d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
216a89d6a4
freebsd-dev/sys/ufs/ffs/ffs_snapshot.c

1077 lines
29 KiB
C
Raw Normal View History

Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
/*
* Copyright 2000 Marshall Kirk McKusick. All Rights Reserved.
*
* Further information about snapshots can be obtained from:
*
* Marshall Kirk McKusick http://www.mckusick.com/softdep/
* 1614 Oxford Street mckusick@mckusick.com
* Berkeley, CA 94709-1608 +1-510-843-9542
* USA
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
* @(#)ffs_snapshot.c 8.11 (McKusick) 7/23/00
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
* $FreeBSD$
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bio.h>
#include <sys/buf.h>
#include <sys/proc.h>
#include <sys/namei.h>
#include <sys/stat.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/resource.h>
#include <sys/resourcevar.h>
#include <sys/vnode.h>
#include <ufs/ufs/extattr.h>
#include <ufs/ufs/quota.h>
#include <ufs/ufs/ufsmount.h>
#include <ufs/ufs/inode.h>
#include <ufs/ufs/ufs_extern.h>
#include <ufs/ffs/fs.h>
#include <ufs/ffs/ffs_extern.h>
#define KERNCRED proc0.p_ucred
Silence a warning.
2000-09-17 19:41:26 +00:00
#define DEBUG 1
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
static int indiracct __P((struct vnode *, struct vnode *, int, ufs_daddr_t,
int, int, int, int));
static int snapacct __P((struct vnode *, ufs_daddr_t *, ufs_daddr_t *));
static int readblock __P((struct buf *, daddr_t));
#ifdef DEBUG
#include <sys/sysctl.h>
int snapdebug = 0;
SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, "");
#endif /* DEBUG */
/*
* Create a snapshot file and initialize it for the filesystem.
*/
int
ffs_snapshot(mp, snapfile)
struct mount *mp;
char *snapfile;
{
ufs_daddr_t rlbn;
ufs_daddr_t lbn, blkno, copyblkno, inoblks[FSMAXSNAP];
int error, cg, snaploc, indiroff, numblks;
int i, size, base, len, loc, inoblkcnt;
int blksperindir, flag = mp->mnt_flag;
The ffs superblock includes a 128-byte region for use by temporary in-core pointers to summary information. An array in this region (fs_csp) could overflow on filesystems with a very large number of cylinder groups (~16000 on i386 with 8k blocks). When this happens, other fields in the superblock get corrupted, and fsck refuses to check the filesystem. Solve this problem by replacing the fs_csp array in 'struct fs' with a single pointer, and add padding to keep the length of the 128-byte region fixed. Update the kernel and userland utilities to use just this single pointer. With this change, the kernel no longer makes use of the superblock fields 'fs_csshift' and 'fs_csmask'. Add a comment to newfs/mkfs.c to indicate that these fields must be calculated for compatibility with older kernels. Reviewed by: mckusick
2001-01-15 18:30:40 +00:00
void *space;
Several small but important fixes for snapshots: 1) Be more tolerant of missing snapshot files by only trying to decrement their reference count if they are registered as active. 2) Fix for snapshots of filesystems with block sizes larger than 8K (from Ollivier Robert <roberto@eurocontrol.fr>). 3) Fix to avoid losing last block in snapshot file when calculating blocks that need to be copied (from Don Coleman <coleman@coleman.org>).
2000-12-19 04:41:09 +00:00
struct fs *copy_fs, *fs = VFSTOUFS(mp)->um_fs;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
struct proc *p = CURPROC;
struct inode *devip, *ip, *xp;
struct buf *bp, *nbp, *ibp;
struct vnode *vp, *devvp;
struct nameidata nd;
struct mount *wrtmp;
struct dinode *dip;
struct vattr vat;
struct cg *cgp;
/*
* Need to serialize access to snapshot code per filesystem.
*/
/*
* Assign a snapshot slot in the superblock.
*/
for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
if (fs->fs_snapinum[snaploc] == 0)
break;
if (snaploc == FSMAXSNAP)
return (ENOSPC);
/*
* Create the snapshot file.
*/
restart:
NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_USERSPACE, snapfile, p);
if ((error = namei(&nd)) != 0)
return (error);
if (nd.ni_vp != NULL) {
vput(nd.ni_vp);
error = EEXIST;
}
if (nd.ni_dvp->v_mount != mp)
error = EXDEV;
if (error) {
NDFREE(&nd, NDF_ONLY_PNBUF);
if (nd.ni_dvp == nd.ni_vp)
vrele(nd.ni_dvp);
else
vput(nd.ni_dvp);
return (error);
}
VATTR_NULL(&vat);
vat.va_type = VREG;
vat.va_mode = S_IRUSR;
vat.va_vaflags |= VA_EXCLUSIVE;
if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp))
wrtmp = NULL;
if (wrtmp != mp)
panic("ffs_snapshot: mount mismatch");
Brain fault, forgot to update ffs_snapshot.c with the new calling convention for vn_start_write.
2000-07-12 00:27:27 +00:00
if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) {
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
NDFREE(&nd, NDF_ONLY_PNBUF);
vput(nd.ni_dvp);
Brain fault, forgot to update ffs_snapshot.c with the new calling convention for vn_start_write.
