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 #include <linux/capability.h>
  81 #include <linux/quotaops.h>
  82
  83 #include <asm/uaccess.h>
  84
  85 #define __DQUOT_PARANOIA
  86
  87 /*
  88  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
  89  * and quota formats and also dqstats structure containing statistics about the
  90  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
  91  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
  92  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
  93  * in inode_add_bytes() and inode_sub_bytes().
  94  *
  95  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
  96  *
  97  * Note that some things (eg. sb pointer, type, id) doesn't change during
  98  * the life of the dquot structure and so needn't to be protected by a lock
  99  *
 100  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
 101  * operation is just reading pointers from inode (or not using them at all) the
 102  * read lock is enough. If pointers are altered function must hold write lock
 103  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
 104  * for altering the flag i_mutex is also needed).  If operation is holding
 105  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
 106  * dqonoff_mutex.
 107  * This locking assures that:
 108  *   a) update/access to dquot pointers in inode is serialized
 109  *   b) everyone is guarded against invalidate_dquots()
 110  *
 111  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
 112  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
 113  * Currently dquot is locked only when it is being read to memory (or space for
 114  * it is being allocated) on the first dqget() and when it is being released on
 115  * the last dqput(). The allocation and release oparations are serialized by
 116  * the dq_lock and by checking the use count in dquot_release().  Write
 117  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
 118  * spinlock to internal buffers before writing.
 119  *
 120  * Lock ordering (including related VFS locks) is the following:
 121  *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
 122  *   dqio_mutex
 123  * i_mutex on quota files is special (it's below dqio_mutex)
 124  */
 125
 126 static DEFINE_SPINLOCK(dq_list_lock);
 127 DEFINE_SPINLOCK(dq_data_lock);
 128
 129 static char *quotatypes[] = INITQFNAMES;
 130 static struct quota_format_type *quota_formats; /* List of registered formats */
 131 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 132
 133 /* SLAB cache for dquot structures */
 134 static struct kmem_cache *dquot_cachep;
 135
 136 int register_quota_format(struct quota_format_type *fmt)
 137 {
 138         spin_lock(&dq_list_lock);
 139         fmt->qf_next = quota_formats;
 140         quota_formats = fmt;
 141         spin_unlock(&dq_list_lock);
 142         return 0;
 143 }
 144
 145 void unregister_quota_format(struct quota_format_type *fmt)
 146 {
 147         struct quota_format_type **actqf;
 148
 149         spin_lock(&dq_list_lock);
 150         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
 151         if (*actqf)
 152                 *actqf = (*actqf)->qf_next;
 153         spin_unlock(&dq_list_lock);
 154 }
 155
 156 static struct quota_format_type *find_quota_format(int id)
 157 {
 158         struct quota_format_type *actqf;
 159
 160         spin_lock(&dq_list_lock);
 161         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 162         if (!actqf || !try_module_get(actqf->qf_owner)) {
 163                 int qm;
 164
 165                 spin_unlock(&dq_list_lock);
 166
 167                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
 168                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
 169                         return NULL;
 170
 171                 spin_lock(&dq_list_lock);
 172                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 173                 if (actqf && !try_module_get(actqf->qf_owner))
 174                         actqf = NULL;
 175         }
 176         spin_unlock(&dq_list_lock);
 177         return actqf;
 178 }
 179
 180 static void put_quota_format(struct quota_format_type *fmt)
 181 {
 182         module_put(fmt->qf_owner);
 183 }
 184
 185 /*
 186  * Dquot List Management:
 187  * The quota code uses three lists for dquot management: the inuse_list,
 188  * free_dquots, and dquot_hash[] array. A single dquot structure may be
 189  * on all three lists, depending on its current state.
 190  *
 191  * All dquots are placed to the end of inuse_list when first created, and this
 192  * list is used for invalidate operation, which must look at every dquot.
 193  *
 194  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 195  * and this list is searched whenever we need an available dquot.  Dquots are
 196  * removed from the list as soon as they are used again, and
 197  * dqstats.free_dquots gives the number of dquots on the list. When
 198  * dquot is invalidated it's completely released from memory.
 199  *
 200  * Dquots with a specific identity (device, type and id) are placed on
 201  * one of the dquot_hash[] hash chains. The provides an efficient search
 202  * mechanism to locate a specific dquot.
 203  */
 204
 205 static LIST_HEAD(inuse_list);
 206 static LIST_HEAD(free_dquots);
 207 static unsigned int dq_hash_bits, dq_hash_mask;
 208 static struct hlist_head *dquot_hash;
 209
 210 struct dqstats dqstats;
 211
 212 static void dqput(struct dquot *dquot);
 213
 214 static inline unsigned int
 215 hashfn(const struct super_block *sb, unsigned int id, int type)
 216 {
 217         unsigned long tmp;
 218
 219         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 220         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 221 }
 222
 223 /*
 224  * Following list functions expect dq_list_lock to be held
 225  */
 226 static inline void insert_dquot_hash(struct dquot *dquot)
 227 {
 228         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
 229         hlist_add_head(&dquot->dq_hash, head);
 230 }
 231
 232 static inline void remove_dquot_hash(struct dquot *dquot)
 233 {
 234         hlist_del_init(&dquot->dq_hash);
 235 }
 236
 237 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
 238 {
 239         struct hlist_node *node;
 240         struct dquot *dquot;
 241
 242         hlist_for_each (node, dquot_hash+hashent) {
 243                 dquot = hlist_entry(node, struct dquot, dq_hash);
 244                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
 245                         return dquot;
 246         }
 247         return NODQUOT;
 248 }
 249
 250 /* Add a dquot to the tail of the free list */
 251 static inline void put_dquot_last(struct dquot *dquot)
 252 {
 253         list_add_tail(&dquot->dq_free, &free_dquots);
 254         dqstats.free_dquots++;
 255 }
 256
 257 static inline void remove_free_dquot(struct dquot *dquot)
 258 {
 259         if (list_empty(&dquot->dq_free))
 260                 return;
 261         list_del_init(&dquot->dq_free);
 262         dqstats.free_dquots--;
 263 }
 264
 265 static inline void put_inuse(struct dquot *dquot)
 266 {
 267         /* We add to the back of inuse list so we don't have to restart
 268          * when traversing this list and we block */
 269         list_add_tail(&dquot->dq_inuse, &inuse_list);
 270         dqstats.allocated_dquots++;
 271 }
 272
 273 static inline void remove_inuse(struct dquot *dquot)
 274 {
 275         dqstats.allocated_dquots--;
 276         list_del(&dquot->dq_inuse);
 277 }
 278 /*
 279  * End of list functions needing dq_list_lock
 280  */
 281
 282 static void wait_on_dquot(struct dquot *dquot)
 283 {
 284         mutex_lock(&dquot->dq_lock);
 285         mutex_unlock(&dquot->dq_lock);
 286 }
 287
 288 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
 289
 290 int dquot_mark_dquot_dirty(struct dquot *dquot)
 291 {
 292         spin_lock(&dq_list_lock);
 293         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
 294                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 295                                 info[dquot->dq_type].dqi_dirty_list);
 296         spin_unlock(&dq_list_lock);
 297         return 0;
 298 }
 299
 300 /* This function needs dq_list_lock */
 301 static inline int clear_dquot_dirty(struct dquot *dquot)
 302 {
 303         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
 304                 return 0;
 305         list_del_init(&dquot->dq_dirty);
 306         return 1;
 307 }
 308
 309 void mark_info_dirty(struct super_block *sb, int type)
 310 {
 311         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
 312 }
 313 EXPORT_SYMBOL(mark_info_dirty);
 314
 315 /*
 316  *      Read dquot from disk and alloc space for it
 317  */
 318
 319 int dquot_acquire(struct dquot *dquot)
 320 {
 321         int ret = 0, ret2 = 0;
 322         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 323
 324         mutex_lock(&dquot->dq_lock);
 325         mutex_lock(&dqopt->dqio_mutex);
