[LibOS] Implement POSIX locks

LibOS/shim/include/shim_fs.h

@@ -115,6 +115,8 @@ struct shim_fs_ops {
  * pretends to have many files in a directory. */
 #define DENTRY_MAX_CHILDREN 1000000
 
+struct fs_lock_info;
+
 DEFINE_LIST(shim_dentry);
 DEFINE_LISTP(shim_dentry);
 struct shim_dentry {
@@ -152,6 +154,15 @@ struct shim_dentry {
     /* Filesystem-specific data. Protected by `lock`. */
     void* data;
 
+    /* File lock information, stored only in the main process. Protected by `lock`. See
+     * `shim_fs_lock.c`. */
+    struct fs_lock* fs_lock;
+
+    /* True if the file might have locks placed by current process. Used in processes other than
+     * main process, to prevent unnecessary IPC calls on handle close. Protected by `lock`. See
+     * `shim_fs_lock.c`. */
+    bool maybe_has_fs_locks;
+
     struct shim_lock lock;
     REFTYPE ref_count;
 };

LibOS/shim/include/shim_fs_lock.h

@@ -0,0 +1,128 @@
+/* SPDX-License-Identifier: LGPL-3.0-or-later */
+/* Copyright (C) 2021 Intel Corporation
+ *                    Paweł Marczewski <[email protected]>
+ */
+
+/*
+ * File locks. Currently only POSIX locks are implemented.
+ */
+
+#ifndef SHIM_FS_LOCK_H_
+#define SHIM_FS_LOCK_H_
+
+
+#include <stdbool.h>
+
+#include "list.h"
+#include "shim_types.h"
+
+#define FS_LOCK_EOF ((uint64_t)-1)
+
+struct shim_dentry;
+
+/* Initialize the file locking subsystem. */
+int init_fs_lock(void);
+
+/*
+ * POSIX locks (also known as advisory record locks). See `man fcntl` for details.
+ *
+ * The current implementation works over IPC and handles all requests in the main process. It has
+ * the following caveats:
+ *
+ * - Lock requests from other processes will always have the overhead of IPC round-trip, even if the
+ *   lock is uncontested.
+ * - The main process has to be able to look up the same file, so locking will not work for files in
+ *   local-process-only filesystems (tmpfs).
+ * - There is no deadlock detection (EDEADLK).
+ * - The lock requests cannot be interrupted (EINTR).
+ * - The locks work only on files that have a dentry (no pipes, sockets etc.)
+ */
+
+DEFINE_LISTP(posix_lock);
+DEFINE_LIST(posix_lock);
+struct posix_lock {
+    /* Lock type: F_RDLCK, F_WRLCK, F_UNLCK */
+    int type;
+
+    /* First byte of range */
+    uint64_t start;
+
+    /* Last byte of range (use FS_LOCK_EOF for a range until end of file) */
+    uint64_t end;
+
+    /* PID of process taking the lock */
+    IDTYPE pid;
+
+    /* List node, used internally */
+    LIST_TYPE(posix_lock) list;
+};
+
+/*!
+ * \brief Set or remove a lock on a file
+ *
+ * \param dent the dentry for a file
+ * \param pl parameters of new lock
+ * \param wait if true, will wait until a lock can be taken
+ *
+ * This is the equivalent of `fnctl(F_SETLK/F_SETLKW)`.
+ *
+ * If `pl->type` is `F_UNLCK`, the function will remove any locks held by the given PID for the
+ * given range. Removing a lock never waits.
+ *
+ * If `pl->type` is `F_RDLCK` or `F_WRLCK`, the function will create a new lock for the given PID
+ * and range, replacing the existing locks held by the given PID for that range. If there are
+ * conflicting locks, the function either waits (if `wait` is true), or fails with `-EAGAIN` (if
+ * `wait` is false).
+ */
+int posix_lock_set(struct shim_dentry* dent, struct posix_lock* pl, bool wait);
+
+/*!
+ * \brief Check for conflicting locks on a file
+ *
+ * \param dent the dentry for a file
+ * \param pl parameters of new lock (type cannot be `F_UNLCK`)
+ * \param[out] out_pl on success, set to `F_UNLCK` or details of a conflicting lock
+ *
+ * This is the equivalent of `fcntl(F_GETLK)`.
+ *
+ * The function checks if there are locks by other PIDs preventing the proposed lock from being
+ * placed. If the lock could be placed, `out_pl->type` is set to `F_UNLCK`. Otherwise, `out_pl`
+ * fields (`type`, `start, `end`, `pid`) are set to details of a conflicting lock.
+ */
+int posix_lock_get(struct shim_dentry* dent, struct posix_lock* pl, struct posix_lock* out_pl);
+
+/* Removes all locks for a given PID. Should be called before process exit. */
+int posix_lock_clear_pid(IDTYPE pid);
+
+/*!
+ * \brief Set or remove a lock on a file (IPC handler)
+ *
+ * \param path absolute path for a file
+ * \param pl parameters of new lock
+ * \param wait if true, will postpone the response until a lock can be taken
+ * \param vmid target process for IPC response
+ * \param seq sequence number for IPC response
+ *
+ * This is a version of `posix_lock_set` called from an IPC callback. This function is responsible
+ * for either sending an IPC response immediately, or scheduling one for later (if `wait` is true
+ * and the lock cannot be taken immediately).
+ *
+ * This function will only return a negative error code when failing to send a response. A failure
+ * to add a lock (-EAGAIN, -ENOMEM etc.) will be sent in the response instead.
+ */
+int posix_lock_set_from_ipc(const char* path, struct posix_lock* pl, bool wait, IDTYPE vmid,
+                            unsigned long seq);
+
+/*!
+ * \brief Check for conflicting locks on a file (IPC handler)
+ *
+ * \param path absolute path for a file
+ * \param pl parameters of new lock (type cannot be `F_UNLCK`)
+ * \param[out] out_pl on success, set to `F_UNLCK` or details of a conflicting lock
+ *
+ * This is a version of `posix_lock_get` called from an IPC callback. The caller is responsible to
+ * send the returned value and `out_pl` in an IPC response.
+ */
+int posix_lock_get_from_ipc(const char* path, struct posix_lock* pl, struct posix_lock* out_pl);
+
+#endif /* SHIM_FS_LOCK_H_ */

