From: Christian Brauner <[email protected]> [ Upstream commit e5804b1024193c91ff2648403beda882cb4f8b6c ]
set_layers_via_fds.c calls open_tree() directly, but that wrapper is not provided by all C libraries, so the test fails to link with "undefined reference to open_tree" on toolchains without it. Use sys_open_tree() from ../wrappers.h, the syscall wrapper already used for move_mount(), fsmount() and friends here. This is also a prerequisite for building the new idmapped_mounts test added to this directory. Link: https://patch.msgid.link/[email protected] Reviewed-by: Amir Goldstein <[email protected]> Signed-off-by: Christian Brauner (Amutable) <[email protected]> Signed-off-by: Sasha Levin <[email protected]> --- LLM Generated explanations, may be completely bogus: I have enough to decide: this is a selftest link fix for code that already exists on 6.18.y, with a clean apply and no kernel runtime change. ## PHASE 1: COMMIT MESSAGE FORENSICS **Step 1.1: PARSE THE SUBJECT LINE** - Record: `[selftests/filesystems/overlayfs] [fix] [replace libc open_tree() with sys_open_tree() so the test links]` **Step 1.2: PARSE ALL COMMIT MESSAGE TAGS** - `Link: https://patch.msgid.link/20260615-work-idmapped- [email protected]` — lore/msgid for patch 8/10 of the idmapped overlayfs series - `Reviewed-by: Amir Goldstein <[email protected]>` — overlayfs maintainer - `Signed-off-by: Christian Brauner (Amutable) <[email protected]>` — VFS maintainer / author - Pipeline-only (ignored for selection): `[Upstream commit e5804b1024193c91ff2648403beda882cb4f8b6c]`, `Signed-off-by: Sasha Levin` - Absent (expected): `Fixes:`, `Cc: [email protected]`, `Reported- by:`, `Tested-by:`, syzbot **Step 1.3: ANALYZE THE COMMIT BODY TEXT** - Record: `[Bug] set_layers_via_fds.c calls libc open_tree(), which some C libraries do not provide. [Symptom] link failure: "undefined reference to open_tree". [Version info] none. [Root cause] the test already uses sys_fsopen/sys_fsmount/sys_move_mount from wrappers.h, but this one test used the libc name instead of sys_open_tree(). The message also says the change is a prerequisite for a new idmapped_mounts test.` **Step 1.4: DETECT HIDDEN BUG FIXES** - Record: `[Not a hidden fix] The subject and body state a link error. That is a real userspace build failure of an existing kselftest, not a disguised kernel runtime bug. The "prerequisite for idmapped_mounts" line is extra context for a later new test, not the bug being fixed here.` ## PHASE 2: DIFF ANALYSIS **Step 2.1: INVENTORY THE CHANGES** - Record: `[Files] tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c (+8/-8). [Function] TEST_F(set_layers_via_fds, set_layers_via_detached_mount_fds) only. [Scope] single-file surgical rename, 8 call sites.]` **Step 2.2: UNDERSTAND THE CODE FLOW CHANGE** - Record: `[Before] eight open_tree(dfd, path, OPEN_TREE_CLONE|OPEN_TREE_CLOEXEC) calls (libc symbol). [After] the same arguments via sys_open_tree(), the local syscall wrapper. [Path] userspace selftest setup for detached overlay layer mounts. No kernel path, no error-path logic change, no flag change.]` **Step 2.3: IDENTIFY THE BUG MECHANISM** - Record: `[Category] build / link error (portability). [Mechanism] unresolved libc symbol on C libraries without open_tree(); sys_open_tree() already exists in wrappers.h and issues syscall(__NR_open_tree, ...).]` **Step 2.4: ASSESS THE FIX QUALITY** - Record: `[Quality] obviously correct: same arguments, wrapper already used for the other new-mount syscalls in this file, OPEN_TREE_* already defined in wrappers.h. [Regression risk] very low. syscall() and libc open_tree() both return fd or -1/errno; the test already uses ASSERT_GE(fd, 0). No kernel ABI, locking, or API change.]