| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "errors" |
| 5 | "fmt" |
| 6 | "os" |
| 7 | |
| 8 | "reasonix/internal/pathidentity" |
| 9 | "reasonix/internal/store" |
| 10 | ) |
| 11 | |
| 12 | // SessionRemovalGuard holds a session's save lock and lease lock for the |
| 13 | // duration of a destructive operation (trash, purge, permanent delete). While |
| 14 | // held, no other runtime can acquire the session lease and no saver can write |
| 15 | // the transcript, so artifacts can be moved or deleted without racing a live |
| 16 | // owner; the lock files themselves are then deleted atomically with the |
| 17 | // release (unlink-under-flock on Unix, delete-disposition on Windows), so a |
| 18 | // later acquirer can never lock an inode that survived the deletion. |
| 19 | // |
| 20 | // This closes the probe-then-delete window: a one-shot busy check followed by |
| 21 | // a plain RemoveAll lets another process acquire the lease between the two |
| 22 | // steps and then loses its lock file, breaking cross-process mutual exclusion. |
| 23 | type SessionRemovalGuard struct { |
| 24 | path string |
| 25 | key string |
| 26 | saveLock *sessionLockFile |
| 27 | leaseLock *sessionLockFile |
| 28 | legacyLeaseLock *sessionLockFile |
| 29 | legacyPath string |
| 30 | restoreOwner uint64 |
| 31 | } |
| 32 | |
| 33 | func tryTakeSessionLeaseLock(path string) (*sessionLockFile, error) { |
| 34 | lock, err := tryTakeSessionLeaseLockFile(store.SessionLeaseLock(path)) |
| 35 | if errors.Is(err, ErrSessionFileLockHeld) { |
| 36 | return nil, ErrSessionLeaseHeld |
| 37 | } |
| 38 | return lock, err |
| 39 | } |
| 40 | |
| 41 | // TryAcquireSessionRemovalGuard takes both locks without blocking. A live |
| 42 | // holder of either — including a lease held elsewhere in this process — |
| 43 | // surfaces as ErrSessionLeaseHeld so callers report the session as busy |
| 44 | // instead of deleting files out from under a running owner. |
| 45 | func TryAcquireSessionRemovalGuard(path string) (*SessionRemovalGuard, error) { |
| 46 | identity, err := resolveSessionPathIdentity(path) |
| 47 | if err != nil { |
| 48 | return nil, err |
| 49 | } |
| 50 | path, key := identity.PhysicalPath, identity.Key |
| 51 | legacyPath := pathidentity.Canonical(identity.AccessPath) |
| 52 | if sessionLeaseHeldLocally(key) { |
| 53 | info, _ := LoadSessionLeaseInfo(path) |
| 54 | return nil, &SessionLeaseError{Path: key, Info: info} |
| 55 | } |
| 56 | leaseLock, legacyLeaseLock, err := tryTakeCompatibleSessionLeaseLocks(path, legacyPath) |
| 57 | if err != nil { |
| 58 | if errors.Is(err, ErrSessionLeaseHeld) { |
| 59 | info, _ := LoadSessionLeaseInfo(path) |
| 60 | return nil, &SessionLeaseError{Path: key, Info: info} |
| 61 | } |
| 62 | return nil, err |
| 63 | } |
| 64 | saveLock, err := tryTakeSessionLockFile(store.SessionLockFile(path)) |
| 65 | if err != nil { |
| 66 | unlockSessionLeaseLocks(leaseLock, legacyLeaseLock) |
| 67 | if errors.Is(err, ErrSessionFileLockHeld) { |
| 68 | // A save is in flight; deleting mid-write would race it. |
| 69 | return nil, &SessionLeaseError{Path: key} |
| 70 | } |
| 71 | return nil, err |
| 72 | } |
| 73 | return &SessionRemovalGuard{ |
| 74 | path: path, key: key, legacyPath: legacyPath, |
| 75 | saveLock: saveLock, leaseLock: leaseLock, legacyLeaseLock: legacyLeaseLock, |
| 76 | }, nil |
| 77 | } |
| 78 | |
| 79 | // TryConvertToRemovalGuard transfers a live lease into destructive ownership |
| 80 | // without ever releasing its lease lock. The save lock is acquired first, so a |
| 81 | // failed conversion leaves the original lease fully active and reusable. |
| 82 | func (l *SessionLease) TryConvertToRemovalGuard() (*SessionRemovalGuard, error) { |
| 83 | if l == nil { |
| 84 | return nil, fmt.Errorf("nil session lease") |
| 85 | } |
| 86 | l.mu.Lock() |
| 87 | defer l.mu.Unlock() |
| 88 | if l.released || l.leaseLock == nil { |
| 89 | return nil, &SessionLeaseError{Path: l.path} |
| 90 | } |
| 91 | saveLock, err := tryTakeSessionLockFile(store.SessionLockFile(l.accessPath)) |
| 92 | if err != nil { |
| 93 | if errors.Is(err, ErrSessionFileLockHeld) { |
| 94 | return nil, &SessionLeaseError{Path: l.path} |
| 95 | } |
| 96 | return nil, err |
| 97 | } |
| 98 | // Keep the OS lock and original process-local reservation while revoking |
| 99 | // active runtime ownership. Both block competing acquisition and make |
| 100 | // rollback immune to an in-process lease race. |
