root/ipc/shm.c

/* */

DEFINITIONS

1. shm_init
2. findkey
3. newseg
4. sys_shmget
5. killseg
6. sys_shmctl
7. shm_map
8. sys_shmat
9. shm_open
10. shm_close
11. sys_shmdt
12. shm_swap_in
13. shm_swap

   1 /*
   2  * linux/ipc/shm.c
   3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4  *         Many improvements/fixes by Bruno Haible.
   5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6  */
   7 
   8 #include <linux/errno.h>
   9 #include <asm/segment.h>
  10 #include <linux/sched.h>
  11 #include <linux/mm.h>
  12 #include <linux/ipc.h>
  13 #include <linux/shm.h>
  14 #include <linux/stat.h>
  15 #include <linux/malloc.h>
  16 
  17 extern int ipcperms (struct ipc_perm *ipcp, short shmflg);
  18 extern unsigned int get_swap_page (void);
  19 static int findkey (key_t key);
  20 static int newseg (key_t key, int shmflg, int size);
  21 static int shm_map (struct vm_area_struct *shmd, int remap);
  22 static void killseg (int id);
  23 static void shm_open (struct vm_area_struct *shmd);
  24 static void shm_close (struct vm_area_struct *shmd);
  25 static unsigned long shm_swap_in (struct vm_area_struct *, unsigned long);
  26 
  27 static int shm_tot = 0; /* total number of shared memory pages */
  28 static int shm_rss = 0; /* number of shared memory pages that are in memory */
  29 static int shm_swp = 0; /* number of shared memory pages that are in swap */
  30 static int max_shmid = 0; /* every used id is <= max_shmid */
  31 static struct wait_queue *shm_lock = NULL; /* calling findkey() may need to wait */
  32 static struct shmid_ds *shm_segs[SHMMNI];
  33 
  34 static unsigned short shm_seq = 0; /* incremented, for recognizing stale ids */
  35 
  36 /* some statistics */
  37 static ulong swap_attempts = 0;
  38 static ulong swap_successes = 0;
  39 static ulong used_segs = 0;
  40 
  41 void shm_init (void)
     /*  */
  42 {
  43         int id;
  44 
  45         for (id = 0; id < SHMMNI; id++)
  46                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
  47         shm_tot = shm_rss = shm_seq = max_shmid = used_segs = 0;
  48         shm_lock = NULL;
  49         return;
  50 }
  51 
  52 static int findkey (key_t key)
     /*  */
  53 {
  54         int id;
  55         struct shmid_ds *shp;
  56 
  57         for (id = 0; id <= max_shmid; id++) {
  58                 while ((shp = shm_segs[id]) == IPC_NOID)
  59                         sleep_on (&shm_lock);
  60                 if (shp == IPC_UNUSED)
  61                         continue;
  62                 if (key == shp->shm_perm.key)
  63                         return id;
  64         }
  65         return -1;
  66 }
  67 
  68 /*
  69  * allocate new shmid_ds and pgtable. protected by shm_segs[id] = NOID.
