kernel/vserver/dlimit.c

   1 /*
   2  *  linux/kernel/vserver/dlimit.c
   3  *
   4  *  Virtual Server: Context Disk Limits
   5  *
   6  *  Copyright (C) 2004-2005  Herbert Pötzl
   7  *
   8  *  V0.01  initial version
   9  *  V0.02  compat32 splitup
  10  *
  11  */
  12
  13 #include <linux/fs.h>
  14 #include <linux/namespace.h>
  15 #include <linux/namei.h>
  16 #include <linux/statfs.h>
  17 #include <linux/compat.h>
  18 #include <linux/vserver/switch.h>
  19 #include <linux/vs_context.h>
  20 #include <linux/vs_dlimit.h>
  21 #include <linux/vserver/dlimit_cmd.h>
  22
  23 #include <asm/errno.h>
  24 #include <asm/uaccess.h>
  25
  26 /*      __alloc_dl_info()
  27
  28         * allocate an initialized dl_info struct
  29         * doesn't make it visible (hash)                        */
  30
  31 static struct dl_info *__alloc_dl_info(struct super_block *sb, xid_t xid)
  32 {
  33         struct dl_info *new = NULL;
  34
  35         vxdprintk(VXD_CBIT(dlim, 5),
  36                 "alloc_dl_info(%p,%d)*", sb, xid);
  37
  38         /* would this benefit from a slab cache? */
  39         new = kmalloc(sizeof(struct dl_info), GFP_KERNEL);
  40         if (!new)
  41                 return 0;
  42
  43         memset (new, 0, sizeof(struct dl_info));
  44         new->dl_xid = xid;
  45         new->dl_sb = sb;
  46         INIT_RCU_HEAD(&new->dl_rcu);
  47         INIT_HLIST_NODE(&new->dl_hlist);
  48         spin_lock_init(&new->dl_lock);
  49         atomic_set(&new->dl_refcnt, 0);
  50         atomic_set(&new->dl_usecnt, 0);
  51
  52         /* rest of init goes here */
  53
  54         vxdprintk(VXD_CBIT(dlim, 4),
  55                 "alloc_dl_info(%p,%d) = %p", sb, xid, new);
  56         return new;
  57 }
  58
  59 /*      __dealloc_dl_info()
  60
  61         * final disposal of dl_info                             */
  62
  63 static void __dealloc_dl_info(struct dl_info *dli)
  64 {
  65         vxdprintk(VXD_CBIT(dlim, 4),
  66                 "dealloc_dl_info(%p)", dli);
  67
  68         dli->dl_hlist.next = LIST_POISON1;
  69         dli->dl_xid = -1;
  70         dli->dl_sb = 0;
  71
  72         BUG_ON(atomic_read(&dli->dl_usecnt));
  73         BUG_ON(atomic_read(&dli->dl_refcnt));
  74
  75         kfree(dli);
  76 }
  77
  78
  79 /*      hash table for dl_info hash */
  80
  81 #define DL_HASH_SIZE    13
  82
  83 struct hlist_head dl_info_hash[DL_HASH_SIZE];
  84
  85 static spinlock_t dl_info_hash_lock = SPIN_LOCK_UNLOCKED;
  86
  87
  88 static inline unsigned int __hashval(struct super_block *sb, xid_t xid)
  89 {
  90         return ((xid ^ (unsigned long)sb) % DL_HASH_SIZE);
  91 }
  92
  93
  94
  95 /*      __hash_dl_info()
  96
  97         * add the dli to the global hash table
  98         * requires the hash_lock to be held                     */
  99
 100 static inline void __hash_dl_info(struct dl_info *dli)
 101 {
 102         struct hlist_head *head;
 103
 104         vxdprintk(VXD_CBIT(dlim, 6),
 105                 "__hash_dl_info: %p[#%d]", dli, dli->dl_xid);
 106         get_dl_info(dli);
 107         head = &dl_info_hash[__hashval(dli->dl_sb, dli->dl_xid)];
 108         hlist_add_head_rcu(&dli->dl_hlist, head);
 109 }
 110
 111 /*      __unhash_dl_info()
 112
 113         * remove the dli from the global hash table
 114         * requires the hash_lock to be held                     */
 115
 116 static inline void __unhash_dl_info(struct dl_info *dli)
 117 {
 118         vxdprintk(VXD_CBIT(dlim, 6),
 119                 "__unhash_dl_info: %p[#%d]", dli, dli->dl_xid);
 120         hlist_del_rcu(&dli->dl_hlist);
 121         put_dl_info(dli);
