fs/dquot.c

   1 /*
   2  * Implementation of the diskquota system for the LINUX operating system. QUOTA
   3  * is implemented using the BSD system call interface as the means of
   4  * communication with the user level. This file contains the generic routines
   5  * called by the different filesystems on allocation of an inode or block.
   6  * These routines take care of the administration needed to have a consistent
   7  * diskquota tracking system. The ideas of both user and group quotas are based
   8  * on the Melbourne quota system as used on BSD derived systems. The internal
   9  * implementation is based on one of the several variants of the LINUX
  10  * inode-subsystem with added complexity of the diskquota system.
  11  *
  12  * Version: $Id: dquot.c,v 6.3 1996/11/17 18:35:34 mvw Exp mvw $
  13  *
  14  * Author:      Marco van Wieringen <mvw@planets.elm.net>
  15  *
  16  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
  17  *
  18  *              Revised list management to avoid races
  19  *              -- Bill Hawes, <whawes@star.net>, 9/98
  20  *
  21  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
  22  *              As the consequence the locking was moved from dquot_decr_...(),
  23  *              dquot_incr_...() to calling functions.
  24  *              invalidate_dquots() now writes modified dquots.
  25  *              Serialized quota_off() and quota_on() for mount point.
  26  *              Fixed a few bugs in grow_dquots().
  27  *              Fixed deadlock in write_dquot() - we no longer account quotas on
  28  *              quota files
  29  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
  30  *              add_dquot_ref() restarts after blocking
  31  *              Added check for bogus uid and fixed check for group in quotactl.
  32  *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
  33  *
  34  *              Used struct list_head instead of own list struct
  35  *              Invalidation of referenced dquots is no longer possible
  36  *              Improved free_dquots list management
  37  *              Quota and i_blocks are now updated in one place to avoid races
  38  *              Warnings are now delayed so we won't block in critical section
  39  *              Write updated not to require dquot lock
  40  *              Jan Kara, <jack@suse.cz>, 9/2000
  41  *
  42  *              Added dynamic quota structure allocation
  43  *              Jan Kara <jack@suse.cz> 12/2000
  44  *
  45  *              Rewritten quota interface. Implemented new quota format and
  46  *              formats registering.
  47  *              Jan Kara, <jack@suse.cz>, 2001,2002
  48  *
  49  *              New SMP locking.
  50  *              Jan Kara, <jack@suse.cz>, 10/2002
  51  *
  52  *              Added journalled quota support
  53  *              Jan Kara, <jack@suse.cz>, 2003,2004
  54  *
  55  * (C) Copyright 1994 - 1997 Marco van Wieringen
  56  */
  57
  58 #include <linux/errno.h>
  59 #include <linux/kernel.h>
  60 #include <linux/fs.h>
  61 #include <linux/mount.h>
  62 #include <linux/mm.h>
  63 #include <linux/time.h>
  64 #include <linux/types.h>
  65 #include <linux/string.h>
  66 #include <linux/fcntl.h>
  67 #include <linux/stat.h>
  68 #include <linux/tty.h>
  69 #include <linux/file.h>
  70 #include <linux/slab.h>
  71 #include <linux/sysctl.h>
  72 #include <linux/smp_lock.h>
  73 #include <linux/init.h>
  74 #include <linux/module.h>
  75 #include <linux/proc_fs.h>
  76 #include <linux/security.h>
  77 #include <linux/kmod.h>
  78 #include <linux/pagemap.h>
  79
  80 #include <asm/uaccess.h>
  81
  82 #define __DQUOT_PARANOIA
  83
  84 /*
  85  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
  86  * and quota formats and also dqstats structure containing statistics about the
  87  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
  88  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
  89  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
  90  * in inode_add_bytes() and inode_sub_bytes().
  91  *
  92  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
  93  *
  94  * Note that some things (eg. sb pointer, type, id) doesn't change during
  95  * the life of the dquot structure and so needn't to be protected by a lock
  96  *
  97  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
  98  * operation is just reading pointers from inode (or not using them at all) the
  99  * read lock is enough. If pointers are altered function must hold write lock
 100  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
 101  * for altering the flag i_sem is also needed).  If operation is holding
 102  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
 103  * dqonoff_sem.
 104  * This locking assures that:
 105  *   a) update/access to dquot pointers in inode is serialized
 106  *   b) everyone is guarded against invalidate_dquots()
 107  *
 108  * Each dquot has its dq_lock semaphore. Locked dquots might not be referenced
 109  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
 110  * Currently dquot is locked only when it is being read to memory (or space for
 111  * it is being allocated) on the first dqget() and when it is being released on
 112  * the last dqput(). The allocation and release oparations are serialized by
 113  * the dq_lock and by checking the use count in dquot_release().  Write
 114  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
 115  * spinlock to internal buffers before writing.
 116  *
 117  * Lock ordering (including related VFS locks) is following:
 118  *   i_sem > dqonoff_sem > iprune_sem > journal_lock > dqptr_sem >
 119  *   > dquot->dq_lock > dqio_sem
 120  * i_sem on quota files is special (it's below dqio_sem)
 121  */
 122
 123 static spinlock_t dq_list_lock = SPIN_LOCK_UNLOCKED;
 124 spinlock_t dq_data_lock = SPIN_LOCK_UNLOCKED;
 125
 126 static char *quotatypes[] = INITQFNAMES;
 127 static struct quota_format_type *quota_formats; /* List of registered formats */
 128 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 129
 130 int register_quota_format(struct quota_format_type *fmt)
 131 {
 132         spin_lock(&dq_list_lock);
 133         fmt->qf_next = quota_formats;
 134         quota_formats = fmt;
 135         spin_unlock(&dq_list_lock);
 136         return 0;
 137 }
 138
 139 void unregister_quota_format(struct quota_format_type *fmt)
 140 {
 141         struct quota_format_type **actqf;
 142
 143         spin_lock(&dq_list_lock);
 144         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
 145         if (*actqf)
 146                 *actqf = (*actqf)->qf_next;
 147         spin_unlock(&dq_list_lock);
 148 }
 149
 150 static struct quota_format_type *find_quota_format(int id)
 151 {
 152         struct quota_format_type *actqf;
 153
 154         spin_lock(&dq_list_lock);
 155         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 156         if (!actqf || !try_module_get(actqf->qf_owner)) {
 157                 int qm;
 158
 159                 spin_unlock(&dq_list_lock);
 160
 161                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
 162                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
 163                         return NULL;
 164
 165                 spin_lock(&dq_list_lock);
 166                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 167                 if (actqf && !try_module_get(actqf->qf_owner))
 168                         actqf = NULL;
 169         }
 170         spin_unlock(&dq_list_lock);
 171         return actqf;
 172 }
 173
 174 static void put_quota_format(struct quota_format_type *fmt)
 175 {
 176         module_put(fmt->qf_owner);
 177 }
 178
 179 /*
 180  * Dquot List Management:
 181  * The quota code uses three lists for dquot management: the inuse_list,
 182  * free_dquots, and dquot_hash[] array. A single dquot structure may be
 183  * on all three lists, depending on its current state.
 184  *
 185  * All dquots are placed to the end of inuse_list when first created, and this
 186  * list is used for the sync and invalidate operations, which must look
 187  * at every dquot.
 188  *
 189  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 190  * and this list is searched whenever we need an available dquot.  Dquots are
 191  * removed from the list as soon as they are used again, and
 192  * dqstats.free_dquots gives the number of dquots on the list. When
 193  * dquot is invalidated it's completely released from memory.
 194  *
 195  * Dquots with a specific identity (device, type and id) are placed on
 196  * one of the dquot_hash[] hash chains. The provides an efficient search
 197  * mechanism to locate a specific dquot.
