If upperrootvp is doomed by a concurrent unmount, unionfs_nodeget()
may return without a reference or lock on it. unionfs_domount() must
prevent the vnode from being recycled for use by a different file until
it is finished with the vnode, namely once vfs_register_upper_from_vp()
fails. Accomplish this by holding the reference returned by namei()
a bit longer.
Reviewed by: kib, markj
Tested by: pho
Differential Revision: https://reviews.freebsd.org/D39767
If either the lower or upper vnode is found to be doomed after
locking it, the newly-created unionfs node won't be associated
with it and its lock will be dropped. In that case, clear the
uppervp and lowervp locals as necessary to avoid further use
of the vnode in unionfs_nodeget(). If the upper vnode is doomed
but the lower vnode remains valid, additionally reset the unionfs
node's v_vnlock field to point to the lower vnode lock.
Reviewed by: kib, markj
Tested by: pho
Differential Revision: https://reviews.freebsd.org/D39767
If the i-node number (d_fileno) for a file on the server did
not fit in 32bits, it would be truncated to the low order 32bits
for the NFSv3 Readdir and ReaddirPlus RPC replies.
This is no longer correct, given that ino_t is now 64bits.
This patch fixes this by sending the full 64bits of d_fileno
on the wire in the NFSv3 Readdir/ReaddirPlus RPC reply.
PR: 271174
Reported by: bmueller@panasas.com
Tested by: bmueller@panasas.com
MFC after: 2 weeks
The pseudo file systems (devfs, fdescfs, procfs, etc.) report total
and available blocks and inodes despite being synthetic with no
underlying storage device to which those values could be applied.
The current code of these file systems tends to report a fixed number
of total blocks but no free blocks, and in the case of procfs,
libprocfs, linsysfs also no free inodes.
This can be irritating in e.g. the "df" output, since 100% of the
resources seem to be in use, but it can also create warnings in
monitoring tools used for capacity management.
This patch makes these file systems return the same value for the
total and free parameters, leading to 0% in use being displayed by
"df". Since there is no resource that can be exhausted, this appears
to be a sensible result.
Reviewed by: mckusick
Differential Revision: https://reviews.freebsd.org/D39442
Some combinations of FAT12 file system parameters could cause a kernel
panic due to an unmapped access if the size of the FAT was larger than
the CPU page size. The reason is that FAT12 uses 3 bytes to store
2 FAT pointers, leading to partial FAT pointers at the end of buffers
of a size that is not a multiple of 3.
With a typical page size of 4 KB, this caused the FAT entry at byte
offsets 4095 and 4096 to cross the page boundary, with only the first
page mapped. This was fixed by adjusting the mapping to always cover
both bytes of each FAT entry.
Testing revealed 2 other inconsistencies that are fixed by this commit:
1) The calculation of the size of the data area did not take into
account the fact that the first two data block numbers are reserved
and that the data area starts with block 2. This could cause a
FAT12 file system created with the maximum supported number of
blocks to be incorrectly identified as FAT16.
2) The root directory does not take up space in the data area of a
FAT12 or FAT16 file system, since it is placed into a reserved
area outside of that data area. This commits makes stat() report
the logical size of the root directory, but with 0 blocks allocated
from the data area.
PR: 270587
Reviewed by: mckusick
Differential Revision: https://reviews.freebsd.org/D39386
The underlying VOP_MKDIR() implementation may temporarily drop the
parent directory vnode's lock. If the vnode is reclaimed during that
window, the unionfs vnode will effectively become unlocked because
the its v_vnlock field will be reset. To uphold the locking
requirements of VOP_MKDIR() and to avoid triggering various VFS
assertions, explicitly re-lock the unionfs vnode before returning
in this case.
Note that there are almost certainly other cases in which we'll
similarly need to handle vnode relocking by the underlying FS; this
is the only one that's caused problems in stress testing so far.
A more general solution, such as that employed for nullfs in
null_bypass(), will likely need to be implemented.
Tested by: pho
Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D39272
The implementation is racy; if the unionfs vnode is not in fact
locked, vnode private data may be concurrently altered or freed.
Instead, simply rely upon the standard implementation to query the
v_vnlock field, which is type-stable and will reflect the correct
lower/upper vnode configuration for the unionfs node.
Tested by: pho
Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D39272
We hold the vnode interlock, so vnode private data cannot suddenly
become NULL.
Tested by: pho
Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D39272
The LK_UPGRADE operation may have temporarily dropped the upper or
lower vnode's lock. If the unionfs vnode was reclaimed during that
window, its lock field will be reset to no longer point at the
upper/lower vnode lock, so the lock operation will use the standard
lock stored in v_lock. Remove LK_UPGRADE from the flags in this case
to avoid a lockmgr assertion, as this lock has not been previously
owned by the calling thread.
Reported by: pho
Tested by: pho
Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D39272
Use it when possible, instead of separated flags.
No functional change intended.
Reviewed by: hselasky, erj
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D39466
If either of vnodes is shared locked, lock must not be recursed.
Requested by: rmacklem
Reviewed by: markj, rmacklem
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D39444
Although the NFS client does not currently perform Null RPCs,
this fix is needed if/when it might do so.
