mm/mmap.c

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

DEFINITIONS
