Re: git: 076002f24d35 - main - Do comprehensive UFS/FFS superblock integrity checks when reading a superblock.
Date: Mon, 30 May 2022 22:07:33 UTC
In message <20220530215552.EE5432C@slippy.cwsent.com>, Cy Schubert writes:
> In message <202205271922.24RJMOJ2039923@gitrepo.freebsd.org>, Kirk McKusick
> wri
> tes:
> > The branch main has been updated by mckusick:
> >
> > URL: https://cgit.FreeBSD.org/src/commit/?id=076002f24d35962f0d21f44bfddd34
> ee
> > 4d7f015d
> >
> > commit 076002f24d35962f0d21f44bfddd34ee4d7f015d
> > Author: Kirk McKusick <mckusick@FreeBSD.org>
> > AuthorDate: 2022-05-27 19:21:11 +0000
> > Commit: Kirk McKusick <mckusick@FreeBSD.org>
> > CommitDate: 2022-05-27 19:22:07 +0000
> >
> > Do comprehensive UFS/FFS superblock integrity checks when reading a sup
> er
> > block.
> >
> > Historically only minimal checks were made of a superblock when it
> > was read in as it was assumed that fsck would have been run to
> > correct any errors before attempting to use the filesystem. Recently
> > several bug reports have been submitted reporting kernel panics
> > that can be triggered by deliberately corrupting filesystem superblocks
> ,
> > see Bug 263979 - [meta] UFS / FFS / GEOM crash (panic) tracking
> > which is tracking the reported corruption bugs.
> >
> > This change upgrades the checks that are performed. These additional
> > checks should prevent panics from a corrupted superblock. Although
> > it appears in only one place, the new code will apply to the kernel
> > modules and (through libufs) user applications that read in superblocks
> .
> >
> > Reported by: Robert Morris and Neeraj
> > Reviewed by: kib
> > Tested by: Peter Holm
> > PR: 263979
> > MFC after: 1 month
> > Differential Revision: https://reviews.freebsd.org/D35219
> > ---
> > sys/ufs/ffs/ffs_subr.c | 163 +++++++++++++++++++++++++++++++++++++++++++--
> --
> > --
> > 1 file changed, 146 insertions(+), 17 deletions(-)
> >
> > diff --git a/sys/ufs/ffs/ffs_subr.c b/sys/ufs/ffs/ffs_subr.c
> > index 01e9f45e1205..28c2fee1cb37 100644
> > --- a/sys/ufs/ffs/ffs_subr.c
> > +++ b/sys/ufs/ffs/ffs_subr.c
> > @@ -35,6 +35,7 @@
> > __FBSDID("$FreeBSD$");
> >
> > #include <sys/param.h>
> > +#include <sys/limits.h>
> >
> > #ifndef _KERNEL
> > #include <stdio.h>
> > @@ -50,6 +51,7 @@ uint32_t ffs_calc_sbhash(struct fs *);
> > struct malloc_type;
> > #define UFS_MALLOC(size, type, flags) malloc(size)
> > #define UFS_FREE(ptr, type) free(ptr)
> > +#define maxphys MAXPHYS
> >
> > #else /* _KERNEL */
> > #include <sys/systm.h>
> > @@ -125,6 +127,7 @@ ffs_update_dinode_ckhash(struct fs *fs, struct ufs2_din
> od
> > e *dip)
> > static off_t sblock_try[] = SBLOCKSEARCH;
> > static int readsuper(void *, struct fs **, off_t, int, int,
> > int (*)(void *, off_t, void **, int));
> > +static int validate_sblock(struct fs *, int);
> >
> > /*
> > * Read a superblock from the devfd device.
> > @@ -141,7 +144,7 @@ static int readsuper(void *, struct fs **, off_t, int,
> in
> > t,
> > * EIO: non-existent or truncated superblock.
> > * EIO: error reading summary information.
> > * ENOENT: no usable known superblock found.
> > - * ENOSPC: failed to allocate space for the superblock.
> > + * ENOMEM: failed to allocate space for the superblock.
> > * EINVAL: The previous newfs operation on this volume did not complet
> e.
> > * The administrator must complete newfs before using this volume.
