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, fix lock inversion problems
  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/namei.h>
  79 #include <linux/buffer_head.h>
  80
  81 #include <asm/uaccess.h>
  82
  83 #define __DQUOT_PARANOIA
  84
  85 /*
  86  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
  87  * and quota formats and also dqstats structure containing statistics about the
  88  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
  89  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
  90  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
  91  * in inode_add_bytes() and inode_sub_bytes().
  92  *
  93  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
  94  *
  95  * Note that some things (eg. sb pointer, type, id) doesn't change during
  96  * the life of the dquot structure and so needn't to be protected by a lock
  97  *
  98  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
  99  * operation is just reading pointers from inode (or not using them at all) the
 100  * read lock is enough. If pointers are altered function must hold write lock
 101  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
 102  * for altering the flag i_sem is also needed).  If operation is holding
 103  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
 104  * dqonoff_sem.
 105  * This locking assures that:
 106  *   a) update/access to dquot pointers in inode is serialized
 107  *   b) everyone is guarded against invalidate_dquots()
 108  *
 109  * Each dquot has its dq_lock semaphore. Locked dquots might not be referenced
 110  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
 111  * Currently dquot is locked only when it is being read to memory (or space for
 112  * it is being allocated) on the first dqget() and when it is being released on
 113  * the last dqput(). The allocation and release oparations are serialized by
 114  * the dq_lock and by checking the use count in dquot_release().  Write
 115  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
 116  * spinlock to internal buffers before writing.
 117  *
 118  * Lock ordering (including related VFS locks) is the following:
 119  *   i_sem > dqonoff_sem > iprune_sem > journal_lock > dqptr_sem >
 120  *   > dquot->dq_lock > dqio_sem
 121  * i_sem on quota files is special (it's below dqio_sem)
 122  */
 123
 124 static DEFINE_SPINLOCK(dq_list_lock);
 125 DEFINE_SPINLOCK(dq_data_lock);
 126
 127 static char *quotatypes[] = INITQFNAMES;
 128 static struct quota_format_type *quota_formats; /* List of registered formats */
 129 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 130
 131 /* SLAB cache for dquot structures */
 132 static kmem_cache_t *dquot_cachep;
 133
 134 int register_quota_format(struct quota_format_type *fmt)
 135 {
 136         spin_lock(&dq_list_lock);
 137         fmt->qf_next = quota_formats;
 138         quota_formats = fmt;
 139         spin_unlock(&dq_list_lock);
 140         return 0;
 141 }
 142
 143 void unregister_quota_format(struct quota_format_type *fmt)
 144 {
 145         struct quota_format_type **actqf;
 146
 147         spin_lock(&dq_list_lock);
 148         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
 149         if (*actqf)
 150                 *actqf = (*actqf)->qf_next;
 151         spin_unlock(&dq_list_lock);
 152 }
 153
 154 static struct quota_format_type *find_quota_format(int id)
 155 {
 156         struct quota_format_type *actqf;
 157
 158         spin_lock(&dq_list_lock);
 159         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 160         if (!actqf || !try_module_get(actqf->qf_owner)) {
 161                 int qm;
 162
 163                 spin_unlock(&dq_list_lock);
 164
 165                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
 166                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
 167                         return NULL;
 168
 169                 spin_lock(&dq_list_lock);
 170                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 171                 if (actqf && !try_module_get(actqf->qf_owner))
 172                         actqf = NULL;
 173         }
 174         spin_unlock(&dq_list_lock);
 175         return actqf;
 176 }
 177
 178 static void put_quota_format(struct quota_format_type *fmt)
 179 {
 180         module_put(fmt->qf_owner);
 181 }
 182
 183 /*
 184  * Dquot List Management:
 185  * The quota code uses three lists for dquot management: the inuse_list,
 186  * free_dquots, and dquot_hash[] array. A single dquot structure may be
 187  * on all three lists, depending on its current state.
 188  *
 189  * All dquots are placed to the end of inuse_list when first created, and this
 190  * list is used for invalidate operation, which must look at every dquot.
 191  *
 192  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 193  * and this list is searched whenever we need an available dquot.  Dquots are
 194  * removed from the list as soon as they are used again, and
 195  * dqstats.free_dquots gives the number of dquots on the list. When
 196  * dquot is invalidated it's completely released from memory.
 197  *
 198  * Dquots with a specific identity (device, type and id) are placed on
 199  * one of the dquot_hash[] hash chains. The provides an efficient search
 200  * mechanism to locate a specific dquot.
 201  */
 202
 203 static LIST_HEAD(inuse_list);
 204 static LIST_HEAD(free_dquots);
 205 static unsigned int dq_hash_bits, dq_hash_mask;
 206 static struct hlist_head *dquot_hash;
 207
 208 struct dqstats dqstats;
 209
 210 static void dqput(struct dquot *dquot);
 211
 212 static inline unsigned int
 213 hashfn(const struct super_block *sb, unsigned int id, int type)
 214 {
 215         unsigned long tmp;
 216
 217         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 218         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 219 }
 220
 221 /*
 222  * Following list functions expect dq_list_lock to be held
 223  */
 224 static inline void insert_dquot_hash(struct dquot *dquot)
 225 {
 226         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
 227         hlist_add_head(&dquot->dq_hash, head);
 228 }
 229
 230 static inline void remove_dquot_hash(struct dquot *dquot)
 231 {
 232         hlist_del_init(&dquot->dq_hash);
 233 }
 234
 235 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
 236 {
 237         struct hlist_node *node;
 238         struct dquot *dquot;
 239
 240         hlist_for_each (node, dquot_hash+hashent) {
 241                 dquot = hlist_entry(node, struct dquot, dq_hash);
 242                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
 243                         return dquot;
 244         }
 245         return NODQUOT;
 246 }
 247
 248 /* Add a dquot to the tail of the free list */
 249 static inline void put_dquot_last(struct dquot *dquot)
 250 {
 251         list_add(&dquot->dq_free, free_dquots.prev);
 252         dqstats.free_dquots++;
 253 }
 254
 255 static inline void remove_free_dquot(struct dquot *dquot)
 256 {
 257         if (list_empty(&dquot->dq_free))
 258                 return;
 259         list_del_init(&dquot->dq_free);
 260         dqstats.free_dquots--;
 261 }
 262
 263 static inline void put_inuse(struct dquot *dquot)
 264 {
 265         /* We add to the back of inuse list so we don't have to restart
 266          * when traversing this list and we block */
 267         list_add(&dquot->dq_inuse, inuse_list.prev);
 268         dqstats.allocated_dquots++;
 269 }
 270
 271 static inline void remove_inuse(struct dquot *dquot)
 272 {
 273         dqstats.allocated_dquots--;
 274         list_del(&dquot->dq_inuse);
 275 }
 276 /*
 277  * End of list functions needing dq_list_lock
 278  */
 279
 280 static void wait_on_dquot(struct dquot *dquot)
 281 {
 282         down(&dquot->dq_lock);
 283         up(&dquot->dq_lock);
 284 }
 285
 286 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
 287
 288 int dquot_mark_dquot_dirty(struct dquot *dquot)
 289 {
 290         spin_lock(&dq_list_lock);
 291         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
 292                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 293                                 info[dquot->dq_type].dqi_dirty_list);
 294         spin_unlock(&dq_list_lock);
 295         return 0;
 296 }
 297
 298 /* This function needs dq_list_lock */
 299 static inline int clear_dquot_dirty(struct dquot *dquot)
 300 {
