swap: Building blocks for swap support

Includes swap parsing, mkswap, swapon(1) and swapon(2).
Does _not_ include actual swapping yet.

Signed-off-by: Pedro Falcato <[email protected]>

kernel/arch/x86_64/syscall_table.json

@@ -2797,6 +2797,35 @@
         ],
         "return_type": "int",
         "abi": "c"
+    },
+    {
+        "name": "swapon",
+        "nr": 157,
+        "nr_args": 2,
+        "args": [
+            [
+                "const char *",
+                "path"
+            ],
+            [
+                "int",
+                "flags"
+            ]
+        ],
+        "return_type": "int",
+        "abi": "c"
+    },
+    {
+        "name": "swapoff",
+        "nr": 158,
+        "nr_args": 1,
+        "args": [
+            [
+                "const char *",
+                "path"
+            ]
+        ],
+        "return_type": "int",
+        "abi": "c"
     }
-
 ]

kernel/include/onyx/block.h

@@ -199,6 +199,7 @@ void block_queue_pending_io_queue(io_queue *queue);
  */
 void bio_queue_pending_req(struct request *req);
 
+__BEGIN_CDECLS
 /**
  * @brief Set the block device's block size
  *
@@ -208,7 +209,6 @@ void bio_queue_pending_req(struct request *req);
  */
 int block_set_bsize(struct blockdev *bdev, unsigned int block_size);
 
-__BEGIN_CDECLS
 int bdev_do_open(struct blockdev *bdev, bool exclusive);
 void bdev_release(struct blockdev *bdev);
 __END_CDECLS

kernel/include/onyx/buffer.h

@@ -158,6 +158,8 @@ void page_remove_block_buf(struct page *page, size_t offset, size_t end);
 
 void buffer_free_page(struct vm_object *vmo, struct page *page);
 
+struct block_buf *bdev_read_block(struct blockdev *bdev, unsigned long block);
+
 __END_CDECLS
 
 #ifdef __cplusplus

kernel/include/onyx/namei.h

@@ -15,14 +15,6 @@
 
 #include <uapi/fcntl.h>
 
-#include <onyx/string_view.hpp>
-
-enum class fs_token_type : uint8_t
-{
-    REGULAR_TOKEN = 0,
-    LAST_NAME_IN_PATH
-};
-
 #define LOOKUP_NOFOLLOW                (1 << 0)
 #define LOOKUP_FAIL_IF_LINK            (1 << 1)
 #define LOOKUP_MUST_BE_DIR             (1 << 2)
@@ -31,83 +23,12 @@ enum class fs_token_type : uint8_t
 #define LOOKUP_DONT_DO_LAST_NAME       (1 << 5)
 #define LOOKUP_INTERNAL_SAW_LAST_NAME  (1U << 31)
 
-/**
- * @brief Represents a path during a lookup
- *
- */
-struct lookup_path
-{
-    std::string_view view;
-    void *ownbuf{nullptr};
-    fs_token_type token_type{fs_token_type::REGULAR_TOKEN};
-    size_t pos{0};
-
-    lookup_path() = default;
-
-    lookup_path(std::string_view view) : view{view}
-    {
-    }
-
-    constexpr bool trailing_slash() const
-    {
-        return view[view.length() - 1] == '/';
-    }
-};
-
-struct nameidata
-{
-    /* Data needed to resolve filesystem names:
-     * view - Contains the pathname;
-     * pos - Contains the offset in the parsing of the pathname;
-     * root - Contains the lookup's filesystem root;
-     * cur - Contains the current relative location and
-     * starts at whatever was passed as the relative dir (controlled with
-     * chdir or *at, or purely through kernel-side use).
-     */
-    /* Note: root and location always hold a reference to the underlying object */
-    struct path root;
-    struct path cur;
-    /* Keeps the parent of cur, *if* we walked once */
-    struct path parent;
-
-    static constexpr const size_t max_loops = SYMLOOP_MAX;
-    /* Number of symbolic links found while looking up -
-     * if it reaches max_symlinks, the lookup fails with -ELOOP.
-     */
-    int nloops{0};
-    int pdepth{0};
-    struct lookup_path paths[SYMLOOP_MAX];
-
-    unsigned int lookup_flags{};
-    int dirfd{AT_FDCWD};
-
-    nameidata(std::string_view view)
-    {
-        paths[0] = lookup_path{view};
-        path_init(&root);
-        path_init(&cur);
-        path_init(&parent);
-    }
-
-    ~nameidata();
-
-    void setcur(struct path newcur)
-    {
-        DCHECK(!path_is_null(&newcur));
-        path_put(&parent);
-        parent = cur;
-        cur = newcur;
-    }
-
-    path getcur()
-    {
-        DCHECK(!path_is_null(&cur));
-        auto ret = cur;
-        path_init(&cur);
-        return ret;
-    }
-};
+__BEGIN_CDECLS
+struct file *c_vfs_open(int dirfd, const char *name, unsigned int open_flags, mode_t mode);
+__END_CDECLS
 
