git.oblomov.eu Git - linux-2.6/blob - 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, *tmp;
 413
 414         spin_lock(&dq_list_lock);
 415         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 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, gfp_t gfp_mask)
 504 {
 505         if (nr) {
 506                 spin_lock(&dq_list_lock);
 507                 prune_dqcache(nr);
 508                 spin_unlock(&dq_list_lock);
 509         }
 510         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
 511 }
 512
 513 /*
 514  * Put reference to dquot
 515  * NOTE: If you change this function please check whether dqput_blocks() works right...
 516  * MUST be called with either dqptr_sem or dqonoff_sem held
 517  */
 518 static void dqput(struct dquot *dquot)
 519 {
 520         if (!dquot)
 521                 return;
 522 #ifdef __DQUOT_PARANOIA
 523         if (!atomic_read(&dquot->dq_count)) {
 524                 printk("VFS: dqput: trying to free free dquot\n");
 525                 printk("VFS: device %s, dquot of %s %d\n",
 526                         dquot->dq_sb->s_id,
 527                         quotatypes[dquot->dq_type],
 528                         dquot->dq_id);
 529                 BUG();
 530         }
 531 #endif
 532
 533         spin_lock(&dq_list_lock);
 534         dqstats.drops++;
 535         spin_unlock(&dq_list_lock);
 536 we_slept:
 537         spin_lock(&dq_list_lock);
 538         if (atomic_read(&dquot->dq_count) > 1) {
 539                 /* We have more than one user... nothing to do */
 540                 atomic_dec(&dquot->dq_count);
 541                 spin_unlock(&dq_list_lock);
 542                 return;
 543         }
 544         /* Need to release dquot? */
 545         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 546                 spin_unlock(&dq_list_lock);
 547                 /* Commit dquot before releasing */
 548                 dquot->dq_sb->dq_op->write_dquot(dquot);
 549                 goto we_slept;
 550         }
 551         /* Clear flag in case dquot was inactive (something bad happened) */
 552         clear_dquot_dirty(dquot);
 553         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 554                 spin_unlock(&dq_list_lock);
 555                 dquot->dq_sb->dq_op->release_dquot(dquot);
 556                 goto we_slept;
 557         }
 558         atomic_dec(&dquot->dq_count);
 559 #ifdef __DQUOT_PARANOIA
 560         /* sanity check */
 561         if (!list_empty(&dquot->dq_free))
 562                 BUG();
 563 #endif
 564         put_dquot_last(dquot);
 565         spin_unlock(&dq_list_lock);
 566 }
 567
 568 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 569 {
 570         struct dquot *dquot;
 571
 572         dquot = kmem_cache_alloc(dquot_cachep, SLAB_NOFS);
 573         if(!dquot)
 574                 return NODQUOT;
 575
 576         memset((caddr_t)dquot, 0, sizeof(struct dquot));
 577         sema_init(&dquot->dq_lock, 1);
 578         INIT_LIST_HEAD(&dquot->dq_free);
 579         INIT_LIST_HEAD(&dquot->dq_inuse);
 580         INIT_HLIST_NODE(&dquot->dq_hash);
 581         INIT_LIST_HEAD(&dquot->dq_dirty);
 582         dquot->dq_sb = sb;
 583         dquot->dq_type = type;
 584         atomic_set(&dquot->dq_count, 1);
 585
 586         return dquot;
 587 }
 588
 589 /*
 590  * Get reference to dquot
 591  * MUST be called with either dqptr_sem or dqonoff_sem held
 592  */
 593 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 594 {
 595         unsigned int hashent = hashfn(sb, id, type);
 596         struct dquot *dquot, *empty = NODQUOT;
 597
 598         if (!sb_has_quota_enabled(sb, type))
 599                 return NODQUOT;
 600 we_slept:
 601         spin_lock(&dq_list_lock);
 602         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 603                 if (empty == NODQUOT) {
 604                         spin_unlock(&dq_list_lock);
 605                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 606                                 schedule();     /* Try to wait for a moment... */
 607                         goto we_slept;
 608                 }
 609                 dquot = empty;
 610                 dquot->dq_id = id;
 611                 /* all dquots go on the inuse_list */
 612                 put_inuse(dquot);
 613                 /* hash it first so it can be found */
 614                 insert_dquot_hash(dquot);
 615                 dqstats.lookups++;
 616                 spin_unlock(&dq_list_lock);
 617         } else {
 618                 if (!atomic_read(&dquot->dq_count))
 619                         remove_free_dquot(dquot);
 620                 atomic_inc(&dquot->dq_count);
 621                 dqstats.cache_hits++;
 622                 dqstats.lookups++;
 623                 spin_unlock(&dq_list_lock);
 624                 if (empty)
 625                         kmem_cache_free(dquot_cachep, empty);
 626         }
 627         /* Wait for dq_lock - after this we know that either dquot_release() is already
 628          * finished or it will be canceled due to dq_count > 1 test */
 629         wait_on_dquot(dquot);
 630         /* Read the dquot and instantiate it (everything done only if needed) */
 631         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 632                 dqput(dquot);
 633                 return NODQUOT;
 634         }
 635 #ifdef __DQUOT_PARANOIA
 636         if (!dquot->dq_sb)      /* Has somebody invalidated entry under us? */
 637                 BUG();
 638 #endif
 639
 640         return dquot;
 641 }
 642
 643 static int dqinit_needed(struct inode *inode, int type)
 644 {
 645         int cnt;
 646
 647         if (IS_NOQUOTA(inode))
 648                 return 0;
 649         if (type != -1)
 650                 return inode->i_dquot[type] == NODQUOT;
 651         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 652                 if (inode->i_dquot[cnt] == NODQUOT)
 653                         return 1;
 654         return 0;
 655 }
 656
 657 /* This routine is guarded by dqonoff_sem semaphore */
 658 static void add_dquot_ref(struct super_block *sb, int type)
 659 {
 660         struct list_head *p;
 661
 662 restart:
 663         file_list_lock();
 664         list_for_each(p, &sb->s_files) {
 665                 struct file *filp = list_entry(p, struct file, f_u.fu_list);
