libstore/build: block io_uring

Change-Id: I45d3895f95abe1bc103a63969f444c334dbbf50d

doc/manual/change-authors.yml

@@ -96,6 +96,10 @@ puck:
   forgejo: puck
   github: puckipedia
 
+quantumjump:
+  display_name: Quantum Jump
+  github: QuantumBJump
+
 r-vdp:
   github: r-vdp
 

doc/manual/rl-next/fix-gc-dry-run.md

@@ -0,0 +1,29 @@
+---
+synopsis: Fix nix-collect-garbage --dry-run
+issues: [fj#432]
+cls: [1566]
+category: Fixes
+credits: [quantumjump]
+---
+
+`nix-collect-garbage --dry-run` did not previously give any output - it simply
+exited without even checking to see what paths would be deleted.
+
+```
+$ nix-collect-garbage --dry-run
+$
+```
+
+We updated the behaviour of the flag such that instead it prints out how many
+paths it *would* delete, but doesn't actually delete them.
+
+```
+$ nix-collect-garbage --dry-run
+finding garbage collector roots...
+determining live/dead paths...
+...
+<nix store paths>
+...
+2670 store paths deleted, 0.00MiB freed
+$
+```

flake.nix

@@ -166,6 +166,7 @@
           nixUnstable = prev.nixUnstable;
 
           check-headers = final.buildPackages.callPackage ./maintainers/check-headers.nix { };
+          check-syscalls = final.buildPackages.callPackage ./maintainers/check-syscalls.nix { };
 
           default-busybox-sandbox-shell = final.busybox.override {
             useMusl = true;

maintainers/check-syscalls.nix

@@ -0,0 +1,16 @@
+{
+  runCommandNoCC,
+  lib,
+  libseccomp,
+  writeShellScriptBin,
+}:
+let
+  syscalls-csv = runCommandNoCC "syscalls.csv" { } ''
+    echo ${lib.escapeShellArg libseccomp.src}
+    tar -xf ${lib.escapeShellArg libseccomp.src} --strip-components=2 ${libseccomp.name}/src/syscalls.csv
+    mv syscalls.csv "$out"
+  '';
+in
+writeShellScriptBin "check-syscalls" ''
+  ${./check-syscalls.sh} ${syscalls-csv}
+''

maintainers/check-syscalls.sh

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+
+set -e
+
+diff -u <(awk < src/libstore/build/local-derivation-goal.cc '/BEGIN extract-syscalls/ { extracting = 1; next }
+match($0, /allowSyscall\(ctx, SCMP_SYS\(([^)]*)\)\);|\/\/ skip ([^ ]*)/, result) { print result[1] result[2] }
+/END extract-syscalls/ { extracting = 0; next }') <(tail -n+2 "$1" | cut -d, -f 1)

package.nix

@@ -412,6 +412,7 @@ stdenv.mkDerivation (finalAttrs: {
         # Lix specific packages
         pre-commit-checks,
         contribNotice,
+        check-syscalls,
       }:
       let
         glibcFix = lib.optionalAttrs (buildPlatform.isLinux && glibcLocales != null) {
@@ -465,6 +466,7 @@ stdenv.mkDerivation (finalAttrs: {
               pythonEnv
               # docker image tool
               skopeo
+              check-syscalls
               just
               nixfmt
               # Included above when internalApiDocs is true, but we set that to

src/libexpr/get-drvs.cc

@@ -45,7 +45,7 @@ DrvInfo::DrvInfo(EvalState & state, ref<Store> store, const std::string & drvPat
 }
 
 
-std::string DrvInfo::queryName() const
+std::string DrvInfo::queryName()
 {
     if (name == "" && attrs) {
         auto i = attrs->find(state->sName);
@@ -56,7 +56,7 @@ std::string DrvInfo::queryName() const
 }
 
 
-std::string DrvInfo::querySystem() const
+std::string DrvInfo::querySystem()
 {
     if (system == "" && attrs) {
         auto i = attrs->find(state->sSystem);
@@ -66,7 +66,7 @@ std::string DrvInfo::querySystem() const
 }
 
 
-std::optional<StorePath> DrvInfo::queryDrvPath() const
+std::optional<StorePath> DrvInfo::queryDrvPath()
 {
     if (!drvPath && attrs) {
         Bindings::iterator i = attrs->find(state->sDrvPath);
@@ -80,7 +80,7 @@ std::optional<StorePath> DrvInfo::queryDrvPath() const
 }
 
 
-StorePath DrvInfo::requireDrvPath() const
+StorePath DrvInfo::requireDrvPath()
 {
     if (auto drvPath = queryDrvPath())
         return *drvPath;
@@ -88,7 +88,7 @@ StorePath DrvInfo::requireDrvPath() const
 }
 
 
-StorePath DrvInfo::queryOutPath() const
+StorePath DrvInfo::queryOutPath()
 {
     if (!outPath && attrs) {
         Bindings::iterator i = attrs->find(state->sOutPath);
@@ -164,7 +164,7 @@ DrvInfo::Outputs DrvInfo::queryOutputs(bool withPaths, bool onlyOutputsToInstall
 }
 
 
-std::string DrvInfo::queryOutputName() const
+std::string DrvInfo::queryOutputName()
 {
     if (outputName == "" && attrs) {
         Bindings::iterator i = attrs->find(state->sOutputName);

src/libexpr/get-drvs.hh

@@ -19,11 +19,11 @@ public:
 private:
     EvalState * state;
 
-    mutable std::string name;
+    std::string name;
-    mutable std::string system;
+    std::string system;
-    mutable std::optional<std::optional<StorePath>> drvPath;
+    std::optional<std::optional<StorePath>> drvPath;
-    mutable std::optional<StorePath> outPath;
+    std::optional<StorePath> outPath;
-    mutable std::string outputName;
+    std::string outputName;
     Outputs outputs;
 
     /**
@@ -47,12 +47,12 @@ public:
     DrvInfo(EvalState & state, std::string attrPath, Bindings * attrs);
     DrvInfo(EvalState & state, ref<Store> store, const std::string & drvPathWithOutputs);
 
-    std::string queryName() const;
+    std::string queryName();
-    std::string querySystem() const;
+    std::string querySystem();
-    std::optional<StorePath> queryDrvPath() const;
+    std::optional<StorePath> queryDrvPath();
-    StorePath requireDrvPath() const;
+    StorePath requireDrvPath();
-    StorePath queryOutPath() const;
+    StorePath queryOutPath();
-    std::string queryOutputName() const;
+    std::string queryOutputName();
     /**
      * Return the unordered map of output names to (optional) output paths.
      * The "outputs to install" are determined by `meta.outputsToInstall`.

src/libmain/progress-bar.cc

@@ -356,7 +356,7 @@ std::chrono::milliseconds ProgressBar::draw(State & state, const std::optional<s
         line += "]";
     }
     if (printMultiline && !line.empty()) {
-        writeToStderr(filterANSIEscapes(line, false, width) + "\n");
+        writeToStderr(filterANSIEscapes(line, false, width) + ANSI_NORMAL "\n");
         state.lastLines++;
     }
 
