mm/mmap.c

/* */
This source file includes following definitions.
do_mmap
get_unmapped_area
find_vma
find_vma_intersection
avl_neighbours
avl_rebalance
avl_insert
avl_insert_neighbours
avl_remove
printk_list
printk_avl
avl_checkheights
avl_checkleft
avl_checkright
avl_checkorder
avl_check
unmap_fixup
sys_munmap
do_munmap
build_mmap_avl
exit_mmap
insert_vm_struct
remove_shared_vm_struct
merge_segments
anon_map
   1 /*
   2  *      linux/mm/mmap.c
   3  *
   4  * Written by obz.
   5  */
   6 #include <linux/stat.h>
   7 #include <linux/sched.h>
   8 #include <linux/kernel.h>
   9 #include <linux/mm.h>
  10 #include <linux/shm.h>
  11 #include <linux/errno.h>
  12 #include <linux/mman.h>
  13 #include <linux/string.h>
  14 #include <linux/malloc.h>
  15 
  16 #include <asm/segment.h>
  17 #include <asm/system.h>
  18 #include <asm/pgtable.h>
  19 
  20 static int anon_map(struct inode *, struct file *, struct vm_area_struct *);
  21 
  22 /*
  23  * description of effects of mapping type and prot in current implementation.
  24  * this is due to the limited x86 page protection hardware.  The expected
  25  * behavior is in parens:
  26  *
  27  * map_type     prot
  28  *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
  29  * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  30  *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
  31  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  32  *              
  33  * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  34  *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
  35  *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  36  *
  37  */
  38 
  39 pgprot_t protection_map[16] = {
  40         __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
  41         __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
  42 };
  43 
  44 unsigned long do_mmap(struct file * file, unsigned long addr, unsigned long len,
     /*  */
  45         unsigned long prot, unsigned long flags, unsigned long off)
  46 {
  47         int error;
  48         struct vm_area_struct * vma;
  49 
  50         if ((len = PAGE_ALIGN(len)) == 0)
  51                 return addr;
  52 
  53         if (addr > TASK_SIZE || len > TASK_SIZE || addr > TASK_SIZE-len)
  54                 return -EINVAL;
  55 
  56         /* offset overflow? */
  57         if (off + len < off)
  58                 return -EINVAL;
  59 
  60         /*
  61          * do simple checking here so the lower-level routines won't have
  62          * to. we assume access permissions have been handled by the open
  63          * of the memory object, so we don't do any here.
  64          */
  65 
  66         if (file != NULL) {
  67                 switch (flags & MAP_TYPE) {
  68                 case MAP_SHARED:
  69                         if ((prot & PROT_WRITE) && !(file->f_mode & 2))
  70                                 return -EACCES;
  71                         /* fall through */
  72                 case MAP_PRIVATE:
  73                         if (!(file->f_mode & 1))
  74                                 return -EACCES;
  75                         break;
  76 
  77                 default:
  78                         return -EINVAL;
  79                 }
  80                 if ((flags & MAP_DENYWRITE) && (file->f_inode->i_wcount > 0))
  81                         return -ETXTBSY;
  82         } else if ((flags & MAP_TYPE) != MAP_PRIVATE)
  83                 return -EINVAL;
  84 
  85         /*
  86          * obtain the address to map to. we verify (or select) it and ensure
  87          * that it represents a valid section of the address space.
  88          */
  89 
  90         if (flags & MAP_FIXED) {
  91                 if (addr & ~PAGE_MASK)
  92                         return -EINVAL;
  93                 if (len > TASK_SIZE || addr > TASK_SIZE - len)
  94                         return -EINVAL;
  95         } else {
  96                 addr = get_unmapped_area(addr, len);
  97                 if (!addr)
  98                         return -ENOMEM;
  99         }
 100 
 101         /*
 102          * determine the object being mapped and call the appropriate
 103          * specific mapper. the address has already been validated, but
 104          * not unmapped, but the maps are removed from the list.
 105          */
 106         if (file && (!file->f_op || !file->f_op->mmap))
 107                 return -ENODEV;
 108 
 109         vma = (struct vm_area_struct *)kmalloc(sizeof(struct vm_area_struct),
 110                 GFP_KERNEL);
 111         if (!vma)
 112                 return -ENOMEM;
 113 
 114         vma->vm_task = current;
 115         vma->vm_start = addr;
 116         vma->vm_end = addr + len;
 117         vma->vm_flags = prot & (VM_READ | VM_WRITE | VM_EXEC);
 118         vma->vm_flags |= flags & (VM_GROWSDOWN | VM_DENYWRITE | VM_EXECUTABLE);
 119 
 120         if (file) {
 121                 if (file->f_mode & 1)
 122                         vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 123                 if (flags & MAP_SHARED) {
 124                         vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
 125                         /*
 126                          * This looks strange, but when we don't have the file open
 127                          * for writing, we can demote the shared mapping to a simpler
 128                          * private mapping. That also takes care of a security hole
 129                          * with ptrace() writing to a shared mapping without write
 130                          * permissions.
 131                          *
 132                          * We leave the VM_MAYSHARE bit on, just to get correct output
 133                          * from /proc/xxx/maps..
