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_zalloc(dquot_cachep, GFP_NOFS);
 604         if(!dquot)
 605                 return NODQUOT;
 606
 607         mutex_init(&dquot->dq_lock);
 608         INIT_LIST_HEAD(&dquot->dq_free);
 609         INIT_LIST_HEAD(&dquot->dq_inuse);
 610         INIT_HLIST_NODE(&dquot->dq_hash);
 611         INIT_LIST_HEAD(&dquot->dq_dirty);
 612         init_waitqueue_head(&dquot->dq_wait_unused);
 613         dquot->dq_sb = sb;
 614         dquot->dq_type = type;
 615         atomic_set(&dquot->dq_count, 1);
 616
 617         return dquot;
 618 }
 619
 620 /*
 621  * Get reference to dquot
 622  * MUST be called with either dqptr_sem or dqonoff_mutex held
 623  */
 624 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 625 {
 626         unsigned int hashent = hashfn(sb, id, type);
 627         struct dquot *dquot, *empty = NODQUOT;
 628
 629         if (!sb_has_quota_enabled(sb, type))
 630                 return NODQUOT;
 631 we_slept:
 632         spin_lock(&dq_list_lock);
 633         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 634                 if (empty == NODQUOT) {
 635                         spin_unlock(&dq_list_lock);
 636                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 637                                 schedule();     /* Try to wait for a moment... */
 638                         goto we_slept;
 639                 }
 640                 dquot = empty;
 641                 dquot->dq_id = id;
 642                 /* all dquots go on the inuse_list */
 643                 put_inuse(dquot);
 644                 /* hash it first so it can be found */
 645                 insert_dquot_hash(dquot);
 646                 dqstats.lookups++;
 647                 spin_unlock(&dq_list_lock);
 648         } else {
 649                 if (!atomic_read(&dquot->dq_count))
 650                         remove_free_dquot(dquot);
 651                 atomic_inc(&dquot->dq_count);
 652                 dqstats.cache_hits++;
 653                 dqstats.lookups++;
 654                 spin_unlock(&dq_list_lock);
 655                 if (empty)
 656                         kmem_cache_free(dquot_cachep, empty);
 657         }
 658         /* Wait for dq_lock - after this we know that either dquot_release() is already
 659          * finished or it will be canceled due to dq_count > 1 test */
 660         wait_on_dquot(dquot);
 661         /* Read the dquot and instantiate it (everything done only if needed) */
 662         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 663                 dqput(dquot);
 664                 return NODQUOT;
 665         }
 666 #ifdef __DQUOT_PARANOIA
 667         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 668 #endif
 669
 670         return dquot;
 671 }
 672
 673 static int dqinit_needed(struct inode *inode, int type)
 674 {
 675         int cnt;
 676
 677         if (IS_NOQUOTA(inode))
 678                 return 0;
 679         if (type != -1)
 680                 return inode->i_dquot[type] == NODQUOT;
 681         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 682                 if (inode->i_dquot[cnt] == NODQUOT)
 683                         return 1;
 684         return 0;
 685 }
 686
 687 /* This routine is guarded by dqonoff_mutex mutex */
 688 static void add_dquot_ref(struct super_block *sb, int type)
 689 {
 690         struct list_head *p;
 691
 692 restart:
 693         file_list_lock();
 694         list_for_each(p, &sb->s_files) {
 695                 struct file *filp = list_entry(p, struct file, f_u.fu_list);
 696                 struct inode *inode = filp->f_path.dentry->d_inode;
 697                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
 698                         struct dentry *dentry = dget(filp->f_path.dentry);
 699                         file_list_unlock();
 700                         sb->dq_op->initialize(inode, type);
 701                         dput(dentry);
 702                         /* As we may have blocked we had better restart... */
 703                         goto restart;
 704                 }
 705         }
 706         file_list_unlock();
 707 }
 708
 709 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 710 static inline int dqput_blocks(struct dquot *dquot)
 711 {
 712         if (atomic_read(&dquot->dq_count) <= 1)
 713                 return 1;
 714         return 0;
 715 }
 716
 717 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 718 /* We can't race with anybody because we hold dqptr_sem for writing... */
 719 int remove_inode_dquot_ref(struct inode *inode, int type, struct list_head *tofree_head)
 720 {
 721         struct dquot *dquot = inode->i_dquot[type];
 722
 723         inode->i_dquot[type] = NODQUOT;
 724         if (dquot != NODQUOT) {
 725                 if (dqput_blocks(dquot)) {
 726 #ifdef __DQUOT_PARANOIA
 727                         if (atomic_read(&dquot->dq_count) != 1)
 728                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 729 #endif
 730                         spin_lock(&dq_list_lock);
 731                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 732                         spin_unlock(&dq_list_lock);
 733                         return 1;
 734                 }
 735                 else
 736                         dqput(dquot);   /* We have guaranteed we won't block */
 737         }
 738         return 0;
 739 }
 740
 741 /* Free list of dquots - called from inode.c */
 742 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 743 static void put_dquot_list(struct list_head *tofree_head)
 744 {
 745         struct list_head *act_head;
 746         struct dquot *dquot;
 747
 748         act_head = tofree_head->next;
 749         /* So now we have dquots on the list... Just free them */
 750         while (act_head != tofree_head) {
 751                 dquot = list_entry(act_head, struct dquot, dq_free);
 752                 act_head = act_head->next;
 753                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 754                 dqput(dquot);
 755         }
 756 }
 757
 758 /* Gather all references from inodes and drop them */
 759 static void drop_dquot_ref(struct super_block *sb, int type)
 760 {
 761         LIST_HEAD(tofree_head);
 762
 763         down_write(&sb_dqopt(sb)->dqptr_sem);
 764         remove_dquot_ref(sb, type, &tofree_head);
 765         up_write(&sb_dqopt(sb)->dqptr_sem);
 766         put_dquot_list(&tofree_head);
 767 }
 768