Found during testing of experimental code that uses Null RPCs
to maintain/monitor TCP connections for "nconnect" mounts.
MFC after: 3 months
Commit f4179ad46f added support for operation bitmaps for
NFSv4.1/4.2. This commit uses those to implement the SP4_MACH_CRED
case for the NFSv4.1/4.2 ExchangeID operation since the Linux
NFSv4.1/4.2 client is now using this for Kerberized mounts.
The Linux Kerberized NFSv4.1/4.2 mounts currently work without
support for this because Linux will fall back to SP4_NONE,
but there is no guarantee this fallback will work forever.
This commit only affects Kerberized NFSv4.1/4.2 mounts from
Linux at this time.
MFC after: 3 months
This will be used later in the linsysfs module to filter out VNETs.
Reviewed by: des
Differential revision: https://reviews.freebsd.org/D39382
MFC after: 1 month
Since 81167243b the size of struct pfs_node is 280 bytes, so the kernel
memory allocator takes memory from 384 bytes sized bucket. However, the
length of the node name is mostly short, e.g., for Linux emulation layer
it is up to 16 bytes. The size of struct pfs_node w/o pfs_name is 152
bytes, i.e., we have 104 bytes left to fit the node name into the 256
bytes-sized bucket.
Reviewed by: des
Differential revision: https://reviews.freebsd.org/D39381
MFC after: 1 month
NFSv4.1/4.2 uses operation bitmaps for various operations,
such as the SP4_MACH_CRED case for ExchangeID.
This patch adds support for operation bitmaps so that
support for SP4_MACH_CRED can be added to the NFSv4.1/4.2
server in a future commit.
This commit should not change any NFSv4.1/4.2 semantics.
MFC after: 3 months
This update implements tallying of free directory entries during
create, delete, or rename operations on FAT12 and FAT16 file systems.
Prior to this change, the total number of root directory entries
was reported as number of inodes, but 0 as the number of free
inodes, causing system health monitoring software to warn about
a suspected disk full issue.
The FAT12 and FAT16 file systems provide a limited number of
root directory entries, e.g. 512 on typical hard disk formats.
The valid range of values is 1 to 65535, but the msdosfs code
will effectively round up "odd" values to the next multiple of 16
(e.g. 513 would allow for 528 root directory entries).
This update implements tracking of directory entries during create,
delete, or rename operations, with initial values determined by
scanning the directory when the file system is mounted.
Total and free directory entries are reported in the f_files and
f_ffree elements of struct statfs, despite differences in semantics
of these values:
- There is no limit on the number of files and directories that can
be created on a FAT file system. Only the root directory of FAT12
and FAT16 file systems is limited, any number of files can still be
created in sub-directories, even when 0 free "inodes" are reported.
- A single file can require 1 to 21 directory entries, depending on
the character set, structure, and length of the name. The DOS 8.3
style file name takes up 1 entry, and if the name does not comply
with the syntax of a DOS 8.3 file name, 1 additional entry is used
for each 13 characters of the file name. Since all these entries
have to be contiguous, it is possible that a file or directory with
a long name can not be created, despite a sufficient total number of
free directory entries.
- Renaming a file can require more directory entries than currently
allocated to store its long name, which may prevent an in-place
update of the name if more entries are needed. This may cause a
rename operation to fail if no contiguous range of free entries for
the new name can be found.
- The volume label is stored in a directory entry. An empty FAT file
system with a volume label will therefore show 1 used "inode" in
df.
- The perceentage of free inodes shown in df or monitoring tools does
only represent the state of the root directory of a FAT12 or FAT16
file system. Neither does a reported value of 0% free inodes does
prevent files from being created in sub-directories, nor does a
value of 50% free inodes guarantee that even a single file with
a "long" name can be created in the root directory (if every other
directory entry is occupied and there are no 2 contiguous entries).
The statfs(2) and df(1) man pages have been updated with a notice
regarding the possibly different semantics of values reported as
total and free inodes for non-Unix file systems.
PR: 270053
Reported by: Ben Woods <woodsb02@freebsd.org>
Approved by: mckusick
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D38987
Coverity does not like code that checks a function's
return value sometimes. Add "(void)" in front of the
function when the return value does not matter to try
and make it happy.
A recent commit deleted "(void)"s in front of nfsm_fhtom().
This commit puts them back in.
Reported by: emaste
MFC after: 3 months
We default to passing the path of the tar file to vfs_mountedfrom
so we can tell where a filesystem was mounted from.
However this can make the output of mount(8) hard to read.
Allow things like:
mount -t tarfs -o as=`basename $tar` $tar /mnt
so "as" is recorded instead of $tar
Reviewed by: des
Sponsored by: Juniper Networks
Differential Revision: https://reviews.freebsd.org/D39273
Coverity does not like code that checks a function's
return value sometimes. Add "(void)" in front of the
function when the return value does not matter to try
and make it happy.