 198  */
 199
 200 static LIST_HEAD(inuse_list);
 201 static LIST_HEAD(free_dquots);
 202 unsigned int dq_hash_bits, dq_hash_mask;
 203 static struct hlist_head *dquot_hash;
 204
 205 struct dqstats dqstats;
 206
 207 static void dqput(struct dquot *dquot);
 208
 209 static inline unsigned int
 210 hashfn(const struct super_block *sb, unsigned int id, int type)
 211 {
 212         unsigned long tmp;
 213
 214         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 215         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 216 }
 217
 218 /*
 219  * Following list functions expect dq_list_lock to be held
 220  */
 221 static inline void insert_dquot_hash(struct dquot *dquot)
 222 {
 223         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
 224         hlist_add_head(&dquot->dq_hash, head);
 225 }
 226
 227 static inline void remove_dquot_hash(struct dquot *dquot)
 228 {
 229         hlist_del_init(&dquot->dq_hash);
 230 }
 231
 232 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
 233 {
 234         struct hlist_node *node;
 235         struct dquot *dquot;
 236
 237         hlist_for_each (node, dquot_hash+hashent) {
 238                 dquot = hlist_entry(node, struct dquot, dq_hash);
 239                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
 240                         return dquot;
 241         }
 242         return NODQUOT;
 243 }
 244
 245 /* Add a dquot to the tail of the free list */
 246 static inline void put_dquot_last(struct dquot *dquot)
 247 {
 248         list_add(&dquot->dq_free, free_dquots.prev);
 249         dqstats.free_dquots++;
 250 }
 251
 252 static inline void remove_free_dquot(struct dquot *dquot)
 253 {
 254         if (list_empty(&dquot->dq_free))
 255                 return;
 256         list_del_init(&dquot->dq_free);
 257         dqstats.free_dquots--;
 258 }
 259
 260 static inline void put_inuse(struct dquot *dquot)
 261 {
 262         /* We add to the back of inuse list so we don't have to restart
 263          * when traversing this list and we block */
 264         list_add(&dquot->dq_inuse, inuse_list.prev);
 265         dqstats.allocated_dquots++;
 266 }
 267
 268 static inline void remove_inuse(struct dquot *dquot)
 269 {
 270         dqstats.allocated_dquots--;
 271         list_del(&dquot->dq_inuse);
 272 }
 273 /*
 274  * End of list functions needing dq_list_lock
 275  */
 276
 277 static void wait_on_dquot(struct dquot *dquot)
 278 {
 279         down(&dquot->dq_lock);
 280         up(&dquot->dq_lock);
 281 }
 282
 283 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
 284
 285 int dquot_mark_dquot_dirty(struct dquot *dquot)
 286 {
 287         spin_lock(&dq_list_lock);
 288         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
 289                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 290                                 info[dquot->dq_type].dqi_dirty_list);
 291         spin_unlock(&dq_list_lock);
 292         return 0;
 293 }
 294
 295 /* This function needs dq_list_lock */
 296 static inline int clear_dquot_dirty(struct dquot *dquot)
 297 {
 298         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
 299                 return 0;
 300         list_del_init(&dquot->dq_dirty);
 301         return 1;
 302 }
 303
 304 void mark_info_dirty(struct super_block *sb, int type)
 305 {
 306         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
 307 }
 308 EXPORT_SYMBOL(mark_info_dirty);
 309
 310 /*
 311  *      Read dquot from disk and alloc space for it
 312  */
 313
 314 int dquot_acquire(struct dquot *dquot)
 315 {
 316         int ret = 0, ret2 = 0;
 317         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 318
 319         down(&dquot->dq_lock);
 320         down(&dqopt->dqio_sem);
 321         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 322                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
 323         if (ret < 0)
 324                 goto out_iolock;
 325         set_bit(DQ_READ_B, &dquot->dq_flags);
 326         /* Instantiate dquot if needed */
 327         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 328                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 329                 /* Write the info if needed */
 330                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 331                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 332                 if (ret < 0)
 333                         goto out_iolock;
 334                 if (ret2 < 0) {
 335                         ret = ret2;
 336                         goto out_iolock;
 337                 }
 338         }
 339         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 340 out_iolock:
 341         up(&dqopt->dqio_sem);
 342         up(&dquot->dq_lock);
 343         return ret;
 344 }
 345
 346 /*
 347  *      Write dquot to disk
 348  */
 349 int dquot_commit(struct dquot *dquot)
 350 {
 351         int ret = 0, ret2 = 0;
 352         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 353
 354         down(&dqopt->dqio_sem);
 355         spin_lock(&dq_list_lock);
 356         if (!clear_dquot_dirty(dquot)) {
 357                 spin_unlock(&dq_list_lock);
 358                 goto out_sem;
 359         }
 360         spin_unlock(&dq_list_lock);
 361         /* Inactive dquot can be only if there was error during read/init
 362          * => we have better not writing it */
 363         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 364                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 365                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 366                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 367                 if (ret >= 0)
 368                         ret = ret2;
 369         }
 370 out_sem:
 371         up(&dqopt->dqio_sem);
 372         return ret;
 373 }
 374
 375 /*
 376  *      Release dquot
 377  */
 378 int dquot_release(struct dquot *dquot)
 379 {
 380         int ret = 0, ret2 = 0;
 381         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 382
 383         down(&dquot->dq_lock);
 384         /* Check whether we are not racing with some other dqget() */
 385         if (atomic_read(&dquot->dq_count) > 1)
 386                 goto out_dqlock;
 387         down(&dqopt->dqio_sem);
 388         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
 389                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
 390                 /* Write the info */
 391                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 392                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 393                 if (ret >= 0)
 394                         ret = ret2;
 395         }
 396         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 397         up(&dqopt->dqio_sem);
 398 out_dqlock:
 399         up(&dquot->dq_lock);
 400         return ret;
 401 }
 402
 403 /* Invalidate all dquots on the list. Note that this function is called after
 404  * quota is disabled and pointers from inodes removed so there cannot be new
 405  * quota users. Also because we hold dqonoff_sem there can be no quota users
 406  * for this sb+type at all. */
 407 static void invalidate_dquots(struct super_block *sb, int type)
 408 {
 409         struct dquot *dquot;
 410         struct list_head *head;
 411
 412         spin_lock(&dq_list_lock);
 413         for (head = inuse_list.next; head != &inuse_list;) {
 414                 dquot = list_entry(head, struct dquot, dq_inuse);
 415                 head = head->next;
 416                 if (dquot->dq_sb != sb)
 417                         continue;
 418                 if (dquot->dq_type != type)
 419                         continue;
 420 #ifdef __DQUOT_PARANOIA
 421                 if (atomic_read(&dquot->dq_count))
 422                         BUG();
 423 #endif
 424                 /* Quota now has no users and it has been written on last dqput() */
 425                 remove_dquot_hash(dquot);
 426                 remove_free_dquot(dquot);
 427                 remove_inuse(dquot);
 428                 kmem_cache_free(dquot_cachep, dquot);
 429         }
 430         spin_unlock(&dq_list_lock);
 431 }
 432
 433 int vfs_quota_sync(struct super_block *sb, int type)
 434 {
 435         struct list_head *dirty;
 436         struct dquot *dquot;
 437         struct quota_info *dqopt = sb_dqopt(sb);
 438         int cnt;
 439
 440         down(&dqopt->dqonoff_sem);
 441         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 442                 if (type != -1 && cnt != type)
 443                         continue;
 444                 if (!sb_has_quota_enabled(sb, cnt))
 445                         continue;
 446                 spin_lock(&dq_list_lock);
 447                 dirty = &dqopt->info[cnt].dqi_dirty_list;
 448                 while (!list_empty(dirty)) {
 449                         dquot = list_entry(dirty->next, struct dquot, dq_dirty);
 450                         /* Dirty and inactive can be only bad dquot... */
 451                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 452                                 clear_dquot_dirty(dquot);
 453                                 continue;
 454                         }
 455                         /* Now we have active dquot from which someone is
 456                          * holding reference so we can safely just increase
 457                          * use count */
 458                         atomic_inc(&dquot->dq_count);
 459                         dqstats.lookups++;
 460                         spin_unlock(&dq_list_lock);
 461                         sb->dq_op->write_dquot(dquot);
 462                         dqput(dquot);
 463                         spin_lock(&dq_list_lock);
 464                 }
 465                 spin_unlock(&dq_list_lock);
 466         }
 467
 468         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 469                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
 470                         && info_dirty(&dqopt->info[cnt]))
 471                         sb->dq_op->write_info(sb, cnt);
 472         spin_lock(&dq_list_lock);
 473         dqstats.syncs++;
 474         spin_unlock(&dq_list_lock);
 475         up(&dqopt->dqonoff_sem);
 476
 477         return 0;
 478 }
 479
 480 /* Free unused dquots from cache */
 481 static void prune_dqcache(int count)
 482 {
 483         struct list_head *head;
 484         struct dquot *dquot;
 485
 486         head = free_dquots.prev;
 487         while (head != &free_dquots && count) {
 488                 dquot = list_entry(head, struct dquot, dq_free);
 489                 remove_dquot_hash(dquot);
 490                 remove_free_dquot(dquot);
 491                 remove_inuse(dquot);
 492                 kmem_cache_free(dquot_cachep, dquot);
 493                 count--;
 494                 head = free_dquots.prev;
 495         }
 496 }
 497
 498 /*
 499  * This is called from kswapd when we think we need some
 500  * more memory
 501  */
 502
 503 static int shrink_dqcache_memory(int nr, unsigned int gfp_mask)
 504 {
 505         int ret;
 506
 507         spin_lock(&dq_list_lock);
 508         if (nr)
 509                 prune_dqcache(nr);
 510         ret = dqstats.allocated_dquots;
 511         spin_unlock(&dq_list_lock);
 512         return ret;
 513 }
 514
 515 /*
 516  * Put reference to dquot
 517  * NOTE: If you change this function please check whether dqput_blocks() works right...