2000-07-12 00:27:27 +00:00
if ((error = vn_start_write(NULL, &wrtmp,
V_XSLEEP | PCATCH)) != 0)
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
return (error);
goto restart;
}
VOP_LEASE(nd.ni_dvp, p, KERNCRED, LEASE_WRITE);
error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat);
vput(nd.ni_dvp);
if (error) {
NDFREE(&nd, NDF_ONLY_PNBUF);
vn_finished_write(wrtmp);
return (error);
}
vp = nd.ni_vp;
ip = VTOI(vp);
devvp = ip->i_devvp;
devip = VTOI(devvp);
/*
* Allocate and copy the last block contents so as to be able
* to set size to that of the filesystem.
*/
numblks = howmany(fs->fs_size, fs->fs_frag);
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
fs->fs_bsize, KERNCRED, B_CLRBUF, &bp);
if (error)
goto out;
ip->i_size = lblktosize(fs, (off_t)numblks);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if ((error = readblock(bp, numblks - 1)) != 0)
goto out;
bawrite(bp);
/*
* Preallocate critical data structures so that we can copy
* them in without further allocation after we suspend all
* operations on the filesystem. We would like to just release
* the allocated buffers without writing them since they will
* be filled in below once we are ready to go, but this upsets
* the soft update code, so we go ahead and write the new buffers.
*
* Allocate all indirect blocks. Also allocate shadow copies
* for each of the indirect blocks.
*/
for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, p->p_ucred, B_METAONLY, &ibp);
if (error)
goto out;
copyblkno = fragstoblks(fs, dbtofsb(fs, ibp->b_blkno));
bdwrite(ibp);
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)copyblkno),
fs->fs_bsize, p->p_ucred, 0, &nbp);
if (error)
goto out;
bawrite(nbp);
}
/*
* Allocate shadow blocks to copy all of the other snapshot inodes
* so that we will be able to expunge them from this snapshot.
*/
for (loc = 0, inoblkcnt = 0; loc < snaploc; loc++) {
blkno = fragstoblks(fs, ino_to_fsba(fs, fs->fs_snapinum[loc]));
for (i = 0; i < inoblkcnt; i++)
if (inoblks[i] == blkno)
break;
if (i == inoblkcnt) {
inoblks[inoblkcnt++] = blkno;
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, KERNCRED, 0, &nbp);
if (error)
goto out;
bawrite(nbp);
}
}
/*
* Allocate all cylinder group blocks.
*/
for (cg = 0; cg < fs->fs_ncg; cg++) {
error = VOP_BALLOC(vp, (off_t)(cgtod(fs, cg)) << fs->fs_fshift,
fs->fs_bsize, KERNCRED, 0, &nbp);
if (error)
goto out;
bawrite(nbp);
}
/*
* Allocate copies for the superblock and its summary information.
*/
Several small but important fixes for snapshots: 1) Be more tolerant of missing snapshot files by only trying to decrement their reference count if they are registered as active. 2) Fix for snapshots of filesystems with block sizes larger than 8K (from Ollivier Robert <roberto@eurocontrol.fr>). 3) Fix to avoid losing last block in snapshot file when calculating blocks that need to be copied (from Don Coleman <coleman@coleman.org>).
2000-12-19 04:41:09 +00:00
if ((error = VOP_BALLOC(vp, (off_t)(SBOFF), SBSIZE, KERNCRED, 0, &nbp)))
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
goto out;
bawrite(nbp);
blkno = fragstoblks(fs, fs->fs_csaddr);
len = howmany(fs->fs_cssize, fs->fs_bsize);
for (loc = 0; loc < len; loc++) {
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)),
fs->fs_bsize, KERNCRED, 0, &nbp);
if (error)
goto out;
bawrite(nbp);
}
/*
* Change inode to snapshot type file.
*/
Clean up the snapshot code so that it no longer depends on the use of the SF_IMMUTABLE flag to prevent writing. Instead put in explicit checking for the SF_SNAPSHOT flag in the appropriate places. With this change, it is now possible to rename and link to snapshot files. It is also possible to set or clear any of the owner, group, or other read bits on the file, though none of the write or execute bits can be set. There is also an explicit test to prevent the setting or clearing of the SF_SNAPSHOT flag via chflags() or fchflags(). Note also that the modify time cannot be changed as it needs to accurately reflect the time that the snapshot was taken. Submitted by: Robert Watson <rwatson@FreeBSD.org>
2000-07-26 23:07:01 +00:00
ip->i_flags |= SF_SNAPSHOT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
ip->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* Ensure that the snapshot is completely on disk.
*/
if ((error = VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p)) != 0)
goto out;
/*
* All allocations are done, so we can now snapshot the system.
*
* Suspend operation on filesystem.
*/
for (;;) {
vn_finished_write(wrtmp);
vfs_write_suspend(vp->v_mount);
if (mp->mnt_kern_flag & MNTK_SUSPENDED)
break;
Brain fault, forgot to update ffs_snapshot.c with the new calling convention for vn_start_write.
2000-07-12 00:27:27 +00:00
vn_start_write(NULL, &wrtmp, V_WAIT);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
}
/*
* First, copy all the cylinder group maps. All the unallocated
* blocks are marked BLK_NOCOPY so that the snapshot knows that
* it need not copy them if they are later written.
