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/init.h>
  73 #include <linux/module.h>
  74 #include <linux/proc_fs.h>
  75 #include <linux/security.h>
  76 #include <linux/kmod.h>
  77 #include <linux/namei.h>
  78 #include <linux/buffer_head.h>
  79 #include <linux/capability.h>
  80 #include <linux/quotaops.h>
  81 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
  82 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
  83 #include <net/netlink.h>
  84 #include <net/genetlink.h>
  85 #endif
  86
  87 #include <asm/uaccess.h>
  88
  89 #define __DQUOT_PARANOIA
  90
  91 /*
  92  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
  93  * and quota formats and also dqstats structure containing statistics about the
  94  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
  95  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
  96  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
  97  * in inode_add_bytes() and inode_sub_bytes().
  98  *
  99  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
 100  *
 101  * Note that some things (eg. sb pointer, type, id) doesn't change during
 102  * the life of the dquot structure and so needn't to be protected by a lock
 103  *
 104  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
 105  * operation is just reading pointers from inode (or not using them at all) the
 106  * read lock is enough. If pointers are altered function must hold write lock
 107  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
 108  * for altering the flag i_mutex is also needed).  If operation is holding
 109  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
 110  * dqonoff_mutex.
 111  * This locking assures that:
 112  *   a) update/access to dquot pointers in inode is serialized
 113  *   b) everyone is guarded against invalidate_dquots()
 114  *
 115  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
 116  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
 117  * Currently dquot is locked only when it is being read to memory (or space for
 118  * it is being allocated) on the first dqget() and when it is being released on
 119  * the last dqput(). The allocation and release oparations are serialized by
 120  * the dq_lock and by checking the use count in dquot_release().  Write
 121  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
 122  * spinlock to internal buffers before writing.
 123  *
 124  * Lock ordering (including related VFS locks) is the following:
 125  *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
 126  *   dqio_mutex
 127  * i_mutex on quota files is special (it's below dqio_mutex)
 128  */
 129
 130 static DEFINE_SPINLOCK(dq_list_lock);
 131 DEFINE_SPINLOCK(dq_data_lock);
 132
 133 static char *quotatypes[] = INITQFNAMES;
 134 static struct quota_format_type *quota_formats; /* List of registered formats */
 135 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 136
 137 /* SLAB cache for dquot structures */
 138 static struct kmem_cache *dquot_cachep;
 139
 140 int register_quota_format(struct quota_format_type *fmt)
 141 {
 142         spin_lock(&dq_list_lock);
 143         fmt->qf_next = quota_formats;
 144         quota_formats = fmt;
 145         spin_unlock(&dq_list_lock);
 146         return 0;
 147 }
 148
 149 void unregister_quota_format(struct quota_format_type *fmt)
 150 {
 151         struct quota_format_type **actqf;
 152
 153         spin_lock(&dq_list_lock);
 154         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
 155         if (*actqf)
 156                 *actqf = (*actqf)->qf_next;
 157         spin_unlock(&dq_list_lock);
 158 }
 159
 160 static struct quota_format_type *find_quota_format(int id)
 161 {
 162         struct quota_format_type *actqf;
 163
 164         spin_lock(&dq_list_lock);
 165         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 166         if (!actqf || !try_module_get(actqf->qf_owner)) {
 167                 int qm;
 168
 169                 spin_unlock(&dq_list_lock);
 170
 171                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
 172                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
 173                         return NULL;
 174
 175                 spin_lock(&dq_list_lock);
 176                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 177                 if (actqf && !try_module_get(actqf->qf_owner))
 178                         actqf = NULL;
 179         }
 180         spin_unlock(&dq_list_lock);
 181         return actqf;
 182 }
 183
 184 static void put_quota_format(struct quota_format_type *fmt)
 185 {
 186         module_put(fmt->qf_owner);
 187 }
 188
 189 /*
 190  * Dquot List Management:
 191  * The quota code uses three lists for dquot management: the inuse_list,
 192  * free_dquots, and dquot_hash[] array. A single dquot structure may be
 193  * on all three lists, depending on its current state.
 194  *
 195  * All dquots are placed to the end of inuse_list when first created, and this
 196  * list is used for invalidate operation, which must look at every dquot.
 197  *
 198  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 199  * and this list is searched whenever we need an available dquot.  Dquots are
 200  * removed from the list as soon as they are used again, and
 201  * dqstats.free_dquots gives the number of dquots on the list. When
 202  * dquot is invalidated it's completely released from memory.
 203  *
 204  * Dquots with a specific identity (device, type and id) are placed on
 205  * one of the dquot_hash[] hash chains. The provides an efficient search
 206  * mechanism to locate a specific dquot.