 326         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 327                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
 328         if (ret < 0)
 329                 goto out_iolock;
 330         set_bit(DQ_READ_B, &dquot->dq_flags);
 331         /* Instantiate dquot if needed */
 332         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 333                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 334                 /* Write the info if needed */
 335                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 336                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 337                 if (ret < 0)
 338                         goto out_iolock;
 339                 if (ret2 < 0) {
 340                         ret = ret2;
 341                         goto out_iolock;
 342                 }
 343         }
 344         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 345 out_iolock:
 346         mutex_unlock(&dqopt->dqio_mutex);
 347         mutex_unlock(&dquot->dq_lock);
 348         return ret;
 349 }
 350
 351 /*
 352  *      Write dquot to disk
 353  */
 354 int dquot_commit(struct dquot *dquot)
 355 {
 356         int ret = 0, ret2 = 0;
 357         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 358
 359         mutex_lock(&dqopt->dqio_mutex);
 360         spin_lock(&dq_list_lock);
 361         if (!clear_dquot_dirty(dquot)) {
 362                 spin_unlock(&dq_list_lock);
 363                 goto out_sem;
 364         }
 365         spin_unlock(&dq_list_lock);
 366         /* Inactive dquot can be only if there was error during read/init
 367          * => we have better not writing it */
 368         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 369                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 370                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 371                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 372                 if (ret >= 0)
 373                         ret = ret2;
 374         }
 375 out_sem:
 376         mutex_unlock(&dqopt->dqio_mutex);
 377         return ret;
 378 }
 379
 380 /*
 381  *      Release dquot
 382  */
 383 int dquot_release(struct dquot *dquot)
 384 {
 385         int ret = 0, ret2 = 0;
 386         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 387
 388         mutex_lock(&dquot->dq_lock);
 389         /* Check whether we are not racing with some other dqget() */
 390         if (atomic_read(&dquot->dq_count) > 1)
 391                 goto out_dqlock;
 392         mutex_lock(&dqopt->dqio_mutex);
 393         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
 394                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
 395                 /* Write the info */
 396                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 397                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 398                 if (ret >= 0)
 399                         ret = ret2;
 400         }
 401         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 402         mutex_unlock(&dqopt->dqio_mutex);
 403 out_dqlock:
 404         mutex_unlock(&dquot->dq_lock);
 405         return ret;
 406 }
 407
 408 /* Invalidate all dquots on the list. Note that this function is called after
 409  * quota is disabled and pointers from inodes removed so there cannot be new
 410  * quota users. There can still be some users of quotas due to inodes being
 411  * just deleted or pruned by prune_icache() (those are not attached to any
 412  * list). We have to wait for such users.
 413  */
 414 static void invalidate_dquots(struct super_block *sb, int type)
 415 {
 416         struct dquot *dquot, *tmp;
 417
 418 restart:
 419         spin_lock(&dq_list_lock);
 420         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 421                 if (dquot->dq_sb != sb)
 422                         continue;
 423                 if (dquot->dq_type != type)
 424                         continue;
 425                 /* Wait for dquot users */
 426                 if (atomic_read(&dquot->dq_count)) {
 427                         DEFINE_WAIT(wait);
 428
 429                         atomic_inc(&dquot->dq_count);
 430                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
 431                                         TASK_UNINTERRUPTIBLE);
 432                         spin_unlock(&dq_list_lock);
 433                         /* Once dqput() wakes us up, we know it's time to free
 434                          * the dquot.
 435                          * IMPORTANT: we rely on the fact that there is always
 436                          * at most one process waiting for dquot to free.
 437                          * Otherwise dq_count would be > 1 and we would never
 438                          * wake up.
 439                          */
 440                         if (atomic_read(&dquot->dq_count) > 1)
 441                                 schedule();
 442                         finish_wait(&dquot->dq_wait_unused, &wait);
 443                         dqput(dquot);
 444                         /* At this moment dquot() need not exist (it could be
 445                          * reclaimed by prune_dqcache(). Hence we must
 446                          * restart. */
 447                         goto restart;
 448                 }
 449                 /*
 450                  * Quota now has no users and it has been written on last
 451                  * dqput()
 452                  */
 453                 remove_dquot_hash(dquot);
 454                 remove_free_dquot(dquot);
 455                 remove_inuse(dquot);
 456                 kmem_cache_free(dquot_cachep, dquot);
 457         }
 458         spin_unlock(&dq_list_lock);
 459 }
 460
 461 int vfs_quota_sync(struct super_block *sb, int type)
 462 {
 463         struct list_head *dirty;
 464         struct dquot *dquot;
 465         struct quota_info *dqopt = sb_dqopt(sb);
 466         int cnt;
 467
 468         mutex_lock(&dqopt->dqonoff_mutex);
 469         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 470                 if (type != -1 && cnt != type)
 471                         continue;
 472                 if (!sb_has_quota_enabled(sb, cnt))
 473                         continue;
 474                 spin_lock(&dq_list_lock);
 475                 dirty = &dqopt->info[cnt].dqi_dirty_list;
 476                 while (!list_empty(dirty)) {
 477                         dquot = list_entry(dirty->next, struct dquot, dq_dirty);
 478                         /* Dirty and inactive can be only bad dquot... */
 479                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 480                                 clear_dquot_dirty(dquot);
 481                                 continue;
 482                         }
 483                         /* Now we have active dquot from which someone is
 484                          * holding reference so we can safely just increase
 485                          * use count */
 486                         atomic_inc(&dquot->dq_count);
 487                         dqstats.lookups++;
 488                         spin_unlock(&dq_list_lock);
 489                         sb->dq_op->write_dquot(dquot);
 490                         dqput(dquot);
 491                         spin_lock(&dq_list_lock);
 492                 }
 493                 spin_unlock(&dq_list_lock);
 494         }
 495
 496         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 497                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
 498                         && info_dirty(&dqopt->info[cnt]))
 499                         sb->dq_op->write_info(sb, cnt);
 500         spin_lock(&dq_list_lock);
 501         dqstats.syncs++;
 502         spin_unlock(&dq_list_lock);
 503         mutex_unlock(&dqopt->dqonoff_mutex);
 504
 505         return 0;
 506 }
 507
 508 /* Free unused dquots from cache */
 509 static void prune_dqcache(int count)
 510 {
 511         struct list_head *head;
 512         struct dquot *dquot;
 513
 514         head = free_dquots.prev;
 515         while (head != &free_dquots && count) {
 516                 dquot = list_entry(head, struct dquot, dq_free);
 517                 remove_dquot_hash(dquot);
 518                 remove_free_dquot(dquot);
 519                 remove_inuse(dquot);
 520                 kmem_cache_free(dquot_cachep, dquot);
 521                 count--;
 522                 head = free_dquots.prev;
 523         }
 524 }
 525
 526 /*
 527  * This is called from kswapd when we think we need some
 528  * more memory
 529  */
 530
 531 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
 532 {
 533         if (nr) {
 534                 spin_lock(&dq_list_lock);
 535                 prune_dqcache(nr);
 536                 spin_unlock(&dq_list_lock);
 537         }
 538         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
 539 }
 540
 541 /*
 542  * Put reference to dquot
 543  * NOTE: If you change this function please check whether dqput_blocks() works right...