 518  * MUST be called with either dqptr_sem or dqonoff_sem held
 519  */
 520 static void dqput(struct dquot *dquot)
 521 {
 522         if (!dquot)
 523                 return;
 524 #ifdef __DQUOT_PARANOIA
 525         if (!atomic_read(&dquot->dq_count)) {
 526                 printk("VFS: dqput: trying to free free dquot\n");
 527                 printk("VFS: device %s, dquot of %s %d\n",
 528                         dquot->dq_sb->s_id,
 529                         quotatypes[dquot->dq_type],
 530                         dquot->dq_id);
 531                 BUG();
 532         }
 533 #endif
 534
 535         spin_lock(&dq_list_lock);
 536         dqstats.drops++;
 537         spin_unlock(&dq_list_lock);
 538 we_slept:
 539         spin_lock(&dq_list_lock);
 540         if (atomic_read(&dquot->dq_count) > 1) {
 541                 /* We have more than one user... nothing to do */
 542                 atomic_dec(&dquot->dq_count);
 543                 spin_unlock(&dq_list_lock);
 544                 return;
 545         }
 546         /* Need to release dquot? */
 547         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 548                 spin_unlock(&dq_list_lock);
 549                 /* Commit dquot before releasing */
 550                 dquot->dq_sb->dq_op->write_dquot(dquot);
 551                 goto we_slept;
 552         }
 553         /* Clear flag in case dquot was inactive (something bad happened) */
 554         clear_dquot_dirty(dquot);
 555         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 556                 spin_unlock(&dq_list_lock);
 557                 dquot->dq_sb->dq_op->release_dquot(dquot);
 558                 goto we_slept;
 559         }
 560         atomic_dec(&dquot->dq_count);
 561 #ifdef __DQUOT_PARANOIA
 562         /* sanity check */
 563         if (!list_empty(&dquot->dq_free))
 564                 BUG();
 565 #endif
 566         put_dquot_last(dquot);
 567         spin_unlock(&dq_list_lock);
 568 }
 569
 570 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 571 {
 572         struct dquot *dquot;
 573
 574         dquot = kmem_cache_alloc(dquot_cachep, SLAB_NOFS);
 575         if(!dquot)
 576                 return NODQUOT;
 577
 578         memset((caddr_t)dquot, 0, sizeof(struct dquot));
 579         sema_init(&dquot->dq_lock, 1);
 580         INIT_LIST_HEAD(&dquot->dq_free);
 581         INIT_LIST_HEAD(&dquot->dq_inuse);
 582         INIT_HLIST_NODE(&dquot->dq_hash);
 583         INIT_LIST_HEAD(&dquot->dq_dirty);
 584         dquot->dq_sb = sb;
 585         dquot->dq_type = type;
 586         atomic_set(&dquot->dq_count, 1);
 587
 588         return dquot;
 589 }
 590
 591 /*
 592  * Get reference to dquot
 593  * MUST be called with either dqptr_sem or dqonoff_sem held
 594  */
 595 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 596 {
 597         unsigned int hashent = hashfn(sb, id, type);
 598         struct dquot *dquot, *empty = NODQUOT;
 599
 600         if (!sb_has_quota_enabled(sb, type))
 601                 return NODQUOT;
 602 we_slept:
 603         spin_lock(&dq_list_lock);
 604         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 605                 if (empty == NODQUOT) {
 606                         spin_unlock(&dq_list_lock);
 607                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 608                                 schedule();     /* Try to wait for a moment... */
 609                         goto we_slept;
 610                 }
 611                 dquot = empty;
 612                 dquot->dq_id = id;
 613                 /* all dquots go on the inuse_list */
 614                 put_inuse(dquot);
 615                 /* hash it first so it can be found */
 616                 insert_dquot_hash(dquot);
 617                 dqstats.lookups++;
 618                 spin_unlock(&dq_list_lock);
 619         } else {
 620                 if (!atomic_read(&dquot->dq_count))
 621                         remove_free_dquot(dquot);
 622                 atomic_inc(&dquot->dq_count);
 623                 dqstats.cache_hits++;
 624                 dqstats.lookups++;
 625                 spin_unlock(&dq_list_lock);
 626                 if (empty)
 627                         kmem_cache_free(dquot_cachep, empty);
 628         }
 629         /* Wait for dq_lock - after this we know that either dquot_release() is already
 630          * finished or it will be canceled due to dq_count > 1 test */
 631         wait_on_dquot(dquot);
 632         /* Read the dquot and instantiate it (everything done only if needed) */
 633         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 634                 dqput(dquot);
 635                 return NODQUOT;
 636         }
 637 #ifdef __DQUOT_PARANOIA
 638         if (!dquot->dq_sb)      /* Has somebody invalidated entry under us? */
 639                 BUG();
 640 #endif
 641
 642         return dquot;
 643 }
 644
 645 static int dqinit_needed(struct inode *inode, int type)
 646 {
 647         int cnt;
 648
 649         if (IS_NOQUOTA(inode))
 650                 return 0;
 651         if (type != -1)
 652                 return inode->i_dquot[type] == NODQUOT;
 653         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 654                 if (inode->i_dquot[cnt] == NODQUOT)
 655                         return 1;
 656         return 0;
 657 }
 658
 659 /* This routine is guarded by dqonoff_sem semaphore */
 660 static void add_dquot_ref(struct super_block *sb, int type)
 661 {
 662         struct list_head *p;
 663
 664 restart:
 665         file_list_lock();
 666         list_for_each(p, &sb->s_files) {
 667                 struct file *filp = list_entry(p, struct file, f_list);
 668                 struct inode *inode = filp->f_dentry->d_inode;
 669                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
 670                         struct dentry *dentry = dget(filp->f_dentry);
 671                         file_list_unlock();
 672                         sb->dq_op->initialize(inode, type);
 673                         dput(dentry);
 674                         /* As we may have blocked we had better restart... */
 675                         goto restart;
 676                 }
 677         }
 678         file_list_unlock();
 679 }
 680
 681 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 682 static inline int dqput_blocks(struct dquot *dquot)
 683 {
 684         if (atomic_read(&dquot->dq_count) <= 1)
 685                 return 1;
 686         return 0;
 687 }
 688
 689 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 690 /* We can't race with anybody because we hold dqptr_sem for writing... */
 691 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
 692 {
 693         struct dquot *dquot = inode->i_dquot[type];
 694
 695         inode->i_dquot[type] = NODQUOT;
 696         if (dquot != NODQUOT) {
 697                 if (dqput_blocks(dquot)) {
 698 #ifdef __DQUOT_PARANOIA
 699                         if (atomic_read(&dquot->dq_count) != 1)
 700                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 701 #endif
 702                         spin_lock(&dq_list_lock);
 703                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 704                         spin_unlock(&dq_list_lock);
 705                         return 1;
 706                 }
 707                 else
 708                         dqput(dquot);   /* We have guaranteed we won't block */
 709         }
 710         return 0;
 711 }
 712
 713 /* Free list of dquots - called from inode.c */
 714 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 715 static void put_dquot_list(struct list_head *tofree_head)
 716 {
 717         struct list_head *act_head;
 718         struct dquot *dquot;
 719
 720         act_head = tofree_head->next;
 721         /* So now we have dquots on the list... Just free them */
 722         while (act_head != tofree_head) {
 723                 dquot = list_entry(act_head, struct dquot, dq_free);
 724                 act_head = act_head->next;
 725                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 726                 dqput(dquot);
 727         }
 728 }
 729
 730 /* Gather all references from inodes and drop them */
 731 static void drop_dquot_ref(struct super_block *sb, int type)
 732 {
 733         LIST_HEAD(tofree_head);
 734
 735         /* We need to be guarded against prune_icache to reach all the