  70  */
  71 static int newseg (key_t key, int shmflg, int size)
     /*  */
  72 {
  73         struct shmid_ds *shp;
  74         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
  75         int id, i;
  76 
  77         if (size < SHMMIN)
  78                 return -EINVAL;
  79         if (shm_tot + numpages >= SHMALL)
  80                 return -ENOSPC;
  81         for (id = 0; id < SHMMNI; id++)
  82                 if (shm_segs[id] == IPC_UNUSED) {
  83                         shm_segs[id] = (struct shmid_ds *) IPC_NOID;
  84                         goto found;
  85                 }
  86         return -ENOSPC;
  87 
  88 found:
  89         shp = (struct shmid_ds *) kmalloc (sizeof (*shp), GFP_KERNEL);
  90         if (!shp) {
  91                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
  92                 if (shm_lock)
  93                         wake_up (&shm_lock);
  94                 return -ENOMEM;
  95         }
  96 
  97         shp->shm_pages = (ulong *) kmalloc (numpages*sizeof(ulong),GFP_KERNEL);
  98         if (!shp->shm_pages) {
  99                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
 100                 if (shm_lock)
 101                         wake_up (&shm_lock);
 102                 kfree(shp);
 103                 return -ENOMEM;
 104         }
 105 
 106         for (i = 0; i < numpages; shp->shm_pages[i++] = 0);
 107         shm_tot += numpages;
 108         shp->shm_perm.key = key;
 109         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 110         shp->shm_perm.cuid = shp->shm_perm.uid = current->euid;
 111         shp->shm_perm.cgid = shp->shm_perm.gid = current->egid;
 112         shp->shm_perm.seq = shm_seq;
 113         shp->shm_segsz = size;
 114         shp->shm_cpid = current->pid;
 115         shp->attaches = NULL;
 116         shp->shm_lpid = shp->shm_nattch = 0;
 117         shp->shm_atime = shp->shm_dtime = 0;
 118         shp->shm_ctime = CURRENT_TIME;
 119         shp->shm_npages = numpages;
 120 
 121         if (id > max_shmid)
 122                 max_shmid = id;
 123         shm_segs[id] = shp;
 124         used_segs++;
 125         if (shm_lock)
 126                 wake_up (&shm_lock);
 127         return (unsigned int) shp->shm_perm.seq * SHMMNI + id;
 128 }
 129 
 130 int sys_shmget (key_t key, int size, int shmflg)
     /*  */
 131 {
 132         struct shmid_ds *shp;
 133         int id = 0;
 134 
 135         if (size < 0 || size > SHMMAX)
 136                 return -EINVAL;
 137         if (key == IPC_PRIVATE)
 138                 return newseg(key, shmflg, size);
 139         if ((id = findkey (key)) == -1) {
 140                 if (!(shmflg & IPC_CREAT))
 141                         return -ENOENT;
 142                 return newseg(key, shmflg, size);
 143         }
 144         if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL))
 145                 return -EEXIST;
 146         shp = shm_segs[id];
 147         if (shp->shm_perm.mode & SHM_DEST)
 148                 return -EIDRM;
 149         if (size > shp->shm_segsz)
 150                 return -EINVAL;
 151         if (ipcperms (&shp->shm_perm, shmflg))
 152                 return -EACCES;
 153         return (unsigned int) shp->shm_perm.seq * SHMMNI + id;
 154 }
 155 
 156 /*
 157  * Only called after testing nattch and SHM_DEST.
 158  * Here pages, pgtable and shmid_ds are freed.
 159  */
 160 static void killseg (int id)
     /*  */
 161 {
 162         struct shmid_ds *shp;
 163         int i, numpages;
 164         ulong page;
 165 
 166         shp = shm_segs[id];
 167         if (shp == IPC_NOID || shp == IPC_UNUSED) {
 168                 printk ("shm nono: killseg called on unused seg id=%d\n", id);
 169                 return;
 170         }
 171         shp->shm_perm.seq++;     /* for shmat */
 172         shm_seq = (shm_seq+1) % ((unsigned)(1<<31)/SHMMNI); /* increment, but avoid overflow */
 173         shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
 174         used_segs--;
 175         if (id == max_shmid)
 176                 while (max_shmid && (shm_segs[--max_shmid] == IPC_UNUSED));
 177         if (!shp->shm_pages) {
 178                 printk ("shm nono: killseg shp->pages=NULL. id=%d\n", id);
