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 
  16 #include <asm/segment.h>
  17 #include <asm/pgtable.h>
  18 
  19 extern int ipcperms (struct ipc_perm *ipcp, short shmflg);
  20 extern unsigned int get_swap_page (void);
  21 static int findkey (key_t key);
  22 static int newseg (key_t key, int shmflg, int size);
  23 static int shm_map (struct vm_area_struct *shmd, int remap);
  24 static void killseg (int id);
  25 static void shm_open (struct vm_area_struct *shmd);
  26 static void shm_close (struct vm_area_struct *shmd);
  27 static pte_t shm_swap_in(struct vm_area_struct *, unsigned long, unsigned long);
  28 
  29 static int shm_tot = 0; /* total number of shared memory pages */
  30 static int shm_rss = 0; /* number of shared memory pages that are in memory */
  31 static int shm_swp = 0; /* number of shared memory pages that are in swap */
  32 static int max_shmid = 0; /* every used id is <= max_shmid */
  33 static struct wait_queue *shm_lock = NULL; /* calling findkey() may need to wait */
  34 static struct shmid_ds *shm_segs[SHMMNI];
  35 
  36 static unsigned short shm_seq = 0; /* incremented, for recognizing stale ids */
  37 
  38 /* some statistics */
  39 static ulong swap_attempts = 0;
  40 static ulong swap_successes = 0;
  41 static ulong used_segs = 0;
  42 
  43 void shm_init (void)
     /*  */
  44 {
  45         int id;
  46 
  47         for (id = 0; id < SHMMNI; id++)
  48                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
  49         shm_tot = shm_rss = shm_seq = max_shmid = used_segs = 0;
  50         shm_lock = NULL;
  51         return;
  52 }
  53 
  54 static int findkey (key_t key)
     /*  */
  55 {
  56         int id;
  57         struct shmid_ds *shp;
  58 
  59         for (id = 0; id <= max_shmid; id++) {
  60                 while ((shp = shm_segs[id]) == IPC_NOID)
  61                         sleep_on (&shm_lock);
  62                 if (shp == IPC_UNUSED)
  63                         continue;
  64                 if (key == shp->shm_perm.key)
  65                         return id;
  66         }
  67         return -1;
  68 }
  69 
  70 /*
  71  * allocate new shmid_ds and pgtable. protected by shm_segs[id] = NOID.
  72  */
  73 static int newseg (key_t key, int shmflg, int size)
     /*  */
  74 {
  75         struct shmid_ds *shp;
  76         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
  77         int id, i;
  78 
  79         if (size < SHMMIN)
  80                 return -EINVAL;
  81         if (shm_tot + numpages >= SHMALL)
  82                 return -ENOSPC;
  83         for (id = 0; id < SHMMNI; id++)
  84                 if (shm_segs[id] == IPC_UNUSED) {
  85                         shm_segs[id] = (struct shmid_ds *) IPC_NOID;
  86                         goto found;
  87                 }
  88         return -ENOSPC;
  89 
  90 found:
  91         shp = (struct shmid_ds *) kmalloc (sizeof (*shp), GFP_KERNEL);
  92         if (!shp) {
  93                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
  94                 if (shm_lock)
  95                         wake_up (&shm_lock);
  96                 return -ENOMEM;
  97         }
  98 
  99         shp->shm_pages = (ulong *) kmalloc (numpages*sizeof(ulong),GFP_KERNEL);
 100         if (!shp->shm_pages) {
 101                 shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
 102                 if (shm_lock)
 103                         wake_up (&shm_lock);
 104                 kfree(shp);
 105                 return -ENOMEM;
 106         }
 107 
 108         for (i = 0; i < numpages; shp->shm_pages[i++] = 0);
 109         shm_tot += numpages;
 110         shp->shm_perm.key = key;
 111         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 112         shp->shm_perm.cuid = shp->shm_perm.uid = current->euid;
 113         shp->shm_perm.cgid = shp->shm_perm.gid = current->egid;
 114         shp->shm_perm.seq = shm_seq;
 115         shp->shm_segsz = size;
 116         shp->shm_cpid = current->pid;
 117         shp->attaches = NULL;
 118         shp->shm_lpid = shp->shm_nattch = 0;
 119         shp->shm_atime = shp->shm_dtime = 0;
 120         shp->shm_ctime = CURRENT_TIME;
 121         shp->shm_npages = numpages;
 122 
 123         if (id > max_shmid)
 124                 max_shmid = id;
 125         shm_segs[id] = shp;
 126         used_segs++;
 127         if (shm_lock)
 128                 wake_up (&shm_lock);
 129         return (unsigned int) shp->shm_perm.seq * SHMMNI + id;
 130 }
 131 
 132 int sys_shmget (key_t key, int size, int shmflg)
