ipc/shm.c

   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  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7  *
   8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9  */
  10
  11 #include <linux/malloc.h>
  12 #include <linux/shm.h>
  13 #include <linux/swap.h>
  14 #include <linux/smp_lock.h>
  15 #include <linux/init.h>
  16 #include <linux/vmalloc.h>
  17 #include <linux/proc_fs.h>
  18
  19 #include <asm/uaccess.h>
  20 #include <asm/pgtable.h>
  21
  22 extern int ipcperms (struct ipc_perm *ipcp, short shmflg);
  23 extern unsigned long get_swap_page (void);
  24 static int findkey (key_t key);
  25 static int newseg (key_t key, int shmflg, int size);
  26 static int shm_map (struct vm_area_struct *shmd);
  27 static void killseg (int id);
  28 static void shm_open (struct vm_area_struct *shmd);
  29 static void shm_close (struct vm_area_struct *shmd);
  30 static unsigned long shm_nopage(struct vm_area_struct *, unsigned long, int);
  31 static int shm_swapout(struct vm_area_struct *, struct page *);
  32 #ifdef CONFIG_PROC_FS
  33 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
  34 #endif
  35
  36 static int shm_tot = 0; /* total number of shared memory pages */
  37 static int shm_rss = 0; /* number of shared memory pages that are in memory */
  38 static int shm_swp = 0; /* number of shared memory pages that are in swap */
  39 static int max_shmid = 0; /* every used id is <= max_shmid */
  40 static DECLARE_WAIT_QUEUE_HEAD(shm_lock); /* calling findkey() may need to wait */
  41 static struct shmid_kernel *shm_segs[SHMMNI];
  42
  43 static unsigned short shm_seq = 0; /* incremented, for recognizing stale ids */
  44
  45 /* some statistics */
  46 static ulong swap_attempts = 0;
  47 static ulong swap_successes = 0;
  48 static ulong used_segs = 0;
  49
  50 void __init shm_init (void)
  51 {
  52         int id;
  53 #ifdef CONFIG_PROC_FS
  54         struct proc_dir_entry *ent;
  55 #endif
  56
  57         for (id = 0; id < SHMMNI; id++)
  58                 shm_segs[id] = (struct shmid_kernel *) IPC_UNUSED;
  59         shm_tot = shm_rss = shm_seq = max_shmid = used_segs = 0;
  60         init_waitqueue_head(&shm_lock);
  61 #ifdef CONFIG_PROC_FS
  62         ent = create_proc_entry("sysvipc/shm", 0, 0);
  63         ent->read_proc = sysvipc_shm_read_proc;
  64 #endif
  65         return;
  66 }
  67
  68 static int findkey (key_t key)
  69 {
  70         int id;
  71         struct shmid_kernel *shp;
  72
  73         for (id = 0; id <= max_shmid; id++) {
  74                 while ((shp = shm_segs[id]) == IPC_NOID)
  75                         sleep_on (&shm_lock);
  76                 if (shp == IPC_UNUSED)
  77                         continue;
  78                 if (key == shp->u.shm_perm.key)
  79                         return id;
  80         }
  81         return -1;
  82 }
  83
  84 /*
  85  * allocate new shmid_kernel and pgtable. protected by shm_segs[id] = NOID.