 134                          */
 135                         if (!(file->f_mode & 2))
 136                                 vma->vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
 137                 }
 138         } else
 139                 vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 140         vma->vm_page_prot = protection_map[vma->vm_flags & 0x0f];
 141         vma->vm_ops = NULL;
 142         vma->vm_offset = off;
 143         vma->vm_inode = NULL;
 144         vma->vm_pte = 0;
 145 
 146         do_munmap(addr, len);   /* Clear old maps */
 147 
 148         if (file)
 149                 error = file->f_op->mmap(file->f_inode, file, vma);
 150         else
 151                 error = anon_map(NULL, NULL, vma);
 152         
 153         if (error) {
 154                 kfree(vma);
 155                 return error;
 156         }
 157         insert_vm_struct(current, vma);
 158         merge_segments(current, vma->vm_start, vma->vm_end);
 159         return addr;
 160 }
 161 
 162 /*
 163  * Get an address range which is currently unmapped.
 164  * For mmap() without MAP_FIXED and shmat() with addr=0.
 165  * Return value 0 means ENOMEM.
 166  */
 167 unsigned long get_unmapped_area(unsigned long addr, unsigned long len)
     /*  */
 168 {
 169         struct vm_area_struct * vmm;
 170 
 171         if (len > TASK_SIZE)
 172                 return 0;
 173         if (!addr)
 174                 addr = TASK_SIZE / 3;
 175         addr = PAGE_ALIGN(addr);
 176 
 177         for (vmm = current->mm->mmap; ; vmm = vmm->vm_next) {
 178                 if (TASK_SIZE - len < addr)
 179                         return 0;
 180                 if (!vmm)
 181                         return addr;
 182                 if (addr > vmm->vm_end)
 183                         continue;
 184                 if (addr + len > vmm->vm_start) {
 185                         addr = vmm->vm_end;
 186                         continue;
 187                 }
 188                 return addr;
 189         }
 190 }
 191 
 192 /*
 193  * Searching a VMA in the linear list task->mm->mmap is horribly slow.
 194  * Use an AVL (Adelson-Velskii and Landis) tree to speed up this search
 195  * from O(n) to O(log n), where n is the number of VMAs of the task
 196  * (typically around 6, but may reach 3000 in some cases).
 197  * Written by Bruno Haible <haible@ma2s2.mathematik.uni-karlsruhe.de>.
 198  */
 199 
 200 /* We keep the list and tree sorted by address. */
 201 #define vm_avl_key      vm_end
 202 #define vm_avl_key_t    unsigned long   /* typeof(vma->avl_key) */
 203 
 204 /*
 205  * task->mm->mmap_avl is the AVL tree corresponding to task->mm->mmap
 206  * or, more exactly, its root.
 207  * A vm_area_struct has the following fields:
 208  *   vm_avl_left     left son of a tree node
 209  *   vm_avl_right    right son of a tree node
 210  *   vm_avl_height   1+max(heightof(left),heightof(right))
 211  * The empty tree is represented as NULL.
 212  */
 213 #define avl_empty       (struct vm_area_struct *) NULL
 214 
 215 /* Since the trees are balanced, their height will never be large. */
 216 #define avl_maxheight   41      /* why this? a small exercise */
 217 #define heightof(tree)  ((tree) == avl_empty ? 0 : (tree)->vm_avl_height)
 218 /*
 219  * Consistency and balancing rules:
 220  * 1. tree->vm_avl_height == 1+max(heightof(tree->vm_avl_left),heightof(tree->vm_avl_right))
 221  * 2. abs( heightof(tree->vm_avl_left) - heightof(tree->vm_avl_right) ) <= 1
 222  * 3. foreach node in tree->vm_avl_left: node->vm_avl_key <= tree->vm_avl_key,
 223  *    foreach node in tree->vm_avl_right: node->vm_avl_key >= tree->vm_avl_key.
