Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / arch / um / sys-i386 / ldt.c
index dc755b0..fe0877b 100644 (file)
  * Licensed under the GPL
  */
 
+#include "linux/stddef.h"
 #include "linux/config.h"
+#include "linux/sched.h"
 #include "linux/slab.h"
+#include "linux/types.h"
+#include "linux/errno.h"
 #include "asm/uaccess.h"
-#include "asm/ptrace.h"
+#include "asm/smp.h"
+#include "asm/ldt.h"
+#include "asm/unistd.h"
 #include "choose-mode.h"
 #include "kern.h"
+#include "mode_kern.h"
+#include "os.h"
 
-#ifdef CONFIG_MODE_TT
 extern int modify_ldt(int func, void *ptr, unsigned long bytecount);
 
-/* XXX this needs copy_to_user and copy_from_user */
-
-int sys_modify_ldt_tt(int func, void __user *ptr, unsigned long bytecount)
-{
-       if (!access_ok(VERIFY_READ, ptr, bytecount))
-               return -EFAULT;
-
-       return modify_ldt(func, ptr, bytecount);
-}
-#endif
-
-#ifdef CONFIG_MODE_SKAS
-extern int userspace_pid[];
-
-#include "skas_ptrace.h"
+#ifdef CONFIG_MODE_TT
 
-int sys_modify_ldt_skas(int func, void __user *ptr, unsigned long bytecount)
+static long do_modify_ldt_tt(int func, void __user *ptr,
+                             unsigned long bytecount)
 {
-       struct ptrace_ldt ldt;
-       void *buf;
-       int res, n;
-
-       buf = kmalloc(bytecount, GFP_KERNEL);
-       if(buf == NULL)
-               return(-ENOMEM);
-
-       res = 0;
+       struct user_desc info;
+       int res = 0;
+       void *buf = NULL;
+       void *p = NULL; /* What we pass to host. */
 
        switch(func){
        case 1:
-       case 0x11:
-               res = copy_from_user(buf, ptr, bytecount);
+       case 0x11: /* write_ldt */
+               /* Do this check now to avoid overflows. */
+               if (bytecount != sizeof(struct user_desc)) {
+                       res = -EINVAL;
+                       goto out;
+               }
+
+               if(copy_from_user(&info, ptr, sizeof(info))) {
+                       res = -EFAULT;
+                       goto out;
+               }
+
+               p = &info;
                break;
-       }
+       case 0:
+       case 2: /* read_ldt */
 
-       if(res != 0){
-               res = -EFAULT;
+               /* The use of info avoids kmalloc on the write case, not on the
+                * read one. */
+               buf = kmalloc(bytecount, GFP_KERNEL);
+               if (!buf) {
+                       res = -ENOMEM;
+                       goto out;
+               }
+               p = buf;
+               break;
+       default:
+               res = -ENOSYS;
                goto out;
        }
 
-       ldt = ((struct ptrace_ldt) { .func      = func,
-                                    .ptr       = buf,
-                                    .bytecount = bytecount });
-#warning Need to look up userspace_pid by cpu
-       res = ptrace(PTRACE_LDT, userspace_pid[0], 0, (unsigned long) &ldt);
+       res = modify_ldt(func, p, bytecount);
        if(res < 0)
                goto out;
 
        switch(func){
        case 0:
        case 2:
-               n = res;
-               res = copy_to_user(ptr, buf, n);
-               if(res != 0)
+               /* Modify_ldt was for reading and returned the number of read
+                * bytes.*/
+               if(copy_to_user(ptr, p, res))
                        res = -EFAULT;
-               else 
-                       res = n;
                break;
        }
 
- out:
+out:
        kfree(buf);
-       return(res);
+       return res;
 }
+
 #endif
 
