X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=fs%2Fntfs%2Fdir.c;h=93577561cdbe6b72d9271a4f37ba1e4d476fb5ef;hb=6a77f38946aaee1cd85eeec6cf4229b204c15071;hp=1f0de905442e07497b93c550f2e5c2944f1a3752;hpb=5273a3df6485dc2ad6aa7ddd441b9a21970f003b;p=linux-2.6.git diff --git a/fs/ntfs/dir.c b/fs/ntfs/dir.c index 1f0de9054..93577561c 100644 --- a/fs/ntfs/dir.c +++ b/fs/ntfs/dir.c @@ -21,15 +21,20 @@ */ #include -#include "ntfs.h" +#include + #include "dir.h" +#include "aops.h" +#include "attrib.h" +#include "mft.h" +#include "debug.h" +#include "ntfs.h" /** * The little endian Unicode string $I30 as a global constant. */ -uchar_t I30[5] = { const_cpu_to_le16('$'), const_cpu_to_le16('I'), - const_cpu_to_le16('3'), const_cpu_to_le16('0'), - const_cpu_to_le16(0) }; +ntfschar I30[5] = { const_cpu_to_le16('$'), const_cpu_to_le16('I'), + const_cpu_to_le16('3'), const_cpu_to_le16('0'), 0 }; /** * ntfs_lookup_inode_by_name - find an inode in a directory given its name @@ -63,8 +68,16 @@ uchar_t I30[5] = { const_cpu_to_le16('$'), const_cpu_to_le16('I'), * This is to avoid polluting the dcache with short file names. We want them to * work but we don't care for how quickly one can access them. This also fixes * the dcache aliasing issues. + * + * Locking: - Caller must hold i_sem on the directory. + * - Each page cache page in the index allocation mapping must be + * locked whilst being accessed otherwise we may find a corrupt + * page due to it being under ->writepage at the moment which + * applies the mst protection fixups before writing out and then + * removes them again after the write is complete after which it + * unlocks the page. */ -MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, +MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const ntfschar *uname, const int uname_len, ntfs_name **res) { ntfs_volume *vol = dir_ni->vol; @@ -75,7 +88,7 @@ MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, INDEX_ALLOCATION *ia; u8 *index_end; u64 mref; - attr_search_context *ctx; + ntfs_attr_search_ctx *ctx; int err, rc; VCN vcn, old_vcn; struct address_space *ia_mapping; @@ -83,24 +96,30 @@ MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, u8 *kaddr; ntfs_name *name = NULL; + BUG_ON(!S_ISDIR(VFS_I(dir_ni)->i_mode)); + BUG_ON(NInoAttr(dir_ni)); /* Get hold of the mft record for the directory. */ m = map_mft_record(dir_ni); - if (unlikely(IS_ERR(m))) { + if (IS_ERR(m)) { ntfs_error(sb, "map_mft_record() failed with error code %ld.", -PTR_ERR(m)); return ERR_MREF(PTR_ERR(m)); } - ctx = get_attr_search_ctx(dir_ni, m); + ctx = ntfs_attr_get_search_ctx(dir_ni, m); if (unlikely(!ctx)) { err = -ENOMEM; goto err_out; } /* Find the index root attribute in the mft record. */ - if (!lookup_attr(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL, 0, - ctx)) { - ntfs_error(sb, "Index root attribute missing in directory " - "inode 0x%lx.", dir_ni->mft_no); - err = -EIO; + err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL, + 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) { + ntfs_error(sb, "Index root attribute missing in " + "directory inode 0x%lx.", + dir_ni->mft_no); + err = -EIO; + } goto err_out; } /* Get to the index root value (it's been verified in read_inode). */ @@ -135,7 +154,7 @@ MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, * returning. */ if (ntfs_are_names_equal(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, CASE_SENSITIVE, vol->upcase, vol->upcase_len)) { found_it: @@ -169,7 +188,7 @@ found_it: *res = NULL; } mref = le64_to_cpu(ie->data.dir.indexed_file); - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); unmap_mft_record(dir_ni); return mref; } @@ -186,7 +205,7 @@ found_it: if (!NVolCaseSensitive(vol) && ie->key.file_name.file_name_type && ntfs_are_names_equal(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, IGNORE_CASE, vol->upcase, vol->upcase_len)) { int name_size = sizeof(ntfs_name); @@ -206,7 +225,7 @@ found_it: } if (type != FILE_NAME_DOS) - name_size += len * sizeof(uchar_t); + name_size += len * sizeof(ntfschar); name = kmalloc(name_size, GFP_NOFS); if (!name) { err = -ENOMEM; @@ -217,7 +236,7 @@ found_it: if (type != FILE_NAME_DOS) { name->len = len; memcpy(name->name, ie->key.file_name.file_name, - len * sizeof(uchar_t)); + len * sizeof(ntfschar)); } else name->len = 0; *res = name; @@ -227,7 +246,7 @@ found_it: * know which way in the B+tree we have to go. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, IGNORE_CASE, vol->upcase, vol->upcase_len); /* @@ -246,7 +265,7 @@ found_it: * collation. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, CASE_SENSITIVE, vol->upcase, vol->upcase_len); if (rc == -1) @@ -268,7 +287,7 @@ found_it: */ if (!(ie->flags & INDEX_ENTRY_NODE)) { if (name) { - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); unmap_mft_record(dir_ni); return name->mref; } @@ -281,17 +300,16 @@ found_it: ntfs_error(sb, "No index allocation attribute but index entry " "requires one. Directory inode 0x%lx is " "corrupt or driver bug.", dir_ni->mft_no); - err = -EIO; goto err_out; } /* Get the starting vcn of the index_block holding the child node. */ - vcn = sle64_to_cpup((u8*)ie + le16_to_cpu(ie->length) - 8); + vcn = sle64_to_cpup((sle64*)((u8*)ie + le16_to_cpu(ie->length) - 8)); ia_mapping = VFS_I(dir_ni)->i_mapping; /* * We are done with the index root and the mft record. Release them, * otherwise we deadlock with ntfs_map_page(). */ - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); unmap_mft_record(dir_ni); m = NULL; ctx = NULL; @@ -309,6 +327,7 @@ descend_into_child_node: err = PTR_ERR(page); goto err_out; } + lock_page(page); kaddr = (u8*)page_address(page); fast_descend_into_child_node: /* Get to the index allocation block. */ @@ -318,7 +337,13 @@ fast_descend_into_child_node: if ((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_CACHE_SIZE) { ntfs_error(sb, "Out of bounds check failed. Corrupt directory " "inode 0x%lx or driver bug.", dir_ni->mft_no); - err = -EIO; + goto unm_err_out; + } + /* Catch multi sector transfer fixup errors. */ + if (unlikely(!ntfs_is_indx_record(ia->magic))) { + ntfs_error(sb, "Directory index record with vcn 0x%llx is " + "corrupt. Corrupt inode 0x%lx. Run chkdsk.", + (unsigned long long)vcn, dir_ni->mft_no); goto unm_err_out; } if (sle64_to_cpu(ia->index_block_vcn) != vcn) { @@ -328,7 +353,6 @@ fast_descend_into_child_node: "bug.", (unsigned long long) sle64_to_cpu(ia->index_block_vcn), (unsigned long long)vcn, dir_ni->mft_no); - err = -EIO; goto unm_err_out; } if (le32_to_cpu(ia->index.allocated_size) + 0x18 != @@ -340,7 +364,6 @@ fast_descend_into_child_node: (unsigned long long)vcn, dir_ni->mft_no, le32_to_cpu(ia->index.allocated_size) + 0x18, dir_ni->itype.index.block_size); - err = -EIO; goto unm_err_out; } index_end = (u8*)ia + dir_ni->itype.index.block_size; @@ -350,7 +373,6 @@ fast_descend_into_child_node: "Cannot access! This is probably a bug in the " "driver.", (unsigned long long)vcn, dir_ni->mft_no); - err = -EIO; goto unm_err_out; } index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length); @@ -358,7 +380,6 @@ fast_descend_into_child_node: ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of directory " "inode 0x%lx exceeds maximum size.", (unsigned long long)vcn, dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* The first index entry. */ @@ -378,7 +399,6 @@ fast_descend_into_child_node: ntfs_error(sb, "Index entry out of bounds in " "directory inode 0x%lx.", dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* @@ -395,7 +415,7 @@ fast_descend_into_child_node: * returning. */ if (ntfs_are_names_equal(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, CASE_SENSITIVE, vol->upcase, vol->upcase_len)) { found_it2: @@ -429,6 +449,7 @@ found_it2: *res = NULL; } mref = le64_to_cpu(ie->data.dir.indexed_file); + unlock_page(page); ntfs_unmap_page(page); return mref; } @@ -445,7 +466,7 @@ found_it2: if (!NVolCaseSensitive(vol) && ie->key.file_name.file_name_type && ntfs_are_names_equal(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, IGNORE_CASE, vol->upcase, vol->upcase_len)) { int name_size = sizeof(ntfs_name); @@ -461,12 +482,13 @@ found_it2: "this message to " "linux-ntfs-dev@lists." "sourceforge.net."); + unlock_page(page); ntfs_unmap_page(page); goto dir_err_out; } if (type != FILE_NAME_DOS) - name_size += len * sizeof(uchar_t); + name_size += len * sizeof(ntfschar); name = kmalloc(name_size, GFP_NOFS); if (!name) { err = -ENOMEM; @@ -477,7 +499,7 @@ found_it2: if (type != FILE_NAME_DOS) { name->len = len; memcpy(name->name, ie->key.file_name.file_name, - len * sizeof(uchar_t)); + len * sizeof(ntfschar)); } else name->len = 0; *res = name; @@ -487,7 +509,7 @@ found_it2: * know which way in the B+tree we have to go. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, IGNORE_CASE, vol->upcase, vol->upcase_len); /* @@ -506,7 +528,7 @@ found_it2: * collation. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, CASE_SENSITIVE, vol->upcase, vol->upcase_len); if (rc == -1) @@ -529,12 +551,12 @@ found_it2: ntfs_error(sb, "Index entry with child node found in " "a leaf node in directory inode 0x%lx.", dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* Child node present, descend into it. */ old_vcn = vcn; - vcn = sle64_to_cpup((u8*)ie + le16_to_cpu(ie->length) - 8); + vcn = sle64_to_cpup((sle64*)((u8*)ie + + le16_to_cpu(ie->length) - 8)); if (vcn >= 0) { /* If vcn is in the same page cache page as old_vcn we * recycle the mapped page. */ @@ -543,12 +565,12 @@ found_it2: vol->cluster_size_bits >> PAGE_CACHE_SHIFT) goto fast_descend_into_child_node; + unlock_page(page); ntfs_unmap_page(page); goto descend_into_child_node; } ntfs_error(sb, "Negative child node vcn in directory inode " "0x%lx.", dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* @@ -557,16 +579,20 @@ found_it2: * associated with it. */ if (name) { + unlock_page(page); ntfs_unmap_page(page); return name->mref; } ntfs_debug("Entry not found."); err = -ENOENT; unm_err_out: + unlock_page(page); ntfs_unmap_page(page); err_out: + if (!err) + err = -EIO; if (ctx) - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); if (m) unmap_mft_record(dir_ni); if (name) { @@ -575,8 +601,7 @@ err_out: } return ERR_MREF(err); dir_err_out: - ntfs_error(sb, "Corrupt directory. Aborting lookup."); - err = -EIO; + ntfs_error(sb, "Corrupt directory. Aborting lookup."); goto err_out; } @@ -607,7 +632,7 @@ dir_err_out: * * Note, @uname_len does not include the (optional) terminating NULL character. */ -u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, +u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const ntfschar *uname, const int uname_len) { ntfs_volume *vol = dir_ni->vol; @@ -618,7 +643,7 @@ u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, INDEX_ALLOCATION *ia; u8 *index_end; u64 mref; - attr_search_context *ctx; + ntfs_attr_search_ctx *ctx; int err, rc; IGNORE_CASE_BOOL ic; VCN vcn, old_vcn; @@ -633,17 +658,21 @@ u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, -PTR_ERR(m)); return ERR_MREF(PTR_ERR(m)); } - ctx = get_attr_search_ctx(dir_ni, m); + ctx = ntfs_attr_get_search_ctx(dir_ni, m); if (!ctx) { err = -ENOMEM; goto err_out; } /* Find the index root attribute in the mft record. */ - if (!lookup_attr(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL, 0, - ctx)) { - ntfs_error(sb, "Index root attribute missing in directory " - "inode 0x%lx.", dir_ni->mft_no); - err = -EIO; + err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL, + 0, ctx); + if (unlikely(err)) { + if (err == -ENOENT) { + ntfs_error(sb, "Index root attribute missing in " + "directory inode 0x%lx.", + dir_ni->mft_no); + err = -EIO; + } goto err_out; } /* Get to the index root value (it's been verified in read_inode). */ @@ -689,12 +718,12 @@ u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const uchar_t *uname, * convert it to cpu format before returning. */ if (ntfs_are_names_equal(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, ic, vol->upcase, vol->upcase_len)) { found_it: mref = le64_to_cpu(ie->data.dir.indexed_file); - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); unmap_mft_record(dir_ni); return mref; } @@ -703,7 +732,7 @@ found_it: * know which way in the B+tree we have to go. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, IGNORE_CASE, vol->upcase, vol->upcase_len); /* @@ -722,7 +751,7 @@ found_it: * collation. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, CASE_SENSITIVE, vol->upcase, vol->upcase_len); if (rc == -1) @@ -750,7 +779,6 @@ found_it: ntfs_error(sb, "No index allocation attribute but index entry " "requires one. Directory inode 0x%lx is " "corrupt or driver bug.", dir_ni->mft_no); - err = -EIO; goto err_out; } /* Get the starting vcn of the index_block holding the child node. */ @@ -760,7 +788,7 @@ found_it: * We are done with the index root and the mft record. Release them, * otherwise we deadlock with ntfs_map_page(). */ - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); unmap_mft_record(dir_ni); m = NULL; ctx = NULL; @@ -778,6 +806,7 @@ descend_into_child_node: err = PTR_ERR(page); goto err_out; } + lock_page(page); kaddr = (u8*)page_address(page); fast_descend_into_child_node: /* Get to the index allocation block. */ @@ -787,7 +816,13 @@ fast_descend_into_child_node: if ((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_CACHE_SIZE) { ntfs_error(sb, "Out of bounds check failed. Corrupt directory " "inode 0x%lx or driver bug.", dir_ni->mft_no); - err = -EIO; + goto unm_err_out; + } + /* Catch multi sector transfer fixup errors. */ + if (unlikely(!ntfs_is_indx_record(ia->magic))) { + ntfs_error(sb, "Directory index record with vcn 0x%llx is " + "corrupt. Corrupt inode 0x%lx. Run chkdsk.", + (unsigned long long)vcn, dir_ni->mft_no); goto unm_err_out; } if (sle64_to_cpu(ia->index_block_vcn) != vcn) { @@ -797,7 +832,6 @@ fast_descend_into_child_node: "bug.", (unsigned long long) sle64_to_cpu(ia->index_block_vcn), (unsigned long long)vcn, dir_ni->mft_no); - err = -EIO; goto unm_err_out; } if (le32_to_cpu(ia->index.allocated_size) + 0x18 != @@ -809,7 +843,6 @@ fast_descend_into_child_node: (unsigned long long)vcn, dir_ni->mft_no, le32_to_cpu(ia->index.allocated_size) + 0x18, dir_ni->itype.index.block_size); - err = -EIO; goto unm_err_out; } index_end = (u8*)ia + dir_ni->itype.index.block_size; @@ -819,7 +852,6 @@ fast_descend_into_child_node: "Cannot access! This is probably a bug in the " "driver.", (unsigned long long)vcn, dir_ni->mft_no); - err = -EIO; goto unm_err_out; } index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length); @@ -827,7 +859,6 @@ fast_descend_into_child_node: ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of directory " "inode 0x%lx exceeds maximum size.", (unsigned long long)vcn, dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* The first index entry. */ @@ -847,7 +878,6 @@ fast_descend_into_child_node: ntfs_error(sb, "Index entry out of bounds in " "directory inode 0x%lx.", dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* @@ -875,11 +905,12 @@ fast_descend_into_child_node: * convert it to cpu format before returning. */ if (ntfs_are_names_equal(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, ic, vol->upcase, vol->upcase_len)) { found_it2: mref = le64_to_cpu(ie->data.dir.indexed_file); + unlock_page(page); ntfs_unmap_page(page); return mref; } @@ -888,7 +919,7 @@ found_it2: * know which way in the B+tree we have to go. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, IGNORE_CASE, vol->upcase, vol->upcase_len); /* @@ -907,7 +938,7 @@ found_it2: * collation. */ rc = ntfs_collate_names(uname, uname_len, - (uchar_t*)&ie->key.file_name.file_name, + (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, 1, CASE_SENSITIVE, vol->upcase, vol->upcase_len); if (rc == -1) @@ -930,7 +961,6 @@ found_it2: ntfs_error(sb, "Index entry with child node found in " "a leaf node in directory inode 0x%lx.", dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* Child node present, descend into it. */ @@ -944,50 +974,41 @@ found_it2: vol->cluster_size_bits >> PAGE_CACHE_SHIFT) goto fast_descend_into_child_node; + unlock_page(page); ntfs_unmap_page(page); goto descend_into_child_node; } ntfs_error(sb, "Negative child