@@ -398,7 +398,7 @@ std::chrono::milliseconds ProgressBar::draw(State & state, const std::optional<s
 
             if (printMultiline) {
                 if (state.lastLines < (height -1)) {
-                    writeToStderr(filterANSIEscapes(activity_line, false, width) + "\n");
+                    writeToStderr(filterANSIEscapes(activity_line, false, width) + ANSI_NORMAL "\n");
                     state.lastLines++;
                 } else moreActivities++;
             }

src/libstore/build/local-derivation-goal.cc

@@ -45,7 +45,6 @@
 #include <sys/prctl.h>
 #include <sys/syscall.h>
 #if HAVE_SECCOMP
-#include "linux/fchmodat2-compat.hh"
 #include <seccomp.h>
 #endif
 #define pivot_root(new_root, put_old) (syscall(SYS_pivot_root, new_root, put_old))
@@ -1363,6 +1362,12 @@ void LocalDerivationGoal::chownToBuilder(const Path & path)
         throw SysError("cannot change ownership of '%1%'", path);
 }
 
+#if HAVE_SECCOMP
+void allowSyscall(scmp_filter_ctx ctx, int syscall) {
+    if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW, syscall, 0) != 0)
+        throw SysError("unable to add seccomp rule");
+}
+#endif
 
 void setupSeccomp()
 {
@@ -1370,7 +1375,9 @@ void setupSeccomp()
 #if HAVE_SECCOMP
     scmp_filter_ctx ctx;
 
-    if (!(ctx = seccomp_init(SCMP_ACT_ALLOW)))
+    // Pretend that syscalls we don't yet know about don't exist.
+    // This is the best option for compatibility: after all, they did in fact not exist not too long ago.
+    if (!(ctx = seccomp_init(SCMP_ACT_ERRNO(ENOSYS))))
         throw SysError("unable to initialize seccomp mode 2");
 
     Finally cleanup([&]() {
@@ -1405,28 +1412,520 @@ void setupSeccomp()
         seccomp_arch_add(ctx, SCMP_ARCH_MIPSEL64N32) != 0)
         printError("unable to add mips64el-*abin32 seccomp architecture");
 