     /*  */
 133 {
 134         struct shmid_ds *shp;
 135         int id = 0;
 136 
 137         if (size < 0 || size > SHMMAX)
 138                 return -EINVAL;
 139         if (key == IPC_PRIVATE)
 140                 return newseg(key, shmflg, size);
 141         if ((id = findkey (key)) == -1) {
 142                 if (!(shmflg & IPC_CREAT))
 143                         return -ENOENT;
 144                 return newseg(key, shmflg, size);
 145         }
 146         if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL))
 147                 return -EEXIST;
 148         shp = shm_segs[id];
 149         if (shp->shm_perm.mode & SHM_DEST)
 150                 return -EIDRM;
 151         if (size > shp->shm_segsz)
 152                 return -EINVAL;
 153         if (ipcperms (&shp->shm_perm, shmflg))
 154                 return -EACCES;
 155         return (unsigned int) shp->shm_perm.seq * SHMMNI + id;
 156 }
 157 
 158 /*
 159  * Only called after testing nattch and SHM_DEST.
 160  * Here pages, pgtable and shmid_ds are freed.
 161  */
 162 static void killseg (int id)
     /*  */
 163 {
 164         struct shmid_ds *shp;
 165         int i, numpages;
 166 
 167         shp = shm_segs[id];
 168         if (shp == IPC_NOID || shp == IPC_UNUSED) {
 169                 printk ("shm nono: killseg called on unused seg id=%d\n", id);
 170                 return;
 171         }
 172         shp->shm_perm.seq++;     /* for shmat */
 173         shm_seq = (shm_seq+1) % ((unsigned)(1<<31)/SHMMNI); /* increment, but avoid overflow */
 174         shm_segs[id] = (struct shmid_ds *) IPC_UNUSED;
 175         used_segs--;
 176         if (id == max_shmid)
 177                 while (max_shmid && (shm_segs[--max_shmid] == IPC_UNUSED));
 178         if (!shp->shm_pages) {
 179                 printk ("shm nono: killseg shp->pages=NULL. id=%d\n", id);
 180                 return;
 181         }
 182         numpages = shp->shm_npages;
 183         for (i = 0; i < numpages ; i++) {
 184                 pte_t pte;
 185                 pte_val(pte) = shp->shm_pages[i];
 186                 if (pte_none(pte))
 187                         continue;
 188                 if (pte_present(pte)) {
 189                         free_page (pte_page(pte));
 190                         shm_rss--;
 191                 } else {
 192                         swap_free(pte_val(pte));
 193                         shm_swp--;
 194                 }
 195         }
 196         kfree(shp->shm_pages);
 197         shm_tot -= numpages;
 198         kfree(shp);
 199         return;
 200 }
 201 
 202 int sys_shmctl (int shmid, int cmd, struct shmid_ds *buf)
     /*  */
 203 {
 204         struct shmid_ds tbuf;
 205         struct shmid_ds *shp;
 206         struct ipc_perm *ipcp;
 207         int id, err;
 208 
 209         if (cmd < 0 || shmid < 0)
 210                 return -EINVAL;
 211         if (cmd == IPC_SET) {
 212                 if (!buf)
 213                         return -EFAULT;
 214                 err = verify_area (VERIFY_READ, buf, sizeof (*buf));
 215                 if (err)
 216                         return err;
 217                 memcpy_fromfs (&tbuf, buf, sizeof (*buf));
 218         }
 219 
 220         switch (cmd) { /* replace with proc interface ? */
 221         case IPC_INFO:
 222         {
 223                 struct shminfo shminfo;
 224                 if (!buf)
 225                         return -EFAULT;
 226                 shminfo.shmmni = SHMMNI;
 227                 shminfo.shmmax = SHMMAX;
 228                 shminfo.shmmin = SHMMIN;
 229                 shminfo.shmall = SHMALL;
 230                 shminfo.shmseg = SHMSEG;
 231                 err = verify_area (VERIFY_WRITE, buf, sizeof (struct shminfo));
 232                 if (err)
 233                         return err;
 234                 memcpy_tofs (buf, &shminfo, sizeof(struct shminfo));
 235                 return max_shmid;
 236         }
 237         case SHM_INFO:
 238         {
 239                 struct shm_info shm_info;
 240                 if (!buf)
 241                         return -EFAULT;
 242                 err = verify_area (VERIFY_WRITE, buf, sizeof (shm_info));
 243                 if (err)
 244                         return err;
 245                 shm_info.used_ids = used_segs;
 246                 shm_info.shm_rss = shm_rss;
 247                 shm_info.shm_tot = shm_tot;
 248                 shm_info.shm_swp = shm_swp;
 249                 shm_info.swap_attempts = swap_attempts;
