root/ipc/shm.c

/* */

DEFINITIONS

1. shm_init
2. findkey
3. newseg
4. sys_shmget
5. killseg
6. sys_shmctl
7. insert_attach
8. remove_attach
9. shm_map
10. sys_shmat
11. shm_open
12. shm_close
13. sys_shmdt
14. shm_swap_in
15. 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 <linux/sched.h>
  10 #include <linux/mm.h>
  11 #include <linux/ipc.h>
  12 #include <linux/shm.h>
  13 #include <linux/stat.h>
  14 #include <linux/malloc.h>
  15 #include <linux/swap.h>
  16 
  17 #include <asm/segment.h>
  18 #include <asm/pgtable.h>
  19 
  20 extern int ipcperms (struct ipc_perm *ipcp, short shmflg);
  21 extern unsigned long get_swap_page (void);
  22 static int findkey (key_t key);
  23 static int newseg (key_t key, int shmflg, int size);
  24 static int shm_map (struct vm_area_struct *shmd);
  25 static void killseg (int id);
  26 static void shm_open (struct vm_area_struct *shmd);
  27 static void shm_close (struct vm_area_struct *shmd);
  28 static pte_t shm_swap_in(struct vm_area_struct *, unsigned long, unsigned long);
  29 
  30 static int shm_tot = 0; /* total number of shared memory pages */
  31 static int shm_rss = 0; /* number of shared memory pages that are in memory */
  32 static int shm_swp = 0; /* number of shared memory pages that are in swap */
  33 static int max_shmid = 0; /* every used id is <= max_shmid */
  34 static struct wait_queue *shm_lock = NULL; /* calling findkey() may need to wait */
  35 static struct shmid_ds *shm_segs[SHMMNI];
  36 
  37 static unsigned short shm_seq = 0; /* incremented, for recognizing stale ids */
  38 
  39 /* some statistics */
  40 static ulong swap_attempts = 0;
  41 static ulong swap_successes = 0;
  42 static ulong used_segs = 0;
  43 
  44 void shm_init (void)
     /*  */
  45 {
  46         int id;
  47 
  48         for (id = 0; id < SHMMNI; id++)
  49                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
  50         shm_tot = shm_rss = shm_seq = max_shmid = used_segs = 0;
  51         shm_lock = NULL;
  52         return;
  53 }
  54 
  55 static int findkey (key_t key)
     /*  */
  56 {
  57         int id;
  58         struct shmid_ds *shp;
  59 
  60         for (id = 0; id <= max_shmid; id++) {
  61                 while ((shp = shm_segs[id]) == IPC_NOID)
  62                         sleep_on (&shm_lock);
  63                 if (shp == IPC_UNUSED)
  64                         continue;
  65                 if (key == shp->shm_perm.key)
  66                         return id;
  67         }
  68         return -1;
  69 }
  70 
  71 /*
  72  * allocate new shmid_ds and pgtable. protected by shm_segs[id] = NOID.