 736          * inodes - otherwise some can be on the local list of prune_icache */
 737         down(&iprune_sem);
 738         down_write(&sb_dqopt(sb)->dqptr_sem);
 739         remove_dquot_ref(sb, type, &tofree_head);
 740         up_write(&sb_dqopt(sb)->dqptr_sem);
 741         up(&iprune_sem);
 742         put_dquot_list(&tofree_head);
 743 }
 744
 745 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
 746 {
 747         dquot->dq_dqb.dqb_curinodes += number;
 748 }
 749
 750 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 751 {
 752         dquot->dq_dqb.dqb_curspace += number;
 753 }
 754
 755 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
 756 {
 757         if (dquot->dq_dqb.dqb_curinodes > number)
 758                 dquot->dq_dqb.dqb_curinodes -= number;
 759         else
 760                 dquot->dq_dqb.dqb_curinodes = 0;
 761         if (dquot->dq_dqb.dqb_curinodes < dquot->dq_dqb.dqb_isoftlimit)
 762                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 763         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 764 }
 765
 766 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 767 {
 768         if (dquot->dq_dqb.dqb_curspace > number)
 769                 dquot->dq_dqb.dqb_curspace -= number;
 770         else
 771                 dquot->dq_dqb.dqb_curspace = 0;
 772         if (toqb(dquot->dq_dqb.dqb_curspace) < dquot->dq_dqb.dqb_bsoftlimit)
 773                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 774         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 775 }
 776
 777 static inline int need_print_warning(struct dquot *dquot)
 778 {
 779         switch (dquot->dq_type) {
 780                 case USRQUOTA:
 781                         return current->fsuid == dquot->dq_id;
 782                 case GRPQUOTA:
 783                         return in_group_p(dquot->dq_id);
 784         }
 785         return 0;
 786 }
 787
 788 /* Values of warnings */
 789 #define NOWARN 0
 790 #define IHARDWARN 1
 791 #define ISOFTLONGWARN 2
 792 #define ISOFTWARN 3
 793 #define BHARDWARN 4
 794 #define BSOFTLONGWARN 5
 795 #define BSOFTWARN 6
 796
 797 /* Print warning to user which exceeded quota */
 798 static void print_warning(struct dquot *dquot, const char warntype)
 799 {
 800         char *msg = NULL;
 801         int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
 802           ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
 803
 804         if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
 805                 return;
 806         tty_write_message(current->signal->tty, dquot->dq_sb->s_id);
 807         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
 808                 tty_write_message(current->signal->tty, ": warning, ");
 809         else
 810                 tty_write_message(current->signal->tty, ": write failed, ");
 811         tty_write_message(current->signal->tty, quotatypes[dquot->dq_type]);
 812         switch (warntype) {
 813                 case IHARDWARN:
 814                         msg = " file limit reached.\n";
 815                         break;
 816                 case ISOFTLONGWARN:
 817                         msg = " file quota exceeded too long.\n";
 818                         break;
 819                 case ISOFTWARN:
 820                         msg = " file quota exceeded.\n";
 821                         break;
 822                 case BHARDWARN:
 823                         msg = " block limit reached.\n";
 824                         break;
 825                 case BSOFTLONGWARN:
 826                         msg = " block quota exceeded too long.\n";
 827                         break;
 828                 case BSOFTWARN:
 829                         msg = " block quota exceeded.\n";
 830                         break;
 831         }
 832         tty_write_message(current->signal->tty, msg);
 833 }
 834
 835 static inline void flush_warnings(struct dquot **dquots, char *warntype)
 836 {
 837         int i;
 838
 839         for (i = 0; i < MAXQUOTAS; i++)
 840                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
 841                         print_warning(dquots[i], warntype[i]);
 842 }
 843
 844 static inline char ignore_hardlimit(struct dquot *dquot)
 845 {
 846         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
 847
 848         return capable(CAP_SYS_RESOURCE) &&
 849             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
 850 }
 851
 852 /* needs dq_data_lock */
 853 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
 854 {
 855         *warntype = NOWARN;
 856         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 857                 return QUOTA_OK;
 858
 859         if (dquot->dq_dqb.dqb_ihardlimit &&
 860            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
 861             !ignore_hardlimit(dquot)) {
 862                 *warntype = IHARDWARN;
 863                 return NO_QUOTA;
 864         }
 865
 866         if (dquot->dq_dqb.dqb_isoftlimit &&
 867            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 868             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
 869             !ignore_hardlimit(dquot)) {
 870                 *warntype = ISOFTLONGWARN;
 871                 return NO_QUOTA;
 872         }
 873
 874         if (dquot->dq_dqb.dqb_isoftlimit &&
 875            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 876             dquot->dq_dqb.dqb_itime == 0) {
 877                 *warntype = ISOFTWARN;
 878                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
 879         }
 880
 881         return QUOTA_OK;
 882 }
 883
 884 /* needs dq_data_lock */
 885 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
 886 {
 887         *warntype = 0;
 888         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 889                 return QUOTA_OK;
 890
 891         if (dquot->dq_dqb.dqb_bhardlimit &&
 892            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
 893             !ignore_hardlimit(dquot)) {
 894                 if (!prealloc)
 895                         *warntype = BHARDWARN;
 896                 return NO_QUOTA;
 897         }
 898
 899         if (dquot->dq_dqb.dqb_bsoftlimit &&
 900            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 901             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
 902             !ignore_hardlimit(dquot)) {
 903                 if (!prealloc)
 904                         *warntype = BSOFTLONGWARN;
 905                 return NO_QUOTA;
 906         }
 907
 908         if (dquot->dq_dqb.dqb_bsoftlimit &&
 909            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 910             dquot->dq_dqb.dqb_btime == 0) {
 911                 if (!prealloc) {
 912                         *warntype = BSOFTWARN;
 913                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
 914                 }
 915                 else
 916                         /*
 917                          * We don't allow preallocation to exceed softlimit so exceeding will
 918                          * be always printed
 919                          */
 920                         return NO_QUOTA;
 921         }
 922
 923         return QUOTA_OK;
 924 }
 925
 926 /*
 927  *      Initialize quota pointers in inode
 928  *      Transaction must be started at entry
 929  */
 930 int dquot_initialize(struct inode *inode, int type)
 931 {
 932         unsigned int id = 0;
 933         int cnt, ret = 0;
 934
 935         /* First test before acquiring semaphore - solves deadlocks when we
 936          * re-enter the quota code and are already holding the semaphore */
 937         if (IS_NOQUOTA(inode))
 938                 return 0;
 939         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 940         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
 941         if (IS_NOQUOTA(inode))
 942                 goto out_err;
 943         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 944                 if (type != -1 && cnt != type)
 945                         continue;
 946                 if (inode->i_dquot[cnt] == NODQUOT) {
 947                         switch (cnt) {
 948                                 case USRQUOTA:
 949                                         id = inode->i_uid;
 950                                         break;
 951                                 case GRPQUOTA:
 952                                         id = inode->i_gid;
 953                                         break;
 954                         }
 955                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
 956                 }
 957         }
 958 out_err:
 959         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 960         return ret;
 961 }
 962
 963 /*
 964  *      Release all quotas referenced by inode
 965  *      Transaction must be started at an entry
 966  */
 967 int dquot_drop(struct inode *inode)
 968 {
 969         int cnt;
 970
 971         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 972         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 973                 if (inode->i_dquot[cnt] != NODQUOT) {
 974                         dqput(inode->i_dquot[cnt]);
 975                         inode->i_dquot[cnt] = NODQUOT;
 976                 }
 977         }
 978         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 979         return 0;
 980 }
 981
 982 /*
 983  * Following four functions update i_blocks+i_bytes fields and
 984  * quota information (together with appropriate checks)
 985  * NOTE: We absolutely rely on the fact that caller dirties
 986  * the inode (usually macros in quotaops.h care about this) and
 987  * holds a handle for the current transaction so that dquot write and
 988  * inode write go into the same transaction.
 989  */
 990
 991 /*
 992  * This operation can block, but only after everything is updated
 993  */
 994 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
 995 {
 996         int cnt, ret = NO_QUOTA;
 997         char warntype[MAXQUOTAS];
 998
 999         /* First test before acquiring semaphore - solves deadlocks when we
1000          * re-enter the quota code and are already holding the semaphore */
1001         if (IS_NOQUOTA(inode)) {
1002 out_add:
1003                 inode_add_bytes(inode, number);
1004                 return QUOTA_OK;
1005         }
1006         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1007                 warntype[cnt] = NOWARN;
1008
1009         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1010         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1011                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1012                 goto out_add;
1013         }
1014         spin_lock(&dq_data_lock);
1015         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1016                 if (inode->i_dquot[cnt] == NODQUOT)
1017                         continue;
1018                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1019                         goto warn_put_all;
1020         }
1021         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1022                 if (inode->i_dquot[cnt] == NODQUOT)
1023                         continue;
1024                 dquot_incr_space(inode->i_dquot[cnt], number);
1025         }
1026         inode_add_bytes(inode, number);
1027         ret = QUOTA_OK;
1028 warn_put_all:
1029         spin_unlock(&dq_data_lock);
1030         if (ret == QUOTA_OK)
1031                 /* Dirtify all the dquots - this can block when journalling */
1032                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1033                         if (inode->i_dquot[cnt])
1034                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1035         flush_warnings(inode->i_dquot, warntype);
1036         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1037         return ret;
1038 }
1039
1040 /*
1041  * This operation can block, but only after everything is updated
1042  */
1043 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1044 {
1045         int cnt, ret = NO_QUOTA;
1046         char warntype[MAXQUOTAS];
1047
1048         /* First test before acquiring semaphore - solves deadlocks when we
1049          * re-enter the quota code and are already holding the semaphore */
1050         if (IS_NOQUOTA(inode))
1051                 return QUOTA_OK;
1052         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1053                 warntype[cnt] = NOWARN;
1054         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1055         if (IS_NOQUOTA(inode)) {
1056                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1057                 return QUOTA_OK;
1058         }
1059         spin_lock(&dq_data_lock);
1060         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1061                 if (inode->i_dquot[cnt] == NODQUOT)
1062                         continue;
1063                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1064                         goto warn_put_all;
1065         }
1066
1067         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1068                 if (inode->i_dquot[cnt] == NODQUOT)
1069                         continue;
1070                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1071         }
1072         ret = QUOTA_OK;
1073 warn_put_all:
1074         spin_unlock(&dq_data_lock);
1075         if (ret == QUOTA_OK)
1076                 /* Dirtify all the dquots - this can block when journalling */
1077                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1078                         if (inode->i_dquot[cnt])
1079                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1080         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1081         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1082         return ret;
1083 }
1084
1085 /*
1086  * This is a non-blocking operation.
1087  */
1088 int dquot_free_space(struct inode *inode, qsize_t number)
1089 {
1090         unsigned int cnt;
1091
1092         /* First test before acquiring semaphore - solves deadlocks when we
1093          * re-enter the quota code and are already holding the semaphore */
1094         if (IS_NOQUOTA(inode)) {
1095 out_sub:
1096                 inode_sub_bytes(inode, number);
1097                 return QUOTA_OK;
1098         }
1099         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1100         /* Now recheck reliably when holding dqptr_sem */
1101         if (IS_NOQUOTA(inode)) {
1102                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1103                 goto out_sub;
1104         }
1105         spin_lock(&dq_data_lock);
1106         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1107                 if (inode->i_dquot[cnt] == NODQUOT)
1108                         continue;
1109                 dquot_decr_space(inode->i_dquot[cnt], number);
1110         }
1111         inode_sub_bytes(inode, number);
1112         spin_unlock(&dq_data_lock);
1113         /* Dirtify all the dquots - this can block when journalling */
1114         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1115                 if (inode->i_dquot[cnt])
1116                         mark_dquot_dirty(inode->i_dquot[cnt]);
1117         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1118         return QUOTA_OK;
1119 }
1120
1121 /*
1122  * This is a non-blocking operation.
1123  */
1124 int dquot_free_inode(const struct inode *inode, unsigned long number)
1125 {
1126         unsigned int cnt;
1127
1128         /* First test before acquiring semaphore - solves deadlocks when we
1129          * re-enter the quota code and are already holding the semaphore */
1130         if (IS_NOQUOTA(inode))
1131                 return QUOTA_OK;
1132         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1133         /* Now recheck reliably when holding dqptr_sem */
1134         if (IS_NOQUOTA(inode)) {
1135                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1136                 return QUOTA_OK;
1137         }
1138         spin_lock(&dq_data_lock);
1139         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1140                 if (inode->i_dquot[cnt] == NODQUOT)
1141                         continue;
1142                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1143         }
1144         spin_unlock(&dq_data_lock);
1145         /* Dirtify all the dquots - this can block when journalling */
1146         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1147                 if (inode->i_dquot[cnt])
1148                         mark_dquot_dirty(inode->i_dquot[cnt]);
1149         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1150         return QUOTA_OK;
1151 }
1152
1153 /*
1154  * Transfer the number of inode and blocks from one diskquota to an other.
