git.oblomov.eu Git - linux-2.6/blob - fs/dquot.c

   1 /*
   2  * Implementation of the diskquota system for the LINUX operating system. QUOTA
   3  * is implemented using the BSD system call interface as the means of
   4  * communication with the user level. This file contains the generic routines
   5  * called by the different filesystems on allocation of an inode or block.
   6  * These routines take care of the administration needed to have a consistent
   7  * diskquota tracking system. The ideas of both user and group quotas are based
   8  * on the Melbourne quota system as used on BSD derived systems. The internal
   9  * implementation is based on one of the several variants of the LINUX
  10  * inode-subsystem with added complexity of the diskquota system.
  11  *
  12  * Version: $Id: dquot.c,v 6.3 1996/11/17 18:35:34 mvw Exp mvw $
  13  *
  14  * Author:      Marco van Wieringen <mvw@planets.elm.net>
  15  *
  16  * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
  17  *
  18  *              Revised list management to avoid races
  19  *              -- Bill Hawes, <whawes@star.net>, 9/98
  20  *
  21  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
  22  *              As the consequence the locking was moved from dquot_decr_...(),
  23  *              dquot_incr_...() to calling functions.
  24  *              invalidate_dquots() now writes modified dquots.
  25  *              Serialized quota_off() and quota_on() for mount point.
  26  *              Fixed a few bugs in grow_dquots().
  27  *              Fixed deadlock in write_dquot() - we no longer account quotas on
  28  *              quota files
  29  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
  30  *              add_dquot_ref() restarts after blocking
  31  *              Added check for bogus uid and fixed check for group in quotactl.
  32  *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
  33  *
  34  *              Used struct list_head instead of own list struct
  35  *              Invalidation of referenced dquots is no longer possible
  36  *              Improved free_dquots list management
  37  *              Quota and i_blocks are now updated in one place to avoid races
  38  *              Warnings are now delayed so we won't block in critical section
  39  *              Write updated not to require dquot lock
  40  *              Jan Kara, <jack@suse.cz>, 9/2000
  41  *
  42  *              Added dynamic quota structure allocation
  43  *              Jan Kara <jack@suse.cz> 12/2000
  44  *
  45  *              Rewritten quota interface. Implemented new quota format and
  46  *              formats registering.
  47  *              Jan Kara, <jack@suse.cz>, 2001,2002
  48  *
  49  *              New SMP locking.
  50  *              Jan Kara, <jack@suse.cz>, 10/2002
  51  *
  52  *              Added journalled quota support, fix lock inversion problems
  53  *              Jan Kara, <jack@suse.cz>, 2003,2004
  54  *
  55  * (C) Copyright 1994 - 1997 Marco van Wieringen
  56  */
  57
  58 #include <linux/errno.h>
  59 #include <linux/kernel.h>
  60 #include <linux/fs.h>
  61 #include <linux/mount.h>
  62 #include <linux/mm.h>
  63 #include <linux/time.h>
  64 #include <linux/types.h>
  65 #include <linux/string.h>
  66 #include <linux/fcntl.h>
  67 #include <linux/stat.h>
  68 #include <linux/tty.h>
  69 #include <linux/file.h>
  70 #include <linux/slab.h>
  71 #include <linux/sysctl.h>
  72 #include <linux/smp_lock.h>
  73 #include <linux/init.h>
  74 #include <linux/module.h>
  75 #include <linux/proc_fs.h>
  76 #include <linux/security.h>
  77 #include <linux/kmod.h>
  78 #include <linux/namei.h>
  79 #include <linux/buffer_head.h>
  80 #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 kmem_cache_t *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, SLAB_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_dentry->d_inode;
 698                 if (filp->f_mode & FMODE_WRITE && dqinit_needed(inode, type)) {
 699                         struct dentry *dentry = dget(filp->f_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         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         tty_write_message(current->signal->tty, dquot->dq_sb->s_id);
 838         if (warntype == ISOFTWARN || warntype == BSOFTWARN)
 839                 tty_write_message(current->signal->tty, ": warning, ");
 840         else
 841                 tty_write_message(current->signal->tty, ": write failed, ");
 842         tty_write_message(current->signal->tty, quotatypes[dquot->dq_type]);
 843         switch (warntype) {
 844                 case IHARDWARN:
 845                         msg = " file limit reached.\r\n";
 846                         break;
 847                 case ISOFTLONGWARN:
 848                         msg = " file quota exceeded too long.\r\n";
 849                         break;
 850                 case ISOFTWARN:
 851                         msg = " file quota exceeded.\r\n";
 852                         break;
 853                 case BHARDWARN:
 854                         msg = " block limit reached.\r\n";
 855                         break;
 856                 case BSOFTLONGWARN:
 857                         msg = " block quota exceeded too long.\r\n";
 858                         break;
 859                 case BSOFTWARN:
 860                         msg = " block quota exceeded.\r\n";
 861                         break;
 862         }
 863         tty_write_message(current->signal->tty, msg);
 864 }
 865
 866 static inline void flush_warnings(struct dquot **dquots, char *warntype)
 867 {
 868         int i;
 869
 870         for (i = 0; i < MAXQUOTAS; i++)
 871                 if (dquots[i] != NODQUOT && warntype[i] != NOWARN)