| 101 | sessionLeaseActiveOwners.CompareAndDelete(l.path, l.ownerID) |
| 102 | leaseLock := l.leaseLock |
| 103 | legacyLeaseLock := l.legacyLeaseLock |
| 104 | l.leaseLock = nil |
| 105 | l.legacyLeaseLock = nil |
| 106 | l.released = true |
| 107 | return &SessionRemovalGuard{ |
| 108 | path: l.accessPath, |
| 109 | key: l.path, |
| 110 | saveLock: saveLock, |
| 111 | leaseLock: leaseLock, |
| 112 | legacyLeaseLock: legacyLeaseLock, |
| 113 | legacyPath: l.legacyAccessPath, |
| 114 | restoreOwner: l.ownerID, |
| 115 | }, nil |
| 116 | } |
| 117 | |
| 118 | // RestoreSessionLease aborts a converted removal guard before the destructive |
| 119 | // commit point. The original lease metadata remains untouched and both locks |
| 120 | // stay held until its owner generation is republished, so rollback cannot fail |
| 121 | // on a second disk write and has no writer gap. |
| 122 | func (g *SessionRemovalGuard) RestoreSessionLease() (*SessionLease, error) { |
| 123 | if g == nil || g.restoreOwner == 0 || g.leaseLock == nil || g.saveLock == nil { |
| 124 | return nil, fmt.Errorf("removal guard cannot restore a session lease") |
| 125 | } |
| 126 | ownerID := g.restoreOwner |
| 127 | current, ok := sessionLeaseOwners.Load(g.key) |
| 128 | if !ok || current != ownerID { |
| 129 | return nil, &SessionLeaseError{Path: g.path} |
| 130 | } |
| 131 | sessionLeaseActiveOwners.Delete(g.key) |
| 132 | sessionLeaseActiveOwners.Store(g.key, ownerID) |
| 133 | lease := &SessionLease{ |
| 134 | path: g.key, accessPath: g.path, legacyAccessPath: g.legacyPath, |
| 135 | ownerID: ownerID, leaseLock: g.leaseLock, legacyLeaseLock: g.legacyLeaseLock, |
| 136 | } |
| 137 | g.leaseLock = nil |
| 138 | g.legacyLeaseLock = nil |
| 139 | g.saveLock.Unlock() |
| 140 | g.saveLock = nil |
| 141 | g.restoreOwner = 0 |
| 142 | return lease, nil |
| 143 | } |
| 144 | |
| 145 | // Release ends the guard without deleting the lock files — the abort path |
| 146 | // when the destructive operation did not happen. Safe to call after |
| 147 | // RemoveSidecarsAndRelease (it becomes a no-op). |
| 148 | func (g *SessionRemovalGuard) Release() { |
| 149 | if g == nil { |
| 150 | return |
| 151 | } |
| 152 | if g.restoreOwner != 0 { |
| 153 | // A converted lease no longer has an active owner. Do not leave its |
| 154 | // identity sidecar naming a writer after an abort that chose not to |
| 155 | // restore the runtime lease. |
| 156 | sessionLeaseActiveOwners.CompareAndDelete(g.key, g.restoreOwner) |
| 157 | sessionLeaseOwners.CompareAndDelete(g.key, g.restoreOwner) |
| 158 | _ = os.Remove(store.SessionLeaseInfo(g.path)) |
| 159 | g.restoreOwner = 0 |
| 160 | } |
| 161 | if g.saveLock != nil { |
| 162 | g.saveLock.Unlock() |
| 163 | g.saveLock = nil |
| 164 | } |
| 165 | if g.leaseLock != nil { |
| 166 | g.leaseLock.Unlock() |
| 167 | g.leaseLock = nil |
| 168 | } |
| 169 | if g.legacyLeaseLock != nil { |
| 170 | g.legacyLeaseLock.Unlock() |
| 171 | g.legacyLeaseLock = nil |
| 172 | } |
| 173 | } |
| 174 | |
| 175 | // RemoveSidecarsAndRelease deletes the lease info and both lock files |
| 176 | // atomically with the release, then ends the guard. The lease info goes first, |
| 177 | // while the lease lock is still held, so no probe can adopt it mid-removal. |
| 178 | func (g *SessionRemovalGuard) RemoveSidecarsAndRelease() error { |
| 179 | if g == nil { |
| 180 | return nil |
| 181 | } |
| 182 | var errs []error |
| 183 | if g.restoreOwner != 0 { |
| 184 | sessionLeaseActiveOwners.CompareAndDelete(g.key, g.restoreOwner) |
| 185 | sessionLeaseOwners.CompareAndDelete(g.key, g.restoreOwner) |
| 186 | } |
| 187 | if err := os.Remove(store.SessionLeaseInfo(g.path)); err != nil && !os.IsNotExist(err) { |
| 188 | errs = append(errs, err) |
| 189 | } |
| 190 | if g.saveLock != nil { |
| 191 | if err := g.saveLock.RemoveAndUnlock(); err != nil { |
| 192 | errs = append(errs, err) |
| 193 | } |
| 194 | g.saveLock = nil |
| 195 | } |
| 196 | if g.leaseLock != nil { |
| 197 | if err := g.leaseLock.RemoveAndUnlock(); err != nil { |
| 198 | errs = append(errs, err) |
| 199 | } |
| 200 | g.leaseLock = nil |
| 201 | } |
| 202 | if g.legacyLeaseLock != nil { |
| 203 | if err := g.legacyLeaseLock.RemoveAndUnlock(); err != nil { |
| 204 | errs = append(errs, err) |
| 205 | } |
| 206 | g.legacyLeaseLock = nil |
| 207 | } |
| 208 | g.restoreOwner = 0 |
| 209 | return errors.Join(errs...) |
| 210 | } |
| 211 |