just have to be careful in readpage/writepage and all other - helpers not running under i_sem that we play nice... - Also need to be careful with initialized_size extention in - ntfs_prepare_write. Basically, just be _very_ careful in this code... - UPDATE: The only things that need to be checked are read/writepage - which do not hold i_sem. Note writepage cannot change i_size but it - needs to cope with a concurrent i_size change, just like readpage. - Also both need to cope with concurrent changes to the other sizes, - i.e. initialized/allocated/compressed size, as well. + - The only places in the kernel where a file is resized are + ntfs_file_write*() and ntfs_truncate() for both of which i_sem is + held. Just have to be careful in read-/writepage and other helpers + not running under i_sem that we play nice... Also need to be careful + with initialized_size extension in ntfs_file_write*() and writepage. + UPDATE: The only things that need to be checked are the compressed + write and the other attribute resize/write cases like index + attributes, etc. For now none of these are implemented so are safe. + - Implement filling in of holes in aops.c::ntfs_writepage() and its + helpers. - Implement mft.c::sync_mft_mirror_umount(). We currently will just leave the volume dirty on umount if the final iput(vol->mft_ino) causes a write of any mirrored mft records due to the mft mirror @@ -22,6 +19,64 @@ ToDo/Notes: - Enable the code for setting the NT4 compatibility flag when we start making NTFS 1.2 specific modifications. +2.1.25 - (Almost) fully implement write(2) and truncate(2). + + - Change ntfs_map_runlist_nolock(), ntfs_attr_find_vcn_nolock() and + {__,}ntfs_cluster_free() to also take an optional attribute search + context as argument. This allows calling these functions with the + mft record mapped. Update all callers. + - Fix potential deadlock in ntfs_mft_data_extend_allocation_nolock() + error handling by passing in the active search context when calling + ntfs_cluster_free(). + - Change ntfs_cluster_alloc() to take an extra boolean parameter + specifying whether the cluster are being allocated to extend an + attribute or to fill a hole. + - Change ntfs_attr_make_non_resident() to call ntfs_cluster_alloc() + with @is_extension set to TRUE and remove the runlist terminator + fixup code as this is now done by ntfs_cluster_alloc(). + - Change ntfs_attr_make_non_resident to take the attribute value size + as an extra parameter. This is needed since we need to know the size + before we can map the mft record and our callers always know it. The + reason we cannot simply read the size from the vfs inode i_size is + that this is not necessarily uptodate. This happens when + ntfs_attr_make_non_resident() is called in the ->truncate call path. + - Fix ntfs_attr_make_non_resident() to update the vfs inode i_blocks + which is zero for a resident attribute but should no longer be zero + once the attribute is non-resident as it then has real clusters + allocated. + - Add fs/ntfs/attrib.[hc]::ntfs_attr_extend_allocation(), a function to + extend the allocation of an attributes. Optionally, the data size, + but not the initialized size can be extended, too. + - Implement fs/ntfs/inode.[hc]::ntfs_truncate(). It only supports + uncompressed and unencrypted files and it never creates sparse files + at least for the moment (making a file sparse requires us to modify + its directory entries and we do not support directory operations at + the moment). Also, support for highly fragmented files, i.e. ones + whose data attribute is split across multiple extents, is severly + limited. When such a case is encountered, EOPNOTSUPP is returned. + - Enable ATTR_SIZE attribute changes in ntfs_setattr(). This completes + the initial implementation of file truncation. Now both open(2)ing + a file with the O_TRUNC flag and the {,f}truncate(2) system calls + will resize a file appropriately. The limitations are that only + uncompressed and unencrypted files are supported. Also, there is + only very limited support for highly fragmented files (the ones whose + $DATA attribute is split into multiple attribute extents). + - In attrib.c::ntfs_attr_set() call balance_dirty_pages_ratelimited() + and cond_resched() in the main loop as we could be dirtying a lot of + pages and this ensures we play nice with the VM and the system as a + whole. + - Implement file operations ->write, ->aio_write, ->writev for regular + files. This replaces the old use of generic_file_write(), et al and + the address space operations ->prepare_write and ->commit_write. + This means that both sparse and non-sparse (unencrypted and + uncompressed) files can now be extended using the normal write(2) + code path. There are two limitations at present and these are that + we never create sparse files and that we only have limited support + for highly fragmented files, i.e. ones whose data attribute is split + across multiple extents. When such a case is encountered, + EOPNOTSUPP is returned. + - $EA attributes can be both resident non-resident. + 2.1.24 - Lots of bug fixes and support more clean journal states. - Support journals ($LogFile) which have been modified by chkdsk. This diff --git a/fs/ntfs/Makefile b/fs/ntfs/Makefile --- a/fs/ntfs/Makefile +++ b/fs/ntfs/Makefile @@ -6,7 +6,7 @@ ntfs-objs := aops.o attrib.o collate.o c index.o inode.o mft.o mst.o namei.o runlist.o super.o sysctl.o \ unistr.o upcase.o -EXTRA_CFLAGS = -DNTFS_VERSION=\"2.1.24\" +EXTRA_CFLAGS = -DNTFS_VERSION=\"2.1.25\" ifeq ($(CONFIG_NTFS_DEBUG),y) EXTRA_CFLAGS += -DDEBUG diff --git a/fs/ntfs/aops.c b/fs/ntfs/aops.c --- a/fs/ntfs/aops.c +++ b/fs/ntfs/aops.c @@ -1391,8 +1391,7 @@ retry_writepage: if (NInoEncrypted(ni)) { unlock_page(page); BUG_ON(ni->type != AT_DATA); - ntfs_debug("Denying write access to encrypted " - "file."); + ntfs_debug("Denying write access to encrypted file."); return -EACCES; } /* Compressed data streams are handled in compress.c. */ @@ -1508,8 +1507,8 @@ retry_writepage: /* Zero out of bounds area in the page cache page. */ memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len); kunmap_atomic(kaddr, KM_USER0); - flush_dcache_mft_record_page(ctx->ntfs_ino); flush_dcache_page(page); + flush_dcache_mft_record_page(ctx->ntfs_ino); /* We are done with the page. */ end_page_writeback(page); /* Finally, mark the mft record dirty, so it gets written back. */ @@ -1542,830 +1541,6 @@ err_out: return err; } -/** - * ntfs_prepare_nonresident_write - - * - */ -static int ntfs_prepare_nonresident_write(struct page *page, - unsigned from, unsigned to) -{ - VCN vcn; - LCN lcn; - s64 initialized_size; - loff_t i_size; - sector_t block, ablock, iblock; - struct inode *vi; - ntfs_inode *ni; - ntfs_volume *vol; - runlist_element *rl; - struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; - unsigned long flags; - unsigned int vcn_ofs, block_start, block_end, blocksize; - int err; - BOOL is_retry; - unsigned char blocksize_bits; - - vi = page->mapping->host; - ni = NTFS_I(vi); - vol = ni->vol; - - ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index " - "0x%lx, from = %u, to = %u.", ni->mft_no, ni->type, - page->index, from, to); - - BUG_ON(!NInoNonResident(ni)); - - blocksize_bits = vi->i_blkbits; - blocksize = 1 << blocksize_bits; - - /* - * create_empty_buffers() will create uptodate/dirty buffers if the - * page is uptodate/dirty. - */ - if (!page_has_buffers(page)) - create_empty_buffers(page, blocksize, 0); - bh = head = page_buffers(page); - if (unlikely(!bh)) - return -ENOMEM; - - /* The first block in the page. */ - block = (s64)page->index << (PAGE_CACHE_SHIFT - blocksize_bits); - - read_lock_irqsave(&ni->size_lock, flags); - /* - * The first out of bounds block for the allocated size. No need to - * round up as allocated_size is in multiples of cluster size and the - * minimum cluster size is 512 bytes, which is equal to the smallest - * blocksize. - */ - ablock = ni->allocated_size >> blocksize_bits; - i_size = i_size_read(vi); - initialized_size = ni->initialized_size; - read_unlock_irqrestore(&ni->size_lock, flags); - - /* The last (fully or partially) initialized block. */ - iblock = initialized_size >> blocksize_bits; - - /* Loop through all the buffers in the page. */ - block_start = 0; - rl = NULL; - err = 0; - do { - block_end = block_start + blocksize; - /* - * If buffer @bh is outside the write, just mark it uptodate - * if the page is uptodate and continue with the next buffer. - */ - if (block_end <= from || block_start >= to) { - if (PageUptodate(page)) { - if (!buffer_uptodate(bh)) - set_buffer_uptodate(bh); - } - continue; - } - /* - * @bh is at least partially being written to. - * Make sure it is not marked as new. - */ - //if (buffer_new(bh)) - // clear_buffer_new(bh); - - if (block >= ablock) { - // TODO: block is above allocated_size, need to - // allocate it. Best done in one go to accommodate not - // only block but all above blocks up to and including: - // ((page->index << PAGE_CACHE_SHIFT) + to + blocksize - // - 1) >> blobksize_bits. Obviously will need to round - // up to next cluster boundary, too. This should be - // done with a helper function, so it can be reused. - ntfs_error(vol->sb, "Writing beyond allocated size " - "is not supported yet. Sorry."); - err = -EOPNOTSUPP; - goto err_out; - // Need to update ablock. - // Need to set_buffer_new() on all block bhs that are - // newly allocated. - } - /* - * Now we have enough allocated size to fulfill the whole - * request, i.e. block < ablock is true. - */ - if (unlikely((block >= iblock) && - (initialized_size < i_size))) { - /* - * If this page is fully outside initialized size, zero - * out all pages between the current initialized size - * and the current page. Just use ntfs_readpage() to do - * the zeroing transparently. - */ - if (block > iblock) { - // TODO: - // For each page do: - // - read_cache_page() - // Again for each page do: - // - wait_on_page_locked() - // - Check (PageUptodate(page) && - // !PageError(page)) - // Update initialized size in the attribute and - // in the inode. - // Again, for each page do: - // __set_page_dirty_buffers(); - // page_cache_release() - // We don't need to wait on the writes. - // Update iblock. - } - /* - * The current page straddles initialized size. Zero - * all non-uptodate buffers and set them uptodate (and - * dirty?). Note, there aren't any non-uptodate buffers - * if the page is uptodate. - * FIXME: For an uptodate page, the buffers may need to - * be written out because they were not initialized on - * disk before. - */ - if (!PageUptodate(page)) { - // TODO: - // Zero any non-uptodate buffers up to i_size. - // Set them uptodate and dirty. - } - // TODO: - // Update initialized size in the attribute and in the - // inode (up to i_size). - // Update iblock. - // FIXME: This is inefficient. Try to batch the two - // size changes to happen in one go. - ntfs_error(vol->sb, "Writing beyond initialized size " - "is not supported yet. Sorry."); - err = -EOPNOTSUPP; - goto err_out; - // Do NOT set_buffer_new() BUT DO clear buffer range - // outside write request range. - // set_buffer_uptodate() on complete buffers as well as - // set_buffer_dirty(). - } - - /* Need to map unmapped buffers. */ - if (!buffer_mapped(bh)) { - /* Unmapped buffer. Need to map it. */ - bh->b_bdev = vol->sb->s_bdev; - - /* Convert block into corresponding vcn and offset. */ - vcn = (VCN)block << blocksize_bits >> - vol->cluster_size_bits; - vcn_ofs = ((VCN)block << blocksize_bits) & - vol->cluster_size_mask; - - is_retry = FALSE; - if (!rl) { -lock_retry_remap: - down_read(&ni->runlist.lock); - rl = ni->runlist.rl; - } - if (likely(rl != NULL)) { - /* Seek to element containing target vcn. */ - while (rl->length && rl[1].vcn <= vcn) - rl++; - lcn = ntfs_rl_vcn_to_lcn(rl, vcn); - } else - lcn = LCN_RL_NOT_MAPPED; - if (unlikely(lcn < 0)) { - /* - * We extended the attribute allocation above. - * If we hit an ENOENT here it means that the - * allocation was insufficient which is a bug. - */ - BUG_ON(lcn == LCN_ENOENT); - - /* It is a hole, need to instantiate it. */ - if (lcn == LCN_HOLE) { - // TODO: Instantiate the hole. - // clear_buffer_new(bh); - // unmap_underlying_metadata(bh->b_bdev, - // bh->b_blocknr); - // For non-uptodate buffers, need to - // zero out the region outside the - // request in this bh or all bhs, - // depending on what we implemented - // above. - // Need to flush_dcache_page(). - // Or could use set_buffer_new() - // instead? - ntfs_error(vol->sb, "Writing into " - "sparse regions is " - "not supported yet. " - "Sorry."); - err = -EOPNOTSUPP; - if (!rl) - up_read(&ni->runlist.lock); - goto err_out; - } else if (!is_retry && - lcn == LCN_RL_NOT_MAPPED) { - is_retry = TRUE; - /* - * Attempt to map runlist, dropping - * lock for the duration. - */ - up_read(&ni->runlist.lock); - err = ntfs_map_runlist(ni, vcn); - if (likely(!err)) - goto lock_retry_remap; - rl = NULL; - } else if (!rl) - up_read(&ni->runlist.lock); - /* - * Failed to map the buffer, even after - * retrying. - */ - if (!err) - err = -EIO; - bh->b_blocknr = -1; - ntfs_error(vol->sb, "Failed to write to inode " - "0x%lx, attribute type 0x%x, " - "vcn 0x%llx, offset 0x%x " - "because its location on disk " - "could not be determined%s " - "(error code %i).", - ni->mft_no, ni->type, - (unsigned long long)vcn, - vcn_ofs, is_retry ? " even " - "after retrying" : "", err); - goto err_out; - } - /* We now have a successful remap, i.e. lcn >= 0. */ - - /* Setup buffer head to correct block. */ - bh->b_blocknr = ((lcn << vol->cluster_size_bits) - + vcn_ofs) >> blocksize_bits; - set_buffer_mapped(bh); - - // FIXME: Something analogous to this is needed for - // each newly allocated block, i.e. BH_New. - // FIXME: Might need to take this out of the - // if (!buffer_mapped(bh)) {}, depending on how we - // implement things during the allocated_size and - // initialized_size extension code above. - if (buffer_new(bh)) { - clear_buffer_new(bh); - unmap_underlying_metadata(bh->b_bdev, - bh->b_blocknr); - if (PageUptodate(page)) { - set_buffer_uptodate(bh); - continue; - } - /* - * Page is _not_ uptodate, zero surrounding - * region. NOTE: This is how we decide if to - * zero or not! - */ - if (block_end > to || block_start < from) { - void *kaddr; - - kaddr = kmap_atomic(page, KM_USER0); - if (block_end > to) - memset(kaddr + to, 0, - block_end - to); - if (block_start < from) - memset(kaddr + block_start, 0, - from - - block_start); - flush_dcache_page(page); - kunmap_atomic(kaddr, KM_USER0); - } - continue; - } - } - /* @bh is mapped, set it uptodate if the page is uptodate. */ - if (PageUptodate(page)) { - if (!buffer_uptodate(bh)) - set_buffer_uptodate(bh); - continue; - } - /* - * The page is not uptodate. The buffer is mapped. If it is not - * uptodate, and it is only partially being written to, we need - * to read the buffer in before the write, i.e. right now. - */ - if (!buffer_uptodate(bh) && - (block_start < from || block_end > to)) { - ll_rw_block(READ, 1, &bh); - *wait_bh++ = bh; - } - } while (block++, block_start = block_end, - (bh = bh->b_this_page) != head); - - /* Release the lock if we took it. */ - if (rl) { - up_read(&ni->runlist.lock); - rl = NULL; - } - - /* If we issued read requests, let them complete. */ - while (wait_bh > wait) { - wait_on_buffer(*--wait_bh); - if (!buffer_uptodate(*wait_bh)) - return -EIO; - } - - ntfs_debug("Done."); - return 0; -err_out: - /* - * Zero out any newly allocated blocks to avoid exposing stale data. - * If BH_New is set, we know that the block was newly allocated in the - * above loop. - * FIXME: What about initialized_size increments? Have we done all the - * required zeroing above? If not this error handling is broken, and - * in particular the if (block_end <= from) check is completely bogus. - */ - bh = head; - block_start = 0; - is_retry = FALSE; - do { - block_end = block_start + blocksize; - if (block_end <= from) - continue; - if (block_start >= to) - break; - if (buffer_new(bh)) { - void *kaddr; - - clear_buffer_new(bh); - kaddr = kmap_atomic(page, KM_USER0); - memset(kaddr + block_start, 0, bh->b_size); - kunmap_atomic(kaddr, KM_USER0); - set_buffer_uptodate(bh); - mark_buffer_dirty(bh); - is_retry = TRUE; - } - } while (block_start = block_end, (bh = bh->b_this_page) != head); - if (is_retry) - flush_dcache_page(page); - if (rl) - up_read(&ni->runlist.lock); - return err; -} - -/** - * ntfs_prepare_write - prepare a page for receiving data - * - * This is called from generic_file_write() with i_sem held on the inode - * (@page->mapping->host). The @page is locked but not kmap()ped. The source - * data has not yet been copied into the @page. - * - * Need to extend the attribute/fill in holes if necessary, create blocks and - * make partially overwritten blocks uptodate, - * - * i_size is not to be modified yet. - * - * Return 0 on success or -errno on error. - * - * Should be using block_prepare_write() [support for sparse files] or - * cont_prepare_write() [no support for sparse files]. Cannot do that due to - * ntfs specifics but can look at them for implementation guidance. - * - * Note: In the range, @from is inclusive and @to is exclusive, i.e. @from is - * the first byte in the page that will be written to and @to is the first byte - * after the last byte that will be written to. - */ -static int ntfs_prepare_write(struct file *file, struct page *page, - unsigned from, unsigned to) -{ - s64 new_size; - loff_t i_size; - struct inode *vi = page->mapping->host; - ntfs_inode *base_ni = NULL, *ni = NTFS_I(vi); - ntfs_volume *vol = ni->vol; - ntfs_attr_search_ctx *ctx = NULL; - MFT_RECORD *m = NULL; - ATTR_RECORD *a; - u8 *kaddr; - u32 attr_len; - int err; - - ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index " - "0x%lx, from = %u, to = %u.", vi->i_ino, ni->type, - page->index, from, to); - BUG_ON(!PageLocked(page)); - BUG_ON(from > PAGE_CACHE_SIZE); - BUG_ON(to > PAGE_CACHE_SIZE); - BUG_ON(from > to); - BUG_ON(NInoMstProtected(ni)); - /* - * If a previous ntfs_truncate() failed, repeat it and abort if it - * fails again. - */ - if (unlikely(NInoTruncateFailed(ni))) { - down_write(&vi->i_alloc_sem); - err = ntfs_truncate(vi); - up_write(&vi->i_alloc_sem); - if (err || NInoTruncateFailed(ni)) { - if (!err) - err = -EIO; - goto err_out; - } - } - /* If the attribute is not resident, deal with it elsewhere. */ - if (NInoNonResident(ni)) { - /* - * Only unnamed $DATA attributes can be compressed, encrypted, - * and/or sparse. - */ - if (ni->type == AT_DATA && !ni->name_len) { - /* If file is encrypted, deny access, just like NT4. */ - if (NInoEncrypted(ni)) { - ntfs_debug("Denying write access to encrypted " - "file."); - return -EACCES; - } - /* Compressed data streams are handled in compress.c. */ - if (NInoCompressed(ni)) { - // TODO: Implement and replace this check with - // return ntfs_write_compressed_block(page); - ntfs_error(vi->i_sb, "Writing to compressed " - "files is not supported yet. " - "Sorry."); - return -EOPNOTSUPP; - } - // TODO: Implement and remove this check. - if (NInoSparse(ni)) { - ntfs_error(vi->i_sb, "Writing to sparse files " - "is not supported yet. Sorry."); - return -EOPNOTSUPP; - } - } - /* Normal data stream. */ - return ntfs_prepare_nonresident_write(page, from, to); - } - /* - * Attribute is resident, implying it is not compressed, encrypted, or - * sparse. - */ - BUG_ON(page_has_buffers(page)); - new_size = ((s64)page->index << PAGE_CACHE_SHIFT) + to; - /* If we do not need to resize the attribute allocation we are done. */ - if (new_size <= i_size_read(vi)) - goto done; - /* Map, pin, and lock the (base) mft record. */ - if (!NInoAttr(ni)) - base_ni = ni; - else - base_ni = ni->ext.base_ntfs_ino; - m = map_mft_record(base_ni); - if (IS_ERR(m)) { - err = PTR_ERR(m); - m = NULL; - ctx = NULL; - goto err_out; - } - ctx = ntfs_attr_get_search_ctx(base_ni, m); - if (unlikely(!ctx)) { - err = -ENOMEM; - goto err_out; - } - err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, - CASE_SENSITIVE, 0, NULL, 0, ctx); - if (unlikely(err)) { - if (err == -ENOENT) - err = -EIO; - goto err_out; - } - m = ctx->mrec; - a = ctx->attr; - /* The total length of the attribute value. */ - attr_len = le32_to_cpu(a->data.resident.value_length); - /* Fix an eventual previous failure of ntfs_commit_write(). */ - i_size = i_size_read(vi); - if (unlikely(attr_len > i_size)) { - attr_len = i_size; - a->data.resident.value_length = cpu_to_le32(attr_len); - } - /* If we do not need to resize the attribute allocation we are done. */ - if (new_size <= attr_len) - goto done_unm; - /* Check if new size is allowed in $AttrDef. */ - err = ntfs_attr_size_bounds_check(vol, ni->type, new_size); - if (unlikely(err)) { - if (err == -ERANGE) { - ntfs_error(vol->sb, "Write would cause the inode " - "0x%lx to exceed the maximum size for " - "its attribute type (0x%x). Aborting " - "write.", vi->i_ino, - le32_to_cpu(ni->type)); - } else { - ntfs_error(vol->sb, "Inode 0x%lx has unknown " - "attribute type 0x%x. Aborting " - "write.", vi->i_ino, - le32_to_cpu(ni->type)); - err = -EIO; - } - goto err_out2; - } - /* - * Extend the attribute record to be able to store the new attribute - * size. - */ - if (new_size >= vol->mft_record_size || ntfs_attr_record_resize(m, a, - le16_to_cpu(a->data.resident.value_offset) + - new_size)) { - /* Not enough space in the mft record. */ - ntfs_error(vol->sb, "Not enough space in the mft record for " - "the resized attribute value. This is not " - "supported yet. Aborting write."); - err = -EOPNOTSUPP; - goto err_out2; - } - /* - * We have enough space in the mft record to fit the write. This - * implies the attribute is smaller than the mft record and hence the - * attribute must be in a single page and hence page->index must be 0. - */ - BUG_ON(page->index); - /* - * If the beginning of the write is past the old size, enlarge the - * attribute value up to the beginning of the write and fill it with - * zeroes. - */ - if (from > attr_len) { - memset((u8*)a + le16_to_cpu(a->data.resident.value_offset) + - attr_len, 0, from - attr_len); - a->data.resident.value_length = cpu_to_le32(from); - /* Zero the corresponding area in the page as well. */ - if (PageUptodate(page)) { - kaddr = kmap_atomic(page, KM_USER0); - memset(kaddr + attr_len, 0, from - attr_len); - kunmap_atomic(kaddr, KM_USER0); - flush_dcache_page(page); - } - } - flush_dcache_mft_record_page(ctx->ntfs_ino); - mark_mft_record_dirty(ctx->ntfs_ino); -done_unm: - ntfs_attr_put_search_ctx(ctx); - unmap_mft_record(base_ni); - /* - * Because resident attributes are handled by memcpy() to/from the - * corresponding MFT record, and because this form of i/o is byte - * aligned rather than block aligned, there is no need to bring the - * page uptodate here as in the non-resident case where we need to - * bring the buffers straddled by the write uptodate before - * generic_file_write() does the copying from userspace. - * - * We thus defer the uptodate bringing of the page region outside the - * region written to to ntfs_commit_write(), which makes the code - * simpler and saves one atomic kmap which is good. - */ -done: - ntfs_debug("Done."); - return 0; -err_out: - if (err == -ENOMEM) - ntfs_warning(vi->i_sb, "Error allocating memory required to " - "prepare the write."); - else { - ntfs_error(vi->i_sb, "Resident attribute prepare write failed " - "with error %i.", err); - NVolSetErrors(vol); - make_bad_inode(vi); - } -err_out2: - if (ctx) - ntfs_attr_put_search_ctx(ctx); - if (m) - unmap_mft_record(base_ni); - return err; -} - -/** - * ntfs_commit_nonresident_write - - * - */ -static int ntfs_commit_nonresident_write(struct page *page, - unsigned from, unsigned to) -{ - s64 pos = ((s64)page->index << PAGE_CACHE_SHIFT) + to; - struct inode *vi = page->mapping->host; - struct buffer_head *bh, *head; - unsigned int block_start, block_end, blocksize; - BOOL partial; - - ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index " - "0x%lx, from = %u, to = %u.", vi->i_ino, - NTFS_I(vi)->type, page->index, from, to); - blocksize = 1 << vi->i_blkbits; - - // FIXME: We need a whole slew of special cases in here for compressed - // files for example... - // For now, we know ntfs_prepare_write() would have failed so we can't - // get here in any of the cases which we have to special case, so we - // are just a ripped off, unrolled generic_commit_write(). - - bh = head = page_buffers(page); - block_start = 0; - partial = FALSE; - do { - block_end = block_start + blocksize; - if (block_end <= from || block_start >= to) { - if (!buffer_uptodate(bh)) - partial = TRUE; - } else { - set_buffer_uptodate(bh); - mark_buffer_dirty(bh); - } - } while (block_start = block_end, (bh = bh->b_this_page) != head); - /* - * If this is a partial write which happened to make all buffers - * uptodate then we can optimize away a bogus ->readpage() for the next - * read(). Here we 'discover' whether the page went uptodate as a - * result of this (potentially partial) write. - */ - if (!partial) - SetPageUptodate(page); - /* - * Not convinced about this at all. See disparity comment above. For - * now we know ntfs_prepare_write() would have failed in the write - * exceeds i_size case, so this will never trigger which is fine. - */ - if (pos > i_size_read(vi)) { - ntfs_error(vi->i_sb, "Writing beyond the existing file size is " - "not supported yet. Sorry."); - return -EOPNOTSUPP; - // vi->i_size = pos; - // mark_inode_dirty(vi); - } - ntfs_debug("Done."); - return 0; -} - -/** - * ntfs_commit_write - commit the received data - * - * This is called from generic_file_write() with i_sem held on the inode - * (@page->mapping->host). The @page is locked but not kmap()ped. The source - * data has already been copied into the @page. ntfs_prepare_write() has been - * called before the data copied and it returned success so we can take the - * results of various BUG checks and some error handling for granted. - * - * Need to mark modified blocks dirty so they get written out later when - * ntfs_writepage() is invoked by the VM. - * - * Return 0 on success or -errno on error. - * - * Should be using generic_commit_write(). This marks buffers uptodate and - * dirty, sets the page uptodate if all buffers in the page are uptodate, and - * updates i_size if the end of io is beyond i_size. In that case, it also - * marks the inode dirty. - * - * Cannot use generic_commit_write() due to ntfs specialities but can look at - * it for implementation guidance. - * - * If things have gone as outlined in ntfs_prepare_write(), then we do not - * need to do any page content modifications here at all, except in the write - * to resident attribute case, where we need to do the uptodate bringing here - * which we combine with the copying into the mft record which means we save - * one atomic kmap. - */ -static int ntfs_commit_write(struct file *file, struct page *page, - unsigned from, unsigned to) -{ - struct inode *vi = page->mapping->host; - ntfs_inode *base_ni, *ni = NTFS_I(vi); - char *kaddr, *kattr; - ntfs_attr_search_ctx *ctx; - MFT_RECORD *m; - ATTR_RECORD *a; - u32 attr_len; - int err; - - ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index " - "0x%lx, from = %u, to = %u.", vi->i_ino, ni->type, - page->index, from, to); - /* If the attribute is not resident, deal with it elsewhere. */ - if (NInoNonResident(ni)) { - /* Only unnamed $DATA attributes can be compressed/encrypted. */ - if (ni->type == AT_DATA && !ni->name_len) { - /* Encrypted files need separate handling. */ - if (NInoEncrypted(ni)) { - // We never get here at present! - BUG(); - } - /* Compressed data streams are handled in compress.c. */ - if (NInoCompressed(ni)) { - // TODO: Implement this! - // return ntfs_write_compressed_block(page); - // We never get here at present! - BUG(); - } - } - /* Normal data stream. */ - return ntfs_commit_nonresident_write(page, from, to); - } - /* - * Attribute is resident, implying it is not compressed, encrypted, or - * sparse. - */ - if (!NInoAttr(ni)) - base_ni = ni; - else - base_ni = ni->ext.base_ntfs_ino; - /* Map, pin, and lock the mft record. */ - m = map_mft_record(base_ni); - if (IS_ERR(m)) { - err = PTR_ERR(m); - m = NULL; - ctx = NULL; - goto err_out; - } - ctx = ntfs_attr_get_search_ctx(base_ni, m); - if (unlikely(!ctx)) { - err = -ENOMEM; - goto err_out; - } - err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, - CASE_SENSITIVE, 0, NULL, 0, ctx); - if (unlikely(err)) { - if (err == -ENOENT) - err = -EIO; - goto err_out; - } - a = ctx->attr; - /* The total length of the attribute value. */ - attr_len = le32_to_cpu(a->data.resident.value_length); - BUG_ON(from > attr_len); - kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset); - kaddr = kmap_atomic(page, KM_USER0); - /* Copy the received data from the page to the mft record. */ - memcpy(kattr + from, kaddr + from, to - from); - /* Update the attribute length if necessary. */ - if (to > attr_len) { - attr_len = to; - a->data.resident.value_length = cpu_to_le32(attr_len); - } - /* - * If the page is not uptodate, bring the out of bounds area(s) - * uptodate by copying data from the mft record to the page. - */ - if (!PageUptodate(page)) { - if (from > 0) - memcpy(kaddr, kattr, from); - if (to < attr_len) - memcpy(kaddr + to, kattr + to, attr_len - to); - /* Zero the region outside the end of the attribute value. */ - if (attr_len < PAGE_CACHE_SIZE) - memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len); - /* - * The probability of not having done any of the above is - * extremely small, so we just flush unconditionally. - */ - flush_dcache_page(page); - SetPageUptodate(page); - } - kunmap_atomic(kaddr, KM_USER0); - /* Update i_size if necessary. */ - if (i_size_read(vi) < attr_len) { - unsigned long flags; - - write_lock_irqsave(&ni->size_lock, flags); - ni->allocated_size = ni->initialized_size = attr_len; - i_size_write(vi, attr_len); - write_unlock_irqrestore(&ni->size_lock, flags); - } - /* Mark the mft record dirty, so it gets written back. */ - flush_dcache_mft_record_page(ctx->ntfs_ino); - mark_mft_record_dirty(ctx->ntfs_ino); - ntfs_attr_put_search_ctx(ctx); - unmap_mft_record(base_ni); - ntfs_debug("Done."); - return 0; -err_out: - if (err == -ENOMEM) { - ntfs_warning(vi->i_sb, "Error allocating memory required to " - "commit the write."); - if (PageUptodate(page)) { - ntfs_warning(vi->i_sb, "Page is uptodate, setting " - "dirty so the write will be retried " - "later on by the VM."); - /* - * Put the page on mapping->dirty_pages, but leave its - * buffers' dirty state as-is. - */ - __set_page_dirty_nobuffers(page); - err = 0; - } else - ntfs_error(vi->i_sb, "Page is not uptodate. Written " - "data has been lost."); - } else { - ntfs_error(vi->i_sb, "Resident attribute commit write failed " - "with error %i.", err); - NVolSetErrors(ni->vol); - make_bad_inode(vi); - } - if (ctx) - ntfs_attr_put_search_ctx(ctx); - if (m) - unmap_mft_record(base_ni); - return err; -} - #endif /* NTFS_RW */ /** @@ -2377,9 +1552,6 @@ struct address_space_operations ntfs_aop disk request queue. */ #ifdef NTFS_RW .writepage = ntfs_writepage, /* Write dirty page to disk. */ - .prepare_write = ntfs_prepare_write, /* Prepare page and buffers - ready to receive data. */ - .commit_write = ntfs_commit_write, /* Commit received data. */ #endif /* NTFS_RW */ }; diff --git a/fs/ntfs/attrib.c b/fs/ntfs/attrib.c --- a/fs/ntfs/attrib.c +++ b/fs/ntfs/attrib.c @@ -21,7 +21,9 @@ */ #include +#include #include +#include #include "attrib.h" #include "debug.h" @@ -36,9 +38,27 @@ * ntfs_map_runlist_nolock - map (a part of) a runlist of an ntfs inode * @ni: ntfs inode for which to map (part of) a runlist * @vcn: map runlist part containing this vcn + * @ctx: active attribute search context if present or NULL if not * * Map the part of a runlist containing the @vcn of the ntfs inode @ni. * + * If @ctx is specified, it is an active search context of @ni and its base mft + * record. This is needed when ntfs_map_runlist_nolock() encounters unmapped + * runlist fragments and allows their mapping. If you do not have the mft + * record mapped, you can specify @ctx as NULL and ntfs_map_runlist_nolock() + * will perform the necessary mapping and unmapping. + * + * Note, ntfs_map_runlist_nolock() saves the state of @ctx on entry and + * restores it before returning. Thus, @ctx will be left pointing to the same + * attribute on return as on entry. However, the actual pointers in @ctx may + * point to different memory locations on return, so you must remember to reset + * any cached pointers from the @ctx, i.e. after the call to + * ntfs_map_runlist_nolock(), you will probably want to do: + * m = ctx->mrec; + * a = ctx->attr; + * Assuming you cache ctx->attr in a variable @a of type ATTR_RECORD * and that + * you cache ctx->mrec in a variable @m of type MFT_RECORD *. + * * Return 0 on success and -errno on error. There is one special error code * which is not an error as such. This is -ENOENT. It means that @vcn is out * of bounds of the runlist. @@ -46,19 +66,32 @@ * Note the runlist can be NULL after this function returns if @vcn is zero and * the attribute has zero allocated size, i.e. there simply is no runlist. * - * Locking: - The runlist must be locked for writing. - * - This function modifies the runlist. + * WARNING: If @ctx is supplied, regardless of whether success or failure is + * returned, you need to check IS_ERR(@ctx->mrec) and if TRUE the @ctx + * is no longer valid, i.e. you need to either call + * ntfs_attr_reinit_search_ctx() or ntfs_attr_put_search_ctx() on it. + * In that case PTR_ERR(@ctx->mrec) will give you the error code for + * why the mapping of the old inode failed. + * + * Locking: - The runlist described by @ni must be locked for writing on entry + * and is locked on return. Note the runlist will be modified. + * - If @ctx is NULL, the base mft record of @ni must not be mapped on + * entry and it will be left unmapped on return. + * - If @ctx is not NULL, the base mft record must be mapped on entry + * and it will be left mapped on return. */ -int ntfs_map_runlist_nolock(ntfs_inode *ni, VCN vcn) +int ntfs_map_runlist_nolock(ntfs_inode *ni, VCN vcn, ntfs_attr_search_ctx *ctx) { VCN end_vcn; + unsigned long flags; ntfs_inode *base_ni; MFT_RECORD *m; ATTR_RECORD *a; - ntfs_attr_search_ctx *ctx; runlist_element *rl; - unsigned long flags; + struct page *put_this_page = NULL; int err = 0; + BOOL ctx_is_temporary, ctx_needs_reset; + ntfs_attr_search_ctx old_ctx; ntfs_debug("Mapping runlist part containing vcn 0x%llx.", (unsigned long long)vcn); @@ -66,20 +99,77 @@ int ntfs_map_runlist_nolock(ntfs_inode * base_ni = ni; else base_ni = ni->ext.base_ntfs_ino; - m = map_mft_record(base_ni); - if (IS_ERR(m)) - return PTR_ERR(m); - ctx = ntfs_attr_get_search_ctx(base_ni, m); - if (unlikely(!ctx)) { - err = -ENOMEM; - goto err_out; + if (!ctx) { + ctx_is_temporary = ctx_needs_reset = TRUE; + m = map_mft_record(base_ni); + if (IS_ERR(m)) + return PTR_ERR(m); + ctx = ntfs_attr_get_search_ctx(base_ni, m); + if (unlikely(!ctx)) { + err = -ENOMEM; + goto err_out; + } + } else { + VCN allocated_size_vcn; + + BUG_ON(IS_ERR(ctx->mrec)); + a = ctx->attr; + BUG_ON(!a->non_resident); + ctx_is_temporary = FALSE; + end_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn); + read_lock_irqsave(&ni->size_lock, flags); + allocated_size_vcn = ni->allocated_size >> + ni->vol->cluster_size_bits; + read_unlock_irqrestore(&ni->size_lock, flags); + if (!a->data.non_resident.lowest_vcn && end_vcn <= 0) + end_vcn = allocated_size_vcn - 1; + /* + * If we already have the attribute extent containing @vcn in + * @ctx, no need to look it up again. We slightly cheat in + * that if vcn exceeds the allocated size, we will refuse to + * map the runlist below, so there is definitely no need to get + * the right attribute extent. + */ + if (vcn >= allocated_size_vcn || (a->type == ni->type && + a->name_length == ni->name_len && + !memcmp((u8*)a + le16_to_cpu(a->name_offset), + ni->name, ni->name_len) && + sle64_to_cpu(a->data.non_resident.lowest_vcn) + <= vcn && end_vcn >= vcn)) + ctx_needs_reset = FALSE; + else { + /* Save the old search context. */ + old_ctx = *ctx; + /* + * If the currently mapped (extent) inode is not the + * base inode we will unmap it when we reinitialize the + * search context which means we need to get a + * reference to the page containing the mapped mft + * record so we do not accidentally drop changes to the + * mft record when it has not been marked dirty yet. + */ + if (old_ctx.base_ntfs_ino && old_ctx.ntfs_ino != + old_ctx.base_ntfs_ino) { + put_this_page = old_ctx.ntfs_ino->page; + page_cache_get(put_this_page); + } + /* + * Reinitialize the search context so we can lookup the + * needed attribute extent. + */ + ntfs_attr_reinit_search_ctx(ctx); + ctx_needs_reset = TRUE; + } } - err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, - CASE_SENSITIVE, vcn, NULL, 0, ctx); - if (unlikely(err)) { - if (err == -ENOENT) - err = -EIO; - goto err_out; + if (ctx_needs_reset) { + err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, vcn, NULL, 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) + err = -EIO; + goto err_out; + } + BUG_ON(!ctx->attr->non_resident); } a = ctx->attr; /* @@ -89,11 +179,9 @@ int ntfs_map_runlist_nolock(ntfs_inode * * ntfs_mapping_pairs_decompress() fails. */ end_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn) + 1; - if (unlikely(!a->data.non_resident.lowest_vcn && end_vcn <= 1)) { - read_lock_irqsave(&ni->size_lock, flags); - end_vcn = ni->allocated_size >> ni->vol->cluster_size_bits; - read_unlock_irqrestore(&ni->size_lock, flags); - } + if (!a->data.non_resident.lowest_vcn && end_vcn == 1) + end_vcn = sle64_to_cpu(a->data.non_resident.allocated_size) >> + ni->vol->cluster_size_bits; if (unlikely(vcn >= end_vcn)) { err = -ENOENT; goto err_out; @@ -104,9 +192,93 @@ int ntfs_map_runlist_nolock(ntfs_inode * else ni->runlist.rl = rl; err_out: - if (likely(ctx)) - ntfs_attr_put_search_ctx(ctx); - unmap_mft_record(base_ni); + if (ctx_is_temporary) { + if (likely(ctx)) + ntfs_attr_put_search_ctx(ctx); + unmap_mft_record(base_ni); + } else if (ctx_needs_reset) { + /* + * If there is no attribute list, restoring the search context + * is acomplished simply by copying the saved context back over + * the caller supplied context. If there is an attribute list, + * things are more complicated as we need to deal with mapping + * of mft records and resulting potential changes in pointers. + */ + if (NInoAttrList(base_ni)) { + /* + * If the currently mapped (extent) inode is not the + * one we had before, we need to unmap it and map the + * old one. + */ + if (ctx->ntfs_ino != old_ctx.ntfs_ino) { + /* + * If the currently mapped inode is not the + * base inode, unmap it. + */ + if (ctx->base_ntfs_ino && ctx->ntfs_ino != + ctx->base_ntfs_ino) { + unmap_extent_mft_record(ctx->ntfs_ino); + ctx->mrec = ctx->base_mrec; + BUG_ON(!ctx->mrec); + } + /* + * If the old mapped inode is not the base + * inode, map it. + */ + if (old_ctx.base_ntfs_ino && + old_ctx.ntfs_ino != + old_ctx.base_ntfs_ino) { +retry_map: + ctx->mrec = map_mft_record( + old_ctx.ntfs_ino); + /* + * Something bad has happened. If out + * of memory retry till it succeeds. + * Any other errors are fatal and we + * return the error code in ctx->mrec. + * Let the caller deal with it... We + * just need to fudge things so the + * caller can reinit and/or put the + * search context safely. + */ + if (IS_ERR(ctx->mrec)) { + if (PTR_ERR(ctx->mrec) == + -ENOMEM) { + schedule(); + goto retry_map; + } else + old_ctx.ntfs_ino = + old_ctx. + base_ntfs_ino; + } + } + } + /* Update the changed pointers in the saved context. */ + if (ctx->mrec != old_ctx.mrec) { + if (!IS_ERR(ctx->mrec)) + old_ctx.attr = (ATTR_RECORD*)( + (u8*)ctx->mrec + + ((u8*)old_ctx.attr - + (u8*)old_ctx.mrec)); + old_ctx.mrec = ctx->mrec; + } + } + /* Restore the search context to the saved one. */ + *ctx = old_ctx; + /* + * We drop the reference on the page we took earlier. In the + * case that IS_ERR(ctx->mrec) is true this means we might lose + * some changes to the mft record that had been made between + * the last time it was marked dirty/written out and now. This + * at this stage is not a problem as the mapping error is fatal + * enough that the mft record cannot be written out anyway and + * the caller is very likely to shutdown the whole inode + * immediately and mark the volume dirty for chkdsk to pick up + * the pieces anyway. + */ + if (put_this_page) + page_cache_release(put_this_page); + } return err; } @@ -122,8 +294,8 @@ err_out: * of bounds of the runlist. * * Locking: - The runlist must be unlocked on entry and is unlocked on return. - * - This function takes the runlist lock for writing and modifies the - * runlist. + * - This function takes the runlist lock for writing and may modify + * the runlist. */ int ntfs_map_runlist(ntfs_inode *ni, VCN vcn) { @@ -133,7 +305,7 @@ int ntfs_map_runlist(ntfs_inode *ni, VCN /* Make sure someone else didn't do the work while we were sleeping. */ if (likely(ntfs_rl_vcn_to_lcn(ni->runlist.rl, vcn) <= LCN_RL_NOT_MAPPED)) - err = ntfs_map_runlist_nolock(ni, vcn); + err = ntfs_map_runlist_nolock(ni, vcn, NULL); up_write(&ni->runlist.lock); return err; } @@ -212,7 +384,7 @@ retry_remap: goto retry_remap; } } - err = ntfs_map_runlist_nolock(ni, vcn); + err = ntfs_map_runlist_nolock(ni, vcn, NULL); if (!write_locked) { up_write(&ni->runlist.lock); down_read(&ni->runlist.lock); @@ -236,9 +408,9 @@ retry_remap: /** * ntfs_attr_find_vcn_nolock - find a vcn in the runlist of an ntfs inode - * @ni: ntfs inode describing the runlist to search - * @vcn: vcn to find - * @write_locked: true if the runlist is locked for writing + * @ni: ntfs inode describing the runlist to search + * @vcn: vcn to find + * @ctx: active attribute search context if present or NULL if not * * Find the virtual cluster number @vcn in the runlist described by the ntfs * inode @ni and return the address of the runlist element containing the @vcn. @@ -246,9 +418,22 @@ retry_remap: * If the @vcn is not mapped yet, the attempt is made to map the attribute * extent containing the @vcn and the vcn to lcn conversion is retried. * - * If @write_locked is true the caller has locked the runlist for writing and - * if false for reading. - * + * If @ctx is specified, it is an active search context of @ni and its base mft + * record. This is needed when ntfs_attr_find_vcn_nolock() encounters unmapped + * runlist fragments and allows their mapping. If you do not have the mft + * record mapped, you can specify @ctx as NULL and ntfs_attr_find_vcn_nolock() + * will perform the necessary mapping and unmapping. + * + * Note, ntfs_attr_find_vcn_nolock() saves the state of @ctx on entry and + * restores it before returning. Thus, @ctx will be left pointing to the same + * attribute on return as on entry. However, the actual pointers in @ctx may + * point to different memory locations on return, so you must remember to reset + * any cached pointers from the @ctx, i.e. after the call to + * ntfs_attr_find_vcn_nolock(), you will probably want to do: + * m = ctx->mrec; + * a = ctx->attr; + * Assuming you cache ctx->attr in a variable @a of type ATTR_RECORD * and that + * you cache ctx->mrec in a variable @m of type MFT_RECORD *. * Note you need to distinguish between the lcn of the returned runlist element * being >= 0 and LCN_HOLE. In the later case you have to return zeroes on * read and allocate clusters on write. @@ -263,22 +448,31 @@ retry_remap: * -ENOMEM - Not enough memory to map runlist. * -EIO - Critical error (runlist/file is corrupt, i/o error, etc). * - * Locking: - The runlist must be locked on entry and is left locked on return. - * - If @write_locked is FALSE, i.e. the runlist is locked for reading, - * the lock may be dropped inside the function so you cannot rely on - * the runlist still being the same when this function returns. + * WARNING: If @ctx is supplied, regardless of whether success or failure is + * returned, you need to check IS_ERR(@ctx->mrec) and if TRUE the @ctx + * is no longer valid, i.e. you need to either call + * ntfs_attr_reinit_search_ctx() or ntfs_attr_put_search_ctx() on it. + * In that case PTR_ERR(@ctx->mrec) will give you the error code for + * why the mapping of the old inode failed. + * + * Locking: - The runlist described by @ni must be locked for writing on entry + * and is locked on return. Note the runlist may be modified when + * needed runlist fragments need to be mapped. + * - If @ctx is NULL, the base mft record of @ni must not be mapped on + * entry and it will be left unmapped on return. + * - If @ctx is not NULL, the base mft record must be mapped on entry + * and it will be left mapped on return. */ runlist_element *ntfs_attr_find_vcn_nolock(ntfs_inode *ni, const