*/
len = howmany(fs->fs_fpg, fs->fs_frag);
for (cg = 0; cg < fs->fs_ncg; cg++) {
error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
(int)fs->fs_cgsize, KERNCRED, &bp);
if (error) {
brelse(bp);
goto out1;
}
cgp = (struct cg *)bp->b_data;
if (!cg_chkmagic(cgp)) {
brelse(bp);
error = EIO;
goto out1;
}
error = bread(vp, fragstoblks(fs, cgtod(fs, cg)), fs->fs_bsize,
KERNCRED, &nbp);
if (error) {
brelse(bp);
brelse(nbp);
goto out1;
}
bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize);
if (fs->fs_cgsize < fs->fs_bsize)
bzero(&nbp->b_data[fs->fs_cgsize],
fs->fs_bsize - fs->fs_cgsize);
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
nbp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bawrite(nbp);
base = cg * fs->fs_fpg / fs->fs_frag;
Several small but important fixes for snapshots: 1) Be more tolerant of missing snapshot files by only trying to decrement their reference count if they are registered as active. 2) Fix for snapshots of filesystems with block sizes larger than 8K (from Ollivier Robert <roberto@eurocontrol.fr>). 3) Fix to avoid losing last block in snapshot file when calculating blocks that need to be copied (from Don Coleman <coleman@coleman.org>).
2000-12-19 04:41:09 +00:00
if (base + len >= numblks)
len = numblks - base - 1;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
loc = 0;
if (base < NDADDR) {
for ( ; loc < NDADDR; loc++) {
if (!ffs_isblock(fs, cg_blksfree(cgp), loc))
continue;
ip->i_db[loc] = BLK_NOCOPY;
}
}
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)),
fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
if (error) {
brelse(bp);
goto out1;
}
indiroff = (base + loc - NDADDR) % NINDIR(fs);
for ( ; loc < len; loc++, indiroff++) {
if (indiroff >= NINDIR(fs)) {
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
ibp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bawrite(ibp);
error = VOP_BALLOC(vp,
lblktosize(fs, (off_t)(base + loc)),
fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
if (error) {
brelse(bp);
goto out1;
}
indiroff = 0;
}
if (!ffs_isblock(fs, cg_blksfree(cgp), loc))
continue;
Several small but important fixes for snapshots: 1) Be more tolerant of missing snapshot files by only trying to decrement their reference count if they are registered as active. 2) Fix for snapshots of filesystems with block sizes larger than 8K (from Ollivier Robert <roberto@eurocontrol.fr>). 3) Fix to avoid losing last block in snapshot file when calculating blocks that need to be copied (from Don Coleman <coleman@coleman.org>).
2000-12-19 04:41:09 +00:00
if (((ufs_daddr_t *)(ibp->b_data))[indiroff] != 0)
panic("ffs_snapshot: lost block");
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
((ufs_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
}
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(bp);
ibp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bdwrite(ibp);
}
/*
* Snapshot the superblock and its summary information.
*/
Several small but important fixes for snapshots: 1) Be more tolerant of missing snapshot files by only trying to decrement their reference count if they are registered as active. 2) Fix for snapshots of filesystems with block sizes larger than 8K (from Ollivier Robert <roberto@eurocontrol.fr>). 3) Fix to avoid losing last block in snapshot file when calculating blocks that need to be copied (from Don Coleman <coleman@coleman.org>).
2000-12-19 04:41:09 +00:00
if ((error = VOP_BALLOC(vp, SBOFF, SBSIZE, KERNCRED, 0, &nbp)) != 0)
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
goto out1;
Several small but important fixes for snapshots: 1) Be more tolerant of missing snapshot files by only trying to decrement their reference count if they are registered as active. 2) Fix for snapshots of filesystems with block sizes larger than 8K (from Ollivier Robert <roberto@eurocontrol.fr>). 3) Fix to avoid losing last block in snapshot file when calculating blocks that need to be copied (from Don Coleman <coleman@coleman.org>).
2000-12-19 04:41:09 +00:00
copy_fs = (struct fs *)(nbp->b_data + blkoff(fs, SBOFF));
bcopy(fs, copy_fs, fs->fs_sbsize);
copy_fs->fs_clean = 1;
if (fs->fs_sbsize < SBSIZE)
bzero(&nbp->b_data[blkoff(fs, SBOFF) + fs->fs_sbsize],
SBSIZE - fs->fs_sbsize);
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
nbp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bawrite(nbp);
blkno = fragstoblks(fs, fs->fs_csaddr);
len = howmany(fs->fs_cssize, fs->fs_bsize) - 1;
size = fs->fs_bsize;
The ffs superblock includes a 128-byte region for use by temporary in-core pointers to summary information. An array in this region (fs_csp) could overflow on filesystems with a very large number of cylinder groups (~16000 on i386 with 8k blocks). When this happens, other fields in the superblock get corrupted, and fsck refuses to check the filesystem. Solve this problem by replacing the fs_csp array in 'struct fs' with a single pointer, and add padding to keep the length of the 128-byte region fixed. Update the kernel and userland utilities to use just this single pointer. With this change, the kernel no longer makes use of the superblock fields 'fs_csshift' and 'fs_csmask'. Add a comment to newfs/mkfs.c to indicate that these fields must be calculated for compatibility with older kernels. Reviewed by: mckusick
2001-01-15 18:30:40 +00:00
space = fs->fs_csp;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
for (loc = 0; loc <= len; loc++) {
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)),
fs->fs_bsize, KERNCRED, 0, &nbp);
if (error)
goto out1;
if (loc == len) {
readblock(nbp, blkno + loc);
Properly compute the size of the final block of superblock summary information. Submitted by: Ian Dowse <iedowse@maths.tcd.ie>
2001-01-12 21:56:55 +00:00
size = fs->fs_cssize - loc * fs->fs_bsize;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
}
The ffs superblock includes a 128-byte region for use by temporary in-core pointers to summary information. An array in this region (fs_csp) could overflow on filesystems with a very large number of cylinder groups (~16000 on i386 with 8k blocks). When this happens, other fields in the superblock get corrupted, and fsck refuses to check the filesystem. Solve this problem by replacing the fs_csp array in 'struct fs' with a single pointer, and add padding to keep the length of the 128-byte region fixed. Update the kernel and userland utilities to use just this single pointer. With this change, the kernel no longer makes use of the superblock fields 'fs_csshift' and 'fs_csmask'. Add a comment to newfs/mkfs.c to indicate that these fields must be calculated for compatibility with older kernels. Reviewed by: mckusick
2001-01-15 18:30:40 +00:00
bcopy(space, nbp->b_data, size);
space = (char *)space + size;
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
nbp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bawrite(nbp);
}
/*
* Copy the shadow blocks for the snapshot inodes so that
* the copies can can be expunged.