 250                 shm_info.swap_successes = swap_successes;
 251                 memcpy_tofs (buf, &shm_info, sizeof(shm_info));
 252                 return max_shmid;
 253         }
 254         case SHM_STAT:
 255                 if (!buf)
 256                         return -EFAULT;
 257                 err = verify_area (VERIFY_WRITE, buf, sizeof (*buf));
 258                 if (err)
 259                         return err;
 260                 if (shmid > max_shmid)
 261                         return -EINVAL;
 262                 shp = shm_segs[shmid];
 263                 if (shp == IPC_UNUSED || shp == IPC_NOID)
 264                         return -EINVAL;
 265                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 266                         return -EACCES;
 267                 id = (unsigned int) shp->shm_perm.seq * SHMMNI + shmid;
 268                 tbuf.shm_perm   = shp->shm_perm;
 269                 tbuf.shm_segsz  = shp->shm_segsz;
 270                 tbuf.shm_atime  = shp->shm_atime;
 271                 tbuf.shm_dtime  = shp->shm_dtime;
 272                 tbuf.shm_ctime  = shp->shm_ctime;
 273                 tbuf.shm_cpid   = shp->shm_cpid;
 274                 tbuf.shm_lpid   = shp->shm_lpid;
 275                 tbuf.shm_nattch = shp->shm_nattch;
 276                 memcpy_tofs (buf, &tbuf, sizeof(*buf));
 277                 return id;
 278         }
 279 
 280         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 281         if (shp == IPC_UNUSED || shp == IPC_NOID)
 282                 return -EINVAL;
 283         if (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)
 284                 return -EIDRM;
 285         ipcp = &shp->shm_perm;
 286 
 287         switch (cmd) {
 288         case SHM_UNLOCK:
 289                 if (!suser())
 290                         return -EPERM;
 291                 if (!(ipcp->mode & SHM_LOCKED))
 292                         return -EINVAL;
 293                 ipcp->mode &= ~SHM_LOCKED;
 294                 break;
 295         case SHM_LOCK:
 296 /* Allow superuser to lock segment in memory */
 297 /* Should the pages be faulted in here or leave it to user? */
 298 /* need to determine interaction with current->swappable */
 299                 if (!suser())
 300                         return -EPERM;
 301                 if (ipcp->mode & SHM_LOCKED)
 302                         return -EINVAL;
 303                 ipcp->mode |= SHM_LOCKED;
 304                 break;
 305         case IPC_STAT:
 306                 if (ipcperms (ipcp, S_IRUGO))
 307                         return -EACCES;
 308                 if (!buf)
 309                         return -EFAULT;
 310                 err = verify_area (VERIFY_WRITE, buf, sizeof (*buf));
 311                 if (err)
 312                         return err;
 313                 tbuf.shm_perm   = shp->shm_perm;
 314                 tbuf.shm_segsz  = shp->shm_segsz;
 315                 tbuf.shm_atime  = shp->shm_atime;
 316                 tbuf.shm_dtime  = shp->shm_dtime;
 317                 tbuf.shm_ctime  = shp->shm_ctime;
 318                 tbuf.shm_cpid   = shp->shm_cpid;
 319                 tbuf.shm_lpid   = shp->shm_lpid;
 320                 tbuf.shm_nattch = shp->shm_nattch;
 321                 memcpy_tofs (buf, &tbuf, sizeof(*buf));
 322                 break;
 323         case IPC_SET:
 324                 if (suser() || current->euid == shp->shm_perm.uid ||
 325                     current->euid == shp->shm_perm.cuid) {
 326                         ipcp->uid = tbuf.shm_perm.uid;
 327                         ipcp->gid = tbuf.shm_perm.gid;
 328                         ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
 329                                 | (tbuf.shm_perm.mode & S_IRWXUGO);
 330                         shp->shm_ctime = CURRENT_TIME;
 331                         break;
 332                 }
 333                 return -EPERM;
 334         case IPC_RMID:
 335                 if (suser() || current->euid == shp->shm_perm.uid ||
 336                     current->euid == shp->shm_perm.cuid) {
 337                         shp->shm_perm.mode |= SHM_DEST;
 338                         if (shp->shm_nattch <= 0)
 339                                 killseg (id);
 340                         break;
 341                 }
 342                 return -EPERM;
 343         default:
 344                 return -EINVAL;
 345         }
 346         return 0;
 347 }
 348 
 349 /*
 350  * The per process internal structure for managing segments is
 351  * `struct vm_area_struct'.
