git://scm.lunaixsky.com / lunaix-os.git / blobdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | inline | side by side
refactor: vmm_set_map has option to ignore existed mapping.
[lunaix-os.git] / lunaix-os / kernel / process.c
diff --git a/lunaix-os/kernel/process.c b/lunaix-os/kernel/process.c
index 29cff62c78f16805077a2bbe277ffbc7b2125440..c4e348f19b9dc96844f5cd71d33bd92136dbb2c0 100644 (file)
--- a/lunaix-os/kernel/process.c
+++ b/lunaix-os/kernel/process.c
@@ -1,136 +1,239 @@
-#include <lunaix/process.h>
-#include <lunaix/mm/vmm.h>
-#include <lunaix/mm/region.h>
+#include <klibc/string.h>
 #include <lunaix/clock.h>
 #include <lunaix/clock.h>
-#include <lunaix/syslog.h>
 #include <lunaix/common.h>
 #include <lunaix/common.h>
-#include <lunaix/syscall.h>
+#include <lunaix/mm/pmm.h>
+#include <lunaix/mm/region.h>
+#include <lunaix/mm/vmm.h>
+#include <lunaix/process.h>
 #include <lunaix/spike.h>
 #include <lunaix/spike.h>
+#include <lunaix/status.h>
+#include <lunaix/syscall.h>
+#include <lunaix/syslog.h>
 
 LOG_MODULE("PROC")
 
 
 LOG_MODULE("PROC")
 
-void* __dup_pagetable(pid_t pid, uintptr_t mount_point) {
+void*
+__dup_pagetable(pid_t pid, uintptr_t mount_point)
+{
     void* ptd_pp = pmm_alloc_page(pid, PP_FGPERSIST);
     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);
-    x86_page_table* pptd = (x86_page_table*) (mount_point | (0x3FF << 12));
+    vmm_set_mapping(PD_REFERENCED, PG_MOUNT_1, ptd_pp, PG_PREM_RW, VMAP_NULL);
 
 
-    for (size_t i = 0; i < PG_MAX_ENTRIES - 1; i++)
-    {
+    x86_page_table* ptd = PG_MOUNT_1;
+    x86_page_table* pptd = (x86_page_table*)(mount_point | (0x3FF << 12));
+
+    for (size_t i = 0; i < PG_MAX_ENTRIES - 1; i++) {
         x86_pte_t ptde = pptd->entry[i];
         if (!ptde || !(ptde & PG_PRESENT)) {
             ptd->entry[i] = ptde;
             continue;
         }
         x86_pte_t ptde = pptd->entry[i];
         if (!ptde || !(ptde & PG_PRESENT)) {
             ptd->entry[i] = ptde;
             continue;
         }
-        
-        x86_page_table* ppt = (x86_page_table*) (mount_point | (i << 12));
+
         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);
 
 
-        for (size_t j = 0; j < PG_MAX_ENTRIES; j++)
-        {
+        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];
             x86_pte_t pte = ppt->entry[j];
-            pmm_ref_page(pid, pte & ~0xfff);
+            pmm_ref_page(pid, PG_ENTRY_ADDR(pte));
             pt->entry[j] = pte;
         }
 
         ptd->entry[i] = (uintptr_t)pt_pp | PG_PREM_RW;
     }
             pt->entry[j] = pte;
         }
 
         ptd->entry[i] = (uintptr_t)pt_pp | PG_PREM_RW;
     }
-    
+
     ptd->entry[PG_MAX_ENTRIES - 1] = NEW_L1_ENTRY(T_SELF_REF_PERM, ptd_pp);
 
     return ptd_pp;
 }
 
     ptd->entry[PG_MAX_ENTRIES - 1] = NEW_L1_ENTRY(T_SELF_REF_PERM, ptd_pp);
 
     return ptd_pp;
 }
 
-void* dup_pagetable(pid_t pid) {
+void
+__del_pagetable(pid_t pid, uintptr_t mount_point)
+{
+    x86_page_table* pptd = (x86_page_table*)(mount_point | (0x3FF << 12));
+
+    for (size_t i = 0; i < PG_MAX_ENTRIES - 1; i++) {
+        x86_pte_t ptde = pptd->entry[i];
+        if (!ptde || !(ptde & PG_PRESENT)) {
+            continue;
+        }
+
+        x86_page_table* ppt = (x86_page_table*)(mount_point | (i << 12));
+
+        for (size_t j = 0; j < PG_MAX_ENTRIES; j++) {
+            x86_pte_t pte = ppt->entry[j];
+            // free the 4KB data page
+            if ((pte & PG_PRESENT)) {
+                pmm_free_page(pid, PG_ENTRY_ADDR(pte));
+            }
+        }
+        // free the L2 page table
+        pmm_free_page(pid, PG_ENTRY_ADDR(ptde));
+    }
+    // free the L1 directory
+    pmm_free_page(pid, PG_ENTRY_ADDR(pptd->entry[PG_MAX_ENTRIES - 1]));
+}
+
+void*
+dup_pagetable(pid_t pid)
+{
     return __dup_pagetable(pid, PD_REFERENCED);
 }
 
     return __dup_pagetable(pid, PD_REFERENCED);
 }
 
-__DEFINE_LXSYSCALL(void, fork) {
-    dup_proc();
+__DEFINE_LXSYSCALL(pid_t, fork)
+{
+    return dup_proc();
 }
 