 544  * MUST be called with either dqptr_sem or dqonoff_mutex held
 545  */
 546 static void dqput(struct dquot *dquot)
 547 {
 548         if (!dquot)
 549                 return;
 550 #ifdef __DQUOT_PARANOIA
 551         if (!atomic_read(&dquot->dq_count)) {
 552                 printk("VFS: dqput: trying to free free dquot\n");
 553                 printk("VFS: device %s, dquot of %s %d\n",
 554                         dquot->dq_sb->s_id,
 555                         quotatypes[dquot->dq_type],
 556                         dquot->dq_id);
 557                 BUG();
 558         }
 559 #endif
 560
 561         spin_lock(&dq_list_lock);
 562         dqstats.drops++;
 563         spin_unlock(&dq_list_lock);
 564 we_slept:
 565         spin_lock(&dq_list_lock);
 566         if (atomic_read(&dquot->dq_count) > 1) {
 567                 /* We have more than one user... nothing to do */
 568                 atomic_dec(&dquot->dq_count);
 569                 /* Releasing dquot during quotaoff phase? */
 570                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
 571                     atomic_read(&dquot->dq_count) == 1)
 572                         wake_up(&dquot->dq_wait_unused);
 573                 spin_unlock(&dq_list_lock);
 574                 return;
 575         }
 576         /* Need to release dquot? */
 577         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 578                 spin_unlock(&dq_list_lock);
 579                 /* Commit dquot before releasing */
 580                 dquot->dq_sb->dq_op->write_dquot(dquot);
 581                 goto we_slept;
 582         }
 583         /* Clear flag in case dquot was inactive (something bad happened) */
 584         clear_dquot_dirty(dquot);
 585         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 586                 spin_unlock(&dq_list_lock);
 587                 dquot->dq_sb->dq_op->release_dquot(dquot);
 588                 goto we_slept;
 589         }
 590         atomic_dec(&dquot->dq_count);
 591 #ifdef __DQUOT_PARANOIA
 592         /* sanity check */
 593         BUG_ON(!list_empty(&dquot->dq_free));
 594 #endif
 595         put_dquot_last(dquot);
 596         spin_unlock(&dq_list_lock);
 597 }
 598
 599 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 600 {
 601         struct dquot *dquot;
 602
 603         dquot = kmem_cache_alloc(dquot_cachep, GFP_NOFS);
 604         if(!dquot)
 605                 return NODQUOT;
 606
 607         memset((caddr_t)dquot, 0, sizeof(struct dquot));
 608         mutex_init(&dquot->dq_lock);
 609         INIT_LIST_HEAD(&dquot->dq_free);
 610         INIT_LIST_HEAD(&dquot->dq_inuse);
 611         INIT_HLIST_NODE(&dquot->dq_hash);
 612         INIT_LIST_HEAD(&dquot->dq_dirty);
 613         init_waitqueue_head(&dquot->dq_wait_unused);
 614         dquot->dq_sb = sb;
 615         dquot->dq_type = type;
 616         atomic_set(&dquot->dq_count, 1);
 617
 618         return dquot;
 619 }
 620
 621 /*
 622  * Get reference to dquot
 623  * MUST be called with either dqptr_sem or dqonoff_mutex held
 624  */
 625 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 626 {
 627         unsigned int hashent = hashfn(sb, id, type);
 628         struct dquot *dquot, *empty = NODQUOT;
 629
 630         if (!sb_has_quota_enabled(sb, type))
 631                 return NODQUOT;
 632 we_slept:
 633         spin_lock(&dq_list_lock);
 634         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 635                 if (empty == NODQUOT) {
 636                         spin_unlock(&dq_list_lock);
 637                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 638                                 schedule();     /* Try to wait for a moment... */
 639                         goto we_slept;
 640                 }
 641                 dquot = empty;
 642                 dquot->dq_id = id;
 643                 /* all dquots go on the inuse_list */
 644                 put_inuse(dquot);
 645                 /* hash it first so it can be found */
 646                 insert_dquot_hash(dquot);
 647                 dqstats.lookups++;
 648                 spin_unlock(&dq_list_lock);
 649         } else {
 650                 if (!atomic_read(&dquot->dq_count))
 651                         remove_free_dquot(dquot);
 652                 atomic_inc(&dquot->dq_count);
 653                 dqstats.cache_hits++;
 654                 dqstats.lookups++;
 655                 spin_unlock(&dq_list_lock);
 656                 if (empty)
 657                         kmem_cache_free(dquot_cachep, empty);
 658         }
 659         /* Wait for dq_lock - after this we know that either dquot_release() is already
 660          * finished or it will be canceled due to dq_count > 1 test */
 661         wait_on_dquot(dquot);
 662         /* Read the dquot and instantiate it (everything done only if needed) */
 663         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 664                 dqput(dquot);
 665                 return NODQUOT;
 666         }
 667 #ifdef __DQUOT_PARANOIA
 668         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 669 #endif
 670
 671         return dquot;
 672 }
 673
 674 static int dqinit_needed(struct inode *inode, int type)
 675 {
 676         int cnt;
 677
 678         if (IS_NOQUOTA(inode))
 679                 return 0;
 680         if (type != -1)
 681                 return inode->i_dquot[type] == NODQUOT;
 682         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 683                 if (inode->i_dquot[cnt] == NODQUOT)
 684                         return 1;
 685         return 0;
 686 }
 687
 688 /* This routine is guarded by dqonoff_mutex mutex */
 689 static void add_dquot_ref(struct super_block *sb, int type)
 690 {
 691         struct list_head *p;
 692
 693 restart:
 694         file_list_lock();
 695         list_for_each(p, &sb->s_files) {
 696                 struct file *filp = list_entry(p, struct file, f_u.fu_list);
 697                 struct inode *inode = filp->f_path.dentry->d_inode;
 698                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
 699                         struct dentry *dentry = dget(filp->f_path.dentry);
 700                         file_list_unlock();
 701                         sb->dq_op->initialize(inode, type);
 702                         dput(dentry);
 703                         /* As we may have blocked we had better restart... */
 704                         goto restart;
 705                 }
 706         }
 707         file_list_unlock();
 708 }
 709
 710 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 711 static inline int dqput_blocks(struct dquot *dquot)
 712 {
 713         if (atomic_read(&dquot->dq_count) <= 1)
 714                 return 1;
 715         return 0;
 716 }
 717
 718 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 719 /* We can't race with anybody because we hold dqptr_sem for writing... */
 720 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
 721 {
 722         struct dquot *dquot = inode->i_dquot[type];
 723
 724         inode->i_dquot[type] = NODQUOT;
 725         if (dquot != NODQUOT) {
 726                 if (dqput_blocks(dquot)) {
 727 #ifdef __DQUOT_PARANOIA
 728                         if (atomic_read(&dquot->dq_count) != 1)
 729                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 730 #endif
 731                         spin_lock(&dq_list_lock);
 732                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 733                         spin_unlock(&dq_list_lock);
 734                         return 1;
 735                 }
 736                 else
 737                         dqput(dquot);   /* We have guaranteed we won't block */
 738         }
 739         return 0;
 740 }
 741
 742 /* Free list of dquots - called from inode.c */
 743 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 744 static void put_dquot_list(struct list_head *tofree_head)
 745 {
 746         struct list_head *act_head;
 747         struct dquot *dquot;
 748
 749         act_head = tofree_head->next;
 750         /* So now we have dquots on the list... Just free them */
 751         while (act_head != tofree_head) {
 752                 dquot = list_entry(act_head, struct dquot, dq_free);
 753                 act_head = act_head->next;
 754                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 755                 dqput(dquot);
 756         }
 757 }
 758
 759 /* Gather all references from inodes and drop them */
 760 static void drop_dquot_ref(struct super_block *sb, int type)
 761 {
 762         LIST_HEAD(tofree_head);
 763
 764         down_write(&sb_dqopt(sb)->dqptr_sem);
 765         remove_dquot_ref(sb, type, &tofree_head);
 766         up_write(&sb_dqopt(sb)->dqptr_sem);
 767         put_dquot_list(&tofree_head);
 768 }
 769
 770 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
 771 {
 772         dquot->dq_dqb.dqb_curinodes += number;
 773 }
 774
 775 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 776 {
 777         dquot->dq_dqb.dqb_curspace += number;
 778 }
 779
 780 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
 781 {
 782         if (dquot->dq_dqb.dqb_curinodes > number)
 783                 dquot->dq_dqb.dqb_curinodes -= number;
 784         else
