PSARC 2006/370 ZFS Boot Support
5008936 ZFS and/or zvol should support dumps
5070124 dumpadm -d /dev/... does not enforce block device requirement for savecore
6521468 ZFS Boot support Phase 2
6553503 bfu can't find 'rootdev' from /etc/vfstab on a zfs root filesystem
6574993 zfs_mountroot() may need to call clkset() to set the boot_time kstat
6633197 zvol should not permit newfs or createpool while it's in use by swap or dump
6661127 zfs_name_valid() does not support ZFS_TYPE_POOL
6684121 The changes to smf scripts for supporting canmount=noauto will cause a boot failure.
\ ident "%Z%%M% %I% %E% SMI"
\ Copyright 2008 Sun Microsystems, Inc. All rights reserved.
\ Use is subject to license terms.
\
\ CDDL HEADER START
\
\ The contents of this file are subject to the terms of the
\ Common Development and Distribution License (the "License").
\ You may not use this file except in compliance with the License.
\
\ You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
\ or http://www.opensolaris.org/os/licensing.
\ See the License for the specific language governing permissions
\ and limitations under the License.
\
\ When distributing Covered Code, include this CDDL HEADER in each
\ file and include the License file at usr/src/OPENSOLARIS.LICENSE.
\ If applicable, add the following below this CDDL HEADER, with the
\ fields enclosed by brackets "[]" replaced with your own identifying
\ information: Portions Copyright [yyyy] [name of copyright owner]
\
\ CDDL HEADER END
\
\
[ifdef] doheaders
headers
[else]
headerless
[then]
id: %Z%%M% %I% %E% SMI
purpose: ZFS file system support package
copyright: Copyright 2006 Sun Microsystems, Inc. All Rights Reserved
" /packages" get-package push-package
new-device
fs-pkg$ device-name diag-cr?
0 instance value temp-space
\ 64b ops
\ fcode is still 32b on 64b sparc-v9, so
\ we need to override some arithmetic ops
\ stack ops and logical ops (dup, and, etc) are 64b
: xcmp ( x1 x2 -- -1|0|1 )
xlsplit rot xlsplit ( x2.lo x2.hi x1.lo x1.hi )
rot 2dup u< if ( x2.lo x1.lo x1.hi x2.hi )
2drop 2drop -1 ( lt )
else u> if ( x2.lo x1.lo )
2drop 1 ( gt )
else swap 2dup u< if ( x1.lo x2.lo )
2drop -1 ( lt )
else u> if ( )
1 ( gt )
else ( )
0 ( eq )
then then then then ( -1|0|1 )
;
: x< ( x1 x2 -- <? ) xcmp -1 = ;
: x> ( x1 x2 -- >? ) xcmp 1 = ;
\ : x= ( x1 x2 -- =? ) xcmp 0= ;
: x<> ( x1 x2 -- <>? ) xcmp 0<> ;
: x0= ( x -- 0=? ) xlsplit 0= swap 0= and ;
/buf-len instance buffer: numbuf
: (xu.) ( u -- u$ )
numbuf /buf-len + swap ( adr u )
begin
d# 10 /mod swap ( adr u' rem )
ascii 0 + ( adr u' c )
rot 1- tuck c! ( u adr' )
swap dup 0= ( adr u done? )
until drop ( adr )
dup numbuf - /buf-len swap - ( adr len )
;
\ pool name
/buf-len instance buffer: bootprop-buf
: bootprop$ ( -- prop$ ) bootprop-buf cscount ;
\ decompression
\
\ uts/common/os/compress.c has a definitive theory of operation comment
\ on lzjb, but here's the reader's digest version:
\
\ repeated phrases are replaced by referenced to the original
\ e.g.,
\ y a d d a _ y a d d a _ y a d d a , _ b l a h _ b l a h _ b l a h
\ becomes
\ y a d d a _ 6 11 , _ b l a h 5 10
\ where 6 11 means memmove(ptr, ptr - 6, 11)
\
\ data is separated from metadata with embedded copymap entries
\ every 8 items e.g.,
\ 0x40 y a d d a _ 6 11 , 0x20 _ b l a h 5 10