  86  */
  87 static int newseg (key_t key, int shmflg, int size)
  88 {
  89         struct shmid_kernel *shp;
  90         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
  91         int id, i;
  92
  93         if (size < SHMMIN)
  94                 return -EINVAL;
  95         if (shm_tot + numpages >= SHMALL)
  96                 return -ENOSPC;
  97         for (id = 0; id < SHMMNI; id++)
  98                 if (shm_segs[id] == IPC_UNUSED) {
  99                         shm_segs[id] = (struct shmid_kernel *) IPC_NOID;
 100                         goto found;
 101                 }
 102         return -ENOSPC;
 103
 104 found:
 105         shp = (struct shmid_kernel *) kmalloc (sizeof (*shp), GFP_KERNEL);
 106         if (!shp) {
 107                 shm_segs[id] = (struct shmid_kernel *) IPC_UNUSED;
 108                 wake_up (&shm_lock);
 109                 return -ENOMEM;
 110         }
 111
 112         shp->shm_pages = (ulong *) vmalloc (numpages*sizeof(ulong));
 113         if (!shp->shm_pages) {
 114                 shm_segs[id] = (struct shmid_kernel *) IPC_UNUSED;
 115                 wake_up (&shm_lock);
 116                 kfree(shp);
 117                 return -ENOMEM;
 118         }
 119
 120         for (i = 0; i < numpages; shp->shm_pages[i++] = 0);
 121         shm_tot += numpages;
 122         shp->u.shm_perm.key = key;
 123         shp->u.shm_perm.mode = (shmflg & S_IRWXUGO);
 124         shp->u.shm_perm.cuid = shp->u.shm_perm.uid = current->euid;
 125         shp->u.shm_perm.cgid = shp->u.shm_perm.gid = current->egid;
 126         shp->u.shm_perm.seq = shm_seq;
 127         shp->u.shm_segsz = size;
 128         shp->u.shm_cpid = current->pid;
 129         shp->attaches = NULL;
 130         shp->u.shm_lpid = shp->u.shm_nattch = 0;
 131         shp->u.shm_atime = shp->u.shm_dtime = 0;
 132         shp->u.shm_ctime = CURRENT_TIME;
 133         shp->shm_npages = numpages;
 134
 135         if (id > max_shmid)
 136                 max_shmid = id;
 137         shm_segs[id] = shp;
 138         used_segs++;
 139         wake_up (&shm_lock);
 140         return (unsigned int) shp->u.shm_perm.seq * SHMMNI + id;
 141 }
 142
 143 int shmmax = SHMMAX;
 144
 145 asmlinkage int sys_shmget (key_t key, int size, int shmflg)
 146 {
 147         struct shmid_kernel *shp;
 148         int err, id = 0;
 149
 150         down(&current->mm->mmap_sem);
 151         lock_kernel();
 152         if (size < 0 || size > shmmax) {
 153                 err = -EINVAL;
 154         } else if (key == IPC_PRIVATE) {
 155                 err = newseg(key, shmflg, size);
 156         } else if ((id = findkey (key)) == -1) {
 157                 if (!(shmflg & IPC_CREAT))
 158                         err = -ENOENT;
 159                 else
 160                         err = newseg(key, shmflg, size);
 161         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 162                 err = -EEXIST;
 163         } else {
 164                 shp = shm_segs[id];
 165                 if (shp->u.shm_perm.mode & SHM_DEST)
 166                         err = -EIDRM;
 167                 else if (size > shp->u.shm_segsz)
 168                         err = -EINVAL;
 169                 else if (ipcperms (&shp->u.shm_perm, shmflg))
 170                         err = -EACCES;
 171                 else
 172                         err = (int) shp->u.shm_perm.seq * SHMMNI + id;
 173         }
 174         unlock_kernel();
 175         up(&current->mm->mmap_sem);
 176         return err;
 177 }
 178
 179 /*
 180  * Only called after testing nattch and SHM_DEST.
 181  * Here pages, pgtable and shmid_kernel are freed.
