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/vs_base.h>
  19 #include <linux/vs_context.h>
  20 #include <linux/vs_dlimit.h>
  21 #include <linux/vserver/switch.h>
  22 #include <linux/vserver/dlimit_cmd.h>
  23
  24 #include <asm/errno.h>
  25 #include <asm/uaccess.h>
  26
  27 /*      __alloc_dl_info()
  28
  29         * allocate an initialized dl_info struct
  30         * doesn't make it visible (hash)                        */
  31
  32 static struct dl_info *__alloc_dl_info(struct super_block *sb, xid_t xid)
  33 {
  34         struct dl_info *new = NULL;
  35
  36         vxdprintk(VXD_CBIT(dlim, 5),
  37                 "alloc_dl_info(%p,%d)*", sb, xid);
  38
  39         /* would this benefit from a slab cache? */
  40         new = kmalloc(sizeof(struct dl_info), GFP_KERNEL);
  41         if (!new)
  42                 return 0;
  43
  44         memset (new, 0, sizeof(struct dl_info));
  45         new->dl_xid = xid;
  46         new->dl_sb = sb;
  47         INIT_RCU_HEAD(&new->dl_rcu);
  48         INIT_HLIST_NODE(&new->dl_hlist);
  49         spin_lock_init(&new->dl_lock);
  50         atomic_set(&new->dl_refcnt, 0);
  51         atomic_set(&new->dl_usecnt, 0);
  52
  53         /* rest of init goes here */
  54
  55         vxdprintk(VXD_CBIT(dlim, 4),
  56                 "alloc_dl_info(%p,%d) = %p", sb, xid, new);
  57         return new;
  58 }
  59
  60 /*      __dealloc_dl_info()
  61
  62         * final disposal of dl_info                             */
  63
  64 static void __dealloc_dl_info(struct dl_info *dli)
  65 {
  66         vxdprintk(VXD_CBIT(dlim, 4),
  67                 "dealloc_dl_info(%p)", dli);
  68
  69         dli->dl_hlist.next = LIST_POISON1;
  70         dli->dl_xid = -1;
  71         dli->dl_sb = 0;
  72
  73         BUG_ON(atomic_read(&dli->dl_usecnt));
  74         BUG_ON(atomic_read(&dli->dl_refcnt));
  75
  76         kfree(dli);
  77 }
  78
  79
  80 /*      hash table for dl_info hash */
  81
  82 #define DL_HASH_SIZE    13
  83
  84 struct hlist_head dl_info_hash[DL_HASH_SIZE];
  85
  86 static spinlock_t dl_info_hash_lock = SPIN_LOCK_UNLOCKED;
  87
  88
  89 static inline unsigned int __hashval(struct super_block *sb, xid_t xid)
  90 {
  91         return ((xid ^ (unsigned long)sb) % DL_HASH_SIZE);
  92 }
  93
  94
  95
  96 /*      __hash_dl_info()
  97
  98         * add the dli to the global hash table
  99         * requires the hash_lock to be held                     */
 100
 101 static inline void __hash_dl_info(struct dl_info *dli)
 102 {
 103         struct hlist_head *head;
 104
 105         vxdprintk(VXD_CBIT(dlim, 6),
 106                 "__hash_dl_info: %p[#%d]", dli, dli->dl_xid);
 107         get_dl_info(dli);
 108         head = &dl_info_hash[__hashval(dli->dl_sb, dli->dl_xid)];
 109         hlist_add_head_rcu(&dli->dl_hlist, head);
 110 }
 111
 112 /*      __unhash_dl_info()
 113
 114         * remove the dli from the global hash table
 115         * requires the hash_lock to be held                     */
 116
 117 static inline void __unhash_dl_info(struct dl_info *dli)
 118 {
 119         vxdprintk(VXD_CBIT(dlim, 6),
 120                 "__unhash_dl_info: %p[#%d]", dli, dli->dl_xid);
 121         hlist_del_rcu(&dli->dl_hlist);
 122         put_dl_info(dli);
 123 }
 124
 125
 126 /*      __lookup_dl_info()
 127
 128         * requires the rcu_read_lock()
 129         * doesn't increment the dl_refcnt                       */
 130
 131 static inline struct dl_info *__lookup_dl_info(struct super_block *sb, xid_t xid)
 132 {