  73  */
  74 static int newseg (key_t key, int shmflg, int size)
     /*  */
  75 {
  76         struct shmid_ds *shp;
  77         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
  78         int id, i;
  79 
  80         if (size < SHMMIN)
  81                 return -EINVAL;
  82         if (shm_tot + numpages >= SHMALL)
  83                 return -ENOSPC;
  84         for (id = 0; id < SHMMNI; id++)
  85                 if (shm_segs[id] == IPC_UNUSED) {
  86                         shm_segs[id] = (struct shmid_ds *) IPC_NOID;
  87                         goto found;
  88                 }
  89         return -ENOSPC;
  90 
  91 found:
  92         shp = (struct shmid_ds *) kmalloc (sizeof (*shp), GFP_KERNEL);
  93         if (!shp) {
  94                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
  95                 if (shm_lock)
  96                         wake_up (&shm_lock);
  97                 return -ENOMEM;
  98         }
  99 
 100         shp->shm_pages = (ulong *) kmalloc (numpages*sizeof(ulong),GFP_KERNEL);
 101         if (!shp->shm_pages) {
 102                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
 103                 if (shm_lock)
 104                         wake_up (&shm_lock);
 105                 kfree(shp);
 106                 return -ENOMEM;
 107         }
 108 
 109         for (i = 0; i < numpages; shp->shm_pages[i++] = 0);
 110         shm_tot += numpages;
 111         shp->shm_perm.key = key;
 112         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 113         shp->shm_perm.cuid = shp->shm_perm.uid = current->euid;
 114         shp->shm_perm.cgid = shp->shm_perm.gid = current->egid;
 115         shp->shm_perm.seq = shm_seq;
 116         shp->shm_segsz = size;
 117         shp->shm_cpid = current->pid;
 118         shp->attaches = NULL;
 119         shp->shm_lpid = shp->shm_nattch = 0;
 120         shp->shm_atime = shp->shm_dtime = 0;
 121         shp->shm_ctime = CURRENT_TIME;
 122         shp->shm_npages = numpages;
 123 
 124         if (id > max_shmid)
 125                 max_shmid = id;
 126         shm_segs[id] = shp;
 127         used_segs++;
 128         if (shm_lock)
 129                 wake_up (&shm_lock);
 130         return (unsigned int) shp->shm_perm.seq * SHMMNI + id;
 131 }
 132 
 133 asmlinkage int sys_shmget (key_t key, int size, int shmflg)
     /*  */
 134 {
 135         struct shmid_ds *shp;
 136         int id = 0;
 137 
 138         if (size < 0 || size > SHMMAX)
 139                 return -EINVAL;
 140         if (key == IPC_PRIVATE)
 141                 return newseg(key, shmflg, size);
 142         if ((id = findkey (key)) == -1) {
 143                 if (!(shmflg & IPC_CREAT))
 144                         return -ENOENT;
 145                 return newseg(key, shmflg, size);
 146         }
 147         if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL))
 148                 return -EEXIST;
 149         shp = shm_segs[id];
 150         if (shp->shm_perm.mode & SHM_DEST)
 151                 return -EIDRM;
 152         if (size > shp->shm_segsz)
 153                 return -EINVAL;
 154         if (ipcperms (&shp->shm_perm, shmflg))
 155                 return -EACCES;
 156         return (unsigned int) shp->shm_perm.seq * SHMMNI + id;
 157 }
 158 
 159 /*
 160  * Only called after testing nattch and SHM_DEST.
 161  * Here pages, pgtable and shmid_ds are freed.
 162  */
 163 static void killseg (int id)
     /*  */
 164 {
 165         struct shmid_ds *shp;
 166         int i, numpages;
 167 
 168         shp = shm_segs[id];
 169         if (shp == IPC_NOID || shp == IPC_UNUSED) {
 170                 printk ("shm nono: killseg called on unused seg id=%d\n", id);
 171                 return;
 172         }
 173         shp->shm_perm.seq++;     /* for shmat */
 174         shm_seq = (shm_seq+1) % ((unsigned)(1<<31)/SHMMNI); /* increment, but avoid overflow */
 175         shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
 176         used_segs--;
 177         if (id == max_shmid)
 178                 while (max_shmid && (shm_segs[--max_shmid] == IPC_UNUSED));
 179         if (!shp->shm_pages) {
 180                 printk ("shm nono: killseg shp->pages=NULL. id=%d\n", id);
 181                 return;
 182         }
 183         numpages = shp->shm_npages;
 184         for (i = 0; i < numpages ; i++) {
 185                 pte_t pte;
 186                 pte_val(pte) = shp->shm_pages[i];
 187                 if (pte_none(pte))
 188                         continue;
 189                 if (pte_present(pte)) {
 190                         free_page (pte_page(pte));
 191                         shm_rss--;
 192                 } else {
 193                         swap_free(pte_val(pte));
 194                         shm_swp--;
 195                 }
 196         }
 197         kfree(shp->shm_pages);
 198         shm_tot -= numpages;