 182  */
 183 static void killseg (int id)
 184 {
 185         struct shmid_kernel *shp;
 186         int i, numpages;
 187
 188         shp = shm_segs[id];
 189         if (shp == IPC_NOID || shp == IPC_UNUSED) {
 190                 printk ("shm nono: killseg called on unused seg id=%d\n", id);
 191                 return;
 192         }
 193         shp->u.shm_perm.seq++;     /* for shmat */
 194         shm_seq = (shm_seq+1) % ((unsigned)(1<<31)/SHMMNI); /* increment, but avoid overflow */
 195         shm_segs[id] = (struct shmid_kernel *) IPC_UNUSED;
 196         used_segs--;
 197         if (id == max_shmid)
 198                 while (max_shmid && (shm_segs[--max_shmid] == IPC_UNUSED));
 199         if (!shp->shm_pages) {
 200                 printk ("shm nono: killseg shp->pages=NULL. id=%d\n", id);
 201                 return;
 202         }
 203         numpages = shp->shm_npages;
 204         for (i = 0; i < numpages ; i++) {
 205                 pte_t pte;
 206                 pte = __pte(shp->shm_pages[i]);
 207                 if (pte_none(pte))
 208                         continue;
 209                 if (pte_present(pte)) {
 210                         free_page (pte_page(pte));
 211                         shm_rss--;
 212                 } else {
 213                         swap_free(pte_val(pte));
 214                         shm_swp--;
 215                 }
 216         }
 217         vfree(shp->shm_pages);
 218         shm_tot -= numpages;
 219         kfree(shp);
 220         return;
 221 }
 222
 223 asmlinkage int sys_shmctl (int shmid, int cmd, struct shmid_ds *buf)
 224 {
 225         struct shmid_ds tbuf;
 226         struct shmid_kernel *shp;
 227         struct ipc_perm *ipcp;
 228         int id, err = -EINVAL;
 229
 230         lock_kernel();
 231         if (cmd < 0 || shmid < 0)
 232                 goto out;
 233         if (cmd == IPC_SET) {
 234                 err = -EFAULT;
 235                 if(copy_from_user (&tbuf, buf, sizeof (*buf)))
 236                         goto out;
 237         }
 238
 239         switch (cmd) { /* replace with proc interface ? */
 240         case IPC_INFO:
 241         {
 242                 struct shminfo shminfo;
 243                 err = -EFAULT;
 244                 if (!buf)
 245                         goto out;
 246                 shminfo.shmmni = SHMMNI;
 247                 shminfo.shmmax = shmmax;
 248                 shminfo.shmmin = SHMMIN;
 249                 shminfo.shmall = SHMALL;
 250                 shminfo.shmseg = SHMSEG;
 251                 if(copy_to_user (buf, &shminfo, sizeof(struct shminfo)))
 252                         goto out;
 253                 err = max_shmid;
 254                 goto out;
 255         }
 256         case SHM_INFO:
 257         {
 258                 struct shm_info shm_info;
 259                 err = -EFAULT;
 260                 shm_info.used_ids = used_segs;
 261                 shm_info.shm_rss = shm_rss;
 262                 shm_info.shm_tot = shm_tot;
 263                 shm_info.shm_swp = shm_swp;
 264                 shm_info.swap_attempts = swap_attempts;
 265                 shm_info.swap_successes = swap_successes;
 266                 if(copy_to_user (buf, &shm_info, sizeof(shm_info)))
 267                         goto out;
 268                 err = max_shmid;
 269                 goto out;
 270         }
 271         case SHM_STAT:
 272                 err = -EINVAL;
 273                 if (shmid > max_shmid)
 274                         goto out;
 275                 shp = shm_segs[shmid];
 276                 if (shp == IPC_UNUSED || shp == IPC_NOID)
 277                         goto out;
 278                 if (ipcperms (&shp->u.shm_perm, S_IRUGO))
 279                         goto out;
 280                 id = (unsigned int) shp->u.shm_perm.seq * SHMMNI + shmid;
 281                 err = -EFAULT;
 282                 if(copy_to_user (buf, &shp->u, sizeof(*buf)))
 283                         goto out;
 284                 err = id;
 285                 goto out;
 286         }
 287
 288         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 289         err = -EINVAL;
 290         if (shp == IPC_UNUSED || shp == IPC_NOID)
 291                 goto out;
 292         err = -EIDRM;
 293         if (shp->u.shm_perm.seq != (unsigned int) shmid / SHMMNI)
 294                 goto out;
 295         ipcp = &shp->u.shm_perm;
 296
 297         switch (cmd) {
 298         case SHM_UNLOCK:
 299                 err = -EPERM;
