| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "errors" |
| 5 | "fmt" |
| 6 | "log" |
| 7 | "os" |
| 8 | "os/user" |
| 9 | "path/filepath" |
| 10 | "strconv" |
| 11 | "strings" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/registry" |
| 14 | ) |
| 15 | |
| 16 | // Root passes keep SQLite in a child running as the service user. |
| 17 | func openServiceRegistry(path string) (*registry.Registry, error) { |
| 18 | return openRegistryConverging(os.Geteuid(), serviceUser, path) |
| 19 | } |
| 20 | |
| 21 | // openRegistryConverging is openServiceRegistry with the two facts it reads |
| 22 | // from the process passed in; see chownRegistryFiles. |
| 23 | func openRegistryConverging(euid int, svcUser, path string) (*registry.Registry, error) { |
| 24 | if euid != 0 { |
| 25 | return registry.Open(path) |
| 26 | } |
| 27 | u, err := user.Lookup(svcUser) |
| 28 | if err != nil { |
| 29 | return nil, err |
| 30 | } |
| 31 | uid, err := strconv.Atoi(u.Uid) |
| 32 | if err != nil { |
| 33 | return nil, err |
| 34 | } |
| 35 | gid, err := strconv.Atoi(u.Gid) |
| 36 | if err != nil { |
| 37 | return nil, err |
| 38 | } |
| 39 | if uid <= 0 || gid < 0 { |
| 40 | return nil, fmt.Errorf("registry: invalid service credentials") |
| 41 | } |
| 42 | if err := chownRegistryFiles(euid, svcUser, path); err != nil { |
| 43 | return nil, err |
| 44 | } |
| 45 | return openRegistryChild(path, uid, gid) |
| 46 | } |
| 47 | |
| 48 | // chownRegistryToService repairs old root-owned files through checked descriptors. |
| 49 | func chownRegistryToService(path string) { |
| 50 | // Geteuid reports -1 on platforms without the concept (the CLI |
| 51 | // cross-compiles to windows), which takes the same early exit. |
| 52 | chownRegistryFiles(os.Geteuid(), serviceUser, path) |
| 53 | } |
| 54 | |
| 55 | // chownRegistryFiles is chownRegistryToService with the process facts — the |
| 56 | // effective uid and the service user's name — made arguments, so a test can |
| 57 | // drive the root arm, and both of its failure arms, from a test binary that |
| 58 | // is not root on a box that has no carlos user. |
| 59 | func chownRegistryFiles(euid int, svcUser, path string) error { |
| 60 | if euid != 0 { |
| 61 | return nil |
| 62 | } |
| 63 | u, err := user.Lookup(svcUser) |
| 64 | if err != nil { |
| 65 | log.Printf("adopt: registry ownership: no %s user: %v", svcUser, err) |
| 66 | return err |
| 67 | } |
| 68 | uid, err := strconv.Atoi(u.Uid) |
| 69 | if err != nil { |
| 70 | return err |
| 71 | } |
| 72 | gid, err := strconv.Atoi(u.Gid) |
| 73 | if err != nil { |
| 74 | return err |
| 75 | } |
| 76 | var refused error |
| 77 | for _, p := range registryFiles(path) { |
| 78 | changed, err := ownRegistryFile(p, uid, gid) |
| 79 | if errors.Is(err, os.ErrNotExist) { |
| 80 | continue |
| 81 | } |
| 82 | if err != nil { |
| 83 | log.Printf("adopt: ALERT: chown %s to %s: %v", p, svcUser, err) |
| 84 | refused = errors.Join(refused, err) |
| 85 | continue |
| 86 | } |
| 87 | if changed { |
| 88 | log.Printf("adopt: %s was not owned by %s — fixed; the edge runs as %s and could not have written it", p, svcUser, svcUser) |
| 89 | } |
| 90 | } |
| 91 | return refused |
| 92 | } |
| 93 | |
| 94 | // registryFiles is the database and the two sidecars SQLite keeps beside it |
| 95 | // in WAL mode. Ownership has to cover all three: a carlos-owned database with |
| 96 | // a root-owned -wal is just as unwritable as a root-owned database. |
| 97 | func registryFiles(path string) []string { |
| 98 | return []string{path, path + "-wal", path + "-shm"} |
| 99 | } |
| 100 | |
| 101 | // registryOpenHint turns SQLite's words into the box's words. |
| 102 | // |
| 103 | // "attempt to write a readonly database (8)" is what the edge printed on all |
| 104 | // three boxes, and it sent the operator looking at disk space and mount |
| 105 | // options before file ownership. The hint exists so the next person does not |
| 106 | // lose that time — which means it has to be RIGHT and not merely helpful: a |
| 107 | // hint that names ownership where ownership is not the cause sends them |
| 108 | // somewhere just as wrong, with more confidence behind it. |
| 109 | // |
| 110 | // So it says one of three things, and the process's own euid picks which: |
| 111 | // |
| 112 | // - As ROOT, ownership is never the answer. The root units drop no |
| 113 | // capabilities, so root holds CAP_DAC_OVERRIDE and writes the file |
| 114 | // whatever its owner and mode; a chown is the one action that cannot |
| 115 | // help. What remains possible for root is a read-only remount, an |
| 116 | // immutable attribute, a hot journal, or a MAC denial — and those are |
| 117 | // what it names. |
