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