lix/src/libstore/store.cc

667 lines
18 KiB
C++
Raw Normal View History

#include <iostream>
2003-12-22 17:40:46 +01:00
#include <algorithm>
#include <sys/wait.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <utime.h>
2003-07-07 09:43:58 +02:00
#include "store.hh"
#include "globals.hh"
#include "db.hh"
#include "archive.hh"
#include "pathlocks.hh"
#include "gc.hh"
/* Nix database. */
static Database nixDB;
/* Database tables. */
/* dbValidPaths :: Path -> ()
The existence of a key $p$ indicates that path $p$ is valid (that
is, produced by a succesful build). */
static TableId dbValidPaths = 0;
/* dbReferences :: Path -> [Path]
This table lists the outgoing file system references for each
output path that has been built by a Nix derivation. These are
found by scanning the path for the hash components of input
paths. */
static TableId dbReferences = 0;
/* dbReferers :: Path -> [Path]
This table is just the reverse mapping of dbReferences. */
static TableId dbReferers = 0;
/* dbSubstitutes :: Path -> [[Path]]
Each pair $(p, subs)$ tells Nix that it can use any of the
substitutes in $subs$ to build path $p$. Each substitute defines a
command-line invocation of a program (i.e., the first list element
is the full path to the program, the remaining elements are
arguments).
The main purpose of this is for distributed caching of derivates.
One system can compute a derivate and put it on a website (as a Nix
archive), for instance, and then another system can register a
substitute for that derivate. The substitute in this case might be
2005-01-20 17:01:07 +01:00
a Nix derivation that fetches the Nix archive.
*/
static TableId dbSubstitutes = 0;
/* dbDerivers :: Path -> [Path]
This table lists the derivation used to build a path. There can
only be multiple such paths for fixed-output derivations (i.e.,
derivations specifying an expected hash). */
static TableId dbDerivers = 0;
bool Substitute::operator == (const Substitute & sub)
{
return program == sub.program
&& args == sub.args;
}
void openDB()
{
if (readOnlyMode) return;
try {
nixDB.open(nixDBPath);
} catch (DbNoPermission & e) {
printMsg(lvlTalkative, "cannot access Nix database; continuing anyway");
readOnlyMode = true;
return;
}
dbValidPaths = nixDB.openTable("validpaths");
dbReferences = nixDB.openTable("references");
dbReferers = nixDB.openTable("referers");
dbSubstitutes = nixDB.openTable("substitutes");
dbDerivers = nixDB.openTable("derivers");
}
void initDB()
{
}
void createStoreTransaction(Transaction & txn)
{
Transaction txn2(nixDB);
txn2.moveTo(txn);
}
/* Path copying. */
struct CopySink : DumpSink
{
int fd;
virtual void operator () (const unsigned char * data, unsigned int len)
{
writeFull(fd, data, len);
}
};
struct CopySource : RestoreSource
{
int fd;
virtual void operator () (unsigned char * data, unsigned int len)
{
readFull(fd, data, len);
}
};
void copyPath(const Path & src, const Path & dst)
{
debug(format("copying `%1%' to `%2%'") % src % dst);
/* Unfortunately C++ doesn't support coprocedures, so we have no
nice way to chain CopySink and CopySource together. Instead we
fork off a child to run the sink. (Fork-less platforms should
use a thread). */
/* Create a pipe. */
2004-06-22 11:51:44 +02:00
Pipe pipe;
pipe.create();
/* Fork. */
2004-06-22 11:51:44 +02:00
Pid pid;
pid = fork();
switch (pid) {
case -1:
throw SysError("unable to fork");
case 0: /* child */
try {
2004-06-22 11:51:44 +02:00
pipe.writeSide.close();
CopySource source;
2004-06-22 11:51:44 +02:00
source.fd = pipe.readSide;
restorePath(dst, source);
_exit(0);
} catch (exception & e) {
cerr << "error: " << e.what() << endl;
}
_exit(1);
}
/* Parent. */
2004-06-22 11:51:44 +02:00
pipe.readSide.close();
CopySink sink;
2004-06-22 11:51:44 +02:00
sink.fd = pipe.writeSide;
{
SwitchToOriginalUser sw;
dumpPath(src, sink);
}