 666                 struct inode *inode = filp->f_dentry->d_inode;
 667                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
 668                         struct dentry *dentry = dget(filp->f_dentry);
 669                         file_list_unlock();
 670                         sb->dq_op->initialize(inode, type);
 671                         dput(dentry);
 672                         /* As we may have blocked we had better restart... */
 673                         goto restart;
 674                 }
 675         }
 676         file_list_unlock();
 677 }
 678
 679 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 680 static inline int dqput_blocks(struct dquot *dquot)
 681 {
 682         if (atomic_read(&dquot->dq_count) <= 1)
 683                 return 1;
 684         return 0;
 685 }
 686
 687 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 688 /* We can't race with anybody because we hold dqptr_sem for writing... */
 689 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
 690 {
 691         struct dquot *dquot = inode->i_dquot[type];
 692
 693         inode->i_dquot[type] = NODQUOT;
 694         if (dquot != NODQUOT) {
 695                 if (dqput_blocks(dquot)) {
 696 #ifdef __DQUOT_PARANOIA
 697                         if (atomic_read(&dquot->dq_count) != 1)
 698                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 699 #endif
 700                         spin_lock(&dq_list_lock);
 701                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 702                         spin_unlock(&dq_list_lock);
 703                         return 1;
 704                 }
 705                 else
 706                         dqput(dquot);   /* We have guaranteed we won't block */
 707         }
 708         return 0;
 709 }
 710
 711 /* Free list of dquots - called from inode.c */
 712 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 713 static void put_dquot_list(struct list_head *tofree_head)
 714 {
 715         struct list_head *act_head;
 716         struct dquot *dquot;
 717
 718         act_head = tofree_head->next;
 719         /* So now we have dquots on the list... Just free them */
 720         while (act_head != tofree_head) {
 721                 dquot = list_entry(act_head, struct dquot, dq_free);
 722                 act_head = act_head->next;
 723                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 724                 dqput(dquot);
 725         }
 726 }
 727
 728 /* Gather all references from inodes and drop them */
 729 static void drop_dquot_ref(struct super_block *sb, int type)
 730 {
 731         LIST_HEAD(tofree_head);
 732
 733         /* We need to be guarded against prune_icache to reach all the
 734          * inodes - otherwise some can be on the local list of prune_icache */
 735         down(&iprune_sem);
 736         down_write(&sb_dqopt(sb)->dqptr_sem);
 737         remove_dquot_ref(sb, type, &tofree_head);
 738         up_write(&sb_dqopt(sb)->dqptr_sem);
 739         up(&iprune_sem);
 740         put_dquot_list(&tofree_head);
 741 }
 742
 743 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
 744 {
 745         dquot->dq_dqb.dqb_curinodes += number;
 746 }
 747
 748 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 749 {
 750         dquot->dq_dqb.dqb_curspace += number;
 751 }
 752
 753 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
 754 {
 755         if (dquot->dq_dqb.dqb_curinodes > number)
 756                 dquot->dq_dqb.dqb_curinodes -= number;
 757         else
 758                 dquot->dq_dqb.dqb_curinodes = 0;
 759         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
 760                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 761         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 762 }
 763
 764 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 765 {
 766         if (dquot->dq_dqb.dqb_curspace > number)
 767                 dquot->dq_dqb.dqb_curspace -= number;
 768         else
 769                 dquot->dq_dqb.dqb_curspace = 0;
 770         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
 771                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 772         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 773 }
 774
 775 static int flag_print_warnings = 1;
 776
 777 static inline int need_print_warning(struct dquot *dquot)
 778 {
 779         if (!flag_print_warnings)
 780                 return 0;
 781
 782         switch (dquot->dq_type) {
 783                 case USRQUOTA:
 784                         return current->fsuid == dquot->dq_id;
 785                 case GRPQUOTA:
 786                         return in_group_p(dquot->dq_id);
 787         }
 788         return 0;
 789 }
 790
 791 /* Values of warnings */
 792 #define NOWARN 0
 793 #define IHARDWARN 1
 794 #define ISOFTLONGWARN 2
 795 #define ISOFTWARN 3
 796 #define BHARDWARN 4
 797 #define BSOFTLONGWARN 5
 798 #define BSOFTWARN 6
 799
 800 /* Print warning to user which exceeded quota */
 801 static void print_warning(struct dquot *dquot, const char warntype)
 802 {
 803         char *msg = NULL;
 804         int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
 805           ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
 806
 807         if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
 808                 return;
 809
 810         tty_write_message(current->signal->tty, dquot->dq_sb->s_id);
 811         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
 812                 tty_write_message(current->signal->tty, ": warning, ");
 813         else
 814                 tty_write_message(current->signal->tty, ": write failed, ");
 815         tty_write_message(current->signal->tty, quotatypes[dquot->dq_type]);
 816         switch (warntype) {
 817                 case IHARDWARN:
 818                         msg = " file limit reached.\r\n";
 819                         break;
 820                 case ISOFTLONGWARN:
 821                         msg = " file quota exceeded too long.\r\n";
 822                         break;
 823                 case ISOFTWARN:
 824                         msg = " file quota exceeded.\r\n";
 825                         break;
 826                 case BHARDWARN:
 827                         msg = " block limit reached.\r\n";