 122 }
 123
 124
 125 /*      __lookup_dl_info()
 126
 127         * requires the rcu_read_lock()
 128         * doesn't increment the dl_refcnt                       */
 129
 130 static inline struct dl_info *__lookup_dl_info(struct super_block *sb, xid_t xid)
 131 {
 132         struct hlist_head *head = &dl_info_hash[__hashval(sb, xid)];
 133         struct hlist_node *pos;
 134         struct dl_info *dli;
 135
 136         hlist_for_each_entry_rcu(dli, pos, head, dl_hlist) {
 137
 138                 if (dli->dl_xid == xid && dli->dl_sb == sb) {
 139                         return dli;
 140                 }
 141         }
 142         return NULL;
 143 }
 144
 145
 146 struct dl_info *locate_dl_info(struct super_block *sb, xid_t xid)
 147 {
 148         struct dl_info *dli;
 149
 150         rcu_read_lock();
 151         dli = get_dl_info(__lookup_dl_info(sb, xid));
 152         vxdprintk(VXD_CBIT(dlim, 7),
 153                 "locate_dl_info(%p,#%d) = %p", sb, xid, dli);
 154         rcu_read_unlock();
 155         return dli;
 156 }
 157
 158 void rcu_free_dl_info(struct rcu_head *head)
 159 {
 160         struct dl_info *dli = container_of(head, struct dl_info, dl_rcu);
 161         int usecnt, refcnt;
 162
 163         BUG_ON(!dli || !head);
 164
 165         usecnt = atomic_read(&dli->dl_usecnt);
 166         BUG_ON(usecnt < 0);
 167
 168         refcnt = atomic_read(&dli->dl_refcnt);
 169         BUG_ON(refcnt < 0);
 170
 171         vxdprintk(VXD_CBIT(dlim, 3),
 172                 "rcu_free_dl_info(%p)", dli);
 173         if (!usecnt)
 174                 __dealloc_dl_info(dli);
 175         else
 176                 printk("!!! rcu didn't free\n");
 177 }
 178
 179
 180
 181
 182 static int do_addrem_dlimit(uint32_t id, const char __user *name,
 183         uint32_t flags, int add)
 184 {
 185         struct nameidata nd;
 186         int ret;
 187
 188         ret = user_path_walk_link(name, &nd);
 189         if (!ret) {
 190                 struct super_block *sb;
 191                 struct dl_info *dli;
 192
 193                 ret = -EINVAL;
 194                 if (!nd.dentry->d_inode)
 195                         goto out_release;
 196                 if (!(sb = nd.dentry->d_inode->i_sb))
 197                         goto out_release;
 198
 199                 if (add) {
 200                         dli = __alloc_dl_info(sb, id);
 201                         spin_lock(&dl_info_hash_lock);
 202
 203                         ret = -EEXIST;
 204                         if (__lookup_dl_info(sb, id))
 205                                 goto out_unlock;
 206                         __hash_dl_info(dli);
 207                         dli = NULL;
 208                 } else {
 209                         spin_lock(&dl_info_hash_lock);
 210                         dli = __lookup_dl_info(sb, id);
 211
 212                         ret = -ESRCH;
 213                         if (!dli)
 214                                 goto out_unlock;
 215                         __unhash_dl_info(dli);
 216                 }
 217                 ret = 0;
 218         out_unlock:
 219                 spin_unlock(&dl_info_hash_lock);
 220                 if (add && dli)
 221                         __dealloc_dl_info(dli);
 222         out_release:
 223                 path_release(&nd);
 224         }
 225         return ret;
 226 }
 227
 228 int vc_add_dlimit(uint32_t id, void __user *data)
 229 {
 230         struct vcmd_ctx_dlimit_base_v0 vc_data;
 231
 232         if (!vx_check(0, VX_ADMIN))
 233                 return -ENOSYS;
 234         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 235                 return -EFAULT;