 133         struct hlist_head *head = &dl_info_hash[__hashval(sb, xid)];
 134         struct hlist_node *pos;
 135         struct dl_info *dli;
 136
 137         hlist_for_each_entry_rcu(dli, pos, head, dl_hlist) {
 138
 139                 if (dli->dl_xid == xid && dli->dl_sb == sb) {
 140                         return dli;
 141                 }
 142         }
 143         return NULL;
 144 }
 145
 146
 147 struct dl_info *locate_dl_info(struct super_block *sb, xid_t xid)
 148 {
 149         struct dl_info *dli;
 150
 151         rcu_read_lock();
 152         dli = get_dl_info(__lookup_dl_info(sb, xid));
 153         vxdprintk(VXD_CBIT(dlim, 7),
 154                 "locate_dl_info(%p,#%d) = %p", sb, xid, dli);
 155         rcu_read_unlock();
 156         return dli;
 157 }
 158
 159 void rcu_free_dl_info(struct rcu_head *head)
 160 {
 161         struct dl_info *dli = container_of(head, struct dl_info, dl_rcu);
 162         int usecnt, refcnt;
 163
 164         BUG_ON(!dli || !head);
 165
 166         usecnt = atomic_read(&dli->dl_usecnt);
 167         BUG_ON(usecnt < 0);
 168
 169         refcnt = atomic_read(&dli->dl_refcnt);
 170         BUG_ON(refcnt < 0);
 171
 172         vxdprintk(VXD_CBIT(dlim, 3),
 173                 "rcu_free_dl_info(%p)", dli);
 174         if (!usecnt)
 175                 __dealloc_dl_info(dli);
 176         else
 177                 printk("!!! rcu didn't free\n");
 178 }
 179
 180
 181
 182
 183 static int do_addrem_dlimit(uint32_t id, const char __user *name,
 184         uint32_t flags, int add)
 185 {
 186         struct nameidata nd;
 187         int ret;
 188
 189         ret = user_path_walk_link(name, &nd);
 190         if (!ret) {
 191                 struct super_block *sb;
 192                 struct dl_info *dli;
 193
 194                 ret = -EINVAL;
 195                 if (!nd.dentry->d_inode)
 196                         goto out_release;
 197                 if (!(sb = nd.dentry->d_inode->i_sb))
 198                         goto out_release;
 199
 200                 if (add) {
 201                         dli = __alloc_dl_info(sb, id);
 202                         spin_lock(&dl_info_hash_lock);
 203
 204                         ret = -EEXIST;
 205                         if (__lookup_dl_info(sb, id))
 206                                 goto out_unlock;
 207                         __hash_dl_info(dli);
 208                         dli = NULL;
 209                 } else {
 210                         spin_lock(&dl_info_hash_lock);
 211                         dli = __lookup_dl_info(sb, id);
 212
 213                         ret = -ESRCH;
 214                         if (!dli)
 215                                 goto out_unlock;
 216                         __unhash_dl_info(dli);
 217                 }
 218                 ret = 0;
 219         out_unlock:
 220                 spin_unlock(&dl_info_hash_lock);
 221                 if (add && dli)
 222                         __dealloc_dl_info(dli);
 223         out_release:
 224                 path_release(&nd);
 225         }
 226         return ret;
 227 }
 228
 229 int vc_add_dlimit(uint32_t id, void __user *data)
 230 {
 231         struct vcmd_ctx_dlimit_base_v0 vc_data;
 232
 233         if (!vx_check(0, VX_ADMIN))
 234                 return -ENOSYS;
 235         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 236                 return -EFAULT;
 237
 238         return do_addrem_dlimit(id, vc_data.name, vc_data.flags, 1);
 239 }
 240
 241 int vc_rem_dlimit(uint32_t id, void __user *data)
 242 {
 243         struct vcmd_ctx_dlimit_base_v0 vc_data;