 199         kfree(shp);
 200         return;
 201 }
 202 
 203 asmlinkage int sys_shmctl (int shmid, int cmd, struct shmid_ds *buf)
     /*  */
 204 {
 205         struct shmid_ds tbuf;
 206         struct shmid_ds *shp;
 207         struct ipc_perm *ipcp;
 208         int id, err;
 209 
 210         if (cmd < 0 || shmid < 0)
 211                 return -EINVAL;
 212         if (cmd == IPC_SET) {
 213                 if (!buf)
 214                         return -EFAULT;
 215                 err = verify_area (VERIFY_READ, buf, sizeof (*buf));
 216                 if (err)
 217                         return err;
 218                 memcpy_fromfs (&tbuf, buf, sizeof (*buf));
 219         }
 220 
 221         switch (cmd) { /* replace with proc interface ? */
 222         case IPC_INFO:
 223         {
 224                 struct shminfo shminfo;
 225                 if (!buf)
 226                         return -EFAULT;
 227                 shminfo.shmmni = SHMMNI;
 228                 shminfo.shmmax = SHMMAX;
 229                 shminfo.shmmin = SHMMIN;
 230                 shminfo.shmall = SHMALL;
 231                 shminfo.shmseg = SHMSEG;
 232                 err = verify_area (VERIFY_WRITE, buf, sizeof (struct shminfo));
 233                 if (err)
 234                         return err;
 235                 memcpy_tofs (buf, &shminfo, sizeof(struct shminfo));
 236                 return max_shmid;
 237         }
 238         case SHM_INFO:
 239         {
 240                 struct shm_info shm_info;
 241                 if (!buf)
 242                         return -EFAULT;
 243                 err = verify_area (VERIFY_WRITE, buf, sizeof (shm_info));
 244                 if (err)
 245                         return err;
 246                 shm_info.used_ids = used_segs;
 247                 shm_info.shm_rss = shm_rss;
 248                 shm_info.shm_tot = shm_tot;
 249                 shm_info.shm_swp = shm_swp;
 250                 shm_info.swap_attempts = swap_attempts;
 251                 shm_info.swap_successes = swap_successes;
 252                 memcpy_tofs (buf, &shm_info, sizeof(shm_info));
 253                 return max_shmid;
 254         }
 255         case SHM_STAT:
 256                 if (!buf)
 257                         return -EFAULT;
 258                 err = verify_area (VERIFY_WRITE, buf, sizeof (*buf));
 259                 if (err)
 260                         return err;
 261                 if (shmid > max_shmid)
 262                         return -EINVAL;
 263                 shp = shm_segs[shmid];
 264                 if (shp == IPC_UNUSED || shp == IPC_NOID)
 265                         return -EINVAL;
 266                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 267                         return -EACCES;
 268                 id = (unsigned int) shp->shm_perm.seq * SHMMNI + shmid;
 269                 tbuf.shm_perm   = shp->shm_perm;
 270                 tbuf.shm_segsz  = shp->shm_segsz;
 271                 tbuf.shm_atime  = shp->shm_atime;
 272                 tbuf.shm_dtime  = shp->shm_dtime;
 273                 tbuf.shm_ctime  = shp->shm_ctime;
 274                 tbuf.shm_cpid   = shp->shm_cpid;
 275                 tbuf.shm_lpid   = shp->shm_lpid;
 276                 tbuf.shm_nattch = shp->shm_nattch;
 277                 memcpy_tofs (buf, &tbuf, sizeof(*buf));
 278                 return id;
 279         }
 280 
 281         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 282         if (shp == IPC_UNUSED || shp == IPC_NOID)
 283                 return -EINVAL;
 284         if (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)
 285                 return -EIDRM;
 286         ipcp = &shp->shm_perm;
 287 
 288         switch (cmd) {
 289         case SHM_UNLOCK:
 290                 if (!suser())
 291                         return -EPERM;
 292                 if (!(ipcp->mode & SHM_LOCKED))
 293                         return -EINVAL;
 294                 ipcp->mode &= ~SHM_LOCKED;
 295                 break;
 296         case SHM_LOCK:
 297 /* Allow superuser to lock segment in memory */
 298 /* Should the pages be faulted in here or leave it to user? */
 299 /* need to determine interaction with current->swappable */
 300                 if (!suser())
 301                         return -EPERM;
 302                 if (ipcp->mode & SHM_LOCKED)
 303                         return -EINVAL;
 304                 ipcp->mode |= SHM_LOCKED;
 305                 break;
 306         case IPC_STAT:
 307                 if (ipcperms (ipcp, S_IRUGO))
 308                         return -EACCES;
 309                 if (!buf)
 310                         return -EFAULT;
 311                 err = verify_area (VERIFY_WRITE, buf, sizeof (*buf));
 312                 if (err)
 313                         return err;
 314                 tbuf.shm_perm   = shp->shm_perm;