 769 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
 770 {
 771         dquot->dq_dqb.dqb_curinodes += number;
 772 }
 773
 774 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 775 {
 776         dquot->dq_dqb.dqb_curspace += number;
 777 }
 778
 779 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
 780 {
 781         if (dquot->dq_dqb.dqb_curinodes > number)
 782                 dquot->dq_dqb.dqb_curinodes -= number;
 783         else
 784                 dquot->dq_dqb.dqb_curinodes = 0;
 785         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
 786                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 787         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 788 }
 789
 790 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 791 {
 792         if (dquot->dq_dqb.dqb_curspace > number)
 793                 dquot->dq_dqb.dqb_curspace -= number;
 794         else
 795                 dquot->dq_dqb.dqb_curspace = 0;
 796         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
 797                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 798         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 799 }
 800
 801 static int flag_print_warnings = 1;
 802
 803 static inline int need_print_warning(struct dquot *dquot)
 804 {
 805         if (!flag_print_warnings)
 806                 return 0;
 807
 808         switch (dquot->dq_type) {
 809                 case USRQUOTA:
 810                         return current->fsuid == dquot->dq_id;
 811                 case GRPQUOTA:
 812                         return in_group_p(dquot->dq_id);
 813         }
 814         return 0;
 815 }
 816
 817 /* Values of warnings */
 818 #define NOWARN 0
 819 #define IHARDWARN 1
 820 #define ISOFTLONGWARN 2
 821 #define ISOFTWARN 3
 822 #define BHARDWARN 4
 823 #define BSOFTLONGWARN 5
 824 #define BSOFTWARN 6
 825
 826 /* Print warning to user which exceeded quota */
 827 static void print_warning(struct dquot *dquot, const char warntype)
 828 {
 829         char *msg = NULL;
 830         struct tty_struct *tty;
 831         int flag = (warntype == BHARDWARN || warntype == BSOFTLONGWARN) ? DQ_BLKS_B :
 832           ((warntype == IHARDWARN || warntype == ISOFTLONGWARN) ? DQ_INODES_B : 0);
 833
 834         if (!need_print_warning(dquot) || (flag && test_and_set_bit(flag, &dquot->dq_flags)))
 835                 return;
 836
 837         mutex_lock(&tty_mutex);
 838         tty = get_current_tty();
 839         if (!tty)
 840                 goto out_lock;
 841         tty_write_message(tty, dquot->dq_sb->s_id);
 842         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
 843                 tty_write_message(tty, ": warning, ");
 844         else
 845                 tty_write_message(tty, ": write failed, ");
 846         tty_write_message(tty, quotatypes[dquot->dq_type]);
 847         switch (warntype) {
 848                 case IHARDWARN:
 849                         msg = " file limit reached.\r\n";
 850                         break;
 851                 case ISOFTLONGWARN:
 852                         msg = " file quota exceeded too long.\r\n";
 853                         break;
 854                 case ISOFTWARN:
 855                         msg = " file quota exceeded.\r\n";
 856                         break;
 857                 case BHARDWARN:
 858                         msg = " block limit reached.\r\n";
 859                         break;
 860                 case BSOFTLONGWARN:
 861                         msg = " block quota exceeded too long.\r\n";
 862                         break;
 863                 case BSOFTWARN:
 864                         msg = " block quota exceeded.\r\n";
 865                         break;
 866         }
 867         tty_write_message(tty, msg);
 868 out_lock:
 869         mutex_unlock(&tty_mutex);
 870 }
 871
 872 static inline void flush_warnings(struct dquot **dquots, char *warntype)
 873 {
 874         int i;
 875
 876         for (i = 0; i < MAXQUOTAS; i++)
 877                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
 878                         print_warning(dquots[i], warntype[i]);
 879 }
 880
 881 static inline char ignore_hardlimit(struct dquot *dquot)
 882 {
 883         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
 884
 885         return capable(CAP_SYS_RESOURCE) &&
 886             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
 887 }
 888
 889 /* needs dq_data_lock */
 890 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
 891 {
 892         *warntype = NOWARN;
 893         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 894                 return QUOTA_OK;
 895
 896         if (dquot->dq_dqb.dqb_ihardlimit &&
 897            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
 898             !ignore_hardlimit(dquot)) {
 899                 *warntype = IHARDWARN;
 900                 return NO_QUOTA;
 901         }
 902
 903         if (dquot->dq_dqb.dqb_isoftlimit &&
 904            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 905             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
 906             !ignore_hardlimit(dquot)) {
 907                 *warntype = ISOFTLONGWARN;
 908                 return NO_QUOTA;
 909         }
 910
 911         if (dquot->dq_dqb.dqb_isoftlimit &&
 912            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 913             dquot->dq_dqb.dqb_itime == 0) {
 914                 *warntype = ISOFTWARN;
 915                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
 916         }
 917
 918         return QUOTA_OK;
 919 }
 920
 921 /* needs dq_data_lock */
 922 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
 923 {
 924         *warntype = 0;
 925         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 926                 return QUOTA_OK;
 927
 928         if (dquot->dq_dqb.dqb_bhardlimit &&
 929            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
 930             !ignore_hardlimit(dquot)) {
 931                 if (!prealloc)
 932                         *warntype = BHARDWARN;
 933                 return NO_QUOTA;
 934         }
 935
 936         if (dquot->dq_dqb.dqb_bsoftlimit &&
 937            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 938             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
 939             !ignore_hardlimit(dquot)) {
 940                 if (!prealloc)
 941                         *warntype = BSOFTLONGWARN;
 942                 return NO_QUOTA;
 943         }
 944
