refactor: mount system reworked

[lunaix-os.git] / lunaix-os / kernel / process.c

diff --git a/lunaix-os/kernel/process.c b/lunaix-os/kernel/process.c
index 020dfb47fe3fe8a5a92c8b404f11005bf3a5ec03..086a5e3105a184ca3ccf784f3f8fb7cba46445f9 100644 (file)
--- a/lunaix-os/kernel/process.c
+++ b/lunaix-os/kernel/process.c
@@ -1,7 +1,9 @@
 #include <klibc/string.h>
 #include <lunaix/clock.h>
 #include <lunaix/common.h>
 #include <klibc/string.h>
 #include <lunaix/clock.h>
 #include <lunaix/common.h>

+#include <lunaix/mm/pmm.h>

 #include <lunaix/mm/region.h>
 #include <lunaix/mm/region.h>

+#include <lunaix/mm/valloc.h>

 #include <lunaix/mm/vmm.h>
 #include <lunaix/process.h>
 #include <lunaix/spike.h>
 #include <lunaix/mm/vmm.h>
 #include <lunaix/process.h>
 #include <lunaix/spike.h>


@@ -15,19 +17,30 @@ void*
 __dup_pagetable(pid_t pid, uintptr_t mount_point)
 {
     void* ptd_pp = pmm_alloc_page(pid, PP_FGPERSIST);
 __dup_pagetable(pid_t pid, uintptr_t mount_point)
 {
     void* ptd_pp = pmm_alloc_page(pid, PP_FGPERSIST);

-    x86_page_table* ptd = vmm_fmap_page(pid, PG_MOUNT_1, ptd_pp, PG_PREM_RW);
+    vmm_set_mapping(PD_REFERENCED, PG_MOUNT_1, ptd_pp, PG_PREM_RW, VMAP_NULL);
+
+    x86_page_table* ptd = PG_MOUNT_1;

     x86_page_table* pptd = (x86_page_table*)(mount_point | (0x3FF << 12));
 
     x86_page_table* pptd = (x86_page_table*)(mount_point | (0x3FF << 12));
 

+    size_t kspace_l1inx = L1_INDEX(KERNEL_MM_BASE);
+

     for (size_t i = 0; i < PG_MAX_ENTRIES - 1; i++) {
     for (size_t i = 0; i < PG_MAX_ENTRIES - 1; i++) {

+

         x86_pte_t ptde = pptd->entry[i];
         x86_pte_t ptde = pptd->entry[i];

-        if (!ptde || !(ptde & PG_PRESENT)) {
+        // 空或者是未在内存中的L1页表项直接照搬过去。
+        // 内核地址空间直接共享过去。
+        if (!ptde || i >= kspace_l1inx || !(ptde & PG_PRESENT)) {

             ptd->entry[i] = ptde;
             continue;
         }
 
             ptd->entry[i] = ptde;
             continue;
         }
 

-        x86_page_table* ppt = (x86_page_table*)(mount_point | (i << 12));
+        // 复制L2页表

         void* pt_pp = pmm_alloc_page(pid, PP_FGPERSIST);
         void* pt_pp = pmm_alloc_page(pid, PP_FGPERSIST);

-        x86_page_table* pt = vmm_fmap_page(pid, PG_MOUNT_2, pt_pp, PG_PREM_RW);
+        vmm_set_mapping(
+          PD_REFERENCED, PG_MOUNT_2, pt_pp, PG_PREM_RW, VMAP_NULL);
+
+        x86_page_table* ppt = (x86_page_table*)(mount_point | (i << 12));
+        x86_page_table* pt = PG_MOUNT_2;

 
         for (size_t j = 0; j < PG_MAX_ENTRIES; j++) {
             x86_pte_t pte = ppt->entry[j];
 
         for (size_t j = 0; j < PG_MAX_ENTRIES; j++) {
             x86_pte_t pte = ppt->entry[j];