 315                 tbuf.shm_segsz  = shp->shm_segsz;
 316                 tbuf.shm_atime  = shp->shm_atime;
 317                 tbuf.shm_dtime  = shp->shm_dtime;
 318                 tbuf.shm_ctime  = shp->shm_ctime;
 319                 tbuf.shm_cpid   = shp->shm_cpid;
 320                 tbuf.shm_lpid   = shp->shm_lpid;
 321                 tbuf.shm_nattch = shp->shm_nattch;
 322                 memcpy_tofs (buf, &tbuf, sizeof(*buf));
 323                 break;
 324         case IPC_SET:
 325                 if (suser() || current->euid == shp->shm_perm.uid ||
 326                     current->euid == shp->shm_perm.cuid) {
 327                         ipcp->uid = tbuf.shm_perm.uid;
 328                         ipcp->gid = tbuf.shm_perm.gid;
 329                         ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
 330                                 | (tbuf.shm_perm.mode & S_IRWXUGO);
 331                         shp->shm_ctime = CURRENT_TIME;
 332                         break;
 333                 }
 334                 return -EPERM;
 335         case IPC_RMID:
 336                 if (suser() || current->euid == shp->shm_perm.uid ||
 337                     current->euid == shp->shm_perm.cuid) {
 338                         shp->shm_perm.mode |= SHM_DEST;
 339                         if (shp->shm_nattch <= 0)
 340                                 killseg (id);
 341                         break;
 342                 }
 343                 return -EPERM;
 344         default:
 345                 return -EINVAL;
 346         }
 347         return 0;
 348 }
 349 
 350 /*
 351  * The per process internal structure for managing segments is
 352  * `struct vm_area_struct'.
 353  * A shmat will add to and shmdt will remove from the list.
 354  * shmd->vm_mm          the attacher
 355  * shmd->vm_start       virt addr of attach, multiple of SHMLBA
 356  * shmd->vm_end         multiple of SHMLBA
 357  * shmd->vm_next        next attach for task
 358  * shmd->vm_next_share  next attach for segment
 359  * shmd->vm_offset      offset into segment
 360  * shmd->vm_pte         signature for this attach
 361  */
 362 
 363 static struct vm_operations_struct shm_vm_ops = {
 364         shm_open,               /* open - callback for a new vm-area open */
 365         shm_close,              /* close - callback for when the vm-area is released */
 366         NULL,                   /* no need to sync pages at unmap */
 367         NULL,                   /* protect */
 368         NULL,                   /* sync */
 369         NULL,                   /* advise */
 370         NULL,                   /* nopage (done with swapin) */
 371         NULL,                   /* wppage */
 372         NULL,                   /* swapout (hardcoded right now) */
 373         shm_swap_in             /* swapin */
 374 };
 375 
 376 /* Insert shmd into the circular list shp->attaches */
 377 static inline void insert_attach (struct shmid_ds * shp, struct vm_area_struct * shmd)
     /*  */
 378 {
 379         struct vm_area_struct * attaches;
 380 
 381         if ((attaches = shp->attaches)) {
 382                 shmd->vm_next_share = attaches;
 383                 shmd->vm_prev_share = attaches->vm_prev_share;
 384                 shmd->vm_prev_share->vm_next_share = shmd;
 385                 attaches->vm_prev_share = shmd;
 386         } else
 387                 shp->attaches = shmd->vm_next_share = shmd->vm_prev_share = shmd;
 388 }
 389 
 390 /* Remove shmd from circular list shp->attaches */
 391 static inline void remove_attach (struct shmid_ds * shp, struct vm_area_struct * shmd)
     /*  */
 392 {
 393         if (shmd->vm_next_share == shmd) {
 394                 if (shp->attaches != shmd) {
 395                         printk("shm_close: shm segment (id=%ld) attach list inconsistent\n",
 396                                SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK);
 397                         printk("shm_close: %08lx-%08lx %c%c%c%c %08lx %08lx\n",
 398                                 shmd->vm_start, shmd->vm_end,
 399                                 shmd->vm_flags & VM_READ ? 'r' : '-',
 400                                 shmd->vm_flags & VM_WRITE ? 'w' : '-',
 401                                 shmd->vm_flags & VM_EXEC ? 'x' : '-',
 402                                 shmd->vm_flags & VM_MAYSHARE ? 's' : 'p',
 403                                 shmd->vm_offset, shmd->vm_pte);
 404                 }
 405                 shp->attaches = NULL;
 406         } else {
 407                 if (shp->attaches == shmd)
 408                         shp->attaches = shmd->vm_next_share;
 409                 shmd->vm_prev_share->vm_next_share = shmd->vm_next_share;
 410                 shmd->vm_next_share->vm_prev_share = shmd->vm_prev_share;
 411         }
 412 }
 413 
 414 /*
 415  * ensure page tables exist
 416  * mark page table entries with shm_sgn.