/* Wait for the child to finish. */
2004-06-22 11:51:44 +02:00
int status = pid.wait(true);
if (!statusOk(status))
2004-06-22 10:50:25 +02:00
throw Error(format("cannot copy `%1% to `%2%': child %3%")
% src % dst % statusToString(status));
}
bool isInStore(const Path & path)
{
return path[0] == '/'
&& path.compare(0, nixStore.size(), nixStore) == 0
&& path.size() >= nixStore.size() + 2
&& path[nixStore.size()] == '/';
}
bool isStorePath(const Path & path)
{
return isInStore(path)
&& path.find('/', nixStore.size() + 1) == Path::npos;
}
void assertStorePath(const Path & path)
{
if (!isStorePath(path))
throw Error(format("path `%1%' is not in the Nix store") % path);
}
Path toStorePath(const Path & path)
{
if (!isInStore(path))
throw Error(format("path `%1%' is not in the Nix store") % path);
unsigned int slash = path.find('/', nixStore.size() + 1);
if (slash == Path::npos)
return path;
else
return Path(path, 0, slash);
}
void canonicalisePathMetaData(const Path & path)
{
checkInterrupt();
struct stat st;
if (lstat(path.c_str(), &st))
throw SysError(format("getting attributes of path `%1%'") % path);
if (!S_ISLNK(st.st_mode)) {
/* Mask out all type related bits. */
mode_t mode = st.st_mode & ~S_IFMT;
if (mode != 0444 && mode != 0555) {
mode = (st.st_mode & S_IFMT)
| 0444
| (st.st_mode & S_IXUSR ? 0111 : 0);
if (chmod(path.c_str(), mode) == -1)
throw SysError(format("changing mode of `%1%' to %2$o") % path % mode);
}
if (st.st_uid != getuid() || st.st_gid != getgid()) {
if (chown(path.c_str(), getuid(), getgid()) == -1)
throw SysError(format("changing owner/group of `%1%' to %2%/%3%")
% path % getuid() % getgid());
}
if (st.st_mtime != 0) {
struct utimbuf utimbuf;
utimbuf.actime = st.st_atime;
utimbuf.modtime = 0;
if (utime(path.c_str(), &utimbuf) == -1)
throw SysError(format("changing modification time of `%1%'") % path);
}
}
if (S_ISDIR(st.st_mode)) {
Strings names = readDirectory(path);
for (Strings::iterator i = names.begin(); i != names.end(); ++i)
canonicalisePathMetaData(path + "/" + *i);
}
}
static bool isValidPathTxn(const Transaction & txn, const Path & path)
{
string s;
return nixDB.queryString(txn, dbValidPaths, path, s);
}
bool isValidPath(const Path & path)
{
return isValidPathTxn(noTxn, path);
}
static Substitutes readSubstitutes(const Transaction & txn,
const Path & srcPath);
static bool isRealisablePath(const Transaction & txn, const Path & path)
{
return isValidPathTxn(txn, path)
|| readSubstitutes(txn, path).size() > 0;
}
static PathSet getReferers(const Transaction & txn, const Path & storePath)
{
Paths referers;
nixDB.queryStrings(txn, dbReferers, storePath, referers);
return PathSet(referers.begin(), referers.end());
}
void setReferences(const Transaction & txn, const Path & storePath,
const PathSet & references)
{
if (!isRealisablePath(txn, storePath))
throw Error(
format("cannot set references for path `%1%' which is invalid and has no substitutes")
% storePath);
Paths oldReferences;
2005-01-27 18:48:14 +01:00
nixDB.queryStrings(txn, dbReferences, storePath, oldReferences);
nixDB.setStrings(txn, dbReferences, storePath,
Paths(references.begin(), references.end()));
/* Update the referers mappings of all referenced paths. */
for (PathSet::const_iterator i = references.begin();
i != references.end(); ++i)
{
PathSet referers = getReferers(txn, *i);
referers.insert(storePath);
nixDB.setStrings(txn, dbReferers, *i,
Paths(referers.begin(), referers.end()));
}
/* Remove referer mappings from paths that are no longer
references. */
for (Paths::iterator i = oldReferences.begin();
i != oldReferences.end(); ++i)
if (references.find(*i) == references.end()) {
PathSet referers = getReferers(txn, *i);
referers.erase(storePath);
nixDB.setStrings(txn, dbReferers, *i,
Paths(referers.begin(), referers.end()));
}
}
void queryReferences(const Path & storePath, PathSet & references)