\ the copymap has a set bit for copy refercences
\ and a clear bit for bytes to be copied directly
\
\ the reference marks are encoded with match-bits and match-min
\ e.g.,
\ byte[0] = ((mlen - MATCH_MIN) << (NBBY - MATCH_BITS) | (off >> NBBY)
\ byte[1] = (uint8_t)off
\
: pow2 ( n -- 2**n ) 1 swap lshift ;
\ assume MATCH_BITS=6 and MATCH_MIN=3
6 constant mbits
3 constant mmin
8 mbits - constant mshift
d# 16 mbits - pow2 1- constant mmask
: decode-src ( src -- mlen off )
dup c@ swap 1+ c@ ( c[0] c[1] )
over mshift rshift mmin + ( c[0] c[1] mlen )
-rot swap bwjoin mmask and ( mlen off )
;
\ equivalent of memmove(dst, dst - off, len)
\ src points to a copy reference to be decoded
: mcopy ( dend dst src -- dend dst' )
decode-src ( dend dst mlen off )
2 pick swap - >r ( dent dst mlen r: cpy )
begin
1- dup 0>= ( dend dst mlen' any? r: cpy )
2over > and ( dend dst mlen !done? r : cpy )
while ( dend dst mlen r: cpy )
swap r> dup 1+ >r c@ ( dend mlen dst c r: cpy' )
over c! 1+ swap ( dend dst' mlen r: cpy )
repeat ( dend dst' mlen r: cpy )
r> 2drop ( dend dst )
;
: lzjb ( src dst len -- )
over + swap ( src dend dst )
rot >r ( dend dst r: src )
\ setup mask so 1st while iteration fills map
0 7 pow2 2swap ( map mask dend dst r: src )
begin 2dup > while
2swap 1 lshift ( dend dst map mask' r: src )
dup 8 pow2 = if
\ fetch next copymap
2drop ( dend dst r: src )
r> dup 1+ >r c@ 1 ( dend dst map' mask' r: src' )
then ( dend dst map mask r: src' )
\ if (map & mask) we hit a copy reference
\ else just copy 1 byte
2swap 2over and if ( map mask dend dst r: src )
r> dup 2+ >r ( map mask dend dst src r: src' )
mcopy ( map mask dend dst' r: src )
else
r> dup 1+ >r c@ ( map mask dend dst c r: src' )
over c! 1+ ( map mask dend dst' r: src )
then
repeat ( map mask dend dst r: src )
2drop 2drop r> drop ( )
;
\
\ ZFS block (SPA) routines
\
2 constant no-comp#
h# 2.0000 constant /max-bsize
d# 512 constant /disk-block
d# 128 constant /blkp
: blk_offset ( bp -- n ) h# 8 + x@ -1 h# 8fff.ffff lxjoin and ;
: blk_gang ( bp -- n ) h# 8 + x@ xlsplit nip d# 31 rshift ;
: blk_comp ( bp -- n ) h# 33 + c@ ;
: blk_psize ( bp -- n ) h# 34 + w@ ;
: blk_lsize ( bp -- n ) h# 36 + w@ ;
: blk_birth ( bp -- n ) h# 50 + x@ ;
0 instance value dev-ih
0 instance value blk-space
0 instance value gang-space
: foff>doff ( fs-off -- disk-off ) /disk-block * h# 40.0000 + ;
: fsz>dsz ( fs-size -- disk-size ) 1+ /disk-block * ;
: bp-dsize ( bp -- dsize ) blk_psize fsz>dsz ;
: bp-lsize ( bp -- lsize ) blk_lsize fsz>dsz ;
: (read-bp) ( adr len bp -- )
blk_offset foff>doff dev-ih read-disk
;
: gang-read ( adr len bp -- )
\ read gang block
gang-space /disk-block rot ( adr len gb-adr gb-len bp )
(read-bp) ( adr len )
\ read gang indirected blocks to blk-space
\ and copy requested len from there
blk-space gang-space ( adr len tmp-adr bp0 )
dup /blkp 3 * + bounds do ( adr len tmp-adr )
i blk_offset x0= ?leave
i bp-dsize ( adr len tmp-adr rd-len )
2dup i (read-bp)
+ ( adr len tmp-adr' )
/blkp +loop
drop ( adr len )
blk-space -rot move ( )
;
\ block read that check for holes, gangs, compression, etc
: read-bp ( adr len bp -- )
\ sparse block?
dup blk_birth x0= if
drop erase exit ( )
then
\ gang block?
dup blk_gang if
gang-read exit ( )
then
\ compression?
