freebsd-skq/sys/ufs/ffs/ffs_snapshot.c

1113 lines
31 KiB
C
Raw Normal View History

/*
* 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
* $FreeBSD$
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/conf.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
2000-09-17 19:41:26 +00:00
#define DEBUG 1
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 ffs_copyonwrite __P((struct vnode *, struct buf *));
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, iblkno, inoblks[FSMAXSNAP];
int error, cg, snaploc, indiroff, numblks;
int i, size, base, len, loc, inoblkcnt;
int blksperindir, flag = mp->mnt_flag;
void *space;
struct fs *copy_fs, *fs = VFSTOUFS(mp)->um_fs;
struct snaphead *snaphead;
struct proc *p = CURPROC;
struct inode *ip, *xp;
struct buf *bp, *nbp, *ibp;
struct nameidata nd;
struct mount *wrtmp;
struct dinode *dip;
struct vattr vat;
struct vnode *vp;
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");
if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) {
NDFREE(&nd, NDF_ONLY_PNBUF);
vput(nd.ni_dvp);
if ((error = vn_start_write(NULL, &wrtmp,
V_XSLEEP | PCATCH)) != 0)
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);
/*
* 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 = UFS_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 and mark all of them as not
* needing to be copied.
*/
for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, p->p_ucred, B_METAONLY, &ibp);
if (error)
goto out;
iblkno = fragstoblks(fs, dbtofsb(fs, ibp->b_blkno));
bdwrite(ibp);
if (iblkno < NDADDR) {
if (ip->i_db[iblkno] != 0)
panic("ffs_snapshot: lost direct block");
ip->i_db[iblkno] = BLK_NOCOPY;
} else {
error = UFS_BALLOC(vp, lblktosize(fs, (off_t)iblkno),
fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
if (error)
goto out;
indiroff = (iblkno - NDADDR) % NINDIR(fs);
if (((ufs_daddr_t *)(ibp->b_data))[indiroff] != 0)
panic("ffs_snapshot: lost indirect block");
((ufs_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
bdwrite(ibp);
}
}
/*
* Allocate shadow blocks to copy all of the other snapshot inodes
* so that we will be able to expunge them from this snapshot. Also
* include a copy of ourselves so that we do not deadlock trying
* to copyonwrite ourselves when VOP_FSYNC'ing below.
*/
fs->fs_snapinum[snaploc] = ip->i_number;
for (loc = snaploc, inoblkcnt = 0; loc >= 0; loc--) {
blkno = fragstoblks(fs, ino_to_fsba(fs, fs->fs_snapinum[loc]));
fs->fs_snapinum[snaploc] = 0;
for (i = 0; i < inoblkcnt; i++)
if (inoblks[i] == blkno)
break;
if (i == inoblkcnt) {
inoblks[inoblkcnt++] = blkno;
error = UFS_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 = UFS_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.
*/
if ((error = UFS_BALLOC(vp, (off_t)(SBOFF), SBSIZE, KERNCRED, 0, &nbp)))
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 = UFS_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.
*/
ip->i_flags |= SF_SNAPSHOT;
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;
vn_start_write(NULL, &wrtmp, V_WAIT);
}
/*
* 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;
bawrite(nbp);
base = cg * fs->fs_fpg / fs->fs_frag;
if (base + len >= numblks)
len = numblks - base - 1;
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 = UFS_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;
bawrite(ibp);
error = UFS_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;
if (((ufs_daddr_t *)(ibp->b_data))[indiroff] != 0)
panic("ffs_snapshot: lost block");
((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;
bdwrite(ibp);
}
/*
* Snapshot the superblock and its summary information.