 224  */
 225 
 226 /* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
 227 struct vm_area_struct * find_vma (struct task_struct * task, unsigned long addr)
     /*  */
 228 {
 229 #if 0 /* equivalent, but slow */
 230         struct vm_area_struct * vma;
 231 
 232         for (vma = task->mm->mmap ; ; vma = vma->vm_next) {
 233                 if (!vma)
 234                         return NULL;
 235                 if (vma->vm_end > addr)
 236                         return vma;
 237         }
 238 #else
 239         struct vm_area_struct * result = NULL;
 240         struct vm_area_struct * tree;
 241 
 242         for (tree = task->mm->mmap_avl ; ; ) {
 243                 if (tree == avl_empty)
 244                         return result;
 245                 if (tree->vm_end > addr) {
 246                         if (tree->vm_start <= addr)
 247                                 return tree;
 248                         result = tree;
 249                         tree = tree->vm_avl_left;
 250                 } else
 251                         tree = tree->vm_avl_right;
 252         }
 253 #endif
 254 }
 255 
 256 /* Look up the first VMA which intersects the interval start_addr..end_addr-1,
 257    NULL if none.  Assume start_addr < end_addr. */
 258 struct vm_area_struct * find_vma_intersection (struct task_struct * task, unsigned long start_addr, unsigned long end_addr)
     /*  */
 259 {
 260         struct vm_area_struct * vma;
 261 
 262 #if 0 /* equivalent, but slow */
 263         for (vma = task->mm->mmap; vma; vma = vma->vm_next) {
 264                 if (end_addr <= vma->vm_start)
 265                         break;
 266                 if (start_addr < vma->vm_end)
 267                         return vma;
 268         }
 269         return NULL;
 270 #else
 271         vma = find_vma(task,start_addr);
 272         if (!vma || end_addr <= vma->vm_start)
 273                 return NULL;
 274         return vma;
 275 #endif
 276 }
 277 
 278 /* Look up the nodes at the left and at the right of a given node. */
 279 static void avl_neighbours (struct vm_area_struct * node, struct vm_area_struct * tree, struct vm_area_struct ** to_the_left, struct vm_area_struct ** to_the_right)
     /*  */
 280 {
 281         vm_avl_key_t key = node->vm_avl_key;
 282 
 283         *to_the_left = *to_the_right = NULL;
 284         for (;;) {
 285                 if (tree == avl_empty) {
 286                         printk("avl_neighbours: node not found in the tree\n");
 287                         return;
 288                 }
 289                 if (key == tree->vm_avl_key)
 290                         break;
 291                 if (key < tree->vm_avl_key) {
 292                         *to_the_right = tree;
 293                         tree = tree->vm_avl_left;
 294                 } else {
 295                         *to_the_left = tree;
 296                         tree = tree->vm_avl_right;
 297                 }
 298         }
 299         if (tree != node) {
 300                 printk("avl_neighbours: node not exactly found in the tree\n");
 301                 return;
 302         }
 303         if (tree->vm_avl_left != avl_empty) {
 304                 struct vm_area_struct * node;
 305                 for (node = tree->vm_avl_left; node->vm_avl_right != avl_empty; node = node->vm_avl_right)
 306                         continue;
 307                 *to_the_left = node;
 308         }
 309         if (tree->vm_avl_right != avl_empty) {
 310                 struct vm_area_struct * node;
 311                 for (node = tree->vm_avl_right; node->vm_avl_left != avl_empty; node = node->vm_avl_left)
 312                         continue;
 313                 *to_the_right = node;
 314         }
 315         if ((*to_the_left && ((*to_the_left)->vm_next != node)) || (node->vm_next != *to_the_right))
 316                 printk("avl_neighbours: tree inconsistent with list\n");
 317 }
 318 
 319 /*
 320  * Rebalance a tree.
 321  * After inserting or deleting a node of a tree we have a sequence of subtrees
 322  * nodes[0]..nodes[k-1] such that
 323  * nodes[0] is the root and nodes[i+1] = nodes[i]->{vm_avl_left|vm_avl_right}.
 324  */
 325 static void avl_rebalance (struct vm_area_struct *** nodeplaces_ptr, int count)
     /*  */
 326 {
 327         for ( ; count > 0 ; count--) {
 328                 struct vm_area_struct ** nodeplace = *--nodeplaces_ptr;
 329                 struct vm_area_struct * node = *nodeplace;
 330                 struct vm_area_struct * nodeleft = node->vm_avl_left;
 331                 struct vm_area_struct * noderight = node->vm_avl_right;
 332                 int heightleft = heightof(nodeleft);
 333                 int heightright = heightof(noderight);
 334                 if (heightright + 1 < heightleft) {
 335                         /*                                                      */
 336                         /*                            *                         */
 337                         /*                          /   \                       */
 338                         /*                       n+2      n                     */
 339                         /*                                                      */
 340                         struct vm_area_struct * nodeleftleft = nodeleft->vm_avl_left;
 341                         struct vm_area_struct * nodeleftright = nodeleft->vm_avl_right;
 342                         int heightleftright = heightof(nodeleftright);
 343                         if (heightof(nodeleftleft) >= heightleftright) {
 344                                 /*                                                        */
 345                                 /*                *                    n+2|n+3            */
 346                                 /*              /   \                  /    \             */
 347                                 /*           n+2      n      -->      /   n+1|n+2         */
 348                                 /*           / \                      |    /    \         */
 349                                 /*         n+1 n|n+1                 n+1  n|n+1  n        */
 350                                 /*                                                        */
 351                                 node->vm_avl_left = nodeleftright; nodeleft->vm_avl_right = node;
 352                                 nodeleft->vm_avl_height = 1 + (node->vm_avl_height = 1 + heightleftright);