 872                         print_warning(dquots[i], warntype[i]);
 873 }
 874
 875 static inline char ignore_hardlimit(struct dquot *dquot)
 876 {
 877         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
 878
 879         return capable(CAP_SYS_RESOURCE) &&
 880             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
 881 }
 882
 883 /* needs dq_data_lock */
 884 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
 885 {
 886         *warntype = NOWARN;
 887         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 888                 return QUOTA_OK;
 889
 890         if (dquot->dq_dqb.dqb_ihardlimit &&
 891            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
 892             !ignore_hardlimit(dquot)) {
 893                 *warntype = IHARDWARN;
 894                 return NO_QUOTA;
 895         }
 896
 897         if (dquot->dq_dqb.dqb_isoftlimit &&
 898            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 899             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
 900             !ignore_hardlimit(dquot)) {
 901                 *warntype = ISOFTLONGWARN;
 902                 return NO_QUOTA;
 903         }
 904
 905         if (dquot->dq_dqb.dqb_isoftlimit &&
 906            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
 907             dquot->dq_dqb.dqb_itime == 0) {
 908                 *warntype = ISOFTWARN;
 909                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
 910         }
 911
 912         return QUOTA_OK;
 913 }
 914
 915 /* needs dq_data_lock */
 916 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
 917 {
 918         *warntype = 0;
 919         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
 920                 return QUOTA_OK;
 921
 922         if (dquot->dq_dqb.dqb_bhardlimit &&
 923            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
 924             !ignore_hardlimit(dquot)) {
 925                 if (!prealloc)
 926                         *warntype = BHARDWARN;
 927                 return NO_QUOTA;
 928         }
 929
 930         if (dquot->dq_dqb.dqb_bsoftlimit &&
 931            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 932             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
 933             !ignore_hardlimit(dquot)) {
 934                 if (!prealloc)
 935                         *warntype = BSOFTLONGWARN;
 936                 return NO_QUOTA;
 937         }
 938
 939         if (dquot->dq_dqb.dqb_bsoftlimit &&
 940            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
 941             dquot->dq_dqb.dqb_btime == 0) {
 942                 if (!prealloc) {
 943                         *warntype = BSOFTWARN;
 944                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
 945                 }
 946                 else
 947                         /*
 948                          * We don't allow preallocation to exceed softlimit so exceeding will
 949                          * be always printed
 950                          */
 951                         return NO_QUOTA;
 952         }
 953
 954         return QUOTA_OK;
 955 }
 956
 957 /*
 958  *      Initialize quota pointers in inode
 959  *      Transaction must be started at entry
 960  */
 961 int dquot_initialize(struct inode *inode, int type)
 962 {
 963         unsigned int id = 0;
 964         int cnt, ret = 0;
 965
 966         /* First test before acquiring mutex - solves deadlocks when we
 967          * re-enter the quota code and are already holding the mutex */
 968         if (IS_NOQUOTA(inode))
 969                 return 0;
 970         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 971         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
 972         if (IS_NOQUOTA(inode))
 973                 goto out_err;
 974         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 975                 if (type != -1 && cnt != type)
 976                         continue;
 977                 if (inode->i_dquot[cnt] == NODQUOT) {
 978                         switch (cnt) {
 979                                 case USRQUOTA:
 980                                         id = inode->i_uid;
 981                                         break;
 982                                 case GRPQUOTA:
 983                                         id = inode->i_gid;
 984                                         break;
 985                         }
 986                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
 987                 }
 988         }
 989 out_err:
 990         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
 991         return ret;
 992 }
 993
 994 /*
 995  *      Release all quotas referenced by inode
 996  *      Transaction must be started at an entry
 997  */
 998 int dquot_drop(struct inode *inode)
 999 {
1000         int cnt;
1001
1002         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1003         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1004                 if (inode->i_dquot[cnt] != NODQUOT) {
1005                         dqput(inode->i_dquot[cnt]);
1006                         inode->i_dquot[cnt] = NODQUOT;
1007                 }
1008         }
1009         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1010         return 0;
1011 }
1012
1013 /*
1014  * Following four functions update i_blocks+i_bytes fields and
1015  * quota information (together with appropriate checks)
1016  * NOTE: We absolutely rely on the fact that caller dirties
1017  * the inode (usually macros in quotaops.h care about this) and
1018  * holds a handle for the current transaction so that dquot write and
1019  * inode write go into the same transaction.