 179                 return;
 180         }
 181         numpages = shp->shm_npages;
 182         for (i = 0; i < numpages ; i++) {
 183                 if (!(page = shp->shm_pages[i]))
 184                         continue;
 185                 if (page & PAGE_PRESENT) {
 186                         free_page (page & PAGE_MASK);
 187                         shm_rss--;
 188                 } else {
 189                         swap_free (page);
 190                         shm_swp--;
 191                 }
 192         }
 193         kfree(shp->shm_pages);
 194         shm_tot -= numpages;
 195         kfree(shp);
 196         return;
 197 }
 198 
 199 int sys_shmctl (int shmid, int cmd, struct shmid_ds *buf)
     /*  */
 200 {
 201         struct shmid_ds tbuf;
 202         struct shmid_ds *shp;
 203         struct ipc_perm *ipcp;
 204         int id, err;
 205 
 206         if (cmd < 0 || shmid < 0)
 207                 return -EINVAL;
 208         if (cmd == IPC_SET) {
 209                 if (!buf)
 210                         return -EFAULT;
 211                 err = verify_area (VERIFY_READ, buf, sizeof (*buf));
 212                 if (err)
 213                         return err;
 214                 memcpy_fromfs (&tbuf, buf, sizeof (*buf));
 215         }
 216 
 217         switch (cmd) { /* replace with proc interface ? */
 218         case IPC_INFO:
 219         {
 220                 struct shminfo shminfo;
 221                 if (!buf)
 222                         return -EFAULT;
 223                 shminfo.shmmni = SHMMNI;
 224                 shminfo.shmmax = SHMMAX;
 225                 shminfo.shmmin = SHMMIN;
 226                 shminfo.shmall = SHMALL;
 227                 shminfo.shmseg = SHMSEG;
 228                 err = verify_area (VERIFY_WRITE, buf, sizeof (struct shminfo));
 229                 if (err)
 230                         return err;
 231                 memcpy_tofs (buf, &shminfo, sizeof(struct shminfo));
 232                 return max_shmid;
 233         }
 234         case SHM_INFO:
 235         {
 236                 struct shm_info shm_info;
 237                 if (!buf)
 238                         return -EFAULT;
 239                 err = verify_area (VERIFY_WRITE, buf, sizeof (shm_info));
 240                 if (err)
 241                         return err;
 242                 shm_info.used_ids = used_segs;
 243                 shm_info.shm_rss = shm_rss;
 244                 shm_info.shm_tot = shm_tot;
 245                 shm_info.shm_swp = shm_swp;
 246                 shm_info.swap_attempts = swap_attempts;
 247                 shm_info.swap_successes = swap_successes;
 248                 memcpy_tofs (buf, &shm_info, sizeof(shm_info));
 249                 return max_shmid;
 250         }
 251         case SHM_STAT:
 252                 if (!buf)
 253                         return -EFAULT;
 254                 err = verify_area (VERIFY_WRITE, buf, sizeof (*buf));
 255                 if (err)
 256                         return err;
 257                 if (shmid > max_shmid)
 258                         return -EINVAL;
 259                 shp = shm_segs[shmid];
 260                 if (shp == IPC_UNUSED || shp == IPC_NOID)
 261                         return -EINVAL;
 262                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 263                         return -EACCES;
 264                 id = (unsigned int) shp->shm_perm.seq * SHMMNI + shmid;
 265                 tbuf.shm_perm   = shp->shm_perm;
 266                 tbuf.shm_segsz  = shp->shm_segsz;
 267                 tbuf.shm_atime  = shp->shm_atime;
 268                 tbuf.shm_dtime  = shp->shm_dtime;
 269                 tbuf.shm_ctime  = shp->shm_ctime;
 270                 tbuf.shm_cpid   = shp->shm_cpid;
 271                 tbuf.shm_lpid   = shp->shm_lpid;
 272                 tbuf.shm_nattch = shp->shm_nattch;
 273                 memcpy_tofs (buf, &tbuf, sizeof(*buf));
 274                 return id;
 275         }
 276 
 277         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 278         if (shp == IPC_UNUSED || shp == IPC_NOID)
 279                 return -EINVAL;