node vcn in directory inode " "0x%lx.", dir_ni->mft_no); - err = -EIO; goto unm_err_out; } /* No child node, return -ENOENT. */ ntfs_debug("Entry not found."); err = -ENOENT; unm_err_out: + unlock_page(page); ntfs_unmap_page(page); err_out: + if (!err) + err = -EIO; if (ctx) - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); if (m) unmap_mft_record(dir_ni); return ERR_MREF(err); dir_err_out: ntfs_error(sb, "Corrupt directory. Aborting lookup."); - err = -EIO; goto err_out; } #endif -typedef union { - INDEX_ROOT *ir; - INDEX_ALLOCATION *ia; -} index_union __attribute__ ((__transparent_union__)); - -typedef enum { - INDEX_TYPE_ROOT, /* index root */ - INDEX_TYPE_ALLOCATION, /* index allocation */ -} INDEX_TYPE; - /** * ntfs_filldir - ntfs specific filldir method * @vol: current ntfs volume * @fpos: position in the directory * @ndir: ntfs inode of current directory - * @index_type: specifies whether @iu is an index root or an index allocation - * @iu: index root or index allocation attribute to which @ie belongs + * @ia_page: page in which the index allocation buffer @ie is in resides * @ie: current index entry * @name: buffer to use for the converted name * @dirent: vfs filldir callback context @@ -995,24 +1016,25 @@ typedef enum { * * Convert the Unicode @name to the loaded NLS and pass it to the @filldir * callback. + * + * If @ia_page is not NULL it is the locked page containing the index + * allocation block containing the index entry @ie. + * + * Note, we drop (and then reacquire) the page lock on @ia_page across the + * @filldir() call otherwise we would deadlock with NFSd when it calls ->lookup + * since ntfs_lookup() will lock the same page. As an optimization, we do not + * retake the lock if we are returning a non-zero value as ntfs_readdir() + * would need to drop the lock immediately anyway. */ -static inline int ntfs_filldir(ntfs_volume *vol, loff_t *fpos, - ntfs_inode *ndir, const INDEX_TYPE index_type, - index_union iu, INDEX_ENTRY *ie, u8 *name, - void *dirent, filldir_t filldir) +static inline int ntfs_filldir(ntfs_volume *vol, loff_t fpos, + ntfs_inode *ndir, struct page *ia_page, INDEX_ENTRY *ie, + u8 *name, void *dirent, filldir_t filldir) { - int name_len; + unsigned long mref; + int name_len, rc; unsigned dt_type; FILE_NAME_TYPE_FLAGS name_type; - /* Advance the position even if going to skip the entry. */ - if (index_type == INDEX_TYPE_ALLOCATION) - *fpos = (u8*)ie - (u8*)iu.ia + - (sle64_to_cpu(iu.ia->index_block_vcn) << - ndir->itype.index.vcn_size_bits) + - vol->mft_record_size; - else /* if (index_type == INDEX_TYPE_ROOT) */ - *fpos = (u8*)ie - (u8*)iu.ir; name_type = ie->key.file_name.file_name_type; if (name_type == FILE_NAME_DOS) { ntfs_debug("Skipping DOS name space entry."); @@ -1027,7 +1049,7 @@ static inline int ntfs_filldir(ntfs_volume *vol, loff_t *fpos, ntfs_debug("Skipping system file."); return 0; } - name_len = ntfs_ucstonls(vol, (uchar_t*)&ie->key.file_name.file_name, + name_len = ntfs_ucstonls(vol, (ntfschar*)&ie->key.file_name.file_name, ie->key.file_name.file_name_length, &name, NTFS_MAX_NAME_LEN * NLS_MAX_CHARSET_SIZE + 1); if (name_len <= 0) { @@ -1039,24 +1061,42 @@ static inline int ntfs_filldir(ntfs_volume *vol, loff_t *fpos, dt_type = DT_DIR; else dt_type = DT_REG; + mref = MREF_LE(ie->data.dir.indexed_file); + /* + * Drop the page lock otherwise we deadlock with NFS when it calls + * ->lookup since ntfs_lookup() will lock the same page. + */ + if (ia_page) + unlock_page(ia_page); ntfs_debug("Calling filldir for %s with len %i, fpos 0x%llx, inode " - "0x%lx, DT_%s.", name, name_len, *fpos, - MREF_LE(ie->data.dir.indexed_file), + "0x%lx, DT_%s.", name, name_len, fpos, mref, dt_type == DT_DIR ? "DIR" : "REG"); - return filldir(dirent, name, name_len, *fpos, - MREF_LE(ie->data.dir.indexed_file), dt_type); + rc = filldir(dirent, name, name_len, fpos, mref, dt_type); + /* Relock the page but not if we are aborting ->readdir. */ + if (!rc && ia_page) + lock_page(ia_page); + return rc; } /* - * VFS calls readdir without BKL but with i_sem