-    /* Prevent builders from creating setuid/setgid binaries. */
+    // This list is intended for machine consumption.
-    for (int perm : { S_ISUID, S_ISGID }) {
+    // Please keep its format, order and BEGIN/END markers.
-        if (seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(chmod), 1,
+    //
-                SCMP_A1(SCMP_CMP_MASKED_EQ, (scmp_datum_t) perm, (scmp_datum_t) perm)) != 0)
+    // Currently, it is up to date with libseccomp 2.5.5 and glibc 2.38.
-            throw SysError("unable to add seccomp rule");
+    // Run check-syscalls to determine which new syscalls should be added.
+    // New syscalls must be audited and handled in a way that blocks the following dangerous operations:
+    // * Creation of non-empty setuid/setgid files
+    // * Creation of extended attributes (including ACLs)
+    //
+    // BEGIN extract-syscalls
+    allowSyscall(ctx, SCMP_SYS(accept));
+    allowSyscall(ctx, SCMP_SYS(accept4));
+    allowSyscall(ctx, SCMP_SYS(access));
+    allowSyscall(ctx, SCMP_SYS(acct));
+    allowSyscall(ctx, SCMP_SYS(add_key));
+    allowSyscall(ctx, SCMP_SYS(adjtimex));
+    allowSyscall(ctx, SCMP_SYS(afs_syscall));
+    allowSyscall(ctx, SCMP_SYS(alarm));
+    allowSyscall(ctx, SCMP_SYS(arch_prctl));
+    allowSyscall(ctx, SCMP_SYS(arm_fadvise64_64));
+    allowSyscall(ctx, SCMP_SYS(arm_sync_file_range));
+    allowSyscall(ctx, SCMP_SYS(bdflush));
+    allowSyscall(ctx, SCMP_SYS(bind));
+    allowSyscall(ctx, SCMP_SYS(bpf));
+    allowSyscall(ctx, SCMP_SYS(break));
+    allowSyscall(ctx, SCMP_SYS(breakpoint));
+    allowSyscall(ctx, SCMP_SYS(brk));
+    allowSyscall(ctx, SCMP_SYS(cachectl));
+    allowSyscall(ctx, SCMP_SYS(cacheflush));
+    allowSyscall(ctx, SCMP_SYS(cachestat));
+    allowSyscall(ctx, SCMP_SYS(capget));
+    allowSyscall(ctx, SCMP_SYS(capset));
+    allowSyscall(ctx, SCMP_SYS(chdir));
+    // skip chmod (dangerous)
+    allowSyscall(ctx, SCMP_SYS(chown));
+    allowSyscall(ctx, SCMP_SYS(chown32));
+    allowSyscall(ctx, SCMP_SYS(chroot));
+    allowSyscall(ctx, SCMP_SYS(clock_adjtime));
+    allowSyscall(ctx, SCMP_SYS(clock_adjtime64));
+    allowSyscall(ctx, SCMP_SYS(clock_getres));
+    allowSyscall(ctx, SCMP_SYS(clock_getres_time64));
+    allowSyscall(ctx, SCMP_SYS(clock_gettime));
+    allowSyscall(ctx, SCMP_SYS(clock_gettime64));
+    allowSyscall(ctx, SCMP_SYS(clock_nanosleep));
+    allowSyscall(ctx, SCMP_SYS(clock_nanosleep_time64));
+    allowSyscall(ctx, SCMP_SYS(clock_settime));
+    allowSyscall(ctx, SCMP_SYS(clock_settime64));
+    allowSyscall(ctx, SCMP_SYS(clone));
+    allowSyscall(ctx, SCMP_SYS(clone3));
+    allowSyscall(ctx, SCMP_SYS(close));
+    allowSyscall(ctx, SCMP_SYS(close_range));
+    allowSyscall(ctx, SCMP_SYS(connect));
+    allowSyscall(ctx, SCMP_SYS(copy_file_range));
+    allowSyscall(ctx, SCMP_SYS(creat));
+    allowSyscall(ctx, SCMP_SYS(create_module));
+    allowSyscall(ctx, SCMP_SYS(delete_module));
+    allowSyscall(ctx, SCMP_SYS(dup));
+    allowSyscall(ctx, SCMP_SYS(dup2));
+    allowSyscall(ctx, SCMP_SYS(dup3));
+    allowSyscall(ctx, SCMP_SYS(epoll_create));
+    allowSyscall(ctx, SCMP_SYS(epoll_create1));
+    allowSyscall(ctx, SCMP_SYS(epoll_ctl));
+    allowSyscall(ctx, SCMP_SYS(epoll_ctl_old));
+    allowSyscall(ctx, SCMP_SYS(epoll_pwait));
+    allowSyscall(ctx, SCMP_SYS(epoll_pwait2));
+    allowSyscall(ctx, SCMP_SYS(epoll_wait));
+    allowSyscall(ctx, SCMP_SYS(epoll_wait_old));
+    allowSyscall(ctx, SCMP_SYS(eventfd));
+    allowSyscall(ctx, SCMP_SYS(eventfd2));
+    allowSyscall(ctx, SCMP_SYS(execve));
+    allowSyscall(ctx, SCMP_SYS(execveat));
+    allowSyscall(ctx, SCMP_SYS(exit));
+    allowSyscall(ctx, SCMP_SYS(exit_group));
+    allowSyscall(ctx, SCMP_SYS(faccessat));
+    allowSyscall(ctx, SCMP_SYS(faccessat2));
+    allowSyscall(ctx, SCMP_SYS(fadvise64));
+    allowSyscall(ctx, SCMP_SYS(fadvise64_64));
+    allowSyscall(ctx, SCMP_SYS(fallocate));
+    allowSyscall(ctx, SCMP_SYS(fanotify_init));
+    allowSyscall(ctx, SCMP_SYS(fanotify_mark));
+    allowSyscall(ctx, SCMP_SYS(fchdir));
+    // skip fchmod (dangerous)
+    // skip fchmodat (dangerous)
+    // skip fchmodat2 (requires glibc 2.39, dangerous)
+    allowSyscall(ctx, SCMP_SYS(fchown));
+    allowSyscall(ctx, SCMP_SYS(fchown32));
+    allowSyscall(ctx, SCMP_SYS(fchownat));
+    allowSyscall(ctx, SCMP_SYS(fcntl));
+    allowSyscall(ctx, SCMP_SYS(fcntl64));
+    allowSyscall(ctx, SCMP_SYS(fdatasync));
+    allowSyscall(ctx, SCMP_SYS(fgetxattr));
+    allowSyscall(ctx, SCMP_SYS(finit_module));
+    allowSyscall(ctx, SCMP_SYS(flistxattr));
+    allowSyscall(ctx, SCMP_SYS(flock));
+    allowSyscall(ctx, SCMP_SYS(fork));
+    allowSyscall(ctx, SCMP_SYS(fremovexattr));
+    allowSyscall(ctx, SCMP_SYS(fsconfig));
+    // skip fsetxattr (dangerous)
+    allowSyscall(ctx, SCMP_SYS(fsmount));
+    allowSyscall(ctx, SCMP_SYS(fsopen));
+    allowSyscall(ctx, SCMP_SYS(fspick));
+    allowSyscall(ctx, SCMP_SYS(fstat));
+    allowSyscall(ctx, SCMP_SYS(fstat64));
+    allowSyscall(ctx, SCMP_SYS(fstatat64));
+    allowSyscall(ctx, SCMP_SYS(fstatfs));
+    allowSyscall(ctx, SCMP_SYS(fstatfs64));
+    allowSyscall(ctx, SCMP_SYS(fsync));
+    allowSyscall(ctx, SCMP_SYS(ftime));
+    allowSyscall(ctx, SCMP_SYS(ftruncate));
+    allowSyscall(ctx, SCMP_SYS(ftruncate64));
+    allowSyscall(ctx, SCMP_SYS(futex));
+    // skip futex_requeue (requires glibc 2.39)
+    allowSyscall(ctx, SCMP_SYS(futex_time64));
+    // skip futex_wait (requires glibc 2.39)
+    allowSyscall(ctx, SCMP_SYS(futex_waitv));
+    // skip futex_wake (requires glibc 2.39)
+    allowSyscall(ctx, SCMP_SYS(futimesat));
+    allowSyscall(ctx, SCMP_SYS(getcpu));
+    allowSyscall(ctx, SCMP_SYS(getcwd));
+    allowSyscall(ctx, SCMP_SYS(getdents));
+    allowSyscall(ctx, SCMP_SYS(getdents64));
+    allowSyscall(ctx, SCMP_SYS(getegid));
+    allowSyscall(ctx, SCMP_SYS(getegid32));
+    allowSyscall(ctx, SCMP_SYS(geteuid));
+    allowSyscall(ctx, SCMP_SYS(geteuid32));
+    allowSyscall(ctx, SCMP_SYS(getgid));
+    allowSyscall(ctx, SCMP_SYS(getgid32));
+    allowSyscall(ctx, SCMP_SYS(getgroups));
+    allowSyscall(ctx, SCMP_SYS(getgroups32));
+    allowSyscall(ctx, SCMP_SYS(getitimer));
+    allowSyscall(ctx, SCMP_SYS(get_kernel_syms));
+    allowSyscall(ctx, SCMP_SYS(get_mempolicy));
+    allowSyscall(ctx, SCMP_SYS(getpeername));
+    allowSyscall(ctx, SCMP_SYS(getpgid));
+    allowSyscall(ctx, SCMP_SYS(getpgrp));
+    allowSyscall(ctx, SCMP_SYS(getpid));
+    allowSyscall(ctx, SCMP_SYS(getpmsg));
+    allowSyscall(ctx, SCMP_SYS(getppid));
+    allowSyscall(ctx, SCMP_SYS(getpriority));
+    allowSyscall(ctx, SCMP_SYS(getrandom));
+    allowSyscall(ctx, SCMP_SYS(getresgid));
+    allowSyscall(ctx, SCMP_SYS(getresgid32));
+    allowSyscall(ctx, SCMP_SYS(getresuid));
+    allowSyscall(ctx, SCMP_SYS(getresuid32));
+    allowSyscall(ctx, SCMP_SYS(getrlimit));
+    allowSyscall(ctx, SCMP_SYS(get_robust_list));
+    allowSyscall(ctx, SCMP_SYS(getrusage));
+    allowSyscall(ctx, SCMP_SYS(getsid));
+    allowSyscall(ctx, SCMP_SYS(getsockname));
+    allowSyscall(ctx, SCMP_SYS(getsockopt));
+    allowSyscall(ctx, SCMP_SYS(get_thread_area));
+    allowSyscall(ctx, SCMP_SYS(gettid));
+    allowSyscall(ctx, SCMP_SYS(gettimeofday));
+    allowSyscall(ctx, SCMP_SYS(get_tls));
+    allowSyscall(ctx, SCMP_SYS(getuid));
+    allowSyscall(ctx, SCMP_SYS(getuid32));
+    allowSyscall(ctx, SCMP_SYS(getxattr));
+    allowSyscall(ctx, SCMP_SYS(gtty));
+    allowSyscall(ctx, SCMP_SYS(idle));
+    allowSyscall(ctx, SCMP_SYS(init_module));
+    allowSyscall(ctx, SCMP_SYS(inotify_add_watch));
+    allowSyscall(ctx, SCMP_SYS(inotify_init));
+    allowSyscall(ctx, SCMP_SYS(inotify_init1));
+    allowSyscall(ctx, SCMP_SYS(inotify_rm_watch));
+    allowSyscall(ctx, SCMP_SYS(io_cancel));
+    allowSyscall(ctx, SCMP_SYS(ioctl));
+    allowSyscall(ctx, SCMP_SYS(io_destroy));
+    allowSyscall(ctx, SCMP_SYS(io_getevents));
+    allowSyscall(ctx, SCMP_SYS(ioperm));
+    allowSyscall(ctx, SCMP_SYS(io_pgetevents));
+    allowSyscall(ctx, SCMP_SYS(io_pgetevents_time64));
+    allowSyscall(ctx, SCMP_SYS(iopl));
+    allowSyscall(ctx, SCMP_SYS(ioprio_get));
+    allowSyscall(ctx, SCMP_SYS(ioprio_set));
+    allowSyscall(ctx, SCMP_SYS(io_setup));
+    allowSyscall(ctx, SCMP_SYS(io_submit));
+    // skip io_uring_enter (may become dangerous)
+    // skip io_uring_register (may become dangerous)
+    // skip io_uring_setup (may become dangerous)
+    allowSyscall(ctx, SCMP_SYS(ipc));
+    allowSyscall(ctx, SCMP_SYS(kcmp));
+    allowSyscall(ctx, SCMP_SYS(kexec_file_load));
+    allowSyscall(ctx, SCMP_SYS(kexec_load));
+    allowSyscall(ctx, SCMP_SYS(keyctl));
+    allowSyscall(ctx, SCMP_SYS(kill));
+    allowSyscall(ctx, SCMP_SYS(landlock_add_rule));
+    allowSyscall(ctx, SCMP_SYS(landlock_create_ruleset));
+    allowSyscall(ctx, SCMP_SYS(landlock_restrict_self));
+    allowSyscall(ctx, SCMP_SYS(lchown));
+    allowSyscall(ctx, SCMP_SYS(lchown32));
+    allowSyscall(ctx, SCMP_SYS(lgetxattr));
+    allowSyscall(ctx, SCMP_SYS(link));
+    allowSyscall(ctx, SCMP_SYS(linkat));
+    allowSyscall(ctx, SCMP_SYS(listen));
+    allowSyscall(ctx, SCMP_SYS(listxattr));
+    allowSyscall(ctx, SCMP_SYS(llistxattr));
+    allowSyscall(ctx, SCMP_SYS(_llseek));
+    allowSyscall(ctx, SCMP_SYS(lock));
+    allowSyscall(ctx, SCMP_SYS(lookup_dcookie));
+    allowSyscall(ctx, SCMP_SYS(lremovexattr));
+    allowSyscall(ctx, SCMP_SYS(lseek));
+    // skip lsetxattr (dangerous)
+    allowSyscall(ctx, SCMP_SYS(lstat));
+    allowSyscall(ctx, SCMP_SYS(lstat64));
+    allowSyscall(ctx, SCMP_SYS(madvise));
+    // skip map_shadow_stack (requires glibc 2.39)
+    allowSyscall(ctx, SCMP_SYS(mbind));
+    allowSyscall(ctx, SCMP_SYS(membarrier));
+    allowSyscall(ctx, SCMP_SYS(memfd_create));
+    allowSyscall(ctx, SCMP_SYS(memfd_secret));
+    allowSyscall(ctx, SCMP_SYS(migrate_pages));
+    allowSyscall(ctx, SCMP_SYS(mincore));
+    allowSyscall(ctx, SCMP_SYS(mkdir));
+    allowSyscall(ctx, SCMP_SYS(mkdirat));
+    allowSyscall(ctx, SCMP_SYS(mknod));
+    allowSyscall(ctx, SCMP_SYS(mknodat));
+    allowSyscall(ctx, SCMP_SYS(mlock));
+    allowSyscall(ctx, SCMP_SYS(mlock2));
+    allowSyscall(ctx, SCMP_SYS(mlockall));
+    allowSyscall(ctx, SCMP_SYS(mmap));
+    allowSyscall(ctx, SCMP_SYS(mmap2));
+    allowSyscall(ctx, SCMP_SYS(modify_ldt));
+    allowSyscall(ctx, SCMP_SYS(mount));
+    allowSyscall(ctx, SCMP_SYS(mount_setattr));
+    allowSyscall(ctx, SCMP_SYS(move_mount));
+    allowSyscall(ctx, SCMP_SYS(move_pages));
+    allowSyscall(ctx, SCMP_SYS(mprotect));
+    allowSyscall(ctx, SCMP_SYS(mpx));
+    allowSyscall(ctx, SCMP_SYS(mq_getsetattr));
+    allowSyscall(ctx, SCMP_SYS(mq_notify));
+    allowSyscall(ctx, SCMP_SYS(mq_open));
+    allowSyscall(ctx, SCMP_SYS(mq_timedreceive));
+    allowSyscall(ctx, SCMP_SYS(mq_timedreceive_time64));
+    allowSyscall(ctx, SCMP_SYS(mq_timedsend));
+    allowSyscall(ctx, SCMP_SYS(mq_timedsend_time64));
+    allowSyscall(ctx, SCMP_SYS(mq_unlink));
+    allowSyscall(ctx, SCMP_SYS(mremap));
+    allowSyscall(ctx, SCMP_SYS(msgctl));
+    allowSyscall(ctx, SCMP_SYS(msgget));
+    allowSyscall(ctx, SCMP_SYS(msgrcv));
+    allowSyscall(ctx, SCMP_SYS(msgsnd));
+    allowSyscall(ctx, SCMP_SYS(msync));
+    allowSyscall(ctx, SCMP_SYS(multiplexer));
+    allowSyscall(ctx, SCMP_SYS(munlock));
+    allowSyscall(ctx, SCMP_SYS(munlockall));
+    allowSyscall(ctx, SCMP_SYS(munmap));