 785                 dquot->dq_dqb.dqb_curinodes = 0;
 786         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
 787                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 788         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 789 }
 790
 791 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 792 {
 793         if (dquot->dq_dqb.dqb_curspace > number)
 794                 dquot->dq_dqb.dqb_curspace -= number;
 795         else
 796                 dquot->dq_dqb.dqb_curspace = 0;
 797         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
 798                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 799         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 800 }
 801
 802 static int flag_print_warnings = 1;
 803
 804 static inline int need_print_warning(struct dquot *dquot)
 805 {
 806         if (!flag_print_warnings)
 807                 return 0;
 808
 809         switch (dquot->dq_type) {
 810                 case USRQUOTA:
 811                         return current->fsuid == dquot->dq_id;
 812                 case GRPQUOTA:
 813                         return in_group_p(dquot->dq_id);
 814         }
 815         return 0;
 816 }
 817
 818 /* Values of warnings */
 819 #define NOWARN 0
 820 #define IHARDWARN 1
 821 #define ISOFTLONGWARN 2
 822 #define ISOFTWARN 3
 823 #define BHARDWARN 4
 824 #define BSOFTLONGWARN 5
 825 #define BSOFTWARN 6
 826
 827 /* Print warning to user which exceeded quota */
 828 static void print_warning(struct dquot *dquot, const char warntype)
 829 {
 830         char *msg = NULL;
 831         struct tty_struct *tty;
 832         int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
 833           ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
 834
 835         if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
 836                 return;
 837
 838         mutex_lock(&tty_mutex);
 839         tty = get_current_tty();
 840         if (!tty)
 841                 goto out_lock;
 842         tty_write_message(tty, dquot->dq_sb->s_id);
 843         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
 844                 tty_write_message(tty, ": warning, ");
 845         else
 846                 tty_write_message(tty, ": write failed, ");
 847         tty_write_message(tty, quotatypes[dquot->dq_type]);
 848         switch (warntype) {
 849                 case IHARDWARN:
 850                         msg = " file limit reached.\r\n";
 851                         break;
 852                 case ISOFTLONGWARN:
 853                         msg = " file quota exceeded too long.\r\n";
 854                         break;
 855                 case ISOFTWARN:
 856                         msg = " file quota exceeded.\r\n";
 857                         break;
 858                 case BHARDWARN:
 859                         msg = " block limit reached.\r\n";
 860                         break;
 861                 case BSOFTLONGWARN:
 862                         msg = " block quota exceeded too long.\r\n";
 863                         break;
 864                 case BSOFTWARN:
 865                         msg = " block quota exceeded.\r\n";
 866                         break;
 867         }
 868         tty_write_message(tty, msg);
 869 out_lock:
 870         mutex_unlock(&tty_mutex);
 871 }
 872
 873 static inline void flush_warnings(struct dquot **dquots, char *warntype)
 874 {
 875         int i;
 876
 877         for (i = 0; i < MAXQUOTAS; i++)
 878                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
 879                         print_warning(dquots[i], warntype[i]);
 880 }
 881
 882 static inline char ignore_hardlimit(struct dquot *dquot)
 883 {
 884         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
 885
 886         return capable(CAP_SYS_RESOURCE) &&
 887             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
 888 }
 889
 890 /* needs dq_data_lock */
 891 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
 892 {
 893         *warntype = NOWARN;
 894         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 895                 return QUOTA_OK;
 896
 897         if (dquot->dq_dqb.dqb_ihardlimit &&
 898            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
 899             !ignore_hardlimit(dquot)) {
 900                 *warntype = IHARDWARN;
 901                 return NO_QUOTA;
 902         }
 903
 904         if (dquot->dq_dqb.dqb_isoftlimit &&
 905            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 906             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
 907             !ignore_hardlimit(dquot)) {
 908                 *warntype = ISOFTLONGWARN;
 909                 return NO_QUOTA;
 910         }
 911
 912         if (dquot->dq_dqb.dqb_isoftlimit &&
 913            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 914             dquot->dq_dqb.dqb_itime == 0) {
 915                 *warntype = ISOFTWARN;
 916                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
 917         }
 918
 919         return QUOTA_OK;
 920 }
 921
 922 /* needs dq_data_lock */
 923 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
 924 {
 925         *warntype = 0;
 926         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 927                 return QUOTA_OK;
 928
 929         if (dquot->dq_dqb.dqb_bhardlimit &&
 930            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
 931             !ignore_hardlimit(dquot)) {
 932                 if (!prealloc)
 933                         *warntype = BHARDWARN;
 934                 return NO_QUOTA;
 935         }
 936
 937         if (dquot->dq_dqb.dqb_bsoftlimit &&
 938            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 939             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
 940             !ignore_hardlimit(dquot)) {
 941                 if (!prealloc)
 942                         *warntype = BSOFTLONGWARN;
 943                 return NO_QUOTA;
 944         }
 945
 946         if (dquot->dq_dqb.dqb_bsoftlimit &&
 947            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 948             dquot->dq_dqb.dqb_btime == 0) {
 949                 if (!prealloc) {
 950                         *warntype = BSOFTWARN;
 951                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
 952                 }
 953                 else
 954                         /*
 955                          * We don't allow preallocation to exceed softlimit so exceeding will
 956                          * be always printed
 957                          */
 958                         return NO_QUOTA;
 959         }
 960
 961         return QUOTA_OK;
 962 }
 963
 964 /*
 965  *      Initialize quota pointers in inode
 966  *      Transaction must be started at entry
 967  */
 968 int dquot_initialize(struct inode *inode, int type)
 969 {
 970         unsigned int id = 0;
 971         int cnt, ret = 0;
 972
 973         /* First test before acquiring mutex - solves deadlocks when we
 974          * re-enter the quota code and are already holding the mutex */
 975         if (IS_NOQUOTA(inode))
 976                 return 0;
 977         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 978         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
 979         if (IS_NOQUOTA(inode))
 980                 goto out_err;
 981         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 982                 if (type != -1 && cnt != type)
 983                         continue;
 984                 if (inode->i_dquot[cnt] == NODQUOT) {
 985                         switch (cnt) {
 986                                 case USRQUOTA:
 987                                         id = inode->i_uid;
 988                                         break;
 989                                 case GRPQUOTA:
 990                                         id = inode->i_gid;
 991                                         break;
 992                         }
 993                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
 994                 }
 995         }
 996 out_err:
 997         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 998         return ret;
 999 }
1000
1001 /*
1002  *      Release all quotas referenced by inode
1003  *      Transaction must be started at an entry
1004  */
1005 int dquot_drop(struct inode *inode)
1006 {
1007         int cnt;
1008
1009         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1010         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1011                 if (inode->i_dquot[cnt] != NODQUOT) {
1012                         dqput(inode->i_dquot[cnt]);
1013                         inode->i_dquot[cnt] = NODQUOT;
1014                 }
1015         }
1016         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1017         return 0;
1018 }
1019
1020 /*
1021  * Following four functions update i_blocks+i_bytes fields and
1022  * quota information (together with appropriate checks)
1023  * NOTE: We absolutely rely on the fact that caller dirties
1024  * the inode (usually macros in quotaops.h care about this) and
1025  * holds a handle for the current transaction so that dquot write and
1026  * inode write go into the same transaction.