dup blk_comp no-comp# <> if
blk-space over bp-dsize ( adr len bp b-adr rd-len )
rot (read-bp) ( adr len )
blk-space -rot lzjb exit ( )
then
\ boring direct block
(read-bp) ( )
;
\
\ ZFS vdev routines
\
h# 1.c000 constant /nvpairs
h# 4000 constant nvpairs-off
\
\ xdr packed nvlist
\
\ 12B header
\ array of xdr packed nvpairs
\ 4B encoded nvpair size
\ 4B decoded nvpair size
\ 4B name string size
\ name string
\ 4B data type
\ 4B # of data elements
\ data
\ 8B of 0
\
d# 12 constant /nvhead
: >nvsize ( nv -- size ) l@ ;
: >nvname ( nv -- name$ )
/l 2* + dup /l + swap l@
;
: >nvdata ( nv -- data )
>nvname + /l roundup
;
\ convert nvdata to 64b int or string
: nvdata>x ( nvdata -- x )
/l 2* + ( ptr )
dup /l + l@ swap l@ ( x.lo x.hi )
lxjoin ( x )
;
alias nvdata>$ >nvname
: nv-lookup ( nv name$ -- nvdata false | true )
rot /nvhead + ( name$ nvpair )
begin dup >nvsize while
dup >r >nvname ( name$ nvname$ r: nvpair )
2over $= if ( name$ r: nvpair )
2drop r> >nvdata ( nvdata )
false exit ( nvdata found )
then ( name$ r: nvpair )
r> dup >nvsize + ( name$ nvpair' )
repeat
3drop true ( not-found )
;
: scan-vdev ( -- )
temp-space /nvpairs nvpairs-off ( adr len off )
dev-ih read-disk ( )
temp-space " txg" nv-lookup if
" no txg nvpair" die
then nvdata>x ( txg )
x0= if
" detached mirror" die
then ( )
temp-space " name" nv-lookup if
" no name nvpair" die
then nvdata>$ ( pool$ )
bootprop-buf swap move ( )
;
\
\ ZFS ueber-block routines
\
d# 1024 constant /uber-block
d# 128 constant #ub/label
#ub/label /uber-block * constant /ub-ring
h# 2.0000 constant ubring-off
: ub_magic ( ub -- n ) x@ ;
: ub_txg ( ub -- n ) h# 10 + x@ ;
: ub_timestamp ( ub -- n ) h# 20 + x@ ;
: ub_rootbp ( ub -- p ) h# 28 + ;
0 instance value uber-block
: ub-cmp ( ub1 ub2 -- best-ub )
\ ub1 wins if ub2 isn't valid
dup ub_magic h# 00bab10c x<> if
drop exit ( ub1 )
then
\ if ub1 is 0, ub2 wins by default
over 0= if nip exit then ( ub2 )
\ 2 valid ubs, compare transaction groups
over ub_txg over ub_txg ( ub1 ub2 txg1 txg2 )
2dup x< if
2drop nip exit ( ub2 )
then ( ub1 ub2 txg1 txg2 )
x> if drop exit then ( ub1 )
\ same txg, check timestamps
over ub_timestamp over ub_timestamp x> if
nip ( ub2 )
else
drop ( ub1 )
then
;
\ find best uber-block in ring, and copy it to uber-block
: get-ub ( -- )
temp-space /ub-ring ubring-off ( adr len off )
dev-ih read-disk ( )
0 temp-space /ub-ring ( null-ub adr len )
bounds do ( ub )
i ub-cmp ( best-ub )
/uber-block +loop
\ make sure we found a valid ub
dup 0= if " no ub found" die then
uber-block /uber-block move ( )
;
\
\ ZFS dnode (DMU) routines
\
d# 512 constant /dnode
: dn_indblkshift ( dn -- n ) h# 1 + c@ ;
: dn_nlevels ( dn -- n ) h# 2 + c@ ;
: dn_datablkszsec ( dn -- n ) h# 8 + w@ ;
: dn_blkptr ( dn -- p ) h# 40 + ;
: dn_bonus ( dn -- p ) h# c0 + ;
0 instance value dnode
\ indirect cache
\
\ ind-cache is a 1 block indirect block cache from dnode ic-dn
\
\ ic-bp and ic-bplim point into the ic-dn's block ptr array,
\ either in dn_blkptr or in ind-cache ic-bp is the ic-blk#'th
\ block ptr, and ic-bplim is limit of the current bp array
\
\ the assumption is that reads will be sequential, so we can
\ just increment ic-bp
\
0 instance value ind-cache
0 instance value ic-dn
0 instance value ic-blk#
0 instance value ic-bp
0 instance value ic-bplim
: dn-bsize ( dn -- bsize ) dn_datablkszsec /disk-block * ;
: dn-indsize ( dn -- indsize ) dn_indblkshift pow2 ;
: dn-indmask ( dn -- mask ) dn-indsize 1- ;
\ recursively climb the block tree from the leaf to the root
: blk@lvl>bp ( dn blk# lvl -- bp ) tokenizer[ reveal ]tokenizer
>r /blkp * over dn_nlevels ( dn bp-off #lvls r: lvl )