+    allowSyscall(ctx, SCMP_SYS(name_to_handle_at));
+    allowSyscall(ctx, SCMP_SYS(nanosleep));
+    allowSyscall(ctx, SCMP_SYS(newfstatat));
+    allowSyscall(ctx, SCMP_SYS(_newselect));
+    allowSyscall(ctx, SCMP_SYS(nfsservctl));
+    allowSyscall(ctx, SCMP_SYS(nice));
+    allowSyscall(ctx, SCMP_SYS(oldfstat));
+    allowSyscall(ctx, SCMP_SYS(oldlstat));
+    allowSyscall(ctx, SCMP_SYS(oldolduname));
+    allowSyscall(ctx, SCMP_SYS(oldstat));
+    allowSyscall(ctx, SCMP_SYS(olduname));
+    allowSyscall(ctx, SCMP_SYS(open));
+    allowSyscall(ctx, SCMP_SYS(openat));
+    allowSyscall(ctx, SCMP_SYS(openat2));
+    allowSyscall(ctx, SCMP_SYS(open_by_handle_at));
+    allowSyscall(ctx, SCMP_SYS(open_tree));
+    allowSyscall(ctx, SCMP_SYS(pause));
+    allowSyscall(ctx, SCMP_SYS(pciconfig_iobase));
+    allowSyscall(ctx, SCMP_SYS(pciconfig_read));
+    allowSyscall(ctx, SCMP_SYS(pciconfig_write));
+    allowSyscall(ctx, SCMP_SYS(perf_event_open));
+    allowSyscall(ctx, SCMP_SYS(personality));
+    allowSyscall(ctx, SCMP_SYS(pidfd_getfd));
+    allowSyscall(ctx, SCMP_SYS(pidfd_open));
+    allowSyscall(ctx, SCMP_SYS(pidfd_send_signal));
+    allowSyscall(ctx, SCMP_SYS(pipe));
+    allowSyscall(ctx, SCMP_SYS(pipe2));
+    allowSyscall(ctx, SCMP_SYS(pivot_root));
+    allowSyscall(ctx, SCMP_SYS(pkey_alloc));
+    allowSyscall(ctx, SCMP_SYS(pkey_free));
+    allowSyscall(ctx, SCMP_SYS(pkey_mprotect));
+    allowSyscall(ctx, SCMP_SYS(poll));
+    allowSyscall(ctx, SCMP_SYS(ppoll));
+    allowSyscall(ctx, SCMP_SYS(ppoll_time64));
+    allowSyscall(ctx, SCMP_SYS(prctl));
+    allowSyscall(ctx, SCMP_SYS(pread64));
+    allowSyscall(ctx, SCMP_SYS(preadv));
+    allowSyscall(ctx, SCMP_SYS(preadv2));
+    allowSyscall(ctx, SCMP_SYS(prlimit64));
+    allowSyscall(ctx, SCMP_SYS(process_madvise));
+    allowSyscall(ctx, SCMP_SYS(process_mrelease));
+    allowSyscall(ctx, SCMP_SYS(process_vm_readv));
+    allowSyscall(ctx, SCMP_SYS(process_vm_writev));
+    allowSyscall(ctx, SCMP_SYS(prof));
+    allowSyscall(ctx, SCMP_SYS(profil));
+    allowSyscall(ctx, SCMP_SYS(pselect6));
+    allowSyscall(ctx, SCMP_SYS(pselect6_time64));
+    allowSyscall(ctx, SCMP_SYS(ptrace));
+    allowSyscall(ctx, SCMP_SYS(putpmsg));
+    allowSyscall(ctx, SCMP_SYS(pwrite64));
+    allowSyscall(ctx, SCMP_SYS(pwritev));
+    allowSyscall(ctx, SCMP_SYS(pwritev2));
+    allowSyscall(ctx, SCMP_SYS(query_module));
+    allowSyscall(ctx, SCMP_SYS(quotactl));
+    allowSyscall(ctx, SCMP_SYS(quotactl_fd));
+    allowSyscall(ctx, SCMP_SYS(read));
+    allowSyscall(ctx, SCMP_SYS(readahead));
+    allowSyscall(ctx, SCMP_SYS(readdir));
+    allowSyscall(ctx, SCMP_SYS(readlink));
+    allowSyscall(ctx, SCMP_SYS(readlinkat));
+    allowSyscall(ctx, SCMP_SYS(readv));
+    allowSyscall(ctx, SCMP_SYS(reboot));
+    allowSyscall(ctx, SCMP_SYS(recv));
+    allowSyscall(ctx, SCMP_SYS(recvfrom));
+    allowSyscall(ctx, SCMP_SYS(recvmmsg));
+    allowSyscall(ctx, SCMP_SYS(recvmmsg_time64));
+    allowSyscall(ctx, SCMP_SYS(recvmsg));
+    allowSyscall(ctx, SCMP_SYS(remap_file_pages));
+    allowSyscall(ctx, SCMP_SYS(removexattr));
+    allowSyscall(ctx, SCMP_SYS(rename));
+    allowSyscall(ctx, SCMP_SYS(renameat));
+    allowSyscall(ctx, SCMP_SYS(renameat2));
+    allowSyscall(ctx, SCMP_SYS(request_key));
+    allowSyscall(ctx, SCMP_SYS(restart_syscall));
+    allowSyscall(ctx, SCMP_SYS(riscv_flush_icache));
+    allowSyscall(ctx, SCMP_SYS(rmdir));
+    allowSyscall(ctx, SCMP_SYS(rseq));
+    allowSyscall(ctx, SCMP_SYS(rtas));
+    allowSyscall(ctx, SCMP_SYS(rt_sigaction));
+    allowSyscall(ctx, SCMP_SYS(rt_sigpending));
+    allowSyscall(ctx, SCMP_SYS(rt_sigprocmask));
+    allowSyscall(ctx, SCMP_SYS(rt_sigqueueinfo));
+    allowSyscall(ctx, SCMP_SYS(rt_sigreturn));
+    allowSyscall(ctx, SCMP_SYS(rt_sigsuspend));
+    allowSyscall(ctx, SCMP_SYS(rt_sigtimedwait));
+    allowSyscall(ctx, SCMP_SYS(rt_sigtimedwait_time64));
+    allowSyscall(ctx, SCMP_SYS(rt_tgsigqueueinfo));
+    allowSyscall(ctx, SCMP_SYS(s390_guarded_storage));
+    allowSyscall(ctx, SCMP_SYS(s390_pci_mmio_read));
+    allowSyscall(ctx, SCMP_SYS(s390_pci_mmio_write));
+    allowSyscall(ctx, SCMP_SYS(s390_runtime_instr));
+    allowSyscall(ctx, SCMP_SYS(s390_sthyi));
+    allowSyscall(ctx, SCMP_SYS(sched_getaffinity));
+    allowSyscall(ctx, SCMP_SYS(sched_getattr));
+    allowSyscall(ctx, SCMP_SYS(sched_getparam));
+    allowSyscall(ctx, SCMP_SYS(sched_get_priority_max));
+    allowSyscall(ctx, SCMP_SYS(sched_get_priority_min));
+    allowSyscall(ctx, SCMP_SYS(sched_getscheduler));
+    allowSyscall(ctx, SCMP_SYS(sched_rr_get_interval));
+    allowSyscall(ctx, SCMP_SYS(sched_rr_get_interval_time64));
+    allowSyscall(ctx, SCMP_SYS(sched_setaffinity));
+    allowSyscall(ctx, SCMP_SYS(sched_setattr));
+    allowSyscall(ctx, SCMP_SYS(sched_setparam));
+    allowSyscall(ctx, SCMP_SYS(sched_setscheduler));
+    allowSyscall(ctx, SCMP_SYS(sched_yield));
+    allowSyscall(ctx, SCMP_SYS(seccomp));
+    allowSyscall(ctx, SCMP_SYS(security));
+    allowSyscall(ctx, SCMP_SYS(select));
+    allowSyscall(ctx, SCMP_SYS(semctl));
+    allowSyscall(ctx, SCMP_SYS(semget));
+    allowSyscall(ctx, SCMP_SYS(semop));
+    allowSyscall(ctx, SCMP_SYS(semtimedop));
+    allowSyscall(ctx, SCMP_SYS(semtimedop_time64));
+    allowSyscall(ctx, SCMP_SYS(send));
+    allowSyscall(ctx, SCMP_SYS(sendfile));
+    allowSyscall(ctx, SCMP_SYS(sendfile64));
+    allowSyscall(ctx, SCMP_SYS(sendmmsg));
+    allowSyscall(ctx, SCMP_SYS(sendmsg));
+    allowSyscall(ctx, SCMP_SYS(sendto));
+    allowSyscall(ctx, SCMP_SYS(setdomainname));
+    allowSyscall(ctx, SCMP_SYS(setfsgid));
+    allowSyscall(ctx, SCMP_SYS(setfsgid32));
+    allowSyscall(ctx, SCMP_SYS(setfsuid));
+    allowSyscall(ctx, SCMP_SYS(setfsuid32));
+    allowSyscall(ctx, SCMP_SYS(setgid));
+    allowSyscall(ctx, SCMP_SYS(setgid32));
+    allowSyscall(ctx, SCMP_SYS(setgroups));
+    allowSyscall(ctx, SCMP_SYS(setgroups32));
+    allowSyscall(ctx, SCMP_SYS(sethostname));
+    allowSyscall(ctx, SCMP_SYS(setitimer));
+    allowSyscall(ctx, SCMP_SYS(set_mempolicy));
+    allowSyscall(ctx, SCMP_SYS(set_mempolicy_home_node));
+    allowSyscall(ctx, SCMP_SYS(setns));
+    allowSyscall(ctx, SCMP_SYS(setpgid));
+    allowSyscall(ctx, SCMP_SYS(setpriority));
+    allowSyscall(ctx, SCMP_SYS(setregid));
+    allowSyscall(ctx, SCMP_SYS(setregid32));
+    allowSyscall(ctx, SCMP_SYS(setresgid));
+    allowSyscall(ctx, SCMP_SYS(setresgid32));
+    allowSyscall(ctx, SCMP_SYS(setresuid));
+    allowSyscall(ctx, SCMP_SYS(setresuid32));
+    allowSyscall(ctx, SCMP_SYS(setreuid));
+    allowSyscall(ctx, SCMP_SYS(setreuid32));
+    allowSyscall(ctx, SCMP_SYS(setrlimit));
+    allowSyscall(ctx, SCMP_SYS(set_robust_list));
+    allowSyscall(ctx, SCMP_SYS(setsid));
+    allowSyscall(ctx, SCMP_SYS(setsockopt));
+    allowSyscall(ctx, SCMP_SYS(set_thread_area));
+    allowSyscall(ctx, SCMP_SYS(set_tid_address));
+    allowSyscall(ctx, SCMP_SYS(settimeofday));
+    allowSyscall(ctx, SCMP_SYS(set_tls));
+    allowSyscall(ctx, SCMP_SYS(setuid));
+    allowSyscall(ctx, SCMP_SYS(setuid32));
+    // skip setxattr (dangerous)
+    allowSyscall(ctx, SCMP_SYS(sgetmask));
+    allowSyscall(ctx, SCMP_SYS(shmat));
+    allowSyscall(ctx, SCMP_SYS(shmctl));
+    allowSyscall(ctx, SCMP_SYS(shmdt));
+    allowSyscall(ctx, SCMP_SYS(shmget));
+    allowSyscall(ctx, SCMP_SYS(shutdown));
+    allowSyscall(ctx, SCMP_SYS(sigaction));
+    allowSyscall(ctx, SCMP_SYS(sigaltstack));
+    allowSyscall(ctx, SCMP_SYS(signal));
+    allowSyscall(ctx, SCMP_SYS(signalfd));
+    allowSyscall(ctx, SCMP_SYS(signalfd4));
+    allowSyscall(ctx, SCMP_SYS(sigpending));
+    allowSyscall(ctx, SCMP_SYS(sigprocmask));
+    allowSyscall(ctx, SCMP_SYS(sigreturn));
+    allowSyscall(ctx, SCMP_SYS(sigsuspend));
+    allowSyscall(ctx, SCMP_SYS(socket));
+    allowSyscall(ctx, SCMP_SYS(socketcall));
+    allowSyscall(ctx, SCMP_SYS(socketpair));
+    allowSyscall(ctx, SCMP_SYS(splice));
+    allowSyscall(ctx, SCMP_SYS(spu_create));
+    allowSyscall(ctx, SCMP_SYS(spu_run));
+    allowSyscall(ctx, SCMP_SYS(ssetmask));
+    allowSyscall(ctx, SCMP_SYS(stat));
+    allowSyscall(ctx, SCMP_SYS(stat64));
+    allowSyscall(ctx, SCMP_SYS(statfs));
+    allowSyscall(ctx, SCMP_SYS(statfs64));
+    allowSyscall(ctx, SCMP_SYS(statx));
+    allowSyscall(ctx, SCMP_SYS(stime));
+    allowSyscall(ctx, SCMP_SYS(stty));
+    allowSyscall(ctx, SCMP_SYS(subpage_prot));
+    allowSyscall(ctx, SCMP_SYS(swapcontext));
+    allowSyscall(ctx, SCMP_SYS(swapoff));
+    allowSyscall(ctx, SCMP_SYS(swapon));
+    allowSyscall(ctx, SCMP_SYS(switch_endian));
+    allowSyscall(ctx, SCMP_SYS(symlink));
+    allowSyscall(ctx, SCMP_SYS(symlinkat));
+    allowSyscall(ctx, SCMP_SYS(sync));
+    allowSyscall(ctx, SCMP_SYS(sync_file_range));
+    allowSyscall(ctx, SCMP_SYS(sync_file_range2));
+    allowSyscall(ctx, SCMP_SYS(syncfs));
+    allowSyscall(ctx, SCMP_SYS(syscall));
+    allowSyscall(ctx, SCMP_SYS(_sysctl));
+    allowSyscall(ctx, SCMP_SYS(sys_debug_setcontext));
+    allowSyscall(ctx, SCMP_SYS(sysfs));
+    allowSyscall(ctx, SCMP_SYS(sysinfo));
+    allowSyscall(ctx, SCMP_SYS(syslog));
+    allowSyscall(ctx, SCMP_SYS(sysmips));
+    allowSyscall(ctx, SCMP_SYS(tee));
+    allowSyscall(ctx, SCMP_SYS(tgkill));
+    allowSyscall(ctx, SCMP_SYS(time));
+    allowSyscall(ctx, SCMP_SYS(timer_create));
+    allowSyscall(ctx, SCMP_SYS(timer_delete));
+    allowSyscall(ctx, SCMP_SYS(timerfd));
+    allowSyscall(ctx, SCMP_SYS(timerfd_create));
+    allowSyscall(ctx, SCMP_SYS(timerfd_gettime));
+    allowSyscall(ctx, SCMP_SYS(timerfd_gettime64));
+    allowSyscall(ctx, SCMP_SYS(timerfd_settime));
+    allowSyscall(ctx, SCMP_SYS(timerfd_settime64));
+    allowSyscall(ctx, SCMP_SYS(timer_getoverrun));
+    allowSyscall(ctx, SCMP_SYS(timer_gettime));
+    allowSyscall(ctx, SCMP_SYS(timer_gettime64));
+    allowSyscall(ctx, SCMP_SYS(timer_settime));
+    allowSyscall(ctx, SCMP_SYS(timer_settime64));
+    allowSyscall(ctx, SCMP_SYS(times));
+    allowSyscall(ctx, SCMP_SYS(tkill));
+    allowSyscall(ctx, SCMP_SYS(truncate));
+    allowSyscall(ctx, SCMP_SYS(truncate64));
+    allowSyscall(ctx, SCMP_SYS(tuxcall));
+    allowSyscall(ctx, SCMP_SYS(ugetrlimit));
+    allowSyscall(ctx, SCMP_SYS(ulimit));
+    allowSyscall(ctx, SCMP_SYS(umask));
+    allowSyscall(ctx, SCMP_SYS(umount));
+    allowSyscall(ctx, SCMP_SYS(umount2));
+    allowSyscall(ctx, SCMP_SYS(uname));
+    allowSyscall(ctx, SCMP_SYS(unlink));
+    allowSyscall(ctx, SCMP_SYS(unlinkat));
+    allowSyscall(ctx, SCMP_SYS(unshare));
+    allowSyscall(ctx, SCMP_SYS(uselib));
+    allowSyscall(ctx, SCMP_SYS(userfaultfd));
+    allowSyscall(ctx, SCMP_SYS(usr26));
+    allowSyscall(ctx, SCMP_SYS(usr32));
+    allowSyscall(ctx, SCMP_SYS(ustat));
+    allowSyscall(ctx, SCMP_SYS(utime));
+    allowSyscall(ctx, SCMP_SYS(utimensat));
+    allowSyscall(ctx, SCMP_SYS(utimensat_time64));
+    allowSyscall(ctx, SCMP_SYS(utimes));
+    allowSyscall(ctx, SCMP_SYS(vfork));
+    allowSyscall(ctx, SCMP_SYS(vhangup));
+    allowSyscall(ctx, SCMP_SYS(vm86));
+    allowSyscall(ctx, SCMP_SYS(vm86old));
+    allowSyscall(ctx, SCMP_SYS(vmsplice));
+    allowSyscall(ctx, SCMP_SYS(vserver));
+    allowSyscall(ctx, SCMP_SYS(wait4));
+    allowSyscall(ctx, SCMP_SYS(waitid));
+    allowSyscall(ctx, SCMP_SYS(waitpid));
+    allowSyscall(ctx, SCMP_SYS(write));
+    allowSyscall(ctx, SCMP_SYS(writev));
+    // END extract-syscalls
 
