XFS for Linux Release 1.2 User Space Change Log

The following changes have been made in user space to XFS for Linux for the 1.2 release.

acl-2.0.19

• Correct the description of the access check algorithm in acl(5).
• Add --numeric option to getfacl (so all uid's/gid's are displayed numerically rather than by symbolic name). Also enlarge the static buffer used for numeric identifiers from 12 to 22 characters, which is enough for 64 bit id's.
• Add a test case for setting just a base ACL.
• Make the test/run script return a proper status code.
• Check for ENOATTR and for ENODATA error numbers in places where ENOATTR is expected, so if ENOATTR is assigned a separate number we'll only need to rebuild.
• Setfacl failed to parse ACLs with whitespace around user/group names or ids (e.g., 'u: bin :rw', 'u: 0 :rwx').
• Setfacl tried to retrieve the default ACL of regular files, which resulted in errno =3D EACCES since 2.0.13, and broke setfacl's --restore operation.
• Fix an IA64 alignment issue affecting several libacl functions.
• Updated libacl version number from libacl.so.1.0.2 to libacl.so.1.0.3
• Fix configure warning (autoconf 2.13) - namely: "AC_TRY_RUN called without default to allow cross compiling"
• Fix a couple of typos in the documentation
• Fix a Debian-specific build issue
• Build infrastructure updates so that configure options can be used to specify paths rather than semi-hard-coded path names controlled by ROOT/ROOT_PREFIX env variables; eg. now allows /lib64 and /lib32.
• Make acl_get_file() fail for regular files and type ACL_TYPE_DEFAULT, as required by 1003.1e draft 17. Update manual page accordingly.
• Make acl_set_file() fail for regular files and type ACL_TYPE_DEFAULT, as required by 1003.1e draft 17. Update manual page accordingly.
• Updated libacl version number from libacl.so.1.0.1 to libacl.so.1.0.2.
• Updated chacl(1) command to behave correctly with this libacl change.
• Increment the libacl version so that its possible for other packages like Samba to easily tell which library version has the libacl build bug fixed (without having to play configure games, etc).
• In other words, libacl.so.1.0.0 is now libacl.so.1.0.1; nothing else has changed in this release.
• Fix a bug when reallocating space in __acl_to_any_text.c (the bug causes ACL entries to get skipped!)
• Minor clean-up in the test/setfacl.test script
• Add accidentally removed dependency of libacl on libattr.

attr-2.0.12

• Fix: the m68k architecture is identified by __mc68000__ rather than by __m68k__.
• Correction in the attr(5) manual page.
• Add test cases for empty EA values, and for a bad ACL extended attribute name (there were bugs with these).
• Make the test/run script return a proper status code.
• Fix up the error message returned for ENODATA rather than ENOATTR, so if ENOATR is assigned a separate number we don't mess with ENOATTR's text message.
• Add system call numbers for the m68k architecture
• Add missing "enabling" #define for the Alpha architecture
• Build updates to allow rpmbuild to build rpms (newer versions of rpm require this, as rpm itself can no longer build rpms)
• Several section 3 man page corrections (minor)
• Build infrastructure updates so that configure options can be used to specify paths rather than semi-hard-coded path names controlled by the PREFIX/ROOT_PREFIX environment variables; eg. now allows /lib64 and /lib32
• Add system call numbers for the Alpha architecture
• Minor wording change in the attr(5) manual page
• Additional test cases for malformed attribute names
• Remove leftover build targets html and ps

dmapi-2.0.5

• Build infrastructure updates so that configure options can be used to specify paths rather than semi-hard-coded path names controlled by the ROOT/ROOT_PREFIX environment variables; eg. now allows /lib64 and /lib32
• kernel<>library interface for DM_OPEN_BY_HANDLE changed
• kernel-library interface has changed, now 64bit safe.
• Minor build system updates