*/
for (loc = 0; loc < inoblkcnt; loc++) {
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)inoblks[loc]),
fs->fs_bsize, KERNCRED, 0, &nbp);
if (error)
goto out1;
readblock(nbp, inoblks[loc]);
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
nbp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bdwrite(nbp);
}
/*
* Copy allocation information from other snapshots and then
* expunge them from the view of the current snapshot.
*/
for (xp = devip->i_copyonwrite; xp; xp = xp->i_copyonwrite) {
/*
* Before expunging a snapshot inode, note all the
* blocks that it claims with BLK_SNAP so that fsck will
* be able to account for those blocks properly and so
* that this snapshot knows that it need not copy them
* if the other snapshot holding them is freed.
*/
if ((error = snapacct(vp, &xp->i_db[0], &xp->i_ib[NIADDR])) !=0)
goto out1;
blksperindir = 1;
lbn = -NDADDR;
len = numblks - NDADDR;
rlbn = NDADDR;
for (i = 0; len > 0 && i < NIADDR; i++) {
error = indiracct(vp, ITOV(xp), i, xp->i_ib[i], lbn,
rlbn, len, blksperindir);
if (error)
goto out1;
blksperindir *= NINDIR(fs);
lbn -= blksperindir + 1;
len -= blksperindir;
rlbn += blksperindir;
}
/*
* Set copied snapshot inode to be a zero length file.
*/
blkno = fragstoblks(fs, ino_to_fsba(fs, xp->i_number));
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, KERNCRED, 0, &nbp);
if (error)
goto out1;
dip = (struct dinode *)nbp->b_data +
ino_to_fsbo(fs, xp->i_number);
dip->di_size = 0;
dip->di_blocks = 0;
Clean up the snapshot code so that it no longer depends on the use of the SF_IMMUTABLE flag to prevent writing. Instead put in explicit checking for the SF_SNAPSHOT flag in the appropriate places. With this change, it is now possible to rename and link to snapshot files. It is also possible to set or clear any of the owner, group, or other read bits on the file, though none of the write or execute bits can be set. There is also an explicit test to prevent the setting or clearing of the SF_SNAPSHOT flag via chflags() or fchflags(). Note also that the modify time cannot be changed as it needs to accurately reflect the time that the snapshot was taken. Submitted by: Robert Watson <rwatson@FreeBSD.org>
2000-07-26 23:07:01 +00:00
dip->di_flags &= ~SF_SNAPSHOT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs_daddr_t));
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
nbp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bdwrite(nbp);
}
/*
* Copy all indirect blocks to their shadows (allocated above)
* to avoid deadlock in ffs_copyonwrite.
*/
for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, p->p_ucred, B_METAONLY, &ibp);
if (error)
goto out1;
copyblkno = fragstoblks(fs, dbtofsb(fs, ibp->b_blkno));
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(ibp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)copyblkno),
fs->fs_bsize, p->p_ucred, 0, &nbp);
if (error)
goto out1;
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, p->p_ucred, B_METAONLY, &ibp);
if (error) {
brelse(nbp);
goto out1;
}
bcopy(ibp->b_data, nbp->b_data, fs->fs_bsize);
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(ibp);
nbp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bawrite(nbp);
}
/*
* Record snapshot inode. Since this is the newest snapshot,
* it must be placed at the end of the list.
*/
fs->fs_snapinum[snaploc] = ip->i_number;
if (ip->i_copyonwrite != 0)
panic("ffs_snapshot: %d already on list", ip->i_number);
if (devip->i_copyonwrite == 0) {
devvp->v_flag |= VCOPYONWRITE;
devip->i_copyonwrite = ip;
} else {
for (xp = devip->i_copyonwrite; xp->i_copyonwrite != 0; )
xp = xp->i_copyonwrite;
xp->i_copyonwrite = ip;
}
vp->v_flag |= VSYSTEM;
/*
* Resume operation on filesystem.
*/
out1:
vfs_write_resume(vp->v_mount);
Brain fault, forgot to update ffs_snapshot.c with the new calling convention for vn_start_write.
2000-07-12 00:27:27 +00:00
vn_start_write(NULL, &wrtmp, V_WAIT);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
out:
mp->mnt_flag = flag;
(void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p);
if (error)
vput(vp);
else
VOP_UNLOCK(vp, 0, p);
vn_finished_write(wrtmp);
return (error);
}
/*
* Descend an indirect block chain for vnode cancelvp accounting for all
* its indirect blocks in snapvp.