 945         if (dquot->dq_dqb.dqb_bsoftlimit &&
 946            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 947             dquot->dq_dqb.dqb_btime == 0) {
 948                 if (!prealloc) {
 949                         *warntype = BSOFTWARN;
 950                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
 951                 }
 952                 else
 953                         /*
 954                          * We don't allow preallocation to exceed softlimit so exceeding will
 955                          * be always printed
 956                          */
 957                         return NO_QUOTA;
 958         }
 959
 960         return QUOTA_OK;
 961 }
 962
 963 /*
 964  *      Initialize quota pointers in inode
 965  *      Transaction must be started at entry
 966  */
 967 int dquot_initialize(struct inode *inode, int type)
 968 {
 969         unsigned int id = 0;
 970         int cnt, ret = 0;
 971
 972         /* First test before acquiring mutex - solves deadlocks when we
 973          * re-enter the quota code and are already holding the mutex */
 974         if (IS_NOQUOTA(inode))
 975                 return 0;
 976         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 977         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
 978         if (IS_NOQUOTA(inode))
 979                 goto out_err;
 980         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 981                 if (type != -1 && cnt != type)
 982                         continue;
 983                 if (inode->i_dquot[cnt] == NODQUOT) {
 984                         switch (cnt) {
 985                                 case USRQUOTA:
 986                                         id = inode->i_uid;
 987                                         break;
 988                                 case GRPQUOTA:
 989                                         id = inode->i_gid;
 990                                         break;
 991                         }
 992                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
 993                 }
 994         }
 995 out_err:
 996         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 997         return ret;
 998 }
 999
1000 /*
1001  *      Release all quotas referenced by inode
1002  *      Transaction must be started at an entry
1003  */
1004 int dquot_drop(struct inode *inode)
1005 {
1006         int cnt;
1007
1008         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1009         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1010                 if (inode->i_dquot[cnt] != NODQUOT) {
1011                         dqput(inode->i_dquot[cnt]);
1012                         inode->i_dquot[cnt] = NODQUOT;
1013                 }
1014         }
1015         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1016         return 0;
1017 }
1018
1019 /*
1020  * Following four functions update i_blocks+i_bytes fields and
1021  * quota information (together with appropriate checks)
1022  * NOTE: We absolutely rely on the fact that caller dirties
1023  * the inode (usually macros in quotaops.h care about this) and
1024  * holds a handle for the current transaction so that dquot write and
1025  * inode write go into the same transaction.
1026  */
1027
1028 /*
1029  * This operation can block, but only after everything is updated
1030  */
1031 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1032 {
1033         int cnt, ret = NO_QUOTA;
1034         char warntype[MAXQUOTAS];
1035
1036         /* First test before acquiring mutex - solves deadlocks when we
1037          * re-enter the quota code and are already holding the mutex */
1038         if (IS_NOQUOTA(inode)) {
1039 out_add:
1040                 inode_add_bytes(inode, number);
1041                 return QUOTA_OK;
1042         }
1043         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1044                 warntype[cnt] = NOWARN;
1045
1046         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1047         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1048                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1049                 goto out_add;
1050         }
1051         spin_lock(&dq_data_lock);
1052         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1053                 if (inode->i_dquot[cnt] == NODQUOT)
1054                         continue;
1055                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1056                         goto warn_put_all;
1057         }
1058         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1059                 if (inode->i_dquot[cnt] == NODQUOT)
1060                         continue;
1061                 dquot_incr_space(inode->i_dquot[cnt], number);
1062         }
1063         inode_add_bytes(inode, number);
1064         ret = QUOTA_OK;
1065 warn_put_all:
1066         spin_unlock(&dq_data_lock);
1067         if (ret == QUOTA_OK)
1068                 /* Dirtify all the dquots - this can block when journalling */
1069                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1070                         if (inode->i_dquot[cnt])
1071                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1072         flush_warnings(inode->i_dquot, warntype);
1073         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1074         return ret;
1075 }
1076
1077 /*
1078  * This operation can block, but only after everything is updated
1079  */
1080 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1081 {
1082         int cnt, ret = NO_QUOTA;
1083         char warntype[MAXQUOTAS];
1084
1085         /* First test before acquiring mutex - solves deadlocks when we
1086          * re-enter the quota code and are already holding the mutex */
1087         if (IS_NOQUOTA(inode))
1088                 return QUOTA_OK;
1089         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1090                 warntype[cnt] = NOWARN;
1091         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1092         if (IS_NOQUOTA(inode)) {
1093                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1094                 return QUOTA_OK;
1095         }
1096         spin_lock(&dq_data_lock);
1097         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1098                 if (inode->i_dquot[cnt] == NODQUOT)
1099                         continue;
1100                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1101                         goto warn_put_all;
1102         }
1103
