kernel/vserver/dlimit.c

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