 300                 if (!capable(CAP_IPC_LOCK))
 301                         goto out;
 302                 err = -EINVAL;
 303                 if (!(ipcp->mode & SHM_LOCKED))
 304                         goto out;
 305                 ipcp->mode &= ~SHM_LOCKED;
 306                 break;
 307         case SHM_LOCK:
 308 /* Allow superuser to lock segment in memory */
 309 /* Should the pages be faulted in here or leave it to user? */
 310 /* need to determine interaction with current->swappable */
 311                 err = -EPERM;
 312                 if (!capable(CAP_IPC_LOCK))
 313                         goto out;
 314                 err = -EINVAL;
 315                 if (ipcp->mode & SHM_LOCKED)
 316                         goto out;
 317                 ipcp->mode |= SHM_LOCKED;
 318                 break;
 319         case IPC_STAT:
 320                 err = -EACCES;
 321                 if (ipcperms (ipcp, S_IRUGO))
 322                         goto out;
 323                 err = -EFAULT;
 324                 if(copy_to_user (buf, &shp->u, sizeof(shp->u)))
 325                         goto out;
 326                 break;
 327         case IPC_SET:
 328                 if (current->euid == shp->u.shm_perm.uid ||
 329                     current->euid == shp->u.shm_perm.cuid ||
 330                     capable(CAP_SYS_ADMIN)) {
 331                         ipcp->uid = tbuf.shm_perm.uid;
 332                         ipcp->gid = tbuf.shm_perm.gid;
 333                         ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
 334                                 | (tbuf.shm_perm.mode & S_IRWXUGO);
 335                         shp->u.shm_ctime = CURRENT_TIME;
 336                         break;
 337                 }
 338                 err = -EPERM;
 339                 goto out;
 340         case IPC_RMID:
 341                 if (current->euid == shp->u.shm_perm.uid ||
 342                     current->euid == shp->u.shm_perm.cuid ||
 343                     capable(CAP_SYS_ADMIN)) {
 344                         shp->u.shm_perm.mode |= SHM_DEST;
 345                         if (shp->u.shm_nattch <= 0)
 346                                 killseg (id);
 347                         break;
 348                 }
 349                 err = -EPERM;
 350                 goto out;
 351         default:
 352                 err = -EINVAL;
 353                 goto out;
 354         }
 355         err = 0;
 356 out:
 357         unlock_kernel();
 358         return err;
 359 }
 360
 361 /*
 362  * The per process internal structure for managing segments is
 363  * `struct vm_area_struct'.
 364  * A shmat will add to and shmdt will remove from the list.
 365  * shmd->vm_mm          the attacher
 366  * shmd->vm_start       virt addr of attach, multiple of SHMLBA
 367  * shmd->vm_end         multiple of SHMLBA
 368  * shmd->vm_next        next attach for task
 369  * shmd->vm_next_share  next attach for segment
 370  * shmd->vm_offset      offset into segment
 371  * shmd->vm_pte         signature for this attach
 372  */
 373
 374 static struct vm_operations_struct shm_vm_ops = {
 375         shm_open,               /* open - callback for a new vm-area open */
 376         shm_close,              /* close - callback for when the vm-area is released */
 377         NULL,                   /* no need to sync pages at unmap */
 378         NULL,                   /* protect */
 379         NULL,                   /* sync */
 380         NULL,                   /* advise */
 381         shm_nopage,             /* nopage */
 382         NULL,                   /* wppage */
 383         shm_swapout,            /* swapout */
 384         NULL                    /* swapin */
 385 };
 386
 387 /* Insert shmd into the list shp->attaches */
 388 static inline void insert_attach (struct shmid_kernel * shp, struct vm_area_struct * shmd)
 389 {
 390         if((shmd->vm_next_share = shp->attaches) != NULL)
 391                 shp->attaches->vm_pprev_share = &shmd->vm_next_share;
 392         shp->attaches = shmd;
 393         shmd->vm_pprev_share = &shp->attaches;
 394 }
 395
 396 /* Remove shmd from list shp->attaches */
 397 static inline void remove_attach (struct shmid_kernel * shp, struct vm_area_struct * shmd)
 398 {
 399         if(shmd->vm_next_share)
 400                 shmd->vm_next_share->vm_pprev_share = shmd->vm_pprev_share;
 401         *shmd->vm_pprev_share = shmd->vm_next_share;
 402 }
 403
 404 /*
 405  * ensure page tables exist
 406  * mark page table entries with shm_sgn.