VCN vcn, - const BOOL write_locked) + ntfs_attr_search_ctx *ctx) { unsigned long flags; runlist_element *rl; int err = 0; BOOL is_retry = FALSE; - ntfs_debug("Entering for i_ino 0x%lx, vcn 0x%llx, %s_locked.", - ni->mft_no, (unsigned long long)vcn, - write_locked ? "write" : "read"); + ntfs_debug("Entering for i_ino 0x%lx, vcn 0x%llx, with%s ctx.", + ni->mft_no, (unsigned long long)vcn, ctx ? "" : "out"); BUG_ON(!ni); BUG_ON(!NInoNonResident(ni)); BUG_ON(vcn < 0); @@ -312,33 +506,22 @@ retry_remap: } if (!err && !is_retry) { /* - * The @vcn is in an unmapped region, map the runlist and - * retry. + * If the search context is invalid we cannot map the unmapped + * region. */ - if (!write_locked) { - up_read(&ni->runlist.lock); - down_write(&ni->runlist.lock); - if (unlikely(ntfs_rl_vcn_to_lcn(ni->runlist.rl, vcn) != - LCN_RL_NOT_MAPPED)) { - up_write(&ni->runlist.lock); - down_read(&ni->runlist.lock); + if (IS_ERR(ctx->mrec)) + err = PTR_ERR(ctx->mrec); + else { + /* + * The @vcn is in an unmapped region, map the runlist + * and retry. + */ + err = ntfs_map_runlist_nolock(ni, vcn, ctx); + if (likely(!err)) { + is_retry = TRUE; goto retry_remap; } } - err = ntfs_map_runlist_nolock(ni, vcn); - if (!write_locked) { - up_write(&ni->runlist.lock); - down_read(&ni->runlist.lock); - } - if (likely(!err)) { - is_retry = TRUE; - goto retry_remap; - } - /* - * -EINVAL coming from a failed mapping attempt is equivalent - * to i/o error for us as it should not happen in our code - * paths. - */ if (err == -EINVAL) err = -EIO; } else if (!err) @@ -1011,6 +1194,7 @@ int ntfs_attr_lookup(const ATTR_TYPE typ ntfs_inode *base_ni; ntfs_debug("Entering."); + BUG_ON(IS_ERR(ctx->mrec)); if (ctx->base_ntfs_ino) base_ni = ctx->base_ntfs_ino; else @@ -1227,7 +1411,7 @@ int ntfs_attr_can_be_non_resident(const */ int ntfs_attr_can_be_resident(const ntfs_volume *vol, const ATTR_TYPE type) { - if (type == AT_INDEX_ALLOCATION || type == AT_EA) + if (type == AT_INDEX_ALLOCATION) return -EPERM; return 0; } @@ -1319,10 +1503,17 @@ int ntfs_resident_attr_value_resize(MFT_ /** * ntfs_attr_make_non_resident - convert a resident to a non-resident attribute * @ni: ntfs inode describing the attribute to convert + * @data_size: size of the resident data to copy to the non-resident attribute * * Convert the resident ntfs attribute described by the ntfs inode @ni to a * non-resident one. * + * @data_size must be equal to the attribute value size. This is needed since + * we need to know the size before we can map the mft record and our callers + * always know it. The reason we cannot simply read the size from the vfs + * inode i_size is that this is not necessarily uptodate. This happens when + * ntfs_attr_make_non_resident() is called in the ->truncate call path(s). + * * Return 0 on success and -errno on error. The following error return codes * are defined: * -EPERM - The attribute is not allowed to be non-resident. @@ -1343,7 +1534,7 @@ int ntfs_resident_attr_value_resize(MFT_ * * Locking: - The caller must hold i_sem on the inode. */ -int ntfs_attr_make_non_resident(ntfs_inode *ni) +int ntfs_attr_make_non_resident(ntfs_inode *ni, const u32 data_size) { s64 new_size; struct inode *vi = VFS_I(ni); @@ -1381,11 +1572,9 @@ int ntfs_attr_make_non_resident(ntfs_ino * The size needs to be aligned to a cluster boundary for allocation * purposes. */ - new_size = (i_size_read(vi) + vol->cluster_size - 1) & + new_size = (data_size + vol->cluster_size - 1) & ~(vol->cluster_size - 1); if (new_size > 0) { - runlist_element *rl2; - /* * Will need the page later and since the page lock nests * outside all ntfs locks, we need to get the page now. @@ -1396,7 +1585,7 @@ int ntfs_attr_make_non_resident(ntfs_ino return -ENOMEM; /* Start by allocating clusters to hold the attribute value. */ rl = ntfs_cluster_alloc(vol, 0, new_size >> - vol->cluster_size_bits, -1, DATA_ZONE); + vol->cluster_size_bits, -1, DATA_ZONE, TRUE); if (IS_ERR(rl)) { err = PTR_ERR(rl); ntfs_debug("Failed to allocate cluster%s, error code " @@ -1405,12 +1594,6 @@ int ntfs_attr_make_non_resident(ntfs_ino err); goto page_err_out; } - /* Change the runlist terminator to LCN_ENOENT. */ - rl2 = rl; - while (rl2->length) - rl2++; - BUG_ON(rl2->lcn != LCN_RL_NOT_MAPPED); - rl2->lcn = LCN_ENOENT; } else { rl = NULL; page = NULL; @@ -1473,7 +1656,7 @@ int ntfs_attr_make_non_resident(ntfs_ino * attribute value. */ attr_size = le32_to_cpu(a->data.resident.value_length); - BUG_ON(attr_size != i_size_read(vi)); + BUG_ON(attr_size != data_size); if (page && !PageUptodate(page)) { kaddr = kmap_atomic(page, KM_USER0); memcpy(kaddr, (u8*)a + @@ -1538,7 +1721,9 @@ int ntfs_attr_make_non_resident(ntfs_ino ffs(ni->itype.compressed.block_size) - 1; ni->itype.compressed.block_clusters = 1U << a->data.non_resident.compression_unit; - } + vi->i_blocks = ni->itype.compressed.size >> 9; + } else + vi->i_blocks = ni->allocated_size >> 9; write_unlock_irqrestore(&ni->size_lock, flags); /* * This needs to be last since the address space operations ->readpage @@ -1652,6 +1837,640 @@ page_err_out: } /** + * ntfs_attr_extend_allocation - extend the allocated space of an attribute + * @ni: ntfs inode of the attribute whose allocation to extend + * @new_alloc_size: new size in bytes to which to extend the allocation to + * @new_data_size: new size in bytes to which to extend the data to + * @data_start: beginning of region which is required to be non-sparse + * + * Extend the allocated space of an attribute described by the ntfs inode @ni + * to @new_alloc_size bytes. If @data_start is -1, the whole extension may be + * implemented as a hole in the file (as long as both the volume and the ntfs + * inode @ni have sparse support enabled). If @data_start is >= 0, then the + * region between the old allocated size and @data_start - 1 may be made sparse + * but the regions between @data_start and @new_alloc_size must be backed by + * actual clusters. + * + * If @new_data_size is -1, it is ignored. If it is >= 0, then the data size + * of the attribute is extended to @new_data_size. Note that the i_size of the + * vfs inode is not updated. Only the data size in the base attribute record + * is updated. The caller has to update i_size separately if this is required. + * WARNING: It is a BUG() for @new_data_size to be smaller than the old data + * size as well as for @new_data_size to be greater than @new_alloc_size. + * + * For resident attributes this involves resizing the attribute record and if + * necessary moving it and/or other attributes into extent mft records and/or + * converting the attribute to a non-resident attribute which in turn involves + * extending the allocation of a non-resident attribute as described below. + * + * For non-resident attributes this involves allocating clusters in the data + * zone on the volume (except for regions that are being made sparse) and + * extending the run list to describe the allocated clusters as well as + * updating the mapping pairs array of the attribute. This in turn involves + * resizing the attribute record and if necessary moving it and/or other + * attributes into extent mft records and/or splitting the attribute record + * into multiple extent attribute records. + * + * Also, the attribute list attribute is updated if present and in some of the + * above cases (the ones where extent mft records/attributes come into play), + * an attribute list attribute is created if not already present. + * + * Return the new allocated size on success and -errno on error. In the case + * that an error is encountered but a partial extension at least up to + * @data_start (if present) is possible, the allocation is partially extended + * and this is returned. This means the caller must check the returned size to + * determine if the extension was partial. If @data_start is -1 then partial + * allocations are not performed. + * + * WARNING: Do not call ntfs_attr_extend_allocation() for $MFT/$DATA. + * + * Locking: This function takes the runlist lock of @ni for writing as well as + * locking the mft record of the base ntfs inode. These locks are maintained + * throughout execution of the function. These locks are required so that the + * attribute can be resized safely and so that it can for example be converted + * from resident to non-resident safely. + * + * TODO: At present attribute list attribute handling is not implemented. + * + * TODO: At present it is not safe to call this function for anything other + * than the $DATA attribute(s) of an uncompressed and unencrypted file. + */ +s64 ntfs_attr_extend_allocation(ntfs_inode *ni, s64 new_alloc_size, + const s64 new_data_size, const s64 data_start) +{ + VCN vcn; + s64 ll, allocated_size, start = data_start; + struct inode *vi = VFS_I(ni); + ntfs_volume *vol = ni->vol; + ntfs_inode *base_ni; + MFT_RECORD *m; + ATTR_RECORD *a; + ntfs_attr_search_ctx *ctx; + runlist_element *rl, *rl2; + unsigned long flags; + int err, mp_size; + u32 attr_len = 0; /* Silence stupid gcc warning. */ + BOOL mp_rebuilt; + +#ifdef NTFS_DEBUG + read_lock_irqsave(&ni->size_lock, flags); + allocated_size = ni->allocated_size; + read_unlock_irqrestore(&ni->size_lock, flags); + ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, " + "old_allocated_size 0x%llx, " + "new_allocated_size 0x%llx, new_data_size 0x%llx, " + "data_start 0x%llx.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), + (unsigned long long)allocated_size, + (unsigned long long)new_alloc_size, + (unsigned long long)new_data_size, + (unsigned long long)start); +#endif +retry_extend: + /* + * For non-resident attributes, @start and @new_size need to be aligned + * to cluster boundaries for allocation purposes. + */ + if (NInoNonResident(ni)) { + if (start > 0) + start &= ~(s64)vol->cluster_size_mask; + new_alloc_size = (new_alloc_size + vol->cluster_size - 1) & + ~(s64)vol->cluster_size_mask; + } + BUG_ON(new_data_size >= 0 && new_data_size > new_alloc_size); + /* Check if new size is allowed in $AttrDef. */ + err = ntfs_attr_size_bounds_check(vol, ni->type, new_alloc_size); + if (unlikely(err)) { + /* Only emit errors when the write will fail completely. */ + read_lock_irqsave(&ni->size_lock, flags); + allocated_size = ni->allocated_size; + read_unlock_irqrestore(&ni->size_lock, flags); + if (start < 0 || start >= allocated_size) { + if (err == -ERANGE) { + ntfs_error(vol->sb, "Cannot extend allocation " + "of inode 0x%lx, attribute " + "type 0x%x, because the new " + "allocation would exceed the " + "maximum allowed size for " + "this attribute type.", + vi->i_ino, (unsigned) + le32_to_cpu(ni->type)); + } else { + ntfs_error(vol->sb, "Cannot extend allocation " + "of inode 0x%lx, attribute " + "type 0x%x, because this " + "attribute type is not " + "defined on the NTFS volume. " + "Possible corruption! You " + "should run chkdsk!", + vi->i_ino, (unsigned) + le32_to_cpu(ni->type)); + } + } + /* Translate error code to be POSIX conformant for write(2). */ + if (err == -ERANGE) + err = -EFBIG; + else + err = -EIO; + return err; + } + if (!NInoAttr(ni)) + base_ni = ni; + else + base_ni = ni->ext.base_ntfs_ino; + /* + * We will be modifying both the runlist (if non-resident) and the mft + * record so lock them both down. + */ + down_write(&ni->runlist.lock); + m = map_mft_record(base_ni); + if (IS_ERR(m)) { + err = PTR_ERR(m); + m = NULL; + ctx = NULL; + goto err_out; + } + ctx = ntfs_attr_get_search_ctx(base_ni, m); + if (unlikely(!ctx)) { + err = -ENOMEM; + goto err_out; + } + read_lock_irqsave(&ni->size_lock, flags); + allocated_size = ni->allocated_size; + read_unlock_irqrestore(&ni->size_lock, flags); + /* + * If non-resident, seek to the last extent. If resident, there is + * only one extent, so seek to that. + */ + vcn = NInoNonResident(ni) ? allocated_size >> vol->cluster_size_bits : + 0; + /* + * Abort if someone did the work whilst we waited for the locks. If we + * just converted the attribute from resident to non-resident it is + * likely that exactly this has happened already. We cannot quite + * abort if we need to update the data size. + */ + if (unlikely(new_alloc_size <= allocated_size)) { + ntfs_debug("Allocated size already exceeds requested size."); + new_alloc_size = allocated_size; + if (new_data_size < 0) + goto done; + /* + * We want the first attribute extent so that we can update the + * data size. + */ + vcn = 0; + } + err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, vcn, NULL, 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) + err = -EIO; + goto err_out; + } + m = ctx->mrec; + a = ctx->attr; + /* Use goto to reduce indentation. */ + if (a->non_resident) + goto do_non_resident_extend; + BUG_ON(NInoNonResident(ni)); + /* The total length of the attribute value. */ + attr_len = le32_to_cpu(a->data.resident.value_length); + /* + * Extend the attribute record to be able to store the new attribute + * size. ntfs_attr_record_resize() will not do anything if the size is + * not changing. + */ + if (new_alloc_size < vol->mft_record_size && + !ntfs_attr_record_resize(m, a, + le16_to_cpu(a->data.resident.value_offset) + + new_alloc_size)) { + /* The resize succeeded! */ + write_lock_irqsave(&ni->size_lock, flags); + ni->allocated_size = le32_to_cpu(a->length) - + le16_to_cpu(a->data.resident.value_offset); + write_unlock_irqrestore(&ni->size_lock, flags); + if (new_data_size >= 0) { + BUG_ON(new_data_size < attr_len); + a->data.resident.value_length = + cpu_to_le32((u32)new_data_size); + } + goto flush_done; + } + /* + * We have to drop all the locks so we can call + * ntfs_attr_make_non_resident(). This could be optimised by try- + * locking the first page cache page and only if that fails dropping + * the locks, locking the page, and redoing all the locking and + * lookups. While this would be a huge optimisation, it is not worth + * it as this is definitely a slow code path. + */ + ntfs_attr_put_search_ctx(ctx); + unmap_mft_record(base_ni); + up_write(&ni->runlist.lock); + /* + * Not enough space in the mft record, try to make the attribute + * non-resident and if successful restart the extension process. + */ + err = ntfs_attr_make_non_resident(ni, attr_len); + if (likely(!err)) + goto retry_extend; + /* + * Could not make non-resident. If this is due to this not being + * permitted for this attribute type or there not being enough space, + * try to make other attributes non-resident. Otherwise fail. + */ + if (unlikely(err != -EPERM && err != -ENOSPC)) { + /* Only emit errors when the write will fail completely. */ + read_lock_irqsave(&ni->size_lock, flags); + allocated_size = ni->allocated_size; + read_unlock_irqrestore(&ni->size_lock, flags); + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Cannot extend allocation of " + "inode 0x%lx, attribute type 0x%x, " + "because the conversion from resident " + "to non-resident attribute failed " + "with error code %i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), err); + if (err != -ENOMEM) + err = -EIO; + goto