 280         if (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)
 281                 return -EIDRM;
 282         ipcp = &shp->shm_perm;
 283 
 284         switch (cmd) {
 285         case SHM_UNLOCK:
 286                 if (!suser())
 287                         return -EPERM;
 288                 if (!(ipcp->mode & SHM_LOCKED))
 289                         return -EINVAL;
 290                 ipcp->mode &= ~SHM_LOCKED;
 291                 break;
 292         case SHM_LOCK:
 293 /* Allow superuser to lock segment in memory */
 294 /* Should the pages be faulted in here or leave it to user? */
 295 /* need to determine interaction with current->swappable */
 296                 if (!suser())
 297                         return -EPERM;
 298                 if (ipcp->mode & SHM_LOCKED)
 299                         return -EINVAL;
 300                 ipcp->mode |= SHM_LOCKED;
 301                 break;
 302         case IPC_STAT:
 303                 if (ipcperms (ipcp, S_IRUGO))
 304                         return -EACCES;
 305                 if (!buf)
 306                         return -EFAULT;
 307                 err = verify_area (VERIFY_WRITE, buf, sizeof (*buf));
 308                 if (err)
 309                         return err;
 310                 tbuf.shm_perm   = shp->shm_perm;
 311                 tbuf.shm_segsz  = shp->shm_segsz;
 312                 tbuf.shm_atime  = shp->shm_atime;
 313                 tbuf.shm_dtime  = shp->shm_dtime;
 314                 tbuf.shm_ctime  = shp->shm_ctime;
 315                 tbuf.shm_cpid   = shp->shm_cpid;
 316                 tbuf.shm_lpid   = shp->shm_lpid;
 317                 tbuf.shm_nattch = shp->shm_nattch;
 318                 memcpy_tofs (buf, &tbuf, sizeof(*buf));
 319                 break;
 320         case IPC_SET:
 321                 if (suser() || current->euid == shp->shm_perm.uid ||
 322                     current->euid == shp->shm_perm.cuid) {
 323                         ipcp->uid = tbuf.shm_perm.uid;
 324                         ipcp->gid = tbuf.shm_perm.gid;
 325                         ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
 326                                 | (tbuf.shm_perm.mode & S_IRWXUGO);
 327                         shp->shm_ctime = CURRENT_TIME;
 328                         break;
 329                 }
 330                 return -EPERM;
 331         case IPC_RMID:
 332                 if (suser() || current->euid == shp->shm_perm.uid ||
 333                     current->euid == shp->shm_perm.cuid) {
 334                         shp->shm_perm.mode |= SHM_DEST;
 335                         if (shp->shm_nattch <= 0)
 336                                 killseg (id);
 337                         break;
 338                 }
 339                 return -EPERM;
 340         default:
 341                 return -EINVAL;
 342         }
 343         return 0;
 344 }
 345 
 346 /*
 347  * The per process internal structure for managing segments is
 348  * `struct vm_area_struct'.
 349  * A shmat will add to and shmdt will remove from the list.
 350  * shmd->vm_task        the attacher
 351  * shmd->vm_start       virt addr of attach, multiple of SHMLBA
 352  * shmd->vm_end         multiple of SHMLBA
 353  * shmd->vm_next        next attach for task
 354  * shmd->vm_next_share  next attach for segment
 355  * shmd->vm_offset      offset into segment
 356  * shmd->vm_pte         signature for this attach
 357  */
 358 
 359 static struct vm_operations_struct shm_vm_ops = {
 360         shm_open,               /* open */
 361         shm_close,              /* close */
 362         NULL,                   /* unmap */
 363         NULL,                   /* protect */
 364         NULL,                   /* sync */
 365         NULL,                   /* advise */
 366         NULL,                   /* nopage (done with swapin) */
 367         NULL,                   /* wppage */
 368         NULL,                   /* swapout (hardcoded right now) */
 369         shm_swap_in             /* swapin */
 370 };
 371 
 372 /*
 373  * check range is unmapped, ensure page tables exist
 374  * mark page table entries with shm_sgn.
 375  * if remap != 0 the range is remapped.