` ## PHASE 3: GIT HISTORY INVESTIGATION **Step 3.1: BLAME THE CHANGED LINES** - Record: `[Introduced by] ccc829b15d48a ("selftests: add tests for using detached mount with overlayfs", 2025-02-12). [Present in this tree] yes (ancestor of v6.18.53). The test file itself dates to af9199145b197 (2024-10-14).]` **Step 3.2: FOLLOW THE FIXES: TAG** - Record: `[N/A] No Fixes: tag. The intro commit is ccc829b15d48a, confirmed via blame on v6.18.53.]` **Step 3.3: CHECK FILE HISTORY FOR RELATED CHANGES** - Record: `[Related] 0bd92b9fe538b moved wrappers.h to tools/testing/selftests/filesystems/wrappers.h (already in v6.18.53). Next commit 226c30d522a3b adds idmapped_mounts.c (new test, not in v6.18.53). [Standalone] this rename does not need the new test.]` **Step 3.4: CHECK THE AUTHOR'S OTHER COMMITS** - Record: `[Author] Christian Brauner, VFS maintainer, overlayfs contributor. [Related] authored the original detached-mount test and the later idmapped overlayfs series. Not a drive-by change.]` **Step 3.5: CHECK FOR DEPENDENT/PREREQUISITE COMMITS** - Record: `[Dependencies] sys_open_tree() and #include "../wrappers.h" already exist at v6.18.53. Parent file vs v6.18.53 file: 0-line diff. [Standalone] yes for this tree. The new idmapped_mounts test is not required.]` ## PHASE 4: MAILING LIST AND EXTERNAL RESEARCH **Step 4.1: FIND THE ORIGINAL PATCH DISCUSSION** - Record: `[Lore] https://patch.msgid.link/20260615-work-idmapped- [email protected] . [Series] v1 only (10 patches, "ovl: support idmapped overlay mounts"). This is 08/10. [Review] no NAK on this patch. Amir Goldstein Reviewed-by on the whole series ("the entire series looks good to me"). His risk comments were about FS_ALLOW_IDMAP on overlayfs itself, not this selftest rename. [Stable nomination] none in the thread.]` **Step 4.2: CHECK WHO REVIEWED THE PATCH** - Record: `[Recipients] Miklos Szeredi, Amir Goldstein, linux-unionfs, linux-fsdevel. [Reviewer] Amir Goldstein (overlayfs maintainer) gave series-wide Reviewed-by.]` **Step 4.3: SEARCH FOR THE BUG REPORT** - Record: `[No Reported-by / bugzilla / syzbot] The failure is described by the author as a link error. No independent user report found.]` **Step 4.4: CHECK FOR RELATED PATCHES AND SERIES** - Record: `[Series] 01–07/10 and 10/10 are overlayfs idmapped-mount kernel/docs features. 09/10 is the new idmapped_mounts selftest. Those are new functionality and are not in v6.18.53 (no FS_ALLOW_IDMAP on overlay in this tree; no idmapped_mounts.c). This 08/10 patch is independently a fix to an existing test.]` **Step 4.5: CHECK STABLE MAILING LIST HISTORY** - Record: `[UNVERIFIED for lore.kernel.org/stable] WebFetch returned 403. b4 dig of the original thread found no stable@ discussion.]` ## PHASE 5: CODE SEMANTIC ANALYSIS **Step 5.1: IDENTIFY KEY FUNCTIONS** - Record: `[Modified] TEST_F(set_layers_via_fds, set_layers_via_detached_mount_fds) in userspace kselftest. [Added] none. semcode MCP not available; identified from @@ hunks.]` **Step 5.2: TRACE CALLERS** - Record: `[Callers] kselftest harness TEST_HARNESS_MAIN / fixture runner when executing the set_layers_via_fds binary. Not in-kernel. Not interrupt or syscall-kernel context.]` **Step 5.3: TRACE CALLEES** - Record: `[sys_open_tree] syscall(__NR_open_tree, dfd, filename, flags) in wrappers.h. Same as other sys_* helpers already used in this file (sys_fsopen, sys_fsmount, sys_move_mount, sys_fsconfig).]