held. This protects the VFS - * parts (e.g. ->f_pos and ->i_size, and it also protects against directory - * modifications). - * * We use the same basic approach as the old NTFS driver, i.e. we parse the * index root entries and then the index allocation entries that are marked * as in use in the index bitmap. + * * While this will return the names in random order this doesn't matter for - * readdir but OTOH results in a faster readdir. + * ->readdir but OTOH results in a faster ->readdir. + * + * VFS calls ->readdir without BKL but with i_sem held. This protects the VFS + * parts (e.g. ->f_pos and ->i_size, and it also protects against directory + * modifications). + * + * Locking: - Caller must hold i_sem on the directory. + * - Each page cache page in the index allocation mapping must be + * locked whilst being accessed otherwise we may find a corrupt + * page due to it being under ->writepage at the moment which + * applies the mst protection fixups before writing out and then + * removes them again after the write is complete after which it + * unlocks the page. */ static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir) { @@ -1067,7 +1107,7 @@ static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir) ntfs_inode *ndir = NTFS_I(vdir); ntfs_volume *vol = NTFS_SB(sb); MFT_RECORD *m; - INDEX_ROOT *ir; + INDEX_ROOT *ir = NULL; INDEX_ENTRY *ie; INDEX_ALLOCATION *ia; u8 *name = NULL; @@ -1075,7 +1115,7 @@ static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir) struct address_space *ia_mapping, *bmp_mapping; struct page *bmp_page = NULL, *ia_page = NULL; u8 *kaddr, *bmp, *index_end; - attr_search_context *ctx; + ntfs_attr_search_ctx *ctx; fpos = filp->f_pos; ntfs_debug("Entering for inode 0x%lx, fpos 0x%llx.", @@ -1120,12 +1160,12 @@ static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir) goto skip_index_root; /* Get hold of the mft record for the directory. */ m = map_mft_record(ndir); - if (unlikely(IS_ERR(m))) { + if (IS_ERR(m)) { err = PTR_ERR(m); m = NULL; goto err_out; } - ctx = get_attr_search_ctx(ndir, m); + ctx = ntfs_attr_get_search_ctx(ndir, m); if (unlikely(!ctx)) { err = -ENOMEM; goto err_out; @@ -1133,15 +1173,36 @@ static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir) /* Get the offset into the index root attribute. */ ir_pos = (s64)fpos; /* Find the index root attribute in the mft record. */ - if (unlikely(!lookup_attr(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, - NULL, 0, ctx))) { + err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL, + 0, ctx); + if (unlikely(err)) { ntfs_error(sb, "Index root attribute missing in directory " "inode 0x%lx.", vdir->i_ino); goto err_out; } - /* Get to the index root value (it's been verified in read_inode). */ - ir = (INDEX_ROOT*)((u8*)ctx->attr + - le16_to_cpu(ctx->attr->data.resident.value_offset)); + /* + * Copy the index root attribute value to a buffer so that we can put + * the search context and unmap the mft record before calling the + * filldir() callback. We need to do this because of NFSd which calls + * ->lookup() from its filldir callback() and this causes NTFS to + * deadlock as ntfs_lookup() maps the mft record of the directory and + * we have got it mapped here already. The only solution is for us to + * unmap the mft record here so that a call to ntfs_lookup() is able to + * map the mft record without deadlocking. + */ + rc = le32_to_cpu(ctx->attr->data.resident.value_length); + ir = (INDEX_ROOT*)kmalloc(rc, GFP_NOFS); + if (unlikely(!ir)) { + err = -ENOMEM; + goto err_out; + } + /* Copy the index root value (it has been verified in read_inode). */ + memcpy(ir, (u8*)ctx->attr + + le16_to_cpu(ctx->attr->data.resident.value_offset), rc); + ntfs_attr_put_search_ctx(ctx); + unmap_mft_record(ndir); + ctx = NULL; + m = NULL; index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length); /* The first index entry. */ ie = (INDEX_ENTRY*)((u8*)&ir->index + @@ -1152,9 +1213,9 @@ static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir) * or signals an error (both covered by the rc