1027  */
1028
1029 /*
1030  * This operation can block, but only after everything is updated
1031  */
1032 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1033 {
1034         int cnt, ret = NO_QUOTA;
1035         char warntype[MAXQUOTAS];
1036
1037         /* First test before acquiring mutex - solves deadlocks when we
1038          * re-enter the quota code and are already holding the mutex */
1039         if (IS_NOQUOTA(inode)) {
1040 out_add:
1041                 inode_add_bytes(inode, number);
1042                 return QUOTA_OK;
1043         }
1044         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1045                 warntype[cnt] = NOWARN;
1046
1047         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1048         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1049                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1050                 goto out_add;
1051         }
1052         spin_lock(&dq_data_lock);
1053         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1054                 if (inode->i_dquot[cnt] == NODQUOT)
1055                         continue;
1056                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1057                         goto warn_put_all;
1058         }
1059         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1060                 if (inode->i_dquot[cnt] == NODQUOT)
1061                         continue;
1062                 dquot_incr_space(inode->i_dquot[cnt], number);
1063         }
1064         inode_add_bytes(inode, number);
1065         ret = QUOTA_OK;
1066 warn_put_all:
1067         spin_unlock(&dq_data_lock);
1068         if (ret == QUOTA_OK)
1069                 /* Dirtify all the dquots - this can block when journalling */
1070                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1071                         if (inode->i_dquot[cnt])
1072                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1073         flush_warnings(inode->i_dquot, warntype);
1074         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1075         return ret;
1076 }
1077
1078 /*
1079  * This operation can block, but only after everything is updated
1080  */
1081 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1082 {
1083         int cnt, ret = NO_QUOTA;
1084         char warntype[MAXQUOTAS];
1085
1086         /* First test before acquiring mutex - solves deadlocks when we
1087          * re-enter the quota code and are already holding the mutex */
1088         if (IS_NOQUOTA(inode))
1089                 return QUOTA_OK;
1090         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1091                 warntype[cnt] = NOWARN;
1092         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1093         if (IS_NOQUOTA(inode)) {
1094                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1095                 return QUOTA_OK;
1096         }
1097         spin_lock(&dq_data_lock);
1098         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1099                 if (inode->i_dquot[cnt] == NODQUOT)
1100                         continue;
1101                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1102                         goto warn_put_all;
1103         }
1104
1105         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1106                 if (inode->i_dquot[cnt] == NODQUOT)
1107                         continue;
1108                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1109         }
1110         ret = QUOTA_OK;
1111 warn_put_all:
1112         spin_unlock(&dq_data_lock);
1113         if (ret == QUOTA_OK)
1114                 /* Dirtify all the dquots - this can block when journalling */
1115                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1116                         if (inode->i_dquot[cnt])
1117                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1118         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1119         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1120         return ret;
1121 }
1122
1123 /*
1124  * This operation can block, but only after everything is updated
1125  */
1126 int dquot_free_space(struct inode *inode, qsize_t number)
1127 {
1128         unsigned int cnt;
1129
1130         /* First test before acquiring mutex - solves deadlocks when we
1131          * re-enter the quota code and are already holding the mutex */
1132         if (IS_NOQUOTA(inode)) {
1133 out_sub:
1134                 inode_sub_bytes(inode, number);
1135                 return QUOTA_OK;
1136         }
1137         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1138         /* Now recheck reliably when holding dqptr_sem */
1139         if (IS_NOQUOTA(inode)) {
1140                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1141                 goto out_sub;
1142         }
1143         spin_lock(&dq_data_lock);
1144         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1145                 if (inode->i_dquot[cnt] == NODQUOT)
1146                         continue;
1147                 dquot_decr_space(inode->i_dquot[cnt], number);
1148         }
1149         inode_sub_bytes(inode, number);
1150         spin_unlock(&dq_data_lock);
1151         /* Dirtify all the dquots - this can block when journalling */
1152         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1153                 if (inode->i_dquot[cnt])
1154                         mark_dquot_dirty(inode->i_dquot[cnt]);
1155         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1156         return QUOTA_OK;
1157 }
1158
1159 /*
1160  * This operation can block, but only after everything is updated
1161  */
1162 int dquot_free_inode(const struct inode *inode, unsigned long number)
1163 {
1164         unsigned int cnt;
1165
1166         /* First test before acquiring mutex - solves deadlocks when we
1167          * re-enter the quota code and are already holding the mutex */
1168         if (IS_NOQUOTA(inode))
1169                 return QUOTA_OK;
1170         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1171         /* Now recheck reliably when holding dqptr_sem */
1172         if (IS_NOQUOTA(inode)) {
1173                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1174                 return QUOTA_OK;
1175         }
1176         spin_lock(&dq_data_lock);
1177         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1178                 if (inode->i_dquot[cnt] == NODQUOT)
1179                         continue;
1180                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1181         }
1182         spin_unlock(&dq_data_lock);
1183         /* Dirtify all the dquots - this can block when journalling */
1184         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1185                 if (inode->i_dquot[cnt])
1186                         mark_dquot_dirty(inode->i_dquot[cnt]);
1187         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1188         return QUOTA_OK;
1189 }
1190
1191 /*
1192  * Transfer the number of inode and blocks from one diskquota to an other.
1193  *
1194  * This operation can block, but only after everything is updated
1195  * A transaction must be started when entering this function.
1196  */
1197 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1198 {
1199         qsize_t space;
1200         struct dquot *transfer_from[MAXQUOTAS];
1201         struct dquot *transfer_to[MAXQUOTAS];
1202         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1203             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1204         char warntype[MAXQUOTAS];
1205
1206         /* First test before acquiring mutex - solves deadlocks when we
1207          * re-enter the quota code and are already holding the mutex */
1208         if (IS_NOQUOTA(inode))
1209                 return QUOTA_OK;
1210         /* Clear the arrays */
1211         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1212                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1213                 warntype[cnt] = NOWARN;
1214         }
1215         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1216         /* Now recheck reliably when holding dqptr_sem */
1217         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1218                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1219                 return QUOTA_OK;
1220         }
1221         /* First build the transfer_to list - here we can block on
1222          * reading/instantiating of dquots.  We know that the transaction for
1223          * us was already started so we don't violate lock ranking here */
1224         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1225                 switch (cnt) {
1226                         case USRQUOTA:
1227                                 if (!chuid)
1228                                         continue;
1229                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1230                                 break;
1231                         case GRPQUOTA:
1232                                 if (!chgid)
1233                                         continue;
1234                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1235                                 break;
1236                 }
1237         }
1238         spin_lock(&dq_data_lock);
1239         space = inode_get_bytes(inode);
1240         /* Build the transfer_from list and check the limits */
1241         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1242                 if (transfer_to[cnt] == NODQUOT)
1243                         continue;
1244                 transfer_from[cnt] = inode->i_dquot[cnt];
1245                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1246                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1247                         goto warn_put_all;
1248         }
1249
1250         /*
1251          * Finally perform the needed transfer from transfer_from to transfer_to
1252          */
1253         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1254                 /*
1255                  * Skip changes for same uid or gid or for turned off quota-type.