 207  */
 208
 209 static LIST_HEAD(inuse_list);
 210 static LIST_HEAD(free_dquots);
 211 static unsigned int dq_hash_bits, dq_hash_mask;
 212 static struct hlist_head *dquot_hash;
 213
 214 struct dqstats dqstats;
 215
 216 static void dqput(struct dquot *dquot);
 217
 218 static inline unsigned int
 219 hashfn(const struct super_block *sb, unsigned int id, int type)
 220 {
 221         unsigned long tmp;
 222
 223         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 224         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 225 }
 226
 227 /*
 228  * Following list functions expect dq_list_lock to be held
 229  */
 230 static inline void insert_dquot_hash(struct dquot *dquot)
 231 {
 232         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
 233         hlist_add_head(&dquot->dq_hash, head);
 234 }
 235
 236 static inline void remove_dquot_hash(struct dquot *dquot)
 237 {
 238         hlist_del_init(&dquot->dq_hash);
 239 }
 240
 241 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
 242 {
 243         struct hlist_node *node;
 244         struct dquot *dquot;
 245
 246         hlist_for_each (node, dquot_hash+hashent) {
 247                 dquot = hlist_entry(node, struct dquot, dq_hash);
 248                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
 249                         return dquot;
 250         }
 251         return NODQUOT;
 252 }
 253
 254 /* Add a dquot to the tail of the free list */
 255 static inline void put_dquot_last(struct dquot *dquot)
 256 {
 257         list_add_tail(&dquot->dq_free, &free_dquots);
 258         dqstats.free_dquots++;
 259 }
 260
 261 static inline void remove_free_dquot(struct dquot *dquot)
 262 {
 263         if (list_empty(&dquot->dq_free))
 264                 return;
 265         list_del_init(&dquot->dq_free);
 266         dqstats.free_dquots--;
 267 }
 268
 269 static inline void put_inuse(struct dquot *dquot)
 270 {
 271         /* We add to the back of inuse list so we don't have to restart
 272          * when traversing this list and we block */
 273         list_add_tail(&dquot->dq_inuse, &inuse_list);
 274         dqstats.allocated_dquots++;
 275 }
 276
 277 static inline void remove_inuse(struct dquot *dquot)
 278 {
 279         dqstats.allocated_dquots--;
 280         list_del(&dquot->dq_inuse);
 281 }
 282 /*
 283  * End of list functions needing dq_list_lock
 284  */
 285
 286 static void wait_on_dquot(struct dquot *dquot)
 287 {
 288         mutex_lock(&dquot->dq_lock);
 289         mutex_unlock(&dquot->dq_lock);
 290 }
 291
 292 #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
 293
 294 int dquot_mark_dquot_dirty(struct dquot *dquot)
 295 {
 296         spin_lock(&dq_list_lock);
 297         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
 298                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 299                                 info[dquot->dq_type].dqi_dirty_list);
 300         spin_unlock(&dq_list_lock);
 301         return 0;
 302 }
 303
 304 /* This function needs dq_list_lock */
 305 static inline int clear_dquot_dirty(struct dquot *dquot)
 306 {
 307         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
 308                 return 0;
 309         list_del_init(&dquot->dq_dirty);
 310         return 1;
 311 }
 312
 313 void mark_info_dirty(struct super_block *sb, int type)
 314 {
 315         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
 316 }
 317 EXPORT_SYMBOL(mark_info_dirty);
 318
 319 /*
 320  *      Read dquot from disk and alloc space for it
 321  */
 322
 323 int dquot_acquire(struct dquot *dquot)
 324 {
 325         int ret = 0, ret2 = 0;
 326         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 327
 328         mutex_lock(&dquot->dq_lock);
 329         mutex_lock(&dqopt->dqio_mutex);
 330         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 331                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
 332         if (ret < 0)
 333                 goto out_iolock;
 334         set_bit(DQ_READ_B, &dquot->dq_flags);
 335         /* Instantiate dquot if needed */
 336         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 337                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 338                 /* Write the info if needed */
 339                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 340                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 341                 if (ret < 0)
 342                         goto out_iolock;
 343                 if (ret2 < 0) {
 344                         ret = ret2;
 345                         goto out_iolock;
 346                 }
 347         }
 348         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 349 out_iolock:
 350         mutex_unlock(&dqopt->dqio_mutex);
 351         mutex_unlock(&dquot->dq_lock);
 352         return ret;
 353 }
 354
 355 /*
 356  *      Write dquot to disk
 357  */
 358 int dquot_commit(struct dquot *dquot)
 359 {
 360         int ret = 0, ret2 = 0;
 361         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 362
 363         mutex_lock(&dqopt->dqio_mutex);
 364         spin_lock(&dq_list_lock);
 365         if (!clear_dquot_dirty(dquot)) {
 366                 spin_unlock(&dq_list_lock);
 367                 goto out_sem;
 368         }
 369         spin_unlock(&dq_list_lock);
 370         /* Inactive dquot can be only if there was error during read/init
 371          * => we have better not writing it */
 372         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 373                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 374                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 375                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 376                 if (ret >= 0)
 377                         ret = ret2;
 378         }
 379 out_sem:
 380         mutex_unlock(&dqopt->dqio_mutex);
 381         return ret;
 382 }
 383
 384 /*
 385  *      Release dquot
 386  */
 387 int dquot_release(struct dquot *dquot)
 388 {
 389         int ret = 0, ret2 = 0;
 390         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 391
 392         mutex_lock(&dquot->dq_lock);
 393         /* Check whether we are not racing with some other dqget() */
 394         if (atomic_read(&dquot->dq_count) > 1)
 395                 goto out_dqlock;
 396         mutex_lock(&dqopt->dqio_mutex);
 397         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
 398                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
 399                 /* Write the info */
 400                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 401                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 402                 if (ret >= 0)
 403                         ret = ret2;
 404         }
 405         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 406         mutex_unlock(&dqopt->dqio_mutex);
 407 out_dqlock:
 408         mutex_unlock(&dquot->dq_lock);
 409         return ret;
 410 }
 411
 412 /* Invalidate all dquots on the list. Note that this function is called after
 413  * quota is disabled and pointers from inodes removed so there cannot be new
 414  * quota users. There can still be some users of quotas due to inodes being
 415  * just deleted or pruned by prune_icache() (those are not attached to any
 416  * list). We have to wait for such users.
 417  */
 418 static void invalidate_dquots(struct super_block *sb, int type)
 419 {
 420         struct dquot *dquot, *tmp;
 421
 422 restart:
 423         spin_lock(&dq_list_lock);
 424         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 425                 if (dquot->dq_sb != sb)
 426                         continue;
 427                 if (dquot->dq_type != type)
 428                         continue;
 429                 /* Wait for dquot users */
 430                 if (atomic_read(&dquot->dq_count)) {
 431                         DEFINE_WAIT(wait);
 432
 433                         atomic_inc(&dquot->dq_count);
 434                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
 435                                         TASK_UNINTERRUPTIBLE);
 436                         spin_unlock(&dq_list_lock);
 437                         /* Once dqput() wakes us up, we know it's time to free
 438                          * the dquot.
 439                          * IMPORTANT: we rely on the fact that there is always
 440                          * at most one process waiting for dquot to free.
 441                          * Otherwise dq_count would be > 1 and we would never
 442                          * wake up.