 353                                 *nodeplace = nodeleft;
 354                         } else {
 355                                 /*                                                        */
 356                                 /*                *                     n+2               */
 357                                 /*              /   \                 /     \             */
 358                                 /*           n+2      n      -->    n+1     n+1           */
 359                                 /*           / \                    / \     / \           */
 360                                 /*          n  n+1                 n   L   R   n          */
 361                                 /*             / \                                        */
 362                                 /*            L   R                                       */
 363                                 /*                                                        */
 364                                 nodeleft->vm_avl_right = nodeleftright->vm_avl_left;
 365                                 node->vm_avl_left = nodeleftright->vm_avl_right;
 366                                 nodeleftright->vm_avl_left = nodeleft;
 367                                 nodeleftright->vm_avl_right = node;
 368                                 nodeleft->vm_avl_height = node->vm_avl_height = heightleftright;
 369                                 nodeleftright->vm_avl_height = heightleft;
 370                                 *nodeplace = nodeleftright;
 371                         }
 372                 }
 373                 else if (heightleft + 1 < heightright) {
 374                         /* similar to the above, just interchange 'left' <--> 'right' */
 375                         struct vm_area_struct * noderightright = noderight->vm_avl_right;
 376                         struct vm_area_struct * noderightleft = noderight->vm_avl_left;
 377                         int heightrightleft = heightof(noderightleft);
 378                         if (heightof(noderightright) >= heightrightleft) {
 379                                 node->vm_avl_right = noderightleft; noderight->vm_avl_left = node;
 380                                 noderight->vm_avl_height = 1 + (node->vm_avl_height = 1 + heightrightleft);
 381                                 *nodeplace = noderight;
 382                         } else {
 383                                 noderight->vm_avl_left = noderightleft->vm_avl_right;
 384                                 node->vm_avl_right = noderightleft->vm_avl_left;
 385                                 noderightleft->vm_avl_right = noderight;
 386                                 noderightleft->vm_avl_left = node;
 387                                 noderight->vm_avl_height = node->vm_avl_height = heightrightleft;
 388                                 noderightleft->vm_avl_height = heightright;
 389                                 *nodeplace = noderightleft;
 390                         }
 391                 }
 392                 else {
 393                         int height = (heightleft<heightright ? heightright : heightleft) + 1;
 394                         if (height == node->vm_avl_height)
 395                                 break;
 396                         node->vm_avl_height = height;
 397                 }
 398         }
 399 }
 400 
 401 /* Insert a node into a tree. */
 402 static void avl_insert (struct vm_area_struct * new_node, struct vm_area_struct ** ptree)
     /*  */
 403 {
 404         vm_avl_key_t key = new_node->vm_avl_key;
 405         struct vm_area_struct ** nodeplace = ptree;
 406         struct vm_area_struct ** stack[avl_maxheight];
 407         int stack_count = 0;
 408         struct vm_area_struct *** stack_ptr = &stack[0]; /* = &stack[stackcount] */
 409         for (;;) {
 410                 struct vm_area_struct * node = *nodeplace;
 411                 if (node == avl_empty)
 412                         break;
 413                 *stack_ptr++ = nodeplace; stack_count++;
 414                 if (key < node->vm_avl_key)
 415                         nodeplace = &node->vm_avl_left;
 416                 else
 417                         nodeplace = &node->vm_avl_right;
 418         }
 419         new_node->vm_avl_left = avl_empty;
 420         new_node->vm_avl_right = avl_empty;
 421         new_node->vm_avl_height = 1;
 422         *nodeplace = new_node;
 423         avl_rebalance(stack_ptr,stack_count);
 424 }
 425 
 426 /* Insert a node into a tree, and
 427  * return the node to the left of it and the node to the right of it.
 428  */
 429 static void avl_insert_neighbours (struct vm_area_struct * new_node, struct vm_area_struct ** ptree,
     /*  */
 430         struct vm_area_struct ** to_the_left, struct vm_area_struct ** to_the_right)
 431 {
 432         vm_avl_key_t key = new_node->vm_avl_key;
 433         struct vm_area_struct ** nodeplace = ptree;
 434         struct vm_area_struct ** stack[avl_maxheight];
 435         int stack_count = 0;
 436         struct vm_area_struct *** stack_ptr = &stack[0]; /* = &stack[stackcount] */
 437         *to_the_left = *to_the_right = NULL;
 438         for (;;) {
 439                 struct vm_area_struct * node = *nodeplace;
 440                 if (node == avl_empty)
 441                         break;
 442                 *stack_ptr++ = nodeplace; stack_count++;
 443                 if (key < node->vm_avl_key) {
 444                         *to_the_right = node;
 445                         nodeplace = &node->vm_avl_left;
 446                 } else {
 447                         *to_the_left = node;
 448                         nodeplace = &node->vm_avl_right;
 449                 }
 450         }
 451         new_node->vm_avl_left = avl_empty;
 452         new_node->vm_avl_right = avl_empty;
 453         new_node->vm_avl_height = 1;
 454         *nodeplace = new_node;
 455         avl_rebalance(stack_ptr,stack_count);
 456 }
 457 
 458 /* Removes a node out of a tree. */
 459 static void avl_remove (struct vm_area_struct * node_to_delete, struct vm_area_struct ** ptree)
     /*  */
 460 {
 461         vm_avl_key_t key = node_to_delete->vm_avl_key;
 462         struct vm_area_struct ** nodeplace = ptree;
 463         struct vm_area_struct ** stack[avl_maxheight];
 464         int stack_count = 0;