 376  */
 377 static int shm_map (struct vm_area_struct *shmd, int remap)
     /*  */
 378 {
 379         unsigned long *page_table;
 380         unsigned long tmp, shm_sgn;
 381         unsigned long page_dir = shmd->vm_task->tss.cr3;
 382 
 383         /* check that the range is unmapped */
 384         if (!remap)
 385                 for (tmp = shmd->vm_start; tmp < shmd->vm_end; tmp += PAGE_SIZE) {
 386                         page_table = PAGE_DIR_OFFSET(page_dir,tmp);
 387                         if (*page_table & PAGE_PRESENT) {
 388                                 page_table = (ulong *) (PAGE_MASK & *page_table);
 389                                 page_table += ((tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1));
 390                                 if (*page_table) {
 391                                         /* printk("shmat() -> EINVAL because address 0x%lx is already mapped.\n",tmp); */
 392                                         return -EINVAL;
 393                                 }
 394                         }
 395                 }
 396 
 397         /* clear old mappings */
 398         do_munmap(shmd->vm_start, shmd->vm_end - shmd->vm_start);
 399 
 400         /* add new mapping */
 401         insert_vm_struct(current, shmd);
 402         merge_segments(current->mm->mmap);
 403 
 404         /* check that the range has page_tables */
 405         for (tmp = shmd->vm_start; tmp < shmd->vm_end; tmp += PAGE_SIZE) {
 406                 page_table = PAGE_DIR_OFFSET(page_dir,tmp);
 407                 if (*page_table & PAGE_PRESENT) {
 408                         page_table = (ulong *) (PAGE_MASK & *page_table);
 409                         page_table += ((tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1));
 410                         if (*page_table) {
 411                                 if (*page_table & PAGE_PRESENT) {
 412                                         --current->mm->rss;
 413                                         free_page (*page_table & PAGE_MASK);
 414                                 }
 415                                 else
 416                                         swap_free (*page_table);
 417                                 *page_table = 0;
 418                         }
 419                 } else {
 420                         unsigned long new_pt;
 421                         if (!(new_pt = get_free_page(GFP_KERNEL)))
 422                                 return -ENOMEM;
 423                         *page_table = new_pt | PAGE_TABLE;
 424                         tmp |= ((PAGE_SIZE << 10) - PAGE_SIZE);
 425                 }
 426         }
 427 
 428         /* map page range */
 429         shm_sgn = shmd->vm_pte + ((shmd->vm_offset >> PAGE_SHIFT) << SHM_IDX_SHIFT);
 430         for (tmp = shmd->vm_start; tmp < shmd->vm_end; tmp += PAGE_SIZE,
 431              shm_sgn += (1 << SHM_IDX_SHIFT)) {
 432                 page_table = PAGE_DIR_OFFSET(page_dir,tmp);
 433                 page_table = (ulong *) (PAGE_MASK & *page_table);
 434                 page_table += (tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
 435                 *page_table = shm_sgn;
 436         }
 437         invalidate();
 438         return 0;
 439 }
 440 
 441 /*
 442  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 443  */
 444 int sys_shmat (int shmid, char *shmaddr, int shmflg, ulong *raddr)
     /*  */
 445 {
 446         struct shmid_ds *shp;
 447         struct vm_area_struct *shmd;
 448         int err;
 449         unsigned int id;
 450         unsigned long addr;
 451 
 452         if (shmid < 0) {
 453                 /* printk("shmat() -> EINVAL because shmid = %d < 0\n",shmid); */
 454                 return -EINVAL;
 455         }
 456 
 457         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 458         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 459                 /* printk("shmat() -> EINVAL because shmid = %d is invalid\n",shmid); */
 460                 return -EINVAL;
 461         }
 462 
 463         if (!(addr = (ulong) shmaddr)) {