{
Paths references2;
if (!isRealisablePath(noTxn, storePath))
throw Error(format("path `%1%' is not valid") % storePath);
nixDB.queryStrings(noTxn, dbReferences, storePath, references2);
references.insert(references2.begin(), references2.end());
}
void queryReferers(const Path & storePath, PathSet & referers)
{
if (!isRealisablePath(noTxn, storePath))
throw Error(format("path `%1%' is not valid") % storePath);
PathSet referers2 = getReferers(noTxn, storePath);
referers.insert(referers2.begin(), referers2.end());
}
void setDeriver(const Transaction & txn, const Path & storePath,
const Path & deriver)
{
assertStorePath(storePath);
if (deriver == "") return;
assertStorePath(deriver);
if (!isRealisablePath(txn, storePath))
throw Error(format("path `%1%' is not valid") % storePath);
nixDB.setString(txn, dbDerivers, storePath, deriver);
}
Path queryDeriver(const Transaction & txn, const Path & storePath)
{
if (!isRealisablePath(noTxn, storePath))
throw Error(format("path `%1%' is not valid") % storePath);
Path deriver;
if (nixDB.queryString(txn, dbDerivers, storePath, deriver))
return deriver;
else
return "";
}
static Substitutes readSubstitutes(const Transaction & txn,
const Path & srcPath)
{
Strings ss;
nixDB.queryStrings(txn, dbSubstitutes, srcPath, ss);
Substitutes subs;
for (Strings::iterator i = ss.begin(); i != ss.end(); ++i) {
if (i->size() < 4 || (*i)[3] != 0) {
/* Old-style substitute. !!! remove this code
eventually? */
break;
}
Strings ss2 = unpackStrings(*i);
if (ss2.size() == 3) {
/* Another old-style substitute. */
continue;
}
if (ss2.size() != 2) throw Error("malformed substitute");
Strings::iterator j = ss2.begin();
Substitute sub;
sub.program = *j++;
sub.args = unpackStrings(*j++);
subs.push_back(sub);
}
return subs;
}
static void writeSubstitutes(const Transaction & txn,
const Path & srcPath, const Substitutes & subs)
{
Strings ss;
for (Substitutes::const_iterator i = subs.begin();
i != subs.end(); ++i)
{
Strings ss2;
ss2.push_back(i->program);
ss2.push_back(packStrings(i->args));
ss.push_back(packStrings(ss2));
}
nixDB.setStrings(txn, dbSubstitutes, srcPath, ss);
}
void registerSubstitute(const Transaction & txn,
const Path & srcPath, const Substitute & sub)
{
assertStorePath(srcPath);
Substitutes subs = readSubstitutes(txn, srcPath);
/* New substitutes take precedence over old ones. If the
substitute is already present, it's moved to the front. */
remove(subs.begin(), subs.end(), sub);
subs.push_front(sub);
writeSubstitutes(txn, srcPath, subs);
}
Substitutes querySubstitutes(const Transaction & txn, const Path & srcPath)
{
return readSubstitutes(txn, srcPath);
}
void clearSubstitutes()
{
Transaction txn(nixDB);
/* Iterate over all paths for which there are substitutes. */
Paths subKeys;
nixDB.enumTable(txn, dbSubstitutes, subKeys);
for (Paths::iterator i = subKeys.begin(); i != subKeys.end(); ++i) {
/* Delete all substitutes for path *i. */
nixDB.delPair(txn, dbSubstitutes, *i);
}
txn.commit();
}
void registerValidPath(const Transaction & txn,
const Path & _path, const Hash & hash, const PathSet & references,
const Path & deriver)
{
Path path(canonPath(_path));
assertStorePath(path);
assert(hash.type == htSHA256);
debug(format("registering path `%1%'") % path);
nixDB.setString(txn, dbValidPaths, path, "sha256:" + printHash(hash));
setReferences(txn, path, references);
/* Check that all referenced paths are also valid. */
for (PathSet::iterator i = references.begin(); i != references.end(); ++i)
if (!isValidPathTxn(txn, *i))
throw Error(format("cannot register path `%1%' as valid, since its reference `%2%' is invalid")
% path % *i);
setDeriver(txn, path, deriver);
}
/* Invalidate a path. The caller is responsible for checking that
there are no referers. */
static void invalidatePath(const Path & path, Transaction & txn)