+#ifdef __cplusplus
 expected<file *, int> vfs_open(int dirfd, const char *name, unsigned int open_flags, mode_t mode);
+#endif
 
 #endif

kernel/include/uapi/swap.h

@@ -0,0 +1,37 @@
+/*
+ * Copyright (c) 2024 Pedro Falcato
+ * This file is part of Onyx, and is released under the terms of the MIT License
+ * check LICENSE at the root directory for more information
+ *
+ * SPDX-License-Identifier: MIT
+ */
+#ifndef _UAPI_SWAP_H
+#define _UAPI_SWAP_H
+
+#include <onyx/types.h>
+
+#define SWAP_VERSION_CURRENT 0
+/* MAGIC = "ONYXSWAP", backwards on big endian */
+#define SWAP_MAGIC           0x5041575358594E4F
+
+/* Used if we go over BADBLOCKS, or if BADBLOCKS themselves are bad */
+#define SWP_FLAG_BAD (1 << 0)
+
+#define SWP_RESERVED_BADBLOCKS_PAGES 8
+
+typedef __u64 __swap_block_t;
+
+struct swap_super
+{
+    __u64 swp_magic;
+    __u32 swp_version;
+    __u32 swp_pagesize;
+    __swap_block_t swp_nr_pages;
+    __u32 swp_flags;
+    __u32 swp_nr_badblocks;
+    __u8 swp_uuid[16];
+};
+
+#define MIN_SWAP_SIZE_PAGES (SWP_RESERVED_BADBLOCKS_PAGES + 1)
+
+#endif

kernel/kernel/fs/buffer.cpp

@@ -487,6 +487,47 @@ struct block_buf *sb_read_block(const struct superblock *sb, unsigned long block
     return buf;
 }
 
+struct block_buf *bdev_read_block(struct blockdev *bdev, unsigned long block)
+{
+    size_t real_off = bdev->block_size * block;
+    size_t aligned_off = real_off & -PAGE_SIZE;
+
+    struct page *page;
+
+    int st = filemap_find_page(bdev->b_ino, real_off >> PAGE_SHIFT, 0, &page, nullptr);
+
+    if (st < 0)
+        return nullptr;
+
+    auto buf = reinterpret_cast<block_buf *>(page->priv);
+
+    while (buf && buf->block_nr != block)
+        buf = buf->next;
+
+    if (unlikely(!buf))
+    {
+        size_t page_off = real_off - aligned_off;
+        sector_t aligned_block = aligned_off / bdev->block_size;
+        sector_t block_nr = aligned_block + ((real_off - aligned_off) / bdev->block_size);
+
+        if (!(buf = page_add_blockbuf(page, page_off)))
+        {
+            page_unref(page);
+            return nullptr;
+        }
+
+        buf->block_nr = block_nr;
+        buf->block_size = bdev->block_size;
+        buf->dev = bdev;
+    }
+
+    block_buf_get(buf);
+
+    page_unref(page);
+
+    return buf;
+}
+
 void block_buf_dirty(block_buf *buf)
 {
     if (!bb_test_and_set(buf, BLOCKBUF_FLAG_DIRTY))

kernel/kernel/fs/namei.cpp

@@ -7,6 +7,7 @@
  */
 #include <onyx/cred.h>
 #include <onyx/dentry.h>
+#include <onyx/err.h>
 #include <onyx/file.h>
 #include <onyx/mount.h>
 #include <onyx/namei.h>
@@ -17,7 +18,88 @@
 