> > */
> > @@ -152,7 +155,8 @@ ffs_sbget(void *devfd, struct fs **fsp, off_t altsblock
> ,
> > {
> > struct fs *fs;
> > struct fs_summary_info *fs_si;
> > - int i, error, size, blks;
> > + int i, error;
> > + uint64_t size, blks;
> > uint8_t *space;
> > int32_t *lp;
> > char *buf;
> > @@ -190,17 +194,16 @@ ffs_sbget(void *devfd, struct fs **fsp, off_t altsblo
> ck
> > ,
> > if (fs->fs_contigsumsize > 0)
> > size += fs->fs_ncg * sizeof(int32_t);
> > size += fs->fs_ncg * sizeof(u_int8_t);
> > - /* When running in libufs or libsa, UFS_MALLOC may fail */
> > - if ((fs_si = UFS_MALLOC(sizeof(*fs_si), filltype, M_WAITOK)) == NULL) {
> > + if ((fs_si = UFS_MALLOC(sizeof(*fs_si), filltype, M_NOWAIT)) == NULL) {
> > UFS_FREE(fs, filltype);
> > - return (ENOSPC);
> > + return (ENOMEM);
> > }
> > bzero(fs_si, sizeof(*fs_si));
> > fs->fs_si = fs_si;
> > - if ((space = UFS_MALLOC(size, filltype, M_WAITOK)) == NULL) {
> > + if ((space = UFS_MALLOC(size, filltype, M_NOWAIT)) == NULL) {
> > UFS_FREE(fs->fs_si, filltype);
> > UFS_FREE(fs, filltype);
> > - return (ENOSPC);
> > + return (ENOMEM);
> > }
> > fs->fs_csp = (struct csum *)space;
> > for (i = 0; i < blks; i += fs->fs_frag) {
> > @@ -253,16 +256,8 @@ readsuper(void *devfd, struct fs **fsp, off_t sblocklo
> c,
> > int isaltsblk,
> > fs = *fsp;
> > if (fs->fs_magic == FS_BAD_MAGIC)
> > return (EINVAL);
> > - if (!(((fs->fs_magic == FS_UFS1_MAGIC && (isaltsblk ||
> > - sblockloc <= SBLOCK_UFS1)) ||
> > - (fs->fs_magic == FS_UFS2_MAGIC && (isaltsblk ||
> > - sblockloc == fs->fs_sblockloc))) &&
> > - fs->fs_ncg >= 1 &&
> > - fs->fs_bsize >= MINBSIZE &&
> > - fs->fs_bsize <= MAXBSIZE &&
> > - fs->fs_bsize >= roundup(sizeof(struct fs), DEV_BSIZE) &&
> > - fs->fs_sbsize <= SBLOCKSIZE))
> > - return (ENOENT);
> > + if ((error = validate_sblock(fs, isaltsblk)) != 0)
> > + return (error);
> > /*
> > * If the filesystem has been run on a kernel without
> > * metadata check hashes, disable them.
> > @@ -310,6 +305,140 @@ readsuper(void *devfd, struct fs **fsp, off_t sblockl
> oc
> > , int isaltsblk,
> > return (0);
> > }
> >
> > +/*
> > + * Verify the filesystem values.
> > + */
> > +#define ILOG2(num) (fls(num) - 1)
> > +
> > +static int
> > +validate_sblock(struct fs *fs, int isaltsblk)
> > +{
> > + int i, sectorsize;
> > + u_int64_t maxfilesize, minfpg, sizepb;
> > +
> > + sectorsize = dbtob(1);
> > + if (fs->fs_magic == FS_UFS2_MAGIC) {
> > + if ((!isaltsblk && (fs->fs_sblockloc != SBLOCK_UFS2 ||
> > + fs->fs_sblockactualloc != SBLOCK_UFS2)) ||
> > + fs->fs_maxsymlinklen != ((UFS_NDADDR + UFS_NIADDR) *
> > + sizeof(ufs2_daddr_t)) ||
> > + fs->fs_nindir != fs->fs_bsize / sizeof(ufs2_daddr_t) ||
> > + fs->fs_inopb != fs->fs_bsize / sizeof(struct ufs2_dinode))
> > + return (ENOENT);
> > + } else if (fs->fs_magic == FS_UFS1_MAGIC) {
> > + if ((!isaltsblk && (fs->fs_sblockloc > SBLOCK_UFS1 ||
> > + fs->fs_sblockactualloc != SBLOCK_UFS1)) ||
> > + fs->fs_nindir != fs->fs_bsize / sizeof(ufs1_daddr_t) ||
> > + fs->fs_inopb != fs->fs_bsize / sizeof(struct ufs1_dinode) |
> > |
> > + fs->fs_maxsymlinklen != ((UFS_NDADDR + UFS_NIADDR) *