 828                         break;
 829                 case BSOFTLONGWARN:
 830                         msg = " block quota exceeded too long.\r\n";
 831                         break;
 832                 case BSOFTWARN:
 833                         msg = " block quota exceeded.\r\n";
 834                         break;
 835         }
 836         tty_write_message(current->signal->tty, msg);
 837 }
 838
 839 static inline void flush_warnings(struct dquot **dquots, char *warntype)
 840 {
 841         int i;
 842
 843         for (i = 0; i < MAXQUOTAS; i++)
 844                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
 845                         print_warning(dquots[i], warntype[i]);
 846 }
 847
 848 static inline char ignore_hardlimit(struct dquot *dquot)
 849 {
 850         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
 851
 852         return capable(CAP_SYS_RESOURCE) &&
 853             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
 854 }
 855
 856 /* needs dq_data_lock */
 857 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
 858 {
 859         *warntype = NOWARN;
 860         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 861                 return QUOTA_OK;
 862
 863         if (dquot->dq_dqb.dqb_ihardlimit &&
 864            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
 865             !ignore_hardlimit(dquot)) {
 866                 *warntype = IHARDWARN;
 867                 return NO_QUOTA;
 868         }
 869
 870         if (dquot->dq_dqb.dqb_isoftlimit &&
 871            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 872             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
 873             !ignore_hardlimit(dquot)) {
 874                 *warntype = ISOFTLONGWARN;
 875                 return NO_QUOTA;
 876         }
 877
 878         if (dquot->dq_dqb.dqb_isoftlimit &&
 879            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 880             dquot->dq_dqb.dqb_itime == 0) {
 881                 *warntype = ISOFTWARN;
 882                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
 883         }
 884
 885         return QUOTA_OK;
 886 }
 887
 888 /* needs dq_data_lock */
 889 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
 890 {
 891         *warntype = 0;
 892         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 893                 return QUOTA_OK;
 894
 895         if (dquot->dq_dqb.dqb_bhardlimit &&
 896            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
 897             !ignore_hardlimit(dquot)) {
 898                 if (!prealloc)
 899                         *warntype = BHARDWARN;
 900                 return NO_QUOTA;
 901         }
 902
 903         if (dquot->dq_dqb.dqb_bsoftlimit &&
 904            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 905             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
 906             !ignore_hardlimit(dquot)) {
 907                 if (!prealloc)
 908                         *warntype = BSOFTLONGWARN;
 909                 return NO_QUOTA;
 910         }
 911
 912         if (dquot->dq_dqb.dqb_bsoftlimit &&
 913            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 914             dquot->dq_dqb.dqb_btime == 0) {
 915                 if (!prealloc) {
 916                         *warntype = BSOFTWARN;
 917                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
 918                 }
 919                 else
 920                         /*
 921                          * We don't allow preallocation to exceed softlimit so exceeding will
 922                          * be always printed
 923                          */
 924                         return NO_QUOTA;
 925         }
 926
 927         return QUOTA_OK;
 928 }
 929
 930 /*
 931  *      Initialize quota pointers in inode
 932  *      Transaction must be started at entry
 933  */
 934 int dquot_initialize(struct inode *inode, int type)
 935 {
 936         unsigned int id = 0;
 937         int cnt, ret = 0;
 938
 939         /* First test before acquiring semaphore - solves deadlocks when we
 940          * re-enter the quota code and are already holding the semaphore */
 941         if (IS_NOQUOTA(inode))
 942                 return 0;
 943         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 944         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
 945         if (IS_NOQUOTA(inode))
 946                 goto out_err;
 947         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 948                 if (type != -1 && cnt != type)
 949                         continue;
 950                 if (inode->i_dquot[cnt] == NODQUOT) {
 951                         switch (cnt) {
 952                                 case USRQUOTA:
 953                                         id = inode->i_uid;
 954                                         break;
 955                                 case GRPQUOTA:
 956                                         id = inode->i_gid;
 957                                         break;
 958                         }
 959                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
 960                 }
 961         }
 962 out_err:
 963         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 964         return ret;
 965 }
 966
 967 /*
 968  *      Release all quotas referenced by inode
 969  *      Transaction must be started at an entry
 970  */
 971 int dquot_drop(struct inode *inode)
 972 {
 973         int cnt;
 974
 975         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 976         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 977                 if (inode->i_dquot[cnt] != NODQUOT) {
 978                         dqput(inode->i_dquot[cnt]);
 979                         inode->i_dquot[cnt] = NODQUOT;
 980                 }
 981         }
 982         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 983         return 0;
 984 }
 985
 986 /*
 987  * Following four functions update i_blocks+i_bytes fields and
 988  * quota information (together with appropriate checks)
 989  * NOTE: We absolutely rely on the fact that caller dirties
 990  * the inode (usually macros in quotaops.h care about this) and
 991  * holds a handle for the current transaction so that dquot write and
 992  * inode write go into the same transaction.