@@ -35,7 +48,7 @@ __dup_pagetable(pid_t pid, uintptr_t mount_point)
             pt->entry[j] = pte;
         }
 
             pt->entry[j] = pte;
         }
 

-        ptd->entry[i] = (uintptr_t)pt_pp | PG_PREM_RW;
+        ptd->entry[i] = (uintptr_t)pt_pp | PG_ENTRY_FLAGS(ptde);

     }
 
     ptd->entry[PG_MAX_ENTRIES - 1] = NEW_L1_ENTRY(T_SELF_REF_PERM, ptd_pp);
     }
 
     ptd->entry[PG_MAX_ENTRIES - 1] = NEW_L1_ENTRY(T_SELF_REF_PERM, ptd_pp);


@@ -48,7 +61,8 @@ __del_pagetable(pid_t pid, uintptr_t mount_point)
 {
     x86_page_table* pptd = (x86_page_table*)(mount_point | (0x3FF << 12));
 
 {
     x86_page_table* pptd = (x86_page_table*)(mount_point | (0x3FF << 12));
 

-    for (size_t i = 0; i < PG_MAX_ENTRIES - 1; i++) {
+    // only remove user address space
+    for (size_t i = 0; i < L1_INDEX(KERNEL_MM_BASE); i++) {

         x86_pte_t ptde = pptd->entry[i];
         if (!ptde || !(ptde & PG_PRESENT)) {
             continue;
         x86_pte_t ptde = pptd->entry[i];
         if (!ptde || !(ptde & PG_PRESENT)) {
             continue;


@@ -71,7 +85,7 @@ __del_pagetable(pid_t pid, uintptr_t mount_point)
 }
 
 void*
 }
 
 void*

-dup_pagetable(pid_t pid)
+vmm_dup_vmspace(pid_t pid)

 {
     return __dup_pagetable(pid, PD_REFERENCED);
 }
 {
     return __dup_pagetable(pid, PD_REFERENCED);
 }


@@ -101,7 +115,7 @@ __DEFINE_LXSYSCALL2(int, setpgid, pid_t, pid, pid_t, pgid)
     struct proc_info* proc = pid ? get_process(pid) : __current;
 
     if (!proc) {
     struct proc_info* proc = pid ? get_process(pid) : __current;
 
     if (!proc) {

-        __current->k_status = LXINVL;
+        __current->k_status = EINVAL;

         return -1;
     }
 
         return -1;
     }
 


@@ -110,7 +124,7 @@ __DEFINE_LXSYSCALL2(int, setpgid, pid_t, pid, pid_t, pgid)
     struct proc_info* gruppenfuhrer = get_process(pgid);
 
     if (!gruppenfuhrer || proc->pgid == proc->pid) {
     struct proc_info* gruppenfuhrer = get_process(pgid);
 
     if (!gruppenfuhrer || proc->pgid == proc->pid) {

-        __current->k_status = LXINVL;
+        __current->k_status = EINVAL;

         return -1;
     }
 
         return -1;
     }
 


@@ -122,75 +136,94 @@ __DEFINE_LXSYSCALL2(int, setpgid, pid_t, pid, pid_t, pgid)
 }
 
 void
 }
 
 void

-init_proc(struct proc_info* pcb)
+init_proc_user_space(struct proc_info* pcb)

 {
 {

-    memset(pcb, 0, sizeof(*pcb));
+    vmm_mount_pd(PD_MOUNT_1, pcb->page_table);
+
+    /*---  分配用户栈  ---*/

 
 