> > + sizeof(ufs1_daddr_t)) ||
> > + fs->fs_old_inodefmt != FS_44INODEFMT ||
> > + fs->fs_old_cgoffset != 0 ||
> > + fs->fs_old_cgmask != 0xffffffff ||
> > + fs->fs_old_size != fs->fs_size ||
> > + fs->fs_old_rotdelay != 0 ||
> > + fs->fs_old_rps != 60 ||
> > + fs->fs_old_nspf != fs->fs_fsize / sectorsize ||
> > + fs->fs_old_cpg != 1 ||
> > + fs->fs_old_interleave != 1 ||
> > + fs->fs_old_trackskew != 0 ||
> > + fs->fs_old_cpc != 0 ||
> > + fs->fs_old_postblformat != 1 ||
> > + fs->fs_old_nrpos != 1 ||
> > + fs->fs_old_spc != fs->fs_fpg * fs->fs_old_nspf ||
> > + fs->fs_old_nsect != fs->fs_old_spc ||
> > + fs->fs_old_npsect != fs->fs_old_spc ||
> > + fs->fs_old_dsize != fs->fs_dsize ||
> > + fs->fs_old_ncyl != fs->fs_ncg)
> > + return (ENOENT);
> > + } else {
> > + return (ENOENT);
> > + }
> > + if (fs->fs_bsize < MINBSIZE || fs->fs_bsize > MAXBSIZE ||
> > + fs->fs_bsize < roundup(sizeof(struct fs), DEV_BSIZE) ||
> > + fs->fs_sbsize > SBLOCKSIZE || fs->fs_sbsize < fs->fs_fsize ||
> > + !powerof2(fs->fs_bsize))
> > + return (ENOENT);
> > + if (fs->fs_fsize < sectorsize || fs->fs_fsize > fs->fs_bsize ||
> > + fs->fs_fsize * MAXFRAG < fs->fs_bsize || !powerof2(fs->fs_fsize))
> > + return (ENOENT);
> > + if (fs->fs_maxbsize < fs->fs_bsize || !powerof2(fs->fs_maxbsize) ||
> > + fs->fs_maxbsize > FS_MAXCONTIG * fs->fs_bsize)
> > + return (ENOENT);
> > + if (fs->fs_bmask != ~(fs->fs_bsize - 1) ||
> > + fs->fs_fmask != ~(fs->fs_fsize - 1) ||
> > + fs->fs_qbmask != ~fs->fs_bmask ||
> > + fs->fs_qfmask != ~fs->fs_fmask ||
> > + fs->fs_bshift != ILOG2(fs->fs_bsize) ||
> > + fs->fs_fshift != ILOG2(fs->fs_fsize) ||
> > + fs->fs_frag != numfrags(fs, fs->fs_bsize) ||
> > + fs->fs_fragshift != ILOG2(fs->fs_frag) ||
> > + fs->fs_frag > MAXFRAG ||
> > + fs->fs_fsbtodb != ILOG2(fs->fs_fsize / sectorsize))
> > + return (ENOENT);
> > + if (fs->fs_sblkno !=
> > + roundup(howmany(fs->fs_sblockloc + SBLOCKSIZE, fs->fs_fsize),
> > + fs->fs_frag) ||
> > + fs->fs_cblkno != fs->fs_sblkno +
> > + roundup(howmany(SBLOCKSIZE, fs->fs_fsize), fs->fs_frag) ||
> > + fs->fs_iblkno != fs->fs_cblkno + fs->fs_frag ||
> > + fs->fs_dblkno != fs->fs_iblkno + fs->fs_ipg / INOPF(fs) ||
> > + fs->fs_cgsize != fragroundup(fs, CGSIZE(fs)))
> > + return (ENOENT);
> > + if (fs->fs_csaddr != cgdmin(fs, 0) ||
> > + fs->fs_cssize !=
> > + fragroundup(fs, fs->fs_ncg * sizeof(struct csum)) ||
> > + fs->fs_dsize != fs->fs_size - fs->fs_sblkno -
> > + fs->fs_ncg * (fs->fs_dblkno - fs->fs_sblkno) -
> > + howmany(fs->fs_cssize, fs->fs_fsize) ||
> > + fs->fs_metaspace < 0 || fs->fs_metaspace > fs->fs_fpg / 2 ||
> > + fs->fs_minfree > 99)
> > + return (ENOENT);
> > + maxfilesize = fs->fs_bsize * UFS_NDADDR - 1;
> > + for (sizepb = fs->fs_bsize, i = 0; i < UFS_NIADDR; i++) {
> > + sizepb *= NINDIR(fs);
> > + maxfilesize += sizepb;
> > + }
> > + if (fs->fs_maxfilesize != maxfilesize)
> > + return (ENOENT);
> > + /*
> > + * These values have a tight interaction with each other that
> > + * makes it hard to tightly bound them. So we can only check
> > + * that they are within a broader possible range.