\ at top, just add dn_blkptr
r@ = if ( dn bp-off r: lvl )
swap dn_blkptr + ( bp r: lvl )
r> drop exit ( bp )
then ( dn bp-off r: lvl )
\ shift bp-off down and find parent indir blk
2dup over dn_indblkshift rshift ( dn bp-off dn blk# r: lvl )
r> 1+ blk@lvl>bp ( dn bp-off bp )
\ read parent indir blk and index
rot tuck dn-indsize ( bp-off dn bp len )
ind-cache swap rot read-bp ( bp-off dn )
dn-indmask and ( bp-off' )
ind-cache + ( bp )
;
\ return end of current bp array
: bplim ( dn bp -- bp-lim )
over dn_nlevels 1 = if
drop dn_blkptr ( bp0 )
3 /blkp * + ( bplim )
else
1+ swap dn-indsize ( bp+1 indsz )
roundup ( bplim )
then
;
\ return the lblk#'th block ptr from dnode
: lblk#>bp ( dn blk# -- bp )
2dup ( dn blk# dn blk# )
ic-blk# <> swap ic-dn <> or ( dn blk# cache-miss? )
ic-bp ic-bplim = ( dn blk# cache-miss? cache-empty? )
or if ( dn blk# )
2dup 1 blk@lvl>bp ( dn blk# bp )
dup to ic-bp ( dn blk# bp )
swap to ic-blk# ( dn bp )
2dup bplim to ic-bplim ( dn bp )
over to ic-dn
then 2drop ( )
ic-blk# 1+ to ic-blk#
ic-bp dup /blkp + to ic-bp ( bp )
;
\
\ ZFS attribute (ZAP) routines
\
1 constant fzap#
3 constant uzap#
d# 64 constant /uzap
d# 24 constant /lf-chunk
d# 21 constant /lf-arr
h# ffff constant chain-end#
h# 100 constant /lf-buf
/lf-buf instance buffer: leaf-value
/lf-buf instance buffer: leaf-name
: +le ( len off -- n ) + w@ ;
: le_next ( le -- n ) h# 2 +le ;
: le_name_chunk ( le -- n ) h# 4 +le ;
: le_name_length ( le -- n ) h# 6 +le ;
: le_value_chunk ( le -- n ) h# 8 +le ;
: le_value_length ( le -- n ) h# a +le ;
: la_array ( la -- adr ) 1+ ;
: la_next ( la -- n ) h# 16 + w@ ;
0 instance value zap-space
\ setup leaf hash bounds
: >leaf-hash ( dn lh -- hash-adr /hash )
/lf-chunk 2* + ( dn hash-adr )
\ size = (bsize / 32) * 2
swap dn-bsize 4 rshift ( hash-adr /hash )
;
: >leaf-chunks ( lf -- ch0 ) >leaf-hash + ;
\ convert chunk # to leaf chunk
: ch#>lc ( dn ch# -- lc )
/lf-chunk * ( dn lc-off )
swap zap-space >leaf-chunks ( lc-off ch0 )
+ ( lc )
;
\ assemble chunk chain into single buffer
: get-chunk-data ( dn ch# adr -- )
dup >r /lf-buf erase ( dn ch# r: adr )
begin
2dup ch#>lc nip ( dn la r: adr )
dup la_array ( dn la la-arr r: adr )
r@ /lf-arr move ( dn la r: adr )
r> /lf-arr + >r ( dn la r: adr' )
la_next dup chain-end# = ( dn la-ch# end? r: adr )
until r> 3drop ( )
;
\ get leaf entry's name
: entry-name$ ( dn le -- name$ )
2dup le_name_chunk ( dn le dn la-ch# )
leaf-name get-chunk-data ( dn le )
nip le_name_length ( len )
leaf-name swap ( name$ )
;
\ return entry value as int
: entry-int-val ( dn le -- n )
le_value_chunk ( dn la-ch# )
leaf-value get-chunk-data ( )
leaf-value x@ ( n )
;
[ifdef] strlookup
\ get leaf entry's value as string
: entry-val$ ( dn le -- val$ )
2dup le_value_chunk ( dn le dn la-ch# )
leaf-value get-chunk-data ( dn le )
nip le_value_length ( len )
leaf-value swap ( name$ )
;
[then]
\ apply xt to entry
: entry-apply ( xt dn le -- xt dn false | ??? true )
over >r ( xt dn le r: dn )
rot dup >r execute if ( ??? r: xt dn )
r> r> 2drop true ( ??? true )
else ( )
r> r> false ( xt dn false )
then
;
\ apply xt to every entry in chain
: chain-apply ( xt dn ch# -- xt dn false | ??? true )
begin
2dup ch#>lc nip ( xt dn le )
dup >r entry-apply if ( ??? r: le )
r> drop true exit ( ??? found )
then ( xt dn r: le )
r> le_next ( xt dn ch# )
dup chain-end# = ( xt dn ch# end? )
until drop ( xt dn )
false ( xt dn false )
;
\ apply xt to every entry in leaf
: leaf-apply ( xt dn blk# -- xt dn false | ??? true )
\ read zap leaf into zap-space
2dup lblk#>bp ( xt dn blk# bp )
nip over dn-bsize zap-space ( xt dn bp len adr )
swap rot read-bp ( xt dn )