1020  */
1021
1022 /*
1023  * This operation can block, but only after everything is updated
1024  */
1025 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1026 {
1027         int cnt, ret = NO_QUOTA;
1028         char warntype[MAXQUOTAS];
1029
1030         /* First test before acquiring mutex - solves deadlocks when we
1031          * re-enter the quota code and are already holding the mutex */
1032         if (IS_NOQUOTA(inode)) {
1033 out_add:
1034                 inode_add_bytes(inode, number);
1035                 return QUOTA_OK;
1036         }
1037         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1038                 warntype[cnt] = NOWARN;
1039
1040         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1041         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1042                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1043                 goto out_add;
1044         }
1045         spin_lock(&dq_data_lock);
1046         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1047                 if (inode->i_dquot[cnt] == NODQUOT)
1048                         continue;
1049                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1050                         goto warn_put_all;
1051         }
1052         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1053                 if (inode->i_dquot[cnt] == NODQUOT)
1054                         continue;
1055                 dquot_incr_space(inode->i_dquot[cnt], number);
1056         }
1057         inode_add_bytes(inode, number);
1058         ret = QUOTA_OK;
1059 warn_put_all:
1060         spin_unlock(&dq_data_lock);
1061         if (ret == QUOTA_OK)
1062                 /* Dirtify all the dquots - this can block when journalling */
1063                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1064                         if (inode->i_dquot[cnt])
1065                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1066         flush_warnings(inode->i_dquot, warntype);
1067         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1068         return ret;
1069 }
1070
1071 /*
1072  * This operation can block, but only after everything is updated
1073  */
1074 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1075 {
1076         int cnt, ret = NO_QUOTA;
1077         char warntype[MAXQUOTAS];
1078
1079         /* First test before acquiring mutex - solves deadlocks when we
1080          * re-enter the quota code and are already holding the mutex */
1081         if (IS_NOQUOTA(inode))
1082                 return QUOTA_OK;
1083         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1084                 warntype[cnt] = NOWARN;
1085         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1086         if (IS_NOQUOTA(inode)) {
1087                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1088                 return QUOTA_OK;
1089         }
1090         spin_lock(&dq_data_lock);
1091         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1092                 if (inode->i_dquot[cnt] == NODQUOT)
1093                         continue;
1094                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1095                         goto warn_put_all;
1096         }
1097
1098         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1099                 if (inode->i_dquot[cnt] == NODQUOT)
1100                         continue;
1101                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1102         }
1103         ret = QUOTA_OK;
1104 warn_put_all:
1105         spin_unlock(&dq_data_lock);
1106         if (ret == QUOTA_OK)
1107                 /* Dirtify all the dquots - this can block when journalling */
1108                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1109                         if (inode->i_dquot[cnt])
1110                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1111         flush_warnings((struct dquot **)inode->i_dquot, warntype);
1112         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1113         return ret;
1114 }
1115
1116 /*
1117  * This operation can block, but only after everything is updated
1118  */
1119 int dquot_free_space(struct inode *inode, qsize_t number)
1120 {
1121         unsigned int cnt;
1122
1123         /* First test before acquiring mutex - solves deadlocks when we
1124          * re-enter the quota code and are already holding the mutex */
1125         if (IS_NOQUOTA(inode)) {
1126 out_sub:
1127                 inode_sub_bytes(inode, number);
1128                 return QUOTA_OK;
1129         }
1130         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1131         /* Now recheck reliably when holding dqptr_sem */
1132         if (IS_NOQUOTA(inode)) {
1133                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1134                 goto out_sub;
1135         }
1136         spin_lock(&dq_data_lock);
1137         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1138                 if (inode->i_dquot[cnt] == NODQUOT)
1139                         continue;
1140                 dquot_decr_space(inode->i_dquot[cnt], number);
1141         }
1142         inode_sub_bytes(inode, number);
1143         spin_unlock(&dq_data_lock);
1144         /* Dirtify all the dquots - this can block when journalling */
1145         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1146                 if (inode->i_dquot[cnt])
1147                         mark_dquot_dirty(inode->i_dquot[cnt]);
1148         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1149         return QUOTA_OK;
1150 }
1151
1152 /*
1153  * This operation can block, but only after everything is updated
1154  */
1155 int dquot_free_inode(const struct inode *inode, unsigned long number)
1156 {
1157         unsigned int cnt;
1158
1159         /* First test before acquiring mutex - solves deadlocks when we
1160          * re-enter the quota code and are already holding the mutex */
1161         if (IS_NOQUOTA(inode))
1162                 return QUOTA_OK;
1163         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1164         /* Now recheck reliably when holding dqptr_sem */
1165         if (IS_NOQUOTA(inode)) {
1166                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1167                 return QUOTA_OK;
1168         }
1169         spin_lock(&dq_data_lock);
1170         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1171                 if (inode->i_dquot[cnt] == NODQUOT)
1172                         continue;
1173                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1174         }
1175         spin_unlock(&dq_data_lock);
1176         /* Dirtify all the dquots - this can block when journalling */
1177         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1178                 if (inode->i_dquot[cnt])
1179                         mark_dquot_dirty(inode->i_dquot[cnt]);
1180         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1181         return QUOTA_OK;
1182 }
1183
1184 /*
1185  * Transfer the number of inode and blocks from one diskquota to an other.