 352  * A shmat will add to and shmdt will remove from the list.
 353  * shmd->vm_task        the attacher
 354  * shmd->vm_start       virt addr of attach, multiple of SHMLBA
 355  * shmd->vm_end         multiple of SHMLBA
 356  * shmd->vm_next        next attach for task
 357  * shmd->vm_next_share  next attach for segment
 358  * shmd->vm_offset      offset into segment
 359  * shmd->vm_pte         signature for this attach
 360  */
 361 
 362 static struct vm_operations_struct shm_vm_ops = {
 363         shm_open,               /* open */
 364         shm_close,              /* close */
 365         NULL,                   /* unmap */
 366         NULL,                   /* protect */
 367         NULL,                   /* sync */
 368         NULL,                   /* advise */
 369         NULL,                   /* nopage (done with swapin) */
 370         NULL,                   /* wppage */
 371         NULL,                   /* swapout (hardcoded right now) */
 372         shm_swap_in             /* swapin */
 373 };
 374 
 375 /* Insert shmd into the circular list shp->attaches */
 376 static inline void insert_attach (struct shmid_ds * shp, struct vm_area_struct * shmd)
     /*  */
 377 {
 378         struct vm_area_struct * attaches;
 379 
 380         if ((attaches = shp->attaches)) {
 381                 shmd->vm_next_share = attaches;
 382                 shmd->vm_prev_share = attaches->vm_prev_share;
 383                 shmd->vm_prev_share->vm_next_share = shmd;
 384                 attaches->vm_prev_share = shmd;
 385         } else
 386                 shp->attaches = shmd->vm_next_share = shmd->vm_prev_share = shmd;
 387 }
 388 
 389 /* Remove shmd from circular list shp->attaches */
 390 static inline void remove_attach (struct shmid_ds * shp, struct vm_area_struct * shmd)
     /*  */
 391 {
 392         if (shmd->vm_next_share == shmd) {
 393                 if (shp->attaches != shmd) {
 394                         printk("shm_close: shm segment (id=%ld) attach list inconsistent\n",
 395                                 (shmd->vm_pte >> SHM_ID_SHIFT) & SHM_ID_MASK);
 396                         printk("shm_close: %d %08lx-%08lx %c%c%c%c %08lx %08lx\n",
 397                                 shmd->vm_task->pid, shmd->vm_start, shmd->vm_end,
 398                                 shmd->vm_flags & VM_READ ? 'r' : '-',
 399                                 shmd->vm_flags & VM_WRITE ? 'w' : '-',
 400                                 shmd->vm_flags & VM_EXEC ? 'x' : '-',
 401                                 shmd->vm_flags & VM_MAYSHARE ? 's' : 'p',
 402                                 shmd->vm_offset, shmd->vm_pte);
 403                 }
 404                 shp->attaches = NULL;
 405         } else {
 406                 if (shp->attaches == shmd)
 407                         shp->attaches = shmd->vm_next_share;
 408                 shmd->vm_prev_share->vm_next_share = shmd->vm_next_share;
 409                 shmd->vm_next_share->vm_prev_share = shmd->vm_prev_share;
 410         }
 411 }
 412 
 413 /*
 414  * check range is unmapped, ensure page tables exist
 415  * mark page table entries with shm_sgn.
 416  * if remap != 0 the range is remapped.