*/
static int
indiracct(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks, blksperindir)
struct vnode *snapvp;
struct vnode *cancelvp;
int level;
ufs_daddr_t blkno;
int lbn;
int rlbn;
int remblks;
int blksperindir;
{
int subblksperindir, error, last, num, i;
struct indir indirs[NIADDR + 2];
ufs_daddr_t *bap;
struct buf *bp;
struct fs *fs;
if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
return (error);
if (lbn != indirs[num - 1 - level].in_lbn || blkno == 0 || num < 2)
panic("indiracct: botched params");
/*
* We have to expand bread here since it will deadlock looking
* up the block number for any blocks that are not in the cache.
*/
fs = VTOI(cancelvp)->i_fs;
bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0);
bp->b_blkno = fsbtodb(fs, blkno);
if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
(error = readblock(bp, fragstoblks(fs, blkno)))) {
brelse(bp);
return (error);
}
/*
* Account for the block pointers in this indirect block.
*/
last = howmany(remblks, blksperindir);
if (last > NINDIR(fs))
last = NINDIR(fs);
if (snapvp != cancelvp) {
bap = (ufs_daddr_t *)bp->b_data;
} else {
MALLOC(bap, ufs_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK);
bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(bp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
}
error = snapacct(snapvp, &bap[0], &bap[last]);
if (error || level == 0)
goto out;
/*
* Account for the block pointers in each of the indirect blocks
* in the levels below us.
*/
subblksperindir = blksperindir / NINDIR(fs);
for (lbn++, level--, i = 0; i < last; i++) {
error = indiracct(snapvp, cancelvp, level, bap[i], lbn,
rlbn, remblks, subblksperindir);
if (error)
goto out;
rlbn += blksperindir;
lbn -= blksperindir;
remblks -= blksperindir;
}
out:
if (snapvp != cancelvp)
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(bp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
else
FREE(bap, M_DEVBUF);
return (error);
}
/*
* Account for a set of blocks allocated in a snapshot inode.
*/
static int
snapacct(vp, oldblkp, lastblkp)
struct vnode *vp;
ufs_daddr_t *oldblkp, *lastblkp;
{
struct inode *ip = VTOI(vp);
struct fs *fs = ip->i_fs;
ufs_daddr_t lbn, blkno, *blkp;
struct buf *ibp;
int error;
for ( ; oldblkp < lastblkp; oldblkp++) {
blkno = *oldblkp;
if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
continue;
lbn = fragstoblks(fs, blkno);
if (lbn < NDADDR) {
blkp = &ip->i_db[lbn];
ip->i_flag |= IN_CHANGE | IN_UPDATE;
} else {
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)lbn),
fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
if (error)
return (error);
blkp = &((ufs_daddr_t *)(ibp->b_data))
[(lbn - NDADDR) % NINDIR(fs)];
}
if (*blkp != 0)
panic("snapacct: bad block");
*blkp = BLK_SNAP;
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
if (lbn >= NDADDR) {
ibp->b_flags |= B_VALIDSUSPWRT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
bdwrite(ibp);
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
}
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
}
return (0);
}
Several small but important fixes for snapshots: 1) Be more tolerant of missing snapshot files by only trying to decrement their reference count if they are registered as active. 2) Fix for snapshots of filesystems with block sizes larger than 8K (from Ollivier Robert <roberto@eurocontrol.fr>). 3) Fix to avoid losing last block in snapshot file when calculating blocks that need to be copied (from Don Coleman <coleman@coleman.org>).
2000-12-19 04:41:09 +00:00
/*
* Decrement extra reference on snapshot when last name is removed.
* It will not be freed until the last open reference goes away.
*/
void
ffs_snapgone(ip)
struct inode *ip;
{
struct inode *xp;
/*
* Find snapshot in incore list.
*/
for (xp = VTOI(ip->i_devvp); xp; xp = xp->i_copyonwrite)
if (xp->i_copyonwrite == ip)
break;
if (xp == 0)
printf("ffs_snapgone: lost snapshot vnode %d\n",
ip->i_number);
else
vrele(ITOV(ip));
}
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
/*
* Prepare a snapshot file for being removed.
*/
void
ffs_snapremove(vp)
struct vnode *vp;
{
struct inode *ip, *xp;
struct vnode *devvp;
struct buf *ibp;
struct fs *fs;
ufs_daddr_t blkno, dblk;
int error, snaploc, loc, last;
ip = VTOI(vp);
fs = ip->i_fs;
/*
* Delete snapshot inode from superblock. Keep list dense.
*/
for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
if (fs->fs_snapinum[snaploc] == ip->i_number)
break;
if (snaploc < FSMAXSNAP) {
for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
if (fs->fs_snapinum[snaploc] == 0)
break;
fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
}
fs->fs_snapinum[snaploc - 1] = 0;
}
/*
* Delete from incore list.
* Clear copy-on-write flag if last snapshot.
*/
devvp = ip->i_devvp;
for (xp = VTOI(devvp); xp; xp = xp->i_copyonwrite) {
if (xp->i_copyonwrite != ip)
continue;
xp->i_copyonwrite = ip->i_copyonwrite;
ip->i_copyonwrite = 0;
break;
}
Clean up the snapshot code so that it no longer depends on the use of the SF_IMMUTABLE flag to prevent writing. Instead put in explicit checking for the SF_SNAPSHOT flag in the appropriate places. With this change, it is now possible to rename and link to snapshot files. It is also possible to set or clear any of the owner, group, or other read bits on the file, though none of the write or execute bits can be set. There is also an explicit test to prevent the setting or clearing of the SF_SNAPSHOT flag via chflags() or fchflags(). Note also that the modify time cannot be changed as it needs to accurately reflect the time that the snapshot was taken. Submitted by: Robert Watson <rwatson@FreeBSD.org>
2000-07-26 23:07:01 +00:00
if (xp == 0)
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
printf("ffs_snapremove: lost snapshot vnode %d\n",
ip->i_number);
if (VTOI(devvp)->i_copyonwrite == 0)
devvp->v_flag &= ~VCOPYONWRITE;
/*
* Clear all BLK_NOCOPY fields. Pass any block claims to other
* snapshots that want them (see ffs_snapblkfree below).