 465         struct vm_area_struct *** stack_ptr = &stack[0]; /* = &stack[stackcount] */
 466         struct vm_area_struct ** nodeplace_to_delete;
 467         for (;;) {
 468                 struct vm_area_struct * node = *nodeplace;
 469                 if (node == avl_empty) {
 470                         /* what? node_to_delete not found in tree? */
 471                         printk("avl_remove: node to delete not found in tree\n");
 472                         return;
 473                 }
 474                 *stack_ptr++ = nodeplace; stack_count++;
 475                 if (key == node->vm_avl_key)
 476                         break;
 477                 if (key < node->vm_avl_key)
 478                         nodeplace = &node->vm_avl_left;
 479                 else
 480                         nodeplace = &node->vm_avl_right;
 481         }
 482         nodeplace_to_delete = nodeplace;
 483         /* Have to remove node_to_delete = *nodeplace_to_delete. */
 484         if (node_to_delete->vm_avl_left == avl_empty) {
 485                 *nodeplace_to_delete = node_to_delete->vm_avl_right;
 486                 stack_ptr--; stack_count--;
 487         } else {
 488                 struct vm_area_struct *** stack_ptr_to_delete = stack_ptr;
 489                 struct vm_area_struct ** nodeplace = &node_to_delete->vm_avl_left;
 490                 struct vm_area_struct * node;
 491                 for (;;) {
 492                         node = *nodeplace;
 493                         if (node->vm_avl_right == avl_empty)
 494                                 break;
 495                         *stack_ptr++ = nodeplace; stack_count++;
 496                         nodeplace = &node->vm_avl_right;
 497                 }
 498                 *nodeplace = node->vm_avl_left;
 499                 /* node replaces node_to_delete */
 500                 node->vm_avl_left = node_to_delete->vm_avl_left;
 501                 node->vm_avl_right = node_to_delete->vm_avl_right;
 502                 node->vm_avl_height = node_to_delete->vm_avl_height;
 503                 *nodeplace_to_delete = node; /* replace node_to_delete */
 504                 *stack_ptr_to_delete = &node->vm_avl_left; /* replace &node_to_delete->vm_avl_left */
 505         }
 506         avl_rebalance(stack_ptr,stack_count);
 507 }
 508 
 509 #ifdef DEBUG_AVL
 510 
 511 /* print a list */
 512 static void printk_list (struct vm_area_struct * vma)
     /*  */
 513 {
 514         printk("[");
 515         while (vma) {
 516                 printk("%08lX-%08lX", vma->vm_start, vma->vm_end);
 517                 vma = vma->vm_next;
 518                 if (!vma)
 519                         break;
 520                 printk(" ");
 521         }
 522         printk("]");
 523 }
 524 
 525 /* print a tree */
 526 static void printk_avl (struct vm_area_struct * tree)
     /*  */
 527 {
 528         if (tree != avl_empty) {
 529                 printk("(");
 530                 if (tree->vm_avl_left != avl_empty) {
 531                         printk_avl(tree->vm_avl_left);
 532                         printk("<");
 533                 }
 534                 printk("%08lX-%08lX", tree->vm_start, tree->vm_end);
 535                 if (tree->vm_avl_right != avl_empty) {
 536                         printk(">");
 537                         printk_avl(tree->vm_avl_right);
 538                 }
 539                 printk(")");
 540         }
 541 }
 542 
 543 static char *avl_check_point = "somewhere";
 544 
 545 /* check a tree's consistency and balancing */
 546 static void avl_checkheights (struct vm_area_struct * tree)
     /*  */
 547 {
 548         int h, hl, hr;
 549 
 550         if (tree == avl_empty)
 551                 return;
 552         avl_checkheights(tree->vm_avl_left);
 553         avl_checkheights(tree->vm_avl_right);
 554         h = tree->vm_avl_height;
 555         hl = heightof(tree->vm_avl_left);
 556         hr = heightof(tree->vm_avl_right);
 557         if ((h == hl+1) && (hr <= hl) && (hl <= hr+1))
 558                 return;
 559         if ((h == hr+1) && (hl <= hr) && (hr <= hl+1))
 560                 return;
 561         printk("%s: avl_checkheights: heights inconsistent\n",avl_check_point);
 562 }
 563 
 564 /* check that all values stored in a tree are < key */
 565 static void avl_checkleft (struct vm_area_struct * tree, vm_avl_key_t key)
     /*  */
 566 {
 567         if (tree == avl_empty)
 568                 return;
 569         avl_checkleft(tree->vm_avl_left,key);
 570         avl_checkleft(tree->vm_avl_right,key);
 571         if (tree->vm_avl_key < key)
 572                 return;
 573         printk("%s: avl_checkleft: left key %lu >= top key %lu\n",avl_check_point,tree->vm_avl_key,key);
 574 }
 575 
 576 /* check that all values stored in a tree are > key */
 577 static void avl_checkright (struct vm_area_struct * tree, vm_avl_key_t key)
     /*  */
 578 {
 579         if (tree == avl_empty)
 580                 return;
 581         avl_checkright(tree->vm_avl_left,key);
 582         avl_checkright(tree->vm_avl_right,key);
 583         if (tree->vm_avl_key > key)
 584                 return;
 585         printk("%s: avl_checkright: right key %lu <= top key %lu\n",avl_check_point,tree->vm_avl_key,key);
 586 }
 587 
 588 /* check that all values are properly increasing */
 589 static void avl_checkorder (struct vm_area_struct * tree)
     /*  */
 590 {
 591         if (tree == avl_empty)
 592                 return;
 593         avl_checkorder(tree->vm_avl_left);
 594         avl_checkorder(tree->vm_avl_right);
 595         avl_checkleft(tree->vm_avl_left,tree->vm_avl_key);
 596         avl_checkright(tree->vm_avl_right,tree->vm_avl_key);
 597 }
 598 
 599 /* all checks */
 600 static void avl_check (struct task_struct * task, char *caller)
     /*  */
 601 {
 602         avl_check_point = caller;
 603 /*      printk("task \"%s\", %s\n",task->comm,caller); */
 604 /*      printk("task \"%s\" list: ",task->comm); printk_list(task->mm->mmap); printk("\n"); */
 605 /*      printk("task \"%s\" tree: ",task->comm); printk_avl(task->mm->mmap_avl); printk("\n"); */
 606         avl_checkheights(task->mm->mmap_avl);
 607         avl_checkorder(task->mm->mmap_avl);
 608 }
 609 
 610 #endif
 611 
 612 
 613 /*
 614  * Normal function to fix up a mapping
 615  * This function is the default for when an area has no specific
 616  * function.  This may be used as part of a more specific routine.