 301         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
 302                 return 0;
 303         list_del_init(&dquot->dq_dirty);
 304         return 1;
 305 }
 306
 307 void mark_info_dirty(struct super_block *sb, int type)
 308 {
 309         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
 310 }
 311 EXPORT_SYMBOL(mark_info_dirty);
 312
 313 /*
 314  *      Read dquot from disk and alloc space for it
 315  */
 316
 317 int dquot_acquire(struct dquot *dquot)
 318 {
 319         int ret = 0, ret2 = 0;
 320         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 321
 322         down(&dquot->dq_lock);
 323         down(&dqopt->dqio_sem);
 324         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 325                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
 326         if (ret < 0)
 327                 goto out_iolock;
 328         set_bit(DQ_READ_B, &dquot->dq_flags);
 329         /* Instantiate dquot if needed */
 330         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 331                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 332                 /* Write the info if needed */
 333                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 334                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 335                 if (ret < 0)
 336                         goto out_iolock;
 337                 if (ret2 < 0) {
 338                         ret = ret2;
 339                         goto out_iolock;
 340                 }
 341         }
 342         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 343 out_iolock:
 344         up(&dqopt->dqio_sem);
 345         up(&dquot->dq_lock);
 346         return ret;
 347 }
 348
 349 /*
 350  *      Write dquot to disk
 351  */
 352 int dquot_commit(struct dquot *dquot)
 353 {
 354         int ret = 0, ret2 = 0;
 355         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 356
 357         down(&dqopt->dqio_sem);
 358         spin_lock(&dq_list_lock);
 359         if (!clear_dquot_dirty(dquot)) {
 360                 spin_unlock(&dq_list_lock);
 361                 goto out_sem;
 362         }
 363         spin_unlock(&dq_list_lock);
 364         /* Inactive dquot can be only if there was error during read/init
 365          * => we have better not writing it */
 366         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 367                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 368                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 369                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 370                 if (ret >= 0)
 371                         ret = ret2;
 372         }
 373 out_sem:
 374         up(&dqopt->dqio_sem);
 375         return ret;
 376 }
 377
 378 /*
 379  *      Release dquot
 380  */
 381 int dquot_release(struct dquot *dquot)
 382 {
 383         int ret = 0, ret2 = 0;
 384         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 385
 386         down(&dquot->dq_lock);
 387         /* Check whether we are not racing with some other dqget() */
 388         if (atomic_read(&dquot->dq_count) > 1)
 389                 goto out_dqlock;
 390         down(&dqopt->dqio_sem);
 391         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
 392                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
 393                 /* Write the info */
 394                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 395                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 396                 if (ret >= 0)
 397                         ret = ret2;
 398         }
 399         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 400         up(&dqopt->dqio_sem);
 401 out_dqlock:
 402         up(&dquot->dq_lock);
 403         return ret;
 404 }
 405
 406 /* Invalidate all dquots on the list. Note that this function is called after
 407  * quota is disabled and pointers from inodes removed so there cannot be new
 408  * quota users. Also because we hold dqonoff_sem there can be no quota users
 409  * for this sb+type at all. */
 410 static void invalidate_dquots(struct super_block *sb, int type)
 411 {
 412         struct dquot *dquot;
 413         struct list_head *head;
 414
 415         spin_lock(&dq_list_lock);
 416         for (head = inuse_list.next; head != &inuse_list;) {
 417                 dquot = list_entry(head, struct dquot, dq_inuse);
 418                 head = head->next;
 419                 if (dquot->dq_sb != sb)
 420                         continue;
 421                 if (dquot->dq_type != type)
 422                         continue;
 423 #ifdef __DQUOT_PARANOIA
 424                 if (atomic_read(&dquot->dq_count))
 425                         BUG();
 426 #endif
 427                 /* Quota now has no users and it has been written on last dqput() */
 428                 remove_dquot_hash(dquot);
 429                 remove_free_dquot(dquot);
 430                 remove_inuse(dquot);
 431                 kmem_cache_free(dquot_cachep, dquot);
 432         }
 433         spin_unlock(&dq_list_lock);
 434 }
 435
 436 int vfs_quota_sync(struct super_block *sb, int type)
 437 {
 438         struct list_head *dirty;
 439         struct dquot *dquot;
 440         struct quota_info *dqopt = sb_dqopt(sb);
 441         int cnt;
 442
 443         down(&dqopt->dqonoff_sem);
 444         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 445                 if (type != -1 && cnt != type)
 446                         continue;
 447                 if (!sb_has_quota_enabled(sb, cnt))
 448                         continue;
 449                 spin_lock(&dq_list_lock);
 450                 dirty = &dqopt->info[cnt].dqi_dirty_list;
 451                 while (!list_empty(dirty)) {
 452                         dquot = list_entry(dirty->next, struct dquot, dq_dirty);
 453                         /* Dirty and inactive can be only bad dquot... */
 454                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 455                                 clear_dquot_dirty(dquot);
 456                                 continue;
 457                         }
 458                         /* Now we have active dquot from which someone is
 459                          * holding reference so we can safely just increase
 460                          * use count */
 461                         atomic_inc(&dquot->dq_count);
 462                         dqstats.lookups++;
 463                         spin_unlock(&dq_list_lock);
 464                         sb->dq_op->write_dquot(dquot);
 465                         dqput(dquot);
 466                         spin_lock(&dq_list_lock);
 467                 }
 468                 spin_unlock(&dq_list_lock);
 469         }
 470
 471         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 472                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
 473                         && info_dirty(&dqopt->info[cnt]))
 474                         sb->dq_op->write_info(sb, cnt);
 475         spin_lock(&dq_list_lock);
 476         dqstats.syncs++;
 477         spin_unlock(&dq_list_lock);
 478         up(&dqopt->dqonoff_sem);
 479
 480         return 0;
 481 }
 482
 483 /* Free unused dquots from cache */
 484 static void prune_dqcache(int count)
 485 {
 486         struct list_head *head;
 487         struct dquot *dquot;
 488
 489         head = free_dquots.prev;
 490         while (head != &free_dquots && count) {
 491                 dquot = list_entry(head, struct dquot, dq_free);
 492                 remove_dquot_hash(dquot);
 493                 remove_free_dquot(dquot);
 494                 remove_inuse(dquot);
 495                 kmem_cache_free(dquot_cachep, dquot);
 496                 count--;
 497                 head = free_dquots.prev;
 498         }
 499 }
 500
 501 /*
 502  * This is called from kswapd when we think we need some
 503  * more memory
 504  */
 505
 506 static int shrink_dqcache_memory(int nr, unsigned int gfp_mask)
 507 {
 508         if (nr) {
 509                 spin_lock(&dq_list_lock);
 510                 prune_dqcache(nr);
 511                 spin_unlock(&dq_list_lock);
 512         }
 513         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
 514 }
 515
 516 /*
 517  * Put reference to dquot
 518  * NOTE: If you change this function please check whether dqput_blocks() works right...