 417  */
 418 static int shm_map (struct vm_area_struct *shmd)
     /*  */
 419 {
 420         pgd_t *page_dir;
 421         pmd_t *page_middle;
 422         pte_t *page_table;
 423         unsigned long tmp, shm_sgn;
 424 
 425         /* clear old mappings */
 426         do_munmap(shmd->vm_start, shmd->vm_end - shmd->vm_start);
 427 
 428         /* add new mapping */
 429         current->mm->total_vm += (shmd->vm_end - shmd->vm_start) >> PAGE_SHIFT;
 430         insert_vm_struct(current, shmd);
 431         merge_segments(current, shmd->vm_start, shmd->vm_end);
 432 
 433         /* map page range */
 434         shm_sgn = shmd->vm_pte +
 435           SWP_ENTRY(0, (shmd->vm_offset >> PAGE_SHIFT) << SHM_IDX_SHIFT);
 436         for (tmp = shmd->vm_start;
 437              tmp < shmd->vm_end;
 438              tmp += PAGE_SIZE, shm_sgn += SWP_ENTRY(0, 1 << SHM_IDX_SHIFT))
 439         {
 440                 page_dir = pgd_offset(shmd->vm_mm,tmp);
 441                 page_middle = pmd_alloc(page_dir,tmp);
 442                 if (!page_middle)
 443                         return -ENOMEM;
 444                 page_table = pte_alloc(page_middle,tmp);
 445                 if (!page_table)
 446                         return -ENOMEM;
 447                 set_pte(page_table, __pte(shm_sgn));
 448         }
 449         invalidate_range(shmd->vm_mm, shmd->vm_start, shmd->vm_end);
 450         return 0;
 451 }
 452 
 453 /*
 454  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 455  */
 456 asmlinkage int sys_shmat (int shmid, char *shmaddr, int shmflg, ulong *raddr)
     /*  */
 457 {
 458         struct shmid_ds *shp;
 459         struct vm_area_struct *shmd;
 460         int err;
 461         unsigned int id;
 462         unsigned long addr;
 463 
 464         if (shmid < 0) {
 465                 /* printk("shmat() -> EINVAL because shmid = %d < 0\n",shmid); */
 466                 return -EINVAL;
 467         }
 468 
 469         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 470         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 471                 /* printk("shmat() -> EINVAL because shmid = %d is invalid\n",shmid); */
 472                 return -EINVAL;
 473         }
 474 
 475         if (!(addr = (ulong) shmaddr)) {
 476                 if (shmflg & SHM_REMAP)
 477                         return -EINVAL;
 478                 if (!(addr = get_unmapped_area(0, shp->shm_segsz)))
 479                         return -ENOMEM;
 480         } else if (addr & (SHMLBA-1)) {
 481                 if (shmflg & SHM_RND)
 482                         addr &= ~(SHMLBA-1);       /* round down */
 483                 else
 484                         return -EINVAL;
 485         }
 486         /*
 487          * If shm segment goes below stack, make sure there is some
 488          * space left for the stack to grow (presently 4 pages).