*/
for (blkno = 1; blkno < NDADDR; blkno++) {
dblk = ip->i_db[blkno];
if (dblk == BLK_NOCOPY || dblk == BLK_SNAP ||
(dblk == blkstofrags(fs, blkno) &&
ffs_snapblkfree(ip, dblk, fs->fs_bsize)))
ip->i_db[blkno] = 0;
}
for (blkno = NDADDR; blkno < fs->fs_size; blkno += NINDIR(fs)) {
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
if (error)
continue;
if ((last = fs->fs_size - blkno) > NINDIR(fs))
last = NINDIR(fs);
for (loc = 0; loc < last; loc++) {
dblk = ((ufs_daddr_t *)(ibp->b_data))[loc];
if (dblk == BLK_NOCOPY || dblk == BLK_SNAP ||
(dblk == blkstofrags(fs, blkno) &&
ffs_snapblkfree(ip, dblk, fs->fs_bsize)))
((ufs_daddr_t *)(ibp->b_data))[loc] = 0;
}
bawrite(ibp);
}
/*
* Clear snapshot flag and drop reference.
*/
Clean up the snapshot code so that it no longer depends on the use of the SF_IMMUTABLE flag to prevent writing. Instead put in explicit checking for the SF_SNAPSHOT flag in the appropriate places. With this change, it is now possible to rename and link to snapshot files. It is also possible to set or clear any of the owner, group, or other read bits on the file, though none of the write or execute bits can be set. There is also an explicit test to prevent the setting or clearing of the SF_SNAPSHOT flag via chflags() or fchflags(). Note also that the modify time cannot be changed as it needs to accurately reflect the time that the snapshot was taken. Submitted by: Robert Watson <rwatson@FreeBSD.org>
2000-07-26 23:07:01 +00:00
ip->i_flags &= ~SF_SNAPSHOT;
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Notification that a block is being freed. Return zero if the free
* should be allowed to proceed. Return non-zero if the snapshot file
* wants to claim the block. The block will be claimed if it is an
* uncopied part of one of the snapshots. It will be freed if it is
* either a BLK_NOCOPY or has already been copied in all of the snapshots.
* If a fragment is being freed, then all snapshots that care about
* it must make a copy since a snapshot file can only claim full sized
* blocks. Note that if more than one snapshot file maps the block,
* we can pick one at random to claim it. Since none of the snapshots
* can change, we are assurred that they will all see the same unmodified
* image. When deleting a snapshot file (see ffs_snapremove above), we
* must push any of these claimed blocks to one of the other snapshots
* that maps it. These claimed blocks are easily identified as they will
* have a block number equal to their logical block number within the
* snapshot. A copied block can never have this property because they
* must always have been allocated from a BLK_NOCOPY location.
*/
int
ffs_snapblkfree(freeip, bno, size)
struct inode *freeip;
ufs_daddr_t bno;
long size;
{
struct buf *ibp, *cbp, *savedcbp = 0;
struct fs *fs = freeip->i_fs;
struct proc *p = CURPROC;
struct inode *ip;
struct vnode *vp;
ufs_daddr_t lbn, blkno;
int indiroff = 0, error = 0, claimedblk = 0;
lbn = fragstoblks(fs, bno);
for (ip = VTOI(freeip->i_devvp)->i_copyonwrite; ip;
ip = ip->i_copyonwrite) {
vp = ITOV(ip);
/*
* Lookup block being written.
*/
if (lbn < NDADDR) {
blkno = ip->i_db[lbn];
} else {
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
p->p_flag |= P_COWINPROGRESS;
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)lbn),
fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
p->p_flag &= ~P_COWINPROGRESS;
VOP_UNLOCK(vp, 0, p);
if (error)
break;
indiroff = (lbn - NDADDR) % NINDIR(fs);
blkno = ((ufs_daddr_t *)(ibp->b_data))[indiroff];
}
/*
* Check to see if block needs to be copied.
*/
switch (blkno) {
/*
* If the snapshot has already copied the block (default),
* or does not care about the block, it is not needed.
*/
default:
case BLK_NOCOPY:
if (lbn >= NDADDR)
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(ibp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
continue;
/*
* No previous snapshot claimed the block, so it will be
* freed and become a BLK_NOCOPY (don't care) for us.
*/
case BLK_SNAP:
if (claimedblk)
panic("snapblkfree: inconsistent block type");
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
if (lbn < NDADDR) {
ip->i_db[lbn] = BLK_NOCOPY;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
} else {
((ufs_daddr_t *)(ibp->b_data))[indiroff] =
BLK_NOCOPY;
bdwrite(ibp);
}
VOP_UNLOCK(vp, 0, p);
continue;
/*
* A block that we map is being freed. If it has not been
* claimed yet, we will claim or copy it (below).
*/
case 0:
claimedblk = 1;
break;
}
/*
* If this is a full size block, we will just grab it
* and assign it to the snapshot inode. Otherwise we
* will proceed to copy it. See explanation for this
* routine as to why only a single snapshot needs to
* claim this block.