conv_err_out; + } + /* TODO: Not implemented from here, abort. */ + read_lock_irqsave(&ni->size_lock, flags); + allocated_size = ni->allocated_size; + read_unlock_irqrestore(&ni->size_lock, flags); + if (start < 0 || start >= allocated_size) { + if (err == -ENOSPC) + ntfs_error(vol->sb, "Not enough space in the mft " + "record/on disk for the non-resident " + "attribute value. This case is not " + "implemented yet."); + else /* if (err == -EPERM) */ + ntfs_error(vol->sb, "This attribute type may not be " + "non-resident. This case is not " + "implemented yet."); + } + err = -EOPNOTSUPP; + goto conv_err_out; +#if 0 + // TODO: Attempt to make other attributes non-resident. + if (!err) + goto do_resident_extend; + /* + * Both the attribute list attribute and the standard information + * attribute must remain in the base inode. Thus, if this is one of + * these attributes, we have to try to move other attributes out into + * extent mft records instead. + */ + if (ni->type == AT_ATTRIBUTE_LIST || + ni->type == AT_STANDARD_INFORMATION) { + // TODO: Attempt to move other attributes into extent mft + // records. + err = -EOPNOTSUPP; + if (!err) + goto do_resident_extend; + goto err_out; + } + // TODO: Attempt to move this attribute to an extent mft record, but + // only if it is not already the only attribute in an mft record in + // which case there would be nothing to gain. + err = -EOPNOTSUPP; + if (!err) + goto do_resident_extend; + /* There is nothing we can do to make enough space. )-: */ + goto err_out; +#endif +do_non_resident_extend: + BUG_ON(!NInoNonResident(ni)); + if (new_alloc_size == allocated_size) { + BUG_ON(vcn); + goto alloc_done; + } + /* + * If the data starts after the end of the old allocation, this is a + * $DATA attribute and sparse attributes are enabled on the volume and + * for this inode, then create a sparse region between the old + * allocated size and the start of the data. Otherwise simply proceed + * with filling the whole space between the old allocated size and the + * new allocated size with clusters. + */ + if ((start >= 0 && start <= allocated_size) || ni->type != AT_DATA || + !NVolSparseEnabled(vol) || NInoSparseDisabled(ni)) + goto skip_sparse; + // TODO: This is not implemented yet. We just fill in with real + // clusters for now... + ntfs_debug("Inserting holes is not-implemented yet. Falling back to " + "allocating real clusters instead."); +skip_sparse: + rl = ni->runlist.rl; + if (likely(rl)) { + /* Seek to the end of the runlist. */ + while (rl->length) + rl++; + } + /* If this attribute extent is not mapped, map it now. */ + if (unlikely(!rl || rl->lcn == LCN_RL_NOT_MAPPED || + (rl->lcn == LCN_ENOENT && rl > ni->runlist.rl && + (rl-1)->lcn == LCN_RL_NOT_MAPPED))) { + if (!rl && !allocated_size) + goto first_alloc; + rl = ntfs_mapping_pairs_decompress(vol, a, ni->runlist.rl); + if (IS_ERR(rl)) { + err = PTR_ERR(rl); + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Cannot extend allocation " + "of inode 0x%lx, attribute " + "type 0x%x, because the " + "mapping of a runlist " + "fragment failed with error " + "code %i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), + err); + if (err != -ENOMEM) + err = -EIO; + goto err_out; + } + ni->runlist.rl = rl; + /* Seek to the end of the runlist. */ + while (rl->length) + rl++; + } + /* + * We now know the runlist of the last extent is mapped and @rl is at + * the end of the runlist. We want to begin allocating clusters + * starting at the last allocated cluster to reduce fragmentation. If + * there are no valid LCNs in the attribute we let the cluster + * allocator choose the starting cluster. + */ + /* If the last LCN is a hole or simillar seek back to last real LCN. */ + while (rl->lcn < 0 && rl > ni->runlist.rl) + rl--; +first_alloc: + // FIXME: Need to implement partial allocations so at least part of the + // write can be performed when start >= 0. (Needed for POSIX write(2) + // conformance.) + rl2 = ntfs_cluster_alloc(vol, allocated_size >> vol->cluster_size_bits, + (new_alloc_size - allocated_size) >> + vol->cluster_size_bits, (rl && (rl->lcn >= 0)) ? + rl->lcn + rl->length : -1, DATA_ZONE, TRUE); + if (IS_ERR(rl2)) { + err = PTR_ERR(rl2); + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Cannot extend allocation of " + "inode 0x%lx, attribute type 0x%x, " + "because the allocation of clusters " + "failed with error code %i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), err); + if (err != -ENOMEM && err != -ENOSPC) + err = -EIO; + goto err_out; + } + rl = ntfs_runlists_merge(ni->runlist.rl, rl2); + if (IS_ERR(rl)) { + err = PTR_ERR(rl); + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Cannot extend allocation of " + "inode 0x%lx, attribute type 0x%x, " + "because the runlist merge failed " + "with error code %i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), err); + if (err != -ENOMEM) + err = -EIO; + if (ntfs_cluster_free_from_rl(vol, rl2)) { + ntfs_error(vol->sb, "Failed to release allocated " + "cluster(s) in error code path. Run " + "chkdsk to recover the lost " + "cluster(s)."); + NVolSetErrors(vol); + } + ntfs_free(rl2); + goto err_out; + } + ni->runlist.rl = rl; + ntfs_debug("Allocated 0x%llx clusters.", (long long)(new_alloc_size - + allocated_size) >> vol->cluster_size_bits); + /* Find the runlist element with which the attribute extent starts. */ + ll = sle64_to_cpu(a->data.non_resident.lowest_vcn); + rl2 = ntfs_rl_find_vcn_nolock(rl, ll); + BUG_ON(!rl2); + BUG_ON(!rl2->length); + BUG_ON(rl2->lcn < LCN_HOLE); + mp_rebuilt = FALSE; + /* Get the size for the new mapping pairs array for this extent. */ + mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1); + if (unlikely(mp_size <= 0)) { + err = mp_size; + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Cannot extend allocation of " + "inode 0x%lx, attribute type 0x%x, " + "because determining the size for the " + "mapping pairs failed with error code " + "%i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), err); + err = -EIO; + goto undo_alloc; + } + /* Extend the attribute record to fit the bigger mapping pairs array. */ + attr_len = le32_to_cpu(a->length); + err = ntfs_attr_record_resize(m, a, mp_size + + le16_to_cpu(a->data.non_resident.mapping_pairs_offset)); + if (unlikely(err)) { + BUG_ON(err != -ENOSPC); + // TODO: Deal with this by moving this extent to a new mft + // record or by starting a new extent in a new mft record, + // possibly by extending this extent partially and filling it + // and creating a new extent for the remainder, or by making + // other attributes non-resident and/or by moving other + // attributes out of this mft record. + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Not enough space in the mft " + "record for the extended attribute " + "record. This case is not " + "implemented yet."); + err = -EOPNOTSUPP; + goto undo_alloc; + } + mp_rebuilt = TRUE; + /* Generate the mapping pairs array directly into the attr record. */ + err = ntfs_mapping_pairs_build(vol, (u8*)a + + le16_to_cpu(a->data.non_resident.mapping_pairs_offset), + mp_size, rl2, ll, -1, NULL); + if (unlikely(err)) { + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Cannot extend allocation of " + "inode 0x%lx, attribute type 0x%x, " + "because building the mapping pairs " + "failed with error code %i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), err); + err = -EIO; + goto undo_alloc; + } + /* Update the highest_vcn. */ + a->data.non_resident.highest_vcn = cpu_to_sle64((new_alloc_size >> + vol->cluster_size_bits) - 1); + /* + * We now have extended the allocated size of the attribute. Reflect + * this in the ntfs_inode structure and the attribute record. + */ + if (a->data.non_resident.lowest_vcn) { + /* + * We are not in the first attribute extent, switch to it, but + * first ensure the changes will make it to disk later. + */ + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + ntfs_attr_reinit_search_ctx(ctx); + err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, 0, NULL, 0, ctx); + if (unlikely(err)) + goto restore_undo_alloc; + /* @m is not used any more so no need to set it. */ + a = ctx->attr; + } + write_lock_irqsave(&ni->size_lock, flags); + ni->allocated_size = new_alloc_size; + a->data.non_resident.allocated_size = cpu_to_sle64(new_alloc_size); + /* + * FIXME: This would fail if @ni is a directory, $MFT, or an index, + * since those can have sparse/compressed set. For example can be + * set compressed even though it is not compressed itself and in that + * case the bit means that files are to be created compressed in the + * directory... At present this is ok as this code is only called for + * regular files, and only for their $DATA attribute(s). + * FIXME: The calculation is wrong if we created a hole above. For now + * it does not matter as we never create holes. + */ + if (NInoSparse(ni) || NInoCompressed(ni)) { + ni->itype.compressed.size += new_alloc_size - allocated_size; + a->data.non_resident.compressed_size = + cpu_to_sle64(ni->itype.compressed.size); + vi->i_blocks = ni->itype.compressed.size >> 9; + } else + vi->i_blocks = new_alloc_size >> 9; + write_unlock_irqrestore(&ni->size_lock, flags); +alloc_done: + if (new_data_size >= 0) { + BUG_ON(new_data_size < + sle64_to_cpu(a->data.non_resident.data_size)); + a->data.non_resident.data_size = cpu_to_sle64(new_data_size); + } +flush_done: + /* Ensure the changes make it to disk. */ + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); +done: + ntfs_attr_put_search_ctx(ctx); + unmap_mft_record(base_ni); + up_write(&ni->runlist.lock); + ntfs_debug("Done, new_allocated_size 0x%llx.", + (unsigned long long)new_alloc_size); + return new_alloc_size; +restore_undo_alloc: + if (start < 0 || start >= allocated_size) + ntfs_error(vol->sb, "Cannot complete extension of allocation " + "of inode 0x%lx, attribute type 0x%x, because " + "lookup of first attribute extent failed with " + "error code %i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), err); + if (err == -ENOENT) + err = -EIO; + ntfs_attr_reinit_search_ctx(ctx); + if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len, CASE_SENSITIVE, + allocated_size >> vol->cluster_size_bits, NULL, 0, + ctx)) { + ntfs_error(vol->sb, "Failed to find last attribute extent of " + "attribute in error code path. Run chkdsk to " + "recover."); + write_lock_irqsave(&ni->size_lock, flags); + ni->allocated_size = new_alloc_size; + /* + * FIXME: This would fail if @ni is a directory... See above. + * FIXME: The calculation is wrong if we created a hole above. + * For now it does not matter as we never create holes. + */ + if (NInoSparse(ni) || NInoCompressed(ni)) { + ni->itype.compressed.size += new_alloc_size - + allocated_size; + vi->i_blocks = ni->itype.compressed.size >> 9; + } else + vi->i_blocks = new_alloc_size >> 9; + write_unlock_irqrestore(&ni->size_lock, flags); + ntfs_attr_put_search_ctx(ctx); + unmap_mft_record(base_ni); + up_write(&ni->runlist.lock); + /* + * The only thing that is now wrong is the allocated size of the + * base attribute extent which chkdsk should be able to fix. + */ + NVolSetErrors(vol); + return err; + } + ctx->attr->data.non_resident.highest_vcn = cpu_to_sle64( + (allocated_size >> vol->cluster_size_bits) - 1); +undo_alloc: + ll = allocated_size >> vol->cluster_size_bits; + if (ntfs_cluster_free(ni, ll, -1, ctx) < 0) { + ntfs_error(vol->sb, "Failed to release allocated cluster(s) " + "in error code path. Run chkdsk to recover " + "the lost cluster(s)."); + NVolSetErrors(vol); + } + m = ctx->mrec; + a = ctx->attr; + /* + * If the runlist truncation fails and/or the search context is no + * longer valid, we cannot resize the attribute record or build the + * mapping pairs array thus we mark the inode bad so that no access to + * the freed clusters can happen. + */ + if (ntfs_rl_truncate_nolock(vol, &ni->runlist, ll) || IS_ERR(m)) { + ntfs_error(vol->sb, "Failed to %s in error code path. Run " + "chkdsk to recover.", IS_ERR(m) ? + "restore attribute search context" : + "truncate attribute runlist"); + make_bad_inode(vi); + make_bad_inode(VFS_I(base_ni)); + NVolSetErrors(vol); + } else if (mp_rebuilt) { + if (ntfs_attr_record_resize(m, a, attr_len)) { + ntfs_error(vol->sb, "Failed to restore attribute " + "record in error code path. Run " + "chkdsk to recover."); + make_bad_inode(vi); + make_bad_inode(VFS_I(base_ni)); + NVolSetErrors(vol); + } else /* if (success) */ { + if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( + a->data.non_resident. + mapping_pairs_offset), attr_len - + le16_to_cpu(a->data.non_resident. + mapping_pairs_offset), rl2, ll, -1, + NULL)) { + ntfs_error(vol->sb, "Failed to restore " + "mapping pairs array in error " + "code path. Run chkdsk to " + "recover."); + make_bad_inode(vi); + make_bad_inode(VFS_I(base_ni)); + NVolSetErrors(vol); + } + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + } + } +err_out: + if (ctx) + ntfs_attr_put_search_ctx(ctx); + if (m) + unmap_mft_record(base_ni); + up_write(&ni->runlist.lock); +conv_err_out: + ntfs_debug("Failed. Returning error code %i.", err); + return err; +} + +/** * ntfs_attr_set - fill (a part of) an attribute with a byte * @ni: ntfs inode describing the attribute to fill * @ofs: offset inside the attribute at which to start to fill @@ -1773,6 +2592,8 @@ int ntfs_attr_set(ntfs_inode *ni, const /* Finally unlock and release the page. */ unlock_page(page); page_cache_release(page); + balance_dirty_pages_ratelimited(mapping); + cond_resched(); } /* If there is a last partial page, need to do it the slow way. */ if (end_ofs) { diff --git a/fs/ntfs/attrib.h b/fs/ntfs/attrib.h --- a/fs/ntfs/attrib.h +++ b/fs/ntfs/attrib.h @@ -60,14 +60,15 @@ typedef struct { ATTR_RECORD *base_attr; } ntfs_attr_search_ctx; -extern int ntfs_map_runlist_nolock(ntfs_inode *ni, VCN vcn); +extern int ntfs_map_runlist_nolock(ntfs_inode *ni, VCN vcn, + ntfs_attr_search_ctx *ctx); extern int ntfs_map_runlist(ntfs_inode *ni, VCN vcn); extern LCN ntfs_attr_vcn_to_lcn_nolock(ntfs_inode *ni, const VCN vcn, const BOOL write_locked); extern runlist_element *ntfs_attr_find_vcn_nolock(ntfs_inode *ni, - const VCN vcn, const BOOL write_locked); + const VCN vcn, ntfs_attr_search_ctx *ctx); int ntfs_attr_lookup(const ATTR_TYPE type, const ntfschar *name, const u32 name_len, const IGNORE_CASE_BOOL ic, @@ -102,7 +103,10 @@ extern int ntfs_attr_record_resize(MFT_R extern int ntfs_resident_attr_value_resize(MFT_RECORD *m, ATTR_RECORD *a, const u32 new_size); -extern int ntfs_attr_make_non_resident(ntfs_inode *ni); +extern int ntfs_attr_make_non_resident(ntfs_inode *ni, const u32 data_size); + +extern s64 ntfs_attr_extend_allocation(ntfs_inode *ni, s64 new_alloc_size, + const s64 new_data_size, const s64 data_start); extern int ntfs_attr_set(ntfs_inode *ni, const s64 ofs, const s64 cnt, const u8 val); diff --git a/fs/ntfs/file.c b/fs/ntfs/file.c --- a/fs/ntfs/file.c +++ b/fs/ntfs/file.c @@ -19,11 +19,24 @@ * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ -#include #include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "attrib.h" +#include "bitmap.h" #include "inode.h" #include "debug.h" +#include "lcnalloc.h" +#include "malloc.h" +#include "mft.h" #include "ntfs.h" /** @@ -56,6 +69,2191 @@ static int ntfs_file_open(struct inode * #ifdef NTFS_RW /** + * ntfs_attr_extend_initialized - extend the initialized size of an attribute + * @ni: ntfs inode of the attribute to extend + * @new_init_size: requested new initialized size in bytes + * @cached_page: store any allocated but unused page here + * @lru_pvec: lru-buffering pagevec of the caller + * + * Extend the initialized size of an attribute described by the ntfs inode @ni + * to @new_init_size bytes. This involves zeroing any non-sparse space between + * the old initialized size and @new_init_size both in the page cache and on + * disk (if relevant complete pages are zeroed in the page cache then these may + * simply be marked dirty for later writeout). There is one caveat and that is + * that if any uptodate page cache pages between the old initialized size and + * the smaller of @new_init_size and the file size (vfs inode->i_size) are in + * memory, these need to be marked dirty without being zeroed since they could + * be non-zero due to mmap() based writes. + * + * As a side-effect, the file size (vfs inode->i_size) may be incremented as, + * in the resident attribute case, it is tied to the initialized size and, in + * the non-resident attribute case, it may not fall below the initialized size. + * + * Note that if the attribute is resident, we do not need to touch the page + * cache at all. This is because if the page cache page is not uptodate we + * bring it uptodate later, when doing the write to the mft record since we + * then already have the page mapped. And if the page is uptodate, the + * non-initialized region will already have been zeroed when the page was + * brought uptodate and the region may in fact already have been overwritten + * with new data via mmap() based writes, so we cannot just zero it. And since + * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped + * is unspecified, we choose not to do zeroing and thus we do not need to touch + * the page at all. For a more detailed explanation see ntfs_truncate() which + * is in fs/ntfs/inode.c. + * + * @cached_page and @lru_pvec are just optimisations for dealing with multiple + * pages. + * + * Return 0 on success and -errno on error. In the case that an error is + * encountered it is possible that the initialized size will already have been + * incremented some way towards @new_init_size but it is guaranteed that if + * this is the case, the necessary zeroing will also have happened and that all + * metadata is self-consistent. + * + * Locking: This function locks the mft record of the base ntfs inode and + * maintains the lock throughout execution of the function. This is required + * so that the initialized size of the attribute can be modified safely. + */ +static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size, + struct page **cached_page, struct pagevec *lru_pvec) +{ + s64 old_init_size; + loff_t old_i_size; + pgoff_t index, end_index; + unsigned long flags; + struct inode *vi = VFS_I(ni); + ntfs_inode *base_ni; + MFT_RECORD *m = NULL; + ATTR_RECORD *a; + ntfs_attr_search_ctx *ctx = NULL; + struct address_space *mapping; + struct page *page = NULL; + u8 *kattr; + int err; + u32 attr_len; + + read_lock_irqsave(&ni->size_lock, flags); + old_init_size = ni->initialized_size; + old_i_size = i_size_read(vi); + BUG_ON(new_init_size > ni->allocated_size); + read_unlock_irqrestore(&ni->size_lock, flags); + ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, " + "old_initialized_size 0x%llx, " + "new_initialized_size 0x%llx, i_size 0x%llx.", + vi->i_ino, (unsigned)le32_to_cpu(ni->type), + (unsigned long long)old_init_size, + (unsigned long long)new_init_size, old_i_size); + if (!NInoAttr(ni)) + base_ni = ni; + else + base_ni = ni->ext.base_ntfs_ino; + /* Use goto to reduce indentation and we need the label below anyway. */ + if (NInoNonResident(ni)) + goto do_non_resident_extend; + BUG_ON(old_init_size != old_i_size); + m = map_mft_record(base_ni); + if (IS_ERR(m)) { + err = PTR_ERR(m); + m = NULL; + goto err_out; + } + ctx = ntfs_attr_get_search_ctx(base_ni, m); + if (unlikely(!ctx)) { + err = -ENOMEM; + goto err_out; + } + err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, 0, NULL, 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) + err = -EIO; + goto err_out; + } + m = ctx->mrec; + a = ctx->attr; + BUG_ON(a->non_resident); + /* The total length of the attribute value. */ + attr_len = le32_to_cpu(a->data.resident.value_length); + BUG_ON(old_i_size != (loff_t)attr_len); + /* + * Do the zeroing in the mft record and update the attribute size in + * the mft record. + */ + kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset); + memset(kattr + attr_len, 0, new_init_size - attr_len); + a->data.resident.value_length = cpu_to_le32((u32)new_init_size); + /* Finally, update the sizes in the vfs and ntfs inodes. */ + write_lock_irqsave(&ni->size_lock, flags); + i_size_write(vi, new_init_size); + ni->initialized_size = new_init_size; + write_unlock_irqrestore(&ni->size_lock, flags); + goto done; +do_non_resident_extend: + /* + * If the new initialized size @new_init_size exceeds the current file + * size (vfs inode->i_size), we need to extend the file size to the + * new initialized size. + */ + if (new_init_size > old_i_size) { + m = map_mft_record(base_ni); + if (IS_ERR(m)) { + err = PTR_ERR(m); + m = NULL; + goto err_out; + } + ctx = ntfs_attr_get_search_ctx(base_ni, m); + if (unlikely(!ctx)) { + err = -ENOMEM; + goto err_out; + } + err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, 0, NULL, 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) + err = -EIO; + goto err_out; + } + m = ctx->mrec; + a = ctx->attr; + BUG_ON(!a->non_resident); + BUG_ON(old_i_size != (loff_t) + sle64_to_cpu(a->data.non_resident.data_size)); + a->data.non_resident.data_size = cpu_to_sle64(new_init_size); + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + /* Update the file size in the vfs inode. */ + i_size_write(vi, new_init_size); + ntfs_attr_put_search_ctx(ctx); + ctx = NULL; + unmap_mft_record(base_ni); + m = NULL; + } + mapping = vi->i_mapping; + index = old_init_size >> PAGE_CACHE_SHIFT; + end_index = (new_init_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + do { + /* + * Read the page. If the page is not present, this will zero + * the uninitialized regions for us. + */ + page = read_cache_page(mapping, index, + (filler_t*)mapping->a_ops->readpage, NULL); + if (IS_ERR(page)) { + err = PTR_ERR(page); + goto init_err_out; + } + wait_on_page_locked(page); + if (unlikely(!PageUptodate(page) || PageError(page))) { + page_cache_release(page); + err = -EIO; + goto init_err_out; + } + /* + * Update the initialized size in the ntfs inode. This is + * enough to make ntfs_writepage() work. + */ + write_lock_irqsave(&ni->size_lock, flags); + ni->initialized_size = (index + 1) << PAGE_CACHE_SHIFT; + if (ni->initialized_size > new_init_size) + ni->initialized_size = new_init_size; + write_unlock_irqrestore(&ni->size_lock, flags); + /* Set the page dirty so it gets written out. */ + set_page_dirty(page); + page_cache_release(page); + /* + * Play nice with the vm and the rest of the system. This is + * very much needed as we can potentially be modifying the + * initialised size from a very small value to a really huge + * value, e.g. + * f = open(somefile, O_TRUNC); + * truncate(f, 10GiB); + * seek(f, 10GiB); + * write(f, 1); + * And this would mean we would be marking dirty hundreds of + * thousands of pages or as in the above example more than + * two and a half million pages! + * + * TODO: For sparse pages could optimize this workload by using + * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This + * would be set in readpage for sparse pages and here we would + * not need to mark dirty any pages which have this bit set. + * The only caveat is that we have to clear the bit everywhere + * where we allocate any clusters that lie in the page or that + * contain the page. + * + * TODO: An even greater optimization would be for us to only + * call readpage() on pages which are not in sparse regions as + * determined from the runlist. This would greatly reduce the + * number of pages we read and make dirty in the case of sparse + * files. + */ + balance_dirty_pages_ratelimited(mapping); + cond_resched(); + } while (++index < end_index); + read_lock_irqsave(&ni->size_lock, flags); + BUG_ON(ni->initialized_size != new_init_size); + read_unlock_irqrestore(&ni->size_lock, flags); + /* Now bring in sync the initialized_size in the mft record. */ + m = map_mft_record(base_ni); + if (IS_ERR(m)) { + err = PTR_ERR(m); + m = NULL; + goto init_err_out; + } + ctx = ntfs_attr_get_search_ctx(base_ni, m); + if (unlikely(!ctx)) { + err = -ENOMEM; + goto init_err_out; + } + err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, 0, NULL, 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) + err = -EIO; + goto init_err_out; + } + m = ctx->mrec; + a = ctx->attr; + BUG_ON(!a->non_resident); + a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size); +done: + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + if (ctx) + ntfs_attr_put_search_ctx(ctx); + if (m) + unmap_mft_record(base_ni); + ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.", + (unsigned long long)new_init_size, i_size_read(vi)); + return 0; +init_err_out: + write_lock_irqsave(&ni->size_lock, flags); + ni->initialized_size = old_init_size; + write_unlock_irqrestore(&ni->size_lock, flags); +err_out: + if (ctx) + ntfs_attr_put_search_ctx(ctx); + if (m) + unmap_mft_record(base_ni); + ntfs_debug("Failed. Returning error code %i.", err); + return err; +} + +/** + * ntfs_fault_in_pages_readable - + * + * Fault a number of userspace pages into pagetables. + * + * Unlike include/linux/pagemap.h::fault_in_pages_readable(), this one copes + * with more than two userspace pages as well as handling the single page case + * elegantly. + * + * If you find this difficult to understand, then think of the while loop being + * the following code, except that we do without the integer variable ret: + * + * do { + * ret = __get_user(c, uaddr); + * uaddr += PAGE_SIZE; + * } while (!ret && uaddr < end); + * + * Note, the final __get_user() may well run out-of-bounds of the user buffer, + * but _not_ out-of-bounds of the page the user buffer belongs to, and since + * this is only a read and not a write, and since it is still in the same page, + * it should not matter and this makes the code much simpler. + */ +static inline void ntfs_fault_in_pages_readable(const char __user *uaddr, + int bytes) +{ + const char __user *end; + volatile char c; + + /* Set @end to the first byte outside the last page we care about. */ + end = (const char __user*)PAGE_ALIGN((ptrdiff_t __user)uaddr + bytes); + + while (!__get_user(c, uaddr) && (uaddr += PAGE_SIZE, uaddr < end)) + ; +} + +/** + * ntfs_fault_in_pages_readable_iovec - + * + * Same as ntfs_fault_in_pages_readable() but operates on an array of iovecs. + */ +static inline void ntfs_fault_in_pages_readable_iovec(const struct iovec *iov, + size_t iov_ofs, int bytes) +{ + do { + const char __user *buf; + unsigned len; + + buf = iov->iov_base + iov_ofs; + len = iov->iov_len - iov_ofs; + if (len > bytes) + len = bytes; + ntfs_fault_in_pages_readable(buf, len); + bytes -= len; + iov++; + iov_ofs = 0; + } while (bytes); +} + +/** + * __ntfs_grab_cache_pages - obtain a number of locked pages + * @mapping: address space mapping from which to obtain page cache pages + * @index: starting index in @mapping at which to begin obtaining pages + * @nr_pages: number of page cache pages to obtain + * @pages: array of pages in which to return the obtained page cache pages + * @cached_page: allocated but as yet unused page + * @lru_pvec: lru-buffering pagevec of caller + * + * Obtain @nr_pages locked page cache pages from the mapping @maping and + * starting at index @index. + * + * If a page is newly created, increment its refcount and add it to the + * caller's lru-buffering pagevec @lru_pvec. + * + * This is the same as mm/filemap.c::__grab_cache_page(), except that @nr_pages + * are obtained at once instead of just one page and that 0 is returned on + * success and -errno on error. + * + * Note, the page locks are obtained in ascending page index order. + */ +static inline int __ntfs_grab_cache_pages(struct address_space *mapping, + pgoff_t index, const unsigned nr_pages, struct page **pages, + struct page **cached_page, struct pagevec *lru_pvec) +{ + int err, nr; + + BUG_ON(!nr_pages); + err = nr = 0; + do { + pages[nr] = find_lock_page(mapping, index); + if (!pages[nr]) { + if (!*cached_page) { + *cached_page = page_cache_alloc(mapping); + if (unlikely(!*cached_page)) { + err = -ENOMEM; + goto err_out; + } + } + err = add_to_page_cache(*cached_page, mapping, index, + GFP_KERNEL); + if (unlikely(err)) { + if (err == -EEXIST) + continue; + goto err_out; + } + pages[nr] = *cached_page; + page_cache_get(*cached_page); + if (unlikely(!pagevec_add(lru_pvec, *cached_page))) + __pagevec_lru_add(lru_pvec); + *cached_page = NULL; + } + index++; + nr++; + } while (nr < nr_pages); +out: + return err; +err_out: + while (nr > 0) { + unlock_page(pages[--nr]); + page_cache_release(pages[nr]); + } + goto out; +} + +static inline int ntfs_submit_bh_for_read(struct buffer_head *bh) +{ + lock_buffer(bh); + get_bh(bh); + bh->b_end_io = end_buffer_read_sync; + return submit_bh(READ, bh); +} + +/** + * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data + * @pages: array of destination pages + * @nr_pages: number of pages in @pages + * @pos: byte position in file at which the write begins + * @bytes: number of bytes to be written + * + * This is called for non-resident attributes from ntfs_file_buffered_write() + * with i_sem held on the inode (@pages[0]->mapping->host). There are + * @nr_pages pages in @pages which are locked but not kmap()ped. The source + * data has not yet been copied into the @pages. + * + * Need to fill any holes with actual clusters, allocate buffers if necessary, + * ensure all the buffers are mapped, and bring uptodate any buffers that are + * only partially being written to. + * + * If @nr_pages is greater than one, we are guaranteed that the cluster size is + * greater than PAGE_CACHE_SIZE, that all pages in @pages are entirely inside + * the same cluster and that they are the entirety of that cluster, and that + * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole. + * + * i_size is not to be modified yet. + * + * Return 0 on success or -errno on error. + */ +static int ntfs_prepare_pages_for_non_resident_write(struct page **pages, + unsigned nr_pages, s64 pos, size_t bytes) +{ + VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend; + LCN lcn; + s64 bh_pos, vcn_len, end, initialized_size; + sector_t lcn_block; + struct page *page; + struct inode *vi; + ntfs_inode *ni, *base_ni = NULL; + ntfs_volume *vol; + runlist_element *rl, *rl2; + struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; + ntfs_attr_search_ctx *ctx = NULL; + MFT_RECORD *m = NULL; + ATTR_RECORD *a = NULL; + unsigned long flags; + u32 attr_rec_len = 0; + unsigned blocksize, u; + int err, mp_size; + BOOL rl_write_locked, was_hole, is_retry; + unsigned char blocksize_bits; + struct { + u8 runlist_merged:1; + u8 mft_attr_mapped:1; + u8 mp_rebuilt:1; + u8 attr_switched:1; + } status = { 0, 0, 0, 0 }; + + BUG_ON(!nr_pages); + BUG_ON(!pages); + BUG_ON(!