 993  */
 994
 995 /*
 996  * This operation can block, but only after everything is updated
 997  */
 998 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
 999 {
1000         int cnt, ret = NO_QUOTA;
1001         char warntype[MAXQUOTAS];
1002
1003         /* First test before acquiring semaphore - solves deadlocks when we
1004          * re-enter the quota code and are already holding the semaphore */
1005         if (IS_NOQUOTA(inode)) {
1006 out_add:
1007                 inode_add_bytes(inode, number);
1008                 return QUOTA_OK;
1009         }
1010         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1011                 warntype[cnt] = NOWARN;
1012
1013         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1014         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1015                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1016                 goto out_add;
1017         }
1018         spin_lock(&dq_data_lock);
1019         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1020                 if (inode->i_dquot[cnt] == NODQUOT)
1021                         continue;
1022                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1023                         goto warn_put_all;
1024         }
1025         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1026                 if (inode->i_dquot[cnt] == NODQUOT)
1027                         continue;
1028                 dquot_incr_space(inode->i_dquot[cnt], number);
1029         }
1030         inode_add_bytes(inode, number);
1031         ret = QUOTA_OK;
1032 warn_put_all:
1033         spin_unlock(&dq_data_lock);
1034         if (ret == QUOTA_OK)
1035                 /* Dirtify all the dquots - this can block when journalling */
1036                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1037                         if (inode->i_dquot[cnt])
1038                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1039         flush_warnings(inode->i_dquot, warntype);
1040         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1041         return ret;
1042 }
1043
1044 /*
1045  * This operation can block, but only after everything is updated
1046  */
1047 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1048 {
1049         int cnt, ret = NO_QUOTA;
1050         char warntype[MAXQUOTAS];
1051
1052         /* First test before acquiring semaphore - solves deadlocks when we
1053          * re-enter the quota code and are already holding the semaphore */
1054         if (IS_NOQUOTA(inode))
1055                 return QUOTA_OK;
1056         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1057                 warntype[cnt] = NOWARN;
1058         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1059         if (IS_NOQUOTA(inode)) {
1060                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1061                 return QUOTA_OK;
1062         }
1063         spin_lock(&dq_data_lock);
1064         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1065                 if (inode->i_dquot[cnt] == NODQUOT)
1066                         continue;
1067                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1068                         goto warn_put_all;
1069         }
1070
1071         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1072                 if (inode->i_dquot[cnt] == NODQUOT)
1073                         continue;
1074                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1075         }
1076         ret = QUOTA_OK;
1077 warn_put_all:
1078         spin_unlock(&dq_data_lock);
1079         if (ret == QUOTA_OK)
1080                 /* Dirtify all the dquots - this can block when journalling */
1081                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1082                         if (inode->i_dquot[cnt])
1083                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1084         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1085         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1086         return ret;
1087 }
1088
1089 /*
1090  * This operation can block, but only after everything is updated
1091  */
1092 int dquot_free_space(struct inode *inode, qsize_t number)
1093 {
1094         unsigned int cnt;
1095
1096         /* First test before acquiring semaphore - solves deadlocks when we
1097          * re-enter the quota code and are already holding the semaphore */
1098         if (IS_NOQUOTA(inode)) {
1099 out_sub:
1100                 inode_sub_bytes(inode, number);
1101                 return QUOTA_OK;
1102         }
1103         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1104         /* Now recheck reliably when holding dqptr_sem */
1105         if (IS_NOQUOTA(inode)) {
1106                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1107                 goto out_sub;
1108         }
1109         spin_lock(&dq_data_lock);
1110         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1111                 if (inode->i_dquot[cnt] == NODQUOT)
1112                         continue;
1113                 dquot_decr_space(inode->i_dquot[cnt], number);
1114         }
1115         inode_sub_bytes(inode, number);
1116         spin_unlock(&dq_data_lock);
1117         /* Dirtify all the dquots - this can block when journalling */
1118         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1119                 if (inode->i_dquot[cnt])
1120                         mark_dquot_dirty(inode->i_dquot[cnt]);
1121         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1122         return QUOTA_OK;
1123 }
1124
1125 /*
1126  * This operation can block, but only after everything is updated
1127  */
1128 int dquot_free_inode(const struct inode *inode, unsigned long number)
1129 {
1130         unsigned int cnt;
1131
1132         /* First test before acquiring semaphore - solves deadlocks when we
1133          * re-enter the quota code and are already holding the semaphore */
1134         if (IS_NOQUOTA(inode))
1135                 return QUOTA_OK;
1136         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1137         /* Now recheck reliably when holding dqptr_sem */
1138         if (IS_NOQUOTA(inode)) {
1139                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1140                 return QUOTA_OK;
1141         }
1142         spin_lock(&dq_data_lock);
1143         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1144                 if (inode->i_dquot[cnt] == NODQUOT)
1145                         continue;
1146                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1147         }
1148         spin_unlock(&dq_data_lock);
1149         /* Dirtify all the dquots - this can block when journalling */
1150         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1151                 if (inode->i_dquot[cnt])
1152                         mark_dquot_dirty(inode->i_dquot[cnt]);
1153         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1154         return QUOTA_OK;
1155 }
1156
1157 /*
1158  * Transfer the number of inode and blocks from one diskquota to an other.
1159  *
1160  * This operation can block, but only after everything is updated
1161  * A transaction must be started when entering this function.
1162  */
1163 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1164 {
1165         qsize_t space;
1166         struct dquot *transfer_from[MAXQUOTAS];
1167         struct dquot *transfer_to[MAXQUOTAS];
1168         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1169             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1170         char warntype[MAXQUOTAS];
1171
1172         /* First test before acquiring semaphore - solves deadlocks when we
1173          * re-enter the quota code and are already holding the semaphore */
1174         if (IS_NOQUOTA(inode))
1175                 return QUOTA_OK;
1176         /* Clear the arrays */
1177         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1178                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1179                 warntype[cnt] = NOWARN;
1180         }
1181         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1182         /* Now recheck reliably when holding dqptr_sem */
1183         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1184                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1185                 return QUOTA_OK;
1186         }
1187         /* First build the transfer_to list - here we can block on
1188          * reading/instantiating of dquots.  We know that the transaction for
1189          * us was already started so we don't violate lock ranking here */
1190         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1191                 switch (cnt) {
1192                         case USRQUOTA:
1193                                 if (!chuid)
1194                                         continue;
1195                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1196                                 break;
1197                         case GRPQUOTA:
1198                                 if (!chgid)
1199                                         continue;
1200                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1201                                 break;
1202                 }
1203         }
1204         spin_lock(&dq_data_lock);
1205         space = inode_get_bytes(inode);
1206         /* Build the transfer_from list and check the limits */
1207         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1208                 if (transfer_to[cnt] == NODQUOT)
1209                         continue;
1210                 transfer_from[cnt] = inode->i_dquot[cnt];
1211                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1212                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1213                         goto warn_put_all;
1214         }
1215
1216         /*
1217          * Finally perform the needed transfer from transfer_from to transfer_to
1218          */
1219         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1220                 /*
1221                  * Skip changes for same uid or gid or for turned off quota-type.