 464                 if (shmflg & SHM_REMAP)
 465                         return -EINVAL;
 466                 if (!(addr = get_unmapped_area(shp->shm_segsz)))
 467                         return -ENOMEM;
 468         } else if (addr & (SHMLBA-1)) {
 469                 if (shmflg & SHM_RND)
 470                         addr &= ~(SHMLBA-1);       /* round down */
 471                 else
 472                         return -EINVAL;
 473         }
 474         if ((addr > current->mm->start_stack - 16384 - PAGE_SIZE*shp->shm_npages)) {
 475                 /* printk("shmat() -> EINVAL because segment intersects stack\n"); */
 476                 return -EINVAL;
 477         }
 478         if (!(shmflg & SHM_REMAP))
 479                 for (shmd = current->mm->mmap; shmd; shmd = shmd->vm_next)
 480                         if (!(addr >= shmd->vm_end || addr + shp->shm_segsz <= shmd->vm_start)) {
 481                                 /* printk("shmat() -> EINVAL because the interval [0x%lx,0x%lx) intersects an already mapped interval [0x%lx,0x%lx).\n",
 482                                         addr, addr + shp->shm_segsz, shmd->vm_start, shmd->vm_end); */
 483                                 return -EINVAL;
 484                         }
 485 
 486         if (ipcperms(&shp->shm_perm, shmflg & SHM_RDONLY ? S_IRUGO : S_IRUGO|S_IWUGO))
 487                 return -EACCES;
 488         if (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)
 489                 return -EIDRM;
 490 
 491         shmd = (struct vm_area_struct *) kmalloc (sizeof(*shmd), GFP_KERNEL);
 492         if (!shmd)
 493                 return -ENOMEM;
 494         if ((shp != shm_segs[id]) || (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)) {
 495                 kfree(shmd);
 496                 return -EIDRM;
 497         }
 498 
 499         shmd->vm_pte = (SHM_SWP_TYPE << 1) | (id << SHM_ID_SHIFT) |
 500                 (shmflg & SHM_RDONLY ? SHM_READ_ONLY : 0);
 501         shmd->vm_start = addr;
 502         shmd->vm_end = addr + shp->shm_npages * PAGE_SIZE;
 503         shmd->vm_task = current;
 504         shmd->vm_page_prot = (shmflg & SHM_RDONLY) ? PAGE_READONLY : PAGE_SHARED;
 505         shmd->vm_flags = VM_SHM | VM_MAYSHARE | VM_SHARED
 506                          | VM_MAYREAD | VM_MAYEXEC | VM_READ | VM_EXEC
 507                          | ((shmflg & SHM_RDONLY) ? 0 : VM_MAYWRITE | VM_WRITE);
 508         shmd->vm_next_share = NULL;
 509         shmd->vm_inode = NULL;
 510         shmd->vm_offset = 0;
 511         shmd->vm_ops = &shm_vm_ops;
 512 
 513         shp->shm_nattch++;            /* prevent destruction */
 514         if ((err = shm_map (shmd, shmflg & SHM_REMAP))) {
 515                 if (--shp->shm_nattch <= 0 && shp->shm_perm.mode & SHM_DEST)
 516                         killseg(id);
 517                 kfree(shmd);
 518                 return err;
 519         }
 520 
 521         shmd->vm_next_share = shp->attaches;
 522         shp->attaches = shmd;
 523         shp->shm_lpid = current->pid;
 524         shp->shm_atime = CURRENT_TIME;
 525 
 526         *raddr = addr;
 527         return 0;
 528 }
 529 
 530 /* This is called by fork, once for every shm attach. */
 531 static void shm_open (struct vm_area_struct *shmd)
     /*  */
 532 {
 533         unsigned int id;
 534         struct shmid_ds *shp;
 535 
 536         id = (shmd->vm_pte >> SHM_ID_SHIFT) & SHM_ID_MASK;
 537         shp = shm_segs[id];
 538         if (shp == IPC_UNUSED) {
 539                 printk("shm_open: unused id=%d PANIC\n", id);
 540                 return;
 541         }
 542         shmd->vm_next_share = shp->attaches;
 543         shp->attaches = shmd;
 544         shp->shm_nattch++;
 545         shp->shm_atime = CURRENT_TIME;
 546         shp->shm_lpid = current->pid;
 547 }
 548 
 549 /*
 550  * remove the attach descriptor shmd.
 551  * free memory for segment if it is marked destroyed.
 552  * The descriptor has already been removed from the current->mm->mmap list
 553  * and will later be kfree()d.