 244
 245         if (!vx_check(0, VX_ADMIN))
 246                 return -ENOSYS;
 247         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 248                 return -EFAULT;
 249
 250         return do_addrem_dlimit(id, vc_data.name, vc_data.flags, 0);
 251 }
 252
 253 #ifdef  CONFIG_COMPAT
 254
 255 int vc_add_dlimit_x32(uint32_t id, void __user *data)
 256 {
 257         struct vcmd_ctx_dlimit_base_v0_x32 vc_data;
 258
 259         if (!vx_check(0, VX_ADMIN))
 260                 return -ENOSYS;
 261         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 262                 return -EFAULT;
 263
 264         return do_addrem_dlimit(id,
 265                 compat_ptr(vc_data.name_ptr), vc_data.flags, 1);
 266 }
 267
 268 int vc_rem_dlimit_x32(uint32_t id, void __user *data)
 269 {
 270         struct vcmd_ctx_dlimit_base_v0_x32 vc_data;
 271
 272         if (!vx_check(0, VX_ADMIN))
 273                 return -ENOSYS;
 274         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 275                 return -EFAULT;
 276
 277         return do_addrem_dlimit(id,
 278                 compat_ptr(vc_data.name_ptr), vc_data.flags, 0);
 279 }
 280
 281 #endif  /* CONFIG_COMPAT */
 282
 283
 284 static inline
 285 int do_set_dlimit(uint32_t id, const char __user *name,
 286         uint32_t space_used, uint32_t space_total,
 287         uint32_t inodes_used, uint32_t inodes_total,
 288         uint32_t reserved, uint32_t flags)
 289 {
 290         struct nameidata nd;
 291         int ret;
 292
 293         ret = user_path_walk_link(name, &nd);
 294         if (!ret) {
 295                 struct super_block *sb;
 296                 struct dl_info *dli;
 297
 298                 ret = -EINVAL;
 299                 if (!nd.dentry->d_inode)
 300                         goto out_release;
 301                 if (!(sb = nd.dentry->d_inode->i_sb))
 302                         goto out_release;
 303                 if ((reserved != (uint32_t)CDLIM_KEEP &&
 304                         reserved > 100) ||
 305                         (inodes_used != (uint32_t)CDLIM_KEEP &&
 306                         inodes_used > inodes_total) ||
 307                         (space_used != (uint32_t)CDLIM_KEEP &&
 308                         space_used > space_total))
 309                         goto out_release;
 310
 311                 ret = -ESRCH;
 312                 dli = locate_dl_info(sb, id);
 313                 if (!dli)
 314                         goto out_release;
 315
 316                 spin_lock(&dli->dl_lock);
 317
 318                 if (inodes_used != (uint32_t)CDLIM_KEEP)
 319                         dli->dl_inodes_used = inodes_used;
 320                 if (inodes_total != (uint32_t)CDLIM_KEEP)
 321                         dli->dl_inodes_total = inodes_total;
 322                 if (space_used != (uint32_t)CDLIM_KEEP) {
 323                         dli->dl_space_used = space_used;
 324                         dli->dl_space_used <<= 10;
 325                 }
 326                 if (space_total == (uint32_t)CDLIM_INFINITY)
 327                         dli->dl_space_total = (uint64_t)CDLIM_INFINITY;
 328                 else if (space_total != (uint32_t)CDLIM_KEEP) {
 329                         dli->dl_space_total = space_total;
 330                         dli->dl_space_total <<= 10;
 331                 }
 332                 if (reserved != (uint32_t)CDLIM_KEEP)
 333                         dli->dl_nrlmult = (1 << 10) * (100 - reserved) / 100;
 334
 335                 spin_unlock(&dli->dl_lock);