-    pcb->pid = alloc_pid();
-    pcb->created = clock_systime();
-    pcb->state = PROC_CREATED;
-    pcb->pgid = pcb->pid;
+    // 注册用户栈区域
+    region_add(
+      &pcb->mm.regions, USTACK_END, USTACK_TOP, REGION_RW | REGION_RSHARED);
+
+    // 预留地址空间，具体物理页将由Page Fault Handler按需分配。
+    for (uintptr_t i = PG_ALIGN(USTACK_END); i < USTACK_TOP; i += PG_SIZE) {
+        vmm_set_mapping(PD_MOUNT_1, i, 0, PG_ALLOW_USER | PG_WRITE, VMAP_NULL);
+    }
+
+    // todo: other uspace initialization stuff
+
+    vmm_unmount_pd(PD_MOUNT_1);
+}
+
+void
+__mark_region(uintptr_t start_vpn, uintptr_t end_vpn, int attr)
+{
+    for (size_t i = start_vpn; i <= end_vpn; i++) {
+        x86_pte_t* curproc = &PTE_MOUNTED(PD_REFERENCED, i);
+        x86_pte_t* newproc = &PTE_MOUNTED(PD_MOUNT_1, i);
+        cpu_invplg(newproc);
+
+        if ((attr & REGION_MODE_MASK) == REGION_RSHARED) {
+            // 如果读共享，则将两者的都标注为只读，那么任何写入都将会应用COW策略。
+            cpu_invplg(curproc);
+            cpu_invplg(i << 12);
+            *curproc = *curproc & ~PG_WRITE;
+            *newproc = *newproc & ~PG_WRITE;
+        } else {
+            // 如果是私有页，则将该页从新进程中移除。
+            *newproc = 0;
+        }
+    }
+}
+
+void
+__copy_fdtable(struct proc_info* pcb)
+{
+    for (size_t i = 0; i < VFS_MAX_FD; i++) {
+        struct v_fd* fd = __current->fdtable->fds[i];
+        if (!fd)
+            continue;
+        vfs_dup_fd(fd, &pcb->fdtable->fds[i]);
+    }

 }
 
 pid_t
 dup_proc()
 {
 }
 
 pid_t
 dup_proc()
 {

-    struct proc_info pcb;
-    init_proc(&pcb);
-    pcb.mm = __current->mm;
-    pcb.intr_ctx = __current->intr_ctx;
-    pcb.parent = __current;
-
-#ifdef USE_KERNEL_PG
-    setup_proc_mem(&pcb, PD_MOUNT_1); //挂载点#1是当前进程的页表
-#else
-    setup_proc_mem(&pcb, PD_REFERENCED);
-#endif
+    struct proc_info* pcb = alloc_process();
+    pcb->mm.u_heap = __current->mm.u_heap;
+    pcb->intr_ctx = __current->intr_ctx;
+    pcb->parent = __current;
+
+    __copy_fdtable(pcb);
+    region_copy(&__current->mm.regions, &pcb->mm.regions);
+
+    setup_proc_mem(pcb, PD_REFERENCED);

 
     // 根据 mm_region 进一步配置页表
 
     // 根据 mm_region 进一步配置页表

-    if (!__current->mm.regions) {
-        goto not_copy;
-    }

 
 

-    llist_init_head(&pcb.mm.regions);

     struct mm_region *pos, *n;
     struct mm_region *pos, *n;

-    llist_for_each(pos, n, &__current->mm.regions->head, head)
+    llist_for_each(pos, n, &pcb->mm.regions.head, head)

     {
     {

-        region_add(&pcb, pos->start, pos->end, pos->attr);
-

         // 如果写共享，则不作处理。
         if ((pos->attr & REGION_WSHARED)) {
             continue;
         }
 
         // 如果写共享，则不作处理。
         if ((pos->attr & REGION_WSHARED)) {
             continue;
         }
 

-        uintptr_t start_vpn = PG_ALIGN(pos->start) >> 12;
-        uintptr_t end_vpn = PG_ALIGN(pos->end) >> 12;
-        for (size_t i = start_vpn; i < end_vpn; i++) {
-            x86_pte_t* curproc = &PTE_MOUNTED(PD_MOUNT_1, i);
-            x86_pte_t* newproc = &PTE_MOUNTED(PD_MOUNT_2, i);
-            cpu_invplg(newproc);
-
-            if (pos->attr == REGION_RSHARED) {
-                // 如果读共享，则将两者的都标注为只读，那么任何写入都将会应用COW策略。
-                cpu_invplg(curproc);
-                *curproc = *curproc & ~PG_WRITE;
-                *newproc = *newproc & ~PG_WRITE;
-            } else {
-                // 如果是私有页，则将该页从新进程中移除。
-                *newproc = 0;
-            }
-        }
+        uintptr_t start_vpn = pos->start >> 12;
+        uintptr_t end_vpn = pos->end >> 12;
+        __mark_region(start_vpn, end_vpn, pos->attr);