 519  * MUST be called with either dqptr_sem or dqonoff_sem held
 520  */
 521 static void dqput(struct dquot *dquot)
 522 {
 523         if (!dquot)
 524                 return;
 525 #ifdef __DQUOT_PARANOIA
 526         if (!atomic_read(&dquot->dq_count)) {
 527                 printk("VFS: dqput: trying to free free dquot\n");
 528                 printk("VFS: device %s, dquot of %s %d\n",
 529                         dquot->dq_sb->s_id,
 530                         quotatypes[dquot->dq_type],
 531                         dquot->dq_id);
 532                 BUG();
 533         }
 534 #endif
 535
 536         spin_lock(&dq_list_lock);
 537         dqstats.drops++;
 538         spin_unlock(&dq_list_lock);
 539 we_slept:
 540         spin_lock(&dq_list_lock);
 541         if (atomic_read(&dquot->dq_count) > 1) {
 542                 /* We have more than one user... nothing to do */
 543                 atomic_dec(&dquot->dq_count);
 544                 spin_unlock(&dq_list_lock);
 545                 return;
 546         }
 547         /* Need to release dquot? */
 548         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 549                 spin_unlock(&dq_list_lock);
 550                 /* Commit dquot before releasing */
 551                 dquot->dq_sb->dq_op->write_dquot(dquot);
 552                 goto we_slept;
 553         }
 554         /* Clear flag in case dquot was inactive (something bad happened) */
 555         clear_dquot_dirty(dquot);
 556         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 557                 spin_unlock(&dq_list_lock);
 558                 dquot->dq_sb->dq_op->release_dquot(dquot);
 559                 goto we_slept;
 560         }
 561         atomic_dec(&dquot->dq_count);
 562 #ifdef __DQUOT_PARANOIA
 563         /* sanity check */
 564         if (!list_empty(&dquot->dq_free))
 565                 BUG();
 566 #endif
 567         put_dquot_last(dquot);
 568         spin_unlock(&dq_list_lock);
 569 }
 570
 571 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 572 {
 573         struct dquot *dquot;
 574
 575         dquot = kmem_cache_alloc(dquot_cachep, SLAB_NOFS);
 576         if(!dquot)
 577                 return NODQUOT;
 578
 579         memset((caddr_t)dquot, 0, sizeof(struct dquot));
 580         sema_init(&dquot->dq_lock, 1);
 581         INIT_LIST_HEAD(&dquot->dq_free);
 582         INIT_LIST_HEAD(&dquot->dq_inuse);
 583         INIT_HLIST_NODE(&dquot->dq_hash);
 584         INIT_LIST_HEAD(&dquot->dq_dirty);
 585         dquot->dq_sb = sb;
 586         dquot->dq_type = type;
 587         atomic_set(&dquot->dq_count, 1);
 588
 589         return dquot;
 590 }
 591
 592 /*
 593  * Get reference to dquot
 594  * MUST be called with either dqptr_sem or dqonoff_sem held
 595  */
 596 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 597 {
 598         unsigned int hashent = hashfn(sb, id, type);
 599         struct dquot *dquot, *empty = NODQUOT;
 600
 601         if (!sb_has_quota_enabled(sb, type))
 602                 return NODQUOT;
 603 we_slept:
 604         spin_lock(&dq_list_lock);
 605         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 606                 if (empty == NODQUOT) {
 607                         spin_unlock(&dq_list_lock);
 608                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 609                                 schedule();     /* Try to wait for a moment... */
 610                         goto we_slept;
 611                 }
 612                 dquot = empty;
 613                 dquot->dq_id = id;
 614                 /* all dquots go on the inuse_list */
 615                 put_inuse(dquot);
 616                 /* hash it first so it can be found */
 617                 insert_dquot_hash(dquot);
 618                 dqstats.lookups++;
 619                 spin_unlock(&dq_list_lock);
 620         } else {
 621                 if (!atomic_read(&dquot->dq_count))
 622                         remove_free_dquot(dquot);
 623                 atomic_inc(&dquot->dq_count);
 624                 dqstats.cache_hits++;
 625                 dqstats.lookups++;
 626                 spin_unlock(&dq_list_lock);
 627                 if (empty)
 628                         kmem_cache_free(dquot_cachep, empty);
 629         }
 630         /* Wait for dq_lock - after this we know that either dquot_release() is already
 631          * finished or it will be canceled due to dq_count > 1 test */
 632         wait_on_dquot(dquot);
 633         /* Read the dquot and instantiate it (everything done only if needed) */
 634         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 635                 dqput(dquot);
 636                 return NODQUOT;
 637         }
 638 #ifdef __DQUOT_PARANOIA
 639         if (!dquot->dq_sb)      /* Has somebody invalidated entry under us? */
 640                 BUG();
 641 #endif
 642
 643         return dquot;
 644 }
 645
 646 static int dqinit_needed(struct inode *inode, int type)
 647 {
 648         int cnt;
 649
 650         if (IS_NOQUOTA(inode))
 651                 return 0;
 652         if (type != -1)
 653                 return inode->i_dquot[type] == NODQUOT;
 654         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 655                 if (inode->i_dquot[cnt] == NODQUOT)
 656                         return 1;
 657         return 0;
 658 }
 659
 660 /* This routine is guarded by dqonoff_sem semaphore */
 661 static void add_dquot_ref(struct super_block *sb, int type)
 662 {
 663         struct list_head *p;
 664
 665 restart:
 666         file_list_lock();
 667         list_for_each(p, &sb->s_files) {
 668                 struct file *filp = list_entry(p, struct file, f_list);
 669                 struct inode *inode = filp->f_dentry->d_inode;
 670                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
 671                         struct dentry *dentry = dget(filp->f_dentry);
 672                         file_list_unlock();
 673                         sb->dq_op->initialize(inode, type);
 674                         dput(dentry);
 675                         /* As we may have blocked we had better restart... */
 676                         goto restart;
 677                 }
 678         }
 679         file_list_unlock();
 680 }
 681
 682 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 683 static inline int dqput_blocks(struct dquot *dquot)
 684 {
 685         if (atomic_read(&dquot->dq_count) <= 1)
 686                 return 1;
 687         return 0;
 688 }
 689
 690 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 691 /* We can't race with anybody because we hold dqptr_sem for writing... */
 692 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
 693 {
 694         struct dquot *dquot = inode->i_dquot[type];
 695
 696         inode->i_dquot[type] = NODQUOT;
 697         if (dquot != NODQUOT) {
 698                 if (dqput_blocks(dquot)) {
 699 #ifdef __DQUOT_PARANOIA
 700                         if (atomic_read(&dquot->dq_count) != 1)
 701                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 702 #endif
 703                         spin_lock(&dq_list_lock);
 704                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 705                         spin_unlock(&dq_list_lock);
 706                         return 1;
 707                 }
 708                 else
 709                         dqput(dquot);   /* We have guaranteed we won't block */
 710         }
 711         return 0;
 712 }
 713
 714 /* Free list of dquots - called from inode.c */
 715 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 716 static void put_dquot_list(struct list_head *tofree_head)
 717 {
 718         struct list_head *act_head;
 719         struct dquot *dquot;
 720
 721         act_head = tofree_head->next;
 722         /* So now we have dquots on the list... Just free them */
 723         while (act_head != tofree_head) {
 724                 dquot = list_entry(act_head, struct dquot, dq_free);
 725                 act_head = act_head->next;
 726                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 727                 dqput(dquot);
 728         }
 729 }
 730
 731 /* Gather all references from inodes and drop them */
 732 static void drop_dquot_ref(struct super_block *sb, int type)
 733 {
 734         LIST_HEAD(tofree_head);
 735
 736         /* We need to be guarded against prune_icache to reach all the
 737          * inodes - otherwise some can be on the local list of prune_icache */
 738         down(&iprune_sem);
 739         down_write(&sb_dqopt(sb)->dqptr_sem);
 740         remove_dquot_ref(sb, type, &tofree_head);
 741         up_write(&sb_dqopt(sb)->dqptr_sem);
 742         up(&iprune_sem);
 743         put_dquot_list(&tofree_head);
 744 }
 745
 746 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
 747 {
 748         dquot->dq_dqb.dqb_curinodes += number;
 749 }
 750
 751 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 752 {
 753         dquot->dq_dqb.dqb_curspace += number;
 754 }
 755
 756 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
 757 {
 758         if (dquot->dq_dqb.dqb_curinodes > number)
 759                 dquot->dq_dqb.dqb_curinodes -= number;