1256                  */
1257                 if (transfer_to[cnt] == NODQUOT)
1258                         continue;
1259
1260                 /* Due to IO error we might not have transfer_from[] structure */
1261                 if (transfer_from[cnt]) {
1262                         dquot_decr_inodes(transfer_from[cnt], 1);
1263                         dquot_decr_space(transfer_from[cnt], space);
1264                 }
1265
1266                 dquot_incr_inodes(transfer_to[cnt], 1);
1267                 dquot_incr_space(transfer_to[cnt], space);
1268
1269                 inode->i_dquot[cnt] = transfer_to[cnt];
1270         }
1271         ret = QUOTA_OK;
1272 warn_put_all:
1273         spin_unlock(&dq_data_lock);
1274         /* Dirtify all the dquots - this can block when journalling */
1275         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1276                 if (transfer_from[cnt])
1277                         mark_dquot_dirty(transfer_from[cnt]);
1278                 if (transfer_to[cnt])
1279                         mark_dquot_dirty(transfer_to[cnt]);
1280         }
1281         flush_warnings(transfer_to, warntype);
1282
1283         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1284                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1285                         dqput(transfer_from[cnt]);
1286                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1287                         dqput(transfer_to[cnt]);
1288         }
1289         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1290         return ret;
1291 }
1292
1293 /*
1294  * Write info of quota file to disk
1295  */
1296 int dquot_commit_info(struct super_block *sb, int type)
1297 {
1298         int ret;
1299         struct quota_info *dqopt = sb_dqopt(sb);
1300
1301         mutex_lock(&dqopt->dqio_mutex);
1302         ret = dqopt->ops[type]->write_file_info(sb, type);
1303         mutex_unlock(&dqopt->dqio_mutex);
1304         return ret;
1305 }
1306
1307 /*
1308  * Definitions of diskquota operations.
1309  */
1310 struct dquot_operations dquot_operations = {
1311         .initialize     = dquot_initialize,
1312         .drop           = dquot_drop,
1313         .alloc_space    = dquot_alloc_space,
1314         .alloc_inode    = dquot_alloc_inode,
1315         .free_space     = dquot_free_space,
1316         .free_inode     = dquot_free_inode,
1317         .transfer       = dquot_transfer,
1318         .write_dquot    = dquot_commit,
1319         .acquire_dquot  = dquot_acquire,
1320         .release_dquot  = dquot_release,
1321         .mark_dirty     = dquot_mark_dquot_dirty,
1322         .write_info     = dquot_commit_info
1323 };
1324
1325 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1326 {
1327         switch (type) {
1328                 case USRQUOTA:
1329                         dqopt->flags |= DQUOT_USR_ENABLED;
1330                         break;
1331                 case GRPQUOTA:
1332                         dqopt->flags |= DQUOT_GRP_ENABLED;
1333                         break;
1334         }
1335 }
1336
1337 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1338 {
1339         switch (type) {
1340                 case USRQUOTA:
1341                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1342                         break;
1343                 case GRPQUOTA:
1344                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1345                         break;
1346         }
1347 }
1348
1349 /*
1350  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1351  */
1352 int vfs_quota_off(struct super_block *sb, int type)
1353 {
1354         int cnt;
1355         struct quota_info *dqopt = sb_dqopt(sb);
1356         struct inode *toputinode[MAXQUOTAS];
1357
1358         /* We need to serialize quota_off() for device */
1359         mutex_lock(&dqopt->dqonoff_mutex);
1360         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1361                 toputinode[cnt] = NULL;
1362                 if (type != -1 && cnt != type)
1363                         continue;
1364                 if (!sb_has_quota_enabled(sb, cnt))
1365                         continue;
1366                 reset_enable_flags(dqopt, cnt);
1367
1368                 /* Note: these are blocking operations */
1369                 drop_dquot_ref(sb, cnt);
1370                 invalidate_dquots(sb, cnt);
1371                 /*
1372                  * Now all dquots should be invalidated, all writes done so we should be only
1373                  * users of the info. No locks needed.
1374                  */
1375                 if (info_dirty(&dqopt->info[cnt]))
1376                         sb->dq_op->write_info(sb, cnt);
1377                 if (dqopt->ops[cnt]->free_file_info)
1378                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1379                 put_quota_format(dqopt->info[cnt].dqi_format);
1380
1381                 toputinode[cnt] = dqopt->files[cnt];
1382                 dqopt->files[cnt] = NULL;
1383                 dqopt->info[cnt].dqi_flags = 0;
1384                 dqopt->info[cnt].dqi_igrace = 0;
1385                 dqopt->info[cnt].dqi_bgrace = 0;
1386                 dqopt->ops[cnt] = NULL;
1387         }
1388         mutex_unlock(&dqopt->dqonoff_mutex);
1389         /* Sync the superblock so that buffers with quota data are written to
1390          * disk (and so userspace sees correct data afterwards). */
1391         if (sb->s_op->sync_fs)
1392                 sb->s_op->sync_fs(sb, 1);
1393         sync_blockdev(sb->s_bdev);
1394         /* Now the quota files are just ordinary files and we can set the
1395          * inode flags back. Moreover we discard the pagecache so that
1396          * userspace sees the writes we did bypassing the pagecache. We
1397          * must also discard the blockdev buffers so that we see the
1398          * changes done by userspace on the next quotaon() */
1399         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1400                 if (toputinode[cnt]) {
1401                         mutex_lock(&dqopt->dqonoff_mutex);
1402                         /* If quota was reenabled in the meantime, we have
1403                          * nothing to do */
1404                         if (!sb_has_quota_enabled(sb, cnt)) {
1405                                 mutex_lock(&toputinode[cnt]->i_mutex);
1406                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1407                                   S_NOATIME | S_NOQUOTA);
1408                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1409                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1410                                 mark_inode_dirty(toputinode[cnt]);
1411                                 iput(toputinode[cnt]);
1412                         }
1413                         mutex_unlock(&dqopt->dqonoff_mutex);
1414                 }
1415         if (sb->s_bdev)
1416                 invalidate_bdev(sb->s_bdev, 0);
1417         return 0;
1418 }
1419
1420 /*
1421  *      Turn quotas on on a device
1422  */
1423
1424 /* Helper function when we already have the inode */
1425 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1426 {
1427         struct quota_format_type *fmt = find_quota_format(format_id);
1428         struct super_block *sb = inode->i_sb;
1429         struct quota_info *dqopt = sb_dqopt(sb);
1430         int error;
1431         int oldflags = -1;
1432
1433         if (!fmt)
1434                 return -ESRCH;
1435         if (!S_ISREG(inode->i_mode)) {
1436                 error = -EACCES;
1437                 goto out_fmt;
1438         }
1439         if (IS_RDONLY(inode)) {
1440                 error = -EROFS;
1441                 goto out_fmt;
1442         }
1443         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1444                 error = -EINVAL;
1445                 goto out_fmt;
1446         }
1447
1448         /* As we bypass the pagecache we must now flush the inode so that
1449          * we see all the changes from userspace... */
1450         write_inode_now(inode, 1);
1451         /* And now flush the block cache so that kernel sees the changes */
1452         invalidate_bdev(sb->s_bdev, 0);
1453         mutex_lock(&inode->i_mutex);