 407  */
 408 static int shm_map (struct vm_area_struct *shmd)
 409 {
 410         unsigned long tmp;
 411
 412         /* clear old mappings */
 413         do_munmap(shmd->vm_start, shmd->vm_end - shmd->vm_start);
 414
 415         /* add new mapping */
 416         tmp = shmd->vm_end - shmd->vm_start;
 417         if((current->mm->total_vm << PAGE_SHIFT) + tmp
 418            > (unsigned long) current->rlim[RLIMIT_AS].rlim_cur)
 419                 return -ENOMEM;
 420         current->mm->total_vm += tmp >> PAGE_SHIFT;
 421         insert_vm_struct(current->mm, shmd);
 422         merge_segments(current->mm, shmd->vm_start, shmd->vm_end);
 423
 424         return 0;
 425 }
 426
 427 /*
 428  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 429  */
 430 asmlinkage int sys_shmat (int shmid, char *shmaddr, int shmflg, ulong *raddr)
 431 {
 432         struct shmid_kernel *shp;
 433         struct vm_area_struct *shmd;
 434         int err = -EINVAL;
 435         unsigned int id;
 436         unsigned long addr;
 437         unsigned long len;
 438
 439         down(&current->mm->mmap_sem);
 440         lock_kernel();
 441         if (shmid < 0) {
 442                 /* printk("shmat() -> EINVAL because shmid = %d < 0\n",shmid); */
 443                 goto out;
 444         }
 445
 446         shp = shm_segs[id = (unsigned int) shmid % SHMMNI];
 447         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 448                 /* printk("shmat() -> EINVAL because shmid = %d is invalid\n",shmid); */
 449                 goto out;
 450         }
 451
 452         if (!(addr = (ulong) shmaddr)) {
 453                 if (shmflg & SHM_REMAP)
 454                         goto out;
 455                 err = -ENOMEM;
 456                 addr = 0;
 457         again:
 458                 if (!(addr = get_unmapped_area(addr, shp->u.shm_segsz)))
 459                         goto out;
 460                 if(addr & (SHMLBA - 1)) {
 461                         addr = (addr + (SHMLBA - 1)) & ~(SHMLBA - 1);
 462                         goto again;
 463                 }
 464         } else if (addr & (SHMLBA-1)) {
 465                 if (shmflg & SHM_RND)
 466                         addr &= ~(SHMLBA-1);       /* round down */
 467                 else
 468                         goto out;
 469         }
 470         /*
 471          * Check if addr exceeds TASK_SIZE (from do_mmap)
 472          */
 473         len = PAGE_SIZE*shp->shm_npages;
 474         err = -EINVAL;
 475         if (addr >= TASK_SIZE || len > TASK_SIZE  || addr > TASK_SIZE - len)
 476                 goto out;
 477         /*
 478          * If shm segment goes below stack, make sure there is some
 479          * space left for the stack to grow (presently 4 pages).