1104         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1105                 if (inode->i_dquot[cnt] == NODQUOT)
1106                         continue;
1107                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1108         }
1109         ret = QUOTA_OK;
1110 warn_put_all:
1111         spin_unlock(&dq_data_lock);
1112         if (ret == QUOTA_OK)
1113                 /* Dirtify all the dquots - this can block when journalling */
1114                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1115                         if (inode->i_dquot[cnt])
1116                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1117         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1118         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1119         return ret;
1120 }
1121
1122 /*
1123  * This operation can block, but only after everything is updated
1124  */
1125 int dquot_free_space(struct inode *inode, qsize_t number)
1126 {
1127         unsigned int cnt;
1128
1129         /* First test before acquiring mutex - solves deadlocks when we
1130          * re-enter the quota code and are already holding the mutex */
1131         if (IS_NOQUOTA(inode)) {
1132 out_sub:
1133                 inode_sub_bytes(inode, number);
1134                 return QUOTA_OK;
1135         }
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                 goto out_sub;
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_space(inode->i_dquot[cnt], number);
1147         }
1148         inode_sub_bytes(inode, number);
1149         spin_unlock(&dq_data_lock);
1150         /* Dirtify all the dquots - this can block when journalling */
1151         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1152                 if (inode->i_dquot[cnt])
1153                         mark_dquot_dirty(inode->i_dquot[cnt]);
1154         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1155         return QUOTA_OK;
1156 }
1157
1158 /*
1159  * This operation can block, but only after everything is updated
1160  */
1161 int dquot_free_inode(const struct inode *inode, unsigned long number)
1162 {
1163         unsigned int cnt;
1164
1165         /* First test before acquiring mutex - solves deadlocks when we
1166          * re-enter the quota code and are already holding the mutex */
1167         if (IS_NOQUOTA(inode))
1168                 return QUOTA_OK;
1169         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1170         /* Now recheck reliably when holding dqptr_sem */
1171         if (IS_NOQUOTA(inode)) {
1172                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1173                 return QUOTA_OK;
1174         }
1175         spin_lock(&dq_data_lock);
1176         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1177                 if (inode->i_dquot[cnt] == NODQUOT)
1178                         continue;
1179                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1180         }
1181         spin_unlock(&dq_data_lock);
1182         /* Dirtify all the dquots - this can block when journalling */
1183         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1184                 if (inode->i_dquot[cnt])
1185                         mark_dquot_dirty(inode->i_dquot[cnt]);
1186         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1187         return QUOTA_OK;
1188 }
1189
1190 /*
1191  * Transfer the number of inode and blocks from one diskquota to an other.
1192  *
1193  * This operation can block, but only after everything is updated
1194  * A transaction must be started when entering this function.
1195  */
1196 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1197 {
1198         qsize_t space;
1199         struct dquot *transfer_from[MAXQUOTAS];
1200         struct dquot *transfer_to[MAXQUOTAS];
1201         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1202             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1203         char warntype[MAXQUOTAS];
1204
1205         /* First test before acquiring mutex - solves deadlocks when we
1206          * re-enter the quota code and are already holding the mutex */
1207         if (IS_NOQUOTA(inode))
1208                 return QUOTA_OK;
1209         /* Clear the arrays */
1210         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1211                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1212                 warntype[cnt] = NOWARN;
1213         }
1214         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1215         /* Now recheck reliably when holding dqptr_sem */
1216         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1217                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1218                 return QUOTA_OK;
1219         }
1220         /* First build the transfer_to list - here we can block on
1221          * reading/instantiating of dquots.  We know that the transaction for
1222          * us was already started so we don't violate lock ranking here */
1223         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1224                 switch (cnt) {
1225                         case USRQUOTA:
1226                                 if (!chuid)
1227                                         continue;
1228                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1229                                 break;
1230                         case GRPQUOTA:
1231                                 if (!chgid)
1232                                         continue;
1233                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1234                                 break;
1235                 }
1236         }
1237         spin_lock(&dq_data_lock);
1238         space = inode_get_bytes(inode);
1239         /* Build the transfer_from list and check the limits */
1240         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1241                 if (transfer_to[cnt] == NODQUOT)
1242                         continue;
1243                 transfer_from[cnt] = inode->i_dquot[cnt];
1244                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1245                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1246                         goto warn_put_all;
1247         }
1248
1249         /*
1250          * Finally perform the needed transfer from transfer_from to transfer_to
1251          */
1252         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1253                 /*
1254                  * Skip changes for same uid or gid or for turned off quota-type.