\ call chunk-look for every valid chunk list
dup zap-space >leaf-hash ( xt dn hash-adr /hash )
bounds do ( xt dn )
i w@ dup chain-end# <> if ( xt dn ch# )
chain-apply if ( ??? )
unloop true exit ( ??? found )
then ( xt dn )
else drop then ( xt dn )
/w +loop
false ( xt dn not-found )
;
\ apply xt to every entry in fzap
: fzap-apply ( xt dn fz -- ??? not-found? )
\ blk# 1 is always the 1st leaf
>r 1 leaf-apply if ( ??? r: fz )
r> drop false exit ( ??? found )
then r> ( xt dn fz )
\ call leaf-apply on every non-duplicate hash entry
\ embedded hash is in 2nd half of fzap block
over dn-bsize tuck + ( xt dn bsize hash-eadr )
swap 2dup 2/ - ( xt dn hash-eadr bsize hash-adr )
nip do ( xt dn )
i x@ dup 1 <> if ( xt dn blk# )
leaf-apply if ( ??? )
unloop true exit ( ??? found )
then ( xt dn )
else drop then ( xt dn )
/x +loop
2drop false ( not-found )
;
: mze_value ( uz -- n ) x@ ;
: mze_name ( uz -- p ) h# e + ;
: uzap-name$ ( uz -- name$ ) mze_name cscount ;
\ apply xt to each entry in micro-zap
: uzap-apply ( xt uz len -- ??? not-found? )
bounds do ( xt )
i swap dup >r ( uz xt r: xt )
execute if ( ??? r: xt )
r> drop ( ??? )
unloop true exit ( ??? found )
then r> ( xt )
/uzap +loop
drop false ( not-found )
;
\ match by name
: fz-nmlook ( prop$ dn le -- prop$ false | prop$ dn le true )
2dup entry-name$ ( prop$ dn le name$ )
2rot 2swap ( dn le prop$ name$ )
2over $= if ( dn le prop$ )
2swap true ( prop$ dn le true )
else ( dn le prop$ )
2swap 2drop false ( prop$ false )
then ( prop$ false | prop$ dn le true )
;
\ match by name
: uz-nmlook ( prop$ uz -- prop$ false | prop$ uz true )
dup >r uzap-name$ ( prop$ name$ r: uz )
2over $= if ( prop$ r: uz )
r> true ( prop$ uz true )
else ( prop$ r: uz )
r> drop false ( prop$ false )
then ( prop$ false | prop$ uz true )
;
: zap-type ( zp -- n ) h# 7 + c@ ;
: >uzap-ent ( adr -- ent ) h# 40 + ;
\ read zap block into temp-space
: get-zap ( dn -- zp )
dup 0 lblk#>bp ( dn bp )
swap dn-bsize ( bp len )
temp-space swap ( bp adr len )
rot read-bp ( )
temp-space ( zp )
;
\ find prop in zap dnode
: zap-lookup ( dn prop$ -- [ n ] not-found? )
rot dup get-zap ( prop$ dn zp )
dup zap-type case
uzap# of
>uzap-ent swap dn-bsize ( prop$ uz len )
['] uz-nmlook -rot ( prop$ xt uz len )
uzap-apply if ( prop$ uz )
mze_value -rot 2drop ( n )
false ( n found )
else ( prop$ )
2drop true ( !found )
then ( [ n ] not-found? )
endof
fzap# of
['] fz-nmlook -rot ( prop$ xt dn fz )
fzap-apply if ( prop$ dn le )
entry-int-val ( prop$ n )
-rot 2drop false ( n found )
else ( prop$ )
2drop true ( !found )
then ( [ n ] not-found? )
endof
3drop 2drop true ( !found )
endcase ( [ n ] not-found? )
;
[ifdef] strlookup
: zap-lookup-str ( dn prop$ -- [ val$ ] not-found? )
rot dup get-zap ( prop$ dn zp )
dup zap-type fzap# <> if ( prop$ dn zp )
2drop 2drop true exit ( !found )
then ( prop$ dn zp )
['] fz-nmlook -rot ( prop$ xt dn fz )
fzap-apply if ( prop$ dn le )
entry-val$ 2swap 2drop false ( val$ found )
else ( prop$ )
2drop true ( !found )
then ( [ val$ ] not-found? )
;
[then]
[ifdef] bigbootblk
: fz-print ( dn le -- false )
entry-name$ type cr false
;
: uz-print ( uz -- false )
uzap-name$ type cr false
;
: zap-print ( dn -- )
dup get-zap ( dn zp )
dup zap-type case
uzap# of
>uzap-ent swap dn-bsize ( uz len )
['] uz-print -rot ( xt uz len )
uzap-apply ( false )
endof
fzap# of
['] fz-print -rot ( xt dn fz )
fzap-apply ( false )
endof
3drop false ( false )
endcase ( false )
drop ( )
;
[then]
\
\ ZFS object set (DSL) routines
\
1 constant pool-dir#
: dd_head_dataset_obj ( dd -- n ) h# 8 + x@ ;
: dd_child_dir_zapobj ( dd -- n ) h# 20 + x@ ;
: ds_snapnames_zapobj ( ds -- n ) h# 20 + x@ ;