     }
 
     }
 

-not_copy:
-    vmm_unmount_pd(PD_MOUNT_2);
+    vmm_unmount_pd(PD_MOUNT_1);

 
     // 正如同fork，返回两次。
 
     // 正如同fork，返回两次。

-    pcb.intr_ctx.registers.eax = 0;
+    pcb->intr_ctx.registers.eax = 0;

 
 

-    push_process(&pcb);
+    commit_process(pcb);

 
 

-    return pcb.pid;
+    return pcb->pid;

 }
 
 extern void __kernel_end;
 }
 
 extern void __kernel_end;


@@ -202,11 +235,11 @@ setup_proc_mem(struct proc_info* proc, uintptr_t usedMnt)
     pid_t pid = proc->pid;
     void* pt_copy = __dup_pagetable(pid, usedMnt);
 
     pid_t pid = proc->pid;
     void* pt_copy = __dup_pagetable(pid, usedMnt);
 

-    vmm_mount_pd(PD_MOUNT_2, pt_copy); // 将新进程的页表挂载到挂载点#2
+    vmm_mount_pd(PD_MOUNT_1, pt_copy); // 将新进程的页表挂载到挂载点#2

 
     // copy the kernel stack
     for (size_t i = KSTACK_START >> 12; i <= KSTACK_TOP >> 12; i++) {
 
     // copy the kernel stack
     for (size_t i = KSTACK_START >> 12; i <= KSTACK_TOP >> 12; i++) {

-        volatile x86_pte_t* ppte = &PTE_MOUNTED(PD_MOUNT_2, i);
+        volatile x86_pte_t* ppte = &PTE_MOUNTED(PD_MOUNT_1, i);

 
         /*
             This is a fucking nightmare, the TLB caching keep the rewrite to PTE
 
         /*
             This is a fucking nightmare, the TLB caching keep the rewrite to PTE


@@ -220,18 +253,13 @@ setup_proc_mem(struct proc_info* proc, uintptr_t usedMnt)
 
         x86_pte_t p = *ppte;
         void* ppa = vmm_dup_page(pid, PG_ENTRY_ADDR(p));
 
         x86_pte_t p = *ppte;
         void* ppa = vmm_dup_page(pid, PG_ENTRY_ADDR(p));

+        pmm_free_page(pid, PG_ENTRY_ADDR(p));

         *ppte = (p & 0xfff) | (uintptr_t)ppa;
     }
 
     // 我们不需要分配内核的区域，因为所有的内核代码和数据段只能通过系统调用来访问，任何非法的访问
     // 都会导致eip落在区域外面，从而segmentation fault.
 
         *ppte = (p & 0xfff) | (uintptr_t)ppa;
     }
 
     // 我们不需要分配内核的区域，因为所有的内核代码和数据段只能通过系统调用来访问，任何非法的访问
     // 都会导致eip落在区域外面，从而segmentation fault.
 

-    // 定义用户栈区域，但是不分配实际的物理页。我们会在Page fault
-    // handler里面实现动态分配物理页的逻辑。（虚拟内存的好处！）
-    // FIXME: 这里应该放到spawn_proc里面。
-    // region_add(proc, USTACK_END, USTACK_SIZE, REGION_PRIVATE | REGION_RW);
-

     // 至于其他的区域我们暂时没有办法知道，因为那需要知道用户程序的信息。我们留到之后在处理。
     // 至于其他的区域我们暂时没有办法知道，因为那需要知道用户程序的信息。我们留到之后在处理。

-

     proc->page_table = pt_copy;
 }
\ No newline at end of file
     proc->page_table = pt_copy;
 }
\ No newline at end of file