-        if (seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(fchmod), 1,
+    // chmod family: prevent adding setuid/setgid bits to existing files.
-                SCMP_A1(SCMP_CMP_MASKED_EQ, (scmp_datum_t) perm, (scmp_datum_t) perm)) != 0)
+    // The Nix store does not support setuid/setgid, and even their temporary creation can weaken the security of the sandbox.
-            throw SysError("unable to add seccomp rule");
+    if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(chmod), 1, SCMP_A1(SCMP_CMP_MASKED_EQ, S_ISUID | S_ISGID, 0)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(chmod), 1, SCMP_A1(SCMP_CMP_MASKED_EQ, S_ISUID, S_ISUID)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(chmod), 1, SCMP_A1(SCMP_CMP_MASKED_EQ, S_ISGID, S_ISGID)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(fchmod), 1, SCMP_A1(SCMP_CMP_MASKED_EQ, S_ISUID | S_ISGID, 0)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(fchmod), 1, SCMP_A1(SCMP_CMP_MASKED_EQ, S_ISUID, S_ISUID)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(fchmod), 1, SCMP_A1(SCMP_CMP_MASKED_EQ, S_ISGID, S_ISGID)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(fchmodat), 1, SCMP_A2(SCMP_CMP_MASKED_EQ, S_ISUID | S_ISGID, 0)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(fchmodat), 1, SCMP_A2(SCMP_CMP_MASKED_EQ, S_ISUID, S_ISUID)) != 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(fchmodat), 1, SCMP_A2(SCMP_CMP_MASKED_EQ, S_ISGID, S_ISGID)) != 0)
+        throw SysError("unable to add seccomp rule");
 