 443                          */
 444                         if (atomic_read(&dquot->dq_count) > 1)
 445                                 schedule();
 446                         finish_wait(&dquot->dq_wait_unused, &wait);
 447                         dqput(dquot);
 448                         /* At this moment dquot() need not exist (it could be
 449                          * reclaimed by prune_dqcache(). Hence we must
 450                          * restart. */
 451                         goto restart;
 452                 }
 453                 /*
 454                  * Quota now has no users and it has been written on last
 455                  * dqput()
 456                  */
 457                 remove_dquot_hash(dquot);
 458                 remove_free_dquot(dquot);
 459                 remove_inuse(dquot);
 460                 kmem_cache_free(dquot_cachep, dquot);
 461         }
 462         spin_unlock(&dq_list_lock);
 463 }
 464
 465 int vfs_quota_sync(struct super_block *sb, int type)
 466 {
 467         struct list_head *dirty;
 468         struct dquot *dquot;
 469         struct quota_info *dqopt = sb_dqopt(sb);
 470         int cnt;
 471
 472         mutex_lock(&dqopt->dqonoff_mutex);
 473         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 474                 if (type != -1 && cnt != type)
 475                         continue;
 476                 if (!sb_has_quota_enabled(sb, cnt))
 477                         continue;
 478                 spin_lock(&dq_list_lock);
 479                 dirty = &dqopt->info[cnt].dqi_dirty_list;
 480                 while (!list_empty(dirty)) {
 481                         dquot = list_first_entry(dirty, struct dquot, dq_dirty);
 482                         /* Dirty and inactive can be only bad dquot... */
 483                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 484                                 clear_dquot_dirty(dquot);
 485                                 continue;
 486                         }
 487                         /* Now we have active dquot from which someone is
 488                          * holding reference so we can safely just increase
 489                          * use count */
 490                         atomic_inc(&dquot->dq_count);
 491                         dqstats.lookups++;
 492                         spin_unlock(&dq_list_lock);
 493                         sb->dq_op->write_dquot(dquot);
 494                         dqput(dquot);
 495                         spin_lock(&dq_list_lock);
 496                 }
 497                 spin_unlock(&dq_list_lock);
 498         }
 499
 500         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 501                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
 502                         && info_dirty(&dqopt->info[cnt]))
 503                         sb->dq_op->write_info(sb, cnt);
 504         spin_lock(&dq_list_lock);
 505         dqstats.syncs++;
 506         spin_unlock(&dq_list_lock);
 507         mutex_unlock(&dqopt->dqonoff_mutex);
 508
 509         return 0;
 510 }
 511
 512 /* Free unused dquots from cache */
 513 static void prune_dqcache(int count)
 514 {
 515         struct list_head *head;
 516         struct dquot *dquot;
 517
 518         head = free_dquots.prev;
 519         while (head != &free_dquots && count) {
 520                 dquot = list_entry(head, struct dquot, dq_free);
 521                 remove_dquot_hash(dquot);
 522                 remove_free_dquot(dquot);
 523                 remove_inuse(dquot);
 524                 kmem_cache_free(dquot_cachep, dquot);
 525                 count--;
 526                 head = free_dquots.prev;
 527         }
 528 }
 529
 530 /*
 531  * This is called from kswapd when we think we need some
 532  * more memory
 533  */
 534
 535 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
 536 {
 537         if (nr) {
 538                 spin_lock(&dq_list_lock);
 539                 prune_dqcache(nr);
 540                 spin_unlock(&dq_list_lock);
 541         }
 542         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
 543 }
 544
 545 static struct shrinker dqcache_shrinker = {
 546         .shrink = shrink_dqcache_memory,
 547         .seeks = DEFAULT_SEEKS,
 548 };
 549
 550 /*
 551  * Put reference to dquot
 552  * NOTE: If you change this function please check whether dqput_blocks() works right...
 553  * MUST be called with either dqptr_sem or dqonoff_mutex held
 554  */
 555 static void dqput(struct dquot *dquot)
 556 {
 557         if (!dquot)
 558                 return;
 559 #ifdef __DQUOT_PARANOIA
 560         if (!atomic_read(&dquot->dq_count)) {
 561                 printk("VFS: dqput: trying to free free dquot\n");
 562                 printk("VFS: device %s, dquot of %s %d\n",
 563                         dquot->dq_sb->s_id,
 564                         quotatypes[dquot->dq_type],
 565                         dquot->dq_id);
 566                 BUG();
 567         }
 568 #endif
 569
 570         spin_lock(&dq_list_lock);
 571         dqstats.drops++;
 572         spin_unlock(&dq_list_lock);
 573 we_slept:
 574         spin_lock(&dq_list_lock);
 575         if (atomic_read(&dquot->dq_count) > 1) {
 576                 /* We have more than one user... nothing to do */
 577                 atomic_dec(&dquot->dq_count);
 578                 /* Releasing dquot during quotaoff phase? */
 579                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
 580                     atomic_read(&dquot->dq_count) == 1)
 581                         wake_up(&dquot->dq_wait_unused);
 582                 spin_unlock(&dq_list_lock);
 583                 return;
 584         }
 585         /* Need to release dquot? */
 586         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 587                 spin_unlock(&dq_list_lock);
 588                 /* Commit dquot before releasing */
 589                 dquot->dq_sb->dq_op->write_dquot(dquot);
 590                 goto we_slept;
 591         }
 592         /* Clear flag in case dquot was inactive (something bad happened) */
 593         clear_dquot_dirty(dquot);
 594         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 595                 spin_unlock(&dq_list_lock);
 596                 dquot->dq_sb->dq_op->release_dquot(dquot);
 597                 goto we_slept;
 598         }
 599         atomic_dec(&dquot->dq_count);
 600 #ifdef __DQUOT_PARANOIA
 601         /* sanity check */
 602         BUG_ON(!list_empty(&dquot->dq_free));
 603 #endif
 604         put_dquot_last(dquot);
 605         spin_unlock(&dq_list_lock);
 606 }
 607
 608 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 609 {
 610         struct dquot *dquot;
 611
 612         dquot = kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
 613         if(!dquot)
 614                 return NODQUOT;
 615
 616         mutex_init(&dquot->dq_lock);
 617         INIT_LIST_HEAD(&dquot->dq_free);
 618         INIT_LIST_HEAD(&dquot->dq_inuse);
 619         INIT_HLIST_NODE(&dquot->dq_hash);
 620         INIT_LIST_HEAD(&dquot->dq_dirty);
 621         init_waitqueue_head(&dquot->dq_wait_unused);
 622         dquot->dq_sb = sb;
 623         dquot->dq_type = type;
 624         atomic_set(&dquot->dq_count, 1);
 625
 626         return dquot;
 627 }
 628
 629 /*
 630  * Get reference to dquot
 631  * MUST be called with either dqptr_sem or dqonoff_mutex held
 632  */
 633 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 634 {
 635         unsigned int hashent = hashfn(sb, id, type);
 636         struct dquot *dquot, *empty = NODQUOT;
 637
 638         if (!sb_has_quota_enabled(sb, type))
 639                 return NODQUOT;
 640 we_slept:
 641         spin_lock(&dq_list_lock);
 642         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 643                 if (empty == NODQUOT) {
 644                         spin_unlock(&dq_list_lock);