Reported by: emaste
MFC after: 3 months
The ix number for the fdescfs root is 1, while any fd vnode has the ix
value at least 3.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D39207
It is already referenced by the VOP_LOOKUP() caller, otherwise vdrop()
after vn_lock() is invalid anyway.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D39207
Just owning the interlock is not enough for vget() to operate on the
vnode race-free with vgone(), the vnode should be held. Use
vget_prep()/vget_finish() to avoid vholding the vnode explicitly, and
drop LK_INTERLOCK.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D39207
In fdesc_lookup(), vn_vget_ino_gen() may fail without invoking the
callback, in which case the ref on fp is leaked. This happens if the
fdescfs mount is being concurrently unmounted. Moreover, we cannot
safely drop the ref while the dvp is locked.
So:
- Use a flag variable to indicate whether the ref is dropped.
- Reorganize things to handle the leak.
Reported by: C Turt <ecturt@gmail.com>
Reviewed by: mjg, kib
Tested by: pho
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D39189
The panic() will be called under ext2_dirbad()
function in case of rw mount. It cause user confusion,
like in BZ 265951.
PR: 265951
Reviewed by: pfg, mckusick
MFC after: 2 week
Differential revision: https://reviews.freebsd.org/D38503
Add check that directory entry with ino=EXT2_ROOTINO
have correct namelength and name. It is possible to
create malicious image which will cause panic if root
directory entry have incorrect name.
PR: 259068
Reported by: Robert Morris
Reviewed by: pfg
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D38502
Without this patch, a Kerberized NFSv4.1/4.2 mount must provide
a Kerberos credential for the client at mount time. This credential
is typically referred to as a "machine credential". It can be
created one of two ways:
- The user (usually root) has a valid TGT at the time the mount
is done and this becomes the machine credential.
There are two problems with this.
1 - The user doing the mount must have a valid TGT for a user
principal at mount time. As such, the mount cannot be put
in fstab(5) or similar.
2 - When the TGT expires, the mount breaks.
- The client machine has a service principal in its default keytab
file and this service principal (typically called a host-based
initiator credential) is used as the machine credential.
There are problems with this approach as well:
1 - There is a certain amount of administrative overhead creating
the service principal for the NFS client, creating a keytab
entry for this principal and then copying the keytab entry
into the client's default keytab file via some secure means.
2 - The NFS client must have a fixed, well known, DNS name, since
that FQDN is in the service principal name as the instance.
This patch uses a feature of NFSv4.1/4.2 called SP4_NONE, which
allows the state maintenance operations to be performed by any
authentication mechanism, to do these operations via AUTH_SYS
instead of RPCSEC_GSS (Kerberos). As such, neither of the above
mechanisms is needed.
It is hoped that this option will encourage adoption of Kerberized
NFS mounts using TLS, to provide a more secure NFS mount.
This new NFSv4.1/4.2 mount option, called "syskrb5" must be used
with "sec=krb5[ip]" to avoid the need for either of the above
Kerberos setups to be done by the client.
Note that all file access/modification operations still require
users on the NFS client to have a valid TGT recognized by the
NFSv4.1/4.2 server. As such, this option allows, at most, a
malicious client to do some sort of DOS attack.
Although not required, use of "tls" with this new option is
encouraged, since it provides on-the-wire encryption plus,
optionally, client identity verification via a X.509
certificate provided to the server during TLS handshake.
Alternately, "sec=krb5p" does provide on-the-wire
encryption of file data.
A mount_nfs(8) man page update will be done in a separate commit.
Discussed on: freebsd-current@
MFC after: 3 months
Building with -DMSDOSFS_DEBUG failed due to a format mismatch and
a variable that has been renamed but not updated in the printf()
parameter list.
MFC after: 1 month
The sysctls were in "struct fha_params", making it impractical
to vnet them. This patch moves the sysctls out of "struct fha_params"
and vnet's them, so that they are available in vnet prisons
running nfsd(8). It also avoids the IS_DEFAULT_VNET() macros
in the VNET_SYSINIT() function.
Reviewed by: glebius
MFC after: 3 months
Differential Revision: https://reviews.freebsd.org/D38748
The Kasan tests show the nfsrvd_cleancache() results
in a modify after free. I think this occurs because the
nfsrv_cleanup() function gets executed after nfs_cleanup()
which free's the nfsstatsv1_p.
This patch makes them use the same subsystem and sets
SI_ORDER_FIRST for nfs_cleanup(), so that it will be called
after nfsrv_cleanup() via VNET_SYSUNINIT().
The patch also sets nfsstatsv1_p NULL after free'ng it,
so that a crash will result if it is used after free'ng.
Tested by: markj
Reviewed by: markj
MFC after: 3 months
Differential Revision: https://reviews.freebsd.org/D38750
The nfsd(8) daemon generates an error message that does not
indicate that the nfsd daemon is already running when the nfssvc(2)
syscall fails for the NFSSVC_STABLERESTART. Also, the check for
running nfsd(8) in a vnet prison will return EPERM when it fails.
This patch replaces EPERM with ENXIO so that the nfsd(8) daemon
can generate more reasonable failure messages. The nfsd(8) daemon
will be patched in a future commit.
MFC after: 3 months