1222                  */
1223                 if (transfer_to[cnt] == NODQUOT)
1224                         continue;
1225
1226                 /* Due to IO error we might not have transfer_from[] structure */
1227                 if (transfer_from[cnt]) {
1228                         dquot_decr_inodes(transfer_from[cnt], 1);
1229                         dquot_decr_space(transfer_from[cnt], space);
1230                 }
1231
1232                 dquot_incr_inodes(transfer_to[cnt], 1);
1233                 dquot_incr_space(transfer_to[cnt], space);
1234
1235                 inode->i_dquot[cnt] = transfer_to[cnt];
1236         }
1237         ret = QUOTA_OK;
1238 warn_put_all:
1239         spin_unlock(&dq_data_lock);
1240         /* Dirtify all the dquots - this can block when journalling */
1241         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1242                 if (transfer_from[cnt])
1243                         mark_dquot_dirty(transfer_from[cnt]);
1244                 if (transfer_to[cnt])
1245                         mark_dquot_dirty(transfer_to[cnt]);
1246         }
1247         flush_warnings(transfer_to, warntype);
1248
1249         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1250                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1251                         dqput(transfer_from[cnt]);
1252                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1253                         dqput(transfer_to[cnt]);
1254         }
1255         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1256         return ret;
1257 }
1258
1259 /*
1260  * Write info of quota file to disk
1261  */
1262 int dquot_commit_info(struct super_block *sb, int type)
1263 {
1264         int ret;
1265         struct quota_info *dqopt = sb_dqopt(sb);
1266
1267         down(&dqopt->dqio_sem);
1268         ret = dqopt->ops[type]->write_file_info(sb, type);
1269         up(&dqopt->dqio_sem);
1270         return ret;
1271 }
1272
1273 /*
1274  * Definitions of diskquota operations.
1275  */
1276 struct dquot_operations dquot_operations = {
1277         .initialize     = dquot_initialize,
1278         .drop           = dquot_drop,
1279         .alloc_space    = dquot_alloc_space,
1280         .alloc_inode    = dquot_alloc_inode,
1281         .free_space     = dquot_free_space,
1282         .free_inode     = dquot_free_inode,
1283         .transfer       = dquot_transfer,
1284         .write_dquot    = dquot_commit,
1285         .acquire_dquot  = dquot_acquire,
1286         .release_dquot  = dquot_release,
1287         .mark_dirty     = dquot_mark_dquot_dirty,
1288         .write_info     = dquot_commit_info
1289 };
1290
1291 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1292 {
1293         switch (type) {
1294                 case USRQUOTA:
1295                         dqopt->flags |= DQUOT_USR_ENABLED;
1296                         break;
1297                 case GRPQUOTA:
1298                         dqopt->flags |= DQUOT_GRP_ENABLED;
1299                         break;
1300         }
1301 }
1302
1303 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1304 {
1305         switch (type) {
1306                 case USRQUOTA:
1307                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1308                         break;
1309                 case GRPQUOTA:
1310                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1311                         break;
1312         }
1313 }
1314
1315 /*
1316  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1317  */
1318 int vfs_quota_off(struct super_block *sb, int type)
1319 {
1320         int cnt;
1321         struct quota_info *dqopt = sb_dqopt(sb);
1322         struct inode *toputinode[MAXQUOTAS];
1323         struct vfsmount *toputmnt[MAXQUOTAS];
1324
1325         /* We need to serialize quota_off() for device */
1326         down(&dqopt->dqonoff_sem);
1327         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1328                 toputinode[cnt] = NULL;
1329                 toputmnt[cnt] = NULL;
1330                 if (type != -1 && cnt != type)
1331                         continue;
1332                 if (!sb_has_quota_enabled(sb, cnt))
1333                         continue;
1334                 reset_enable_flags(dqopt, cnt);
1335
1336                 /* Note: these are blocking operations */
1337                 drop_dquot_ref(sb, cnt);
1338                 invalidate_dquots(sb, cnt);
1339                 /*
1340                  * Now all dquots should be invalidated, all writes done so we should be only
1341                  * users of the info. No locks needed.
1342                  */
1343                 if (info_dirty(&dqopt->info[cnt]))
1344                         sb->dq_op->write_info(sb, cnt);
1345                 if (dqopt->ops[cnt]->free_file_info)
1346                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1347                 put_quota_format(dqopt->info[cnt].dqi_format);
1348
1349                 toputinode[cnt] = dqopt->files[cnt];
1350                 toputmnt[cnt] = dqopt->mnt[cnt];
1351                 dqopt->files[cnt] = NULL;
1352                 dqopt->mnt[cnt] = NULL;
1353                 dqopt->info[cnt].dqi_flags = 0;
1354                 dqopt->info[cnt].dqi_igrace = 0;
1355                 dqopt->info[cnt].dqi_bgrace = 0;
1356                 dqopt->ops[cnt] = NULL;
1357         }
1358         up(&dqopt->dqonoff_sem);
1359         /* Sync the superblock so that buffers with quota data are written to
1360          * disk (and so userspace sees correct data afterwards).