 #include <onyx/memory.hpp>
 
-// XXX(heat): lookup root seems to leak
+enum class fs_token_type : uint8_t
+{
+    REGULAR_TOKEN = 0,
+    LAST_NAME_IN_PATH
+};
+
+/**
+ * @brief Represents a path during a lookup
+ *
+ */
+struct lookup_path
+{
+    std::string_view view;
+    void *ownbuf{nullptr};
+    fs_token_type token_type{fs_token_type::REGULAR_TOKEN};
+    size_t pos{0};
+
+    lookup_path() = default;
+
+    lookup_path(std::string_view view) : view{view}
+    {
+    }
+
+    constexpr bool trailing_slash() const
+    {
+        return view[view.length() - 1] == '/';
+    }
+};
+
+struct nameidata
+{
+    /* Data needed to resolve filesystem names:
+     * view - Contains the pathname;
+     * pos - Contains the offset in the parsing of the pathname;
+     * root - Contains the lookup's filesystem root;
+     * cur - Contains the current relative location and
+     * starts at whatever was passed as the relative dir (controlled with
+     * chdir or *at, or purely through kernel-side use).
+     */
+    /* Note: root and location always hold a reference to the underlying object */
+    struct path root;
+    struct path cur;
+    /* Keeps the parent of cur, *if* we walked once */
+    struct path parent;
+
+    static constexpr const size_t max_loops = SYMLOOP_MAX;
+    /* Number of symbolic links found while looking up -
+     * if it reaches max_symlinks, the lookup fails with -ELOOP.
+     */
+    int nloops{0};
+    int pdepth{0};
+    struct lookup_path paths[SYMLOOP_MAX];
+
+    unsigned int lookup_flags{};
+    int dirfd{AT_FDCWD};
+
+    nameidata(std::string_view view)
+    {
+        paths[0] = lookup_path{view};
+        path_init(&root);
+        path_init(&cur);
+        path_init(&parent);
+    }
+
+    ~nameidata();
+
+    void setcur(struct path newcur)
+    {
+        DCHECK(!path_is_null(&newcur));
+        path_put(&parent);
+        parent = cur;
+        cur = newcur;
+    }
+
+    path getcur()
+    {
+        DCHECK(!path_is_null(&cur));
+        auto ret = cur;
+        path_init(&cur);
+        return ret;
+    }
+};
 
 std::string_view get_token_from_path(lookup_path &path, bool no_consume_if_last)
 {
@@ -1366,3 +1448,10 @@ int path_openat(int dirfd, const char *name, unsigned int flags, struct path *pa
     *path = namedata.getcur();
     return 0;
 }
+
+extern "C" struct file *c_vfs_open(int dirfd, const char *name, unsigned int open_flags,
+                                   mode_t mode)
+{
+    auto ex = vfs_open(dirfd, name, open_flags, mode);
+    return ex.has_value() ? ex.value() : (struct file *) ERR_PTR(ex.error());
+}

kernel/kernel/mm/Makefile

@@ -1,4 +1,4 @@
-mm-y:= bootmem.o page.o pagealloc.o vm_object.o vm.o vmalloc.o reclaim.o anon.o mincore.o page_lru.o rmap.o
+mm-y:= bootmem.o page.o pagealloc.o vm_object.o vm.o vmalloc.o reclaim.o anon.o mincore.o page_lru.o swap.o rmap.o
 mm-$(CONFIG_KUNIT)+= vm_tests.o
 mm-$(CONFIG_X86)+= memory.o
 mm-$(CONFIG_RISCV)+= memory.o