 417  */
 418 static int shm_map (struct vm_area_struct *shmd, int remap)
     /*  */
 419 {
 420         pgd_t *page_dir;
 421         pte_t *page_table;
 422         unsigned long tmp, shm_sgn;
 423 
 424         /* check that the range is unmapped */
 425         if (!remap)
 426                 for (tmp = shmd->vm_start; tmp < shmd->vm_end; tmp += PAGE_SIZE) {
 427                         page_dir = PAGE_DIR_OFFSET(shmd->vm_task,tmp);
 428                         if (pgd_none(*page_dir))
 429                                 continue;
 430                         if (pgd_bad(*page_dir)) {
 431                                 printk("bad ipc page directory entry %08lx\n", pgd_val(*page_dir));
 432                                 pgd_clear(page_dir);
 433                                 continue;
 434                         }
 435                         page_table = (pte_t *) pgd_page(*page_dir);     
 436                         page_table += ((tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1));
 437                         if (!pte_none(*page_table)) {
 438                                 /* printk("shmat() -> EINVAL because address 0x%lx is already mapped.\n",tmp); */
 439                                 return -EINVAL;
 440                         }
 441                 }
 442 
 443         /* clear old mappings */
 444         do_munmap(shmd->vm_start, shmd->vm_end - shmd->vm_start);
 445 
 446         /* add new mapping */
 447         insert_vm_struct(current, shmd);
 448         merge_segments(current, shmd->vm_start, shmd->vm_end);
 449 
 450         /* check that the range has page_tables */
 451         for (tmp = shmd->vm_start; tmp < shmd->vm_end; tmp += PAGE_SIZE) {
 452                 page_dir = PAGE_DIR_OFFSET(shmd->vm_task,tmp);
 453                 if (!pgd_none(*page_dir)) {
 454                         page_table = (pte_t *) pgd_page(*page_dir);
 455                         page_table += ((tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1));
 456                         if (!pte_none(*page_table)) {
 457                                 if (pte_present(*page_table)) {
 458                                         --current->mm->rss;
 459                                         free_page (pte_page(*page_table));
 460                                 } else
 461                                         swap_free(pte_val(*page_table));
 462                                 pte_clear(page_table);
 463                         }
 464                 } else {
 465                         if (!(page_table = (pte_t *) get_free_page(GFP_KERNEL)))
 466                                 return -ENOMEM;
 467                         pgd_set(page_dir, page_table);
 468                         tmp |= (PGDIR_SIZE - PAGE_SIZE);
 469                 }
 470         }
 471 
 472         /* map page range */
 473         shm_sgn = shmd->vm_pte + ((shmd->vm_offset >> PAGE_SHIFT) << SHM_IDX_SHIFT);
 474         for (tmp = shmd->vm_start; tmp < shmd->vm_end; tmp += PAGE_SIZE,
 475              shm_sgn += (1 << SHM_IDX_SHIFT)) {
 476                 page_dir = PAGE_DIR_OFFSET(shmd->vm_task,tmp);
 477                 page_table = (pte_t *) pgd_page(*page_dir);
 478                 page_table += (tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
 479                 pte_val(*page_table) = shm_sgn;
 480         }
 481         invalidate();
 482         return 0;
 483 }
 484 
 485 /*
 486  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 487  */
 488 int sys_shmat (int shmid, char *shmaddr, int shmflg, ulong *raddr)
     /*  */
 489 {
 490         struct shmid_ds *shp;
 491         struct vm_area_struct *shmd;
 492         int err;
 493         unsigned int id;
 494         unsigned long addr;
 495 
 496         if (shmid < 0) {
 497                 /* printk("shmat() -> EINVAL because shmid = %d < 0\n",shmid); */
 498                 return -EINVAL;
 499         }
 500 
 501         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 502         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 503                 /* printk("shmat() -> EINVAL because shmid = %d is invalid\n",shmid); */