1361          * The reference to vfsmnt we are still holding protects us from
1362          * umount (we don't have it only when quotas are turned on/off for
1363          * journal replay but in that case we are guarded by the fs anyway). */
1364         if (sb->s_op->sync_fs)
1365                 sb->s_op->sync_fs(sb, 1);
1366         sync_blockdev(sb->s_bdev);
1367         /* Now the quota files are just ordinary files and we can set the
1368          * inode flags back. Moreover we discard the pagecache so that
1369          * userspace sees the writes we did bypassing the pagecache. We
1370          * must also discard the blockdev buffers so that we see the
1371          * changes done by userspace on the next quotaon() */
1372         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1373                 if (toputinode[cnt]) {
1374                         down(&dqopt->dqonoff_sem);
1375                         /* If quota was reenabled in the meantime, we have
1376                          * nothing to do */
1377                         if (!sb_has_quota_enabled(sb, cnt)) {
1378                                 down(&toputinode[cnt]->i_sem);
1379                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1380                                   S_NOATIME | S_NOQUOTA);
1381                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1382                                 up(&toputinode[cnt]->i_sem);
1383                                 mark_inode_dirty(toputinode[cnt]);
1384                                 iput(toputinode[cnt]);
1385                         }
1386                         up(&dqopt->dqonoff_sem);
1387                         /* We don't hold the reference when we turned on quotas
1388                          * just for the journal replay... */
1389                         if (toputmnt[cnt])
1390                                 mntput(toputmnt[cnt]);
1391                 }
1392         if (sb->s_bdev)
1393                 invalidate_bdev(sb->s_bdev, 0);
1394         return 0;
1395 }
1396
1397 /*
1398  *      Turn quotas on on a device
1399  */
1400
1401 /* Helper function when we already have the inode */
1402 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1403 {
1404         struct quota_format_type *fmt = find_quota_format(format_id);
1405         struct super_block *sb = inode->i_sb;
1406         struct quota_info *dqopt = sb_dqopt(sb);
1407         int error;
1408         int oldflags = -1;
1409
1410         if (!fmt)
1411                 return -ESRCH;
1412         if (!S_ISREG(inode->i_mode)) {
1413                 error = -EACCES;
1414                 goto out_fmt;
1415         }
1416         if (IS_RDONLY(inode)) {
1417                 error = -EROFS;
1418                 goto out_fmt;
1419         }
1420         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1421                 error = -EINVAL;
1422                 goto out_fmt;
1423         }
1424
1425         /* As we bypass the pagecache we must now flush the inode so that
1426          * we see all the changes from userspace... */
1427         write_inode_now(inode, 1);
1428         /* And now flush the block cache so that kernel sees the changes */
1429         invalidate_bdev(sb->s_bdev, 0);
1430         down(&inode->i_sem);
1431         down(&dqopt->dqonoff_sem);
1432         if (sb_has_quota_enabled(sb, type)) {
1433                 error = -EBUSY;
1434                 goto out_lock;
1435         }
1436         /* We don't want quota and atime on quota files (deadlocks possible)
1437          * Also nobody should write to the file - we use special IO operations
1438          * which ignore the immutable bit. */
1439         down_write(&dqopt->dqptr_sem);
1440         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1441         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1442         up_write(&dqopt->dqptr_sem);
1443         sb->dq_op->drop(inode);
1444
1445         error = -EIO;
1446         dqopt->files[type] = igrab(inode);
1447         if (!dqopt->files[type])
1448                 goto out_lock;
1449         error = -EINVAL;
1450         if (!fmt->qf_ops->check_quota_file(sb, type))
1451                 goto out_file_init;
1452
1453         dqopt->ops[type] = fmt->qf_ops;
1454         dqopt->info[type].dqi_format = fmt;
1455         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1456         down(&dqopt->dqio_sem);
1457         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1458                 up(&dqopt->dqio_sem);
1459                 goto out_file_init;
1460         }
1461         up(&dqopt->dqio_sem);
1462         up(&inode->i_sem);
1463         set_enable_flags(dqopt, type);
1464
1465         add_dquot_ref(sb, type);
1466         up(&dqopt->dqonoff_sem);
1467
1468         return 0;
1469
1470 out_file_init:
1471         dqopt->files[type] = NULL;
1472         iput(inode);
1473 out_lock:
1474         up(&dqopt->dqonoff_sem);
1475         if (oldflags != -1) {
1476                 down_write(&dqopt->dqptr_sem);
1477                 /* Set the flags back (in the case of accidental quotaon()
1478                  * on a wrong file we don't want to mess up the flags) */
1479                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1480                 inode->i_flags |= oldflags;
1481                 up_write(&dqopt->dqptr_sem);
1482         }
1483         up(&inode->i_sem);
1484 out_fmt:
1485         put_quota_format(fmt);
1486
1487         return error;
1488 }
1489
1490 /* Actual function called from quotactl() */
1491 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1492 {
1493         struct nameidata nd;
1494         int error;
1495
1496         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1497         if (error < 0)
1498                 return error;
1499         error = security_quota_on(nd.dentry);
1500         if (error)
1501                 goto out_path;
1502         /* Quota file not on the same filesystem? */
1503         if (nd.mnt->mnt_sb != sb)
1504                 error = -EXDEV;
1505         else {
1506                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1507                 if (!error)
1508                         sb_dqopt(sb)->mnt[type] = mntget(nd.mnt);
1509         }
1510 out_path:
1511         path_release(&nd);
1512         return error;
1513 }
1514
1515 /*
1516  * This function is used when filesystem needs to initialize quotas
1517  * during mount time.