 480          */
 481         if (addr < current->mm->start_stack &&
 482             addr > current->mm->start_stack - PAGE_SIZE*(shp->shm_npages + 4))
 483         {
 484                 /* printk("shmat() -> EINVAL because segment intersects stack\n"); */
 485                 goto out;
 486         }
 487         if (!(shmflg & SHM_REMAP))
 488                 if ((shmd = find_vma_intersection(current->mm, addr, addr + shp->u.shm_segsz))) {
 489                         /* printk("shmat() -> EINVAL because the interval [0x%lx,0x%lx) intersects an already mapped interval [0x%lx,0x%lx).\n",
 490                                 addr, addr + shp->shm_segsz, shmd->vm_start, shmd->vm_end); */
 491                         goto out;
 492                 }
 493
 494         err = -EACCES;
 495         if (ipcperms(&shp->u.shm_perm, shmflg & SHM_RDONLY ? S_IRUGO : S_IRUGO|S_IWUGO))
 496                 goto out;
 497         err = -EIDRM;
 498         if (shp->u.shm_perm.seq != (unsigned int) shmid / SHMMNI)
 499                 goto out;
 500
 501         err = -ENOMEM;
 502         shmd = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 503         if (!shmd)
 504                 goto out;
 505         if ((shp != shm_segs[id]) || (shp->u.shm_perm.seq != (unsigned int) shmid / SHMMNI)) {
 506                 kmem_cache_free(vm_area_cachep, shmd);
 507                 err = -EIDRM;
 508                 goto out;
 509         }
 510
 511         shmd->vm_pte = SWP_ENTRY(SHM_SWP_TYPE, id);
 512         shmd->vm_start = addr;
 513         shmd->vm_end = addr + shp->shm_npages * PAGE_SIZE;
 514         shmd->vm_mm = current->mm;
 515         shmd->vm_page_prot = (shmflg & SHM_RDONLY) ? PAGE_READONLY : PAGE_SHARED;
 516         shmd->vm_flags = VM_SHM | VM_MAYSHARE | VM_SHARED
 517                          | VM_MAYREAD | VM_MAYEXEC | VM_READ | VM_EXEC
 518                          | ((shmflg & SHM_RDONLY) ? 0 : VM_MAYWRITE | VM_WRITE);
 519         shmd->vm_file = NULL;
 520         shmd->vm_offset = 0;
 521         shmd->vm_ops = &shm_vm_ops;
 522
 523         shp->u.shm_nattch++;            /* prevent destruction */
 524         if ((err = shm_map (shmd))) {
 525                 if (--shp->u.shm_nattch <= 0 && shp->u.shm_perm.mode & SHM_DEST)
 526                         killseg(id);
 527                 kmem_cache_free(vm_area_cachep, shmd);
 528                 goto out;
 529         }
 530
 531         insert_attach(shp,shmd);  /* insert shmd into shp->attaches */
 532
 533         shp->u.shm_lpid = current->pid;
 534         shp->u.shm_atime = CURRENT_TIME;
 535
 536         *raddr = addr;
 537         err = 0;
 538 out:
 539         unlock_kernel();
 540         up(&current->mm->mmap_sem);
 541         return err;
 542 }
 543
 544 /* This is called by fork, once for every shm attach. */
 545 static void shm_open (struct vm_area_struct *shmd)
 546 {
 547         unsigned int id;
 548         struct shmid_kernel *shp;
 549
 550         id = SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK;
 551         shp = shm_segs[id];
 552         if (shp == IPC_UNUSED) {
 553                 printk("shm_open: unused id=%d PANIC\n", id);
 554                 return;
 555         }
 556         insert_attach(shp,shmd);  /* insert shmd into shp->attaches */
 557         shp->u.shm_nattch++;
 558         shp->u.shm_atime = CURRENT_TIME;
 559         shp->u.shm_lpid = current->pid;
 560 }
 561
 562 /*
 563  * remove the attach descriptor shmd.
 564  * free memory for segment if it is marked destroyed.
 565  * The descriptor has already been removed from the current->mm->mmap list
 566  * and will later be kfree()d.
 567  */
 568 static void shm_close (struct vm_area_struct *shmd)
 569 {
 570         struct shmid_kernel *shp;
 571         int id;
 572
 573         /* remove from the list of attaches of the shm segment */
 574         id = SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK;
 575         shp = shm_segs[id];
 576         remove_attach(shp,shmd);  /* remove from shp->attaches */
 577         shp->u.shm_lpid = current->pid;
 578         shp->u.shm_dtime = CURRENT_TIME;
 579         if (--shp->u.shm_nattch <= 0 && shp->u.shm_perm.mode & SHM_DEST)
 580                 killseg (id);
 581 }
 582
 583 /*
 584  * detach and kill segment if marked destroyed.
 585  * The work is done in shm_close.
 586  */
 587 asmlinkage int sys_shmdt (char *shmaddr)
 588 {
 589         struct vm_area_struct *shmd, *shmdnext;
 590
 591         down(&current->mm->mmap_sem);
 592         lock_kernel();
 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         unlock_kernel();
 600         up(&current->mm->mmap_sem);
 601         return 0;
 602 }
 603
 604 /*
 605  * Enter the shm page into the SHM data structures.