 760         else
 761                 dquot->dq_dqb.dqb_curinodes = 0;
 762         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
 763                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 764         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 765 }
 766
 767 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 768 {
 769         if (dquot->dq_dqb.dqb_curspace > number)
 770                 dquot->dq_dqb.dqb_curspace -= number;
 771         else
 772                 dquot->dq_dqb.dqb_curspace = 0;
 773         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
 774                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 775         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 776 }
 777
 778 static int flag_print_warnings = 1;
 779
 780 static inline int need_print_warning(struct dquot *dquot)
 781 {
 782         if (!flag_print_warnings)
 783                 return 0;
 784
 785         switch (dquot->dq_type) {
 786                 case USRQUOTA:
 787                         return current->fsuid == dquot->dq_id;
 788                 case GRPQUOTA:
 789                         return in_group_p(dquot->dq_id);
 790         }
 791         return 0;
 792 }
 793
 794 /* Values of warnings */
 795 #define NOWARN 0
 796 #define IHARDWARN 1
 797 #define ISOFTLONGWARN 2
 798 #define ISOFTWARN 3
 799 #define BHARDWARN 4
 800 #define BSOFTLONGWARN 5
 801 #define BSOFTWARN 6
 802
 803 /* Print warning to user which exceeded quota */
 804 static void print_warning(struct dquot *dquot, const char warntype)
 805 {
 806         char *msg = NULL;
 807         int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
 808           ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
 809
 810         if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
 811                 return;
 812
 813         tty_write_message(current->signal->tty, dquot->dq_sb->s_id);
 814         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
 815                 tty_write_message(current->signal->tty, ": warning, ");
 816         else
 817                 tty_write_message(current->signal->tty, ": write failed, ");
 818         tty_write_message(current->signal->tty, quotatypes[dquot->dq_type]);
 819         switch (warntype) {
 820                 case IHARDWARN:
 821                         msg = " file limit reached.\r\n";
 822                         break;
 823                 case ISOFTLONGWARN:
 824                         msg = " file quota exceeded too long.\r\n";
 825                         break;
 826                 case ISOFTWARN:
 827                         msg = " file quota exceeded.\r\n";
 828                         break;
 829                 case BHARDWARN:
 830                         msg = " block limit reached.\r\n";
 831                         break;
 832                 case BSOFTLONGWARN:
 833                         msg = " block quota exceeded too long.\r\n";
 834                         break;
 835                 case BSOFTWARN:
 836                         msg = " block quota exceeded.\r\n";
 837                         break;
 838         }
 839         tty_write_message(current->signal->tty, msg);
 840 }
 841
 842 static inline void flush_warnings(struct dquot **dquots, char *warntype)
 843 {
 844         int i;
 845
 846         for (i = 0; i < MAXQUOTAS; i++)
 847                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
 848                         print_warning(dquots[i], warntype[i]);
 849 }
 850
 851 static inline char ignore_hardlimit(struct dquot *dquot)
 852 {
 853         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
 854
 855         return capable(CAP_SYS_RESOURCE) &&
 856             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
 857 }
 858
 859 /* needs dq_data_lock */
 860 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
 861 {
 862         *warntype = NOWARN;
 863         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 864                 return QUOTA_OK;
 865
 866         if (dquot->dq_dqb.dqb_ihardlimit &&
 867            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
 868             !ignore_hardlimit(dquot)) {
 869                 *warntype = IHARDWARN;
 870                 return NO_QUOTA;
 871         }
 872
 873         if (dquot->dq_dqb.dqb_isoftlimit &&
 874            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 875             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
 876             !ignore_hardlimit(dquot)) {
 877                 *warntype = ISOFTLONGWARN;
 878                 return NO_QUOTA;
 879         }
 880
 881         if (dquot->dq_dqb.dqb_isoftlimit &&
 882            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 883             dquot->dq_dqb.dqb_itime == 0) {
 884                 *warntype = ISOFTWARN;
 885                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
 886         }
 887
 888         return QUOTA_OK;
 889 }
 890
 891 /* needs dq_data_lock */
 892 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
 893 {
 894         *warntype = 0;
 895         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 896                 return QUOTA_OK;
 897
 898         if (dquot->dq_dqb.dqb_bhardlimit &&
 899            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
 900             !ignore_hardlimit(dquot)) {
 901                 if (!prealloc)
 902                         *warntype = BHARDWARN;
 903                 return NO_QUOTA;
 904         }
 905
 906         if (dquot->dq_dqb.dqb_bsoftlimit &&
 907            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 908             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
 909             !ignore_hardlimit(dquot)) {
 910                 if (!prealloc)
 911                         *warntype = BSOFTLONGWARN;
 912                 return NO_QUOTA;
 913         }
 914
 915         if (dquot->dq_dqb.dqb_bsoftlimit &&
 916            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 917             dquot->dq_dqb.dqb_btime == 0) {
 918                 if (!prealloc) {
 919                         *warntype = BSOFTWARN;
 920                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
 921                 }
 922                 else
 923                         /*
 924                          * We don't allow preallocation to exceed softlimit so exceeding will
 925                          * be always printed
 926                          */
 927                         return NO_QUOTA;
 928         }
 929
 930         return QUOTA_OK;
 931 }
 932
 933 /*
 934  *      Initialize quota pointers in inode
 935  *      Transaction must be started at entry
 936  */
 937 int dquot_initialize(struct inode *inode, int type)
 938 {
 939         unsigned int id = 0;
 940         int cnt, ret = 0;
 941
 942         /* First test before acquiring semaphore - solves deadlocks when we
 943          * re-enter the quota code and are already holding the semaphore */
 944         if (IS_NOQUOTA(inode))
 945                 return 0;
 946         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 947         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
 948         if (IS_NOQUOTA(inode))
 949                 goto out_err;
 950         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 951                 if (type != -1 && cnt != type)
 952                         continue;
 953                 if (inode->i_dquot[cnt] == NODQUOT) {
 954                         switch (cnt) {
 955                                 case USRQUOTA:
 956                                         id = inode->i_uid;
 957                                         break;
 958                                 case GRPQUOTA:
 959                                         id = inode->i_gid;
 960                                         break;
 961                         }
 962                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
 963                 }
 964         }
 965 out_err:
 966         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 967         return ret;
 968 }
 969
 970 /*
 971  *      Release all quotas referenced by inode
 972  *      Transaction must be started at an entry
 973  */
 974 int dquot_drop(struct inode *inode)
 975 {
 976         int cnt;
 977
 978         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 979         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 980                 if (inode->i_dquot[cnt] != NODQUOT) {
 981                         dqput(inode->i_dquot[cnt]);
 982                         inode->i_dquot[cnt] = NODQUOT;
 983                 }
 984         }
 985         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 986         return 0;
 987 }
 988
 989 /*
 990  * Following four functions update i_blocks+i_bytes fields and
 991  * quota information (together with appropriate checks)
 992  * NOTE: We absolutely rely on the fact that caller dirties
 993  * the inode (usually macros in quotaops.h care about this) and
 994  * holds a handle for the current transaction so that dquot write and
 995  * inode write go into the same transaction.