-int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
+#ifdef CONFIG_MODE_SKAS
+
+#include "skas.h"
+#include "skas_ptrace.h"
+#include "asm/mmu_context.h"
+#include "proc_mm.h"
+
+long write_ldt_entry(struct mm_id * mm_idp, int func, struct user_desc * desc,
+                    void **addr, int done)
 {
-       return(CHOOSE_MODE_PROC(sys_modify_ldt_tt, sys_modify_ldt_skas, func, 
-                               ptr, bytecount));
+       long res;
+
+       if(proc_mm){
+               /* This is a special handling for the case, that the mm to
+                * modify isn't current->active_mm.
+                * If this is called directly by modify_ldt,
+                *     (current->active_mm->context.skas.u == mm_idp)
+                * will be true. So no call to switch_mm_skas(mm_idp) is done.
+                * If this is called in case of init_new_ldt or PTRACE_LDT,
+                * mm_idp won't belong to current->active_mm, but child->mm.
+                * So we need to switch child's mm into our userspace, then
+                * later switch back.
+                *
+                * Note: I'm unsure: should interrupts be disabled here?
+                */
+               if(!current->active_mm || current->active_mm == &init_mm ||
+                  mm_idp != &current->active_mm->context.skas.id)
+                       switch_mm_skas(mm_idp);
+       }
+
+       if(ptrace_ldt) {
+               struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
+                       .func = func,
+                       .ptr = desc,
+                       .bytecount = sizeof(*desc)};
+               u32 cpu;
+               int pid;
+
+               if(!proc_mm)
+                       pid = mm_idp->u.pid;
+               else {
+                       cpu = get_cpu();
+                       pid = userspace_pid[cpu];
+               }
+
+               res = os_ptrace_ldt(pid, 0, (unsigned long) &ldt_op);
+
+               if(proc_mm)
+                       put_cpu();
+       }
+       else {
+               void *stub_addr;
+               res = syscall_stub_data(mm_idp, (unsigned long *)desc,
+                                       (sizeof(*desc) + sizeof(long) - 1) &
+                                           ~(sizeof(long) - 1),
+                                       addr, &stub_addr);
+               if(!res){
+                       unsigned long args[] = { func,
+                                                (unsigned long)stub_addr,
+                                                sizeof(*desc),
+                                                0, 0, 0 };
+                       res = run_syscall_stub(mm_idp, __NR_modify_ldt, args,
+                                              0, addr, done);
+               }
+       }
+
+       if(proc_mm){
+               /* This is the second part of special handling, that makes
+                * PTRACE_LDT possible to implement.
+                */
+               if(current->active_mm && current->active_mm != &init_mm &&
+                  mm_idp != &current->active_mm->context.skas.id)
+                       switch_mm_skas(&current->active_mm->context.skas.id);
+       }
+
+       return res;
 }
 
+static long read_ldt_from_host(void __user * ptr, unsigned long bytecount)
+{
+       int res, n;
+       struct ptrace_ldt ptrace_ldt = (struct ptrace_ldt) {
+                       .func = 0,
+                       .bytecount = bytecount,
+                       .ptr = (void *)kmalloc(bytecount, GFP_KERNEL)};
+       u32 cpu;
+
+       if(ptrace_ldt.ptr == NULL)
+               return -ENOMEM;
 
+       /* This is called from sys_modify_ldt only, so userspace_pid gives
+        * us the right number
+        */
+
+       cpu = get_cpu();
+       res = os_ptrace_ldt(userspace_pid[cpu], 0, (unsigned long) &ptrace_ldt);
+       put_cpu();
+       if(res < 0)
+               goto out;
+
+       n = copy_to_user(ptr, ptrace_ldt.ptr, res);
+       if(n != 0)
+               res = -EFAULT;
+
+  out:
+       kfree(ptrace_ldt.ptr);
+
+       return res;
+}
 