1255                  */
1256                 if (transfer_to[cnt] == NODQUOT)
1257                         continue;
1258
1259                 /* Due to IO error we might not have transfer_from[] structure */
1260                 if (transfer_from[cnt]) {
1261                         dquot_decr_inodes(transfer_from[cnt], 1);
1262                         dquot_decr_space(transfer_from[cnt], space);
1263                 }
1264
1265                 dquot_incr_inodes(transfer_to[cnt], 1);
1266                 dquot_incr_space(transfer_to[cnt], space);
1267
1268                 inode->i_dquot[cnt] = transfer_to[cnt];
1269         }
1270         ret = QUOTA_OK;
1271 warn_put_all:
1272         spin_unlock(&dq_data_lock);
1273         /* Dirtify all the dquots - this can block when journalling */
1274         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1275                 if (transfer_from[cnt])
1276                         mark_dquot_dirty(transfer_from[cnt]);
1277                 if (transfer_to[cnt])
1278                         mark_dquot_dirty(transfer_to[cnt]);
1279         }
1280         flush_warnings(transfer_to, warntype);
1281
1282         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1283                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1284                         dqput(transfer_from[cnt]);
1285                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1286                         dqput(transfer_to[cnt]);
1287         }
1288         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1289         return ret;
1290 }
1291
1292 /*
1293  * Write info of quota file to disk
1294  */
1295 int dquot_commit_info(struct super_block *sb, int type)
1296 {
1297         int ret;
1298         struct quota_info *dqopt = sb_dqopt(sb);
1299
1300         mutex_lock(&dqopt->dqio_mutex);
1301         ret = dqopt->ops[type]->write_file_info(sb, type);
1302         mutex_unlock(&dqopt->dqio_mutex);
1303         return ret;
1304 }
1305
1306 /*
1307  * Definitions of diskquota operations.
1308  */
1309 struct dquot_operations dquot_operations = {
1310         .initialize     = dquot_initialize,
1311         .drop           = dquot_drop,
1312         .alloc_space    = dquot_alloc_space,
1313         .alloc_inode    = dquot_alloc_inode,
1314         .free_space     = dquot_free_space,
1315         .free_inode     = dquot_free_inode,
1316         .transfer       = dquot_transfer,
1317         .write_dquot    = dquot_commit,
1318         .acquire_dquot  = dquot_acquire,
1319         .release_dquot  = dquot_release,
1320         .mark_dirty     = dquot_mark_dquot_dirty,
1321         .write_info     = dquot_commit_info
1322 };
1323
1324 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1325 {
1326         switch (type) {
1327                 case USRQUOTA:
1328                         dqopt->flags |= DQUOT_USR_ENABLED;
1329                         break;
1330                 case GRPQUOTA:
1331                         dqopt->flags |= DQUOT_GRP_ENABLED;
1332                         break;
1333         }
1334 }
1335
1336 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1337 {
1338         switch (type) {
1339                 case USRQUOTA:
1340                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1341                         break;
1342                 case GRPQUOTA:
1343                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1344                         break;
1345         }
1346 }
1347
1348 /*
1349  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1350  */
1351 int vfs_quota_off(struct super_block *sb, int type)
1352 {
1353         int cnt;
1354         struct quota_info *dqopt = sb_dqopt(sb);
1355         struct inode *toputinode[MAXQUOTAS];
1356
1357         /* We need to serialize quota_off() for device */
1358         mutex_lock(&dqopt->dqonoff_mutex);
1359         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1360                 toputinode[cnt] = NULL;
1361                 if (type != -1 && cnt != type)
1362                         continue;
1363                 if (!sb_has_quota_enabled(sb, cnt))
1364                         continue;
1365                 reset_enable_flags(dqopt, cnt);
1366
1367                 /* Note: these are blocking operations */
1368                 drop_dquot_ref(sb, cnt);
1369                 invalidate_dquots(sb, cnt);
1370                 /*
1371                  * Now all dquots should be invalidated, all writes done so we should be only
1372                  * users of the info. No locks needed.
1373                  */
1374                 if (info_dirty(&dqopt->info[cnt]))
1375                         sb->dq_op->write_info(sb, cnt);
1376                 if (dqopt->ops[cnt]->free_file_info)
1377                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1378                 put_quota_format(dqopt->info[cnt].dqi_format);
1379
1380                 toputinode[cnt] = dqopt->files[cnt];
1381                 dqopt->files[cnt] = NULL;
1382                 dqopt->info[cnt].dqi_flags = 0;
1383                 dqopt->info[cnt].dqi_igrace = 0;
1384                 dqopt->info[cnt].dqi_bgrace = 0;
1385                 dqopt->ops[cnt] = NULL;
1386         }
1387         mutex_unlock(&dqopt->dqonoff_mutex);
1388         /* Sync the superblock so that buffers with quota data are written to
1389          * disk (and so userspace sees correct data afterwards). */
1390         if (sb->s_op->sync_fs)
1391                 sb->s_op->sync_fs(sb, 1);
1392         sync_blockdev(sb->s_bdev);
1393         /* Now the quota files are just ordinary files and we can set the
1394          * inode flags back. Moreover we discard the pagecache so that