 236
 237         return do_addrem_dlimit(id, vc_data.name, vc_data.flags, 1);
 238 }
 239
 240 int vc_rem_dlimit(uint32_t id, void __user *data)
 241 {
 242         struct vcmd_ctx_dlimit_base_v0 vc_data;
 243
 244         if (!vx_check(0, VX_ADMIN))
 245                 return -ENOSYS;
 246         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 247                 return -EFAULT;
 248
 249         return do_addrem_dlimit(id, vc_data.name, vc_data.flags, 0);
 250 }
 251
 252 #ifdef  CONFIG_COMPAT
 253
 254 int vc_add_dlimit_x32(uint32_t id, void __user *data)
 255 {
 256         struct vcmd_ctx_dlimit_base_v0_x32 vc_data;
 257
 258         if (!vx_check(0, VX_ADMIN))
 259                 return -ENOSYS;
 260         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 261                 return -EFAULT;
 262
 263         return do_addrem_dlimit(id,
 264                 compat_ptr(vc_data.name_ptr), vc_data.flags, 1);
 265 }
 266
 267 int vc_rem_dlimit_x32(uint32_t id, void __user *data)
 268 {
 269         struct vcmd_ctx_dlimit_base_v0_x32 vc_data;
 270
 271         if (!vx_check(0, VX_ADMIN))
 272                 return -ENOSYS;
 273         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 274                 return -EFAULT;
 275
 276         return do_addrem_dlimit(id,
 277                 compat_ptr(vc_data.name_ptr), vc_data.flags, 0);
 278 }
 279
 280 #endif  /* CONFIG_COMPAT */
 281
 282
 283 static inline
 284 int do_set_dlimit(uint32_t id, const char __user *name,
 285         uint32_t space_used, uint32_t space_total,
 286         uint32_t inodes_used, uint32_t inodes_total,
 287         uint32_t reserved, uint32_t flags)
 288 {
 289         struct nameidata nd;
 290         int ret;
 291
 292         ret = user_path_walk_link(name, &nd);
 293         if (!ret) {
 294                 struct super_block *sb;
 295                 struct dl_info *dli;
 296
 297                 ret = -EINVAL;
 298                 if (!nd.dentry->d_inode)
 299                         goto out_release;
 300                 if (!(sb = nd.dentry->d_inode->i_sb))
 301                         goto out_release;
 302                 if ((reserved != (uint32_t)CDLIM_KEEP &&
 303                         reserved > 100) ||
 304                         (inodes_used != (uint32_t)CDLIM_KEEP &&
 305                         inodes_used > inodes_total) ||
 306                         (space_used != (uint32_t)CDLIM_KEEP &&
 307                         space_used > space_total))
 308                         goto out_release;
 309
 310                 ret = -ESRCH;
 311                 dli = locate_dl_info(sb, id);
 312                 if (!dli)
 313                         goto out_release;
 314
 315                 spin_lock(&dli->dl_lock);
 316
 317                 if (inodes_used != (uint32_t)CDLIM_KEEP)
 318                         dli->dl_inodes_used = inodes_used;
 319                 if (inodes_total != (uint32_t)CDLIM_KEEP)
 320                         dli->dl_inodes_total = inodes_total;
 321                 if (space_used != (uint32_t)CDLIM_KEEP) {
 322                         dli->dl_space_used = space_used;
 323                         dli->dl_space_used <<= 10;
 324                 }
 325                 if (space_total == (uint32_t)CDLIM_INFINITY)
 326                         dli->dl_space_total = (uint64_t)CDLIM_INFINITY;
 327                 else if (space_total != (uint32_t)CDLIM_KEEP) {
 328                         dli->dl_space_total = space_total;
 329                         dli->dl_space_total <<= 10;
 330                 }
 331                 if (reserved != (uint32_t)CDLIM_KEEP)
