Support bringing up block device as rootfs

etc/syscalls_linux_aarch64.md

@@ -40,9 +40,9 @@
 | 0x24 (36)   | symlinkat               | (const char *oldname, int newdfd, const char *newname)                                                                                     | __arm64_sys_symlinkat               | true        |
 | 0x25 (37)   | linkat                  | (int olddfd, const char *oldname, int newdfd, const char *newname, int flags)                                                              | __arm64_sys_linkat                  | true        |
 | 0x26 (38)   | renameat                | (int olddfd, const char *oldname, int newdfd, const char *newname)                                                                         | __arm64_sys_renameat                | true        |
-| 0x27 (39)   | umount                  | (char *name, int flags)                                                                                                                    | __arm64_sys_umount                  | false       |
+| 0x27 (39)   | umount                  | (char *name, int flags)                                                                                                                    | __arm64_sys_umount                  | true        |
 | 0x28 (40)   | mount                   | (char *dev_name, char *dir_name, char *type, unsigned long flags, void *data)                                                              | __arm64_sys_mount                   | partial     |
-| 0x29 (41)   | pivot_root              | (const char *new_root, const char *put_old)                                                                                                | __arm64_sys_pivot_root              | false       |
+| 0x29 (41)   | pivot_root              | (const char *new_root, const char *put_old)                                                                                                | __arm64_sys_pivot_root              | partial     |
 | 0x2b (43)   | statfs                  | (const char *pathname, struct statfs *buf)                                                                                                 | __arm64_sys_statfs                  | partial     |
 | 0x2c (44)   | fstatfs                 | (unsigned int fd, struct statfs *buf)                                                                                                      | __arm64_sys_fstatfs                 | partial     |
 | 0x2d (45)   | truncate                | (const char *path, long length)                                                                                                            | __arm64_sys_truncate                | true        |
@@ -158,7 +158,7 @@
 | 0x9b (155)  | getpgid                 | (pid_t pid)                                                                                                                                | __arm64_sys_getpgid                 | true        |
 | 0x9c (156)  | getsid                  | (pid_t pid)                                                                                                                                | __arm64_sys_getsid                  | true        |
 | 0x9d (157)  | setsid                  | ()                                                                                                                                         | __arm64_sys_setsid                  | true        |
-| 0x9e (158)  | getgroups               | (int gidsetsize, gid_t *grouplist)                                                                                                         | __arm64_sys_getgroups               | false       |
+| 0x9e (158)  | getgroups               | (int gidsetsize, gid_t *grouplist)                                                                                                         | __arm64_sys_getgroups               | stub        |
 | 0x9f (159)  | setgroups               | (int gidsetsize, gid_t *grouplist)                                                                                                         | __arm64_sys_setgroups               | false       |
 | 0xa0 (160)  | newuname                | (struct new_utsname *name)                                                                                                                 | __arm64_sys_newuname                | true        |
 | 0xa1 (161)  | sethostname             | (char *name, int len)                                                                                                                      | __arm64_sys_sethostname             | true        |

libkernel/src/arch/arm64/memory/pg_walk.rs

@@ -2,18 +2,20 @@
 
 use super::{
     pg_descriptors::L3Descriptor,
-    pg_tables::{L0Table, L3Table},
+    pg_tables::{L0Table, L1Table, L3Table},
 };
 use crate::{
     error::{MapError, Result},
     memory::{
         PAGE_SIZE,
-        address::{TPA, VA},
+        address::{PA, TPA, VA},
         paging::{
-            NullTlbInvalidator, PageTableEntry, PageTableMapper, PgTable, PgTableArray,
-            walk::{RecursiveWalker, WalkContext},
+            NullTlbInvalidator, PaMapper, PageTableEntry, PageTableMapper, PgTable, PgTableArray,
+            TableMapper,
+            permissions::PtePermissions,
+            walk::{RecursiveWalker, Translator, WalkContext},
         },
-        region::VirtMemoryRegion,
+        region::{PhysMemoryRegion, VirtMemoryRegion},
     },
 };
 