 617  * This function works out what part of an area is affected and
 618  * adjusts the mapping information.  Since the actual page
 619  * manipulation is done in do_mmap(), none need be done here,
 620  * though it would probably be more appropriate.
 621  *
 622  * By the time this function is called, the area struct has been
 623  * removed from the process mapping list, so it needs to be
 624  * reinserted if necessary.
 625  *
 626  * The 4 main cases are:
 627  *    Unmapping the whole area
 628  *    Unmapping from the start of the segment to a point in it
 629  *    Unmapping from an intermediate point to the end
 630  *    Unmapping between to intermediate points, making a hole.
 631  *
 632  * Case 4 involves the creation of 2 new areas, for each side of
 633  * the hole.
 634  */
 635 void unmap_fixup(struct vm_area_struct *area,
     /*  */
 636                  unsigned long addr, size_t len)
 637 {
 638         struct vm_area_struct *mpnt;
 639         unsigned long end = addr + len;
 640 
 641         if (addr < area->vm_start || addr >= area->vm_end ||
 642             end <= area->vm_start || end > area->vm_end ||
 643             end < addr)
 644         {
 645                 printk("unmap_fixup: area=%lx-%lx, unmap %lx-%lx!!\n",
 646                        area->vm_start, area->vm_end, addr, end);
 647                 return;
 648         }
 649 
 650         /* Unmapping the whole area */
 651         if (addr == area->vm_start && end == area->vm_end) {
 652                 if (area->vm_ops && area->vm_ops->close)
 653                         area->vm_ops->close(area);
 654                 if (area->vm_inode)
 655                         iput(area->vm_inode);
 656                 return;
 657         }
 658 
 659         /* Work out to one of the ends */
 660         if (end == area->vm_end)
 661                 area->vm_end = addr;
 662         else
 663         if (addr == area->vm_start) {
 664                 area->vm_offset += (end - area->vm_start);
 665                 area->vm_start = end;
 666         }
 667         else {
 668         /* Unmapping a hole: area->vm_start < addr <= end < area->vm_end */
 669                 /* Add end mapping -- leave beginning for below */
 670                 mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 671 
 672                 if (!mpnt)
 673                         return;
 674                 *mpnt = *area;
 675                 mpnt->vm_offset += (end - area->vm_start);
 676                 mpnt->vm_start = end;
 677                 if (mpnt->vm_inode)
 678                         mpnt->vm_inode->i_count++;
 679                 if (mpnt->vm_ops && mpnt->vm_ops->open)
 680                         mpnt->vm_ops->open(mpnt);
 681                 area->vm_end = addr;    /* Truncate area */
 682                 insert_vm_struct(current, mpnt);
 683         }
 684 
 685         /* construct whatever mapping is needed */
 686         mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
 687         if (!mpnt)
 688                 return;
 689         *mpnt = *area;
 690         if (mpnt->vm_ops && mpnt->vm_ops->open)
 691                 mpnt->vm_ops->open(mpnt);
 692         if (area->vm_ops && area->vm_ops->close) {
 693                 area->vm_end = area->vm_start;
 694                 area->vm_ops->close(area);
 695         }
 696         insert_vm_struct(current, mpnt);
 697 }
 698 
 699 asmlinkage int sys_munmap(unsigned long addr, size_t len)
     /*  */
 700 {
 701         return do_munmap(addr, len);
 702 }
 703 
 704 /*
 705  * Munmap is split into 2 main parts -- this part which finds
 706  * what needs doing, and the areas themselves, which do the
 707  * work.  This now handles partial unmappings.