*/
if (size == fs->fs_bsize) {
#ifdef DEBUG
if (snapdebug)
printf("%s %d lbn %d from inum %d\n",
"Grabonremove: snapino", ip->i_number, lbn,
freeip->i_number);
#endif
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
if (lbn < NDADDR) {
ip->i_db[lbn] = bno;
} else {
((ufs_daddr_t *)(ibp->b_data))[indiroff] = bno;
bdwrite(ibp);
}
ip->i_blocks += btodb(size);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
VOP_UNLOCK(vp, 0, p);
return (1);
}
if (lbn >= NDADDR)
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(ibp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
/*
* Allocate the block into which to do the copy. Note that this
* allocation will never require any additional allocations for
* the snapshot inode.
*/
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
p->p_flag |= P_COWINPROGRESS;
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)lbn),
fs->fs_bsize, KERNCRED, 0, &cbp);
p->p_flag &= ~P_COWINPROGRESS;
VOP_UNLOCK(vp, 0, p);
if (error)
break;
#ifdef DEBUG
if (snapdebug)
printf("%s%d lbn %d for inum %d size %ld to blkno %d\n",
"Copyonremove: snapino ", ip->i_number, lbn,
freeip->i_number, size, cbp->b_blkno);
#endif
/*
* If we have already read the old block contents, then
* simply copy them to the new block.
*/
if (savedcbp != 0) {
bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
bawrite(cbp);
continue;
}
/*
* Otherwise, read the old block contents into the buffer.
*/
if ((error = readblock(cbp, lbn)) != 0)
break;
savedcbp = cbp;
}
if (savedcbp)
bawrite(savedcbp);
/*
* If we have been unable to allocate a block in which to do
* the copy, then return non-zero so that the fragment will
* not be freed. Although space will be lost, the snapshot
* will stay consistent.
*/
return (error);
}
/*
* Associate snapshot files when mounting.
*/
void
ffs_snapshot_mount(mp)
struct mount *mp;
{
struct ufsmount *ump = VFSTOUFS(mp);
struct fs *fs = ump->um_fs;
struct proc *p = CURPROC;
struct inode *ip, **listtailp;
struct vnode *vp;
int error, snaploc, loc;
listtailp = &VTOI(ump->um_devvp)->i_copyonwrite;
for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
if (fs->fs_snapinum[snaploc] == 0)
return;
if ((error = VFS_VGET(mp, fs->fs_snapinum[snaploc], &vp)) != 0){
printf("ffs_snapshot_mount: vget failed %d\n", error);
continue;
}
ip = VTOI(vp);
if ((ip->i_flags & SF_SNAPSHOT) == 0) {
printf("ffs_snapshot_mount: non-snapshot inode %d\n",
fs->fs_snapinum[snaploc]);
vput(vp);
for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
if (fs->fs_snapinum[loc] == 0)
break;
fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
}
fs->fs_snapinum[loc - 1] = 0;
snaploc--;
continue;
}
if (ip->i_copyonwrite != 0)
panic("ffs_snapshot_mount: %d already on list",
ip->i_number);
*listtailp = ip;
listtailp = &ip->i_copyonwrite;
vp->v_flag |= VSYSTEM;
VOP_UNLOCK(vp, 0, p);
ump->um_devvp->v_flag |= VCOPYONWRITE;
}
}
/*
* Disassociate snapshot files when unmounting.
*/
void
ffs_snapshot_unmount(mp)
struct mount *mp;
{
struct ufsmount *ump = VFSTOUFS(mp);
struct inode *devip = VTOI(ump->um_devvp);
struct inode *xp;
while ((xp = devip->i_copyonwrite) != 0) {
devip->i_copyonwrite = xp->i_copyonwrite;
xp->i_copyonwrite = 0;
Clean up the snapshot code so that it no longer depends on the use of the SF_IMMUTABLE flag to prevent writing. Instead put in explicit checking for the SF_SNAPSHOT flag in the appropriate places. With this change, it is now possible to rename and link to snapshot files. It is also possible to set or clear any of the owner, group, or other read bits on the file, though none of the write or execute bits can be set. There is also an explicit test to prevent the setting or clearing of the SF_SNAPSHOT flag via chflags() or fchflags(). Note also that the modify time cannot be changed as it needs to accurately reflect the time that the snapshot was taken. Submitted by: Robert Watson <rwatson@FreeBSD.org>
2000-07-26 23:07:01 +00:00
if (xp->i_effnlink > 0)
vrele(ITOV(xp));
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
}
ump->um_devvp->v_flag &= ~VCOPYONWRITE;
}
/*
* Check for need to copy block that is about to be written,
* copying the block if necessary.
*/
int
ffs_copyonwrite(ap)
struct vop_copyonwrite_args /* {
struct vnode *a_vp;
struct buf *a_bp;
} */ *ap;
{
struct buf *ibp, *cbp, *savedcbp = 0, *bp = ap->a_bp;
struct fs *fs = VTOI(bp->b_vp)->i_fs;
struct proc *p = CURPROC;
struct inode *ip;
struct vnode *vp;
ufs_daddr_t lbn, blkno;
int indiroff, error = 0;
lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
if (p->p_flag & P_COWINPROGRESS)
panic("ffs_copyonwrite: recursive call");
for (ip = VTOI(ap->a_vp)->i_copyonwrite; ip; ip = ip->i_copyonwrite) {
vp = ITOV(ip);
/*
* We ensure that everything of our own that needs to be
* copied will be done at the time that ffs_snapshot is
* called. Thus we can skip the check here which can
* deadlock in doing the lookup in VOP_BALLOC.