-        if (seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), SCMP_SYS(fchmodat), 1,
+    // setxattr family: prevent creation of extended attributes or ACLs.
-                SCMP_A2(SCMP_CMP_MASKED_EQ, (scmp_datum_t) perm, (scmp_datum_t) perm)) != 0)
+    // Not all filesystems support them, and they're incompatible with the NAR format.
-            throw SysError("unable to add seccomp rule");
-
-        if (seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EPERM), NIX_SYSCALL_FCHMODAT2, 1,
-                SCMP_A2(SCMP_CMP_MASKED_EQ, (scmp_datum_t) perm, (scmp_datum_t) perm)) != 0)
-            throw SysError("unable to add seccomp rule");
-    }
-
-    /* Prevent builders from creating EAs or ACLs. Not all filesystems
-       support these, and they're not allowed in the Nix store because
-       they're not representable in the NAR serialisation. */
     if (seccomp_rule_add(ctx, SCMP_ACT_ERRNO(ENOTSUP), SCMP_SYS(setxattr), 0) != 0 ||
         seccomp_rule_add(ctx, SCMP_ACT_ERRNO(ENOTSUP), SCMP_SYS(lsetxattr), 0) != 0 ||
         seccomp_rule_add(ctx, SCMP_ACT_ERRNO(ENOTSUP), SCMP_SYS(fsetxattr), 0) != 0)
@@ -1460,11 +1959,7 @@ void LocalDerivationGoal::runChild()
 