 504                 return -EINVAL;
 505         }
 506 
 507         if (!(addr = (ulong) shmaddr)) {
 508                 if (shmflg & SHM_REMAP)
 509                         return -EINVAL;
 510                 if (!(addr = get_unmapped_area(shp->shm_segsz)))
 511                         return -ENOMEM;
 512         } else if (addr & (SHMLBA-1)) {
 513                 if (shmflg & SHM_RND)
 514                         addr &= ~(SHMLBA-1);       /* round down */
 515                 else
 516                         return -EINVAL;
 517         }
 518         if ((addr > current->mm->start_stack - 16384 - PAGE_SIZE*shp->shm_npages)) {
 519                 /* printk("shmat() -> EINVAL because segment intersects stack\n"); */
 520                 return -EINVAL;
 521         }
 522         if (!(shmflg & SHM_REMAP))
 523                 if ((shmd = find_vma_intersection(current, addr, addr + shp->shm_segsz))) {
 524                         /* printk("shmat() -> EINVAL because the interval [0x%lx,0x%lx) intersects an already mapped interval [0x%lx,0x%lx).\n",
 525                                 addr, addr + shp->shm_segsz, shmd->vm_start, shmd->vm_end); */
 526                         return -EINVAL;
 527                 }
 528 
 529         if (ipcperms(&shp->shm_perm, shmflg & SHM_RDONLY ? S_IRUGO : S_IRUGO|S_IWUGO))
 530                 return -EACCES;
 531         if (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)
 532                 return -EIDRM;
 533 
 534         shmd = (struct vm_area_struct *) kmalloc (sizeof(*shmd), GFP_KERNEL);
 535         if (!shmd)
 536                 return -ENOMEM;
 537         if ((shp != shm_segs[id]) || (shp->shm_perm.seq != (unsigned int) shmid / SHMMNI)) {
 538                 kfree(shmd);
 539                 return -EIDRM;
 540         }
 541 
 542         shmd->vm_pte = (SHM_SWP_TYPE << 1) | (id << SHM_ID_SHIFT);
 543         shmd->vm_start = addr;
 544         shmd->vm_end = addr + shp->shm_npages * PAGE_SIZE;
 545         shmd->vm_task = current;
 546         shmd->vm_page_prot = (shmflg & SHM_RDONLY) ? PAGE_READONLY : PAGE_SHARED;
 547         shmd->vm_flags = VM_SHM | VM_MAYSHARE | VM_SHARED
 548                          | VM_MAYREAD | VM_MAYEXEC | VM_READ | VM_EXEC
 549                          | ((shmflg & SHM_RDONLY) ? 0 : VM_MAYWRITE | VM_WRITE);
 550         shmd->vm_next_share = shmd->vm_prev_share = NULL;
 551         shmd->vm_inode = NULL;
 552         shmd->vm_offset = 0;
 553         shmd->vm_ops = &shm_vm_ops;
 554 
 555         shp->shm_nattch++;            /* prevent destruction */
 556         if ((err = shm_map (shmd, shmflg & SHM_REMAP))) {
 557                 if (--shp->shm_nattch <= 0 && shp->shm_perm.mode & SHM_DEST)
 558                         killseg(id);
 559                 kfree(shmd);
 560                 return err;
 561         }
 562 
 563         insert_attach(shp,shmd);  /* insert shmd into shp->attaches */
 564 
 565         shp->shm_lpid = current->pid;
 566         shp->shm_atime = CURRENT_TIME;
 567 
 568         *raddr = addr;
 569         return 0;
 570 }
 571 
 572 /* This is called by fork, once for every shm attach. */
 573 static void shm_open (struct vm_area_struct *shmd)
     /*  */
 574 {
 575         unsigned int id;
 576         struct shmid_ds *shp;
 577 
 578         id = (shmd->vm_pte >> SHM_ID_SHIFT) & SHM_ID_MASK;
 579         shp = shm_segs[id];
 580         if (shp == IPC_UNUSED) {
 581                 printk("shm_open: unused id=%d PANIC\n", id);
 582                 return;
 583         }
 584         insert_attach(shp,shmd);  /* insert shmd into shp->attaches */
 585         shp->shm_nattch++;
 586         shp->shm_atime = CURRENT_TIME;
 587         shp->shm_lpid = current->pid;
 588 }
 589 
 590 /*
 591  * remove the attach descriptor shmd.
 592  * free memory for segment if it is marked destroyed.
 593  * The descriptor has already been removed from the current->mm->mmap list
 594  * and will later be kfree()d.