 996  */
 997
 998 /*
 999  * This operation can block, but only after everything is updated
1000  */
1001 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1002 {
1003         int cnt, ret = NO_QUOTA;
1004         char warntype[MAXQUOTAS];
1005
1006         /* First test before acquiring semaphore - solves deadlocks when we
1007          * re-enter the quota code and are already holding the semaphore */
1008         if (IS_NOQUOTA(inode)) {
1009 out_add:
1010                 inode_add_bytes(inode, number);
1011                 return QUOTA_OK;
1012         }
1013         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1014                 warntype[cnt] = NOWARN;
1015
1016         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1017         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1018                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1019                 goto out_add;
1020         }
1021         spin_lock(&dq_data_lock);
1022         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1023                 if (inode->i_dquot[cnt] == NODQUOT)
1024                         continue;
1025                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1026                         goto warn_put_all;
1027         }
1028         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1029                 if (inode->i_dquot[cnt] == NODQUOT)
1030                         continue;
1031                 dquot_incr_space(inode->i_dquot[cnt], number);
1032         }
1033         inode_add_bytes(inode, number);
1034         ret = QUOTA_OK;
1035 warn_put_all:
1036         spin_unlock(&dq_data_lock);
1037         if (ret == QUOTA_OK)
1038                 /* Dirtify all the dquots - this can block when journalling */
1039                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1040                         if (inode->i_dquot[cnt])
1041                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1042         flush_warnings(inode->i_dquot, warntype);
1043         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1044         return ret;
1045 }
1046
1047 /*
1048  * This operation can block, but only after everything is updated
1049  */
1050 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1051 {
1052         int cnt, ret = NO_QUOTA;
1053         char warntype[MAXQUOTAS];
1054
1055         /* First test before acquiring semaphore - solves deadlocks when we
1056          * re-enter the quota code and are already holding the semaphore */
1057         if (IS_NOQUOTA(inode))
1058                 return QUOTA_OK;
1059         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1060                 warntype[cnt] = NOWARN;
1061         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1062         if (IS_NOQUOTA(inode)) {
1063                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1064                 return QUOTA_OK;
1065         }
1066         spin_lock(&dq_data_lock);
1067         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1068                 if (inode->i_dquot[cnt] == NODQUOT)
1069                         continue;
1070                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1071                         goto warn_put_all;
1072         }
1073
1074         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1075                 if (inode->i_dquot[cnt] == NODQUOT)
1076                         continue;
1077                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1078         }
1079         ret = QUOTA_OK;
1080 warn_put_all:
1081         spin_unlock(&dq_data_lock);
1082         if (ret == QUOTA_OK)
1083                 /* Dirtify all the dquots - this can block when journalling */
1084                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1085                         if (inode->i_dquot[cnt])
1086                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1087         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1088         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1089         return ret;
1090 }
1091
1092 /*
1093  * This operation can block, but only after everything is updated
1094  */
1095 int dquot_free_space(struct inode *inode, qsize_t number)
1096 {
1097         unsigned int cnt;
1098
1099         /* First test before acquiring semaphore - solves deadlocks when we
1100          * re-enter the quota code and are already holding the semaphore */
1101         if (IS_NOQUOTA(inode)) {
1102 out_sub:
1103                 inode_sub_bytes(inode, number);
1104                 return QUOTA_OK;
1105         }
1106         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1107         /* Now recheck reliably when holding dqptr_sem */
1108         if (IS_NOQUOTA(inode)) {
1109                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1110                 goto out_sub;
1111         }
1112         spin_lock(&dq_data_lock);
1113         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1114                 if (inode->i_dquot[cnt] == NODQUOT)
1115                         continue;
1116                 dquot_decr_space(inode->i_dquot[cnt], number);
1117         }
1118         inode_sub_bytes(inode, number);
1119         spin_unlock(&dq_data_lock);
1120         /* Dirtify all the dquots - this can block when journalling */
1121         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1122                 if (inode->i_dquot[cnt])
1123                         mark_dquot_dirty(inode->i_dquot[cnt]);
1124         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1125         return QUOTA_OK;
1126 }
1127
1128 /*
1129  * This operation can block, but only after everything is updated
1130  */
1131 int dquot_free_inode(const struct inode *inode, unsigned long number)
1132 {
1133         unsigned int cnt;
1134
1135         /* First test before acquiring semaphore - solves deadlocks when we
1136          * re-enter the quota code and are already holding the semaphore */
1137         if (IS_NOQUOTA(inode))
1138                 return QUOTA_OK;
1139         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1140         /* Now recheck reliably when holding dqptr_sem */
1141         if (IS_NOQUOTA(inode)) {
1142                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1143                 return QUOTA_OK;
1144         }
1145         spin_lock(&dq_data_lock);
1146         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1147                 if (inode->i_dquot[cnt] == NODQUOT)
1148                         continue;
1149                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1150         }
1151         spin_unlock(&dq_data_lock);
1152         /* Dirtify all the dquots - this can block when journalling */
1153         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1154                 if (inode->i_dquot[cnt])
1155                         mark_dquot_dirty(inode->i_dquot[cnt]);
1156         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1157         return QUOTA_OK;
1158 }
1159
1160 /*
1161  * Transfer the number of inode and blocks from one diskquota to an other.
1162  *
1163  * This operation can block, but only after everything is updated
1164  * A transaction must be started when entering this function.
1165  */
1166 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1167 {
1168         qsize_t space;
1169         struct dquot *transfer_from[MAXQUOTAS];
1170         struct dquot *transfer_to[MAXQUOTAS];
1171         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1172             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1173         char warntype[MAXQUOTAS];
1174
1175         /* First test before acquiring semaphore - solves deadlocks when we
1176          * re-enter the quota code and are already holding the semaphore */
1177         if (IS_NOQUOTA(inode))
1178                 return QUOTA_OK;
1179         /* Clear the arrays */
1180         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1181                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1182                 warntype[cnt] = NOWARN;
1183         }
1184         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1185         /* Now recheck reliably when holding dqptr_sem */
1186         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1187                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1188                 return QUOTA_OK;
1189         }
1190         /* First build the transfer_to list - here we can block on
1191          * reading/instantiating of dquots.  We know that the transaction for
1192          * us was already started so we don't violate lock ranking here */
1193         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1194                 switch (cnt) {
1195                         case USRQUOTA:
1196                                 if (!chuid)
1197                                         continue;
1198                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1199                                 break;
1200                         case GRPQUOTA:
1201                                 if (!chgid)
1202                                         continue;
1203                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1204                                 break;
1205                 }
1206         }
1207         spin_lock(&dq_data_lock);
1208         space = inode_get_bytes(inode);
1209         /* Build the transfer_from list and check the limits */
1210         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1211                 if (transfer_to[cnt] == NODQUOT)
1212                         continue;
1213                 transfer_from[cnt] = inode->i_dquot[cnt];
1214                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1215                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1216                         goto warn_put_all;
1217         }
1218
1219         /*
1220          * Finally perform the needed transfer from transfer_from to transfer_to
1221          */
1222         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1223                 /*
1224                  * Skip changes for same uid or gid or for turned off quota-type.