@@ -111,6 +113,71 @@ pub fn get_pte<PM: PageTableMapper>(
     Ok(descriptor)
 }
 
+impl Translator for L0Table {
+    fn translate<PM: PageTableMapper>(
+        table_pa: TPA<PgTableArray<Self>>,
+        va: VA,
+        ctx: &mut WalkContext<PM>,
+    ) -> Result<Option<(PA, usize, PtePermissions)>> {
+        let desc = unsafe {
+            ctx.mapper
+                .with_page_table(table_pa, |pgtable| Self::from_ptr(pgtable).get_desc(va))?
+        };
+
+        match desc.next_table_address() {
+            Some(next_pa) => L1Table::translate(next_pa, va, ctx),
+            None if desc.is_valid() => Err(MapError::InvalidDescriptor.into()),
+            None => Ok(None),
+        }
+    }
+}
+
+impl Translator for L3Table {
+    fn translate<PM: PageTableMapper>(
+        table_pa: TPA<PgTableArray<Self>>,
+        va: VA,
+        ctx: &mut WalkContext<PM>,
+    ) -> Result<Option<(PA, usize, PtePermissions)>> {
+        let desc = unsafe {
+            ctx.mapper
+                .with_page_table(table_pa, |pgtable| Self::from_ptr(pgtable).get_desc(va))?
+        };
+
+        match desc.mapped_address() {
+            Some(pa) => Ok(Some((
+                pa,
+                1 << Self::Descriptor::MAP_SHIFT,
+                desc.permissions().unwrap(),
+            ))),
+            None if desc.is_valid() => Err(MapError::InvalidDescriptor.into()),
+            None => Ok(None),
+        }
+    }
+}
+
+/// Translates the VA into a physical region plus an offset and permissions.
+pub fn translate<PM: PageTableMapper>(
+    l0_table: TPA<PgTableArray<L0Table>>,
+    va: VA,
+    mapper: &mut PM,
+) -> Result<Option<(PhysMemoryRegion, usize, PtePermissions)>> {
+    let mut walk_ctx = WalkContext {
+        mapper,
+        // Safe to not invalidate the TLB, as we are not modifying any PTEs.
+        invalidator: &NullTlbInvalidator {},
+    };
+
+    if let Some((pa, blk_sz, perms)) = L0Table::translate(l0_table, va, &mut walk_ctx)? {
+        debug_assert!(blk_sz.is_power_of_two());
+
+        let offset = va.value() & (blk_sz - 1);
+
+        Ok(Some((PhysMemoryRegion::new(pa, blk_sz), offset, perms)))
+    } else {
+        Ok(None)
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

libkernel/src/driver.rs

@@ -8,3 +8,13 @@ pub struct CharDevDescriptor {
     /// The minor device number (identifies the device instance).
     pub minor: u64,
 }
+
+impl CharDevDescriptor {
+    /// Encodes this device descriptor into a Linux-style `dev_t` value.
+    pub const fn dev_t(self) -> u64 {
+        (self.minor & 0xff)
+            | ((self.major & 0xfff) << 8)
+            | ((self.minor & !0xff) << 12)
+            | ((self.major & !0xfff) << 32)
+    }
+}

scripts/qemu_runner.py

@@ -14,6 +14,7 @@
 parser.add_argument("--memory", default="2G")
 parser.add_argument("--debug", action="store_true", help="Enable QEMU debugging")
 parser.add_argument("--display", action="store_true", help="Add a display device to the VM")
+parser.add_argument("--disk", action="store_true", help="Add a disk device to the VM")
 
 
 
@@ -41,6 +42,7 @@
     "-rtc": "base=utc,clock=host",
     "-nographic": None,
     "-s": None,
+    "-snapshot": None,
     "-kernel": bin_executable_location,
     "-append": f"{append_args} --rootfs=ext4fs --automount=/dev,devfs --automount=/tmp,tmpfs --automount=/proc,procfs --automount=/sys,sysfs",
 }
@@ -59,6 +61,10 @@
     default_args["-serial"] = "stdio"
     extra_args += ["-device", "virtio-gpu-device"]
 
+if args.disk:
+    default_args["-drive"] = "file=rootfs.img,format=raw,if=none,cache=none,id=x0"
+    extra_args += ["-device", "virtio-blk-device,drive=x0"]
+
 qemu_command = ["qemu-system-aarch64"]
 
 for key, value in default_args.items():