 489          */
 490         if (addr < current->mm->start_stack &&
 491             addr > current->mm->start_stack - PAGE_SIZE*(shp->shm_npages + 4))
 492         {
 493                 /* printk("shmat() -> EINVAL because segment intersects stack\n"); */
 494                 return -EINVAL;
 495         }
 496         if (!(shmflg & SHM_REMAP))
 497                 if ((shmd = find_vma_intersection(current, addr, addr + shp->shm_segsz))) {
 498                         /* printk("shmat() -> EINVAL because the interval [0x%lx,0x%lx) intersects an already mapped interval [0x%lx,0x%lx).\n",
 499                                 addr, addr + shp->shm_segsz, shmd->vm_start, shmd->vm_end); */
 500                         return -EINVAL;
 501                 }
 502 
 503         if (ipcperms(&shp->shm_perm, shmflg & SHM_RDONLY ? S_IRUGO : S_IRUGO|S_IWUGO))
 504                 return -EACCES;
 505         if (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)
 506                 return -EIDRM;
 507 
 508         shmd = (struct vm_area_struct *) kmalloc (sizeof(*shmd), GFP_KERNEL);
 509         if (!shmd)
 510                 return -ENOMEM;
 511         if ((shp != shm_segs[id]) || (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)) {
 512                 kfree(shmd);
 513                 return -EIDRM;
 514         }
 515 
 516         shmd->vm_pte = SWP_ENTRY(SHM_SWP_TYPE, id);
 517         shmd->vm_start = addr;
 518         shmd->vm_end = addr + shp->shm_npages * PAGE_SIZE;
 519         shmd->vm_mm = current->mm;
 520         shmd->vm_page_prot = (shmflg & SHM_RDONLY) ? PAGE_READONLY : PAGE_SHARED;
 521         shmd->vm_flags = VM_SHM | VM_MAYSHARE | VM_SHARED
 522                          | VM_MAYREAD | VM_MAYEXEC | VM_READ | VM_EXEC
 523                          | ((shmflg & SHM_RDONLY) ? 0 : VM_MAYWRITE | VM_WRITE);
 524         shmd->vm_next_share = shmd->vm_prev_share = NULL;
 525         shmd->vm_inode = NULL;
 526         shmd->vm_offset = 0;
 527         shmd->vm_ops = &shm_vm_ops;
 528 
 529         shp->shm_nattch++;            /* prevent destruction */
 530         if ((err = shm_map (shmd))) {
 531                 if (--shp->shm_nattch <= 0 && shp->shm_perm.mode & SHM_DEST)
 532                         killseg(id);
 533                 kfree(shmd);
 534                 return err;
 535         }
 536 
 537         insert_attach(shp,shmd);  /* insert shmd into shp->attaches */
 538 
 539         shp->shm_lpid = current->pid;
 540         shp->shm_atime = CURRENT_TIME;
 541 
 542         *raddr = addr;
 543         return 0;
 544 }
 545 
 546 /* This is called by fork, once for every shm attach. */
 547 static void shm_open (struct vm_area_struct *shmd)
     /*  */
 548 {
 549         unsigned int id;
 550         struct shmid_ds *shp;
 551 
 552         id = SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK;
 553         shp = shm_segs[id];
 554         if (shp == IPC_UNUSED) {
 555                 printk("shm_open: unused id=%d PANIC\n", id);
 556                 return;
 557         }
 558         insert_attach(shp,shmd);  /* insert shmd into shp->attaches */
 559         shp->shm_nattch++;
 560         shp->shm_atime = CURRENT_TIME;
 561         shp->shm_lpid = current->pid;
 562 }
 563 
 564 /*
 565  * remove the attach descriptor shmd.
 566  * free memory for segment if it is marked destroyed.
 567  * The descriptor has already been removed from the current->mm->mmap list
 568  * and will later be kfree()d.
 569  */
 570 static void shm_close (struct vm_area_struct *shmd)
     /*  */
 571 {
 572         struct shmid_ds *shp;
 573         int id;
 574 
 575         /* remove from the list of attaches of the shm segment */
 576         id = SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK;
 577         shp = shm_segs[id];
 578         remove_attach(shp,shmd);  /* remove from shp->attaches */
 579         shp->shm_lpid = current->pid;
 580         shp->shm_dtime = CURRENT_TIME;
 581         if (--shp->shm_nattch <= 0 && shp->shm_perm.mode & SHM_DEST)
 582                 killseg (id);
 583 }
 584 
 585 /*
 586  * detach and kill segment if marked destroyed.