 }
 
-__DEFINE_LXSYSCALL(pid_t, getpid) {
+__DEFINE_LXSYSCALL(pid_t, getpid)
+{
     return __current->pid;
 }
 
     return __current->pid;
 }
 
-__DEFINE_LXSYSCALL(pid_t, getppid) {
+__DEFINE_LXSYSCALL(pid_t, getppid)
+{
     return __current->parent->pid;
 }
 
     return __current->parent->pid;
 }
 
-void dup_proc() {
-    pid_t pid = alloc_pid();
+__DEFINE_LXSYSCALL(pid_t, getpgid)
+{
+    return __current->pgid;
+}
+
+__DEFINE_LXSYSCALL2(int, setpgid, pid_t, pid, pid_t, pgid)
+{
+    struct proc_info* proc = pid ? get_process(pid) : __current;
+
+    if (!proc) {
+        __current->k_status = LXINVL;
+        return -1;
+    }
+
+    pgid = pgid ? pgid : proc->pid;
+
+    struct proc_info* gruppenfuhrer = get_process(pgid);
+
+    if (!gruppenfuhrer || proc->pgid == proc->pid) {
+        __current->k_status = LXINVL;
+        return -1;
+    }
+
+    llist_delete(&proc->grp_member);
+    llist_append(&gruppenfuhrer->grp_member, &proc->grp_member);
+
+    proc->pgid = pgid;
+    return 0;
+}
+
+void
+init_proc(struct proc_info* pcb)
+{
+    memset(pcb, 0, sizeof(*pcb));
 
 
-    struct proc_info pcb = (struct proc_info) {
-        .created = clock_systime(),
-        .pid = pid,
-        .mm = __current->mm,
-        .intr_ctx = __current->intr_ctx,
-        .parent = __current
-    };
+    pcb->pid = alloc_pid();
+    pcb->created = clock_systime();
+    pcb->state = PROC_CREATED;
+    pcb->pgid = pcb->pid;
+}
 
 
-    setup_proc_mem(&pcb, PD_MOUNT_1);    //挂载点#1是当前进程的页表
+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
 
     // 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) {
-        struct mm_region *pos, *n;
-        llist_for_each(pos, n, &__current->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;
-            }
+    if (!__current->mm.regions) {
+        goto not_copy;
+    }
 
 
-            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 = &((x86_page_table*)(PD_MOUNT_1 | ((i & 0xffc00) << 2)))->entry[i & 0x3ff];
-                x86_pte_t *newproc = &((x86_page_table*)(PD_MOUNT_2 | ((i & 0xffc00) << 2)))->entry[i & 0x3ff];
-
-                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。
-                    *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;
-                }
+    llist_init_head(&pcb.mm.regions);
+    struct mm_region *pos, *n;
+    llist_for_each(pos, n, &__current->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;
+        }
+
+        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;
             }
         }
     }
             }
         }
     }
-    
+
+not_copy:
     vmm_unmount_pd(PD_MOUNT_2);
 
     // 正如同fork，返回两次。
     pcb.intr_ctx.registers.eax = 0;
     vmm_unmount_pd(PD_MOUNT_2);
 
     // 正如同fork，返回两次。
     pcb.intr_ctx.registers.eax = 0;
-    __current->intr_ctx.registers.eax = pid;
 
     push_process(&pcb);
 
     push_process(&pcb);
+
+    return pcb.pid;
 }
 
 extern void __kernel_end;
 
 }
 
 extern void __kernel_end;
 
-void setup_proc_mem(struct proc_info* proc, uintptr_t usedMnt) {
+void
+setup_proc_mem(struct proc_info* proc, uintptr_t usedMnt)
+{
     // copy the entire kernel page table
     pid_t pid = proc->pid;
     void* pt_copy = __dup_pagetable(pid, usedMnt);
 
     // copy the entire kernel page table
     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_2, pt_copy); // 将新进程的页表挂载到挂载点#2
 
     // copy the kernel stack
 
     // copy the kernel stack
-    for (size_t i = KSTACK_START >> 12; i <= KSTACK_TOP >> 12; i++)
-    {
-        x86_pte_t *ppte = &((x86_page_table*)(PD_MOUNT_2 | ((i & 0xffc00) << 2)))->entry[i & 0x3ff];
-        void* ppa = vmm_dup_page(pid, PG_ENTRY_ADDR(*ppte));
-        *ppte = (*ppte & 0xfff) | (uintptr_t)ppa;
+    for (size_t i = KSTACK_START >> 12; i <= KSTACK_TOP >> 12; i++) {
+        volatile x86_pte_t* ppte = &PTE_MOUNTED(PD_MOUNT_2, i);
+
+        /*
+            This is a fucking nightmare, the TLB caching keep the rewrite to PTE
+           from updating. Even the Nightmare Moon the Evil is far less nasty
+           than this. It took me hours of debugging to figure this out.
+
+            In the name of Celestia our glorious goddess, I will fucking HATE
+           the TLB for the rest of my LIFE!
+        */
+        cpu_invplg(ppte);
+
+        x86_pte_t p = *ppte;
+        void* ppa = vmm_dup_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.
     }
 
     // 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！）
+
+    // 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);
 
     // FIXME: 这里应该放到spawn_proc里面。
     // region_add(proc, USTACK_END, USTACK_SIZE, REGION_PRIVATE | REGION_RW);
 
The lunaix project - hobby kernel from scratch