*/
if ((error = UFS_BALLOC(vp, SBOFF, SBSIZE, KERNCRED, 0, &nbp)) != 0)
goto out1;
copy_fs = (struct fs *)(nbp->b_data + blkoff(fs, SBOFF));
bcopy(fs, copy_fs, fs->fs_sbsize);
if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
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;
bawrite(nbp);
blkno = fragstoblks(fs, fs->fs_csaddr);
len = howmany(fs->fs_cssize, fs->fs_bsize) - 1;
size = fs->fs_bsize;
space = fs->fs_csp;
for (loc = 0; loc <= len; loc++) {
error = UFS_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);
size = fs->fs_cssize - loc * fs->fs_bsize;
}
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;
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 = UFS_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;
bdwrite(nbp);
}
/*
* Copy allocation information from other snapshots and then
* expunge them from the view of the current snapshot.
*/
snaphead = &ip->i_devvp->v_rdev->si_snapshots;
TAILQ_FOREACH(xp, snaphead, i_nextsnap) {
/*
* 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 = UFS_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;
dip->di_flags &= ~SF_SNAPSHOT;
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;
bdwrite(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_nextsnap.tqe_prev != 0)
panic("ffs_snapshot: %d already on list", ip->i_number);
TAILQ_INSERT_TAIL(snaphead, ip, i_nextsnap);
ip->i_devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
ip->i_devvp->v_flag |= VCOPYONWRITE;
vp->v_flag |= VSYSTEM;
/*
* Resume operation on filesystem.
*/
out1:
vfs_write_resume(vp->v_mount);
vn_start_write(NULL, &wrtmp, V_WAIT);
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);
}
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);
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 = UFS_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;
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
}
}
return (0);
}
/*
* 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;
struct fs *fs;
int snaploc;
/*
* Find snapshot in incore list.
*/
TAILQ_FOREACH(xp, &ip->i_devvp->v_rdev->si_snapshots, i_nextsnap)
if (xp == ip)
break;
if (xp == 0)
printf("ffs_snapgone: lost snapshot vnode %d\n",
ip->i_number);
else
vrele(ITOV(ip));
/*
* Delete snapshot inode from superblock. Keep list dense.
*/
fs = ip->i_fs;
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;
}
}
/*
* Prepare a snapshot file for being removed.
*/
void
ffs_snapremove(vp)
struct vnode *vp;
{
struct inode *ip;
struct vnode *devvp;
struct buf *ibp;
struct fs *fs;
ufs_daddr_t blkno, dblk;
int error, loc, last;
ip = VTOI(vp);
fs = ip->i_fs;
/*
* If active, delete from incore list (this snapshot may
* already have been in the process of being deleted, so
* would not have been active).
*
* Clear copy-on-write flag if last snapshot.
*/
if (ip->i_nextsnap.tqe_prev != 0) {
devvp = ip->i_devvp;
TAILQ_REMOVE(&devvp->v_rdev->si_snapshots, ip, i_nextsnap);
ip->i_nextsnap.tqe_prev = 0;
if (TAILQ_FIRST(&devvp->v_rdev->si_snapshots) == 0) {
devvp->v_rdev->si_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 = UFS_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.
*/
ip->i_flags &= ~SF_SNAPSHOT;
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;
struct snaphead *snaphead;
lbn = fragstoblks(fs, bno);
snaphead = &freeip->i_devvp->v_rdev->si_snapshots;
TAILQ_FOREACH(ip, snaphead, i_nextsnap) {
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 = UFS_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);
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);
/*
* 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 = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
fs->fs_bsize, KERNCRED, 0, &cbp);
p->p_flag &= ~P_COWINPROGRESS;
if (error) {
VOP_UNLOCK(vp, 0, p);
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. Note that we need
* to synchronously write snapshots that have not been
* unlinked, and hence will be visible after a crash,
* to ensure their integrity.
*/
if (savedcbp != 0) {
bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
bawrite(cbp);
if (ip->i_effnlink > 0)
(void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p);
VOP_UNLOCK(vp, 0, p);
continue;
}
/*
* Otherwise, read the old block contents into the buffer.