 645                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 646                                 schedule();     /* Try to wait for a moment... */
 647                         goto we_slept;
 648                 }
 649                 dquot = empty;
 650                 dquot->dq_id = id;
 651                 /* all dquots go on the inuse_list */
 652                 put_inuse(dquot);
 653                 /* hash it first so it can be found */
 654                 insert_dquot_hash(dquot);
 655                 dqstats.lookups++;
 656                 spin_unlock(&dq_list_lock);
 657         } else {
 658                 if (!atomic_read(&dquot->dq_count))
 659                         remove_free_dquot(dquot);
 660                 atomic_inc(&dquot->dq_count);
 661                 dqstats.cache_hits++;
 662                 dqstats.lookups++;
 663                 spin_unlock(&dq_list_lock);
 664                 if (empty)
 665                         kmem_cache_free(dquot_cachep, empty);
 666         }
 667         /* Wait for dq_lock - after this we know that either dquot_release() is already
 668          * finished or it will be canceled due to dq_count > 1 test */
 669         wait_on_dquot(dquot);
 670         /* Read the dquot and instantiate it (everything done only if needed) */
 671         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 672                 dqput(dquot);
 673                 return NODQUOT;
 674         }
 675 #ifdef __DQUOT_PARANOIA
 676         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 677 #endif
 678
 679         return dquot;
 680 }
 681
 682 static int dqinit_needed(struct inode *inode, int type)
 683 {
 684         int cnt;
 685
 686         if (IS_NOQUOTA(inode))
 687                 return 0;
 688         if (type != -1)
 689                 return inode->i_dquot[type] == NODQUOT;
 690         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 691                 if (inode->i_dquot[cnt] == NODQUOT)
 692                         return 1;
 693         return 0;
 694 }
 695
 696 /* This routine is guarded by dqonoff_mutex mutex */
 697 static void add_dquot_ref(struct super_block *sb, int type)
 698 {
 699         struct inode *inode, *old_inode = NULL;
 700
 701         spin_lock(&inode_lock);
 702         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 703                 if (!atomic_read(&inode->i_writecount))
 704                         continue;
 705                 if (!dqinit_needed(inode, type))
 706                         continue;
 707                 if (inode->i_state & (I_FREEING|I_WILL_FREE))
 708                         continue;
 709
 710                 __iget(inode);
 711                 spin_unlock(&inode_lock);
 712
 713                 iput(old_inode);
 714                 sb->dq_op->initialize(inode, type);
 715                 /* We hold a reference to 'inode' so it couldn't have been
 716                  * removed from s_inodes list while we dropped the inode_lock.
 717                  * We cannot iput the inode now as we can be holding the last
 718                  * reference and we cannot iput it under inode_lock. So we
 719                  * keep the reference and iput it later. */
 720                 old_inode = inode;
 721                 spin_lock(&inode_lock);
 722         }
 723         spin_unlock(&inode_lock);
 724         iput(old_inode);
 725 }
 726
 727 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 728 static inline int dqput_blocks(struct dquot *dquot)
 729 {
 730         if (atomic_read(&dquot->dq_count) <= 1)
 731                 return 1;
 732         return 0;
 733 }
 734
 735 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 736 /* We can't race with anybody because we hold dqptr_sem for writing... */
 737 static int remove_inode_dquot_ref(struct inode *inode, int type,
 738                                   struct list_head *tofree_head)
 739 {
 740         struct dquot *dquot = inode->i_dquot[type];
 741
 742         inode->i_dquot[type] = NODQUOT;
 743         if (dquot != NODQUOT) {
 744                 if (dqput_blocks(dquot)) {
 745 #ifdef __DQUOT_PARANOIA
 746                         if (atomic_read(&dquot->dq_count) != 1)
 747                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 748 #endif
 749                         spin_lock(&dq_list_lock);
 750                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 751                         spin_unlock(&dq_list_lock);
 752                         return 1;
 753                 }
 754                 else
 755                         dqput(dquot);   /* We have guaranteed we won't block */
 756         }
 757         return 0;
 758 }
 759
 760 /* Free list of dquots - called from inode.c */
 761 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 762 static void put_dquot_list(struct list_head *tofree_head)
 763 {
 764         struct list_head *act_head;
 765         struct dquot *dquot;
 766
 767         act_head = tofree_head->next;
 768         /* So now we have dquots on the list... Just free them */
 769         while (act_head != tofree_head) {
 770                 dquot = list_entry(act_head, struct dquot, dq_free);
 771                 act_head = act_head->next;
 772                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 773                 dqput(dquot);
 774         }
 775 }
 776
 777 static void remove_dquot_ref(struct super_block *sb, int type,
 778                 struct list_head *tofree_head)
 779 {
 780         struct inode *inode;
 781
 782         spin_lock(&inode_lock);
 783         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 784                 if (!IS_NOQUOTA(inode))
 785                         remove_inode_dquot_ref(inode, type, tofree_head);
 786         }
 787         spin_unlock(&inode_lock);
 788 }
 789
 790 /* Gather all references from inodes and drop them */
 791 static void drop_dquot_ref(struct super_block *sb, int type)
 792 {
 793         LIST_HEAD(tofree_head);
 794
 795         if (sb->dq_op) {
 796                 down_write(&sb_dqopt(sb)->dqptr_sem);
 797                 remove_dquot_ref(sb, type, &tofree_head);
 798                 up_write(&sb_dqopt(sb)->dqptr_sem);
 799                 put_dquot_list(&tofree_head);
 800         }
 801 }
 802
 803 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
 804 {
 805         dquot->dq_dqb.dqb_curinodes += number;
 806 }
 807
 808 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 809 {
 810         dquot->dq_dqb.dqb_curspace += number;
 811 }
 812
 813 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
 814 {
 815         if (dquot->dq_dqb.dqb_curinodes > number)
 816                 dquot->dq_dqb.dqb_curinodes -= number;
 817         else
 818                 dquot->dq_dqb.dqb_curinodes = 0;
 819         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
 820                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 821         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 822 }
 823
 824 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 825 {
 826         if (dquot->dq_dqb.dqb_curspace > number)
 827                 dquot->dq_dqb.dqb_curspace -= number;
 828         else
 829                 dquot->dq_dqb.dqb_curspace = 0;
 830         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
 831                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 832         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 833 }
 834
 835 static int warning_issued(struct dquot *dquot, const int warntype)
 836 {
 837         int flag = (warntype == QUOTA_NL_BHARDWARN ||
 838                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
 839                 ((warntype == QUOTA_NL_IHARDWARN ||
 840                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
 841
 842         if (!flag)
 843                 return 0;
 844         return test_and_set_bit(flag, &dquot->dq_flags);