 332                         dli->dl_nrlmult = (1 << 10) * (100 - reserved) / 100;
 333
 334                 spin_unlock(&dli->dl_lock);
 335
 336                 put_dl_info(dli);
 337                 ret = 0;
 338
 339         out_release:
 340                 path_release(&nd);
 341         }
 342         return ret;
 343 }
 344
 345 int vc_set_dlimit(uint32_t id, void __user *data)
 346 {
 347         struct vcmd_ctx_dlimit_v0 vc_data;
 348
 349         if (!vx_check(0, VX_ADMIN))
 350                 return -ENOSYS;
 351         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 352                 return -EFAULT;
 353
 354         return do_set_dlimit(id, vc_data.name,
 355                 vc_data.space_used, vc_data.space_total,
 356                 vc_data.inodes_used, vc_data.inodes_total,
 357                 vc_data.reserved, vc_data.flags);
 358 }
 359
 360 #ifdef  CONFIG_COMPAT
 361
 362 int vc_set_dlimit_x32(uint32_t id, void __user *data)
 363 {
 364         struct vcmd_ctx_dlimit_v0_x32 vc_data;
 365
 366         if (!vx_check(0, VX_ADMIN))
 367                 return -ENOSYS;
 368         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 369                 return -EFAULT;
 370
 371         return do_set_dlimit(id, compat_ptr(vc_data.name_ptr),
 372                 vc_data.space_used, vc_data.space_total,
 373                 vc_data.inodes_used, vc_data.inodes_total,
 374                 vc_data.reserved, vc_data.flags);
 375 }
 376
 377 #endif  /* CONFIG_COMPAT */
 378
 379
 380 static inline
 381 int do_get_dlimit(uint32_t id, const char __user *name,
 382         uint32_t *space_used, uint32_t *space_total,
 383         uint32_t *inodes_used, uint32_t *inodes_total,
 384         uint32_t *reserved, uint32_t *flags)
 385 {
 386         struct nameidata nd;
 387         int ret;
 388
 389         ret = user_path_walk_link(name, &nd);
 390         if (!ret) {
 391                 struct super_block *sb;
 392                 struct dl_info *dli;
 393
 394                 ret = -EINVAL;
 395                 if (!nd.dentry->d_inode)
 396                         goto out_release;
 397                 if (!(sb = nd.dentry->d_inode->i_sb))
 398                         goto out_release;
 399
 400                 ret = -ESRCH;
 401                 dli = locate_dl_info(sb, id);
 402                 if (!dli)
 403                         goto out_release;
 404
 405                 spin_lock(&dli->dl_lock);
 406                 *inodes_used = dli->dl_inodes_used;
 407                 *inodes_total = dli->dl_inodes_total;
 408                 *space_used = dli->dl_space_used >> 10;
 409                 if (dli->dl_space_total == (uint64_t)CDLIM_INFINITY)
 410                         *space_total = (uint32_t)CDLIM_INFINITY;
 411                 else
 412                         *space_total = dli->dl_space_total >> 10;
 413
 414                 *reserved = 100 - ((dli->dl_nrlmult * 100 + 512) >> 10);
 415                 spin_unlock(&dli->dl_lock);
 416
 417                 put_dl_info(dli);
 418                 ret = -EFAULT;
 419
 420                 ret = 0;
 421         out_release:
 422                 path_release(&nd);
 423         }
 424         return ret;
 425 }
 426
 427
 428 int vc_get_dlimit(uint32_t id, void __user *data)
 429 {
 430         struct vcmd_ctx_dlimit_v0 vc_data;
 431         int ret;
 432
 433         if (!vx_check(0, VX_ADMIN))
 434                 return -ENOSYS;
 435         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 436                 return -EFAULT;
 437
 438         ret = do_get_dlimit(id, vc_data.name,