1518  */
1519 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1520                 int format_id, int type)
1521 {
1522         struct dentry *dentry;
1523         int error;
1524
1525         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1526         if (IS_ERR(dentry))
1527                 return PTR_ERR(dentry);
1528
1529         error = security_quota_on(dentry);
1530         if (!error)
1531                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1532
1533         dput(dentry);
1534         return error;
1535 }
1536
1537 /* Generic routine for getting common part of quota structure */
1538 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1539 {
1540         struct mem_dqblk *dm = &dquot->dq_dqb;
1541
1542         spin_lock(&dq_data_lock);
1543         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1544         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1545         di->dqb_curspace = dm->dqb_curspace;
1546         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1547         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1548         di->dqb_curinodes = dm->dqb_curinodes;
1549         di->dqb_btime = dm->dqb_btime;
1550         di->dqb_itime = dm->dqb_itime;
1551         di->dqb_valid = QIF_ALL;
1552         spin_unlock(&dq_data_lock);
1553 }
1554
1555 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1556 {
1557         struct dquot *dquot;
1558
1559         down(&sb_dqopt(sb)->dqonoff_sem);
1560         if (!(dquot = dqget(sb, id, type))) {
1561                 up(&sb_dqopt(sb)->dqonoff_sem);
1562                 return -ESRCH;
1563         }
1564         do_get_dqblk(dquot, di);
1565         dqput(dquot);
1566         up(&sb_dqopt(sb)->dqonoff_sem);
1567         return 0;
1568 }
1569
1570 /* Generic routine for setting common part of quota structure */
1571 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1572 {
1573         struct mem_dqblk *dm = &dquot->dq_dqb;
1574         int check_blim = 0, check_ilim = 0;
1575
1576         spin_lock(&dq_data_lock);
1577         if (di->dqb_valid & QIF_SPACE) {
1578                 dm->dqb_curspace = di->dqb_curspace;
1579                 check_blim = 1;
1580         }
1581         if (di->dqb_valid & QIF_BLIMITS) {
1582                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1583                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1584                 check_blim = 1;
1585         }
1586         if (di->dqb_valid & QIF_INODES) {
1587                 dm->dqb_curinodes = di->dqb_curinodes;
1588                 check_ilim = 1;
1589         }
1590         if (di->dqb_valid & QIF_ILIMITS) {
1591                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1592                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1593                 check_ilim = 1;
1594         }
1595         if (di->dqb_valid & QIF_BTIME)
1596                 dm->dqb_btime = di->dqb_btime;
1597         if (di->dqb_valid & QIF_ITIME)
1598                 dm->dqb_itime = di->dqb_itime;
1599
1600         if (check_blim) {
1601                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1602                         dm->dqb_btime = 0;
1603                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1604                 }
1605                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1606                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1607         }
1608         if (check_ilim) {
1609                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1610                         dm->dqb_itime = 0;
1611                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1612                 }
1613                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1614                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1615         }
1616         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1617                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1618         else
1619                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1620         spin_unlock(&dq_data_lock);
1621         mark_dquot_dirty(dquot);
1622 }
1623
1624 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1625 {
1626         struct dquot *dquot;
1627
1628         down(&sb_dqopt(sb)->dqonoff_sem);
1629         if (!(dquot = dqget(sb, id, type))) {
1630                 up(&sb_dqopt(sb)->dqonoff_sem);
1631                 return -ESRCH;
1632         }
1633         do_set_dqblk(dquot, di);
1634         dqput(dquot);
1635         up(&sb_dqopt(sb)->dqonoff_sem);
1636         return 0;
1637 }
1638
1639 /* Generic routine for getting common part of quota file information */
1640 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1641 {
1642         struct mem_dqinfo *mi;
1643
1644         down(&sb_dqopt(sb)->dqonoff_sem);
1645         if (!sb_has_quota_enabled(sb, type)) {
1646                 up(&sb_dqopt(sb)->dqonoff_sem);
1647                 return -ESRCH;
1648         }
1649         mi = sb_dqopt(sb)->info + type;
1650         spin_lock(&dq_data_lock);
1651         ii->dqi_bgrace = mi->dqi_bgrace;
1652         ii->dqi_igrace = mi->dqi_igrace;
1653         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1654         ii->dqi_valid = IIF_ALL;
1655         spin_unlock(&dq_data_lock);
1656         up(&sb_dqopt(sb)->dqonoff_sem);
1657         return 0;
1658 }
1659
1660 /* Generic routine for setting common part of quota file information */
1661 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1662 {
1663         struct mem_dqinfo *mi;
1664
1665         down(&sb_dqopt(sb)->dqonoff_sem);
1666         if (!sb_has_quota_enabled(sb, type)) {
1667                 up(&sb_dqopt(sb)->dqonoff_sem);
1668                 return -ESRCH;
1669         }
1670         mi = sb_dqopt(sb)->info + type;
1671         spin_lock(&dq_data_lock);
1672         if (ii->dqi_valid & IIF_BGRACE)
1673                 mi->dqi_bgrace = ii->dqi_bgrace;
1674         if (ii->dqi_valid & IIF_IGRACE)
1675                 mi->dqi_igrace = ii->dqi_igrace;
1676         if (ii->dqi_valid & IIF_FLAGS)
1677                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1678         spin_unlock(&dq_data_lock);
1679         mark_info_dirty(sb, type);
1680         /* Force write to disk */
1681         sb->dq_op->write_info(sb, type);
1682         up(&sb_dqopt(sb)->dqonoff_sem);
1683         return 0;
1684 }
1685
1686 struct quotactl_ops vfs_quotactl_ops = {