 554  */
 555 static void shm_close (struct vm_area_struct *shmd)
     /*  */
 556 {
 557         struct vm_area_struct **shmdp;
 558         struct shmid_ds *shp;
 559         int id;
 560 
 561         unmap_page_range (shmd->vm_start, shmd->vm_end - shmd->vm_start);
 562 
 563         /* remove from the list of attaches of the shm segment */
 564         id = (shmd->vm_pte >> SHM_ID_SHIFT) & SHM_ID_MASK;
 565         shp = shm_segs[id];
 566         for (shmdp = &shp->attaches; *shmdp; shmdp = &(*shmdp)->vm_next_share)
 567                 if (*shmdp == shmd) {
 568                         *shmdp = shmd->vm_next_share;
 569                         goto found;
 570                 }
 571         printk("shm_close: shm segment (id=%d) attach list inconsistent\n",id);
 572         printk("shm_close: %d %08lx-%08lx %c%c%c%c %08lx %08lx\n",
 573                 shmd->vm_task->pid, shmd->vm_start, shmd->vm_end,
 574                 shmd->vm_flags & VM_READ ? 'r' : '-',
 575                 shmd->vm_flags & VM_WRITE ? 'w' : '-',
 576                 shmd->vm_flags & VM_EXEC ? 'x' : '-',
 577                 shmd->vm_flags & VM_SHARED ? 's' : 'p',
 578                 shmd->vm_offset, shmd->vm_pte);
 579 
 580  found:
 581         shp->shm_lpid = current->pid;
 582         shp->shm_dtime = CURRENT_TIME;
 583         if (--shp->shm_nattch <= 0 && shp->shm_perm.mode & SHM_DEST)
 584                 killseg (id);
 585 }
 586 
 587 /*
 588  * detach and kill segment if marked destroyed.
 589  * The work is done in shm_close.
 590  */
 591 int sys_shmdt (char *shmaddr)
     /*  */
 592 {
 593         struct vm_area_struct *shmd, *shmdnext;
 594 
 595         for (shmd = current->mm->mmap; shmd; shmd = shmdnext) {
 596                 shmdnext = shmd->vm_next;
 597                 if (shmd->vm_ops == &shm_vm_ops
 598                     && shmd->vm_start - shmd->vm_offset == (ulong) shmaddr)
 599                         do_munmap(shmd->vm_start, shmd->vm_end - shmd->vm_start);
 600         }
 601         return 0;
 602 }
 603 
 604 /*
 605  * page not present ... go through shm_pages
 606  */
 607 static unsigned long shm_swap_in(struct vm_area_struct * vma, unsigned long code)
     /*  */
 608 {
 609         unsigned long page;
 610         struct shmid_ds *shp;
 611         unsigned int id, idx;
 612 
 613         id = (code >> SHM_ID_SHIFT) & SHM_ID_MASK;
 614         if (id > max_shmid) {
 615                 printk ("shm_no_page: id=%d too big. proc mem corrupted\n", id);
 616                 return BAD_PAGE | PAGE_SHARED;
 617         }
 618         shp = shm_segs[id];
 619         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 620                 printk ("shm_no_page: id=%d invalid. Race.\n", id);
 621                 return BAD_PAGE | PAGE_SHARED;
 622         }
 623         idx = (code >> SHM_IDX_SHIFT) & SHM_IDX_MASK;
 624         if (idx >= shp->shm_npages) {
 625                 printk ("shm_no_page : too large page index. id=%d\n", id);
 626                 return BAD_PAGE | PAGE_SHARED;
 627         }
 628 
 629         if (!(shp->shm_pages[idx] & PAGE_PRESENT)) {
 630                 if(!(page = get_free_page(GFP_KERNEL))) {
 631                         oom(current);
 632                         return BAD_PAGE | PAGE_SHARED;
 633                 }
 634                 if (shp->shm_pages[idx] & PAGE_PRESENT) {
 635                         free_page (page);
 636                         goto done;
 637                 }
 638                 if (shp->shm_pages[idx]) {
 639                         read_swap_page (shp->shm_pages[idx], (char *) page);
 640                         if (shp->shm_pages[idx] & PAGE_PRESENT)  {
 641                                 free_page (page);
 642                                 goto done;
 643                         }
 644                         swap_free (shp->shm_pages[idx]);
 645                         shm_swp--;
 646                 }
 647                 shm_rss++;
 648                 shp->shm_pages[idx] = page | (PAGE_SHARED | PAGE_DIRTY);
 649         } else
 650                 --current->mm->maj_flt;  /* was incremented in do_no_page */
 651 
 652 done:
 653         current->mm->min_flt++;
 654         page = shp->shm_pages[idx];
 655         if (code & SHM_READ_ONLY)           /* write-protect */
 656                 page &= ~PAGE_RW;
 657         mem_map[MAP_NR(page)]++;
 658         return page;
 659 }
 660 
 661 /*
 662  * Goes through counter = (shm_rss << prio) present shm pages.