 /*
- * Overrides for Emacs so that we follow Linus's tabbing style.
- * Emacs will notice this stuff at the end of the file and automatically
- * adjust the settings for this buffer only.  This must remain at the end
- * of the file.
- * ---------------------------------------------------------------------------
- * Local variables:
- * c-file-style: "linux"
- * End:
+ * In skas mode, we hold our own ldt data in UML.
+ * Thus, the code implementing sys_modify_ldt_skas
+ * is very similar to (and mostly stolen from) sys_modify_ldt
+ * for arch/i386/kernel/ldt.c
+ * The routines copied and modified in part are:
+ * - read_ldt
+ * - read_default_ldt
+ * - write_ldt
+ * - sys_modify_ldt_skas
  */
+
+static int read_ldt(void __user * ptr, unsigned long bytecount)
+{
+       int i, err = 0;
+       unsigned long size;
+       uml_ldt_t * ldt = &current->mm->context.skas.ldt;
+
+       if(!ldt->entry_count)
+               goto out;
+       if(bytecount > LDT_ENTRY_SIZE*LDT_ENTRIES)
+               bytecount = LDT_ENTRY_SIZE*LDT_ENTRIES;
+       err = bytecount;
+
+       if(ptrace_ldt){
+               return read_ldt_from_host(ptr, bytecount);
+       }
+
+       down(&ldt->semaphore);
+       if(ldt->entry_count <= LDT_DIRECT_ENTRIES){
+               size = LDT_ENTRY_SIZE*LDT_DIRECT_ENTRIES;
+               if(size > bytecount)
+                       size = bytecount;
+               if(copy_to_user(ptr, ldt->u.entries, size))
+                       err = -EFAULT;
+               bytecount -= size;
+               ptr += size;
+       }
+       else {
+               for(i=0; i<ldt->entry_count/LDT_ENTRIES_PER_PAGE && bytecount;
+                        i++){
+                       size = PAGE_SIZE;
+                       if(size > bytecount)
+                               size = bytecount;
+                       if(copy_to_user(ptr, ldt->u.pages[i], size)){
+                               err = -EFAULT;
+                               break;
+                       }
+                       bytecount -= size;
+                       ptr += size;
+               }
+       }
+       up(&ldt->semaphore);
+
+       if(bytecount == 0 || err == -EFAULT)
+               goto out;
+
+       if(clear_user(ptr, bytecount))
+               err = -EFAULT;
+
+out:
+       return err;
+}
+
+static int read_default_ldt(void __user * ptr, unsigned long bytecount)
+{
+       int err;
+
+       if(bytecount > 5*LDT_ENTRY_SIZE)
+               bytecount = 5*LDT_ENTRY_SIZE;
+
+       err = bytecount;
+       /* UML doesn't support lcall7 and lcall27.
+        * So, we don't really have a default ldt, but emulate
+        * an empty ldt of common host default ldt size.
+        */
+       if(clear_user(ptr, bytecount))
+               err = -EFAULT;
+
+       return err;
+}
+
+static int write_ldt(void __user * ptr, unsigned long bytecount, int func)
+{
+       uml_ldt_t * ldt = &current->mm->context.skas.ldt;
+       struct mm_id * mm_idp = &current->mm->context.skas.id;
+       int i, err;
+       struct user_desc ldt_info;
+       struct ldt_entry entry0, *ldt_p;
+       void *addr = NULL;
+
+       err = -EINVAL;
+       if(bytecount != sizeof(ldt_info))
+               goto out;
+       err = -EFAULT;
+       if(copy_from_user(&ldt_info, ptr, sizeof(ldt_info)))
+               goto out;
+
+       err = -EINVAL;
+       if(ldt_info.entry_number >= LDT_ENTRIES)
+               goto out;
+       if(ldt_info.contents == 3){
+               if (func == 1)
+                       goto out;
+               if (ldt_info.seg_not_present == 0)
+                       goto out;
+       }
+
+        if(!ptrace_ldt)
+                down(&ldt->semaphore);
+
+       err = write_ldt_entry(mm_idp, func, &ldt_info, &addr, 1);
+       if(err)
+               goto out_unlock;
+        else if(ptrace_ldt) {
+       /* With PTRACE_LDT available, this is used as a flag only */
+                ldt->entry_count = 1;
+                goto out;
+        }
+
+       if(ldt_info.entry_number >= ldt->entry_count &&
+          ldt_info.entry_number >= LDT_DIRECT_ENTRIES){
+               for(i=ldt->entry_count/LDT_ENTRIES_PER_PAGE;
+                   i*LDT_ENTRIES_PER_PAGE <= ldt_info.entry_number;
+                   i++){
+                       if(i == 0)
+                               memcpy(&entry0, ldt->u.entries,
+                                      sizeof(entry0));
+                       ldt->u.pages[i] = (struct ldt_entry *)