 845 }
 846
 847 #ifdef CONFIG_PRINT_QUOTA_WARNING
 848 static int flag_print_warnings = 1;
 849
 850 static inline int need_print_warning(struct dquot *dquot)
 851 {
 852         if (!flag_print_warnings)
 853                 return 0;
 854
 855         switch (dquot->dq_type) {
 856                 case USRQUOTA:
 857                         return current->fsuid == dquot->dq_id;
 858                 case GRPQUOTA:
 859                         return in_group_p(dquot->dq_id);
 860         }
 861         return 0;
 862 }
 863
 864 /* Print warning to user which exceeded quota */
 865 static void print_warning(struct dquot *dquot, const int warntype)
 866 {
 867         char *msg = NULL;
 868         struct tty_struct *tty;
 869
 870         if (!need_print_warning(dquot))
 871                 return;
 872
 873         mutex_lock(&tty_mutex);
 874         tty = get_current_tty();
 875         if (!tty)
 876                 goto out_lock;
 877         tty_write_message(tty, dquot->dq_sb->s_id);
 878         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
 879                 tty_write_message(tty, ": warning, ");
 880         else
 881                 tty_write_message(tty, ": write failed, ");
 882         tty_write_message(tty, quotatypes[dquot->dq_type]);
 883         switch (warntype) {
 884                 case QUOTA_NL_IHARDWARN:
 885                         msg = " file limit reached.\r\n";
 886                         break;
 887                 case QUOTA_NL_ISOFTLONGWARN:
 888                         msg = " file quota exceeded too long.\r\n";
 889                         break;
 890                 case QUOTA_NL_ISOFTWARN:
 891                         msg = " file quota exceeded.\r\n";
 892                         break;
 893                 case QUOTA_NL_BHARDWARN:
 894                         msg = " block limit reached.\r\n";
 895                         break;
 896                 case QUOTA_NL_BSOFTLONGWARN:
 897                         msg = " block quota exceeded too long.\r\n";
 898                         break;
 899                 case QUOTA_NL_BSOFTWARN:
 900                         msg = " block quota exceeded.\r\n";
 901                         break;
 902         }
 903         tty_write_message(tty, msg);
 904 out_lock:
 905         mutex_unlock(&tty_mutex);
 906 }
 907 #endif
 908
 909 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
 910
 911 /* Netlink family structure for quota */
 912 static struct genl_family quota_genl_family = {
 913         .id = GENL_ID_GENERATE,
 914         .hdrsize = 0,
 915         .name = "VFS_DQUOT",
 916         .version = 1,
 917         .maxattr = QUOTA_NL_A_MAX,
 918 };
 919
 920 /* Send warning to userspace about user which exceeded quota */
 921 static void send_warning(const struct dquot *dquot, const char warntype)
 922 {
 923         static atomic_t seq;
 924         struct sk_buff *skb;
 925         void *msg_head;
 926         int ret;
 927         int msg_size = 4 * nla_total_size(sizeof(u32)) +
 928                        2 * nla_total_size(sizeof(u64));
 929
 930         /* We have to allocate using GFP_NOFS as we are called from a
 931          * filesystem performing write and thus further recursion into
 932          * the fs to free some data could cause deadlocks. */
 933         skb = genlmsg_new(msg_size, GFP_NOFS);
 934         if (!skb) {
 935                 printk(KERN_ERR
 936                   "VFS: Not enough memory to send quota warning.\n");
 937                 return;
 938         }
 939         msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
 940                         &quota_genl_family, 0, QUOTA_NL_C_WARNING);
 941         if (!msg_head) {
 942                 printk(KERN_ERR
 943                   "VFS: Cannot store netlink header in quota warning.\n");
 944                 goto err_out;
 945         }
 946         ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
 947         if (ret)
 948                 goto attr_err_out;
 949         ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
 950         if (ret)
 951                 goto attr_err_out;
 952         ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
 953         if (ret)
 954                 goto attr_err_out;
 955         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
 956                 MAJOR(dquot->dq_sb->s_dev));
 957         if (ret)
 958                 goto attr_err_out;
 959         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
 960                 MINOR(dquot->dq_sb->s_dev));
 961         if (ret)
 962                 goto attr_err_out;
 963         ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
 964         if (ret)
 965                 goto attr_err_out;
 966         genlmsg_end(skb, msg_head);
 967
 968         ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
 969         if (ret < 0 && ret != -ESRCH)
 970                 printk(KERN_ERR
 971                         "VFS: Failed to send notification message: %d\n", ret);
 972         return;
 973 attr_err_out:
 974         printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
 975 err_out:
 976         kfree_skb(skb);
 977 }
 978 #endif
 979
 980 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
 981 {
 982         int i;
 983
 984         for (i = 0; i < MAXQUOTAS; i++)
 985                 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
 986                     !warning_issued(dquots[i], warntype[i])) {
 987 #ifdef CONFIG_PRINT_QUOTA_WARNING
 988                         print_warning(dquots[i], warntype[i]);
 989 #endif
 990 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
 991                         send_warning(dquots[i], warntype[i]);
 992 #endif
 993                 }
 994 }
 995
 996 static inline char ignore_hardlimit(struct dquot *dquot)
 997 {
 998         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
 999
1000         return capable(CAP_SYS_RESOURCE) &&
1001             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1002 }
1003
1004 /* needs dq_data_lock */
1005 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
1006 {
1007         *warntype = QUOTA_NL_NOWARN;
1008         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1009                 return QUOTA_OK;
1010
1011         if (dquot->dq_dqb.dqb_ihardlimit &&
1012            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1013             !ignore_hardlimit(dquot)) {
1014                 *warntype = QUOTA_NL_IHARDWARN;
1015                 return NO_QUOTA;
1016         }
1017
1018         if (dquot->dq_dqb.dqb_isoftlimit &&
1019            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1020             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1021             !ignore_hardlimit(dquot)) {
1022                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1023                 return NO_QUOTA;
1024         }
1025
1026         if (dquot->dq_dqb.dqb_isoftlimit &&
1027            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1028             dquot->dq_dqb.dqb_itime == 0) {
1029                 *warntype = QUOTA_NL_ISOFTWARN;
1030                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1031         }
1032
1033         return QUOTA_OK;
1034 }
1035
1036 /* needs dq_data_lock */
1037 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1038 {
1039         *warntype = QUOTA_NL_NOWARN;
1040         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1041                 return QUOTA_OK;
1042
1043         if (dquot->dq_dqb.dqb_bhardlimit &&
1044            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1045             !ignore_hardlimit(dquot)) {
1046                 if (!prealloc)
1047                         *warntype = QUOTA_NL_BHARDWARN;
1048                 return NO_QUOTA;
1049         }
1050
1051         if (dquot->dq_dqb.dqb_bsoftlimit &&