1186  *
1187  * This operation can block, but only after everything is updated
1188  * A transaction must be started when entering this function.
1189  */
1190 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1191 {
1192         qsize_t space;
1193         struct dquot *transfer_from[MAXQUOTAS];
1194         struct dquot *transfer_to[MAXQUOTAS];
1195         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1196             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1197         char warntype[MAXQUOTAS];
1198
1199         /* First test before acquiring mutex - solves deadlocks when we
1200          * re-enter the quota code and are already holding the mutex */
1201         if (IS_NOQUOTA(inode))
1202                 return QUOTA_OK;
1203         /* Clear the arrays */
1204         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1205                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1206                 warntype[cnt] = NOWARN;
1207         }
1208         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1209         /* Now recheck reliably when holding dqptr_sem */
1210         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1211                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1212                 return QUOTA_OK;
1213         }
1214         /* First build the transfer_to list - here we can block on
1215          * reading/instantiating of dquots.  We know that the transaction for
1216          * us was already started so we don't violate lock ranking here */
1217         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1218                 switch (cnt) {
1219                         case USRQUOTA:
1220                                 if (!chuid)
1221                                         continue;
1222                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1223                                 break;
1224                         case GRPQUOTA:
1225                                 if (!chgid)
1226                                         continue;
1227                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1228                                 break;
1229                 }
1230         }
1231         spin_lock(&dq_data_lock);
1232         space = inode_get_bytes(inode);
1233         /* Build the transfer_from list and check the limits */
1234         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1235                 if (transfer_to[cnt] == NODQUOT)
1236                         continue;
1237                 transfer_from[cnt] = inode->i_dquot[cnt];
1238                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1239                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1240                         goto warn_put_all;
1241         }
1242
1243         /*
1244          * Finally perform the needed transfer from transfer_from to transfer_to
1245          */
1246         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1247                 /*
1248                  * Skip changes for same uid or gid or for turned off quota-type.
1249                  */
1250                 if (transfer_to[cnt] == NODQUOT)
1251                         continue;
1252
1253                 /* Due to IO error we might not have transfer_from[] structure */
1254                 if (transfer_from[cnt]) {
1255                         dquot_decr_inodes(transfer_from[cnt], 1);
1256                         dquot_decr_space(transfer_from[cnt], space);
1257                 }
1258
1259                 dquot_incr_inodes(transfer_to[cnt], 1);
1260                 dquot_incr_space(transfer_to[cnt], space);
1261
1262                 inode->i_dquot[cnt] = transfer_to[cnt];
1263         }
1264         ret = QUOTA_OK;
1265 warn_put_all:
1266         spin_unlock(&dq_data_lock);
1267         /* Dirtify all the dquots - this can block when journalling */
1268         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1269                 if (transfer_from[cnt])
1270                         mark_dquot_dirty(transfer_from[cnt]);
1271                 if (transfer_to[cnt])
1272                         mark_dquot_dirty(transfer_to[cnt]);
1273         }
1274         flush_warnings(transfer_to, warntype);
1275
1276         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1277                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1278                         dqput(transfer_from[cnt]);
1279                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1280                         dqput(transfer_to[cnt]);
1281         }
1282         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1283         return ret;
1284 }
1285
1286 /*
1287  * Write info of quota file to disk
1288  */
1289 int dquot_commit_info(struct super_block *sb, int type)
1290 {
1291         int ret;
1292         struct quota_info *dqopt = sb_dqopt(sb);
1293
1294         mutex_lock(&dqopt->dqio_mutex);
1295         ret = dqopt->ops[type]->write_file_info(sb, type);
1296         mutex_unlock(&dqopt->dqio_mutex);
1297         return ret;
1298 }
1299
1300 /*
1301  * Definitions of diskquota operations.