+                               __get_free_page(GFP_KERNEL|__GFP_ZERO);
+                       if(!ldt->u.pages[i]){
+                               err = -ENOMEM;
+                               /* Undo the change in host */
+                               memset(&ldt_info, 0, sizeof(ldt_info));
+                               write_ldt_entry(mm_idp, 1, &ldt_info, &addr, 1);
+                               goto out_unlock;
+                       }
+                       if(i == 0) {
+                               memcpy(ldt->u.pages[0], &entry0,
+                                      sizeof(entry0));
+                               memcpy(ldt->u.pages[0]+1, ldt->u.entries+1,
+                                      sizeof(entry0)*(LDT_DIRECT_ENTRIES-1));
+                       }
+                       ldt->entry_count = (i + 1) * LDT_ENTRIES_PER_PAGE;
+               }
+       }
+       if(ldt->entry_count <= ldt_info.entry_number)
+               ldt->entry_count = ldt_info.entry_number + 1;
+
+       if(ldt->entry_count <= LDT_DIRECT_ENTRIES)
+               ldt_p = ldt->u.entries + ldt_info.entry_number;
+       else
+               ldt_p = ldt->u.pages[ldt_info.entry_number/LDT_ENTRIES_PER_PAGE] +
+                       ldt_info.entry_number%LDT_ENTRIES_PER_PAGE;
+
+       if(ldt_info.base_addr == 0 && ldt_info.limit == 0 &&
+          (func == 1 || LDT_empty(&ldt_info))){
+               ldt_p->a = 0;
+               ldt_p->b = 0;
+       }
+       else{
+               if (func == 1)
+                       ldt_info.useable = 0;
+               ldt_p->a = LDT_entry_a(&ldt_info);
+               ldt_p->b = LDT_entry_b(&ldt_info);
+       }
+       err = 0;
+
+out_unlock:
+       up(&ldt->semaphore);
+out:
+       return err;
+}
+
+static long do_modify_ldt_skas(int func, void __user *ptr,
+                              unsigned long bytecount)
+{
+       int ret = -ENOSYS;
+
+       switch (func) {
+               case 0:
+                       ret = read_ldt(ptr, bytecount);
+                       break;
+               case 1:
+               case 0x11:
+                       ret = write_ldt(ptr, bytecount, func);
+                       break;
+               case 2:
+                       ret = read_default_ldt(ptr, bytecount);
+                       break;
+       }
+       return ret;
+}
+
+short dummy_list[9] = {0, -1};
+short * host_ldt_entries = NULL;
+
+void ldt_get_host_info(void)
+{
+       long ret;
+       struct ldt_entry * ldt;
+       int i, size, k, order;
+
+       host_ldt_entries = dummy_list+1;
+
+       for(i = LDT_PAGES_MAX-1, order=0; i; i>>=1, order++);
+
+       ldt = (struct ldt_entry *)
+             __get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
+       if(ldt == NULL) {
+               printk("ldt_get_host_info: couldn't allocate buffer for host ldt\n");
+               return;
+       }
+
+       ret = modify_ldt(0, ldt, (1<<order)*PAGE_SIZE);
+       if(ret < 0) {
+               printk("ldt_get_host_info: couldn't read host ldt\n");
+               goto out_free;
+       }
+       if(ret == 0) {
+               /* default_ldt is active, simply write an empty entry 0 */
+               host_ldt_entries = dummy_list;
+               goto out_free;
+       }
+
+       for(i=0, size=0; i<ret/LDT_ENTRY_SIZE; i++){
+               if(ldt[i].a != 0 || ldt[i].b != 0)
+                       size++;
+       }
+
+       if(size < sizeof(dummy_list)/sizeof(dummy_list[0])) {
+               host_ldt_entries = dummy_list;
+       }
+       else {
+               size = (size + 1) * sizeof(dummy_list[0]);
+               host_ldt_entries = (short *)kmalloc(size, GFP_KERNEL);
+               if(host_ldt_entries == NULL) {
+                       printk("ldt_get_host_info: couldn't allocate host ldt list\n");
+                       goto out_free;
+               }
+       }
+
+       for(i=0, k=0; i<ret/LDT_ENTRY_SIZE; i++){
+               if(ldt[i].a != 0 || ldt[i].b != 0) {
+                       host_ldt_entries[k++] = i;
+               }
+       }
+       host_ldt_entries[k] = -1;
+
+out_free:
+       free_pages((unsigned long)ldt, order);
+}
+
+long init_new_ldt(struct mmu_context_skas * new_mm,
+                 struct mmu_context_skas * from_mm)
+{
+       struct user_desc desc;
+       short * num_p;
+       int i;
+       long page, err=0;
+       void *addr = NULL;
+       struct proc_mm_op copy;
+
+
+       if(!ptrace_ldt)
+               init_MUTEX(&new_mm->ldt.semaphore);