1225                  */
1226                 if (transfer_to[cnt] == NODQUOT)
1227                         continue;
1228
1229                 /* Due to IO error we might not have transfer_from[] structure */
1230                 if (transfer_from[cnt]) {
1231                         dquot_decr_inodes(transfer_from[cnt], 1);
1232                         dquot_decr_space(transfer_from[cnt], space);
1233                 }
1234
1235                 dquot_incr_inodes(transfer_to[cnt], 1);
1236                 dquot_incr_space(transfer_to[cnt], space);
1237
1238                 inode->i_dquot[cnt] = transfer_to[cnt];
1239         }
1240         ret = QUOTA_OK;
1241 warn_put_all:
1242         spin_unlock(&dq_data_lock);
1243         /* Dirtify all the dquots - this can block when journalling */
1244         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1245                 if (transfer_from[cnt])
1246                         mark_dquot_dirty(transfer_from[cnt]);
1247                 if (transfer_to[cnt])
1248                         mark_dquot_dirty(transfer_to[cnt]);
1249         }
1250         flush_warnings(transfer_to, warntype);
1251
1252         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1253                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1254                         dqput(transfer_from[cnt]);
1255                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1256                         dqput(transfer_to[cnt]);
1257         }
1258         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1259         return ret;
1260 }
1261
1262 /*
1263  * Write info of quota file to disk
1264  */
1265 int dquot_commit_info(struct super_block *sb, int type)
1266 {
1267         int ret;
1268         struct quota_info *dqopt = sb_dqopt(sb);
1269
1270         down(&dqopt->dqio_sem);
1271         ret = dqopt->ops[type]->write_file_info(sb, type);
1272         up(&dqopt->dqio_sem);
1273         return ret;
1274 }
1275
1276 /*
1277  * Definitions of diskquota operations.
1278  */
1279 struct dquot_operations dquot_operations = {
1280         .initialize     = dquot_initialize,
1281         .drop           = dquot_drop,
1282         .alloc_space    = dquot_alloc_space,
1283         .alloc_inode    = dquot_alloc_inode,
1284         .free_space     = dquot_free_space,
1285         .free_inode     = dquot_free_inode,
1286         .transfer       = dquot_transfer,
1287         .write_dquot    = dquot_commit,
1288         .acquire_dquot  = dquot_acquire,
1289         .release_dquot  = dquot_release,
1290         .mark_dirty     = dquot_mark_dquot_dirty,
1291         .write_info     = dquot_commit_info
1292 };
1293
1294 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1295 {
1296         switch (type) {
1297                 case USRQUOTA:
1298                         dqopt->flags |= DQUOT_USR_ENABLED;
1299                         break;
1300                 case GRPQUOTA:
1301                         dqopt->flags |= DQUOT_GRP_ENABLED;
1302                         break;
1303         }
1304 }
1305
1306 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1307 {
1308         switch (type) {
1309                 case USRQUOTA:
1310                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1311                         break;
1312                 case GRPQUOTA:
1313                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1314                         break;
1315         }
1316 }
1317
1318 /*
1319  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1320  */
1321 int vfs_quota_off(struct super_block *sb, int type)
1322 {
1323         int cnt;
1324         struct quota_info *dqopt = sb_dqopt(sb);
1325         struct inode *toputinode[MAXQUOTAS];
1326         struct vfsmount *toputmnt[MAXQUOTAS];
1327
1328         /* We need to serialize quota_off() for device */
1329         down(&dqopt->dqonoff_sem);
1330         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1331                 toputinode[cnt] = NULL;
1332                 toputmnt[cnt] = NULL;
1333                 if (type != -1 && cnt != type)
1334                         continue;
1335                 if (!sb_has_quota_enabled(sb, cnt))
1336                         continue;
1337                 reset_enable_flags(dqopt, cnt);
1338
1339                 /* Note: these are blocking operations */
1340                 drop_dquot_ref(sb, cnt);
1341                 invalidate_dquots(sb, cnt);
1342                 /*
1343                  * Now all dquots should be invalidated, all writes done so we should be only
1344                  * users of the info. No locks needed.
1345                  */
1346                 if (info_dirty(&dqopt->info[cnt]))
1347                         sb->dq_op->write_info(sb, cnt);
1348                 if (dqopt->ops[cnt]->free_file_info)
1349                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1350                 put_quota_format(dqopt->info[cnt].dqi_format);
1351
1352                 toputinode[cnt] = dqopt->files[cnt];
1353                 toputmnt[cnt] = dqopt->mnt[cnt];
1354                 dqopt->files[cnt] = NULL;
1355                 dqopt->mnt[cnt] = NULL;
1356                 dqopt->info[cnt].dqi_flags = 0;
1357                 dqopt->info[cnt].dqi_igrace = 0;
1358                 dqopt->info[cnt].dqi_bgrace = 0;
1359                 dqopt->ops[cnt] = NULL;
1360         }
1361         up(&dqopt->dqonoff_sem);
1362         /* Sync the superblock so that buffers with quota data are written to
1363          * disk (and so userspace sees correct data afterwards).
1364          * The reference to vfsmnt we are still holding protects us from
1365          * umount (we don't have it only when quotas are turned on/off for
1366          * journal replay but in that case we are guarded by the fs anyway). */
1367         if (sb->s_op->sync_fs)
1368                 sb->s_op->sync_fs(sb, 1);
1369         sync_blockdev(sb->s_bdev);
1370         /* Now the quota files are just ordinary files and we can set the
1371          * inode flags back. Moreover we discard the pagecache so that
1372          * userspace sees the writes we did bypassing the pagecache. We
1373          * must also discard the blockdev buffers so that we see the
1374          * changes done by userspace on the next quotaon() */
1375         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1376                 if (toputinode[cnt]) {
1377                         down(&dqopt->dqonoff_sem);
1378                         /* If quota was reenabled in the meantime, we have
1379                          * nothing to do */
1380                         if (!sb_has_quota_enabled(sb, cnt)) {
1381                                 down(&toputinode[cnt]->i_sem);
1382                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1383                                   S_NOATIME | S_NOQUOTA);
1384                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1385                                 up(&toputinode[cnt]->i_sem);
1386                                 mark_inode_dirty(toputinode[cnt]);
1387                                 iput(toputinode[cnt]);
1388                         }
1389                         up(&dqopt->dqonoff_sem);
1390                         /* We don't hold the reference when we turned on quotas
1391                          * just for the journal replay... */
1392                         if (toputmnt[cnt])
1393                                 mntput(toputmnt[cnt]);
1394                 }
1395         if (sb->s_bdev)
1396                 invalidate_bdev(sb->s_bdev, 0);
1397         return 0;
1398 }
1399
1400 /*
1401  *      Turn quotas on on a device
1402  */
1403
1404 /* Helper function when we already have the inode */
1405 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1406 {
1407         struct quota_format_type *fmt = find_quota_format(format_id);
1408         struct super_block *sb = inode->i_sb;
1409         struct quota_info *dqopt = sb_dqopt(sb);
1410         int error;
1411         int oldflags = -1;
1412
1413         if (!fmt)
1414                 return -ESRCH;
1415         if (!S_ISREG(inode->i_mode)) {
1416                 error = -EACCES;
1417                 goto out_fmt;
1418         }