` **Step 5.4: FOLLOW THE CALL CHAIN** - Record: `[Reachability] only when a developer/CI builds and runs overlayfs kselftests (typically root, CONFIG_OVERLAY_FS). Not a production kernel path. The link failure happens at test build time, before the test runs.]` **Step 5.5: SEARCH FOR SIMILAR PATTERNS** - Record: `[This tree at v6.18.53] the only remaining libc open_tree() uses in tools/testing/selftests besides this file are landlock/fs_test.c, which defines its own local open_tree() wrapper. Overlayfs tests were already supposed to use sys_open_tree(). Earlier overlayfs selftest build fixes already in this tree: e8b8c5264d4eb (ppc64 format), 0f05ee447949c (mount.h / sys_mount).]` ## PHASE 6: CROSS-REFERENCING AGAINST THE LOCAL TREE **Step 6.1: DOES THE BUGGY CODE EXIST IN THIS TREE?** - Record: `[Tree] Makefile VERSION=6 PATCHLEVEL=18 SUBLEVEL=53; git describe HEAD = v6.18.53-5049-g0f0bdc2b0dffb; branch autosel. [v6.18.53] set_layers_via_fds.c exists and has all 8 open_tree() calls. wrappers.h already has sys_open_tree() and OPEN_TREE_CLONE/CLOEXEC. Include of ../wrappers.h is already present. [Bug introduced] ccc829b15d48a, in this series.]` **Step 6.2: CHECK FOR BACKPORT COMPLICATIONS** - Record: `[Difficulty] clean apply. `git diff v6.18.53 088318c9d7a6e^ -- set_layers_via_fds.c` is empty. Applying the commit patch onto a copy of the v6.18.53 file: APPLY_CLEAN.]` **Step 6.3: CHECK IF RELATED FIXES ARE ALREADY HERE** - Record: `[No equivalent fix in v6.18.53] still uses open_tree(). Candidate 088318c9d7a6e is on autosel only, not in v6.18.53 (`git merge-base --is-ancestor 088318c9d7a6e v6.18.53` → not an ancestor).]` ## PHASE 7: SUBSYSTEM AND MAINTAINER CONTEXT **Step 7.1: IDENTIFY THE SUBSYSTEM AND ITS CRITICALITY** - Record: `[Subsystem] kselftests for overlayfs (tools/testing/selftests/filesystems/overlayfs/). [Criticality] PERIPHERAL for production runtime; IMPORTANT for overlayfs regression testing of this LTS tree. Overlayfs itself is a widely used fs (containers), but this commit does not touch fs/overlayfs/.]` **Step 7.2: ASSESS SUBSYSTEM ACTIVITY** - Record: `[Activity] overlayfs selftests were added/extended through 6.15 (fd mounts, detached mounts, override_creds). Relatively new tests, still the ones this 6.18.y tree ships. Kernel overlayfs is mature.]` ## PHASE 8: IMPACT AND RISK ASSESSMENT **Step 8.1: DETERMINE WHO IS AFFECTED** - Record: `[Affected] people/CI building overlayfs kselftests on a C library without libc open_tree() (author: "not provided by all C libraries"). Not universal kernel users. UNVERIFIED which specific libc versions still lack it in 2026; glibc added open_tree in 2.36, so musl/older glibc are the plausible cases.]` **Step 8.2: DETERMINE THE TRIGGER CONDITIONS** - Record: `[Trigger] building TEST_GEN_PROGS set_layers_via_fds (Makefile already lists it). Common for anyone running these selftests. Unprivileged users are not a security concern; this is a link failure, not a runtime exploit.]` **Step 8.3: DETERMINE THE FAILURE MODE SEVERITY** - Record: `[Failure] userspace link error; the kernel still builds and boots. Test binary is not produced, so that overlayfs selftest cannot be run. [Severity] LOW for production runtime, MEDIUM for stable- kernel testability. Matches the documented "build error" class, not oops/corruption/deadlock.]` **Step 8.4: CALCULATE RISK-BENEFIT RATIO** - Record: `[Benefit] medium for testers of this tree (existing test becomes buildable everywhere the rest of this directory already builds). [Risk] very low: 8 identifier substitutions, no kernel change, wrapper already in tree. [Ratio] favorable if build fixes for existing selftests are in scope; this one is.]