*pages); + vi = pages[0]->mapping->host; + ni = NTFS_I(vi); + vol = ni->vol; + ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page " + "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%x.", + vi->i_ino, ni->type, pages[0]->index, nr_pages, + (long long)pos, bytes); + blocksize_bits = vi->i_blkbits; + blocksize = 1 << blocksize_bits; + u = 0; + do { + struct page *page = pages[u]; + /* + * create_empty_buffers() will create uptodate/dirty buffers if + * the page is uptodate/dirty. + */ + if (!page_has_buffers(page)) { + create_empty_buffers(page, blocksize, 0); + if (unlikely(!page_has_buffers(page))) + return -ENOMEM; + } + } while (++u < nr_pages); + rl_write_locked = FALSE; + rl = NULL; + err = 0; + vcn = lcn = -1; + vcn_len = 0; + lcn_block = -1; + was_hole = FALSE; + cpos = pos >> vol->cluster_size_bits; + end = pos + bytes; + cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits; + /* + * Loop over each page and for each page over each buffer. Use goto to + * reduce indentation. + */ + u = 0; +do_next_page: + page = pages[u]; + bh_pos = (s64)page->index << PAGE_CACHE_SHIFT; + bh = head = page_buffers(page); + do { + VCN cdelta; + s64 bh_end; + unsigned bh_cofs; + + /* Clear buffer_new on all buffers to reinitialise state. */ + if (buffer_new(bh)) + clear_buffer_new(bh); + bh_end = bh_pos + blocksize; + bh_cpos = bh_pos >> vol->cluster_size_bits; + bh_cofs = bh_pos & vol->cluster_size_mask; + if (buffer_mapped(bh)) { + /* + * The buffer is already mapped. If it is uptodate, + * ignore it. + */ + if (buffer_uptodate(bh)) + continue; + /* + * The buffer is not uptodate. If the page is uptodate + * set the buffer uptodate and otherwise ignore it. + */ + if (PageUptodate(page)) { + set_buffer_uptodate(bh); + continue; + } + /* + * Neither the page nor the buffer are uptodate. If + * the buffer is only partially being written to, we + * need to read it in before the write, i.e. now. + */ + if ((bh_pos < pos && bh_end > pos) || + (bh_pos < end && bh_end > end)) { + /* + * If the buffer is fully or partially within + * the initialized size, do an actual read. + * Otherwise, simply zero the buffer. + */ + read_lock_irqsave(&ni->size_lock, flags); + initialized_size = ni->initialized_size; + read_unlock_irqrestore(&ni->size_lock, flags); + if (bh_pos < initialized_size) { + ntfs_submit_bh_for_read(bh); + *wait_bh++ = bh; + } else { + u8 *kaddr = kmap_atomic(page, KM_USER0); + memset(kaddr + bh_offset(bh), 0, + blocksize); + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + set_buffer_uptodate(bh); + } + } + continue; + } + /* Unmapped buffer. Need to map it. */ + bh->b_bdev = vol->sb->s_bdev; + /* + * If the current buffer is in the same clusters as the map + * cache, there is no need to check the runlist again. The + * map cache is made up of @vcn, which is the first cached file + * cluster, @vcn_len which is the number of cached file + * clusters, @lcn is the device cluster corresponding to @vcn, + * and @lcn_block is the block number corresponding to @lcn. + */ + cdelta = bh_cpos - vcn; + if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) { +map_buffer_cached: + BUG_ON(lcn < 0); + bh->b_blocknr = lcn_block + + (cdelta << (vol->cluster_size_bits - + blocksize_bits)) + + (bh_cofs >> blocksize_bits); + set_buffer_mapped(bh); + /* + * If the page is uptodate so is the buffer. If the + * buffer is fully outside the write, we ignore it if + * it was already allocated and we mark it dirty so it + * gets written out if we allocated it. On the other + * hand, if we allocated the buffer but we are not + * marking it dirty we set buffer_new so we can do + * error recovery. + */ + if (PageUptodate(page)) { + if (!buffer_uptodate(bh)) + set_buffer_uptodate(bh); + if (unlikely(was_hole)) { + /* We allocated the buffer. */ + unmap_underlying_metadata(bh->b_bdev, + bh->b_blocknr); + if (bh_end <= pos || bh_pos >= end) + mark_buffer_dirty(bh); + else + set_buffer_new(bh); + } + continue; + } + /* Page is _not_ uptodate. */ + if (likely(!was_hole)) { + /* + * Buffer was already allocated. If it is not + * uptodate and is only partially being written + * to, we need to read it in before the write, + * i.e. now. + */ + if (!buffer_uptodate(bh) && ((bh_pos < pos && + bh_end > pos) || + (bh_end > end && + bh_end > end))) { + /* + * If the buffer is fully or partially + * within the initialized size, do an + * actual read. Otherwise, simply zero + * the buffer. + */ + read_lock_irqsave(&ni->size_lock, + flags); + initialized_size = ni->initialized_size; + read_unlock_irqrestore(&ni->size_lock, + flags); + if (bh_pos < initialized_size) { + ntfs_submit_bh_for_read(bh); + *wait_bh++ = bh; + } else { + u8 *kaddr = kmap_atomic(page, + KM_USER0); + memset(kaddr + bh_offset(bh), + 0, blocksize); + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + set_buffer_uptodate(bh); + } + } + continue; + } + /* We allocated the buffer. */ + unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); + /* + * If the buffer is fully outside the write, zero it, + * set it uptodate, and mark it dirty so it gets + * written out. If it is partially being written to, + * zero region surrounding the write but leave it to + * commit write to do anything else. Finally, if the + * buffer is fully being overwritten, do nothing. + */ + if (bh_end <= pos || bh_pos >= end) { + if (!buffer_uptodate(bh)) { + u8 *kaddr = kmap_atomic(page, KM_USER0); + memset(kaddr + bh_offset(bh), 0, + blocksize); + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + set_buffer_uptodate(bh); + } + mark_buffer_dirty(bh); + continue; + } + set_buffer_new(bh); + if (!buffer_uptodate(bh) && + (bh_pos < pos || bh_end > end)) { + u8 *kaddr; + unsigned pofs; + + kaddr = kmap_atomic(page, KM_USER0); + if (bh_pos < pos) { + pofs = bh_pos & ~PAGE_CACHE_MASK; + memset(kaddr + pofs, 0, pos - bh_pos); + } + if (bh_end > end) { + pofs = end & ~PAGE_CACHE_MASK; + memset(kaddr + pofs, 0, bh_end - end); + } + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + } + continue; + } + /* + * Slow path: this is the first buffer in the cluster. If it + * is outside allocated size and is not uptodate, zero it and + * set it uptodate. + */ + read_lock_irqsave(&ni->size_lock, flags); + initialized_size = ni->allocated_size; + read_unlock_irqrestore(&ni->size_lock, flags); + if (bh_pos > initialized_size) { + if (PageUptodate(page)) { + if (!buffer_uptodate(bh)) + set_buffer_uptodate(bh); + } else if (!buffer_uptodate(bh)) { + u8 *kaddr = kmap_atomic(page, KM_USER0); + memset(kaddr + bh_offset(bh), 0, blocksize); + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + set_buffer_uptodate(bh); + } + continue; + } + is_retry = FALSE; + if (!rl) { + down_read(&ni->runlist.lock); +retry_remap: + rl = ni->runlist.rl; + } + if (likely(rl != NULL)) { + /* Seek to element containing target cluster. */ + while (rl->length && rl[1].vcn <= bh_cpos) + rl++; + lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos); + if (likely(lcn >= 0)) { + /* + * Successful remap, setup the map cache and + * use that to deal with the buffer. + */ + was_hole = FALSE; + vcn = bh_cpos; + vcn_len = rl[1].vcn - vcn; + lcn_block = lcn << (vol->cluster_size_bits - + blocksize_bits); + cdelta = 0; + /* + * If the number of remaining clusters in the + * @pages is smaller or equal to the number of + * cached clusters, unlock the runlist as the + * map cache will be used from now on. + */ + if (likely(vcn + vcn_len >= cend)) { + if (rl_write_locked) { + up_write(&ni->runlist.lock); + rl_write_locked = FALSE; + } else + up_read(&ni->runlist.lock); + rl = NULL; + } + goto map_buffer_cached; + } + } else + lcn = LCN_RL_NOT_MAPPED; + /* + * If it is not a hole and not out of bounds, the runlist is + * probably unmapped so try to map it now. + */ + if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) { + if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) { + /* Attempt to map runlist. */ + if (!rl_write_locked) { + /* + * We need the runlist locked for + * writing, so if it is locked for + * reading relock it now and retry in + * case it changed whilst we dropped + * the lock. + */ + up_read(&ni->runlist.lock); + down_write(&ni->runlist.lock); + rl_write_locked = TRUE; + goto retry_remap; + } + err = ntfs_map_runlist_nolock(ni, bh_cpos, + NULL); + if (likely(!err)) { + is_retry = TRUE; + goto retry_remap; + } + /* + * If @vcn is out of bounds, pretend @lcn is + * LCN_ENOENT. As long as the buffer is out + * of bounds this will work fine. + */ + if (err == -ENOENT) { + lcn = LCN_ENOENT; + err = 0; + goto rl_not_mapped_enoent; + } + } else + err = -EIO; + /* Failed to map the buffer, even after retrying. */ + bh->b_blocknr = -1; + ntfs_error(vol->sb, "Failed to write to inode 0x%lx, " + "attribute type 0x%x, vcn 0x%llx, " + "vcn offset 0x%x, because its " + "location on disk could not be " + "determined%s (error code %i).", + ni->mft_no, ni->type, + (unsigned long long)bh_cpos, + (unsigned)bh_pos & + vol->cluster_size_mask, + is_retry ? " even after retrying" : "", + err); + break; + } +rl_not_mapped_enoent: + /* + * The buffer is in a hole or out of bounds. We need to fill + * the hole, unless the buffer is in a cluster which is not + * touched by the write, in which case we just leave the buffer + * unmapped. This can only happen when the cluster size is + * less than the page cache size. + */ + if (unlikely(vol->cluster_size < PAGE_CACHE_SIZE)) { + bh_cend = (bh_end + vol->cluster_size - 1) >> + vol->cluster_size_bits; + if ((bh_cend <= cpos || bh_cpos >= cend)) { + bh->b_blocknr = -1; + /* + * If the buffer is uptodate we skip it. If it + * is not but the page is uptodate, we can set + * the buffer uptodate. If the page is not + * uptodate, we can clear the buffer and set it + * uptodate. Whether this is worthwhile is + * debatable and this could be removed. + */ + if (PageUptodate(page)) { + if (!buffer_uptodate(bh)) + set_buffer_uptodate(bh); + } else if (!buffer_uptodate(bh)) { + u8 *kaddr = kmap_atomic(page, KM_USER0); + memset(kaddr + bh_offset(bh), 0, + blocksize); + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + set_buffer_uptodate(bh); + } + continue; + } + } + /* + * Out of bounds buffer is invalid if it was not really out of + * bounds. + */ + BUG_ON(lcn != LCN_HOLE); + /* + * We need the runlist locked for writing, so if it is locked + * for reading relock it now and retry in case it changed + * whilst we dropped the lock. + */ + BUG_ON(!rl); + if (!rl_write_locked) { + up_read(&ni->runlist.lock); + down_write(&ni->runlist.lock); + rl_write_locked = TRUE; + goto retry_remap; + } + /* Find the previous last allocated cluster. */ + BUG_ON(rl->lcn != LCN_HOLE); + lcn = -1; + rl2 = rl; + while (--rl2 >= ni->runlist.rl) { + if (rl2->lcn >= 0) { + lcn = rl2->lcn + rl2->length; + break; + } + } + rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE, + FALSE); + if (IS_ERR(rl2)) { + err = PTR_ERR(rl2); + ntfs_debug("Failed to allocate cluster, error code %i.", + err); + break; + } + lcn = rl2->lcn; + rl = ntfs_runlists_merge(ni->runlist.rl, rl2); + if (IS_ERR(rl)) { + err = PTR_ERR(rl); + if (err != -ENOMEM) + err = -EIO; + if (ntfs_cluster_free_from_rl(vol, rl2)) { + ntfs_error(vol->sb, "Failed to release " + "allocated cluster in error " + "code path. Run chkdsk to " + "recover the lost cluster."); + NVolSetErrors(vol); + } + ntfs_free(rl2); + break; + } + ni->runlist.rl = rl; + status.runlist_merged = 1; + ntfs_debug("Allocated cluster, lcn 0x%llx.", lcn); + /* Map and lock the mft record and get the attribute record. */ + if (!NInoAttr(ni)) + base_ni = ni; + else + base_ni = ni->ext.base_ntfs_ino; + m = map_mft_record(base_ni); + if (IS_ERR(m)) { + err = PTR_ERR(m); + break; + } + ctx = ntfs_attr_get_search_ctx(base_ni, m); + if (unlikely(!ctx)) { + err = -ENOMEM; + unmap_mft_record(base_ni); + break; + } + status.mft_attr_mapped = 1; + err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, bh_cpos, NULL, 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) + err = -EIO; + break; + } + m = ctx->mrec; + a = ctx->attr; + /* + * Find the runlist element with which the attribute extent + * starts. Note, we cannot use the _attr_ version because we + * have mapped the mft record. That is ok because we know the + * runlist fragment must be mapped already to have ever gotten + * here, so we can just use the _rl_ version. + */ + vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn); + rl2 = ntfs_rl_find_vcn_nolock(rl, vcn); + BUG_ON(!rl2); + BUG_ON(!rl2->length); + BUG_ON(rl2->lcn < LCN_HOLE); + highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn); + /* + * If @highest_vcn is zero, calculate the real highest_vcn + * (which can really be zero). + */ + if (!highest_vcn) + highest_vcn = (sle64_to_cpu( + a->data.non_resident.allocated_size) >> + vol->cluster_size_bits) - 1; + /* + * Determine the size of the mapping pairs array for the new + * extent, i.e. the old extent with the hole filled. + */ + mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn, + highest_vcn); + if (unlikely(mp_size <= 0)) { + if (!