1155  *
1156  * This operation can block, but only after everything is updated
1157  */
1158 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1159 {
1160         qsize_t space;
1161         struct dquot *transfer_from[MAXQUOTAS];
1162         struct dquot *transfer_to[MAXQUOTAS];
1163         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1164             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1165         char warntype[MAXQUOTAS];
1166
1167         /* First test before acquiring semaphore - solves deadlocks when we
1168          * re-enter the quota code and are already holding the semaphore */
1169         if (IS_NOQUOTA(inode))
1170                 return QUOTA_OK;
1171         /* Clear the arrays */
1172         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1173                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1174                 warntype[cnt] = NOWARN;
1175         }
1176         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1177         /* Now recheck reliably when holding dqptr_sem */
1178         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1179                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1180                 return QUOTA_OK;
1181         }
1182         /* First build the transfer_to list - here we can block on
1183          * reading/instantiating of dquots.  We know that the transaction for
1184          * us was already started so we don't violate lock ranking here */
1185         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1186                 switch (cnt) {
1187                         case USRQUOTA:
1188                                 if (!chuid)
1189                                         continue;
1190                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1191                                 break;
1192                         case GRPQUOTA:
1193                                 if (!chgid)
1194                                         continue;
1195                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1196                                 break;
1197                 }
1198         }
1199         spin_lock(&dq_data_lock);
1200         space = inode_get_bytes(inode);
1201         /* Build the transfer_from list and check the limits */
1202         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1203                 if (transfer_to[cnt] == NODQUOT)
1204                         continue;
1205                 transfer_from[cnt] = inode->i_dquot[cnt];
1206                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1207                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1208                         goto warn_put_all;
1209         }
1210
1211         /*
1212          * Finally perform the needed transfer from transfer_from to transfer_to
1213          */
1214         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1215                 /*
1216                  * Skip changes for same uid or gid or for turned off quota-type.
1217                  */
1218                 if (transfer_to[cnt] == NODQUOT)
1219                         continue;
1220
1221                 /* Due to IO error we might not have transfer_from[] structure */
1222                 if (transfer_from[cnt]) {
1223                         dquot_decr_inodes(transfer_from[cnt], 1);
1224                         dquot_decr_space(transfer_from[cnt], space);
1225                 }
1226
1227                 dquot_incr_inodes(transfer_to[cnt], 1);
1228                 dquot_incr_space(transfer_to[cnt], space);
1229
1230                 inode->i_dquot[cnt] = transfer_to[cnt];
1231         }
1232         ret = QUOTA_OK;
1233 warn_put_all:
1234         spin_unlock(&dq_data_lock);
1235         /* Dirtify all the dquots - this can block when journalling */
1236         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1237                 if (transfer_from[cnt])
1238                         mark_dquot_dirty(transfer_from[cnt]);
1239                 if (transfer_to[cnt])
1240                         mark_dquot_dirty(transfer_to[cnt]);
1241         }
1242         flush_warnings(transfer_to, warntype);
1243
1244         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1245                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1246                         dqput(transfer_from[cnt]);
1247                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1248                         dqput(transfer_to[cnt]);
1249         }
1250         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1251         return ret;
1252 }
1253
1254 /*
1255  * Write info of quota file to disk
1256  */
1257 int dquot_commit_info(struct super_block *sb, int type)
1258 {
1259         int ret;
1260         struct quota_info *dqopt = sb_dqopt(sb);
1261
1262         down(&dqopt->dqio_sem);
1263         ret = dqopt->ops[type]->write_file_info(sb, type);
1264         up(&dqopt->dqio_sem);
1265         return ret;
1266 }
1267
1268 /*
1269  * Definitions of diskquota operations.
1270  */
1271 struct dquot_operations dquot_operations = {
1272         .initialize     = dquot_initialize,
1273         .drop           = dquot_drop,
1274         .alloc_space    = dquot_alloc_space,
1275         .alloc_inode    = dquot_alloc_inode,
1276         .free_space     = dquot_free_space,
1277         .free_inode     = dquot_free_inode,
1278         .transfer       = dquot_transfer,
1279         .write_dquot    = dquot_commit,
1280         .acquire_dquot  = dquot_acquire,
1281         .release_dquot  = dquot_release,
1282         .mark_dirty     = dquot_mark_dquot_dirty,
1283         .write_info     = dquot_commit_info
1284 };
1285
1286 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1287 {
1288         switch (type) {
1289                 case USRQUOTA:
1290                         dqopt->flags |= DQUOT_USR_ENABLED;
1291                         break;
1292                 case GRPQUOTA:
1293                         dqopt->flags |= DQUOT_GRP_ENABLED;
1294                         break;
1295         }
1296 }
1297
1298 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1299 {
1300         switch (type) {
1301                 case USRQUOTA:
1302                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1303                         break;
1304                 case GRPQUOTA:
1305                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1306                         break;
1307         }
1308 }
1309
1310 /*
1311  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1312  */
1313 int vfs_quota_off(struct super_block *sb, int type)
1314 {
1315         int cnt;
1316         struct quota_info *dqopt = sb_dqopt(sb);
1317
1318         /* We need to serialize quota_off() for device */
1319         down(&dqopt->dqonoff_sem);
1320         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1321                 if (type != -1 && cnt != type)
1322                         continue;
1323                 if (!sb_has_quota_enabled(sb, cnt))
1324                         continue;
1325                 reset_enable_flags(dqopt, cnt);
1326
1327                 /* Note: these are blocking operations */
1328                 drop_dquot_ref(sb, cnt);
1329                 invalidate_dquots(sb, cnt);
1330                 /*
1331                  * Now all dquots should be invalidated, all writes done so we should be only
1332                  * users of the info. No locks needed.
1333                  */
1334                 if (info_dirty(&dqopt->info[cnt]))
1335                         sb->dq_op->write_info(sb, cnt);
1336                 if (dqopt->ops[cnt]->free_file_info)
1337                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1338                 put_quota_format(dqopt->info[cnt].dqi_format);
1339
1340                 fput(dqopt->files[cnt]);
1341                 dqopt->files[cnt] = NULL;
1342                 dqopt->info[cnt].dqi_flags = 0;
1343                 dqopt->info[cnt].dqi_igrace = 0;
1344                 dqopt->info[cnt].dqi_bgrace = 0;
1345                 dqopt->ops[cnt] = NULL;
1346         }
1347         up(&dqopt->dqonoff_sem);
1348         return 0;
1349 }
1350
1351 /*
1352  *      Turn quotas on on a device
1353  */
1354
1355 /* Helper function when we already have file open */
1356 static int vfs_quota_on_file(struct file *f, int type, int format_id)
1357 {
1358         struct quota_format_type *fmt = find_quota_format(format_id);
1359         struct inode *inode;
1360         struct super_block *sb = f->f_dentry->d_sb;
1361         struct quota_info *dqopt = sb_dqopt(sb);
1362         struct dquot *to_drop[MAXQUOTAS];
1363         int error, cnt;
1364         unsigned int oldflags = -1;
1365
1366         if (!fmt)
1367                 return -ESRCH;
1368         error = -EIO;
1369         if (!f->f_op || !f->f_op->read || !f->f_op->write)
1370                 goto out_fmt;
1371         inode = f->f_dentry->d_inode;
1372         error = -EACCES;
1373         if (!S_ISREG(inode->i_mode))
1374                 goto out_fmt;
1375
1376         down(&inode->i_sem);
1377         down(&dqopt->dqonoff_sem);
1378         if (sb_has_quota_enabled(sb, type)) {
1379                 up(&inode->i_sem);
1380                 error = -EBUSY;
1381                 goto out_lock;
1382         }
1383         /* We don't want quota and atime on quota files (deadlocks possible)
1384          * We also need to set GFP mask differently because we cannot recurse
1385          * into filesystem when allocating page for quota inode */
1386         down_write(&dqopt->dqptr_sem);
1387         oldflags = inode->i_flags & (S_NOATIME | S_NOQUOTA);
1388         inode->i_flags |= S_NOQUOTA | S_NOATIME;
1389         up_write(&dqopt->dqptr_sem);
1390         up(&inode->i_sem);
1391
1392         dqopt->files[type] = f;
1393         error = -EINVAL;
1394         if (!fmt->qf_ops->check_quota_file(sb, type))
1395                 goto out_file_init;
1396         /*
1397          * We write to quota files deep within filesystem code.  We don't want
1398          * the VFS to reenter filesystem code when it tries to allocate a
1399          * pagecache page for the quota file write.  So clear __GFP_FS in
1400          * the quota file's allocation flags.