` ## PHASE 9: FINAL SYNTHESIS **Step 9.1: COMPILE THE EVIDENCE** FOR backporting: - Real link failure of an existing test in this 6.18.y tree - Documentation/process/stable-kernel-rules.rst explicitly lists build errors as acceptable - 8-line surgical change, obviously correct - sys_open_tree() and wrappers.h already present; applies cleanly - Overlayfs maintainer Reviewed-by - Same class of fix as e8b8c5264d4eb / 0f05ee447949c already in this lineage - No kernel runtime regression surface AGAINST: - Not a kernel crash, security hole, or data-corruption fix - Lives in a 10-patch new-feature series; 09/10 and the kernel idmap work must not be pulled with it - No independent user/syzbot report - Only testers building kselftests benefit UNRESOLVED: - Exact C libraries that still lack open_tree() in 2026 (author claim, not independently enumerated) - No stable@ thread found (lore stable search 403) **Step 9.2: APPLY THE STABLE RULES CHECKLIST** 1. Obviously correct and tested? **PASS** — rename to an existing wrapper; reviewed by Amir Goldstein 2. Fixes a real bug that affects users? **PASS** — existing kselftest fails to link on some toolchains 3. Important issue (crash, security, corruption, deadlock)? **PASS** under the documented build-error clause, not under crash/security 4. Small and contained? **PASS** — 8 lines, one file, one test 5. No new features or APIs? **PASS** — no new test, no kernel API; the later idmapped_mounts test is a different commit 6. Can apply to the local tree? **PASS** — v6.18.53 file matches the parent; apply check clean **Step 9.3: CHECK FOR EXCEPTION CATEGORIES** - Record: `[Exception] BUILD FIX for an existing in-tree selftest. Not a device ID, quirk, DT, or docs-only change.]` **Step 9.4: MAKE YOUR DECISION** This should be backported to this 6.18.y tree. The bug is in the tree, the wrapper is in the tree, and the patch only makes an existing overlayfs selftest match the sys_* calling convention the rest of the file already uses. The idmapped-overlay feature series around it is irrelevant here: that feature and its new test are not in v6.18.53 and are not part of this diff. ### What the commit solves and why it matters here `set_layers_via_fds` is already built on 6.18.y (`TEST_GEN_PROGS += set_layers_via_fds`). One fixture still calls libc `open_tree()`, so the binary fails to link where that symbol is missing. Switching to `sys_open_tree()` fixes that with no kernel change. ### Stable rules It is small, obviously correct, and a build error. It does not add features or APIs. ### Risk vs benefit Risk is negligible (userspace identifier change). Benefit is that overlayfs kselftests for this LTS tree can actually be built on the same toolchains the wrappers were written for. ### Concerns Do not pull 09/10 (`idmapped_mounts.c`) or the overlay FS_ALLOW_IDMAP kernel patches with this; they implement a new feature this tree does not have. This commit does not require them. Verification: - [Phase 1] Parsed tags: Link to patch 8/10, Reviewed-by Amir Goldstein, SOB Christian Brauner; no Fixes/Cc:stable/Reported-by - [Phase 1] Body: link error "undefined reference to open_tree"; also mentions idmapped_mounts prerequisite - [Phase 2] Diff: 8 open_tree → sys_open_tree in set_layers_via_detached_mount_fds; +8/-8 one file - [Phase 2] wrappers.h: sys_open_tree() = syscall(__NR_open_tree, ...); OPEN_TREE_CLONE/CLOEXEC defined - [Phase 3] git describe HEAD: v6.18.53-5049-g0f0bdc2b0dffb; make kernelversion 6.18.53 - [Phase 3] git merge-base --is-ancestor 088318c9d7a6e v6.18.53: NOT in v6.18.53 - [Phase 3] git blame v6.18.53: open_tree() lines from ccc829b15d48a (2025-02-12), present in this tree - [Phase 3] git diff v6.18.53 vs commit parent