: ds_bp ( ds -- p ) h# 80 + ;
0 instance value mos-dn
0 instance value obj-dir
0 instance value root-dsl
0 instance value root-dsl#
0 instance value fs-dn
\ dn-cache contains dc-dn's contents at dc-blk#
\ dc-dn will be either mos-dn or fs-dn
0 instance value dn-cache
0 instance value dc-dn
0 instance value dc-blk#
alias >dsl-dir dn_bonus
alias >dsl-ds dn_bonus
: #dn/blk ( dn -- n ) dn-bsize /dnode / ;
\ read block into dn-cache
: get-dnblk ( dn blk# -- )
lblk#>bp dn-cache swap ( adr bp )
dup bp-lsize swap read-bp ( )
;
\ read obj# from objset dir dn into dnode
: get-dnode ( dn obj# -- )
\ check dn-cache
2dup swap #dn/blk /mod ( dn obj# off# blk# )
swap >r nip ( dn blk# r: off# )
2dup dc-blk# <> ( dn blk# dn !blk-hit? r: off# )
swap dc-dn <> or if ( dn blk# r: off# )
\ cache miss, fill from dir
2dup get-dnblk
over to dc-dn
dup to dc-blk#
then ( dn blk# r: off# )
\ index and copy
2drop r> /dnode * ( off )
dn-cache + ( dn-adr )
dnode /dnode move ( )
;
\ read meta object set from uber-block
: get-mos ( -- )
mos-dn /dnode ( adr len )
uber-block ub_rootbp read-bp
;
: get-mos-dnode ( obj# -- )
mos-dn swap get-dnode
;
\ get root dataset
: get-root-dsl ( -- )
\ read MOS
get-mos
\ read object dir
pool-dir# get-mos-dnode
dnode obj-dir /dnode move
\ read root dataset
obj-dir " root_dataset" zap-lookup if
" no root_dataset" die
then ( obj# )
dup to root-dsl#
get-mos-dnode ( )
dnode root-dsl /dnode move
;
\ find snapshot of given dataset
: snap-look ( snap$ ds-obj# -- [ss-obj# ] not-found? )
get-mos-dnode dnode >dsl-ds ( snap$ ds )
ds_snapnames_zapobj get-mos-dnode ( snap$ )
dnode -rot zap-lookup ( [ss-obj# ] not-found? )
;
\ dsl dir to dataset
: dir>ds ( dn -- obj# ) >dsl-dir dd_head_dataset_obj ;
\ look thru the dsl hierarchy for path
\ this looks almost exactly like a FS directory lookup
: dsl-lookup ( path$ -- [ ds-obj# ] not-found? )
root-dsl >r ( path$ r: root-dn )
begin
ascii / left-parse-string ( path$ file$ r: dn )
dup while
\ get child dir zap dnode
r> >dsl-dir dd_child_dir_zapobj ( path$ file$ obj# )
get-mos-dnode ( path$ file$ )
\ check for snapshot names
ascii @ left-parse-string ( path$ snap$ file$ )
\ search it
dnode -rot zap-lookup if ( path$ snap$ )
\ not found
2drop 2drop true exit ( not-found )
then ( path$ snap$ obj# )
get-mos-dnode ( path$ snap$ )
\ lookup any snapshot name
dup if
\ must be last path component
2swap nip if ( snap$ )
2drop true exit ( not-found )
then
dnode dir>ds snap-look if ( )
true exit ( not-found )
then ( obj# )
false exit ( obj# found )
else 2drop then ( path$ )
dnode >r ( path$ r: dn )
repeat ( path$ file$ r: dn)
2drop 2drop r> drop ( )
\ found it, return dataset obj#
dnode dir>ds ( ds-obj# )
false ( ds-obj# found )
;
\ get objset from dataset
: get-objset ( adr dn -- )
>dsl-ds ds_bp /dnode swap read-bp
;
\
\ ZFS file-system (ZPL) routines
\
1 constant master-node#
d# 264 constant /znode
d# 56 constant /zn-slink
: zp_mode ( zn -- n ) h# 48 + x@ ;
: zp_size ( zn -- n ) h# 50 + x@ ;
: zp_parent ( zn -- n ) h# 58 + x@ ;
0 instance value bootfs-obj#
0 instance value root-obj#
0 instance value current-obj#
0 instance value search-obj#
alias >znode dn_bonus
: fsize ( dn -- n ) >znode zp_size ;
: ftype ( dn -- n ) >znode zp_mode h# f000 and ;
: dir? ( dn -- flag ) ftype h# 4000 = ;
: symlink? ( dn -- flag ) ftype h# a000 = ;
\ read obj# from fs objset
: get-fs-dnode ( obj# -- )
dup to current-obj#
fs-dn swap get-dnode ( )
;
\ get root-obj# from dataset
: get-rootobj# ( ds-obj# -- fsroot-obj# )
dup to bootfs-obj#
get-mos-dnode ( )
fs-dn dnode get-objset
\ get root obj# from master node
master-node# get-fs-dnode
dnode " ROOT" zap-lookup if