xfsdump-2.2.3

• minor update to Debian packaging dependencies.
• inventory related bugfix.
• Remove fork/setuid from fsr. Just chown temp. file to owner. This gets fsr working again with the restricted RESVSP ioctl. Chris Wedgwood
• minor troff-version-specific fixups for xfsrestore(8).
• force use of version 1 XFS geometry ioctl everywhere in xfsdump package, so that these tools can still work on older kernels (they don't need the additional information in the new ioctl).
• added new config.h.in and configure changes to implement this.
• do some additional cleanup after autoconf has run.
• minor build fix when using relatively old versions of glibc which don't provide the SYS_getdents64 symbol.
• xfsinvutil -m option added (merge from IRIX).
• add missing GPL/copyright headers to several xfsinvutil files.
• correct several uses of uuid_compare in xfsinvutil.
• small xfsinvutil fixups to preserve compatibility with old -n option.
• fixed all new xfsinvutil warnings on 64bit platforms
• xfsinvutil interactive mode
• merge of a dump/restore fix from IRIX - test if a child exited with error and if a child did, set the exit_code to EXIT_ERROR.
• build infrastructure updates so that configure options can be used to specify paths rather than semi-hard-coded path names controlled by the ROOT/ROOT_PREFIX environment variables; eg. now allows /lib64 and /lib32
• new fsinfo system call used, mostly changing version to differentiate this.
• fix xfsrestore/xfsdump version output to report both=20 their version and the respective dump format version. They will report their version from cmd/xfsdump/VERSION. =20
• changed the messages printed out for the xfsrestore=20 interactive mode to emphasize the restore is interactive.
• fix xfsdump so that it does not cause segmentation fault when media specific object id (OPT_MOBJID) is specified=20 with -I option.
• fix similar code for filesystem id (OPT_FSID) option.=20
• in xfsdump, check for ERANGE as well as E2BIG for testing of the existence of the SGI_XFSDUMP_SKIP_FILE extended attribute. This only affects dumping using the -e option.=20 This was motivated by an XFS kernel change from returning E2BIG into returning ERANGE.
• fix "xfsdump -v silent" to really be silent

xfsprogs-2.3.5

• Sync up user/kernel source in lib{xfs,xlog} and headers.
• Fix mkfs (libxfs) bug when using BLKFLSBUF ioctl - we passed in the device number instead of a file descriptor (and EBADF was the end result, which we ignored).
• Thanks to Anton Blanchard for fixing this.
• Fix size calculation bug in xfs_repair's memory allocation, reduces memory usage by a factor of 4.
• Fix mkfs bug when optimizing AG size on striped devices, ASSERT(agcount !=3D 0) tripped on single-ag filesystems.
• Use pread/pwrite instead of lseek + read/write.
• Allow xfs_db to accept negative numbers when writing values.
• Several changes to geometry ioctl callers which will make the tools useable on older kernel versions too.
• Mainly affects xfs_bmap and xfs_growfs.
• Do some additional cleanup after autoconf has run.
• Update libdisk for a few other filesystems.
• Fix call to lvdisplay on LVM volumes so that lvdisplay always gets the full, real path to the volume, as required.
• Fix endian bug in xfs_db, was not flipping 16-bit numbers.
• Make xfs_repair recognize a NULLFSINO and print that in addition to "18446744073709551615" for easy recognition.
• Fix format string that xfs_repair uses when fixing inodes, so that the new inode numbers are printed correctly.
• Fix buglet in libdisk get_driver_block_major() routine which would result in incorrect majors being returned for not-found drivers.
• Fix problem in install process with newer autoconf versions.
• mkfs.xfs defaults to a blocksize of 4096 bytes, and no longer uses getpagesize(2) to dynamically configure the default size.
• Add EVMS stripe unit/width auto-detection support in mkfs.xfs (patch from Luciano Chavez).
• Sync user/kernel headers (arch.h incorporated into xfs_arch.h, and several other minor changes).
• Fix up log stripe unit specification options, and update man page. Also fix suffix handling for data su/sw options (disallow unit suffixes on -d sunit,swidth,sw and -l sunit). Add "s" (512-byte sectors) as valid suffix.
• Automatically select v2 logs if a log stripe unit is specified.
• Build infrastructure updates so that configure options can be used to specify paths rather than semi-hard-coded path names controlled by the ROOT/ROOT_PREFIX environment variables; eg. now allows /lib64 and /lib32
• Support for XFS version 2 log format.
• Fix for xfs_repair mangling i8count for dir2_sf directories
• Minor mkfs.xfs man page update for blocksize limits on Linux
• xfs_cred.h split into xfs_acl.h, xfs_cap.h and xfs_mac.h
• Fix typo in mkfs realtime summary inode alloc failure message
• Fix error returns from log recovery (affects xfs_repair).
• Fix the way mkfs.xfs round downs the device when the last=20 AG is smaller than the minimum AG size.
• Fix bug in xfs_repair da_write() routine, which affects filesystems where the data blocksize !=3D naming blocksize (see xfs_info output).
• Fix a use-after-free bug in xfs_repair code for checking version 1 btree-format directories.
• Add checks of the ACL permissions field into xfs_repair.
• Size AGs so that they do not always start on the same=20 part of a striped disk
• Fix an off-by-one error on rounding down an AG that is too small to be an AG
• Don't auto-grow the log to be larger than an AG
• Change the error philosophy for -d su=3D,sw=3D away from forcing=20 the XFS stripe size to match the volume manager stripe size and instead accept, with a warning, the stripe unit & width supplied on the commandline.
• Update xfs_growfs man page
• Don't build libxlog.a with DEBUG enabled
• xfs_db fixes from ASANO Masahiro at NEC
• Minor update sync'ing with kernel changes (less endian swabbing in libxfs code)

misc