1302  */
1303 struct dquot_operations dquot_operations = {
1304         .initialize     = dquot_initialize,
1305         .drop           = dquot_drop,
1306         .alloc_space    = dquot_alloc_space,
1307         .alloc_inode    = dquot_alloc_inode,
1308         .free_space     = dquot_free_space,
1309         .free_inode     = dquot_free_inode,
1310         .transfer       = dquot_transfer,
1311         .write_dquot    = dquot_commit,
1312         .acquire_dquot  = dquot_acquire,
1313         .release_dquot  = dquot_release,
1314         .mark_dirty     = dquot_mark_dquot_dirty,
1315         .write_info     = dquot_commit_info
1316 };
1317
1318 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1319 {
1320         switch (type) {
1321                 case USRQUOTA:
1322                         dqopt->flags |= DQUOT_USR_ENABLED;
1323                         break;
1324                 case GRPQUOTA:
1325                         dqopt->flags |= DQUOT_GRP_ENABLED;
1326                         break;
1327         }
1328 }
1329
1330 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1331 {
1332         switch (type) {
1333                 case USRQUOTA:
1334                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1335                         break;
1336                 case GRPQUOTA:
1337                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1338                         break;
1339         }
1340 }
1341
1342 /*
1343  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1344  */
1345 int vfs_quota_off(struct super_block *sb, int type)
1346 {
1347         int cnt;
1348         struct quota_info *dqopt = sb_dqopt(sb);
1349         struct inode *toputinode[MAXQUOTAS];
1350
1351         /* We need to serialize quota_off() for device */
1352         mutex_lock(&dqopt->dqonoff_mutex);
1353         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1354                 toputinode[cnt] = NULL;
1355                 if (type != -1 && cnt != type)
1356                         continue;
1357                 if (!sb_has_quota_enabled(sb, cnt))
1358                         continue;
1359                 reset_enable_flags(dqopt, cnt);
1360
1361                 /* Note: these are blocking operations */
1362                 drop_dquot_ref(sb, cnt);
1363                 invalidate_dquots(sb, cnt);
1364                 /*
1365                  * Now all dquots should be invalidated, all writes done so we should be only
1366                  * users of the info. No locks needed.
1367                  */
1368                 if (info_dirty(&dqopt->info[cnt]))
1369                         sb->dq_op->write_info(sb, cnt);
1370                 if (dqopt->ops[cnt]->free_file_info)
1371                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1372                 put_quota_format(dqopt->info[cnt].dqi_format);
1373
1374                 toputinode[cnt] = dqopt->files[cnt];
1375                 dqopt->files[cnt] = NULL;
1376                 dqopt->info[cnt].dqi_flags = 0;
1377                 dqopt->info[cnt].dqi_igrace = 0;
1378                 dqopt->info[cnt].dqi_bgrace = 0;
1379                 dqopt->ops[cnt] = NULL;
1380         }
1381         mutex_unlock(&dqopt->dqonoff_mutex);
1382         /* Sync the superblock so that buffers with quota data are written to
1383          * disk (and so userspace sees correct data afterwards). */
1384         if (sb->s_op->sync_fs)
1385                 sb->s_op->sync_fs(sb, 1);
1386         sync_blockdev(sb->s_bdev);
1387         /* Now the quota files are just ordinary files and we can set the
1388          * inode flags back. Moreover we discard the pagecache so that
1389          * userspace sees the writes we did bypassing the pagecache. We
1390          * must also discard the blockdev buffers so that we see the
1391          * changes done by userspace on the next quotaon() */
1392         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1393                 if (toputinode[cnt]) {
1394                         mutex_lock(&dqopt->dqonoff_mutex);
1395                         /* If quota was reenabled in the meantime, we have
1396                          * nothing to do */
1397                         if (!sb_has_quota_enabled(sb, cnt)) {
1398                                 mutex_lock(&toputinode[cnt]->i_mutex);
1399                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1400                                   S_NOATIME | S_NOQUOTA);
1401                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1402                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1403                                 mark_inode_dirty(toputinode[cnt]);
1404                                 iput(toputinode[cnt]);
1405                         }
1406                         mutex_unlock(&dqopt->dqonoff_mutex);
1407                 }
1408         if (sb->s_bdev)
1409                 invalidate_bdev(sb->s_bdev, 0);
1410         return 0;
1411 }
1412
1413 /*
1414  *      Turn quotas on on a device
1415  */
1416
1417 /* Helper function when we already have the inode */
1418 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1419 {
1420         struct quota_format_type *fmt = find_quota_format(format_id);
1421         struct super_block *sb = inode->i_sb;
1422         struct quota_info *dqopt = sb_dqopt(sb);
1423         int error;
1424         int oldflags = -1;
1425
1426         if (!fmt)
1427                 return -ESRCH;
1428         if (!S_ISREG(inode->i_mode)) {
1429                 error = -EACCES;
1430                 goto out_fmt;
1431         }
1432         if (IS_RDONLY(inode)) {
1433                 error = -EROFS;
1434                 goto out_fmt;