         commonChildInit();
 
-        try {
+        setupSeccomp();
-            setupSeccomp();
-        } catch (...) {
-            if (buildUser) throw;
-        }
 
         bool setUser = true;
 

src/libstore/linux/fchmodat2-compat.hh

@@ -1,35 +0,0 @@
-/*
- * Determine the syscall number for `fchmodat2`.
- *
- * On most platforms this is 452. Exceptions can be found on
- * a glibc git checkout via `rg --pcre2 'define __NR_fchmodat2 (?!452)'`.
- *
- * The problem is that glibc 2.39 and libseccomp 2.5.5 are needed to
- * get the syscall number. However, a Lix built against nixpkgs 23.11
- * (glibc 2.38) should still have the issue fixed without depending
- * on the build environment.
- *
- * To achieve that, the macros below try to determine the platform and
- * set the syscall number which is platform-specific, but
- * in most cases 452.
- *
- * TODO: remove this when 23.11 is EOL and the entire (supported) ecosystem
- * is on glibc 2.39.
- */
-
-#pragma once
-///@file
-
-#if defined(__alpha__)
-#  define NIX_SYSCALL_FCHMODAT2 562
-#elif defined(__x86_64__) && SIZE_MAX == 0xFFFFFFFF // x32
-#  define NIX_SYSCALL_FCHMODAT2 1073742276
-#elif defined(__mips__) && defined(__mips64) && defined(_ABIN64) // mips64/n64
-#  define NIX_SYSCALL_FCHMODAT2 5452
-#elif defined(__mips__) && defined(__mips64) && defined(_ABIN32) // mips64/n32
-#  define NIX_SYSCALL_FCHMODAT2 6452
-#elif defined(__mips__) && defined(_ABIO32) // mips32
-#  define NIX_SYSCALL_FCHMODAT2 4452
-#else
-#  define NIX_SYSCALL_FCHMODAT2 452
-#endif

src/nix-collect-garbage/nix-collect-garbage.cc

@@ -89,12 +89,21 @@ static int main_nix_collect_garbage(int argc, char * * argv)
 