 708  * Jeremy Fitzhardine <jeremy@sw.oz.au>
 709  */
 710 int do_munmap(unsigned long addr, size_t len)
     /*  */
 711 {
 712         struct vm_area_struct *mpnt, *prev, *next, **npp, *free;
 713 
 714         if ((addr & ~PAGE_MASK) || addr > TASK_SIZE || len > TASK_SIZE-addr)
 715                 return -EINVAL;
 716 
 717         if ((len = PAGE_ALIGN(len)) == 0)
 718                 return 0;
 719 
 720         /*
 721          * Check if this memory area is ok - put it on the temporary
 722          * list if so..  The checks here are pretty simple --
 723          * every area affected in some way (by any overlap) is put
 724          * on the list.  If nothing is put on, nothing is affected.
 725          */
 726         mpnt = find_vma(current, addr);
 727         if (!mpnt)
 728                 return 0;
 729         avl_neighbours(mpnt, current->mm->mmap_avl, &prev, &next);
 730         /* we have  prev->vm_next == mpnt && mpnt->vm_next = next */
 731         /* and  addr < mpnt->vm_end  */
 732 
 733         npp = (prev ? &prev->vm_next : &current->mm->mmap);
 734         free = NULL;
 735         for ( ; mpnt && mpnt->vm_start < addr+len; mpnt = *npp) {
 736                 *npp = mpnt->vm_next;
 737                 mpnt->vm_next = free;
 738                 free = mpnt;
 739                 avl_remove(mpnt, &current->mm->mmap_avl);
 740         }
 741 
 742         if (free == NULL)
 743                 return 0;
 744 
 745         /*
 746          * Ok - we have the memory areas we should free on the 'free' list,
 747          * so release them, and unmap the page range..
 748          * If the one of the segments is only being partially unmapped,
 749          * it will put new vm_area_struct(s) into the address space.
 750          */
 751         while (free) {
 752                 unsigned long st, end;
 753 
 754                 mpnt = free;
 755                 free = free->vm_next;
 756 
 757                 remove_shared_vm_struct(mpnt);
 758 
 759                 st = addr < mpnt->vm_start ? mpnt->vm_start : addr;
 760                 end = addr+len;
 761                 end = end > mpnt->vm_end ? mpnt->vm_end : end;
 762 
 763                 if (mpnt->vm_ops && mpnt->vm_ops->unmap)
 764                         mpnt->vm_ops->unmap(mpnt, st, end-st);
 765 
 766                 unmap_fixup(mpnt, st, end-st);
 767                 kfree(mpnt);
 768         }
 769 
 770         unmap_page_range(addr, len);
 771         return 0;
 772 }
 773 
 774 /* Build the AVL tree corresponding to the VMA list. */
 775 void build_mmap_avl(struct task_struct * task)
     /*  */
 776 {
 777         struct vm_area_struct * vma;
 778 
 779         task->mm->mmap_avl = NULL;
 780         for (vma = task->mm->mmap; vma; vma = vma->vm_next)
 781                 avl_insert(vma, &task->mm->mmap_avl);
 782 }
 783 
 784 /* Release all mmaps. */
 785 void exit_mmap(struct task_struct * task)
     /*  */
 786 {
 787         struct vm_area_struct * mpnt;
 788 
 789         mpnt = task->mm->mmap;
 790         task->mm->mmap = NULL;
 791         task->mm->mmap_avl = NULL;
 792         while (mpnt) {
 793                 struct vm_area_struct * next = mpnt->vm_next;
 794                 if (mpnt->vm_ops && mpnt->vm_ops->close)
 795                         mpnt->vm_ops->close(mpnt);
 796                 remove_shared_vm_struct(mpnt);
 797                 if (mpnt->vm_inode)
 798                         iput(mpnt->vm_inode);
 799                 kfree(mpnt);
 800                 mpnt = next;
 801         }
 802 }
 803 
 804 /*
 805  * Insert vm structure into process list sorted by address
 806  * and into the inode's i_mmap ring.
 807  */
 808 void insert_vm_struct(struct task_struct *t, struct vm_area_struct *vmp)
     /*  */
 809 {
 810         struct vm_area_struct *share;
 811         struct inode * inode;
 812 
 813 #if 0 /* equivalent, but slow */
 814         struct vm_area_struct **p, *mpnt;
 815 
 816         p = &t->mm->mmap;
 817         while ((mpnt = *p) != NULL) {
 818                 if (mpnt->vm_start > vmp->vm_start)
 819                         break;
 820                 if (mpnt->vm_end > vmp->vm_start)
 821                         printk("insert_vm_struct: overlapping memory areas\n");
 822                 p = &mpnt->vm_next;
 823         }
 824         vmp->vm_next = mpnt;
 825         *p = vmp;
 826 #else
 827         struct vm_area_struct * prev, * next;
 828 
 829         avl_insert_neighbours(vmp, &t->mm->mmap_avl, &prev, &next);
 830         if ((prev ? prev->vm_next : t->mm->mmap) != next)
 831                 printk("insert_vm_struct: tree inconsistent with list\n");
 832         if (prev)
 833                 prev->vm_next = vmp;
 834         else
 835                 t->mm->mmap = vmp;
 836         vmp->vm_next = next;
 837 #endif
 838 
 839         inode = vmp->vm_inode;
 840         if (!inode)
 841                 return;
 842 
 843         /* insert vmp into inode's circular share list */
 844         if ((share = inode->i_mmap)) {
 845                 vmp->vm_next_share = share->vm_next_share;
 846                 vmp->vm_next_share->vm_prev_share = vmp;
 847                 share->vm_next_share = vmp;
 848                 vmp->vm_prev_share = share;
 849         } else
 850                 inode->i_mmap = vmp->vm_next_share = vmp->vm_prev_share = vmp;
 851 }
 852 
 853 /*
 854  * Remove one vm structure from the inode's i_mmap ring.