 595  */
 596 static void shm_close (struct vm_area_struct *shmd)
     /*  */
 597 {
 598         struct shmid_ds *shp;
 599         int id;
 600 
 601         unmap_page_range (shmd->vm_start, shmd->vm_end - shmd->vm_start);
 602 
 603         /* remove from the list of attaches of the shm segment */
 604         id = (shmd->vm_pte >> SHM_ID_SHIFT) & SHM_ID_MASK;
 605         shp = shm_segs[id];
 606         remove_attach(shp,shmd);  /* remove from shp->attaches */
 607         shp->shm_lpid = current->pid;
 608         shp->shm_dtime = CURRENT_TIME;
 609         if (--shp->shm_nattch <= 0 && shp->shm_perm.mode & SHM_DEST)
 610                 killseg (id);
 611 }
 612 
 613 /*
 614  * detach and kill segment if marked destroyed.
 615  * The work is done in shm_close.
 616  */
 617 int sys_shmdt (char *shmaddr)
     /*  */
 618 {
 619         struct vm_area_struct *shmd, *shmdnext;
 620 
 621         for (shmd = current->mm->mmap; shmd; shmd = shmdnext) {
 622                 shmdnext = shmd->vm_next;
 623                 if (shmd->vm_ops == &shm_vm_ops
 624                     && shmd->vm_start - shmd->vm_offset == (ulong) shmaddr)
 625                         do_munmap(shmd->vm_start, shmd->vm_end - shmd->vm_start);
 626         }
 627         return 0;
 628 }
 629 
 630 /*
 631  * page not present ... go through shm_pages
 632  */
 633 static pte_t shm_swap_in(struct vm_area_struct * shmd, unsigned long offset, unsigned long code)
     /*  */
 634 {
 635         pte_t pte;
 636         struct shmid_ds *shp;
 637         unsigned int id, idx;
 638 
 639         id = (code >> SHM_ID_SHIFT) & SHM_ID_MASK;
 640         if (id != ((shmd->vm_pte >> SHM_ID_SHIFT) & SHM_ID_MASK)) {
 641                 printk ("shm_swap_in: code id = %d and shmd id = %ld differ\n",
 642                         id, (shmd->vm_pte >> SHM_ID_SHIFT) & SHM_ID_MASK);
 643                 return BAD_PAGE;
 644         }
 645         if (id > max_shmid) {
 646                 printk ("shm_swap_in: id=%d too big. proc mem corrupted\n", id);
 647                 return BAD_PAGE;
 648         }
 649         shp = shm_segs[id];
 650         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 651                 printk ("shm_swap_in: id=%d invalid. Race.\n", id);
 652                 return BAD_PAGE;
 653         }
 654         idx = (code >> SHM_IDX_SHIFT) & SHM_IDX_MASK;
 655         if (idx != (offset >> PAGE_SHIFT)) {
 656                 printk ("shm_swap_in: code idx = %u and shmd idx = %lu differ\n",
 657                         idx, offset >> PAGE_SHIFT);
 658                 return BAD_PAGE;
 659         }
 660         if (idx >= shp->shm_npages) {
 661                 printk ("shm_swap_in : too large page index. id=%d\n", id);
 662                 return BAD_PAGE;
 663         }
 664 
 665         pte_val(pte) = shp->shm_pages[idx];
 666         if (!pte_present(pte)) {
 667                 unsigned long page = get_free_page(GFP_KERNEL);
 668                 if (!page) {
 669                         oom(current);
 670                         return BAD_PAGE;
 671                 }
 672                 pte_val(pte) = shp->shm_pages[idx];
 673                 if (pte_present(pte)) {
 674                         free_page (page); /* doesn't sleep */
 675                         goto done;
 676                 }
 677                 if (!pte_none(pte)) {
 678                         read_swap_page(pte_val(pte), (char *) page);
 679                         pte_val(pte) = shp->shm_pages[idx];
 680                         if (pte_present(pte))  {
 681                                 free_page (page); /* doesn't sleep */
 682                                 goto done;
 683                         }
 684                         swap_free(pte_val(pte));
 685                         shm_swp--;
 686                 }
 687                 shm_rss++;
 688                 pte = pte_mkdirty(mk_pte(page, PAGE_SHARED));
 689                 shp->shm_pages[idx] = pte_val(pte);
 690         } else
 691                 --current->mm->maj_flt;  /* was incremented in do_no_page */
 692 
 693 done:   /* pte_val(pte) == shp->shm_pages[idx] */
 694         current->mm->min_flt++;
 695         mem_map[MAP_NR(pte_page(pte))]++;
 696         return pte_modify(pte, shmd->vm_page_prot);
 697 }
 698 
 699 /*
 700  * Goes through counter = (shm_rss << prio) present shm pages.