1435         }
1436         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1437                 error = -EINVAL;
1438                 goto out_fmt;
1439         }
1440
1441         /* As we bypass the pagecache we must now flush the inode so that
1442          * we see all the changes from userspace... */
1443         write_inode_now(inode, 1);
1444         /* And now flush the block cache so that kernel sees the changes */
1445         invalidate_bdev(sb->s_bdev, 0);
1446         mutex_lock(&inode->i_mutex);
1447         mutex_lock(&dqopt->dqonoff_mutex);
1448         if (sb_has_quota_enabled(sb, type)) {
1449                 error = -EBUSY;
1450                 goto out_lock;
1451         }
1452         /* We don't want quota and atime on quota files (deadlocks possible)
1453          * Also nobody should write to the file - we use special IO operations
1454          * which ignore the immutable bit. */
1455         down_write(&dqopt->dqptr_sem);
1456         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1457         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1458         up_write(&dqopt->dqptr_sem);
1459         sb->dq_op->drop(inode);
1460
1461         error = -EIO;
1462         dqopt->files[type] = igrab(inode);
1463         if (!dqopt->files[type])
1464                 goto out_lock;
1465         error = -EINVAL;
1466         if (!fmt->qf_ops->check_quota_file(sb, type))
1467                 goto out_file_init;
1468
1469         dqopt->ops[type] = fmt->qf_ops;
1470         dqopt->info[type].dqi_format = fmt;
1471         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1472         mutex_lock(&dqopt->dqio_mutex);
1473         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1474                 mutex_unlock(&dqopt->dqio_mutex);
1475                 goto out_file_init;
1476         }
1477         mutex_unlock(&dqopt->dqio_mutex);
1478         mutex_unlock(&inode->i_mutex);
1479         set_enable_flags(dqopt, type);
1480
1481         add_dquot_ref(sb, type);
1482         mutex_unlock(&dqopt->dqonoff_mutex);
1483
1484         return 0;
1485
1486 out_file_init:
1487         dqopt->files[type] = NULL;
1488         iput(inode);
1489 out_lock:
1490         mutex_unlock(&dqopt->dqonoff_mutex);
1491         if (oldflags != -1) {
1492                 down_write(&dqopt->dqptr_sem);
1493                 /* Set the flags back (in the case of accidental quotaon()
1494                  * on a wrong file we don't want to mess up the flags) */
1495                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1496                 inode->i_flags |= oldflags;
1497                 up_write(&dqopt->dqptr_sem);
1498         }
1499         mutex_unlock(&inode->i_mutex);
1500 out_fmt:
1501         put_quota_format(fmt);
1502
1503         return error;
1504 }
1505
1506 /* Actual function called from quotactl() */
1507 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1508 {
1509         struct nameidata nd;
1510         int error;
1511
1512         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1513         if (error < 0)
1514                 return error;
1515         error = security_quota_on(nd.dentry);
1516         if (error)
1517                 goto out_path;
1518         /* Quota file not on the same filesystem? */
1519         if (nd.mnt->mnt_sb != sb)
1520                 error = -EXDEV;
1521         else
1522                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1523 out_path:
1524         path_release(&nd);
1525         return error;
1526 }
1527
1528 /*
1529  * This function is used when filesystem needs to initialize quotas
1530  * during mount time.
1531  */
1532 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1533                 int format_id, int type)
1534 {
1535         struct dentry *dentry;
1536         int error;
1537
1538         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1539         if (IS_ERR(dentry))
1540                 return PTR_ERR(dentry);
1541
1542         if (!dentry->d_inode) {
1543                 error = -ENOENT;
1544                 goto out;
1545         }
1546
1547         error = security_quota_on(dentry);
1548         if (!error)
1549                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1550
1551 out:
1552         dput(dentry);
1553         return error;
1554 }
1555
1556 /* Generic routine for getting common part of quota structure */
1557 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1558 {
1559         struct mem_dqblk *dm = &dquot->dq_dqb;
1560
1561         spin_lock(&dq_data_lock);
1562         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1563         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1564         di->dqb_curspace = dm->dqb_curspace;
1565         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1566         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1567         di->dqb_curinodes = dm->dqb_curinodes;
1568         di->dqb_btime = dm->dqb_btime;
1569         di->dqb_itime = dm->dqb_itime;
1570         di->dqb_valid = QIF_ALL;
1571         spin_unlock(&dq_data_lock);
1572 }
1573
1574 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1575 {
1576         struct dquot *dquot;
1577
1578         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1579         if (!(dquot = dqget(sb, id, type))) {
1580                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1581                 return -ESRCH;
1582         }
1583         do_get_dqblk(dquot, di);
1584         dqput(dquot);
1585         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1586         return 0;
1587 }
1588
1589 /* Generic routine for setting common part of quota structure */