*/
if (bp->b_vp == vp)
continue;
/*
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
* Check to see if block needs to be copied. We have to
* be able to do the VOP_BALLOC without blocking, otherwise
* we may get in a deadlock with another process also
* trying to allocate. If we find outselves unable to
* get the buffer lock, we unlock the snapshot vnode,
* sleep briefly, and try again.
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
*/
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
retry:
vn_lock(vp, LK_SHARED | LK_RETRY, p);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
if (lbn < NDADDR) {
blkno = ip->i_db[lbn];
} else {
p->p_flag |= P_COWINPROGRESS;
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)lbn),
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
fs->fs_bsize, KERNCRED, B_METAONLY | B_NOWAIT, &ibp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
p->p_flag &= ~P_COWINPROGRESS;
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
if (error) {
VOP_UNLOCK(vp, 0, p);
if (error != EWOULDBLOCK)
break;
tsleep(vp, p->p_usrpri, "nap", 1);
goto retry;
}
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
indiroff = (lbn - NDADDR) % NINDIR(fs);
blkno = ((ufs_daddr_t *)(ibp->b_data))[indiroff];
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
bqrelse(ibp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
}
#ifdef DIAGNOSTIC
if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
panic("ffs_copyonwrite: bad copy block");
#endif
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
if (blkno != 0) {
VOP_UNLOCK(vp, 0, p);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
continue;
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
}
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
/*
* Allocate the block into which to do the copy. Note that this
* allocation will never require any additional allocations for
* the snapshot inode.
*/
p->p_flag |= P_COWINPROGRESS;
error = VOP_BALLOC(vp, lblktosize(fs, (off_t)lbn),
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
fs->fs_bsize, KERNCRED, B_NOWAIT, &cbp);
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
p->p_flag &= ~P_COWINPROGRESS;
VOP_UNLOCK(vp, 0, p);
This patch corrects the first round of panics and hangs reported with the new snapshot code. Update addaliasu to correctly implement the semantics of the old checkalias function. When a device vnode first comes into existence, check to see if an anonymous vnode for the same device was created at boot time by bdevvp(). If so, adopt the bdevvp vnode rather than creating a new vnode for the device. This corrects a problem which caused the kernel to panic when taking a snapshot of the root filesystem. Change the calling convention of vn_write_suspend_wait() to be the same as vn_start_write(). Split out softdep_flushworklist() from softdep_flushfiles() so that it can be used to clear the work queue when suspending filesystem operations. Access to buffers becomes recursive so that snapshots can recursively traverse their indirect blocks using ffs_copyonwrite() when checking for the need for copy on write when flushing one of their own indirect blocks. This eliminates a deadlock between the syncer daemon and a process taking a snapshot. Ensure that softdep_process_worklist() can never block because of a snapshot being taken. This eliminates a problem with buffer starvation. Cleanup change in ffs_sync() which did not synchronously wait when MNT_WAIT was specified. The result was an unclean filesystem panic when doing forcible unmount with heavy filesystem I/O in progress. Return a zero'ed block when reading a block that was not in use at the time that a snapshot was taken. Normally, these blocks should never be read. However, the readahead code will occationally read them which can cause unexpected behavior. Clean up the debugging code that ensures that no blocks be written on a filesystem while it is suspended. Snapshots must explicitly label the blocks that they are writing during the suspension so that they do not cause a `write on suspended filesystem' panic. Reorganize ffs_copyonwrite() to eliminate a deadlock and also to prevent a race condition that would permit the same block to be copied twice. This change eliminates an unexpected soft updates inconsistency in fsck caused by the double allocation. Use bqrelse rather than brelse for buffers that will be needed soon again by the snapshot code. This improves snapshot performance.
2000-07-24 05:28:33 +00:00
if (error) {
if (error != EWOULDBLOCK)
break;
tsleep(vp, p->p_usrpri, "nap", 1);
goto retry;
}
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
#ifdef DEBUG
if (snapdebug) {
printf("Copyonwrite: snapino %d lbn %d for ",
ip->i_number, lbn);
if (bp->b_vp == ap->a_vp)
printf("fs metadata");
else
printf("inum %d", VTOI(bp->b_vp)->i_number);
printf(" lblkno %d to blkno %d\n", bp->b_lblkno,
cbp->b_blkno);
}
#endif
/*
* If we have already read the old block contents, then
* simply copy them to the new block.
*/
if (savedcbp != 0) {
bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
bawrite(cbp);
continue;
}
/*
* Otherwise, read the old block contents into the buffer.
*/
if ((error = readblock(cbp, lbn)) != 0)
break;
savedcbp = cbp;
}
if (savedcbp)
bawrite(savedcbp);
return (error);
}
/*
* Read the specified block into the given buffer.
* Much of this boiler-plate comes from bwrite().
*/
static int
readblock(bp, lbn)
struct buf *bp;
daddr_t lbn;
{
struct uio auio;
struct iovec aiov;
struct proc *p = CURPROC;
struct inode *ip = VTOI(bp->b_vp);
aiov.iov_base = bp->b_data;
aiov.iov_len = bp->b_bcount;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn)));
auio.uio_resid = bp->b_bcount;
auio.uio_rw = UIO_READ;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_procp = p;
return (physio(ip->i_devvp->v_rdev, &auio, 0));
}
Reference in New Issue Copy Permalink