for this file: 0 lines (identical) - [Phase 3] idmapped_mounts.c: not in v6.18.53; FS_ALLOW_IDMAP not set on overlay in this tree - [Phase 4] b4 dig -c e5804b1024193c91ff2648403beda882cb4f8b6c: matched 08/10 of ovl idmapped series - [Phase 4] b4 dig -a: v1 only - [Phase 4] b4 dig -w: Miklos Szeredi, Amir Goldstein, linux-unionfs, linux-fsdevel - [Phase 4] b4 dig -m: Amir series-wide Reviewed-by; risk comments were about kernel FS_ALLOW_IDMAP, not this patch; no NAK on 08/10 - [Phase 4] lore WebFetch 403; stable list search 403 — UNVERIFIED beyond the saved mbox - [Phase 5] semcode unavailable; callers are kselftest harness only - [Phase 5] git grep v6.18.53: only this file plus landlock (local wrapper) use open_tree() in selftests - [Phase 6] v6.18.53 has the 8 open_tree() calls and ../wrappers.h include; patch apply onto that file: APPLY_CLEAN - [Phase 7] tools/testing/selftests/filesystems/overlayfs; not fs/overlayfs - [Phase 8] Failure mode: userspace link error; severity LOW runtime / build-error class - UNVERIFIED: exact C libraries lacking open_tree() in 2026; no stable@ thread (403) **YES** .../filesystems/overlayfs/set_layers_via_fds.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c b/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c index dc0449fa628f8..d4a645ec88636 100644 --- a/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c +++ b/tools/testing/selftests/filesystems/overlayfs/set_layers_via_fds.c @@ -624,7 +624,7 @@ TEST_F(set_layers_via_fds, set_layers_via_detached_mount_fds) ASSERT_EQ(sys_move_mount(fd_tmpfs, "", -EBADF, "/set_layers_via_fds_tmpfs", MOVE_MOUNT_F_EMPTY_PATH), 0); - fd_tmp = open_tree(fd_tmpfs, "u", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + fd_tmp = sys_open_tree(fd_tmpfs, "u", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(fd_tmp, 0); layer_fds[0] = openat(fd_tmp, "upper", O_CLOEXEC | O_DIRECTORY | O_PATH); @@ -633,25 +633,25 @@ TEST_F(set_layers_via_fds, set_layers_via_detached_mount_fds) layer_fds[1] = openat(fd_tmp, "work", O_CLOEXEC | O_DIRECTORY | O_PATH); ASSERT_GE(layer_fds[1], 0); - layer_fds[2] = open_tree(fd_tmpfs, "l1", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + layer_fds[2] = sys_open_tree(fd_tmpfs, "l1", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(layer_fds[2], 0); - layer_fds[3] = open_tree(fd_tmpfs, "l2", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + layer_fds[3] = sys_open_tree(fd_tmpfs, "l2", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(layer_fds[3], 0); - layer_fds[4] = open_tree(fd_tmpfs, "l3", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + layer_fds[4] = sys_open_tree(fd_tmpfs, "l3", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(layer_fds[4], 0); - layer_fds[5] = open_tree(fd_tmpfs, "l4", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + layer_fds[5] = sys_open_tree(fd_tmpfs, "l4", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(layer_fds[5], 0); - layer_fds[6] = open_tree(fd_tmpfs, "d1", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + layer_fds[6] = sys_open_tree(fd_tmpfs, "d1", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(layer_fds[6], 0); - layer_fds[7] = open_tree(fd_tmpfs, "d2", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + layer_fds[7] = sys_open_tree(fd_tmpfs, "d2", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(layer_fds[7], 0); - layer_fds[8] = open_tree(fd_tmpfs, "d3", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); + layer_fds[8] = sys_open_tree(fd_tmpfs, "d3", OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); ASSERT_GE(layer_fds[8], 0); ASSERT_EQ(close(fd_tmpfs), 0); -- 2.53.0