1419         if (IS_RDONLY(inode)) {
1420                 error = -EROFS;
1421                 goto out_fmt;
1422         }
1423         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1424                 error = -EINVAL;
1425                 goto out_fmt;
1426         }
1427
1428         /* As we bypass the pagecache we must now flush the inode so that
1429          * we see all the changes from userspace... */
1430         write_inode_now(inode, 1);
1431         /* And now flush the block cache so that kernel sees the changes */
1432         invalidate_bdev(sb->s_bdev, 0);
1433         down(&inode->i_sem);
1434         down(&dqopt->dqonoff_sem);
1435         if (sb_has_quota_enabled(sb, type)) {
1436                 error = -EBUSY;
1437                 goto out_lock;
1438         }
1439         /* We don't want quota and atime on quota files (deadlocks possible)
1440          * Also nobody should write to the file - we use special IO operations
1441          * which ignore the immutable bit. */
1442         down_write(&dqopt->dqptr_sem);
1443         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1444         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1445         up_write(&dqopt->dqptr_sem);
1446         sb->dq_op->drop(inode);
1447
1448         error = -EIO;
1449         dqopt->files[type] = igrab(inode);
1450         if (!dqopt->files[type])
1451                 goto out_lock;
1452         error = -EINVAL;
1453         if (!fmt->qf_ops->check_quota_file(sb, type))
1454                 goto out_file_init;
1455
1456         dqopt->ops[type] = fmt->qf_ops;
1457         dqopt->info[type].dqi_format = fmt;
1458         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1459         down(&dqopt->dqio_sem);
1460         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1461                 up(&dqopt->dqio_sem);
1462                 goto out_file_init;
1463         }
1464         up(&dqopt->dqio_sem);
1465         up(&inode->i_sem);
1466         set_enable_flags(dqopt, type);
1467
1468         add_dquot_ref(sb, type);
1469         up(&dqopt->dqonoff_sem);
1470
1471         return 0;
1472
1473 out_file_init:
1474         dqopt->files[type] = NULL;
1475         iput(inode);
1476 out_lock:
1477         up(&dqopt->dqonoff_sem);
1478         if (oldflags != -1) {
1479                 down_write(&dqopt->dqptr_sem);
1480                 /* Set the flags back (in the case of accidental quotaon()
1481                  * on a wrong file we don't want to mess up the flags) */
1482                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1483                 inode->i_flags |= oldflags;
1484                 up_write(&dqopt->dqptr_sem);
1485         }
1486         up(&inode->i_sem);
1487 out_fmt:
1488         put_quota_format(fmt);
1489
1490         return error;
1491 }
1492
1493 /* Actual function called from quotactl() */
1494 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1495 {
1496         struct nameidata nd;
1497         int error;
1498
1499         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1500         if (error < 0)
1501                 return error;
1502         error = security_quota_on(nd.dentry);
1503         if (error)
1504                 goto out_path;
1505         /* Quota file not on the same filesystem? */
1506         if (nd.mnt->mnt_sb != sb)
1507                 error = -EXDEV;
1508         else {
1509                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1510                 if (!error)
1511                         sb_dqopt(sb)->mnt[type] = mntget(nd.mnt);
1512         }
1513 out_path:
1514         path_release(&nd);
1515         return error;
1516 }
1517
1518 /*
1519  * This function is used when filesystem needs to initialize quotas
1520  * during mount time.
1521  */
1522 int vfs_quota_on_mount(int type, int format_id, struct dentry *dentry)
1523 {
1524         int error;
1525
1526         error = security_quota_on(dentry);
1527         if (error)
1528                 return error;
1529         return vfs_quota_on_inode(dentry->d_inode, type, format_id);
1530 }
1531
1532 /* Generic routine for getting common part of quota structure */
1533 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1534 {
1535         struct mem_dqblk *dm = &dquot->dq_dqb;
1536
1537         spin_lock(&dq_data_lock);
1538         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1539         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1540         di->dqb_curspace = dm->dqb_curspace;
1541         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1542         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1543         di->dqb_curinodes = dm->dqb_curinodes;
1544         di->dqb_btime = dm->dqb_btime;
1545         di->dqb_itime = dm->dqb_itime;
1546         di->dqb_valid = QIF_ALL;
1547         spin_unlock(&dq_data_lock);
1548 }
1549
1550 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1551 {
1552         struct dquot *dquot;
1553
1554         down(&sb_dqopt(sb)->dqonoff_sem);
1555         if (!(dquot = dqget(sb, id, type))) {
1556                 up(&sb_dqopt(sb)->dqonoff_sem);
1557                 return -ESRCH;
1558         }
1559         do_get_dqblk(dquot, di);
1560         dqput(dquot);
1561         up(&sb_dqopt(sb)->dqonoff_sem);
1562         return 0;
1563 }
1564
1565 /* Generic routine for setting common part of quota structure */
1566 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1567 {
1568         struct mem_dqblk *dm = &dquot->dq_dqb;
1569         int check_blim = 0, check_ilim = 0;
1570
1571         spin_lock(&dq_data_lock);
1572         if (di->dqb_valid & QIF_SPACE) {
1573                 dm->dqb_curspace = di->dqb_curspace;
1574                 check_blim = 1;
1575         }
1576         if (di->dqb_valid & QIF_BLIMITS) {
1577                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1578                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1579                 check_blim = 1;
1580         }
1581         if (di->dqb_valid & QIF_INODES) {
1582                 dm->dqb_curinodes = di->dqb_curinodes;
1583                 check_ilim = 1;
1584         }
1585         if (di->dqb_valid & QIF_ILIMITS) {
1586                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1587                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1588                 check_ilim = 1;
1589         }
1590         if (di->dqb_valid & QIF_BTIME)
1591                 dm->dqb_btime = di->dqb_btime;
1592         if (di->dqb_valid & QIF_ITIME)
1593                 dm->dqb_itime = di->dqb_itime;
1594
1595         if (check_blim) {
1596                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1597                         dm->dqb_btime = 0;
1598                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1599                 }
1600                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1601                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1602         }
1603         if (check_ilim) {
1604                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1605                         dm->dqb_itime = 0;
1606                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1607                 }
1608                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1609                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1610         }
1611         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1612                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1613         else
1614                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1615         spin_unlock(&dq_data_lock);
1616         mark_dquot_dirty(dquot);
1617 }
1618
1619 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1620 {
1621         struct dquot *dquot;
1622
1623         down(&sb_dqopt(sb)->dqonoff_sem);
1624         if (!(dquot = dqget(sb, id, type))) {
1625                 up(&sb_dqopt(sb)->dqonoff_sem);
1626                 return -ESRCH;
1627         }
1628         do_set_dqblk(dquot, di);
1629         dqput(dquot);
1630         up(&sb_dqopt(sb)->dqonoff_sem);
1631         return 0;
1632 }
1633
1634 /* Generic routine for getting common part of quota file information */