" no ROOT" die
then ( fsroot-obj# )
;
: prop>rootobj# ( -- )
obj-dir " pool_props" zap-lookup if
" no pool_props" die
then ( prop-obj# )
get-mos-dnode ( )
dnode " bootfs" zap-lookup if
" no bootfs" die
then ( ds-obj# )
get-rootobj# ( fsroot-obj# )
;
: fs>rootobj# ( fs$ -- root-obj# not-found? )
\ skip pool name
ascii / left-parse-string 2drop
\ lookup fs in dsl
dsl-lookup if ( )
true exit ( not-found )
then ( ds-obj# )
get-rootobj# ( fsroot-obj# )
false ( fsroot-obj# found )
;
\ lookup file is current directory
: dirlook ( file$ dn -- not-found? )
\ . and .. are magic
-rot 2dup " ." $= if ( dn file$ )
3drop false exit ( found )
then
2dup " .." $= if
2drop >znode zp_parent ( obj# )
else ( dn file$ )
\ search dir
current-obj# to search-obj#
zap-lookup if ( )
true exit ( not-found )
then ( obj# )
then ( obj# )
get-fs-dnode false ( found )
;
/buf-len instance buffer: fpath-buf
: clr-fpath-buf ( -- ) fpath-buf /buf-len erase ;
: fpath-buf$ ( -- path$ ) fpath-buf cscount ;
\ copy symlink target to adr
: readlink ( dst dn -- )
dup fsize tuck /zn-slink > if ( dst size dn )
\ contents in 1st block
temp-space over dn-bsize ( dst size dn t-adr bsize )
rot 0 lblk#>bp read-bp ( dst size )
temp-space ( dst size src )
else ( dst size dn )
\ contents in dnode
>znode /znode + ( dst size src )
then ( dst size src )
-rot move ( )
;
\ modify tail to account for symlink
: follow-symlink ( tail$ -- tail$' )
clr-fpath-buf ( tail$ )
fpath-buf dnode readlink
\ append to current path
?dup if ( tail$ )
" /" fpath-buf$ $append ( tail$ )
fpath-buf$ $append ( )
else drop then ( )
fpath-buf$ ( path$ )
\ get directory that starts changed path
over c@ ascii / = if ( path$ )
str++ root-obj# ( path$' obj# )
else ( path$ )
search-obj# ( path$ obj# )
then ( path$ obj# )
get-fs-dnode ( path$ )
;
\ open dnode at path
: lookup ( path$ -- not-found? )
\ get directory that starts path
over c@ ascii / = if
str++ root-obj# ( path$' obj# )
else
current-obj# ( path$ obj# )
then ( path$ obj# )
get-fs-dnode ( path$ )
\ lookup each path component
begin ( path$ )
ascii / left-parse-string ( path$ file$ )
dup while
dnode dir? 0= if
2drop true exit ( not-found )
then ( path$ file$ )
dnode dirlook if ( path$ )
2drop true exit ( not-found )
then ( path$ )
dnode symlink? if
follow-symlink ( path$' )
then ( path$ )
repeat ( path$ file$ )
2drop 2drop false ( found )
;
\
\ ZFS volume (ZVOL) routines
\
1 constant zvol-data#
2 constant zvol-prop#
0 instance value zv-dn
: get-zvol ( zvol$ -- not-found? )
dsl-lookup if
drop true exit ( failed )
then ( ds-obj# )
\ get zvol objset
get-mos-dnode ( )
zv-dn dnode get-objset
false ( succeeded )
;
\ get zvol data dnode
: zvol-data ( -- )
zv-dn zvol-data# get-dnode
;
: zvol-size ( -- size )
zv-dn zvol-prop# get-dnode
dnode " size" zap-lookup if
" no zvol size" die
then ( size )
;
\
\ ZFS installation routines
\
\ ZFS file interface
struct
/x field >busy
/x field >offset
/x field >fsize
/dnode field >dnode
constant /file-record
d# 10 constant #opens
#opens /file-record * constant /file-records
/file-records instance buffer: file-records
-1 instance value current-fd
: fd>record ( fd -- rec ) /file-record * file-records + ;
: file-offset@ ( -- off ) current-fd fd>record >offset x@ ;
: file-offset! ( off -- ) current-fd fd>record >offset x! ;
: file-dnode ( -- dn ) current-fd fd>record >dnode ;
: file-size ( -- size ) current-fd fd>record >fsize x@ ;
: file-bsize ( -- bsize ) file-dnode dn-bsize ;
\ find free fd slot
: get-slot ( -- fd false | true )
#opens 0 do
i fd>record >busy x@ 0= if
i false unloop exit
then
loop true
;
: free-slot ( fd -- )
0 swap fd>record >busy x!