1401          */
1402         mapping_set_gfp_mask(inode->i_mapping,
1403                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
1404
1405         down_write(&dqopt->dqptr_sem);
1406         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1407                 to_drop[cnt] = inode->i_dquot[cnt];
1408                 inode->i_dquot[cnt] = NODQUOT;
1409         }
1410         up_write(&dqopt->dqptr_sem);
1411         /* We must put dquots outside of dqptr_sem because we may need to
1412          * start transaction for dquot_release() */
1413         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1414                 if (to_drop[cnt])
1415                         dqput(to_drop[cnt]);
1416         }
1417
1418         dqopt->ops[type] = fmt->qf_ops;
1419         dqopt->info[type].dqi_format = fmt;
1420         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1421         down(&dqopt->dqio_sem);
1422         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1423                 up(&dqopt->dqio_sem);
1424                 goto out_file_init;
1425         }
1426         up(&dqopt->dqio_sem);
1427         set_enable_flags(dqopt, type);
1428
1429         add_dquot_ref(sb, type);
1430         up(&dqopt->dqonoff_sem);
1431
1432         return 0;
1433
1434 out_file_init:
1435         dqopt->files[type] = NULL;
1436 out_lock:
1437         up(&dqopt->dqonoff_sem);
1438         if (oldflags != -1) {
1439                 down(&inode->i_sem);
1440                 down_write(&dqopt->dqptr_sem);
1441                 /* Reset the NOATIME flag back. I know it could change in the
1442                  * mean time but playing with NOATIME flags on a quota file is
1443                  * never a good idea */
1444                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA);
1445                 inode->i_flags |= oldflags;
1446                 up_write(&dqopt->dqptr_sem);
1447                 up(&inode->i_sem);
1448         }
1449 out_fmt:
1450         put_quota_format(fmt);
1451
1452         return error;
1453 }
1454
1455 /* Actual function called from quotactl() */
1456 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1457 {
1458         struct file *f;
1459         int error;
1460
1461         f = filp_open(path, O_RDWR, 0600);
1462         if (IS_ERR(f))
1463                 return PTR_ERR(f);
1464         error = security_quota_on(f);
1465         if (error)
1466                 goto out_f;
1467         error = vfs_quota_on_file(f, type, format_id);
1468         if (!error)
1469                 return 0;
1470 out_f:
1471         filp_close(f, NULL);
1472         return error;
1473 }
1474
1475 /*
1476  * Function used by filesystems when filp_open() would fail (filesystem is
1477  * being mounted now). We will use a private file structure. Caller is
1478  * responsible that it's IO functions won't need vfsmnt structure or
1479  * some dentry tricks...
1480  */
1481 int vfs_quota_on_mount(int type, int format_id, struct dentry *dentry)
1482 {
1483         struct file *f;
1484         int error;
1485
1486         dget(dentry);   /* Get a reference for struct file */
1487         f = dentry_open(dentry, NULL, O_RDWR);
1488         if (IS_ERR(f)) {
1489                 error = PTR_ERR(f);
1490                 goto out_dentry;
1491         }
1492         error = vfs_quota_on_file(f, type, format_id);
1493         if (!error)
1494                 return 0;
1495         fput(f);
1496 out_dentry:
1497         dput(dentry);
1498         return error;
1499 }
1500
1501 /* Generic routine for getting common part of quota structure */
1502 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1503 {
1504         struct mem_dqblk *dm = &dquot->dq_dqb;
1505
1506         spin_lock(&dq_data_lock);
1507         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1508         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1509         di->dqb_curspace = dm->dqb_curspace;
1510         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1511         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1512         di->dqb_curinodes = dm->dqb_curinodes;
1513         di->dqb_btime = dm->dqb_btime;
1514         di->dqb_itime = dm->dqb_itime;
1515         di->dqb_valid = QIF_ALL;
1516         spin_unlock(&dq_data_lock);
1517 }
1518
1519 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1520 {
1521         struct dquot *dquot;
1522
1523         down(&sb_dqopt(sb)->dqonoff_sem);
1524         if (!(dquot = dqget(sb, id, type))) {
1525                 up(&sb_dqopt(sb)->dqonoff_sem);
1526                 return -ESRCH;
1527         }
1528         do_get_dqblk(dquot, di);
1529         dqput(dquot);
1530         up(&sb_dqopt(sb)->dqonoff_sem);
1531         return 0;
1532 }
1533
1534 /* Generic routine for setting common part of quota structure */
1535 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1536 {
1537         struct mem_dqblk *dm = &dquot->dq_dqb;
1538         int check_blim = 0, check_ilim = 0;
1539
1540         spin_lock(&dq_data_lock);
1541         if (di->dqb_valid & QIF_SPACE) {
1542                 dm->dqb_curspace = di->dqb_curspace;
1543                 check_blim = 1;
1544         }
1545         if (di->dqb_valid & QIF_BLIMITS) {
1546                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1547                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1548                 check_blim = 1;
1549         }
1550         if (di->dqb_valid & QIF_INODES) {
1551                 dm->dqb_curinodes = di->dqb_curinodes;
1552                 check_ilim = 1;
1553         }
1554         if (di->dqb_valid & QIF_ILIMITS) {
1555                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1556                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1557                 check_ilim = 1;
1558         }
1559         if (di->dqb_valid & QIF_BTIME)
1560                 dm->dqb_btime = di->dqb_btime;
1561         if (di->dqb_valid & QIF_ITIME)
1562                 dm->dqb_itime = di->dqb_itime;
1563
1564         if (check_blim) {
1565                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1566                         dm->dqb_btime = 0;
1567                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1568                 }
1569                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1570                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1571         }
1572         if (check_ilim) {
1573                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1574                         dm->dqb_itime = 0;
1575                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1576                 }
1577                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1578                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1579         }
1580         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1581                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1582         else
1583                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1584         spin_unlock(&dq_data_lock);
1585         mark_dquot_dirty(dquot);
1586 }
1587
1588 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1589 {
1590         struct dquot *dquot;
1591
1592         down(&sb_dqopt(sb)->dqonoff_sem);
1593         if (!(dquot = dqget(sb, id, type))) {
1594                 up(&sb_dqopt(sb)->dqonoff_sem);
1595                 return -ESRCH;
1596         }
1597         do_set_dqblk(dquot, di);
1598         dqput(dquot);
1599         up(&sb_dqopt(sb)->dqonoff_sem);
1600         return 0;
1601 }
1602
1603 /* Generic routine for getting common part of quota file information */
1604 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1605 {
1606         struct mem_dqinfo *mi;
1607
1608         down(&sb_dqopt(sb)->dqonoff_sem);