 587  * The work is done in shm_close.
 588  */
 589 asmlinkage int sys_shmdt (char *shmaddr)
     /*  */
 590 {
 591         struct vm_area_struct *shmd, *shmdnext;
 592 
 593         for (shmd = current->mm->mmap; shmd; shmd = shmdnext) {
 594                 shmdnext = shmd->vm_next;
 595                 if (shmd->vm_ops == &shm_vm_ops
 596                     && shmd->vm_start - shmd->vm_offset == (ulong) shmaddr)
 597                         do_munmap(shmd->vm_start, shmd->vm_end - shmd->vm_start);
 598         }
 599         return 0;
 600 }
 601 
 602 /*
 603  * page not present ... go through shm_pages
 604  */
 605 static pte_t shm_swap_in(struct vm_area_struct * shmd, unsigned long offset, unsigned long code)
     /*  */
 606 {
 607         pte_t pte;
 608         struct shmid_ds *shp;
 609         unsigned int id, idx;
 610 
 611         id = SWP_OFFSET(code) & SHM_ID_MASK;
 612         if (id != (SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK)) {
 613                 printk ("shm_swap_in: code id = %d and shmd id = %ld differ\n",
 614                         id, SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK);
 615                 return BAD_PAGE;
 616         }
 617         if (id > max_shmid) {
 618                 printk ("shm_swap_in: id=%d too big. proc mem corrupted\n", id);
 619                 return BAD_PAGE;
 620         }
 621         shp = shm_segs[id];
 622         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 623                 printk ("shm_swap_in: id=%d invalid. Race.\n", id);
 624                 return BAD_PAGE;
 625         }
 626         idx = (SWP_OFFSET(code) >> SHM_IDX_SHIFT) & SHM_IDX_MASK;
 627         if (idx != (offset >> PAGE_SHIFT)) {
 628                 printk ("shm_swap_in: code idx = %u and shmd idx = %lu differ\n",
 629                         idx, offset >> PAGE_SHIFT);
 630                 return BAD_PAGE;
 631         }
 632         if (idx >= shp->shm_npages) {
 633                 printk ("shm_swap_in : too large page index. id=%d\n", id);
 634                 return BAD_PAGE;
 635         }
 636 
 637         pte_val(pte) = shp->shm_pages[idx];
 638         if (!pte_present(pte)) {
 639                 unsigned long page = get_free_page(GFP_KERNEL);
 640                 if (!page) {
 641                         oom(current);
 642                         return BAD_PAGE;
 643                 }
 644                 pte_val(pte) = shp->shm_pages[idx];
 645                 if (pte_present(pte)) {
 646                         free_page (page); /* doesn't sleep */
 647                         goto done;
 648                 }
 649                 if (!pte_none(pte)) {
 650                         read_swap_page(pte_val(pte), (char *) page);
 651                         pte_val(pte) = shp->shm_pages[idx];
 652                         if (pte_present(pte))  {
 653                                 free_page (page); /* doesn't sleep */
 654                                 goto done;
 655                         }
 656                         swap_free(pte_val(pte));
 657                         shm_swp--;
 658                 }
 659                 shm_rss++;
 660                 pte = pte_mkdirty(mk_pte(page, PAGE_SHARED));
 661                 shp->shm_pages[idx] = pte_val(pte);
 662         } else
 663                 --current->maj_flt;  /* was incremented in do_no_page */
 664 
 665 done:   /* pte_val(pte) == shp->shm_pages[idx] */
 666         current->min_flt++;
 667         mem_map[MAP_NR(pte_page(pte))].count++;
 668         return pte_modify(pte, shmd->vm_page_prot);
 669 }
 670 
 671 /*
 672  * Goes through counter = (shm_rss >> prio) present shm pages.