{
debug(format("unregistering path `%1%'") % path);
/* Clear the `references' entry for this path, as well as the
inverse `referers' entries, and the `derivers' entry; but only
if there are no substitutes for this path. This maintains the
cleanup invariant. */
if (querySubstitutes(txn, path).size() == 0) {
setReferences(txn, path, PathSet());
nixDB.delPair(txn, dbDerivers, path);
}
nixDB.delPair(txn, dbValidPaths, path);
}
Path makeStorePath(const string & type,
* Removed the `id' attribute hack. * Formalise the notion of fixed-output derivations, i.e., derivations for which a cryptographic hash of the output is known in advance. Changes to such derivations should not propagate upwards through the dependency graph. Previously this was done by specifying the hash component of the output path through the `id' attribute, but this is insecure since you can lie about it (i.e., you can specify any hash and then produce a completely different output). Now the responsibility for checking the output is moved from the builder to Nix itself. A fixed-output derivation can be created by specifying the `outputHash' and `outputHashAlgo' attributes, the latter taking values `md5', `sha1', and `sha256', and the former specifying the actual hash in hexadecimal or in base-32 (auto-detected by looking at the length of the attribute value). MD5 is included for compatibility but should be considered deprecated. * Removed the `drvPath' pseudo-attribute in derivation results. It's no longer necessary. * Cleaned up the support for multiple output paths in derivation store expressions. Each output now has a unique identifier (e.g., `out', `devel', `docs'). Previously there was no way to tell output paths apart at the store expression level. * `nix-hash' now has a flag `--base32' to specify that the hash should be printed in base-32 notation. * `fetchurl' accepts parameters `sha256' and `sha1' in addition to `md5'. * `nix-prefetch-url' now prints out a SHA-1 hash in base-32. (TODO: a flag to specify the hash.)
2005-01-17 17:55:19 +01:00
const Hash & hash, const string & suffix)
{
/* e.g., "source:sha256:1abc...:/nix/store:foo.tar.gz" */
string s = type + ":sha256:" + printHash(hash) + ":"
+ nixStore + ":" + suffix;
return nixStore + "/"
* Removed the `id' attribute hack. * Formalise the notion of fixed-output derivations, i.e., derivations for which a cryptographic hash of the output is known in advance. Changes to such derivations should not propagate upwards through the dependency graph. Previously this was done by specifying the hash component of the output path through the `id' attribute, but this is insecure since you can lie about it (i.e., you can specify any hash and then produce a completely different output). Now the responsibility for checking the output is moved from the builder to Nix itself. A fixed-output derivation can be created by specifying the `outputHash' and `outputHashAlgo' attributes, the latter taking values `md5', `sha1', and `sha256', and the former specifying the actual hash in hexadecimal or in base-32 (auto-detected by looking at the length of the attribute value). MD5 is included for compatibility but should be considered deprecated. * Removed the `drvPath' pseudo-attribute in derivation results. It's no longer necessary. * Cleaned up the support for multiple output paths in derivation store expressions. Each output now has a unique identifier (e.g., `out', `devel', `docs'). Previously there was no way to tell output paths apart at the store expression level. * `nix-hash' now has a flag `--base32' to specify that the hash should be printed in base-32 notation. * `fetchurl' accepts parameters `sha256' and `sha1' in addition to `md5'. * `nix-prefetch-url' now prints out a SHA-1 hash in base-32. (TODO: a flag to specify the hash.)