 701  */
 702 static unsigned long swap_id = 0; /* currently being swapped */
 703 static unsigned long swap_idx = 0; /* next to swap */
 704 
 705 int shm_swap (int prio)
     /*  */
 706 {
 707         pte_t page;
 708         struct shmid_ds *shp;
 709         struct vm_area_struct *shmd;
 710         unsigned int swap_nr;
 711         unsigned long id, idx, invalid = 0;
 712         int counter;
 713 
 714         counter = shm_rss >> prio;
 715         if (!counter || !(swap_nr = get_swap_page()))
 716                 return 0;
 717 
 718  check_id:
 719         shp = shm_segs[swap_id];
 720         if (shp == IPC_UNUSED || shp == IPC_NOID || shp->shm_perm.mode & SHM_LOCKED ) {
 721                 swap_idx = 0;
 722                 if (++swap_id > max_shmid)
 723                         swap_id = 0;
 724                 goto check_id;
 725         }
 726         id = swap_id;
 727 
 728  check_table:
 729         idx = swap_idx++;
 730         if (idx >= shp->shm_npages) {
 731                 swap_idx = 0;
 732                 if (++swap_id > max_shmid)
 733                         swap_id = 0;
 734                 goto check_id;
 735         }
 736 
 737         pte_val(page) = shp->shm_pages[idx];
 738         if (!pte_present(page))
 739                 goto check_table;
 740         swap_attempts++;
 741 
 742         if (--counter < 0) { /* failed */
 743                 if (invalid)
 744                         invalidate();
 745                 swap_free (swap_nr);
 746                 return 0;
 747         }
 748         if (shp->attaches)
 749           for (shmd = shp->attaches; ; ) {
 750             do {
 751                 pgd_t *page_dir;
 752                 pte_t *page_table, pte;
 753                 unsigned long tmp;
 754 
 755                 if ((shmd->vm_pte >> SHM_ID_SHIFT & SHM_ID_MASK) != id) {
 756                         printk ("shm_swap: id=%ld does not match shmd->vm_pte.id=%ld\n", id, shmd->vm_pte >> SHM_ID_SHIFT & SHM_ID_MASK);
 757                         continue;
 758                 }
 759                 tmp = shmd->vm_start + (idx << PAGE_SHIFT) - shmd->vm_offset;
 760                 if (!(tmp >= shmd->vm_start && tmp < shmd->vm_end))
 761                         continue;
 762                 page_dir = PAGE_DIR_OFFSET(shmd->vm_task,tmp);
 763                 if (pgd_none(*page_dir) || pgd_bad(*page_dir)) {
 764                         printk("shm_swap: bad pgtbl! id=%ld start=%lx idx=%ld\n",
 765                                         id, shmd->vm_start, idx);
 766                         pgd_clear(page_dir);
 767                         continue;
 768                 }
 769                 page_table = (pte_t *) pgd_page(*page_dir);
 770                 page_table += ((tmp >> PAGE_SHIFT) & (PTRS_PER_PAGE-1));
 771                 pte = *page_table;
 772                 if (!pte_present(pte))
 773                         continue;
 774                 if (pte_young(pte)) {
 775                         *page_table = pte_mkold(pte);
 776                         continue;
 777                 }
 778                 if (pte_page(pte) != pte_page(page))
 779                         printk("shm_swap_out: page and pte mismatch\n");
 780                 pte_val(*page_table) = shmd->vm_pte | idx << SHM_IDX_SHIFT;
 781                 mem_map[MAP_NR(pte_page(pte))]--;
 782                 shmd->vm_task->mm->rss--;
 783                 invalid++;
 784             /* continue looping through circular list */
 785             } while (0);
 786             if ((shmd = shmd->vm_next_share) == shp->attaches)
 787                 break;
 788         }
 789 
 790         if (mem_map[MAP_NR(pte_page(page))] != 1)
 791                 goto check_table;
 792         shp->shm_pages[idx] = swap_nr;
 793         if (invalid)
 794                 invalidate();
 795         write_swap_page (swap_nr, (char *) pte_page(page));
 796         free_page(pte_page(page));
 797         swap_successes++;
 798         shm_swp++;
 799         shm_rss--;
 800         return 1;
 801 }

/* */