 855  */
 856 void remove_shared_vm_struct(struct vm_area_struct *mpnt)
     /*  */
 857 {
 858         struct inode * inode = mpnt->vm_inode;
 859 
 860         if (!inode)
 861                 return;
 862 
 863         if (mpnt->vm_next_share == mpnt) {
 864                 if (inode->i_mmap != mpnt)
 865                         printk("Inode i_mmap ring corrupted\n");
 866                 inode->i_mmap = NULL;
 867                 return;
 868         }
 869 
 870         if (inode->i_mmap == mpnt)
 871                 inode->i_mmap = mpnt->vm_next_share;
 872 
 873         mpnt->vm_prev_share->vm_next_share = mpnt->vm_next_share;
 874         mpnt->vm_next_share->vm_prev_share = mpnt->vm_prev_share;
 875 }
 876 
 877 /*
 878  * Merge the list of memory segments if possible.
 879  * Redundant vm_area_structs are freed.
 880  * This assumes that the list is ordered by address.
 881  * We don't need to traverse the entire list, only those segments
 882  * which intersect or are adjacent to a given interval.
 883  */
 884 void merge_segments (struct task_struct * task, unsigned long start_addr, unsigned long end_addr)
     /*  */
 885 {
 886         struct vm_area_struct *prev, *mpnt, *next;
 887 
 888         mpnt = find_vma(task, start_addr);
 889         if (!mpnt)
 890                 return;
 891         avl_neighbours(mpnt, task->mm->mmap_avl, &prev, &next);
 892         /* we have  prev->vm_next == mpnt && mpnt->vm_next = next */
 893 
 894         if (!prev) {
 895                 prev = mpnt;
 896                 mpnt = next;
 897         }
 898 
 899         /* prev and mpnt cycle through the list, as long as
 900          * start_addr < mpnt->vm_end && prev->vm_start < end_addr
 901          */
 902         for ( ; mpnt && prev->vm_start < end_addr ; prev = mpnt, mpnt = next) {
 903 #if 0
 904                 printk("looping in merge_segments, mpnt=0x%lX\n", (unsigned long) mpnt);
 905 #endif
 906 
 907                 next = mpnt->vm_next;
 908 
 909                 /*
 910                  * To share, we must have the same inode, operations.. 
 911                  */
 912                 if (mpnt->vm_inode != prev->vm_inode)
 913                         continue;
 914                 if (mpnt->vm_pte != prev->vm_pte)
 915                         continue;
 916                 if (mpnt->vm_ops != prev->vm_ops)
 917                         continue;
 918                 if (mpnt->vm_flags != prev->vm_flags)
 919                         continue;
 920                 if (prev->vm_end != mpnt->vm_start)
 921                         continue;
 922                 /*
 923                  * and if we have an inode, the offsets must be contiguous..
 924                  */
 925                 if ((mpnt->vm_inode != NULL) || (mpnt->vm_flags & VM_SHM)) {
 926                         if (prev->vm_offset + prev->vm_end - prev->vm_start != mpnt->vm_offset)
 927                                 continue;
 928                 }
 929 
 930                 /*
 931                  * merge prev with mpnt and set up pointers so the new
 932                  * big segment can possibly merge with the next one.
 933                  * The old unused mpnt is freed.
 934                  */
 935                 avl_remove(mpnt, &task->mm->mmap_avl);
 936                 prev->vm_end = mpnt->vm_end;
 937                 prev->vm_next = mpnt->vm_next;
 938                 if (mpnt->vm_ops && mpnt->vm_ops->close) {
 939                         mpnt->vm_offset += mpnt->vm_end - mpnt->vm_start;
 940                         mpnt->vm_start = mpnt->vm_end;
 941                         mpnt->vm_ops->close(mpnt);
 942                 }
 943                 remove_shared_vm_struct(mpnt);
 944                 if (mpnt->vm_inode)
 945                         mpnt->vm_inode->i_count--;
 946                 kfree_s(mpnt, sizeof(*mpnt));
 947                 mpnt = prev;
 948         }
 949 }
 950 
 951 /*
 952  * Map memory not associated with any file into a process
 953  * address space.  Adjacent memory is merged.
 954  */
 955 static int anon_map(struct inode *ino, struct file * file, struct vm_area_struct * vma)
     /*  */
 956 {
 957         if (zeromap_page_range(vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 958                 return -ENOMEM;
 959         return 0;
 960 }
/* */
root/mm/mmap.c

DEFINITIONS