+
+       if(!from_mm){
+               memset(&desc, 0, sizeof(desc));
+               /*
+                * We have to initialize a clean ldt.
+                */
+               if(proc_mm) {
+                       /*
+                        * If the new mm was created using proc_mm, host's
+                        * default-ldt currently is assigned, which normally
+                        * contains the call-gates for lcall7 and lcall27.
+                        * To remove these gates, we simply write an empty
+                        * entry as number 0 to the host.
+                        */
+                       err = write_ldt_entry(&new_mm->id, 1, &desc,
+                                             &addr, 1);
+               }
+               else{
+                       /*
+                        * Now we try to retrieve info about the ldt, we
+                        * inherited from the host. All ldt-entries found
+                        * will be reset in the following loop
+                        */
+                       if(host_ldt_entries == NULL)
+                               ldt_get_host_info();
+                       for(num_p=host_ldt_entries; *num_p != -1; num_p++){
+                               desc.entry_number = *num_p;
+                               err = write_ldt_entry(&new_mm->id, 1, &desc,
+                                                     &addr, *(num_p + 1) == -1);
+                               if(err)
+                                       break;
+                       }
+               }
+               new_mm->ldt.entry_count = 0;
+
+               goto out;
+       }
+
+       if(proc_mm){
+               /* We have a valid from_mm, so we now have to copy the LDT of
+                * from_mm to new_mm, because using proc_mm an new mm with
+                * an empty/default LDT was created in new_mm()
+                */
+               copy = ((struct proc_mm_op) { .op       = MM_COPY_SEGMENTS,
+                                             .u        =
+                                             { .copy_segments =
+                                                       from_mm->id.u.mm_fd } } );
+               i = os_write_file(new_mm->id.u.mm_fd, &copy, sizeof(copy));
+               if(i != sizeof(copy))
+                       printk("new_mm : /proc/mm copy_segments failed, "
+                              "err = %d\n", -i);
+       }
+
+       if(!ptrace_ldt) {
+               /* Our local LDT is used to supply the data for
+                * modify_ldt(READLDT), if PTRACE_LDT isn't available,
+                * i.e., we have to use the stub for modify_ldt, which
+                * can't handle the big read buffer of up to 64kB.
+                */
+               down(&from_mm->ldt.semaphore);
+               if(from_mm->ldt.entry_count <= LDT_DIRECT_ENTRIES){
+                       memcpy(new_mm->ldt.u.entries, from_mm->ldt.u.entries,
+                              sizeof(new_mm->ldt.u.entries));
+               }
+               else{
+                       i = from_mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
+                       while(i-->0){
+                               page = __get_free_page(GFP_KERNEL|__GFP_ZERO);
+                               if (!page){
+                                       err = -ENOMEM;
+                                       break;
+                               }
+                               new_mm->ldt.u.pages[i] =
+                                       (struct ldt_entry *) page;
+                               memcpy(new_mm->ldt.u.pages[i],
+                                      from_mm->ldt.u.pages[i], PAGE_SIZE);
+                       }
+               }
+               new_mm->ldt.entry_count = from_mm->ldt.entry_count;
+               up(&from_mm->ldt.semaphore);
+       }
+
+    out:
+       return err;
+}
+
+
+void free_ldt(struct mmu_context_skas * mm)
+{
+       int i;
+
+       if(!ptrace_ldt && mm->ldt.entry_count > LDT_DIRECT_ENTRIES){
+               i = mm->ldt.entry_count / LDT_ENTRIES_PER_PAGE;
+               while(i-- > 0){
+                       free_page((long )mm->ldt.u.pages[i]);
+               }
+       }
+       mm->ldt.entry_count = 0;
+}
+#endif
+
+int sys_modify_ldt(int func, void __user *ptr, unsigned long bytecount)
+{
+       return(CHOOSE_MODE_PROC(do_modify_ldt_tt, do_modify_ldt_skas, func,
+                               ptr, bytecount));
+}