1395          * userspace sees the writes we did bypassing the pagecache. We
1396          * must also discard the blockdev buffers so that we see the
1397          * changes done by userspace on the next quotaon() */
1398         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1399                 if (toputinode[cnt]) {
1400                         mutex_lock(&dqopt->dqonoff_mutex);
1401                         /* If quota was reenabled in the meantime, we have
1402                          * nothing to do */
1403                         if (!sb_has_quota_enabled(sb, cnt)) {
1404                                 mutex_lock(&toputinode[cnt]->i_mutex);
1405                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1406                                   S_NOATIME | S_NOQUOTA);
1407                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1408                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1409                                 mark_inode_dirty(toputinode[cnt]);
1410                                 iput(toputinode[cnt]);
1411                         }
1412                         mutex_unlock(&dqopt->dqonoff_mutex);
1413                 }
1414         if (sb->s_bdev)
1415                 invalidate_bdev(sb->s_bdev, 0);
1416         return 0;
1417 }
1418
1419 /*
1420  *      Turn quotas on on a device
1421  */
1422
1423 /* Helper function when we already have the inode */
1424 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1425 {
1426         struct quota_format_type *fmt = find_quota_format(format_id);
1427         struct super_block *sb = inode->i_sb;
1428         struct quota_info *dqopt = sb_dqopt(sb);
1429         int error;
1430         int oldflags = -1;
1431
1432         if (!fmt)
1433                 return -ESRCH;
1434         if (!S_ISREG(inode->i_mode)) {
1435                 error = -EACCES;
1436                 goto out_fmt;
1437         }
1438         if (IS_RDONLY(inode)) {
1439                 error = -EROFS;
1440                 goto out_fmt;
1441         }
1442         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1443                 error = -EINVAL;
1444                 goto out_fmt;
1445         }
1446
1447         /* As we bypass the pagecache we must now flush the inode so that
1448          * we see all the changes from userspace... */
1449         write_inode_now(inode, 1);
1450         /* And now flush the block cache so that kernel sees the changes */
1451         invalidate_bdev(sb->s_bdev, 0);
1452         mutex_lock(&inode->i_mutex);
1453         mutex_lock(&dqopt->dqonoff_mutex);
1454         if (sb_has_quota_enabled(sb, type)) {
1455                 error = -EBUSY;
1456                 goto out_lock;
1457         }
1458         /* We don't want quota and atime on quota files (deadlocks possible)
1459          * Also nobody should write to the file - we use special IO operations
1460          * which ignore the immutable bit. */
1461         down_write(&dqopt->dqptr_sem);
1462         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1463         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1464         up_write(&dqopt->dqptr_sem);
1465         sb->dq_op->drop(inode);
1466
1467         error = -EIO;
1468         dqopt->files[type] = igrab(inode);
1469         if (!dqopt->files[type])
1470                 goto out_lock;
1471         error = -EINVAL;
1472         if (!fmt->qf_ops->check_quota_file(sb, type))
1473                 goto out_file_init;
1474
1475         dqopt->ops[type] = fmt->qf_ops;
1476         dqopt->info[type].dqi_format = fmt;
1477         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1478         mutex_lock(&dqopt->dqio_mutex);
1479         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1480                 mutex_unlock(&dqopt->dqio_mutex);
1481                 goto out_file_init;
1482         }
1483         mutex_unlock(&dqopt->dqio_mutex);
1484         mutex_unlock(&inode->i_mutex);
1485         set_enable_flags(dqopt, type);
1486
1487         add_dquot_ref(sb, type);
1488         mutex_unlock(&dqopt->dqonoff_mutex);
1489
1490         return 0;
1491
1492 out_file_init:
1493         dqopt->files[type] = NULL;
1494         iput(inode);
1495 out_lock:
1496         mutex_unlock(&dqopt->dqonoff_mutex);
1497         if (oldflags != -1) {
1498                 down_write(&dqopt->dqptr_sem);
1499                 /* Set the flags back (in the case of accidental quotaon()
1500                  * on a wrong file we don't want to mess up the flags) */
1501                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1502                 inode->i_flags |= oldflags;
1503                 up_write(&dqopt->dqptr_sem);
1504         }
1505         mutex_unlock(&inode->i_mutex);
1506 out_fmt:
1507         put_quota_format(fmt);
1508
1509         return error;
1510 }
1511
1512 /* Actual function called from quotactl() */
1513 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1514 {
1515         struct nameidata nd;
1516         int error;
1517
1518         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1519         if (error < 0)
1520                 return error;
1521         error = security_quota_on(nd.dentry);
1522         if (error)
1523                 goto out_path;
1524         /* Quota file not on the same filesystem? */
1525         if (nd.mnt->mnt_sb != sb)
1526                 error = -EXDEV;
1527         else
1528                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1529 out_path:
1530         path_release(&nd);
1531         return error;
1532 }
1533
1534 /*
1535  * This function is used when filesystem needs to initialize quotas
1536  * during mount time.