1454         mutex_lock(&dqopt->dqonoff_mutex);
1455         if (sb_has_quota_enabled(sb, type)) {
1456                 error = -EBUSY;
1457                 goto out_lock;
1458         }
1459         /* We don't want quota and atime on quota files (deadlocks possible)
1460          * Also nobody should write to the file - we use special IO operations
1461          * which ignore the immutable bit. */
1462         down_write(&dqopt->dqptr_sem);
1463         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1464         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1465         up_write(&dqopt->dqptr_sem);
1466         sb->dq_op->drop(inode);
1467
1468         error = -EIO;
1469         dqopt->files[type] = igrab(inode);
1470         if (!dqopt->files[type])
1471                 goto out_lock;
1472         error = -EINVAL;
1473         if (!fmt->qf_ops->check_quota_file(sb, type))
1474                 goto out_file_init;
1475
1476         dqopt->ops[type] = fmt->qf_ops;
1477         dqopt->info[type].dqi_format = fmt;
1478         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1479         mutex_lock(&dqopt->dqio_mutex);
1480         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1481                 mutex_unlock(&dqopt->dqio_mutex);
1482                 goto out_file_init;
1483         }
1484         mutex_unlock(&dqopt->dqio_mutex);
1485         mutex_unlock(&inode->i_mutex);
1486         set_enable_flags(dqopt, type);
1487
1488         add_dquot_ref(sb, type);
1489         mutex_unlock(&dqopt->dqonoff_mutex);
1490
1491         return 0;
1492
1493 out_file_init:
1494         dqopt->files[type] = NULL;
1495         iput(inode);
1496 out_lock:
1497         mutex_unlock(&dqopt->dqonoff_mutex);
1498         if (oldflags != -1) {
1499                 down_write(&dqopt->dqptr_sem);
1500                 /* Set the flags back (in the case of accidental quotaon()
1501                  * on a wrong file we don't want to mess up the flags) */
1502                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1503                 inode->i_flags |= oldflags;
1504                 up_write(&dqopt->dqptr_sem);
1505         }
1506         mutex_unlock(&inode->i_mutex);
1507 out_fmt:
1508         put_quota_format(fmt);
1509
1510         return error;
1511 }
1512
1513 /* Actual function called from quotactl() */
1514 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1515 {
1516         struct nameidata nd;
1517         int error;
1518
1519         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1520         if (error < 0)
1521                 return error;
1522         error = security_quota_on(nd.dentry);
1523         if (error)
1524                 goto out_path;
1525         /* Quota file not on the same filesystem? */
1526         if (nd.mnt->mnt_sb != sb)
1527                 error = -EXDEV;
1528         else
1529                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1530 out_path:
1531         path_release(&nd);
1532         return error;
1533 }
1534
1535 /*
1536  * This function is used when filesystem needs to initialize quotas
1537  * during mount time.
1538  */
1539 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1540                 int format_id, int type)
1541 {
1542         struct dentry *dentry;
1543         int error;
1544
1545         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1546         if (IS_ERR(dentry))
1547                 return PTR_ERR(dentry);
1548
1549         if (!dentry->d_inode) {
1550                 error = -ENOENT;
1551                 goto out;
1552         }
1553
1554         error = security_quota_on(dentry);
1555         if (!error)
1556                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1557
1558 out:
1559         dput(dentry);
1560         return error;
1561 }
1562
1563 /* Generic routine for getting common part of quota structure */
1564 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1565 {
1566         struct mem_dqblk *dm = &dquot->dq_dqb;
1567
1568         spin_lock(&dq_data_lock);
1569         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1570         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1571         di->dqb_curspace = dm->dqb_curspace;
1572         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1573         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1574         di->dqb_curinodes = dm->dqb_curinodes;
1575         di->dqb_btime = dm->dqb_btime;
1576         di->dqb_itime = dm->dqb_itime;
1577         di->dqb_valid = QIF_ALL;
1578         spin_unlock(&dq_data_lock);
1579 }
1580
1581 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1582 {
1583         struct dquot *dquot;
1584
1585         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1586         if (!(dquot = dqget(sb, id, type))) {
1587                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1588                 return -ESRCH;
1589         }
1590         do_get_dqblk(dquot, di);
1591         dqput(dquot);
1592         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1593         return 0;
1594 }
1595
1596 /* Generic routine for setting common part of quota structure */
1597 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1598 {
1599         struct mem_dqblk *dm = &dquot->dq_dqb;
1600         int check_blim = 0, check_ilim = 0;
1601
1602         spin_lock(&dq_data_lock);
1603         if (di->dqb_valid & QIF_SPACE) {
1604                 dm->dqb_curspace = di->dqb_curspace;
1605                 check_blim = 1;
1606         }
1607         if (di->dqb_valid & QIF_BLIMITS) {
1608                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1609                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1610                 check_blim = 1;
1611         }
1612         if (di->dqb_valid & QIF_INODES) {
1613                 dm->dqb_curinodes = di->dqb_curinodes;
1614                 check_ilim = 1;
1615         }
1616         if (di->dqb_valid & QIF_ILIMITS) {
1617                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1618                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1619                 check_ilim = 1;
1620         }
1621         if (di->dqb_valid & QIF_BTIME)
1622                 dm->dqb_btime = di->dqb_btime;
1623         if (di->dqb_valid & QIF_ITIME)
1624                 dm->dqb_itime = di->dqb_itime;
1625
1626         if (check_blim) {
1627                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1628                         dm->dqb_btime = 0;
1629                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1630                 }
1631                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1632                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1633         }
1634         if (check_ilim) {
1635                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1636                         dm->dqb_itime = 0;
1637                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1638                 }
1639                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1640                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1641         }
1642         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1643                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1644         else
1645                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1646         spin_unlock(&dq_data_lock);
1647         mark_dquot_dirty(dquot);
1648 }
1649
1650 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1651 {
1652         struct dquot *dquot;
1653
1654         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1655         if (!(dquot = dqget(sb, id, type))) {
1656                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1657                 return -ESRCH;
1658         }
1659         do_set_dqblk(dquot, di);
1660         dqput(dquot);
1661         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1662         return 0;
1663 }
1664
1665 /* Generic routine for getting common part of quota file information */
1666 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1667 {
1668         struct mem_dqinfo *mi;
1669
1670         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1671         if (!sb_has_quota_enabled(sb, type)) {