test). */ for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) { - ntfs_debug("In index root, offset 0x%x.", (u8*)ie - (u8*)ir); + ntfs_debug("In index root, offset 0x%zx.", (u8*)ie - (u8*)ir); /* Bounds checks. */ - if (unlikely((u8*)ie < (u8*)ctx->mrec || (u8*)ie + + if (unlikely((u8*)ie < (u8*)ir || (u8*)ie + sizeof(INDEX_ENTRY_HEADER) > index_end || (u8*)ie + le16_to_cpu(ie->key_length) > index_end)) @@ -1165,24 +1226,19 @@ static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir) /* Skip index root entry if continuing previous readdir. */ if (ir_pos > (u8*)ie - (u8*)ir) continue; + /* Advance the position even if going to skip the entry. */ + fpos = (u8*)ie - (u8*)ir; /* Submit the name to the filldir callback. */ - rc = ntfs_filldir(vol, &fpos, ndir, INDEX_TYPE_ROOT, ir, ie, - name, dirent, filldir); + rc = ntfs_filldir(vol, fpos, ndir, NULL, ie, name, dirent, + filldir); if (rc) { - put_attr_search_ctx(ctx); - unmap_mft_record(ndir); + kfree(ir); goto abort; } } - /* - * We are done with the index root and the mft record for that matter. - * We need to release it, otherwise we deadlock on ntfs_attr_iget() - * and/or ntfs_read_page(). - */ - put_attr_search_ctx(ctx); - unmap_mft_record(ndir); - m = NULL; - ctx = NULL; + /* We are done with the index root and can free the buffer. */ + kfree(ir); + ir = NULL; /* If there is no index allocation attribute we are finished. */ if (!NInoIndexAllocPresent(ndir)) goto EOD; @@ -1196,9 +1252,9 @@ skip_index_root: ia_mapping = vdir->i_mapping; bmp_vi = ndir->itype.index.bmp_ino; if (unlikely(!bmp_vi)) { - ntfs_debug("Inode %lu, regetting index bitmap.", vdir->i_ino); + ntfs_debug("Inode 0x%lx, regetting index bitmap.", vdir->i_ino); bmp_vi = ntfs_attr_iget(vdir, AT_BITMAP, I30, 4); - if (unlikely(IS_ERR(bmp_vi))) { + if (IS_ERR(bmp_vi)) { ntfs_error(sb, "Failed to get bitmap attribute."); err = PTR_ERR(bmp_vi); goto err_out; @@ -1220,10 +1276,10 @@ get_next_bmp_page: ntfs_debug("Reading bitmap with page index 0x%llx, bit ofs 0x%llx", (unsigned long long)bmp_pos >> (3 + PAGE_CACHE_SHIFT), (unsigned long long)bmp_pos & - ((PAGE_CACHE_SIZE * 8) - 1)); + (unsigned long long)((PAGE_CACHE_SIZE * 8) - 1)); bmp_page = ntfs_map_page(bmp_mapping, bmp_pos >> (3 + PAGE_CACHE_SHIFT)); - if (unlikely(IS_ERR(bmp_page))) { + if (IS_ERR(bmp_page)) { ntfs_error(sb, "Reading index bitmap failed."); err = PTR_ERR(bmp_page); bmp_page = NULL; @@ -1255,19 +1311,22 @@ find_next_index_buffer: /* If the current index buffer is in the same page we reuse the page. */ if ((prev_ia_pos & PAGE_CACHE_MASK) != (ia_pos & PAGE_CACHE_MASK)) { prev_ia_pos = ia_pos; - if (likely(ia_page != NULL)) + if (likely(ia_page != NULL)) { + unlock_page(ia_page); ntfs_unmap_page(ia_page); + } /* * Map the page cache page containing the current ia_pos, * reading it from disk if necessary. */ ia_page = ntfs_map_page(ia_mapping, ia_pos >> PAGE_CACHE_SHIFT); - if (unlikely(IS_ERR(ia_page))) { + if (IS_ERR(ia_page)) { ntfs_error(sb, "Reading index allocation data failed."); err = PTR_ERR(ia_page); ia_page = NULL; goto err_out; } + lock_page(ia_page); kaddr = (u8*)page_address(ia_page); } /* Get the current index buffer. */ @@ -1279,6 +1338,14 @@ find_next_index_buffer: "inode 0x%lx or driver bug.", vdir->i_ino); goto err_out; } + /* Catch multi sector transfer fixup errors. */ + if (unlikely(!ntfs_is_indx_record(ia->magic))) { + ntfs_error(sb, "Directory index record with vcn 0x%llx is " + "corrupt. Corrupt inode 0x%lx. Run chkdsk.", + (unsigned long long)ia_pos >> + ndir->itype.index.vcn_size_bits, vdir->i_ino); + goto err_out; + } if (unlikely(sle64_to_cpu(ia->index_block_vcn) != (ia_pos & ~(s64)(ndir->itype.index.block_size - 1)) >> ndir->itype.index.vcn_size_bits)) { @@ -1331,8 +1398,8 @@ find_next_index_buffer: */ for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) { ntfs_debug("In index allocation, offset 0x%llx.", - (unsigned long long)ia_start + ((u8*)ie - - (u8*)ia)); + (unsigned long long)ia_start + + (unsigned long long)((u8*)ie - (u8*)ia)); /* Bounds checks. */ if (unlikely((u8*)ie < (u8*)ia || (u8*)ie + sizeof(INDEX_ENTRY_HEADER) > index_end || @@ -1345,10 +1412,21 @@ find_next_index_buffer: /* Skip index block entry if continuing previous readdir. */ if (ia_pos - ia_start > (u8*)ie - (u8*)ia) continue; - /* Submit the name to the filldir callback. */ - rc = ntfs_filldir(vol, &fpos, ndir, INDEX_TYPE_ALLOCATION, ia, - ie, name, dirent, filldir); + /* Advance the position even if going to skip the entry. */ + fpos = (u8*)ie - (u8*)ia + + (sle64_to_cpu(ia->index_block_vcn) << + ndir->itype.index.vcn_size_bits) + + vol->mft_record_size; + /* + * Submit the name to the @filldir callback. Note, + * ntfs_filldir() drops the lock on @ia_page but it retakes it + * before returning, unless a non-zero value is returned in + * which case the page is left unlocked. + */ + rc = ntfs_filldir(vol, fpos, ndir, ia_page, ie, name, dirent, + filldir); if (rc) { + /* @ia_page is already unlocked in this case. */ ntfs_unmap_page(ia_page); ntfs_unmap_page(bmp_page); goto abort; @@ -1356,8 +1434,10 @@ find_next_index_buffer: } goto find_next_index_buffer; unm_EOD: - if (ia_page) + if (ia_page) { + unlock_page(ia_page); ntfs_unmap_page(ia_page); + } ntfs_unmap_page(bmp_page); EOD: /* We are finished, set fpos to EOD. */ @@ -1377,12 +1457,16 @@ done: err_out: if (bmp_page) ntfs_unmap_page(bmp_page); - if (ia_page) + if (ia_page) { + unlock_page(ia_page); ntfs_unmap_page(ia_page); + } + if (ir) + kfree(ir); if (name) kfree(name); if (ctx) - put_attr_search_ctx(ctx); + ntfs_attr_put_search_ctx(ctx); if (m) unmap_mft_record(ndir); if (!err) @@ -1417,10 +1501,69 @@ static int ntfs_dir_open(struct inode *vi, struct file *filp) return 0; } +#ifdef NTFS_RW + +/** + * ntfs_dir_fsync - sync a directory to disk + * @filp: directory to be synced + * @dentry: dentry describing the directory to sync + * @datasync: if non-zero only flush user data and not metadata + * + * Data integrity sync of a directory to disk. Used for fsync, fdatasync, and + * msync system calls. This function is based on file.c::ntfs_file_fsync(). + * + * Write the mft record and all associated extent mft records as well as the + * $INDEX_ALLOCATION and $BITMAP attributes and then sync the block device. + * + * If @datasync is true, we do not wait on the inode(s) to be written out + * but we always wait on the page cache pages to be written out. + * + * Note: In the past @filp could be NULL so we ignore it as we don't need it + * anyway. + * + * Locking: Caller must hold i_sem on the inode. + * + * TODO: We should probably also write all attribute/index inodes associated + * with this inode but since we have no simple way of getting to them we ignore + * this problem for now. We do write the $BITMAP attribute if it is present + * which is the important one for a directory so things are not too bad. + */ +static int ntfs_dir_fsync(struct file *filp, struct dentry *dentry, + int datasync) +{ + struct inode *vi = dentry->d_inode; + ntfs_inode *ni = NTFS_I(vi); + int err, ret; + + ntfs_debug("Entering for inode 0x%lx.", vi->i_ino); + BUG_ON(!S_ISDIR(vi->i_mode)); + if (NInoIndexAllocPresent(ni) && ni->itype.index.bmp_ino) + write_inode_now(ni->itype.index.bmp_ino, !datasync); + ret = ntfs_write_inode(vi, 1); + write_inode_now(vi, !datasync); + err = sync_blockdev(vi->i_sb->s_bdev); + if (unlikely(err && !ret)) + ret = err; + if (likely(!ret)) + ntfs_debug("Done."); + else + ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx. Error " + "%u.", datasync ? "data" : "", vi->i_ino, -ret); + return ret; +} + +#endif /* NTFS_RW */ + struct file_operations ntfs_dir_ops = { .llseek = generic_file_llseek, /* Seek inside directory. */ .read = generic_read_dir, /* Return -EISDIR. */ .readdir = ntfs_readdir, /* Read directory contents. */ +#ifdef NTFS_RW + .fsync = ntfs_dir_fsync, /* Sync a directory to disk. */ + /*.aio_fsync = ,*/ /* Sync all outstanding async + i/o operations on a kiocb. */ +#endif /* NTFS_RW */ + /*.ioctl = ,*/ /* Perform function on the + mounted filesystem. */ .open = ntfs_dir_open, /* Open directory. */ }; -