(err = mp_size)) + err = -EIO; + ntfs_debug("Failed to get size for mapping pairs " + "array, error code %i.", err); + break; + } + /* + * Resize the attribute record to fit the new mapping pairs + * array. + */ + attr_rec_len = le32_to_cpu(a->length); + err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu( + a->data.non_resident.mapping_pairs_offset)); + if (unlikely(err)) { + BUG_ON(err != -ENOSPC); + // TODO: Deal with this by using the current attribute + // and fill it with as much of the mapping pairs + // array as possible. Then loop over each attribute + // extent rewriting the mapping pairs arrays as we go + // along and if when we reach the end we have not + // enough space, try to resize the last attribute + // extent and if even that fails, add a new attribute + // extent. + // We could also try to resize at each step in the hope + // that we will not need to rewrite every single extent. + // Note, we may need to decompress some extents to fill + // the runlist as we are walking the extents... + ntfs_error(vol->sb, "Not enough space in the mft " + "record for the extended attribute " + "record. This case is not " + "implemented yet."); + err = -EOPNOTSUPP; + break ; + } + status.mp_rebuilt = 1; + /* + * Generate the mapping pairs array directly into the attribute + * record. + */ + err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( + a->data.non_resident.mapping_pairs_offset), + mp_size, rl2, vcn, highest_vcn, NULL); + if (unlikely(err)) { + ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, " + "attribute type 0x%x, because building " + "the mapping pairs failed with error " + "code %i.", vi->i_ino, + (unsigned)le32_to_cpu(ni->type), err); + err = -EIO; + break; + } + /* Update the highest_vcn but only if it was not set. */ + if (unlikely(!a->data.non_resident.highest_vcn)) + a->data.non_resident.highest_vcn = + cpu_to_sle64(highest_vcn); + /* + * If the attribute is sparse/compressed, update the compressed + * size in the ntfs_inode structure and the attribute record. + */ + if (likely(NInoSparse(ni) || NInoCompressed(ni))) { + /* + * If we are not in the first attribute extent, switch + * to it, but first ensure the changes will make it to + * disk later. + */ + if (a->data.non_resident.lowest_vcn) { + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + ntfs_attr_reinit_search_ctx(ctx); + err = ntfs_attr_lookup(ni->type, ni->name, + ni->name_len, CASE_SENSITIVE, + 0, NULL, 0, ctx); + if (unlikely(err)) { + status.attr_switched = 1; + break; + } + /* @m is not used any more so do not set it. */ + a = ctx->attr; + } + write_lock_irqsave(&ni->size_lock, flags); + ni->itype.compressed.size += vol->cluster_size; + a->data.non_resident.compressed_size = + cpu_to_sle64(ni->itype.compressed.size); + write_unlock_irqrestore(&ni->size_lock, flags); + } + /* Ensure the changes make it to disk. */ + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + ntfs_attr_put_search_ctx(ctx); + unmap_mft_record(base_ni); + /* Successfully filled the hole. */ + status.runlist_merged = 0; + status.mft_attr_mapped = 0; + status.mp_rebuilt = 0; + /* Setup the map cache and use that to deal with the buffer. */ + was_hole = TRUE; + vcn = bh_cpos; + vcn_len = 1; + lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits); + cdelta = 0; + /* + * If the number of remaining clusters in the @pages is smaller + * or equal to the number of cached clusters, unlock the + * runlist as the map cache will be used from now on. + */ + if (likely(vcn + vcn_len >= cend)) { + up_write(&ni->runlist.lock); + rl_write_locked = FALSE; + rl = NULL; + } + goto map_buffer_cached; + } while (bh_pos += blocksize, (bh = bh->b_this_page) != head); + /* If there are no errors, do the next page. */ + if (likely(!err && ++u < nr_pages)) + goto do_next_page; + /* If there are no errors, release the runlist lock if we took it. */ + if (likely(!err)) { + if (unlikely(rl_write_locked)) { + up_write(&ni->runlist.lock); + rl_write_locked = FALSE; + } else if (unlikely(rl)) + up_read(&ni->runlist.lock); + rl = NULL; + } + /* If we issued read requests, let them complete. */ + read_lock_irqsave(&ni->size_lock, flags); + initialized_size = ni->initialized_size; + read_unlock_irqrestore(&ni->size_lock, flags); + while (wait_bh > wait) { + bh = *--wait_bh; + wait_on_buffer(bh); + if (likely(buffer_uptodate(bh))) { + page = bh->b_page; + bh_pos = ((s64)page->index << PAGE_CACHE_SHIFT) + + bh_offset(bh); + /* + * If the buffer overflows the initialized size, need + * to zero the overflowing region. + */ + if (unlikely(bh_pos + blocksize > initialized_size)) { + u8 *kaddr; + int ofs = 0; + + if (likely(bh_pos < initialized_size)) + ofs = initialized_size - bh_pos; + kaddr = kmap_atomic(page, KM_USER0); + memset(kaddr + bh_offset(bh) + ofs, 0, + blocksize - ofs); + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + } + } else /* if (unlikely(!buffer_uptodate(bh))) */ + err = -EIO; + } + if (likely(!err)) { + /* Clear buffer_new on all buffers. */ + u = 0; + do { + bh = head = page_buffers(pages[u]); + do { + if (buffer_new(bh)) + clear_buffer_new(bh); + } while ((bh = bh->b_this_page) != head); + } while (++u < nr_pages); + ntfs_debug("Done."); + return err; + } + if (status.attr_switched) { + /* Get back to the attribute extent we modified. */ + ntfs_attr_reinit_search_ctx(ctx); + if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len, + CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) { + ntfs_error(vol->sb, "Failed to find required " + "attribute extent of attribute in " + "error code path. Run chkdsk to " + "recover."); + write_lock_irqsave(&ni->size_lock, flags); + ni->itype.compressed.size += vol->cluster_size; + write_unlock_irqrestore(&ni->size_lock, flags); + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + /* + * The only thing that is now wrong is the compressed + * size of the base attribute extent which chkdsk + * should be able to fix. + */ + NVolSetErrors(vol); + } else { + m = ctx->mrec; + a = ctx->attr; + status.attr_switched = 0; + } + } + /* + * If the runlist has been modified, need to restore it by punching a + * hole into it and we then need to deallocate the on-disk cluster as + * well. Note, we only modify the runlist if we are able to generate a + * new mapping pairs array, i.e. only when the mapped attribute extent + * is not switched. + */ + if (status.runlist_merged && !status.attr_switched) { + BUG_ON(!rl_write_locked); + /* Make the file cluster we allocated sparse in the runlist. */ + if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) { + ntfs_error(vol->sb, "Failed to punch hole into " + "attribute runlist in error code " + "path. Run chkdsk to recover the " + "lost cluster."); + make_bad_inode(vi); + make_bad_inode(VFS_I(base_ni)); + NVolSetErrors(vol); + } else /* if (success) */ { + status.runlist_merged = 0; + /* + * Deallocate the on-disk cluster we allocated but only + * if we succeeded in punching its vcn out of the + * runlist. + */ + down_write(&vol->lcnbmp_lock); + if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) { + ntfs_error(vol->sb, "Failed to release " + "allocated cluster in error " + "code path. Run chkdsk to " + "recover the lost cluster."); + NVolSetErrors(vol); + } + up_write(&vol->lcnbmp_lock); + } + } + /* + * Resize the attribute record to its old size and rebuild the mapping + * pairs array. Note, we only can do this if the runlist has been + * restored to its old state which also implies that the mapped + * attribute extent is not switched. + */ + if (status.mp_rebuilt && !status.runlist_merged) { + if (ntfs_attr_record_resize(m, a, attr_rec_len)) { + ntfs_error(vol->sb, "Failed to restore attribute " + "record in error code path. Run " + "chkdsk to recover."); + make_bad_inode(vi); + make_bad_inode(VFS_I(base_ni)); + NVolSetErrors(vol); + } else /* if (success) */ { + if (ntfs_mapping_pairs_build(vol, (u8*)a + + le16_to_cpu(a->data.non_resident. + mapping_pairs_offset), attr_rec_len - + le16_to_cpu(a->data.non_resident. + mapping_pairs_offset), ni->runlist.rl, + vcn, highest_vcn, NULL)) { + ntfs_error(vol->sb, "Failed to restore " + "mapping pairs array in error " + "code path. Run chkdsk to " + "recover."); + make_bad_inode(vi); + make_bad_inode(VFS_I(base_ni)); + NVolSetErrors(vol); + } + flush_dcache_mft_record_page(ctx->ntfs_ino); + mark_mft_record_dirty(ctx->ntfs_ino); + } + } + /* Release the mft record and the attribute. */ + if (status.mft_attr_mapped) { + ntfs_attr_put_search_ctx(ctx); + unmap_mft_record(base_ni); + } + /* Release the runlist lock. */ + if (rl_write_locked) + up_write(&ni->runlist.lock); + else if (rl) + up_read(&ni->runlist.lock); + /* + * Zero out any newly allocated blocks to avoid exposing stale data. + * If BH_New is set, we know that the block was newly allocated above + * and that it has not been fully zeroed and marked dirty yet. + */ + nr_pages = u; + u = 0; + end = bh_cpos << vol->cluster_size_bits; + do { + page = pages[u]; + bh = head = page_buffers(page); + do { + if (u == nr_pages && + ((s64)page->index << PAGE_CACHE_SHIFT) + + bh_offset(bh) >= end) + break; + if (!buffer_new(bh)) + continue; + clear_buffer_new(bh); + if (!buffer_uptodate(bh)) { + if (PageUptodate(page)) + set_buffer_uptodate(bh); + else { + u8 *kaddr = kmap_atomic(page, KM_USER0); + memset(kaddr + bh_offset(bh), 0, + blocksize); + kunmap_atomic(kaddr, KM_USER0); + flush_dcache_page(page); + set_buffer_uptodate(bh); + } + } + mark_buffer_dirty(bh); + } while ((bh = bh->b_this_page) != head); + } while (++u <= nr_pages); + ntfs_error(vol->sb, "Failed. Returning error code %i.", err); + return err; +} + +/* + * Copy as much as we can into the pages and return the number of bytes which + * were sucessfully copied. If a fault is encountered then clear the pages + * out to (ofs + bytes) and return the number of bytes which were copied. + */ +static inline size_t ntfs_copy_from_user(struct page **pages, + unsigned nr_pages, unsigned ofs, const char __user *buf, + size_t bytes) +{ + struct page **last_page = pages + nr_pages; + char *kaddr; + size_t total = 0; + unsigned len; + int left; + + do { + len = PAGE_CACHE_SIZE - ofs; + if (len > bytes) + len = bytes; + kaddr = kmap_atomic(*pages, KM_USER0); + left = __copy_from_user_inatomic(kaddr + ofs, buf, len); + kunmap_atomic(kaddr, KM_USER0); + if (unlikely(left)) { + /* Do it the slow way. */ + kaddr = kmap(*pages); + left = __copy_from_user(kaddr + ofs, buf, len); + kunmap(*pages); + if (unlikely(left)) + goto err_out; + } + total += len; + bytes -= len; + if (!bytes) + break; + buf += len; + ofs = 0; + } while (++pages < last_page); +out: + return total; +err_out: + total += len - left; + /* Zero the rest of the target like __copy_from_user(). */ + while (++pages < last_page) { + bytes -= len; + if (!bytes) + break; + len = PAGE_CACHE_SIZE; + if (len > bytes) + len = bytes; + kaddr = kmap_atomic(*pages, KM_USER0); + memset(kaddr, 0, len); + kunmap_atomic(kaddr, KM_USER0); + } + goto out; +} + +static size_t __ntfs_copy_from_user_iovec(char *vaddr, + const struct iovec *iov, size_t iov_ofs, size_t bytes) +{ + size_t total = 0; + + while (1) { + const char __user *buf = iov->iov_base + iov_ofs; + unsigned len; + size_t left; + + len = iov->iov_len - iov_ofs; + if (len > bytes) + len = bytes; + left = __copy_from_user_inatomic(vaddr, buf, len); + total += len; + bytes -= len; + vaddr += len; + if (unlikely(left)) { + /* + * Zero the rest of the target like __copy_from_user(). + */ + memset(vaddr, 0, bytes); + total -= left; + break; + } + if (!bytes) + break; + iov++; + iov_ofs = 0; + } + return total; +} + +static inline void ntfs_set_next_iovec(const struct iovec **iovp, + size_t *iov_ofsp, size_t bytes) +{ + const struct iovec *iov = *iovp; + size_t iov_ofs = *iov_ofsp; + + while (bytes) { + unsigned len; + + len = iov->iov_len - iov_ofs; + if (len > bytes) + len = bytes; + bytes -= len; + iov_ofs += len; + if (iov->iov_len == iov_ofs) { + iov++; + iov_ofs = 0; + } + } + *iovp = iov; + *iov_ofsp = iov_ofs; +} + +/* + * This has the same side-effects and return value as ntfs_copy_from_user(). + * The difference is that on a fault we need to memset the remainder of the + * pages (out to offset + bytes), to emulate ntfs_copy_from_user()'s + * single-segment behaviour. + * + * We call the same helper (__ntfs_copy_from_user_iovec()) both when atomic and + * when not atomic. This is ok because __ntfs_copy_from_user_iovec() calls + * __copy_from_user_inatomic() and it is ok to call this when non-atomic. In + * fact, the only difference between __copy_from_user_inatomic() and + * __copy_from_user() is that the latter calls might_sleep(). And on many + * architectures __copy_from_user_inatomic() is just defined to + * __copy_from_user() so it makes no difference at all on those architectures. + */ +static inline size_t ntfs_copy_from_user_iovec(struct page **pages, + unsigned nr_pages, unsigned ofs, const struct iovec **iov, + size_t *iov_ofs, size_t bytes) +{ + struct page **last_page = pages + nr_pages; + char *kaddr; + size_t copied, len, total = 0; + + do { + len = PAGE_CACHE_SIZE - ofs; + if (len > bytes) + len = bytes; + kaddr = kmap_atomic(*pages, KM_USER0); + copied = __ntfs_copy_from_user_iovec(kaddr + ofs, + *iov, *iov_ofs, len); + kunmap_atomic(kaddr, KM_USER0); + if (unlikely(copied != len)) { + /* Do it the slow way. */ + kaddr = kmap(*pages); + copied = __ntfs_copy_from_user_iovec(kaddr + ofs, + *iov, *iov_ofs, len); + kunmap(*pages); + if (unlikely(copied != len)) + goto err_out; + } + total += len; + bytes -= len; + if (!bytes) + break; + ntfs_set_next_iovec(iov, iov_ofs, len); + ofs = 0; + } while (++pages < last_page); +out: + return total; +err_out: + total += copied; + /* Zero the rest of the target like __copy_from_user(). */ + while (++pages < last_page) { + bytes -= len; + if (!bytes) + break; + len = PAGE_CACHE_SIZE; + if (len > bytes) + len = bytes; + kaddr = kmap_atomic(*pages, KM_USER0); + memset(kaddr, 0, len); + kunmap_atomic(kaddr, KM_USER0); + } + goto out; +} + +static inline void ntfs_flush_dcache_pages(struct page **pages, + unsigned nr_pages) +{ + BUG_ON(!nr_pages); + do { + /* + * Warning: Do not do the decrement at the same time as the + * call because flush_dcache_page() is a NULL macro on i386 + * and hence the decrement never happens. + */ + flush_dcache_page(pages[nr_pages]); + } while (--nr_pages > 0); +} + +/** + * ntfs_commit_pages_after_non_resident_write - commit the received data + * @pages: array of destination pages + * @nr_pages: number of pages in @pages + * @pos: byte position in file at which the write begins + * @bytes: number of bytes to be written + * + * See description of ntfs_commit_pages_after_write(), below. + */ +static inline int ntfs_commit_pages_after_non_resident_write( + struct page **pages, const unsigned nr_pages, + s64 pos, size_t bytes) +{ + s64 end, initialized_size; + struct inode *vi; + ntfs_inode *ni, *base_ni; + struct buffer_head *bh, *head; + ntfs_attr_search_ctx *ctx; + MFT_RECORD *m; + ATTR_RECORD *a; + unsigned long flags; + unsigned blocksize, u; + int err; + + vi = pages[0]->mapping->host; + ni = NTFS_I(vi); + blocksize = 1 << vi->i_blkbits; + end = pos + bytes; + u = 0; + do { + s64 bh_pos; + struct page *page; + BOOL partial; + + page = pages[u]; + bh_pos = (s64)page->index << PAGE_CACHE_SHIFT; + bh = head = page_buffers(page); + partial = FALSE; + do { + s64 bh_end; + + bh_end = bh_pos + blocksize; + if (bh_end <= pos || bh_pos >= end) { + if (!buffer_uptodate(bh)) + partial = TRUE; + } else { + set_buffer_uptodate(bh); + mark_buffer_dirty(bh); + } + } while (bh_pos += blocksize, (bh = bh->b_this_page) != head); + /* + * If all buffers are now uptodate but the page is not, set the + * page uptodate. + */ + if (!partial && !PageUptodate(page)) + SetPageUptodate(page); + } while (++u < nr_pages); + /* + * Finally, if we do not need to update initialized_size or i_size we + * are finished. + */ + read_lock_irqsave(&ni->size_lock, flags); + initialized_size = ni->initialized_size; + read_unlock_irqrestore(&ni->size_lock, flags); + if (end <= initialized_size) { + ntfs_debug("Done."); + return 0; + } + /* + * Updat