ref: 4040ea7a5eda976098131794e8df3b162cafa8f4
dir: /sys/src/liboventi/venti.txt/
protocol version id
both client and server send '\n' terminated ascii of the form
venti-<versions>-<software>
<software> = is a software id which is ignored but may be useful for debugging
If the server and client have the same version number then this
version is used. Otherwise backup to the greatest common
major version number, e.g. 1.00 2.00 etc. If no version in
common then abort. The idea is both client and server should
support a continusous range of major versions. The minor version
numbers are used for development purposes.
After protocol negotiation switch to binary format
all numbers sent big endian
strings are sent with 2 byte length without nulls and in utf-8 max size of 1K
RPC protocol
header
op[1] byte op
tid[2] transation id
client manages the tid space. Until session is established tid must equal 0.
i.e. only one outstanding op.
CipherStrength
None,
Auth,
Weak,
Strong
CipherAlgorithms
None
SSL3
TLS1
CompressionAlgorithms
None
Deflate
Thwack?
BlockTypes
Root
Pointer
Data
==============================================
TPing
RPing
RError
error: string
THello
version:string; know to be supported by both
uid: string
uhash[20] use for cipher boot strapping
cipherstrength[1]
ncipher[1];
cipher[ncipher];
ncompressor[1];
compressor[ncompressor];
RHello
sid: string
shash[20]; use for cipher bott strapping
cipher[1];
compressor[1];
use srp style authentication
g=2
N is safe prime 1024 bits - find a good prime!
x = H("venti" H(sid) H(uid) H(password))
v = g^x
a = random
b = random
B = (v + g^b)
A = g^a
u = first four bytes of H(B)
S = (B - g^x) ^ (a + u*x) = (A * v^u) ^ b
K = H(S)
M1 = H(H(versions) H(THello) H(RHello) A B K)
M2 = H(A M1 K)
TAuth0
A[128]
RAuth0
B[128]
TAuth1
M1[20]
RAuth1
M2[20]
push cipher
push compression
TWrite
length[2] max 56*1024
type[1]
data[length]
RWrite
hash[20]
TRead
hash[20]
type[1]
length[2]
RRead
length[2]
data[length]
============================================================
simplified access when trusting the server
reduces network bandwidth since pointer blocks are not
sent to the client - can also enable permission checking
RReadRoot
root[20]
TReadRoot
name: string
type: string
blocksize[2]
nblock[8]
time[4]
uid: string;
gid: string
RReadData
root[20]
block[8]
length[2]
TReadData
length[2]
collision[1]
data[length]
==============================================
maybe some clients should be required to navigate to block via root nodes.
This would enable permission checking via access to the root node.
RTagOpen
tag[2]
root[20]
TTagOpen
QTagRead
tag[2]
type[1]
length[2]
RTagRead
length[2]
data[length]
QTagWalkRead
tag[2]
ntag[2] can reuse a tag to do an implict clunk
index[2]
type[1]
length[2]
RTagWalkRead
length[2]
data[length]
RTagClunk
tag[2]
TTagClunk
============================
Types of blocks
Data
Root
name[128]
type[128]
score[20] - DirBlock
Pointer
score[20]* the number of hashes can be less than fanout when the
tree is not full
DirBlock
DirEntry[32]*
DirEntry
pointersize[2] - pointer block size
datasize[2] - data blocks size
flag[1] directory
size[7] in bytes - determines pointer depth - intermidate truncated block count as full
score[20] root of pointer blocks or data block
============================
mode flags
(1<<0) ModeOtherExec
(1<<1) ModeOtherWrite
(1<<2) ModeOtherRead
(1<<3) ModeGroupExec
(1<<4) ModeGroupWrite
(1<<5) ModeGroupRead
(1<<6) ModeOwnerExec
(1<<7) ModeOwnerWrite
(1<<8) ModeOwnerRead
(1<<9) ModeSticky
(1<<10) ModeSetUid
(1<<11) ModeSetGid
(1<<12) ModeAppend
(1<<13) ModeExclusive
(1<<14) ModeLink
(1<<15) ModeDir - duplicates dir entry
(1<<16) ModeHidden
(1<<17) ModeSystem
(1<<18) ModeArchive
(1<<19) ModeTemporary
(1<<18) ModeCompressed
(1<<19) ModeEncrypted
extraType
2 Plan9
version[4]
muid: string
3 NT Time
createTime[8];
modifyTime[8];
accessTime[8]
MetaEntry
name: string /* must be first */
direntry[4]
id[8]
uid: string
gui: string
mtime[4]
ctime[4]
atime[4]
mode[4]
extratype;
extrasize[2]
extra:[nextra]
MetaEntry Block
magic[4]
size[2]
free[2] - used to determine if work compacting
maxindex[2] - size of index table in bytes
nindex[2]
index[2*nindex]
per OS directory entries?
inode...