 663  */
 664 static unsigned long swap_id = 0; /* currently being swapped */
 665 static unsigned long swap_idx = 0; /* next to swap */
 666 
 667 int shm_swap (int prio)
     /*  */
 668 {
 669         unsigned long page;
 670         struct shmid_ds *shp;
 671         struct vm_area_struct *shmd;
 672         unsigned int swap_nr;
 673         unsigned long id, idx, invalid = 0;
 674         int counter;
 675 
 676         counter = shm_rss >> prio;
 677         if (!counter || !(swap_nr = get_swap_page()))
 678                 return 0;
 679 
 680  check_id:
 681         shp = shm_segs[swap_id];
 682         if (shp == IPC_UNUSED || shp == IPC_NOID || shp->shm_perm.mode & SHM_LOCKED ) {
 683                 swap_idx = 0;
 684                 if (++swap_id > max_shmid)
 685                         swap_id = 0;
 686                 goto check_id;
 687         }
 688         id = swap_id;
 689 
 690  check_table:
 691         idx = swap_idx++;
 692         if (idx >= shp->shm_npages) {
 693                 swap_idx = 0;
 694                 if (++swap_id > max_shmid)
 695                         swap_id = 0;
 696                 goto check_id;
 697         }
 698 
 699         page = shp->shm_pages[idx];
 700         if (!(page & PAGE_PRESENT))
 701                 goto check_table;
 702         swap_attempts++;
 703 
 704         if (--counter < 0) { /* failed */
 705                 if (invalid)
 706                         invalidate();
 707                 swap_free (swap_nr);
 708                 return 0;
 709         }
 710         for (shmd = shp->attaches; shmd; shmd = shmd->vm_next_share) {
 711                 unsigned long tmp, *pte;
 712                 if ((shmd->vm_pte >> SHM_ID_SHIFT & SHM_ID_MASK) != id) {
 713                         printk ("shm_swap: id=%ld does not match shmd->vm_pte.id=%ld\n", id, shmd->vm_pte >> SHM_ID_SHIFT & SHM_ID_MASK);
 714                         continue;
 715                 }
 716                 tmp = shmd->vm_start + (idx << PAGE_SHIFT) - shmd->vm_offset;
 717                 if (!(tmp >= shmd->vm_start && tmp < shmd->vm_end))
 718                         continue;
 719                 pte = PAGE_DIR_OFFSET(shmd->vm_task->tss.cr3,tmp);
 720                 if (!(*pte & PAGE_PRESENT)) {
 721                         printk("shm_swap: bad pgtbl! id=%ld start=%lx idx=%ld\n",
 722                                         id, shmd->vm_start, idx);
 723                         *pte = 0;
 724                         continue;
 725                 }
 726                 pte = (ulong *) (PAGE_MASK & *pte);
 727                 pte += ((tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1));
 728                 tmp = *pte;
 729                 if (!(tmp & PAGE_PRESENT))
 730                         continue;
 731                 if (tmp & PAGE_ACCESSED) {
 732                         *pte &= ~PAGE_ACCESSED;
 733                         continue;
 734                 }
 735                 tmp = shmd->vm_pte | idx << SHM_IDX_SHIFT;
 736                 *pte = tmp;
 737                 mem_map[MAP_NR(page)]--;
 738                 shmd->vm_task->mm->rss--;
 739                 invalid++;
 740         }
 741 
 742         if (mem_map[MAP_NR(page)] != 1)
 743                 goto check_table;
 744         page &= PAGE_MASK;
 745         shp->shm_pages[idx] = swap_nr;
 746         if (invalid)
 747                 invalidate();
 748         write_swap_page (swap_nr, (char *) page);
 749         free_page (page);
 750         swap_successes++;
 751         shm_swp++;
 752         shm_rss--;
 753         return 1;
 754 }

/* */