1537  */
1538 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1539                 int format_id, int type)
1540 {
1541         struct dentry *dentry;
1542         int error;
1543
1544         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1545         if (IS_ERR(dentry))
1546                 return PTR_ERR(dentry);
1547
1548         if (!dentry->d_inode) {
1549                 error = -ENOENT;
1550                 goto out;
1551         }
1552
1553         error = security_quota_on(dentry);
1554         if (!error)
1555                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1556
1557 out:
1558         dput(dentry);
1559         return error;
1560 }
1561
1562 /* Generic routine for getting common part of quota structure */
1563 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1564 {
1565         struct mem_dqblk *dm = &dquot->dq_dqb;
1566
1567         spin_lock(&dq_data_lock);
1568         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1569         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1570         di->dqb_curspace = dm->dqb_curspace;
1571         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1572         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1573         di->dqb_curinodes = dm->dqb_curinodes;
1574         di->dqb_btime = dm->dqb_btime;
1575         di->dqb_itime = dm->dqb_itime;
1576         di->dqb_valid = QIF_ALL;
1577         spin_unlock(&dq_data_lock);
1578 }
1579
1580 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1581 {
1582         struct dquot *dquot;
1583
1584         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1585         if (!(dquot = dqget(sb, id, type))) {
1586                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1587                 return -ESRCH;
1588         }
1589         do_get_dqblk(dquot, di);
1590         dqput(dquot);
1591         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1592         return 0;
1593 }
1594
1595 /* Generic routine for setting common part of quota structure */
1596 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1597 {
1598         struct mem_dqblk *dm = &dquot->dq_dqb;
1599         int check_blim = 0, check_ilim = 0;
1600
1601         spin_lock(&dq_data_lock);
1602         if (di->dqb_valid & QIF_SPACE) {
1603                 dm->dqb_curspace = di->dqb_curspace;
1604                 check_blim = 1;
1605         }
1606         if (di->dqb_valid & QIF_BLIMITS) {
1607                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1608                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1609                 check_blim = 1;
1610         }
1611         if (di->dqb_valid & QIF_INODES) {
1612                 dm->dqb_curinodes = di->dqb_curinodes;
1613                 check_ilim = 1;
1614         }
1615         if (di->dqb_valid & QIF_ILIMITS) {
1616                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1617                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1618                 check_ilim = 1;
1619         }
1620         if (di->dqb_valid & QIF_BTIME)
1621                 dm->dqb_btime = di->dqb_btime;
1622         if (di->dqb_valid & QIF_ITIME)
1623                 dm->dqb_itime = di->dqb_itime;
1624
1625         if (check_blim) {
1626                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1627                         dm->dqb_btime = 0;
1628                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1629                 }
1630                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1631                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1632         }
1633         if (check_ilim) {
1634                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1635                         dm->dqb_itime = 0;
1636                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1637                 }
1638                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1639                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1640         }
1641         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1642                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1643         else
1644                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1645         spin_unlock(&dq_data_lock);
1646         mark_dquot_dirty(dquot);
1647 }
1648
1649 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1650 {
1651         struct dquot *dquot;
1652
1653         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1654         if (!(dquot = dqget(sb, id, type))) {
1655                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1656                 return -ESRCH;
1657         }
1658         do_set_dqblk(dquot, di);
1659         dqput(dquot);
1660         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1661         return 0;
1662 }
1663
1664 /* Generic routine for getting common part of quota file information */
1665 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1666 {
1667         struct mem_dqinfo *mi;
1668
1669         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1670         if (!sb_has_quota_enabled(sb, type)) {
1671                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1672                 return -ESRCH;
1673         }
1674         mi = sb_dqopt(sb)->info + type;
1675         spin_lock(&dq_data_lock);
1676         ii->dqi_bgrace = mi->dqi_bgrace;
1677         ii->dqi_igrace = mi->dqi_igrace;
1678         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1679         ii->dqi_valid = IIF_ALL;
1680         spin_unlock(&dq_data_lock);
1681         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1682         return 0;
1683 }
1684
1685 /* Generic routine for setting common part of quota file information */
1686 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1687 {
1688         struct mem_dqinfo *mi;
1689
1690         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1691         if (!sb_has_quota_enabled(sb, type)) {
1692                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1693                 return -ESRCH;
1694         }
1695         mi = sb_dqopt(sb)->info + type;
1696         spin_lock(&dq_data_lock);
1697         if (ii->dqi_valid & IIF_BGRACE)
1698                 mi->dqi_bgrace = ii->dqi_bgrace;
1699         if (ii->dqi_valid & IIF_IGRACE)
1700                 mi->dqi_igrace = ii->dqi_igrace;
1701         if (ii->dqi_valid & IIF_FLAGS)
1702                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1703         spin_unlock(&dq_data_lock);
1704         mark_info_dirty(sb, type);
1705         /* Force write to disk */
1706         sb->dq_op->write_info(sb, type);