1590 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1591 {
1592         struct mem_dqblk *dm = &dquot->dq_dqb;
1593         int check_blim = 0, check_ilim = 0;
1594
1595         spin_lock(&dq_data_lock);
1596         if (di->dqb_valid & QIF_SPACE) {
1597                 dm->dqb_curspace = di->dqb_curspace;
1598                 check_blim = 1;
1599         }
1600         if (di->dqb_valid & QIF_BLIMITS) {
1601                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1602                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1603                 check_blim = 1;
1604         }
1605         if (di->dqb_valid & QIF_INODES) {
1606                 dm->dqb_curinodes = di->dqb_curinodes;
1607                 check_ilim = 1;
1608         }
1609         if (di->dqb_valid & QIF_ILIMITS) {
1610                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1611                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1612                 check_ilim = 1;
1613         }
1614         if (di->dqb_valid & QIF_BTIME)
1615                 dm->dqb_btime = di->dqb_btime;
1616         if (di->dqb_valid & QIF_ITIME)
1617                 dm->dqb_itime = di->dqb_itime;
1618
1619         if (check_blim) {
1620                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1621                         dm->dqb_btime = 0;
1622                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1623                 }
1624                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1625                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1626         }
1627         if (check_ilim) {
1628                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1629                         dm->dqb_itime = 0;
1630                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1631                 }
1632                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1633                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1634         }
1635         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1636                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1637         else
1638                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1639         spin_unlock(&dq_data_lock);
1640         mark_dquot_dirty(dquot);
1641 }
1642
1643 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1644 {
1645         struct dquot *dquot;
1646
1647         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1648         if (!(dquot = dqget(sb, id, type))) {
1649                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1650                 return -ESRCH;
1651         }
1652         do_set_dqblk(dquot, di);
1653         dqput(dquot);
1654         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1655         return 0;
1656 }
1657
1658 /* Generic routine for getting common part of quota file information */
1659 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1660 {
1661         struct mem_dqinfo *mi;
1662
1663         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1664         if (!sb_has_quota_enabled(sb, type)) {
1665                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1666                 return -ESRCH;
1667         }
1668         mi = sb_dqopt(sb)->info + type;
1669         spin_lock(&dq_data_lock);
1670         ii->dqi_bgrace = mi->dqi_bgrace;
1671         ii->dqi_igrace = mi->dqi_igrace;
1672         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1673         ii->dqi_valid = IIF_ALL;
1674         spin_unlock(&dq_data_lock);
1675         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1676         return 0;
1677 }
1678
1679 /* Generic routine for setting common part of quota file information */
1680 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1681 {
1682         struct mem_dqinfo *mi;
1683
1684         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1685         if (!sb_has_quota_enabled(sb, type)) {
1686                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1687                 return -ESRCH;
1688         }
1689         mi = sb_dqopt(sb)->info + type;
1690         spin_lock(&dq_data_lock);
1691         if (ii->dqi_valid & IIF_BGRACE)
1692                 mi->dqi_bgrace = ii->dqi_bgrace;
1693         if (ii->dqi_valid & IIF_IGRACE)
1694                 mi->dqi_igrace = ii->dqi_igrace;
1695         if (ii->dqi_valid & IIF_FLAGS)
1696                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1697         spin_unlock(&dq_data_lock);
1698         mark_info_dirty(sb, type);
1699         /* Force write to disk */
1700         sb->dq_op->write_info(sb, type);
1701         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1702         return 0;
1703 }
1704
1705 struct quotactl_ops vfs_quotactl_ops = {
1706         .quota_on       = vfs_quota_on,
1707         .quota_off      = vfs_quota_off,
1708         .quota_sync     = vfs_quota_sync,
1709         .get_info       = vfs_get_dqinfo,
1710         .set_info       = vfs_set_dqinfo,
1711         .get_dqblk      = vfs_get_dqblk,
1712         .set_dqblk      = vfs_set_dqblk
1713 };
1714
1715 static ctl_table fs_dqstats_table[] = {
1716         {
1717                 .ctl_name       = FS_DQ_LOOKUPS,
1718                 .procname       = "lookups",
1719                 .data           = &dqstats.lookups,
1720                 .maxlen         = sizeof(int),
1721                 .mode           = 0444,
1722                 .proc_handler   = &proc_dointvec,
1723         },
1724         {
1725                 .ctl_name       = FS_DQ_DROPS,
1726                 .procname       = "drops",