1609         if (!sb_has_quota_enabled(sb, type)) {
1610                 up(&sb_dqopt(sb)->dqonoff_sem);
1611                 return -ESRCH;
1612         }
1613         mi = sb_dqopt(sb)->info + type;
1614         spin_lock(&dq_data_lock);
1615         ii->dqi_bgrace = mi->dqi_bgrace;
1616         ii->dqi_igrace = mi->dqi_igrace;
1617         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1618         ii->dqi_valid = IIF_ALL;
1619         spin_unlock(&dq_data_lock);
1620         up(&sb_dqopt(sb)->dqonoff_sem);
1621         return 0;
1622 }
1623
1624 /* Generic routine for setting common part of quota file information */
1625 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1626 {
1627         struct mem_dqinfo *mi;
1628
1629         down(&sb_dqopt(sb)->dqonoff_sem);
1630         if (!sb_has_quota_enabled(sb, type)) {
1631                 up(&sb_dqopt(sb)->dqonoff_sem);
1632                 return -ESRCH;
1633         }
1634         mi = sb_dqopt(sb)->info + type;
1635         spin_lock(&dq_data_lock);
1636         if (ii->dqi_valid & IIF_BGRACE)
1637                 mi->dqi_bgrace = ii->dqi_bgrace;
1638         if (ii->dqi_valid & IIF_IGRACE)
1639                 mi->dqi_igrace = ii->dqi_igrace;
1640         if (ii->dqi_valid & IIF_FLAGS)
1641                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1642         spin_unlock(&dq_data_lock);
1643         mark_info_dirty(sb, type);
1644         /* Force write to disk */
1645         sb->dq_op->write_info(sb, type);
1646         up(&sb_dqopt(sb)->dqonoff_sem);
1647         return 0;
1648 }
1649
1650 struct quotactl_ops vfs_quotactl_ops = {
1651         .quota_on       = vfs_quota_on,
1652         .quota_off      = vfs_quota_off,
1653         .quota_sync     = vfs_quota_sync,
1654         .get_info       = vfs_get_dqinfo,
1655         .set_info       = vfs_set_dqinfo,
1656         .get_dqblk      = vfs_get_dqblk,
1657         .set_dqblk      = vfs_set_dqblk
1658 };
1659
1660 static ctl_table fs_dqstats_table[] = {
1661         {
1662                 .ctl_name       = FS_DQ_LOOKUPS,
1663                 .procname       = "lookups",
1664                 .data           = &dqstats.lookups,
1665                 .maxlen         = sizeof(int),
1666                 .mode           = 0444,
1667                 .proc_handler   = &proc_dointvec,
1668         },
1669         {
1670                 .ctl_name       = FS_DQ_DROPS,
1671                 .procname       = "drops",
1672                 .data           = &dqstats.drops,
1673                 .maxlen         = sizeof(int),
1674                 .mode           = 0444,
1675                 .proc_handler   = &proc_dointvec,
1676         },
1677         {
1678                 .ctl_name       = FS_DQ_READS,
1679                 .procname       = "reads",
1680                 .data           = &dqstats.reads,
1681                 .maxlen         = sizeof(int),
1682                 .mode           = 0444,
1683                 .proc_handler   = &proc_dointvec,
1684         },
1685         {
1686                 .ctl_name       = FS_DQ_WRITES,
1687                 .procname       = "writes",
1688                 .data           = &dqstats.writes,
1689                 .maxlen         = sizeof(int),
1690                 .mode           = 0444,
1691                 .proc_handler   = &proc_dointvec,
1692         },
1693         {
1694                 .ctl_name       = FS_DQ_CACHE_HITS,
1695                 .procname       = "cache_hits",
1696                 .data           = &dqstats.cache_hits,
1697                 .maxlen         = sizeof(int),
1698                 .mode           = 0444,
1699                 .proc_handler   = &proc_dointvec,
1700         },
1701         {
1702                 .ctl_name       = FS_DQ_ALLOCATED,
1703                 .procname       = "allocated_dquots",
1704                 .data           = &dqstats.allocated_dquots,
1705                 .maxlen         = sizeof(int),
1706                 .mode           = 0444,
1707                 .proc_handler   = &proc_dointvec,
1708         },
1709         {
1710                 .ctl_name       = FS_DQ_FREE,
1711                 .procname       = "free_dquots",
1712                 .data           = &dqstats.free_dquots,
1713                 .maxlen         = sizeof(int),
1714                 .mode           = 0444,
1715                 .proc_handler   = &proc_dointvec,
1716         },
1717         {
1718                 .ctl_name       = FS_DQ_SYNCS,
1719                 .procname       = "syncs",
1720                 .data           = &dqstats.syncs,
1721                 .maxlen         = sizeof(int),
1722                 .mode           = 0444,
1723                 .proc_handler   = &proc_dointvec,
1724         },
1725         { .ctl_name = 0 },
1726 };
1727
1728 static ctl_table fs_table[] = {
1729         {
1730                 .ctl_name       = FS_DQSTATS,
1731                 .procname       = "quota",
1732                 .mode           = 0555,
1733                 .child          = fs_dqstats_table,
1734         },
1735         { .ctl_name = 0 },
1736 };
1737
1738 static ctl_table sys_table[] = {
1739         {
1740                 .ctl_name       = CTL_FS,
1741                 .procname       = "fs",
1742                 .mode           = 0555,
1743                 .child          = fs_table,
1744         },
1745         { .ctl_name = 0 },
1746 };
1747
1748 /* SLAB cache for dquot structures */
1749 kmem_cache_t *dquot_cachep;
1750
1751 static int __init dquot_init(void)
1752 {
1753         int i;
1754         unsigned long nr_hash, order;
1755
1756         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1757
1758         register_sysctl_table(sys_table, 0);
1759
1760         dquot_cachep = kmem_cache_create("dquot",
1761                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1762                         SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1763                         NULL, NULL);
1764
1765         order = 0;
1766         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1767         if (!dquot_hash)
1768                 panic("Cannot create dquot hash table");
1769
1770         /* Find power-of-two hlist_heads which can fit into allocation */
1771         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1772         dq_hash_bits = 0;
1773         do {
1774                 dq_hash_bits++;
1775         } while (nr_hash >> dq_hash_bits);
1776         dq_hash_bits--;
1777
1778         nr_hash = 1UL << dq_hash_bits;
1779         dq_hash_mask = nr_hash - 1;
1780         for (i = 0; i < nr_hash; i++)
1781                 INIT_HLIST_HEAD(dquot_hash + i);
1782
1783         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1784                         nr_hash, order, (PAGE_SIZE << order));
1785
1786         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1787
1788         return 0;
1789 }
1790 module_init(dquot_init);
1791
1792 EXPORT_SYMBOL(register_quota_format);
1793 EXPORT_SYMBOL(unregister_quota_format);
1794 EXPORT_SYMBOL(dqstats);
1795 EXPORT_SYMBOL(dq_data_lock);
1796 EXPORT_SYMBOL(vfs_quota_on);
1797 EXPORT_SYMBOL(vfs_quota_on_mount);
1798 EXPORT_SYMBOL(vfs_quota_off);
1799 EXPORT_SYMBOL(vfs_quota_sync);
1800 EXPORT_SYMBOL(vfs_get_dqinfo);
1801 EXPORT_SYMBOL(vfs_set_dqinfo);
1802 EXPORT_SYMBOL(vfs_get_dqblk);
1803 EXPORT_SYMBOL(vfs_set_dqblk);
1804 EXPORT_SYMBOL(dquot_commit);
1805 EXPORT_SYMBOL(dquot_commit_info);
1806 EXPORT_SYMBOL(dquot_acquire);
1807 EXPORT_SYMBOL(dquot_release);
1808 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1809 EXPORT_SYMBOL(dquot_initialize);
1810 EXPORT_SYMBOL(dquot_drop);
1811 EXPORT_SYMBOL(dquot_alloc_space);
1812 EXPORT_SYMBOL(dquot_alloc_inode);
1813 EXPORT_SYMBOL(dquot_free_space);
1814 EXPORT_SYMBOL(dquot_free_inode);
1815 EXPORT_SYMBOL(dquot_transfer);