LibOS/shim/include/shim_ipc.h

@@ -32,6 +32,9 @@ enum {
     IPC_MSG_SYNC_CONFIRM_UPGRADE,
     IPC_MSG_SYNC_CONFIRM_DOWNGRADE,
     IPC_MSG_SYNC_CONFIRM_CLOSE,
+    IPC_MSG_POSIX_LOCK_SET,
+    IPC_MSG_POSIX_LOCK_GET,
+    IPC_MSG_POSIX_LOCK_CLEAR_PID,
     IPC_MSG_CODE_BOUND,
 };
 
@@ -253,4 +256,41 @@ int ipc_sync_confirm_upgrade_callback(IDTYPE src, void* data, unsigned long seq)
 int ipc_sync_confirm_downgrade_callback(IDTYPE src, void* data, unsigned long seq);
 int ipc_sync_confirm_close_callback(IDTYPE src, void* data, unsigned long seq);
 
+/*
+ * POSIX_LOCK_SET: `struct shim_ipc_posix_lock` -> `int`
+ * POSIX_LOCK_GET: `struct shim_ipc_posix_lock` -> `struct shim_ipc_posix_lock_resp`
+ * POSIX_LOCK_CLEAR_PID: `IDTYPE` -> `int`
+ */
+
+struct shim_ipc_posix_lock {
+    /* see `struct posix_lock` in `shim_fs_lock.h` */
+    int type;
+    uint64_t start;
+    uint64_t end;
+    IDTYPE pid;
+
+    bool wait;
+    char path[]; /* null-terminated */
+};
+
+struct shim_ipc_posix_lock_resp {
+    int result;
+
+    /* see `struct posix_lock` in `shim_fs_lock.h` */
+    int type;
+    uint64_t start;
+    uint64_t end;
+    IDTYPE pid;
+};
+
+struct posix_lock;
+
+int ipc_posix_lock_set(const char* path, struct posix_lock* pl, bool wait);
+int ipc_posix_lock_set_send_response(IDTYPE vmid, unsigned long seq, int result);
+int ipc_posix_lock_get(const char* path, struct posix_lock* pl, struct posix_lock* out_pl);
+int ipc_posix_lock_clear_pid(IDTYPE pid);
+int ipc_posix_lock_set_callback(IDTYPE src, void* data, unsigned long seq);
+int ipc_posix_lock_get_callback(IDTYPE src, void* data, unsigned long seq);
+int ipc_posix_lock_clear_pid_callback(IDTYPE src, void* data, unsigned long seq);
+
 #endif /* SHIM_IPC_H_ */

LibOS/shim/src/bookkeep/shim_handle.c

@@ -9,6 +9,7 @@
 #include "pal_error.h"
 #include "shim_checkpoint.h"
 #include "shim_fs.h"
+#include "shim_fs_lock.h"
 #include "shim_handle.h"
 #include "shim_internal.h"
 #include "shim_lock.h"
@@ -313,6 +314,22 @@ struct shim_handle* detach_fd_handle(FDTYPE fd, int* flags, struct shim_handle_m
         handle = __detach_fd_handle(handle_map->map[fd], flags, handle_map);
 
     unlock(&handle_map->lock);
+
+    if (handle && handle->dentry) {
+        /* Clear POSIX locks for a file. We are required to do that every time a FD is closed, even
+         * if the process holds other handles for that file, or duplicated FDs for the same
+         * handle. */
+        struct posix_lock pl = {
+            .type = F_UNLCK,
+            .start = 0,
+            .end = FS_LOCK_EOF,
+            .pid = g_process.pid,
+        };
+        int ret = posix_lock_set(handle->dentry, &pl, /*block=*/false);
+        if (ret < 0)
+            log_warning("error releasing locks: %d", ret);
+    }
+
     return handle;
 }
 

LibOS/shim/src/fs/shim_dcache.c

@@ -453,6 +453,9 @@ BEGIN_CP_FUNC(dentry) {
             new_dent->data = NULL;
         }
 
+        /* `fs_lock` is used only by process leader. */
+        new_dent->fs_lock = NULL;
+
         DO_CP_IN_MEMBER(qstr, new_dent, name);
 
         if (new_dent->mount)