2005-01-17 17:55:19 +01:00
+ printHash32(compressHash(hashString(htSHA256, s), 20))
+ "-" + suffix;
}
Path addToStore(const Path & _srcPath)
{
Path srcPath(absPath(_srcPath));
debug(format("adding `%1%' to the store") % srcPath);
Hash h(htSHA256);
{
SwitchToOriginalUser sw;
* Removed the `id' attribute hack. * Formalise the notion of fixed-output derivations, i.e., derivations for which a cryptographic hash of the output is known in advance. Changes to such derivations should not propagate upwards through the dependency graph. Previously this was done by specifying the hash component of the output path through the `id' attribute, but this is insecure since you can lie about it (i.e., you can specify any hash and then produce a completely different output). Now the responsibility for checking the output is moved from the builder to Nix itself. A fixed-output derivation can be created by specifying the `outputHash' and `outputHashAlgo' attributes, the latter taking values `md5', `sha1', and `sha256', and the former specifying the actual hash in hexadecimal or in base-32 (auto-detected by looking at the length of the attribute value). MD5 is included for compatibility but should be considered deprecated. * Removed the `drvPath' pseudo-attribute in derivation results. It's no longer necessary. * Cleaned up the support for multiple output paths in derivation store expressions. Each output now has a unique identifier (e.g., `out', `devel', `docs'). Previously there was no way to tell output paths apart at the store expression level. * `nix-hash' now has a flag `--base32' to specify that the hash should be printed in base-32 notation. * `fetchurl' accepts parameters `sha256' and `sha1' in addition to `md5'. * `nix-prefetch-url' now prints out a SHA-1 hash in base-32. (TODO: a flag to specify the hash.)
2005-01-17 17:55:19 +01:00
h = hashPath(htSHA256, srcPath);
}
string baseName = baseNameOf(srcPath);
Path dstPath = makeStorePath("source", h, baseName);
addTempRoot(dstPath);
if (!readOnlyMode && !isValidPath(dstPath)) {
/* The first check above is an optimisation to prevent
unnecessary lock acquisition. */
PathSet lockPaths;
lockPaths.insert(dstPath);
PathLocks outputLock(lockPaths);
if (!isValidPath(dstPath)) {
if (pathExists(dstPath)) deletePath(dstPath);
/* !!! race: srcPath might change between hashPath() and
here! */
copyPath(srcPath, dstPath);
* Removed the `id' attribute hack. * Formalise the notion of fixed-output derivations, i.e., derivations for which a cryptographic hash of the output is known in advance. Changes to such derivations should not propagate upwards through the dependency graph. Previously this was done by specifying the hash component of the output path through the `id' attribute, but this is insecure since you can lie about it (i.e., you can specify any hash and then produce a completely different output). Now the responsibility for checking the output is moved from the builder to Nix itself. A fixed-output derivation can be created by specifying the `outputHash' and `outputHashAlgo' attributes, the latter taking values `md5', `sha1', and `sha256', and the former specifying the actual hash in hexadecimal or in base-32 (auto-detected by looking at the length of the attribute value). MD5 is included for compatibility but should be considered deprecated. * Removed the `drvPath' pseudo-attribute in derivation results. It's no longer necessary. * Cleaned up the support for multiple output paths in derivation store expressions. Each output now has a unique identifier (e.g., `out', `devel', `docs'). Previously there was no way to tell output paths apart at the store expression level. * `nix-hash' now has a flag `--base32' to specify that the hash should be printed in base-32 notation. * `fetchurl' accepts parameters `sha256' and `sha1' in addition to `md5'. * `nix-prefetch-url' now prints out a SHA-1 hash in base-32. (TODO: a flag to specify the hash.)