*/
if ((error = readblock(cbp, lbn)) != 0) {
bzero(cbp->b_data, fs->fs_bsize);
bawrite(cbp);
if (ip->i_effnlink > 0)
(void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p);
VOP_UNLOCK(vp, 0, p);
break;
}
savedcbp = cbp;
}
/*
* Note that we need to synchronously write snapshots that
* have not been unlinked, and hence will be visible after
* a crash, to ensure their integrity.
*/
if (savedcbp) {
vp = savedcbp->b_vp;
bawrite(savedcbp);
if (VTOI(vp)->i_effnlink > 0)
(void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p);
VOP_UNLOCK(vp, 0, p);
}
/*
* 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 snaphead *snaphead;
struct vnode *vp;
struct inode *ip;
int error, snaploc, loc;
snaphead = &ump->um_devvp->v_rdev->si_snapshots;
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_nextsnap.tqe_prev != 0)
panic("ffs_snapshot_mount: %d already on list",
ip->i_number);
else
TAILQ_INSERT_TAIL(snaphead, ip, i_nextsnap);
vp->v_flag |= VSYSTEM;
ump->um_devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
ump->um_devvp->v_flag |= VCOPYONWRITE;
VOP_UNLOCK(vp, 0, p);
}
}
/*
* Disassociate snapshot files when unmounting.
*/
void
ffs_snapshot_unmount(mp)
struct mount *mp;
{
struct ufsmount *ump = VFSTOUFS(mp);
struct snaphead *snaphead = &ump->um_devvp->v_rdev->si_snapshots;
struct inode *xp;
while ((xp = TAILQ_FIRST(snaphead)) != 0) {
TAILQ_REMOVE(snaphead, xp, i_nextsnap);
xp->i_nextsnap.tqe_prev = 0;
if (xp->i_effnlink > 0)
vrele(ITOV(xp));
}
ump->um_devvp->v_rdev->si_copyonwrite = 0;
ump->um_devvp->v_flag &= ~VCOPYONWRITE;
}
/*
* Check for need to copy block that is about to be written,
* copying the block if necessary.
*/
static int
ffs_copyonwrite(devvp, bp)
struct vnode *devvp;
struct buf *bp;
{
struct buf *ibp, *cbp, *savedcbp = 0;
struct proc *p = CURPROC;
struct fs *fs;
struct inode *ip;
struct vnode *vp;
ufs_daddr_t lbn, blkno;
int indiroff, error = 0;
fs = TAILQ_FIRST(&devvp->v_rdev->si_snapshots)->i_fs;
lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
if (p->p_flag & P_COWINPROGRESS)
panic("ffs_copyonwrite: recursive call");
TAILQ_FOREACH(ip, &devvp->v_rdev->si_snapshots, i_nextsnap) {
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 UFS_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 UFS_BALLOC without blocking, otherwise
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
* 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.
*/
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);
if (lbn < NDADDR) {
blkno = ip->i_db[lbn];
} else {
p->p_flag |= P_COWINPROGRESS;
error = UFS_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);
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;
Implement a unified run queue and adjust priority levels accordingly. - All processes go into the same array of queues, with different scheduling classes using different portions of the array. This allows user processes to have their priorities propogated up into interrupt thread range if need be. - I chose 64 run queues as an arbitrary number that is greater than 32. We used to have 4 separate arrays of 32 queues each, so this may not be optimal. The new run queue code was written with this in mind; changing the number of run queues only requires changing constants in runq.h and adjusting the priority levels. - The new run queue code takes the run queue as a parameter. This is intended to be used to create per-cpu run queues. Implement wrappers for compatibility with the old interface which pass in the global run queue structure. - Group the priority level, user priority, native priority (before propogation) and the scheduling class into a struct priority. - Change any hard coded priority levels that I found to use symbolic constants (TTIPRI and TTOPRI). - Remove the curpriority global variable and use that of curproc. This was used to detect when a process' priority had lowered and it should yield. We now effectively yield on every interrupt. - Activate propogate_priority(). It should now have the desired effect without needing to also propogate the scheduling class. - Temporarily comment out the call to vm_page_zero_idle() in the idle loop. It interfered with propogate_priority() because the idle process needed to do a non-blocking acquire of Giant and then other processes would try to propogate their priority onto it. The idle process should not do anything except idle. vm_page_zero_idle() will return in the form of an idle priority kernel thread which is woken up at apprioriate times by the vm system. - Update struct kinfo_proc to the new priority interface. Deliberately change its size by adjusting the spare fields. It remained the same size, but the layout has changed, so userland processes that use it would parse the data incorrectly. The size constraint should really be changed to an arbitrary version number. Also add a debug.sizeof sysctl node for struct kinfo_proc.