 336
 337                 put_dl_info(dli);
 338                 ret = 0;
 339
 340         out_release:
 341                 path_release(&nd);
 342         }
 343         return ret;
 344 }
 345
 346 int vc_set_dlimit(uint32_t id, void __user *data)
 347 {
 348         struct vcmd_ctx_dlimit_v0 vc_data;
 349
 350         if (!vx_check(0, VX_ADMIN))
 351                 return -ENOSYS;
 352         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 353                 return -EFAULT;
 354
 355         return do_set_dlimit(id, vc_data.name,
 356                 vc_data.space_used, vc_data.space_total,
 357                 vc_data.inodes_used, vc_data.inodes_total,
 358                 vc_data.reserved, vc_data.flags);
 359 }
 360
 361 #ifdef  CONFIG_COMPAT
 362
 363 int vc_set_dlimit_x32(uint32_t id, void __user *data)
 364 {
 365         struct vcmd_ctx_dlimit_v0_x32 vc_data;
 366
 367         if (!vx_check(0, VX_ADMIN))
 368                 return -ENOSYS;
 369         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 370                 return -EFAULT;
 371
 372         return do_set_dlimit(id, compat_ptr(vc_data.name_ptr),
 373                 vc_data.space_used, vc_data.space_total,
 374                 vc_data.inodes_used, vc_data.inodes_total,
 375                 vc_data.reserved, vc_data.flags);
 376 }
 377
 378 #endif  /* CONFIG_COMPAT */
 379
 380
 381 static inline
 382 int do_get_dlimit(uint32_t id, const char __user *name,
 383         uint32_t *space_used, uint32_t *space_total,
 384         uint32_t *inodes_used, uint32_t *inodes_total,
 385         uint32_t *reserved, uint32_t *flags)
 386 {
 387         struct nameidata nd;
 388         int ret;
 389
 390         ret = user_path_walk_link(name, &nd);
 391         if (!ret) {
 392                 struct super_block *sb;
 393                 struct dl_info *dli;
 394
 395                 ret = -EINVAL;
 396                 if (!nd.dentry->d_inode)
 397                         goto out_release;
 398                 if (!(sb = nd.dentry->d_inode->i_sb))
 399                         goto out_release;
 400
 401                 ret = -ESRCH;
 402                 dli = locate_dl_info(sb, id);
 403                 if (!dli)
 404                         goto out_release;
 405
 406                 spin_lock(&dli->dl_lock);
 407                 *inodes_used = dli->dl_inodes_used;
 408                 *inodes_total = dli->dl_inodes_total;
 409                 *space_used = dli->dl_space_used >> 10;
 410                 if (dli->dl_space_total == (uint64_t)CDLIM_INFINITY)
 411                         *space_total = (uint32_t)CDLIM_INFINITY;
 412                 else
 413                         *space_total = dli->dl_space_total >> 10;
 414
 415                 *reserved = 100 - ((dli->dl_nrlmult * 100 + 512) >> 10);
 416                 spin_unlock(&dli->dl_lock);
 417
 418                 put_dl_info(dli);
 419                 ret = -EFAULT;
 420
 421                 ret = 0;
 422         out_release:
 423                 path_release(&nd);
 424         }
 425         return ret;
 426 }
 427
 428
 429 int vc_get_dlimit(uint32_t id, void __user *data)
 430 {
 431         struct vcmd_ctx_dlimit_v0 vc_data;
 432         int ret;
 433
 434         if (!vx_check(0, VX_ADMIN))
 435                 return -ENOSYS;
 436         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 437                 return -EFAULT;
 438
 439         ret = do_get_dlimit(id, vc_data.name,
 440                 &vc_data.space_used, &vc_data.space_total,