| 118 | // - Off root, holding a file this process does not own, ownership IS the |
| 119 | // answer and it gives the chown. |
| 120 | // - Off root with everything already owned correctly, ownership is not the |
| 121 | // answer either, so it points at the mount and the unit sandbox rather |
| 122 | // than repeating advice the operator has already taken. |
| 123 | // |
| 124 | // It stats all three of registryFiles rather than the database alone. A |
| 125 | // carlos-owned database with a root-owned -wal reports the identical error, |
| 126 | // and that is exactly the shape a hand chown of the database alone leaves |
| 127 | // behind — the second failure in #8, and why two of the three boxes needed |
| 128 | // the chown twice. Reporting the database alone printed "owned by carlos … |
| 129 | // must be owned by carlos" there: a self-contradiction, on the very case that |
| 130 | // motivated the hint. |
| 131 | // |
| 132 | // It returns "" for any other error, so a genuinely full disk still reads as |
| 133 | // itself. |
| 134 | func registryOpenHint(path string, err error) string { |
| 135 | return registryOpenHintFor(os.Geteuid(), path, err) |
| 136 | } |
| 137 | |
| 138 | // registryOpenHintFor is registryOpenHint with the effective uid passed in, |
| 139 | // so the root and non-root wordings are both reachable from one test binary. |
| 140 | func registryOpenHintFor(euid int, path string, err error) string { |
| 141 | if err == nil || !strings.Contains(strings.ToLower(err.Error()), "readonly database") { |
| 142 | return "" |
| 143 | } |
| 144 | var b strings.Builder |
| 145 | fmt.Fprintf(&b, "the registry is not writable by this process (uid %d).", euid) |
| 146 | |
| 147 | // present counts the files that exist; mine, the ones this process owns. |
| 148 | // They are equal only when there is nothing left for a chown to fix. |
| 149 | var present, mine int |
| 150 | for _, p := range registryFiles(path) { |
| 151 | fi, statErr := os.Stat(p) |
| 152 | if statErr != nil { |
| 153 | // A missing -wal/-shm is the normal state between connections, |
| 154 | // not a finding to report. |
| 155 | continue |
| 156 | } |
| 157 | present++ |
| 158 | uid, gid, ok := fileOwner(fi) |
| 159 | if !ok { |
| 160 | // Off unix there is no owner to read. Counted as present but |
| 161 | // never as mine, so the hint never claims ownership is fine on |
| 162 | // evidence it does not have. |
| 163 | fmt.Fprintf(&b, " %s exists (owner unreadable on this platform).", p) |
| 164 | continue |
| 165 | } |
| 166 | if uid == euid { |
| 167 | mine++ |
| 168 | } |
| 169 | fmt.Fprintf(&b, " %s is owned by %s (uid %d, gid %d).", p, ownerName(uid), uid, gid) |
| 170 | } |
| 171 | |
| 172 | dir := filepath.Dir(path) |
| 173 | switch { |
| 174 | case euid == 0: |
| 175 | fmt.Fprintf(&b, " Running as root, this is NOT an ownership problem: the root units drop"+ |
| 176 | " no capabilities, so root holds CAP_DAC_OVERRIDE and writes the file whatever its"+ |
| 177 | " owner and mode — changing the owner would change nothing. What can still produce"+ |
| 178 | " it for root: the filesystem remounted read-only after an I/O error"+ |
| 179 | " (`dmesg -T | tail`, `findmnt -no OPTIONS %s`), an immutable attribute on one of"+ |
| 180 | " the files (`lsattr %s*`), a hot journal or -wal left by a crash that SQLite"+ |
| 181 | " cannot replay, or a MAC denial (SELinux/AppArmor) on the write.", dir, path) |
| 182 | case present > 0 && present == mine: |
| 183 | fmt.Fprintf(&b, " Ownership is already correct — every file that exists is owned by this"+ |
| 184 | " uid — so setting it again will not help. Look instead at the mount (a read-only"+ |
| 185 | " remount after an I/O error, or a read-only %s: `findmnt -no OPTIONS %s`) and at"+ |
| 186 | " this unit's sandbox (ProtectSystem= or ReadOnlyPaths=, or %s missing from"+ |
| 187 | " ReadWritePaths=).", dir, dir, dir) |
| 188 | default: |
| 189 | fmt.Fprintf(&b, " It and its -wal/-shm sidecars must be owned by %s, the user"+ |
| 190 | " carlos-edge.service runs as. A root `carlos adopt` pass that reached the file"+ |
| 191 | " before the edge did is the usual cause. Fix: chown %s:%s %s*", |
| 192 | serviceUser, serviceUser, serviceUser, path) |
| 193 | } |
| 194 | return b.String() |
| 195 | } |
| 196 | |
| 197 | // ownerName resolves a uid to a username, falling back to the number itself. |
| 198 | // A box whose passwd lookup is broken still gets a usable hint. |
| 199 | func ownerName(uid int) string { |
| 200 | if u, err := user.LookupId(strconv.Itoa(uid)); err == nil && u.Username != "" { |
| 201 | return u.Username |
| 202 | } |
| 203 | return strconv.Itoa(uid) |
| 204 | } |
| 205 | |
amadan