1727                 .data           = &dqstats.drops,
1728                 .maxlen         = sizeof(int),
1729                 .mode           = 0444,
1730                 .proc_handler   = &proc_dointvec,
1731         },
1732         {
1733                 .ctl_name       = FS_DQ_READS,
1734                 .procname       = "reads",
1735                 .data           = &dqstats.reads,
1736                 .maxlen         = sizeof(int),
1737                 .mode           = 0444,
1738                 .proc_handler   = &proc_dointvec,
1739         },
1740         {
1741                 .ctl_name       = FS_DQ_WRITES,
1742                 .procname       = "writes",
1743                 .data           = &dqstats.writes,
1744                 .maxlen         = sizeof(int),
1745                 .mode           = 0444,
1746                 .proc_handler   = &proc_dointvec,
1747         },
1748         {
1749                 .ctl_name       = FS_DQ_CACHE_HITS,
1750                 .procname       = "cache_hits",
1751                 .data           = &dqstats.cache_hits,
1752                 .maxlen         = sizeof(int),
1753                 .mode           = 0444,
1754                 .proc_handler   = &proc_dointvec,
1755         },
1756         {
1757                 .ctl_name       = FS_DQ_ALLOCATED,
1758                 .procname       = "allocated_dquots",
1759                 .data           = &dqstats.allocated_dquots,
1760                 .maxlen         = sizeof(int),
1761                 .mode           = 0444,
1762                 .proc_handler   = &proc_dointvec,
1763         },
1764         {
1765                 .ctl_name       = FS_DQ_FREE,
1766                 .procname       = "free_dquots",
1767                 .data           = &dqstats.free_dquots,
1768                 .maxlen         = sizeof(int),
1769                 .mode           = 0444,
1770                 .proc_handler   = &proc_dointvec,
1771         },
1772         {
1773                 .ctl_name       = FS_DQ_SYNCS,
1774                 .procname       = "syncs",
1775                 .data           = &dqstats.syncs,
1776                 .maxlen         = sizeof(int),
1777                 .mode           = 0444,
1778                 .proc_handler   = &proc_dointvec,
1779         },
1780         {
1781                 .ctl_name       = FS_DQ_WARNINGS,
1782                 .procname       = "warnings",
1783                 .data           = &flag_print_warnings,
1784                 .maxlen         = sizeof(int),
1785                 .mode           = 0644,
1786                 .proc_handler   = &proc_dointvec,
1787         },
1788         { .ctl_name = 0 },
1789 };
1790
1791 static ctl_table fs_table[] = {
1792         {
1793                 .ctl_name       = FS_DQSTATS,
1794                 .procname       = "quota",
1795                 .mode           = 0555,
1796                 .child          = fs_dqstats_table,
1797         },
1798         { .ctl_name = 0 },
1799 };
1800
1801 static ctl_table sys_table[] = {
1802         {
1803                 .ctl_name       = CTL_FS,
1804                 .procname       = "fs",
1805                 .mode           = 0555,
1806                 .child          = fs_table,
1807         },
1808         { .ctl_name = 0 },
1809 };
1810
1811 static int __init dquot_init(void)
1812 {
1813         int i;
1814         unsigned long nr_hash, order;
1815
1816         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1817
1818         register_sysctl_table(sys_table, 0);
1819
1820         dquot_cachep = kmem_cache_create("dquot",
1821                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1822                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1823                                 SLAB_MEM_SPREAD|SLAB_PANIC),
1824                         NULL, NULL);
1825
1826         order = 0;
1827         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1828         if (!dquot_hash)
1829                 panic("Cannot create dquot hash table");
1830
1831         /* Find power-of-two hlist_heads which can fit into allocation */
1832         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1833         dq_hash_bits = 0;
1834         do {
1835                 dq_hash_bits++;
1836         } while (nr_hash >> dq_hash_bits);
1837         dq_hash_bits--;
1838
1839         nr_hash = 1UL << dq_hash_bits;
1840         dq_hash_mask = nr_hash - 1;
1841         for (i = 0; i < nr_hash; i++)
1842                 INIT_HLIST_HEAD(dquot_hash + i);
1843
1844         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1845                         nr_hash, order, (PAGE_SIZE << order));
1846
1847         set_shrinker(DEFAULT_SEEKS, shrink_dqcache_memory);
1848
1849         return 0;
1850 }
1851 module_init(dquot_init);
1852
1853 EXPORT_SYMBOL(register_quota_format);
1854 EXPORT_SYMBOL(unregister_quota_format);
1855 EXPORT_SYMBOL(dqstats);
1856 EXPORT_SYMBOL(dq_data_lock);
1857 EXPORT_SYMBOL(vfs_quota_on);
1858 EXPORT_SYMBOL(vfs_quota_on_mount);
1859 EXPORT_SYMBOL(vfs_quota_off);
1860 EXPORT_SYMBOL(vfs_quota_sync);
1861 EXPORT_SYMBOL(vfs_get_dqinfo);
1862 EXPORT_SYMBOL(vfs_set_dqinfo);
1863 EXPORT_SYMBOL(vfs_get_dqblk);
1864 EXPORT_SYMBOL(vfs_set_dqblk);
1865 EXPORT_SYMBOL(dquot_commit);
1866 EXPORT_SYMBOL(dquot_commit_info);
1867 EXPORT_SYMBOL(dquot_acquire);
1868 EXPORT_SYMBOL(dquot_release);
1869 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1870 EXPORT_SYMBOL(dquot_initialize);
1871 EXPORT_SYMBOL(dquot_drop);
1872 EXPORT_SYMBOL(dquot_alloc_space);
1873 EXPORT_SYMBOL(dquot_alloc_inode);
1874 EXPORT_SYMBOL(dquot_free_space);
1875 EXPORT_SYMBOL(dquot_free_inode);
1876 EXPORT_SYMBOL(dquot_transfer);