 439                 &vc_data.space_used, &vc_data.space_total,
 440                 &vc_data.inodes_used, &vc_data.inodes_total,
 441                 &vc_data.reserved, &vc_data.flags);
 442         if (ret)
 443                 return ret;
 444
 445         if (copy_to_user(data, &vc_data, sizeof(vc_data)))
 446                 return -EFAULT;
 447         return 0;
 448 }
 449
 450 #ifdef  CONFIG_COMPAT
 451
 452 int vc_get_dlimit_x32(uint32_t id, void __user *data)
 453 {
 454         struct vcmd_ctx_dlimit_v0_x32 vc_data;
 455         int ret;
 456
 457         if (!vx_check(0, VX_ADMIN))
 458                 return -ENOSYS;
 459         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 460                 return -EFAULT;
 461
 462         ret = do_get_dlimit(id, compat_ptr(vc_data.name_ptr),
 463                 &vc_data.space_used, &vc_data.space_total,
 464                 &vc_data.inodes_used, &vc_data.inodes_total,
 465                 &vc_data.reserved, &vc_data.flags);
 466         if (ret)
 467                 return ret;
 468
 469         if (copy_to_user(data, &vc_data, sizeof(vc_data)))
 470                 return -EFAULT;
 471         return 0;
 472 }
 473
 474 #endif  /* CONFIG_COMPAT */
 475
 476
 477 void vx_vsi_statfs(struct super_block *sb, struct kstatfs *buf)
 478 {
 479         struct dl_info *dli;
 480         __u64 blimit, bfree, bavail;
 481         __u32 ifree;
 482
 483         dli = locate_dl_info(sb, vx_current_xid());
 484         if (!dli)
 485                 return;
 486
 487         spin_lock(&dli->dl_lock);
 488         if (dli->dl_inodes_total == (uint32_t)CDLIM_INFINITY)
 489                 goto no_ilim;
 490
 491         /* reduce max inodes available to limit */
 492         if (buf->f_files > dli->dl_inodes_total)
 493                 buf->f_files = dli->dl_inodes_total;
 494
 495         /* inode hack for reiserfs */
 496         if ((buf->f_files == 0) && (dli->dl_inodes_total > 0)) {
 497                 buf->f_files = dli->dl_inodes_total;
 498                 buf->f_ffree = dli->dl_inodes_total;
 499         }
 500
 501         ifree = dli->dl_inodes_total - dli->dl_inodes_used;
 502         /* reduce free inodes to min */
 503         if (ifree < buf->f_ffree)
 504                 buf->f_ffree = ifree;
 505
 506 no_ilim:
 507         if (dli->dl_space_total == (uint64_t)CDLIM_INFINITY)
 508                 goto no_blim;
 509
 510         blimit = dli->dl_space_total >> sb->s_blocksize_bits;
 511
 512         if (dli->dl_space_total < dli->dl_space_used)
 513                 bfree = 0;
 514         else
 515                 bfree = (dli->dl_space_total - dli->dl_space_used)
 516                         >> sb->s_blocksize_bits;
 517
 518         bavail = ((dli->dl_space_total >> 10) * dli->dl_nrlmult);
 519         if (bavail < dli->dl_space_used)
 520                 bavail = 0;
 521         else
 522                 bavail = (bavail - dli->dl_space_used)
 523                         >> sb->s_blocksize_bits;
 524
 525         /* reduce max space available to limit */
 526         if (buf->f_blocks > blimit)
 527                 buf->f_blocks = blimit;
 528
 529         /* reduce free space to min */
 530         if (bfree < buf->f_bfree)
 531                 buf->f_bfree = bfree;
 532
 533         /* reduce avail space to min */
 534         if (bavail < buf->f_bavail)
 535                 buf->f_bavail = bavail;
 536
 537 no_blim:
 538         spin_unlock(&dli->dl_lock);
 539         put_dl_info(dli);
 540
 541         return;
 542 }
 543
 544 #include <linux/module.h>
 545
 546 EXPORT_SYMBOL_GPL(locate_dl_info);
 547 EXPORT_SYMBOL_GPL(rcu_free_dl_info);
 548