1672                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1673                 return -ESRCH;
1674         }
1675         mi = sb_dqopt(sb)->info + type;
1676         spin_lock(&dq_data_lock);
1677         ii->dqi_bgrace = mi->dqi_bgrace;
1678         ii->dqi_igrace = mi->dqi_igrace;
1679         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1680         ii->dqi_valid = IIF_ALL;
1681         spin_unlock(&dq_data_lock);
1682         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1683         return 0;
1684 }
1685
1686 /* Generic routine for setting common part of quota file information */
1687 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1688 {
1689         struct mem_dqinfo *mi;
1690
1691         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1692         if (!sb_has_quota_enabled(sb, type)) {
1693                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1694                 return -ESRCH;
1695         }
1696         mi = sb_dqopt(sb)->info + type;
1697         spin_lock(&dq_data_lock);
1698         if (ii->dqi_valid & IIF_BGRACE)
1699                 mi->dqi_bgrace = ii->dqi_bgrace;
1700         if (ii->dqi_valid & IIF_IGRACE)
1701                 mi->dqi_igrace = ii->dqi_igrace;
1702         if (ii->dqi_valid & IIF_FLAGS)
1703                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1704         spin_unlock(&dq_data_lock);
1705         mark_info_dirty(sb, type);
1706         /* Force write to disk */
1707         sb->dq_op->write_info(sb, type);
1708         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1709         return 0;
1710 }
1711
1712 struct quotactl_ops vfs_quotactl_ops = {
1713         .quota_on       = vfs_quota_on,
1714         .quota_off      = vfs_quota_off,
1715         .quota_sync     = vfs_quota_sync,
1716         .get_info       = vfs_get_dqinfo,
1717         .set_info       = vfs_set_dqinfo,
1718         .get_dqblk      = vfs_get_dqblk,
1719         .set_dqblk      = vfs_set_dqblk
1720 };
1721
1722 static ctl_table fs_dqstats_table[] = {
1723         {
1724                 .ctl_name       = FS_DQ_LOOKUPS,
1725                 .procname       = "lookups",
1726                 .data           = &dqstats.lookups,
1727                 .maxlen         = sizeof(int),
1728                 .mode           = 0444,
1729                 .proc_handler   = &proc_dointvec,
1730         },
1731         {
1732                 .ctl_name       = FS_DQ_DROPS,
1733                 .procname       = "drops",
1734                 .data           = &dqstats.drops,
1735                 .maxlen         = sizeof(int),
1736                 .mode           = 0444,
1737                 .proc_handler   = &proc_dointvec,
1738         },
1739         {
1740                 .ctl_name       = FS_DQ_READS,
1741                 .procname       = "reads",
1742                 .data           = &dqstats.reads,
1743                 .maxlen         = sizeof(int),
1744                 .mode           = 0444,
1745                 .proc_handler   = &proc_dointvec,
1746         },
1747         {
1748                 .ctl_name       = FS_DQ_WRITES,
1749                 .procname       = "writes",
1750                 .data           = &dqstats.writes,
1751                 .maxlen         = sizeof(int),
1752                 .mode           = 0444,
1753                 .proc_handler   = &proc_dointvec,
1754         },
1755         {
1756                 .ctl_name       = FS_DQ_CACHE_HITS,
1757                 .procname       = "cache_hits",
1758                 .data           = &dqstats.cache_hits,
1759                 .maxlen         = sizeof(int),
1760                 .mode           = 0444,
1761                 .proc_handler   = &proc_dointvec,
1762         },
1763         {
1764                 .ctl_name       = FS_DQ_ALLOCATED,
1765                 .procname       = "allocated_dquots",
1766                 .data           = &dqstats.allocated_dquots,
1767                 .maxlen         = sizeof(int),
1768                 .mode           = 0444,
1769                 .proc_handler   = &proc_dointvec,
1770         },
1771         {
1772                 .ctl_name       = FS_DQ_FREE,
1773                 .procname       = "free_dquots",
1774                 .data           = &dqstats.free_dquots,
1775                 .maxlen         = sizeof(int),
1776                 .mode           = 0444,
1777                 .proc_handler   = &proc_dointvec,
1778         },
1779         {
1780                 .ctl_name       = FS_DQ_SYNCS,
1781                 .procname       = "syncs",
1782                 .data           = &dqstats.syncs,
1783                 .maxlen         = sizeof(int),
1784                 .mode           = 0444,
1785                 .proc_handler   = &proc_dointvec,
1786         },
1787         {
1788                 .ctl_name       = FS_DQ_WARNINGS,
1789                 .procname       = "warnings",
1790                 .data           = &flag_print_warnings,
1791                 .maxlen         = sizeof(int),
1792                 .mode           = 0644,
1793                 .proc_handler   = &proc_dointvec,
1794         },
1795         { .ctl_name = 0 },
1796 };
1797
1798 static ctl_table fs_table[] = {
1799         {
1800                 .ctl_name       = FS_DQSTATS,
1801                 .procname       = "quota",
1802                 .mode           = 0555,
1803                 .child          = fs_dqstats_table,
1804         },
1805         { .ctl_name = 0 },
1806 };
1807
1808 static ctl_table sys_table[] = {
1809         {
1810                 .ctl_name       = CTL_FS,
1811                 .procname       = "fs",
1812                 .mode           = 0555,
1813                 .child          = fs_table,
1814         },
1815         { .ctl_name = 0 },
1816 };
1817
1818 static int __init dquot_init(void)
1819 {
1820         int i;
1821         unsigned long nr_hash, order;
1822
1823         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1824
1825         register_sysctl_table(sys_table, 0);
1826
1827         dquot_cachep = kmem_cache_create("dquot",
1828                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1829                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1830                                 SLAB_MEM_SPREAD|SLAB_PANIC),
1831                         NULL, NULL);
1832
1833         order = 0;
1834         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1835         if (!dquot_hash)
1836                 panic("Cannot create dquot hash table");
1837
1838         /* Find power-of-two hlist_heads which can fit into allocation */
1839         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1840         dq_hash_bits = 0;
1841         do {
1842                 dq_hash_bits++;
1843         } while (nr_hash >> dq_hash_bits);
1844         dq_hash_bits--;
1845
1846         nr_hash = 1UL << dq_hash_bits;
1847         dq_hash_mask = nr_hash - 1;
1848         for (i = 0; i < nr_hash; i++)
1849                 INIT_HLIST_HEAD(dquot_hash + i);
1850
1851         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1852                         nr_hash, order, (PAGE_SIZE << order));
1853
1854         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1855
1856         return 0;
1857 }
1858 module_init(dquot_init);
1859
1860 EXPORT_SYMBOL(register_quota_format);
1861 EXPORT_SYMBOL(unregister_quota_format);
1862 EXPORT_SYMBOL(dqstats);
1863 EXPORT_SYMBOL(dq_data_lock);
1864 EXPORT_SYMBOL(vfs_quota_on);
1865 EXPORT_SYMBOL(vfs_quota_on_mount);
1866 EXPORT_SYMBOL(vfs_quota_off);
1867 EXPORT_SYMBOL(vfs_quota_sync);
1868 EXPORT_SYMBOL(vfs_get_dqinfo);
1869 EXPORT_SYMBOL(vfs_set_dqinfo);
1870 EXPORT_SYMBOL(vfs_get_dqblk);
1871 EXPORT_SYMBOL(vfs_set_dqblk);
1872 EXPORT_SYMBOL(dquot_commit);
1873 EXPORT_SYMBOL(dquot_commit_info);
1874 EXPORT_SYMBOL(dquot_acquire);
1875 EXPORT_SYMBOL(dquot_release);
1876 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1877 EXPORT_SYMBOL(dquot_initialize);
1878 EXPORT_SYMBOL(dquot_drop);
1879 EXPORT_SYMBOL(dquot_alloc_space);
1880 EXPORT_SYMBOL(dquot_alloc_inode);
1881 EXPORT_SYMBOL(dquot_free_space);
1882 EXPORT_SYMBOL(dquot_free_inode);
1883 EXPORT_SYMBOL(dquot_transfer);