 606  *
 607  * The way "nopage" is done, we don't actually have to
 608  * do anything here: nopage will have filled in the shm
 609  * data structures already, and shm_swap_out() will just
 610  * work off them..
 611  */
 612 static int shm_swapout(struct vm_area_struct * vma, struct page * page)
 613 {
 614         return 0;
 615 }
 616
 617 /*
 618  * page not present ... go through shm_pages
 619  */
 620 static unsigned long shm_nopage(struct vm_area_struct * shmd, unsigned long address, int no_share)
 621 {
 622         pte_t pte;
 623         struct shmid_kernel *shp;
 624         unsigned int id, idx;
 625
 626         id = SWP_OFFSET(shmd->vm_pte) & SHM_ID_MASK;
 627         idx = (address - shmd->vm_start + shmd->vm_offset) >> PAGE_SHIFT;
 628
 629 #ifdef DEBUG_SHM
 630         if (id > max_shmid) {
 631                 printk ("shm_nopage: id=%d too big. proc mem corrupted\n", id);
 632                 return 0;
 633         }
 634 #endif
 635         shp = shm_segs[id];
 636
 637 #ifdef DEBUG_SHM
 638         if (shp == IPC_UNUSED || shp == IPC_NOID) {
 639                 printk ("shm_nopage: id=%d invalid. Race.\n", id);
 640                 return 0;
 641         }
 642         if (idx >= shp->shm_npages) {
 643                 printk ("shm_nopage : too large page index. id=%d\n", id);
 644                 return 0;
 645         }
 646 #endif
 647
 648         pte = __pte(shp->shm_pages[idx]);
 649         if (!pte_present(pte)) {
 650                 unsigned long page = get_free_page(GFP_USER);
 651                 if (!page) {
 652                         oom(current);
 653                         return 0;
 654                 }
 655                 pte = __pte(shp->shm_pages[idx]);
 656                 if (pte_present(pte)) {
 657                         free_page (page); /* doesn't sleep */
 658                         goto done;
 659                 }
 660                 if (!pte_none(pte)) {
 661                         rw_swap_page_nocache(READ, pte_val(pte), (char *)page);
 662                         pte = __pte(shp->shm_pages[idx]);
 663                         if (pte_present(pte))  {
 664                                 free_page (page); /* doesn't sleep */
 665                                 goto done;
 666                         }
 667                         swap_free(pte_val(pte));
 668                         shm_swp--;
 669                 }
 670                 shm_rss++;
 671                 pte = pte_mkdirty(mk_pte(page, PAGE_SHARED));
 672                 shp->shm_pages[idx] = pte_val(pte);
 673         } else
 674                 --current->maj_flt;  /* was incremented in do_no_page */
 675
 676 done:   /* pte_val(pte) == shp->shm_pages[idx] */
 677         current->min_flt++;
 678         atomic_inc(&mem_map[MAP_NR(pte_page(pte))].count);
 679         return pte_page(pte);
 680 }
 681
 682 /*
 683  * Goes through counter = (shm_rss >> prio) present shm pages.
 684  */
 685 static unsigned long swap_id = 0; /* currently being swapped */
 686 static unsigned long swap_idx = 0; /* next to swap */
 687
 688 int shm_swap (int prio, int gfp_mask)
 689 {
 690         pte_t page;
 691         struct shmid_kernel *shp;
 692         unsigned long swap_nr;
 693         unsigned long id, idx;
 694         int loop = 0;
 695         int counter;
 696
 697         counter = shm_rss >> prio;
 698         if (!counter || !(swap_nr = get_swap_page()))
 699                 return 0;
 700
 701  check_id:
 702         shp = shm_segs[swap_id];
 703         if (shp == IPC_UNUSED || shp == IPC_NOID || shp->u.shm_perm.mode & SHM_LOCKED ) {
 704                 next_id:
 705                 swap_idx = 0;
 706                 if (++swap_id > max_shmid) {
 707                         swap_id = 0;
 708                         if (loop)
 709                                 goto failed;
 710                         loop = 1;
 711                 }
 712                 goto check_id;
 713         }
 714         id = swap_id;
 715
 716  check_table:
 717         idx = swap_idx++;
 718         if (idx >= shp->shm_npages)
 719                 goto next_id;
 720
 721         page = __pte(shp->shm_pages[idx]);
 722         if (!pte_present(page))
 723                 goto check_table;
 724         if ((gfp_mask & __GFP_DMA) && !PageDMA(&mem_map[MAP_NR(pte_page(page))]))
 725                 goto check_table;
 726         swap_attempts++;
 727
 728         if (--counter < 0) { /* failed */
 729                 failed:
 730                 swap_free (swap_nr);
 731                 return 0;
 732         }
 733         if (atomic_read(&mem_map[MAP_NR(pte_page(page))].count) != 1)
 734                 goto check_table;
 735         shp->shm_pages[idx] = swap_nr;
 736         rw_swap_page_nocache (WRITE, swap_nr, (char *) pte_page(page));
 737         free_page(pte_page(page));
 738         swap_successes++;
 739         shm_swp++;
 740         shm_rss--;
 741         return 1;
 742 }
 743
 744 /*
 745  * Free the swap entry and set the new pte for the shm page.