1052            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1053             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1054             !ignore_hardlimit(dquot)) {
1055                 if (!prealloc)
1056                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1057                 return NO_QUOTA;
1058         }
1059
1060         if (dquot->dq_dqb.dqb_bsoftlimit &&
1061            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1062             dquot->dq_dqb.dqb_btime == 0) {
1063                 if (!prealloc) {
1064                         *warntype = QUOTA_NL_BSOFTWARN;
1065                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1066                 }
1067                 else
1068                         /*
1069                          * We don't allow preallocation to exceed softlimit so exceeding will
1070                          * be always printed
1071                          */
1072                         return NO_QUOTA;
1073         }
1074
1075         return QUOTA_OK;
1076 }
1077
1078 /*
1079  *      Initialize quota pointers in inode
1080  *      Transaction must be started at entry
1081  */
1082 int dquot_initialize(struct inode *inode, int type)
1083 {
1084         unsigned int id = 0;
1085         int cnt, ret = 0;
1086
1087         /* First test before acquiring mutex - solves deadlocks when we
1088          * re-enter the quota code and are already holding the mutex */
1089         if (IS_NOQUOTA(inode))
1090                 return 0;
1091         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1092         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1093         if (IS_NOQUOTA(inode))
1094                 goto out_err;
1095         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1096                 if (type != -1 && cnt != type)
1097                         continue;
1098                 if (inode->i_dquot[cnt] == NODQUOT) {
1099                         switch (cnt) {
1100                                 case USRQUOTA:
1101                                         id = inode->i_uid;
1102                                         break;
1103                                 case GRPQUOTA:
1104                                         id = inode->i_gid;
1105                                         break;
1106                         }
1107                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1108                 }
1109         }
1110 out_err:
1111         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1112         return ret;
1113 }
1114
1115 /*
1116  *      Release all quotas referenced by inode
1117  *      Transaction must be started at an entry
1118  */
1119 int dquot_drop(struct inode *inode)
1120 {
1121         int cnt;
1122
1123         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1124         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1125                 if (inode->i_dquot[cnt] != NODQUOT) {
1126                         dqput(inode->i_dquot[cnt]);
1127                         inode->i_dquot[cnt] = NODQUOT;
1128                 }
1129         }
1130         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1131         return 0;
1132 }
1133
1134 /*
1135  * Following four functions update i_blocks+i_bytes fields and
1136  * quota information (together with appropriate checks)
1137  * NOTE: We absolutely rely on the fact that caller dirties
1138  * the inode (usually macros in quotaops.h care about this) and
1139  * holds a handle for the current transaction so that dquot write and
1140  * inode write go into the same transaction.
1141  */
1142
1143 /*
1144  * This operation can block, but only after everything is updated
1145  */
1146 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1147 {
1148         int cnt, ret = NO_QUOTA;
1149         char warntype[MAXQUOTAS];
1150
1151         /* First test before acquiring mutex - solves deadlocks when we
1152          * re-enter the quota code and are already holding the mutex */
1153         if (IS_NOQUOTA(inode)) {
1154 out_add:
1155                 inode_add_bytes(inode, number);
1156                 return QUOTA_OK;
1157         }
1158         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1159                 warntype[cnt] = QUOTA_NL_NOWARN;
1160
1161         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1162         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1163                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1164                 goto out_add;
1165         }
1166         spin_lock(&dq_data_lock);
1167         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1168                 if (inode->i_dquot[cnt] == NODQUOT)
1169                         continue;
1170                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1171                         goto warn_put_all;
1172         }
1173         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1174                 if (inode->i_dquot[cnt] == NODQUOT)
1175                         continue;
1176                 dquot_incr_space(inode->i_dquot[cnt], number);
1177         }
1178         inode_add_bytes(inode, number);
1179         ret = QUOTA_OK;
1180 warn_put_all:
1181         spin_unlock(&dq_data_lock);
1182         if (ret == QUOTA_OK)
1183                 /* Dirtify all the dquots - this can block when journalling */
1184                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1185                         if (inode->i_dquot[cnt])
1186                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1187         flush_warnings(inode->i_dquot, warntype);
1188         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1189         return ret;
1190 }
1191
1192 /*
1193  * This operation can block, but only after everything is updated
1194  */
1195 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1196 {
1197         int cnt, ret = NO_QUOTA;
1198         char warntype[MAXQUOTAS];
1199
1200         /* First test before acquiring mutex - solves deadlocks when we
1201          * re-enter the quota code and are already holding the mutex */
1202         if (IS_NOQUOTA(inode))
1203                 return QUOTA_OK;
1204         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1205                 warntype[cnt] = QUOTA_NL_NOWARN;
1206         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1207         if (IS_NOQUOTA(inode)) {
1208                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1209                 return QUOTA_OK;
1210         }
1211         spin_lock(&dq_data_lock);
1212         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1213                 if (inode->i_dquot[cnt] == NODQUOT)
1214                         continue;
1215                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1216                         goto warn_put_all;
1217         }
1218
1219         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1220                 if (inode->i_dquot[cnt] == NODQUOT)
1221                         continue;
1222                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1223         }
1224         ret = QUOTA_OK;
1225 warn_put_all:
1226         spin_unlock(&dq_data_lock);
1227         if (ret == QUOTA_OK)
1228                 /* Dirtify all the dquots - this can block when journalling */
1229                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1230                         if (inode->i_dquot[cnt])
1231                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1232         flush_warnings(inode->i_dquot, warntype);
1233         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1234         return ret;
1235 }
1236
1237 /*
1238  * This operation can block, but only after everything is updated
1239  */
1240 int dquot_free_space(struct inode *inode, qsize_t number)
1241 {
1242         unsigned int cnt;
1243
1244         /* First test before acquiring mutex - solves deadlocks when we