2001-02-12 00:20:08 +00:00
tsleep(vp, p->p_pri.pri_user, "nap", 1);
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
goto retry;
}
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);
}
#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);
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
}
/*
* 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 = UFS_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);
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);
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 != EWOULDBLOCK)
break;
Implement a unified run queue and adjust priority levels accordingly. - All processes go into the same array of queues, with different scheduling classes using different portions of the array. This allows user processes to have their priorities propogated up into interrupt thread range if need be. - I chose 64 run queues as an arbitrary number that is greater than 32. We used to have 4 separate arrays of 32 queues each, so this may not be optimal. The new run queue code was written with this in mind; changing the number of run queues only requires changing constants in runq.h and adjusting the priority levels. - The new run queue code takes the run queue as a parameter. This is intended to be used to create per-cpu run queues. Implement wrappers for compatibility with the old interface which pass in the global run queue structure. - Group the priority level, user priority, native priority (before propogation) and the scheduling class into a struct priority. - Change any hard coded priority levels that I found to use symbolic constants (TTIPRI and TTOPRI). - Remove the curpriority global variable and use that of curproc. This was used to detect when a process' priority had lowered and it should yield. We now effectively yield on every interrupt. - Activate propogate_priority(). It should now have the desired effect without needing to also propogate the scheduling class. - Temporarily comment out the call to vm_page_zero_idle() in the idle loop. It interfered with propogate_priority() because the idle process needed to do a non-blocking acquire of Giant and then other processes would try to propogate their priority onto it. The idle process should not do anything except idle. vm_page_zero_idle() will return in the form of an idle priority kernel thread which is woken up at apprioriate times by the vm system. - Update struct kinfo_proc to the new priority interface. Deliberately change its size by adjusting the spare fields. It remained the same size, but the layout has changed, so userland processes that use it would parse the data incorrectly. The size constraint should really be changed to an arbitrary version number. Also add a debug.sizeof sysctl node for struct kinfo_proc.
2001-02-12 00:20:08 +00:00
tsleep(vp, p->p_pri.pri_user, "nap", 1);
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
goto retry;
}
#ifdef DEBUG
if (snapdebug) {
printf("Copyonwrite: snapino %d lbn %d for ",
ip->i_number, lbn);
if (bp->b_vp == devvp)
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. Note that we need
* to synchronously write snapshots that have not been
* unlinked, and hence will be visible after a crash,
* to ensure their integrity.
*/
if (savedcbp != 0) {
bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
bawrite(cbp);
if (ip->i_effnlink > 0)
(void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p);
VOP_UNLOCK(vp, 0, p);
continue;
}
/*
* Otherwise, read the old block contents into the buffer.
*/
if ((error = readblock(cbp, lbn)) != 0) {
bzero(cbp->b_data, fs->fs_bsize);
bawrite(cbp);
if (ip->i_effnlink > 0)
(void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p);
VOP_UNLOCK(vp, 0, p);
break;
}
savedcbp = cbp;
}
/*
* Note that we need to synchronously write snapshots that
* have not been unlinked, and hence will be visible after
* a crash, to ensure their integrity.
*/
if (savedcbp) {
vp = savedcbp->b_vp;
bawrite(savedcbp);
if (VTOI(vp)->i_effnlink > 0)
(void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, p);
VOP_UNLOCK(vp, 0, p);
}
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));
}