1687         .quota_on       = vfs_quota_on,
1688         .quota_off      = vfs_quota_off,
1689         .quota_sync     = vfs_quota_sync,
1690         .get_info       = vfs_get_dqinfo,
1691         .set_info       = vfs_set_dqinfo,
1692         .get_dqblk      = vfs_get_dqblk,
1693         .set_dqblk      = vfs_set_dqblk
1694 };
1695
1696 static ctl_table fs_dqstats_table[] = {
1697         {
1698                 .ctl_name       = FS_DQ_LOOKUPS,
1699                 .procname       = "lookups",
1700                 .data           = &dqstats.lookups,
1701                 .maxlen         = sizeof(int),
1702                 .mode           = 0444,
1703                 .proc_handler   = &proc_dointvec,
1704         },
1705         {
1706                 .ctl_name       = FS_DQ_DROPS,
1707                 .procname       = "drops",
1708                 .data           = &dqstats.drops,
1709                 .maxlen         = sizeof(int),
1710                 .mode           = 0444,
1711                 .proc_handler   = &proc_dointvec,
1712         },
1713         {
1714                 .ctl_name       = FS_DQ_READS,
1715                 .procname       = "reads",
1716                 .data           = &dqstats.reads,
1717                 .maxlen         = sizeof(int),
1718                 .mode           = 0444,
1719                 .proc_handler   = &proc_dointvec,
1720         },
1721         {
1722                 .ctl_name       = FS_DQ_WRITES,
1723                 .procname       = "writes",
1724                 .data           = &dqstats.writes,
1725                 .maxlen         = sizeof(int),
1726                 .mode           = 0444,
1727                 .proc_handler   = &proc_dointvec,
1728         },
1729         {
1730                 .ctl_name       = FS_DQ_CACHE_HITS,
1731                 .procname       = "cache_hits",
1732                 .data           = &dqstats.cache_hits,
1733                 .maxlen         = sizeof(int),
1734                 .mode           = 0444,
1735                 .proc_handler   = &proc_dointvec,
1736         },
1737         {
1738                 .ctl_name       = FS_DQ_ALLOCATED,
1739                 .procname       = "allocated_dquots",
1740                 .data           = &dqstats.allocated_dquots,
1741                 .maxlen         = sizeof(int),
1742                 .mode           = 0444,
1743                 .proc_handler   = &proc_dointvec,
1744         },
1745         {
1746                 .ctl_name       = FS_DQ_FREE,
1747                 .procname       = "free_dquots",
1748                 .data           = &dqstats.free_dquots,
1749                 .maxlen         = sizeof(int),
1750                 .mode           = 0444,
1751                 .proc_handler   = &proc_dointvec,
1752         },
1753         {
1754                 .ctl_name       = FS_DQ_SYNCS,
1755                 .procname       = "syncs",
1756                 .data           = &dqstats.syncs,
1757                 .maxlen         = sizeof(int),
1758                 .mode           = 0444,
1759                 .proc_handler   = &proc_dointvec,
1760         },
1761         {
1762                 .ctl_name       = FS_DQ_WARNINGS,
1763                 .procname       = "warnings",
1764                 .data           = &flag_print_warnings,
1765                 .maxlen         = sizeof(int),
1766                 .mode           = 0644,
1767                 .proc_handler   = &proc_dointvec,
1768         },
1769         { .ctl_name = 0 },
1770 };
1771
1772 static ctl_table fs_table[] = {
1773         {
1774                 .ctl_name       = FS_DQSTATS,
1775                 .procname       = "quota",
1776                 .mode           = 0555,
1777                 .child          = fs_dqstats_table,
1778         },
1779         { .ctl_name = 0 },
1780 };
1781
1782 static ctl_table sys_table[] = {
1783         {
1784                 .ctl_name       = CTL_FS,
1785                 .procname       = "fs",
1786                 .mode           = 0555,
1787                 .child          = fs_table,
1788         },
1789         { .ctl_name = 0 },
1790 };
1791
1792 static int __init dquot_init(void)
1793 {
1794         int i;
1795         unsigned long nr_hash, order;
1796
1797         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1798
1799         register_sysctl_table(sys_table, 0);
1800
1801         dquot_cachep = kmem_cache_create("dquot",
1802                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1803                         SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1804                         NULL, NULL);
1805
1806         order = 0;
1807         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1808         if (!dquot_hash)
1809                 panic("Cannot create dquot hash table");
1810
1811         /* Find power-of-two hlist_heads which can fit into allocation */
1812         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1813         dq_hash_bits = 0;
1814         do {
1815                 dq_hash_bits++;
1816         } while (nr_hash >> dq_hash_bits);
1817         dq_hash_bits--;
1818
1819         nr_hash = 1UL << dq_hash_bits;
1820         dq_hash_mask = nr_hash - 1;
1821         for (i = 0; i < nr_hash; i++)
1822                 INIT_HLIST_HEAD(dquot_hash + i);
1823
1824         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1825                         nr_hash, order, (PAGE_SIZE << order));
1826
1827         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1828
1829         return 0;
1830 }
1831 module_init(dquot_init);
1832
1833 EXPORT_SYMBOL(register_quota_format);
1834 EXPORT_SYMBOL(unregister_quota_format);
1835 EXPORT_SYMBOL(dqstats);
1836 EXPORT_SYMBOL(dq_data_lock);
1837 EXPORT_SYMBOL(vfs_quota_on);
1838 EXPORT_SYMBOL(vfs_quota_on_mount);
1839 EXPORT_SYMBOL(vfs_quota_off);
1840 EXPORT_SYMBOL(vfs_quota_sync);
1841 EXPORT_SYMBOL(vfs_get_dqinfo);
1842 EXPORT_SYMBOL(vfs_set_dqinfo);
1843 EXPORT_SYMBOL(vfs_get_dqblk);
1844 EXPORT_SYMBOL(vfs_set_dqblk);
1845 EXPORT_SYMBOL(dquot_commit);
1846 EXPORT_SYMBOL(dquot_commit_info);
1847 EXPORT_SYMBOL(dquot_acquire);
1848 EXPORT_SYMBOL(dquot_release);
1849 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1850 EXPORT_SYMBOL(dquot_initialize);
1851 EXPORT_SYMBOL(dquot_drop);
1852 EXPORT_SYMBOL(dquot_alloc_space);
1853 EXPORT_SYMBOL(dquot_alloc_inode);
1854 EXPORT_SYMBOL(dquot_free_space);
1855 EXPORT_SYMBOL(dquot_free_inode);
1856 EXPORT_SYMBOL(dquot_transfer);