;
\ init fd to offset 0 and copy dnode
: init-fd ( fsize fd -- )
fd>record ( fsize rec )
dup >busy 1 swap x!
dup >dnode dnode swap /dnode move
dup >fsize rot swap x! ( rec )
>offset 0 swap x! ( )
;
\ make fd current
: set-fd ( fd -- error? )
dup fd>record >busy x@ 0= if ( fd )
drop true exit ( failed )
then ( fd )
to current-fd false ( succeeded )
;
\ read next fs block
: file-bread ( adr -- )
file-bsize ( adr len )
file-offset@ over / ( adr len blk# )
file-dnode swap lblk#>bp ( adr len bp )
read-bp ( )
;
\ advance file io stack by n
: fio+ ( # adr len n -- #+n adr+n len-n )
dup file-offset@ + file-offset!
dup >r - -rot ( len' # adr r: n )
r@ + -rot ( adr' len' # r: n )
r> + -rot ( #' adr' len' )
;
/max-bsize 5 *
/uber-block +
/dnode 6 * +
/disk-block +
constant alloc-size
: allocate-buffers ( -- )
alloc-size h# a0.0000 vmem-alloc dup 0= if
" no memory" die
then ( adr )
dup to temp-space /max-bsize + ( adr )
dup to dn-cache /max-bsize + ( adr )
dup to blk-space /max-bsize + ( adr )
dup to ind-cache /max-bsize + ( adr )
dup to zap-space /max-bsize + ( adr )
dup to uber-block /uber-block + ( adr )
dup to mos-dn /dnode + ( adr )
dup to obj-dir /dnode + ( adr )
dup to root-dsl /dnode + ( adr )
dup to fs-dn /dnode + ( adr )
dup to zv-dn /dnode + ( adr )
dup to dnode /dnode + ( adr )
to gang-space ( )
\ zero instance buffers
file-records /file-records erase
bootprop-buf /buf-len erase
;
: release-buffers ( -- )
temp-space alloc-size mem-free
;
external
: open ( -- okay? )
my-args dev-open dup 0= if
exit ( failed )
then to dev-ih
allocate-buffers
scan-vdev
get-ub
get-root-dsl
true
;
: open-fs ( fs$ -- okay? )
fs>rootobj# if ( )
false ( failed )
else ( obj# )
to root-obj# true ( succeeded )
then ( okay? )
;
: close ( -- )
dev-ih dev-close
0 to dev-ih
release-buffers
;
: open-file ( path$ -- fd true | false )
\ open default fs if no open-fs
root-obj# 0= if
prop>rootobj# to root-obj#
then
get-slot if
2drop false exit ( failed )
then -rot ( fd path$ )
lookup if ( fd )
drop false exit ( failed )
then ( fd )
dnode fsize over init-fd
true ( fd succeeded )
;
: open-volume ( vol$ -- okay? )
get-slot if
2drop false exit ( failed )
then -rot ( fd vol$ )
get-zvol if ( fd )
drop false exit ( failed )
then
zvol-size over ( fd size fd )
zvol-data init-fd ( fd )
true ( fd succeeded )
;
: close-file ( fd -- )
free-slot ( )
;
: size-file ( fd -- size )
set-fd if 0 else file-size then
;
: seek-file ( off fd -- off true | false )
set-fd if ( off )
drop false exit ( failed )
then ( off )
dup file-size x> if ( off )
drop false exit ( failed )
then ( off )
dup file-offset! true ( off succeeded )
;
: read-file ( adr len fd -- #read )
set-fd if ( adr len )
2drop 0 exit ( 0 )
then ( adr len )
\ adjust len if reading past eof
dup file-offset@ + file-size x> if
dup file-offset@ + file-size - -
then
dup 0= if nip exit then
0 -rot ( #read adr len )
\ initial partial block
file-offset@ file-bsize mod ?dup if ( #read adr len off )
temp-space file-bread
2dup file-bsize swap - min ( #read adr len off cpy-len )
2over drop -rot ( #read adr len adr off cpy-len )
>r temp-space + swap ( #read adr len cpy-src adr r: cpy-len )
r@ move r> fio+ ( #read' adr' len' )
then ( #read adr len )
dup file-bsize / 0 ?do ( #read adr len )
over file-bread
file-bsize fio+ ( #read' adr' len' )
loop ( #read adr len )
\ final partial block
dup if ( #read adr len )
temp-space file-bread
2dup temp-space -rot move ( #read adr len )
dup fio+ ( #read' adr' 0 )
then 2drop ( #read )
;
: cinfo-file ( fd -- bsize fsize comp? )
set-fd if
0 0 0
else
file-bsize file-size ( bsize fsize )
\ zfs does internal compression
0 ( bsize fsize comp? )
then
;
\ read ramdisk fcode at rd-offset
: get-rd ( adr len -- )
rd-offset dev-ih read-disk
;
: bootprop
" /" bootprop$ $append
bootfs-obj# (xu.) bootprop$ $append
bootprop$ encode-string " zfs-bootfs" ( propval propname )
true
;
[ifdef] bigbootblk
: chdir ( dir$ -- )
current-obj# -rot ( obj# dir$ )
lookup if ( obj# )
to current-obj# ( )
." no such dir" cr exit
then ( obj# )
dnode dir? 0= if ( obj# )
to current-obj# ( )
." not a dir" cr exit
then drop ( )
;
: dir ( -- )
current-obj# get-fs-dnode
dnode zap-print
;
[then]
finish-device
pop-package