1245          * re-enter the quota code and are already holding the mutex */
1246         if (IS_NOQUOTA(inode)) {
1247 out_sub:
1248                 inode_sub_bytes(inode, number);
1249                 return QUOTA_OK;
1250         }
1251         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1252         /* Now recheck reliably when holding dqptr_sem */
1253         if (IS_NOQUOTA(inode)) {
1254                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1255                 goto out_sub;
1256         }
1257         spin_lock(&dq_data_lock);
1258         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1259                 if (inode->i_dquot[cnt] == NODQUOT)
1260                         continue;
1261                 dquot_decr_space(inode->i_dquot[cnt], number);
1262         }
1263         inode_sub_bytes(inode, number);
1264         spin_unlock(&dq_data_lock);
1265         /* Dirtify all the dquots - this can block when journalling */
1266         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1267                 if (inode->i_dquot[cnt])
1268                         mark_dquot_dirty(inode->i_dquot[cnt]);
1269         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1270         return QUOTA_OK;
1271 }
1272
1273 /*
1274  * This operation can block, but only after everything is updated
1275  */
1276 int dquot_free_inode(const struct inode *inode, unsigned long number)
1277 {
1278         unsigned int cnt;
1279
1280         /* First test before acquiring mutex - solves deadlocks when we
1281          * re-enter the quota code and are already holding the mutex */
1282         if (IS_NOQUOTA(inode))
1283                 return QUOTA_OK;
1284         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1285         /* Now recheck reliably when holding dqptr_sem */
1286         if (IS_NOQUOTA(inode)) {
1287                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1288                 return QUOTA_OK;
1289         }
1290         spin_lock(&dq_data_lock);
1291         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1292                 if (inode->i_dquot[cnt] == NODQUOT)
1293                         continue;
1294                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1295         }
1296         spin_unlock(&dq_data_lock);
1297         /* Dirtify all the dquots - this can block when journalling */
1298         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1299                 if (inode->i_dquot[cnt])
1300                         mark_dquot_dirty(inode->i_dquot[cnt]);
1301         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1302         return QUOTA_OK;
1303 }
1304
1305 /*
1306  * Transfer the number of inode and blocks from one diskquota to an other.
1307  *
1308  * This operation can block, but only after everything is updated
1309  * A transaction must be started when entering this function.
1310  */
1311 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1312 {
1313         qsize_t space;
1314         struct dquot *transfer_from[MAXQUOTAS];
1315         struct dquot *transfer_to[MAXQUOTAS];
1316         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1317             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1318         char warntype[MAXQUOTAS];
1319
1320         /* First test before acquiring mutex - solves deadlocks when we
1321          * re-enter the quota code and are already holding the mutex */
1322         if (IS_NOQUOTA(inode))
1323                 return QUOTA_OK;
1324         /* Clear the arrays */
1325         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1326                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1327                 warntype[cnt] = QUOTA_NL_NOWARN;
1328         }
1329         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1330         /* Now recheck reliably when holding dqptr_sem */
1331         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1332                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1333                 return QUOTA_OK;
1334         }
1335         /* First build the transfer_to list - here we can block on
1336          * reading/instantiating of dquots.  We know that the transaction for
1337          * us was already started so we don't violate lock ranking here */
1338         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1339                 switch (cnt) {
1340                         case USRQUOTA:
1341                                 if (!chuid)
1342                                         continue;
1343                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1344                                 break;
1345                         case GRPQUOTA:
1346                                 if (!chgid)
1347                                         continue;
1348                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1349                                 break;
1350                 }
1351         }
1352         spin_lock(&dq_data_lock);
1353         space = inode_get_bytes(inode);
1354         /* Build the transfer_from list and check the limits */
1355         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1356                 if (transfer_to[cnt] == NODQUOT)
1357                         continue;
1358                 transfer_from[cnt] = inode->i_dquot[cnt];
1359                 if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
1360                     check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
1361                         goto warn_put_all;
1362         }
1363
1364         /*
1365          * Finally perform the needed transfer from transfer_from to transfer_to
1366          */
1367         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1368                 /*
1369                  * Skip changes for same uid or gid or for turned off quota-type.
1370                  */
1371                 if (transfer_to[cnt] == NODQUOT)
1372                         continue;
1373
1374                 /* Due to IO error we might not have transfer_from[] structure */
1375                 if (transfer_from[cnt]) {
1376                         dquot_decr_inodes(transfer_from[cnt], 1);
1377                         dquot_decr_space(transfer_from[cnt], space);
1378                 }
1379
1380                 dquot_incr_inodes(transfer_to[cnt], 1);
1381                 dquot_incr_space(transfer_to[cnt], space);
1382
1383                 inode->i_dquot[cnt] = transfer_to[cnt];
1384         }
1385         ret = QUOTA_OK;
1386 warn_put_all:
1387         spin_unlock(&dq_data_lock);
1388         /* Dirtify all the dquots - this can block when journalling */
1389         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1390                 if (transfer_from[cnt])
1391                         mark_dquot_dirty(transfer_from[cnt]);
1392                 if (transfer_to[cnt])
1393                         mark_dquot_dirty(transfer_to[cnt]);
1394         }
1395         flush_warnings(transfer_to, warntype);
1396
1397         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1398                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1399                         dqput(transfer_from[cnt]);
1400                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1401                         dqput(transfer_to[cnt]);
1402         }
1403         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1404         return ret;
1405 }
1406
1407 /*
1408  * Write info of quota file to disk
1409  */
1410 int dquot_commit_info(struct super_block *sb, int type)
1411 {
1412         int ret;
1413         struct quota_info *dqopt = sb_dqopt(sb);
1414
1415         mutex_lock(&dqopt->dqio_mutex);
1416         ret = dqopt->ops[type]->write_file_info(sb, type);
1417         mutex_unlock(&dqopt->dqio_mutex);
1418         return ret;
1419 }
1420
1421 /*
1422  * Definitions of diskquota operations.