> > + *
> > + * Calculate minfpg, the minimum number of fragments that can be
> > + * in a cylinder group. The value 12289 is calculated in newfs(8)
> > + * when creating the smallest block size UFS version 1 filesystem
> > + * (4096 block size) with no fragments (4096 fragment size). That
> > + * number may be depressed even further for very small filesystems
> > + * since newfs(8) strives to have at least four cylinder groups.
> > + */
> > + minfpg = MIN(12289, fs->fs_size / 4);
> > + if (fs->fs_ncg < 1 || fs->fs_ncg > (fs->fs_size / minfpg) + 1 ||
> > + fs->fs_fpg < minfpg || fs->fs_fpg > fs->fs_size ||
> > + fs->fs_ipg * fs->fs_ncg > (((int64_t)(1)) << 32) - INOPB(fs) ||
> > + fs->fs_ipg > fs->fs_fpg || fs->fs_size < 8 * fs->fs_frag)
> > + return (ENOENT);
> > + if (fs->fs_size <= (fs->fs_ncg - 1) * fs->fs_fpg ||
> > + fs->fs_size > fs->fs_ncg * fs->fs_fpg)
> > + return (ENOENT);
> > + /*
> > + * Maxcontig sets the default for the maximum number of blocks
> > + * that may be allocated sequentially. With file system clustering
> > + * it is possible to allocate contiguous blocks up to the maximum
> > + * transfer size permitted by the controller or buffering.
> > + */
> > + if (fs->fs_maxcontig < 1 ||
> > + fs->fs_maxcontig > MAX(1, maxphys / fs->fs_bsize))
> > + return (ENOENT);
> > + if (fs->fs_maxcontig < 0 ||
> > + (fs->fs_maxcontig == 0 && fs->fs_contigsumsize != 0) ||
> > + (fs->fs_maxcontig > 1 &&
> > + fs->fs_contigsumsize != MIN(fs->fs_maxcontig, FS_MAXCONTIG)))
> > + return (ENOENT);
> > + return (0);
> > +}
> > +
> > /*
> > * Write a superblock to the devfd device from the memory pointed to by fs
> .
> > * Write out the superblock summary information if it is present.
> >
>
> Hi Kirk,
>
> This patch broke loader on one of my machines.
>
> I get the following error:
>
> LUA ERROR: Cannot open /boot/lua/loader.lua: no such file or directory"
> can't load kernel
>
> Of course booting from my rescue worked.
>
> Copying loader* from /boot on the rescue disk to /boot works around the
> problem.
>
> Backing up, newfs, and restoring the root filesystem, using my rescue disk
> also didn't work around the problem.
>
> Reverting this patch provided permanent relief.
>
> Loader could read the filesystems on my rescue disk, which had blocksizes
> of 16K but not those on my boot disk which had blocksizes of 32K.
>
> My three machines downstairs, all AMD gear with UFS blocksize of 32K, had
> no no problems with this whereas my laptop, an Intel, had the problem.
>
> To repeat, a newly created filesystem restored from backup using my rescue
> disk didn't resolve the problem.
One more data point that I forgot:
While in the loader prompt, ls against any and all UFS filesystems on the
disk listed that / did not exist. Changing currdev= to another filesystem
also did not work. However when in the loader prompt on the boot disk, I
could ls the UFS filesystems on the rescue disk, which was still attached.
This clearly indicated that libsa could not read the UFS filesystems on the
boot disk but could still read the UFS filesystems on the inactive rescue
disk.
--
Cheers,
Cy Schubert <Cy.Schubert@komquats.com> or <Cy.Schubert@cschubert.com>
FreeBSD UNIX: <cy@FreeBSD.org> Web: http://www.FreeBSD.org
NTP: <cy@nwtime.org> Web: https://nwtime.org
e**(i*pi)+1=0