2005-01-17 17:55:19 +01:00
Hash h2 = hashPath(htSHA256, dstPath);
if (h != h2)
throw Error(format("contents of `%1%' changed while copying it to `%2%' (%3% -> %4%)")
% srcPath % dstPath % printHash(h) % printHash(h2));
canonicalisePathMetaData(dstPath);
Transaction txn(nixDB);
registerValidPath(txn, dstPath, h, PathSet(), "");
txn.commit();
}
outputLock.setDeletion(true);
}
return dstPath;
}
Path addTextToStore(const string & suffix, const string & s,
const PathSet & references)
{
* Removed the `id' attribute hack. * Formalise the notion of fixed-output derivations, i.e., derivations for which a cryptographic hash of the output is known in advance. Changes to such derivations should not propagate upwards through the dependency graph. Previously this was done by specifying the hash component of the output path through the `id' attribute, but this is insecure since you can lie about it (i.e., you can specify any hash and then produce a completely different output). Now the responsibility for checking the output is moved from the builder to Nix itself. A fixed-output derivation can be created by specifying the `outputHash' and `outputHashAlgo' attributes, the latter taking values `md5', `sha1', and `sha256', and the former specifying the actual hash in hexadecimal or in base-32 (auto-detected by looking at the length of the attribute value). MD5 is included for compatibility but should be considered deprecated. * Removed the `drvPath' pseudo-attribute in derivation results. It's no longer necessary. * Cleaned up the support for multiple output paths in derivation store expressions. Each output now has a unique identifier (e.g., `out', `devel', `docs'). Previously there was no way to tell output paths apart at the store expression level. * `nix-hash' now has a flag `--base32' to specify that the hash should be printed in base-32 notation. * `fetchurl' accepts parameters `sha256' and `sha1' in addition to `md5'. * `nix-prefetch-url' now prints out a SHA-1 hash in base-32. (TODO: a flag to specify the hash.)
2005-01-17 17:55:19 +01:00
Hash hash = hashString(htSHA256, s);
Path dstPath = makeStorePath("text", hash, suffix);
addTempRoot(dstPath);
if (!readOnlyMode && !isValidPath(dstPath)) {
PathSet lockPaths;
lockPaths.insert(dstPath);
PathLocks outputLock(lockPaths);
if (!isValidPath(dstPath)) {
if (pathExists(dstPath)) deletePath(dstPath);
2003-11-22 16:58:34 +01:00
writeStringToFile(dstPath, s);
canonicalisePathMetaData(dstPath);
Transaction txn(nixDB);
registerValidPath(txn, dstPath,
hashPath(htSHA256, dstPath), references, "");
txn.commit();
}
outputLock.setDeletion(true);
}
return dstPath;
}
void deleteFromStore(const Path & _path)
2003-06-23 16:40:49 +02:00
{
Path path(canonPath(_path));
assertStorePath(path);
Transaction txn(nixDB);
if (isValidPathTxn(txn, path)) {
PathSet referers = getReferers(txn, path);
if (referers.size() > 1 ||
(referers.size() == 1 &&
*referers.begin() != path))
throw Error(format("cannot delete path `%1%' because it is in use") % path);
invalidatePath(path, txn);
}
txn.commit();
deletePath(path);
2003-06-23 16:40:49 +02:00
}
void verifyStore()
{
Transaction txn(nixDB);
Paths paths;
PathSet validPaths;
nixDB.enumTable(txn, dbValidPaths, paths);
for (Paths::iterator i = paths.begin(); i != paths.end(); ++i) {
Path path = *i;
if (!pathExists(path)) {
printMsg(lvlError, format("path `%1%' disappeared") % path);
invalidatePath(path, txn);
} else if (!isStorePath(path)) {
printMsg(lvlError, format("path `%1%' is not in the Nix store") % path);
invalidatePath(path, txn);
} else
validPaths.insert(path);
}
/* !!! the code below does not allow transitive substitutes.
I.e., if B is a substitute of A, then B must be a valid path.
B cannot itself be invalid but have a substitute. */
/* "Usable" paths are those that are valid or have a substitute.
These are the paths that are allowed to appear in the
right-hand side of a sute mapping. */
PathSet usablePaths(validPaths);
/* Check that the values of the substitute mappings are valid
paths. */
Paths subKeys;
nixDB.enumTable(txn, dbSubstitutes, subKeys);
for (Paths::iterator i = subKeys.begin(); i != subKeys.end(); ++i) {
Substitutes subs = readSubstitutes(txn, *i);
if (subs.size() > 0)
usablePaths.insert(*i);
else
nixDB.delPair(txn, dbSubstitutes, *i);
}
txn.commit();
}