1707         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1708         return 0;
1709 }
1710
1711 struct quotactl_ops vfs_quotactl_ops = {
1712         .quota_on       = vfs_quota_on,
1713         .quota_off      = vfs_quota_off,
1714         .quota_sync     = vfs_quota_sync,
1715         .get_info       = vfs_get_dqinfo,
1716         .set_info       = vfs_set_dqinfo,
1717         .get_dqblk      = vfs_get_dqblk,
1718         .set_dqblk      = vfs_set_dqblk
1719 };
1720
1721 static ctl_table fs_dqstats_table[] = {
1722         {
1723                 .ctl_name       = FS_DQ_LOOKUPS,
1724                 .procname       = "lookups",
1725                 .data           = &dqstats.lookups,
1726                 .maxlen         = sizeof(int),
1727                 .mode           = 0444,
1728                 .proc_handler   = &proc_dointvec,
1729         },
1730         {
1731                 .ctl_name       = FS_DQ_DROPS,
1732                 .procname       = "drops",
1733                 .data           = &dqstats.drops,
1734                 .maxlen         = sizeof(int),
1735                 .mode           = 0444,
1736                 .proc_handler   = &proc_dointvec,
1737         },
1738         {
1739                 .ctl_name       = FS_DQ_READS,
1740                 .procname       = "reads",
1741                 .data           = &dqstats.reads,
1742                 .maxlen         = sizeof(int),
1743                 .mode           = 0444,
1744                 .proc_handler   = &proc_dointvec,
1745         },
1746         {
1747                 .ctl_name       = FS_DQ_WRITES,
1748                 .procname       = "writes",
1749                 .data           = &dqstats.writes,
1750                 .maxlen         = sizeof(int),
1751                 .mode           = 0444,
1752                 .proc_handler   = &proc_dointvec,
1753         },
1754         {
1755                 .ctl_name       = FS_DQ_CACHE_HITS,
1756                 .procname       = "cache_hits",
1757                 .data           = &dqstats.cache_hits,
1758                 .maxlen         = sizeof(int),
1759                 .mode           = 0444,
1760                 .proc_handler   = &proc_dointvec,
1761         },
1762         {
1763                 .ctl_name       = FS_DQ_ALLOCATED,
1764                 .procname       = "allocated_dquots",
1765                 .data           = &dqstats.allocated_dquots,
1766                 .maxlen         = sizeof(int),
1767                 .mode           = 0444,
1768                 .proc_handler   = &proc_dointvec,
1769         },
1770         {
1771                 .ctl_name       = FS_DQ_FREE,
1772                 .procname       = "free_dquots",
1773                 .data           = &dqstats.free_dquots,
1774                 .maxlen         = sizeof(int),
1775                 .mode           = 0444,
1776                 .proc_handler   = &proc_dointvec,
1777         },
1778         {
1779                 .ctl_name       = FS_DQ_SYNCS,
1780                 .procname       = "syncs",
1781                 .data           = &dqstats.syncs,
1782                 .maxlen         = sizeof(int),
1783                 .mode           = 0444,
1784                 .proc_handler   = &proc_dointvec,
1785         },
1786         {
1787                 .ctl_name       = FS_DQ_WARNINGS,
1788                 .procname       = "warnings",
1789                 .data           = &flag_print_warnings,
1790                 .maxlen         = sizeof(int),
1791                 .mode           = 0644,
1792                 .proc_handler   = &proc_dointvec,
1793         },
1794         { .ctl_name = 0 },
1795 };
1796
1797 static ctl_table fs_table[] = {
1798         {
1799                 .ctl_name       = FS_DQSTATS,
1800                 .procname       = "quota",
1801                 .mode           = 0555,
1802                 .child          = fs_dqstats_table,
1803         },
1804         { .ctl_name = 0 },
1805 };
1806
1807 static ctl_table sys_table[] = {
1808         {
1809                 .ctl_name       = CTL_FS,
1810                 .procname       = "fs",
1811                 .mode           = 0555,
1812                 .child          = fs_table,
1813         },
1814         { .ctl_name = 0 },
1815 };
1816
1817 static int __init dquot_init(void)
1818 {
1819         int i;
1820         unsigned long nr_hash, order;
1821
1822         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1823
1824         register_sysctl_table(sys_table, 0);
1825
1826         dquot_cachep = kmem_cache_create("dquot",
1827                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1828                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1829                                 SLAB_MEM_SPREAD|SLAB_PANIC),
1830                         NULL, NULL);
1831
1832         order = 0;
1833         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1834         if (!dquot_hash)
1835                 panic("Cannot create dquot hash table");
1836
1837         /* Find power-of-two hlist_heads which can fit into allocation */
1838         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1839         dq_hash_bits = 0;
1840         do {
1841                 dq_hash_bits++;
1842         } while (nr_hash >> dq_hash_bits);
1843         dq_hash_bits--;
1844
1845         nr_hash = 1UL << dq_hash_bits;
1846         dq_hash_mask = nr_hash - 1;
1847         for (i = 0; i < nr_hash; i++)
1848                 INIT_HLIST_HEAD(dquot_hash + i);
1849
1850         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1851                         nr_hash, order, (PAGE_SIZE << order));
1852
1853         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1854
1855         return 0;
1856 }
1857 module_init(dquot_init);
1858
1859 EXPORT_SYMBOL(register_quota_format);
1860 EXPORT_SYMBOL(unregister_quota_format);
1861 EXPORT_SYMBOL(dqstats);
1862 EXPORT_SYMBOL(dq_data_lock);
1863 EXPORT_SYMBOL(vfs_quota_on);
1864 EXPORT_SYMBOL(vfs_quota_on_mount);
1865 EXPORT_SYMBOL(vfs_quota_off);
1866 EXPORT_SYMBOL(vfs_quota_sync);
1867 EXPORT_SYMBOL(vfs_get_dqinfo);
1868 EXPORT_SYMBOL(vfs_set_dqinfo);
1869 EXPORT_SYMBOL(vfs_get_dqblk);
1870 EXPORT_SYMBOL(vfs_set_dqblk);
1871 EXPORT_SYMBOL(dquot_commit);
1872 EXPORT_SYMBOL(dquot_commit_info);
1873 EXPORT_SYMBOL(dquot_acquire);
1874 EXPORT_SYMBOL(dquot_release);
1875 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1876 EXPORT_SYMBOL(dquot_initialize);
1877 EXPORT_SYMBOL(dquot_drop);
1878 EXPORT_SYMBOL(dquot_alloc_space);
1879 EXPORT_SYMBOL(dquot_alloc_inode);
1880 EXPORT_SYMBOL(dquot_free_space);
1881 EXPORT_SYMBOL(dquot_free_inode);
1882 EXPORT_SYMBOL(dquot_transfer);