 673  */
 674 static unsigned long swap_id = 0; /* currently being swapped */
 675 static unsigned long swap_idx = 0; /* next to swap */
 676 
 677 int shm_swap (int prio, unsigned long limit)
     /*  */
 678 {
 679         pte_t page;
 680         struct shmid_ds *shp;
 681         struct vm_area_struct *shmd;
 682         unsigned long swap_nr;
 683         unsigned long id, idx;
 684         int loop = 0;
 685         int counter;
 686         
 687         counter = shm_rss >> prio;
 688         if (!counter || !(swap_nr = get_swap_page()))
 689                 return 0;
 690 
 691  check_id:
 692         shp = shm_segs[swap_id];
 693         if (shp == IPC_UNUSED || shp == IPC_NOID || shp->shm_perm.mode & SHM_LOCKED ) {
 694                 next_id:
 695                 swap_idx = 0;
 696                 if (++swap_id > max_shmid) {
 697                         if (loop)
 698                                 goto failed;
 699                         loop = 1;
 700                         swap_id = 0;
 701                 }
 702                 goto check_id;
 703         }
 704         id = swap_id;
 705 
 706  check_table:
 707         idx = swap_idx++;
 708         if (idx >= shp->shm_npages)
 709                 goto next_id;
 710 
 711         pte_val(page) = shp->shm_pages[idx];
 712         if (!pte_present(page))
 713                 goto check_table;
 714         if (pte_page(page) >= limit)
 715                 goto check_table;
 716         swap_attempts++;
 717 
 718         if (--counter < 0) { /* failed */
 719                 failed:
 720                 swap_free (swap_nr);
 721                 return 0;
 722         }
 723         if (shp->attaches)
 724           for (shmd = shp->attaches; ; ) {
 725             do {
 726                 pgd_t *page_dir;
 727                 pmd_t *page_middle;
 728                 pte_t *page_table, pte;
 729                 unsigned long tmp;
 730 
 731                 if ((SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK) != id) {
 732                         printk ("shm_swap: id=%ld does not match shmd->vm_pte.id=%ld\n",
 733                                 id, SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK);
 734                         continue;
 735                 }
 736                 tmp = shmd->vm_start + (idx << PAGE_SHIFT) - shmd->vm_offset;
 737                 if (!(tmp >= shmd->vm_start && tmp < shmd->vm_end))
 738                         continue;
 739                 page_dir = pgd_offset(shmd->vm_mm,tmp);
 740                 if (pgd_none(*page_dir) || pgd_bad(*page_dir)) {
 741                         printk("shm_swap: bad pgtbl! id=%ld start=%lx idx=%ld\n",
 742                                         id, shmd->vm_start, idx);
 743                         pgd_clear(page_dir);
 744                         continue;
 745                 }
 746                 page_middle = pmd_offset(page_dir,tmp);
 747                 if (pmd_none(*page_middle) || pmd_bad(*page_middle)) {
 748                         printk("shm_swap: bad pgmid! id=%ld start=%lx idx=%ld\n",
 749                                         id, shmd->vm_start, idx);
 750                         pmd_clear(page_middle);
 751                         continue;
 752                 }
 753                 page_table = pte_offset(page_middle,tmp);
 754                 pte = *page_table;
 755                 if (!pte_present(pte))
 756                         continue;
 757                 if (pte_young(pte)) {
 758                         set_pte(page_table, pte_mkold(pte));
 759                         continue;
 760                 }
 761                 if (pte_page(pte) != pte_page(page))
 762                         printk("shm_swap_out: page and pte mismatch\n");
 763                 set_pte(page_table,
 764                   __pte(shmd->vm_pte + SWP_ENTRY(0, idx << SHM_IDX_SHIFT)));
 765                 mem_map[MAP_NR(pte_page(pte))].count--;
 766                 if (shmd->vm_mm->rss > 0)
 767                         shmd->vm_mm->rss--;
 768                 invalidate_page(shmd, tmp);
 769             /* continue looping through circular list */
 770             } while (0);
 771             if ((shmd = shmd->vm_next_share) == shp->attaches)
 772                 break;
 773         }
 774 
 775         if (mem_map[MAP_NR(pte_page(page))].count != 1)
 776                 goto check_table;
 777         shp->shm_pages[idx] = swap_nr;
 778         write_swap_page (swap_nr, (char *) pte_page(page));
 779         free_page(pte_page(page));
 780         swap_successes++;
 781         shm_swp++;
 782         shm_rss--;
 783         return 1;
 784 }

/* */