 441                 &vc_data.inodes_used, &vc_data.inodes_total,
 442                 &vc_data.reserved, &vc_data.flags);
 443         if (ret)
 444                 return ret;
 445
 446         if (copy_to_user(data, &vc_data, sizeof(vc_data)))
 447                 return -EFAULT;
 448         return 0;
 449 }
 450
 451 #ifdef  CONFIG_COMPAT
 452
 453 int vc_get_dlimit_x32(uint32_t id, void __user *data)
 454 {
 455         struct vcmd_ctx_dlimit_v0_x32 vc_data;
 456         int ret;
 457
 458         if (!vx_check(0, VX_ADMIN))
 459                 return -ENOSYS;
 460         if (copy_from_user (&vc_data, data, sizeof(vc_data)))
 461                 return -EFAULT;
 462
 463         ret = do_get_dlimit(id, compat_ptr(vc_data.name_ptr),
 464                 &vc_data.space_used, &vc_data.space_total,
 465                 &vc_data.inodes_used, &vc_data.inodes_total,
 466                 &vc_data.reserved, &vc_data.flags);
 467         if (ret)
 468                 return ret;
 469
 470         if (copy_to_user(data, &vc_data, sizeof(vc_data)))
 471                 return -EFAULT;
 472         return 0;
 473 }
 474
 475 #endif  /* CONFIG_COMPAT */
 476
 477
 478 void vx_vsi_statfs(struct super_block *sb, struct kstatfs *buf)
 479 {
 480         struct dl_info *dli;
 481         __u64 blimit, bfree, bavail;
 482         __u32 ifree;
 483
 484         dli = locate_dl_info(sb, vx_current_xid());
 485         if (!dli)
 486                 return;
 487
 488         spin_lock(&dli->dl_lock);
 489         if (dli->dl_inodes_total == (uint32_t)CDLIM_INFINITY)
 490                 goto no_ilim;
 491
 492         /* reduce max inodes available to limit */
 493         if (buf->f_files > dli->dl_inodes_total)
 494                 buf->f_files = dli->dl_inodes_total;
 495
 496         /* inode hack for reiserfs */
 497         if ((buf->f_files == 0) && (dli->dl_inodes_total > 0)) {
 498                 buf->f_files = dli->dl_inodes_total;
 499                 buf->f_ffree = dli->dl_inodes_total;
 500         }
 501
 502         ifree = dli->dl_inodes_total - dli->dl_inodes_used;
 503         /* reduce free inodes to min */
 504         if (ifree < buf->f_ffree)
 505                 buf->f_ffree = ifree;
 506
 507 no_ilim:
 508         if (dli->dl_space_total == (uint64_t)CDLIM_INFINITY)
 509                 goto no_blim;
 510
 511         blimit = dli->dl_space_total >> sb->s_blocksize_bits;
 512
 513         if (dli->dl_space_total < dli->dl_space_used)
 514                 bfree = 0;
 515         else
 516                 bfree = (dli->dl_space_total - dli->dl_space_used)
 517                         >> sb->s_blocksize_bits;
 518
 519         bavail = ((dli->dl_space_total >> 10) * dli->dl_nrlmult);
 520         if (bavail < dli->dl_space_used)
 521                 bavail = 0;
 522         else
 523                 bavail = (bavail - dli->dl_space_used)
 524                         >> sb->s_blocksize_bits;
 525
 526         /* reduce max space available to limit */
 527         if (buf->f_blocks > blimit)
 528                 buf->f_blocks = blimit;
 529
 530         /* reduce free space to min */
 531         if (bfree < buf->f_bfree)
 532                 buf->f_bfree = bfree;
 533
 534         /* reduce avail space to min */
 535         if (bavail < buf->f_bavail)
 536                 buf->f_bavail = bavail;
 537
 538 no_blim:
 539         spin_unlock(&dli->dl_lock);
 540         put_dl_info(dli);
 541
 542         return;
 543 }
 544
 545 #include <linux/module.h>
 546
 547 EXPORT_SYMBOL_GPL(locate_dl_info);
 548 EXPORT_SYMBOL_GPL(rcu_free_dl_info);
 549