1635 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1636 {
1637         struct mem_dqinfo *mi;
1638
1639         down(&sb_dqopt(sb)->dqonoff_sem);
1640         if (!sb_has_quota_enabled(sb, type)) {
1641                 up(&sb_dqopt(sb)->dqonoff_sem);
1642                 return -ESRCH;
1643         }
1644         mi = sb_dqopt(sb)->info + type;
1645         spin_lock(&dq_data_lock);
1646         ii->dqi_bgrace = mi->dqi_bgrace;
1647         ii->dqi_igrace = mi->dqi_igrace;
1648         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1649         ii->dqi_valid = IIF_ALL;
1650         spin_unlock(&dq_data_lock);
1651         up(&sb_dqopt(sb)->dqonoff_sem);
1652         return 0;
1653 }
1654
1655 /* Generic routine for setting common part of quota file information */
1656 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1657 {
1658         struct mem_dqinfo *mi;
1659
1660         down(&sb_dqopt(sb)->dqonoff_sem);
1661         if (!sb_has_quota_enabled(sb, type)) {
1662                 up(&sb_dqopt(sb)->dqonoff_sem);
1663                 return -ESRCH;
1664         }
1665         mi = sb_dqopt(sb)->info + type;
1666         spin_lock(&dq_data_lock);
1667         if (ii->dqi_valid & IIF_BGRACE)
1668                 mi->dqi_bgrace = ii->dqi_bgrace;
1669         if (ii->dqi_valid & IIF_IGRACE)
1670                 mi->dqi_igrace = ii->dqi_igrace;
1671         if (ii->dqi_valid & IIF_FLAGS)
1672                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1673         spin_unlock(&dq_data_lock);
1674         mark_info_dirty(sb, type);
1675         /* Force write to disk */
1676         sb->dq_op->write_info(sb, type);
1677         up(&sb_dqopt(sb)->dqonoff_sem);
1678         return 0;
1679 }
1680
1681 struct quotactl_ops vfs_quotactl_ops = {
1682         .quota_on       = vfs_quota_on,
1683         .quota_off      = vfs_quota_off,
1684         .quota_sync     = vfs_quota_sync,
1685         .get_info       = vfs_get_dqinfo,
1686         .set_info       = vfs_set_dqinfo,
1687         .get_dqblk      = vfs_get_dqblk,
1688         .set_dqblk      = vfs_set_dqblk
1689 };
1690
1691 static ctl_table fs_dqstats_table[] = {
1692         {
1693                 .ctl_name       = FS_DQ_LOOKUPS,
1694                 .procname       = "lookups",
1695                 .data           = &dqstats.lookups,
1696                 .maxlen         = sizeof(int),
1697                 .mode           = 0444,
1698                 .proc_handler   = &proc_dointvec,
1699         },
1700         {
1701                 .ctl_name       = FS_DQ_DROPS,
1702                 .procname       = "drops",
1703                 .data           = &dqstats.drops,
1704                 .maxlen         = sizeof(int),
1705                 .mode           = 0444,
1706                 .proc_handler   = &proc_dointvec,
1707         },
1708         {
1709                 .ctl_name       = FS_DQ_READS,
1710                 .procname       = "reads",
1711                 .data           = &dqstats.reads,
1712                 .maxlen         = sizeof(int),
1713                 .mode           = 0444,
1714                 .proc_handler   = &proc_dointvec,
1715         },
1716         {
1717                 .ctl_name       = FS_DQ_WRITES,
1718                 .procname       = "writes",
1719                 .data           = &dqstats.writes,
1720                 .maxlen         = sizeof(int),
1721                 .mode           = 0444,
1722                 .proc_handler   = &proc_dointvec,
1723         },
1724         {
1725                 .ctl_name       = FS_DQ_CACHE_HITS,
1726                 .procname       = "cache_hits",
1727                 .data           = &dqstats.cache_hits,
1728                 .maxlen         = sizeof(int),
1729                 .mode           = 0444,
1730                 .proc_handler   = &proc_dointvec,
1731         },
1732         {
1733                 .ctl_name       = FS_DQ_ALLOCATED,
1734                 .procname       = "allocated_dquots",
1735                 .data           = &dqstats.allocated_dquots,
1736                 .maxlen         = sizeof(int),
1737                 .mode           = 0444,
1738                 .proc_handler   = &proc_dointvec,
1739         },
1740         {
1741                 .ctl_name       = FS_DQ_FREE,
1742                 .procname       = "free_dquots",
1743                 .data           = &dqstats.free_dquots,
1744                 .maxlen         = sizeof(int),
1745                 .mode           = 0444,
1746                 .proc_handler   = &proc_dointvec,
1747         },
1748         {
1749                 .ctl_name       = FS_DQ_SYNCS,
1750                 .procname       = "syncs",
1751                 .data           = &dqstats.syncs,
1752                 .maxlen         = sizeof(int),
1753                 .mode           = 0444,
1754                 .proc_handler   = &proc_dointvec,
1755         },
1756         {
1757                 .ctl_name       = FS_DQ_WARNINGS,
1758                 .procname       = "warnings",
1759                 .data           = &flag_print_warnings,
1760                 .maxlen         = sizeof(int),
1761                 .mode           = 0644,
1762                 .proc_handler   = &proc_dointvec,
1763         },
1764         { .ctl_name = 0 },
1765 };
1766
1767 static ctl_table fs_table[] = {
1768         {
1769                 .ctl_name       = FS_DQSTATS,
1770                 .procname       = "quota",
1771                 .mode           = 0555,
1772                 .child          = fs_dqstats_table,
1773         },
1774         { .ctl_name = 0 },
1775 };
1776
1777 static ctl_table sys_table[] = {
1778         {
1779                 .ctl_name       = CTL_FS,
1780                 .procname       = "fs",
1781                 .mode           = 0555,
1782                 .child          = fs_table,
1783         },
1784         { .ctl_name = 0 },
1785 };
1786
1787 static int __init dquot_init(void)
1788 {
1789         int i;
1790         unsigned long nr_hash, order;
1791
1792         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1793
1794         register_sysctl_table(sys_table, 0);
1795
1796         dquot_cachep = kmem_cache_create("dquot",
1797                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1798                         SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1799                         NULL, NULL);
1800
1801         order = 0;
1802         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1803         if (!dquot_hash)
1804                 panic("Cannot create dquot hash table");
1805
1806         /* Find power-of-two hlist_heads which can fit into allocation */
1807         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1808         dq_hash_bits = 0;
1809         do {
1810                 dq_hash_bits++;
1811         } while (nr_hash >> dq_hash_bits);
1812         dq_hash_bits--;
1813
1814         nr_hash = 1UL << dq_hash_bits;
1815         dq_hash_mask = nr_hash - 1;
1816         for (i = 0; i < nr_hash; i++)
1817                 INIT_HLIST_HEAD(dquot_hash + i);
1818
1819         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1820                         nr_hash, order, (PAGE_SIZE << order));
1821
1822         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1823
1824         return 0;
1825 }
1826 module_init(dquot_init);
1827
1828 EXPORT_SYMBOL(register_quota_format);
1829 EXPORT_SYMBOL(unregister_quota_format);
1830 EXPORT_SYMBOL(dqstats);
1831 EXPORT_SYMBOL(dq_data_lock);
1832 EXPORT_SYMBOL(vfs_quota_on);
1833 EXPORT_SYMBOL(vfs_quota_on_mount);
1834 EXPORT_SYMBOL(vfs_quota_off);
1835 EXPORT_SYMBOL(vfs_quota_sync);
1836 EXPORT_SYMBOL(vfs_get_dqinfo);
1837 EXPORT_SYMBOL(vfs_set_dqinfo);
1838 EXPORT_SYMBOL(vfs_get_dqblk);
1839 EXPORT_SYMBOL(vfs_set_dqblk);
1840 EXPORT_SYMBOL(dquot_commit);
1841 EXPORT_SYMBOL(dquot_commit_info);
1842 EXPORT_SYMBOL(dquot_acquire);
1843 EXPORT_SYMBOL(dquot_release);
1844 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1845 EXPORT_SYMBOL(dquot_initialize);
1846 EXPORT_SYMBOL(dquot_drop);
1847 EXPORT_SYMBOL(dquot_alloc_space);
1848 EXPORT_SYMBOL(dquot_alloc_inode);
1849 EXPORT_SYMBOL(dquot_free_space);
1850 EXPORT_SYMBOL(dquot_free_inode);
1851 EXPORT_SYMBOL(dquot_transfer);