src/arch/arm64/exceptions/syscall.rs

@@ -36,6 +36,7 @@ use crate::{
             iov::{sys_preadv, sys_preadv2, sys_pwritev, sys_pwritev2, sys_readv, sys_writev},
             listxattr::{sys_flistxattr, sys_listxattr, sys_llistxattr},
             mount::sys_mount,
+            pivot_root::sys_pivot_root,
             removexattr::{sys_fremovexattr, sys_lremovexattr, sys_removexattr},
             rw::{sys_pread64, sys_pwrite64, sys_read, sys_write},
             seek::sys_lseek,
@@ -45,6 +46,7 @@ use crate::{
             statfs::{sys_fstatfs, sys_statfs},
             sync::{sys_fdatasync, sys_fsync, sys_sync, sys_syncfs},
             trunc::{sys_ftruncate, sys_truncate},
+            umount::sys_umount2,
         },
     },
     kernel::{
@@ -71,9 +73,9 @@ use crate::{
         caps::{sys_capget, sys_capset},
         clone::sys_clone,
         creds::{
-            sys_getegid, sys_geteuid, sys_getgid, sys_getresgid, sys_getresuid, sys_getsid,
-            sys_gettid, sys_getuid, sys_setfsgid, sys_setfsuid, sys_setgid, sys_setregid,
-            sys_setresgid, sys_setresuid, sys_setreuid, sys_setsid, sys_setuid,
+            sys_getegid, sys_geteuid, sys_getgid, sys_getgroups, sys_getresgid, sys_getresuid,
+            sys_getsid, sys_gettid, sys_getuid, sys_setfsgid, sys_setfsuid, sys_setgid,
+            sys_setregid, sys_setresgid, sys_setresuid, sys_setreuid, sys_setsid, sys_setuid,
         },
         epoll::{sys_epoll_create1, sys_epoll_ctl, sys_epoll_pwait},
         exec::sys_execve,
@@ -284,6 +286,7 @@ pub async fn handle_syscall(mut ctx: ProcessCtx) {
             )
             .await
         }
+        0x27 => sys_umount2(&ctx, TUA::from_value(arg1 as _), arg2 as _).await,
         0x28 => {
             sys_mount(
                 &ctx,
@@ -295,6 +298,7 @@ pub async fn handle_syscall(mut ctx: ProcessCtx) {
             )
             .await
         }
+        0x29 => sys_pivot_root(&ctx, TUA::from_value(arg1 as _), TUA::from_value(arg2 as _)).await,
         0x2b => sys_statfs(&ctx, TUA::from_value(arg1 as _), TUA::from_value(arg2 as _)).await,
         0x2c => sys_fstatfs(&ctx, arg1.into(), TUA::from_value(arg2 as _)).await,
         0x2d => sys_truncate(&ctx, TUA::from_value(arg1 as _), arg2 as _).await,
@@ -616,6 +620,7 @@ pub async fn handle_syscall(mut ctx: ProcessCtx) {
         0x9b => sys_getpgid(&ctx, arg1 as _),
         0x9c => sys_getsid(&ctx).await,
         0x9d => sys_setsid(&ctx).await,
+        0x9e => sys_getgroups(&ctx, arg1 as _, TUA::from_value(arg2 as _)).map_err(|e| match e {}),
         0xa0 => sys_uname(TUA::from_value(arg1 as _)).await,
         0xa1 => sys_sethostname(&ctx, TUA::from_value(arg1 as _), arg2 as _).await,
         0xa3 => Err(KernelError::InvalidValue),

src/arch/arm64/memory/address_space.rs

@@ -14,7 +14,7 @@ use libkernel::{
         pg_descriptors::{L3Descriptor, MemoryType},
         pg_tables::{L0Table, MapAttributes, MappingContext, map_range},
         pg_tear_down::tear_down_address_space,
-        pg_walk::{get_pte, walk_and_modify_region},
+        pg_walk::{translate as translate_va, walk_and_modify_region},
     },
     error::{KernelError, MapError, Result},
     memory::{
@@ -137,16 +137,15 @@ impl UserAddressSpace for Arm64ProcessAddressSpace {
     }
 
     fn translate(&self, va: VA) -> Option<PageInfo> {
-        let pte = get_pte(
-            self.l0_table,
-            va.page_aligned(),
-            &mut PageOffsetPgTableMapper {},
-        )
-        .unwrap()?;
+        let (region, offset, perms) =
+            translate_va(self.l0_table, va, &mut PageOffsetPgTableMapper {})
+                .ok()
+                .flatten()?;
+        let pa = region.start_address().add_bytes(offset);
 
         Some(PageInfo {
-            pfn: pte.mapped_address()?.to_pfn(),
-            perms: pte.permissions()?,
+            pfn: pa.to_pfn(),
+            perms,
         })
     }
 

src/arch/arm64/memory/mmu.rs

@@ -4,12 +4,12 @@ use libkernel::{
     arch::arm64::memory::{
         pg_descriptors::MemoryType,
         pg_tables::{L0Table, MapAttributes, MappingContext, map_range},
-        pg_walk::get_pte,
+        pg_walk::translate as translate_va,
     },
     error::Result,
     memory::{
         address::{PA, TPA, VA},
-        paging::{PaMapper, PgTableArray, permissions::PtePermissions},
+        paging::{PgTableArray, permissions::PtePermissions},
         proc_vm::address_space::KernAddressSpace,
         region::{PhysMemoryRegion, VirtMemoryRegion},
     },
@@ -39,16 +39,11 @@ impl Arm64KernelAddressSpace {
         map_range(self.kernel_l0, map_attrs, &mut ctx)
     }
 
-    pub fn translate(&self, va: VA) -> Option<PA> {
-        let pg_offset = va.page_offset();
-
-        let pte = get_pte(self.kernel_l0, va, &mut PageOffsetPgTableMapper {})
+    pub fn translate(&self, va: VA) -> Option<(PhysMemoryRegion, usize)> {
+        translate_va(self.kernel_l0, va, &mut PageOffsetPgTableMapper {})
             .ok()
-            .flatten()?;
-
-        let pa = pte.mapped_address()?;
-
-        Some(pa.add_bytes(pg_offset))
+            .flatten()
+            .map(|(region, offset, _)| (region, offset))
     }
 
     pub fn table_pa(&self) -> PA {