 746  */
 747 static void shm_unuse_page(struct shmid_kernel *shp, unsigned long idx,
 748                            unsigned long page, unsigned long entry)
 749 {
 750         pte_t pte;
 751
 752         pte = pte_mkdirty(mk_pte(page, PAGE_SHARED));
 753         shp->shm_pages[idx] = pte_val(pte);
 754         atomic_inc(&mem_map[MAP_NR(page)].count);
 755         shm_rss++;
 756
 757         swap_free(entry);
 758         shm_swp--;
 759 }
 760
 761 /*
 762  * unuse_shm() search for an eventually swapped out shm page.
 763  */
 764 void shm_unuse(unsigned long entry, unsigned long page)
 765 {
 766         int i, n;
 767
 768         for (i = 0; i < SHMMNI; i++)
 769                 if (shm_segs[i] != IPC_UNUSED && shm_segs[i] != IPC_NOID)
 770                         for (n = 0; n < shm_segs[i]->shm_npages; n++)
 771                                 if (shm_segs[i]->shm_pages[n] == entry)
 772                                 {
 773                                         shm_unuse_page(shm_segs[i], n,
 774                                                        page, entry);
 775                                         return;
 776                                 }
 777 }
 778
 779 #ifdef CONFIG_PROC_FS
 780 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
 781 {
 782         off_t pos = 0;
 783         off_t begin = 0;
 784         int i, len = 0;
 785
 786         len += sprintf(buffer, "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n");
 787
 788         for(i = 0; i < SHMMNI; i++)
 789                 if(shm_segs[i] != IPC_UNUSED) {
 790                         len += sprintf(buffer + len, "%10d %10d  %4o %10d %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n",
 791                         shm_segs[i]->u.shm_perm.key,
 792                         shm_segs[i]->u.shm_perm.seq * SHMMNI + i,
 793                         shm_segs[i]->u.shm_perm.mode,
 794                         shm_segs[i]->u.shm_segsz,
 795                         shm_segs[i]->u.shm_cpid,
 796                         shm_segs[i]->u.shm_lpid,
 797                         shm_segs[i]->u.shm_nattch,
 798                         shm_segs[i]->u.shm_perm.uid,
 799                         shm_segs[i]->u.shm_perm.gid,
 800                         shm_segs[i]->u.shm_perm.cuid,
 801                         shm_segs[i]->u.shm_perm.cgid,
 802                         shm_segs[i]->u.shm_atime,
 803                         shm_segs[i]->u.shm_dtime,
 804                         shm_segs[i]->u.shm_ctime);
 805
 806                         pos += len;
 807                         if(pos < offset) {
 808                                 len = 0;
 809                                 begin = pos;
 810                         }
 811                         if(pos > offset + length)
 812                                 goto done;
 813                 }
 814         *eof = 1;
 815 done:
 816         *start = buffer + (offset - begin);
 817         len -= (offset - begin);
 818         if(len > length)
 819                 len = length;
 820         if(len < 0)
 821                 len = 0;
 822         return len;
 823 }
 824 #endif