         // Run the actual garbage collector.
         if (!dryRun) {
-            auto store = openStore();
-            auto & gcStore = require<GcStore>(*store);
             options.action = GCOptions::gcDeleteDead;
-            GCResults results;
+        } else {
-            PrintFreed freed(true, results);
+            options.action = GCOptions::gcReturnDead;
-            gcStore.collectGarbage(options, results);
+        }
+        auto store = openStore();
+        auto & gcStore = require<GcStore>(*store);
+        GCResults results;
+        PrintFreed freed(true, results);
+        gcStore.collectGarbage(options, results);
+
+        if (dryRun) {
+            // Only print results for dry run; when !dryRun, paths will be printed as they're deleted.
+            for (auto & i : results.paths) {
+                printInfo("%s", i);
+            }
         }
 
         return 0;

src/nix-env/nix-env.cc

@@ -236,10 +236,10 @@ static void checkSelectorUse(DrvNames & selectors)
 
 namespace {
 
-std::set<std::string> searchByPrefix(const DrvInfos & allElems, std::string_view prefix) {
+std::set<std::string> searchByPrefix(DrvInfos & allElems, std::string_view prefix) {
     constexpr std::size_t maxResults = 3;
     std::set<std::string> result;
-    for (const auto & drvInfo : allElems) {
+    for (auto & drvInfo : allElems) {
         const auto drvName = DrvName { drvInfo.queryName() };
         if (drvName.name.starts_with(prefix)) {
             result.emplace(drvName.name);
@@ -319,7 +319,7 @@ std::vector<Match> pickNewestOnly(EvalState & state, std::vector<Match> matches)
 
 } // end namespace
 
-static DrvInfos filterBySelector(EvalState & state, const DrvInfos & allElems,
+static DrvInfos filterBySelector(EvalState & state, DrvInfos & allElems,
     const Strings & args, bool newestOnly)
 {
     DrvNames selectors = drvNamesFromArgs(args);
@@ -331,7 +331,7 @@ static DrvInfos filterBySelector(EvalState & state, const DrvInfos & allElems,
 
     for (auto & selector : selectors) {
         std::vector<Match> matches;
-        for (const auto & [index, drvInfo] : enumerate(allElems)) {
+        for (auto && [index, drvInfo] : enumerate(allElems)) {
             const auto drvName = DrvName { drvInfo.queryName() };
             if (selector.matches(drvName)) {
                 ++selector.hits;
@@ -457,9 +457,12 @@ static void queryInstSources(EvalState & state,
            user environment.  These are then filtered as in the
            `srcNixExprDrvs' case. */
         case srcProfile: {
-            elems = filterBySelector(state,
+            auto installed = queryInstalled(state, instSource.profile);
-                queryInstalled(state, instSource.profile),
+            elems = filterBySelector(
-                args, newestOnly);
+                state,
+                installed,
+                args, newestOnly
+            );
             break;
         }
 
@@ -838,7 +841,7 @@ static bool cmpChars(char a, char b)
 }
 
 
-static bool cmpElemByName(const DrvInfo & a, const DrvInfo & b)
+static bool cmpElemByName(DrvInfo & a, DrvInfo & b)
 {
     auto a_name = a.queryName();
     auto b_name = b.queryName();
@@ -891,7 +894,7 @@ void printTable(Table & table)
 typedef enum { cvLess, cvEqual, cvGreater, cvUnavail } VersionDiff;
 
 static VersionDiff compareVersionAgainstSet(
-    const DrvInfo & elem, const DrvInfos & elems, std::string & version)
+    DrvInfo & elem, DrvInfos & elems, std::string & version)
 {
     DrvName name(elem.queryName());
 

tests/functional/meson.build

@@ -73,6 +73,7 @@ functional_tests_scripts = [
   'flakes/flake-in-submodule.sh',
   'gc.sh',
   'nix-collect-garbage-d.sh',
+  'nix-collect-garbage-dry-run.sh',
   'remote-store.sh',
   'legacy-ssh-store.sh',
   'lang.sh',

tests/functional/nix-collect-garbage-dry-run.sh

@@ -0,0 +1,40 @@
+source common.sh
+
+clearStore
+
+## Test `nix-collect-garbage --dry-run`
+
+
+testCollectGarbageDryRun () {
+    clearProfiles
+    # Install then uninstall a package
+    # This should leave packages ready to be garbage collected.
+    nix-env -f ./user-envs.nix -i foo-1.0
+    nix-env -f ./user-envs.nix -e foo-1.0
+
+
+    nix-env --delete-generations old
+    [[ $(nix-store --gc --print-dead | wc -l) -eq 7 ]]
+
+    nix-collect-garbage --dry-run
+    [[ $(nix-store --gc --print-dead | wc -l) -eq 7 ]]
+
+}
+
+testCollectGarbageDryRun
+
+# Run the same test, but forcing the profiles an arbitrary location.
+rm ~/.nix-profile
+ln -s $TEST_ROOT/blah ~/.nix-profile
+testCollectGarbageDryRun
+
+# Run the same test, but forcing the profiles at their legacy location under
+# /nix/var/nix.
+#
+# Note that we *don't* use the default profile; `nix-collect-garbage` will
+# need to check the legacy conditional unconditionally not just follow
+# `~/.nix-profile` to pass this test.
+#
+# Regression test for #8294
+rm ~/.nix-profile
+testCollectGarbageDryRun --profile "$NIX_STATE_DIR/profiles/per-user/me"

tests/nixos/setuid/fchmodat2-suid.c

@@ -1,21 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <sys/stat.h>
-#include <sys/syscall.h>
-#include <errno.h>
-#include <unistd.h>
-#include <assert.h>
-
-int main(void) {
-    char *name = getenv("out");
-    FILE *fd = fopen(name, "w");
-    fprintf(fd, "henlo :3");
-    fclose(fd);
-
-    // FIXME use something nicer here that's less
-    // platform-dependent as soon as we go to 24.05
-    // and the glibc is new enough to support fchmodat2
-    long rs = syscall(452, NULL, name, S_ISUID, 0);
-    assert(rs == -1);
-    assert(errno == EPERM);
-}

tests/nixos/setuid/setuid.nix

@@ -4,17 +4,6 @@
 
 let
   pkgs = config.nodes.machine.nixpkgs.pkgs;
-
-  fchmodat2-builder = pkgs.runCommandCC "fchmodat2-suid" {
-    passAsFile = [ "code" ];
-    code = builtins.readFile ./fchmodat2-suid.c;
-    # Doesn't work with -O0, shuts up the warning about that.
-    hardeningDisable = [ "fortify" ];
-  } ''
-    mkdir -p $out/bin/
-    $CC -x c "$codePath" -O0 -g -o $out/bin/fchmodat2-suid
-  '';
-
 in
 {
   name = "setuid";
@@ -27,26 +16,13 @@ in
       virtualisation.additionalPaths = [
         pkgs.stdenvNoCC
         pkgs.pkgsi686Linux.stdenvNoCC
-        fchmodat2-builder
       ];
-      # need at least 6.6 to test for fchmodat2
-      boot.kernelPackages = pkgs.linuxKernel.packages.linux_6_6;
-
     };
 
   testScript = { nodes }: ''
     # fmt: off
     start_all()
 
-    with subtest("fchmodat2 suid regression test"):
-      machine.succeed("""
-      nix-build -E '(with import <nixpkgs> {}; runCommand "fchmodat2-suid" {
-        BUILDER = builtins.storePath ${fchmodat2-builder};
-      } "
-        exec \\"$BUILDER\\"/bin/fchmodat2-suid
-      ")'
-      """)
-
     # Copying to /tmp should succeed.
     machine.succeed(r"""
     nix-build --no-sandbox -E '(with import <nixpkgs> {}; runCommand "foo" {} "