1423  */
1424 struct dquot_operations dquot_operations = {
1425         .initialize     = dquot_initialize,
1426         .drop           = dquot_drop,
1427         .alloc_space    = dquot_alloc_space,
1428         .alloc_inode    = dquot_alloc_inode,
1429         .free_space     = dquot_free_space,
1430         .free_inode     = dquot_free_inode,
1431         .transfer       = dquot_transfer,
1432         .write_dquot    = dquot_commit,
1433         .acquire_dquot  = dquot_acquire,
1434         .release_dquot  = dquot_release,
1435         .mark_dirty     = dquot_mark_dquot_dirty,
1436         .write_info     = dquot_commit_info
1437 };
1438
1439 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1440 {
1441         switch (type) {
1442                 case USRQUOTA:
1443                         dqopt->flags |= DQUOT_USR_ENABLED;
1444                         break;
1445                 case GRPQUOTA:
1446                         dqopt->flags |= DQUOT_GRP_ENABLED;
1447                         break;
1448         }
1449 }
1450
1451 static inline void reset_enable_flags(struct quota_info *dqopt, int type)
1452 {
1453         switch (type) {
1454                 case USRQUOTA:
1455                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1456                         break;
1457                 case GRPQUOTA:
1458                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1459                         break;
1460         }
1461 }
1462
1463 /*
1464  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1465  */
1466 int vfs_quota_off(struct super_block *sb, int type)
1467 {
1468         int cnt;
1469         struct quota_info *dqopt = sb_dqopt(sb);
1470         struct inode *toputinode[MAXQUOTAS];
1471
1472         /* We need to serialize quota_off() for device */
1473         mutex_lock(&dqopt->dqonoff_mutex);
1474         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1475                 toputinode[cnt] = NULL;
1476                 if (type != -1 && cnt != type)
1477                         continue;
1478                 if (!sb_has_quota_enabled(sb, cnt))
1479                         continue;
1480                 reset_enable_flags(dqopt, cnt);
1481
1482                 /* Note: these are blocking operations */
1483                 drop_dquot_ref(sb, cnt);
1484                 invalidate_dquots(sb, cnt);
1485                 /*
1486                  * Now all dquots should be invalidated, all writes done so we should be only
1487                  * users of the info. No locks needed.
1488                  */
1489                 if (info_dirty(&dqopt->info[cnt]))
1490                         sb->dq_op->write_info(sb, cnt);
1491                 if (dqopt->ops[cnt]->free_file_info)
1492                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1493                 put_quota_format(dqopt->info[cnt].dqi_format);
1494
1495                 toputinode[cnt] = dqopt->files[cnt];
1496                 dqopt->files[cnt] = NULL;
1497                 dqopt->info[cnt].dqi_flags = 0;
1498                 dqopt->info[cnt].dqi_igrace = 0;
1499                 dqopt->info[cnt].dqi_bgrace = 0;
1500                 dqopt->ops[cnt] = NULL;
1501         }
1502         mutex_unlock(&dqopt->dqonoff_mutex);
1503         /* Sync the superblock so that buffers with quota data are written to
1504          * disk (and so userspace sees correct data afterwards). */
1505         if (sb->s_op->sync_fs)
1506                 sb->s_op->sync_fs(sb, 1);
1507         sync_blockdev(sb->s_bdev);
1508         /* Now the quota files are just ordinary files and we can set the
1509          * inode flags back. Moreover we discard the pagecache so that
1510          * userspace sees the writes we did bypassing the pagecache. We
1511          * must also discard the blockdev buffers so that we see the
1512          * changes done by userspace on the next quotaon() */
1513         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1514                 if (toputinode[cnt]) {
1515                         mutex_lock(&dqopt->dqonoff_mutex);
1516                         /* If quota was reenabled in the meantime, we have
1517                          * nothing to do */
1518                         if (!sb_has_quota_enabled(sb, cnt)) {
1519                                 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1520                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1521                                   S_NOATIME | S_NOQUOTA);
1522                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1523                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1524                                 mark_inode_dirty(toputinode[cnt]);
1525                                 iput(toputinode[cnt]);
1526                         }
1527                         mutex_unlock(&dqopt->dqonoff_mutex);
1528                 }
1529         if (sb->s_bdev)
1530                 invalidate_bdev(sb->s_bdev);
1531         return 0;
1532 }
1533
1534 /*
1535  *      Turn quotas on on a device
1536  */
1537
1538 /* Helper function when we already have the inode */
1539 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1540 {
1541         struct quota_format_type *fmt = find_quota_format(format_id);
1542         struct super_block *sb = inode->i_sb;
1543         struct quota_info *dqopt = sb_dqopt(sb);
1544         int error;
1545         int oldflags = -1;
1546
1547         if (!fmt)
1548                 return -ESRCH;
1549         if (!S_ISREG(inode->i_mode)) {
1550                 error = -EACCES;
1551                 goto out_fmt;
1552         }
1553         if (IS_RDONLY(inode)) {
1554                 error = -EROFS;
1555                 goto out_fmt;
1556         }
1557         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1558                 error = -EINVAL;
1559                 goto out_fmt;
1560         }
1561
1562         /* As we bypass the pagecache we must now flush the inode so that
1563          * we see all the changes from userspace... */
1564         write_inode_now(inode, 1);
1565         /* And now flush the block cache so that kernel sees the changes */
1566         invalidate_bdev(sb->s_bdev);
1567         mutex_lock(&inode->i_mutex);
1568         mutex_lock(&dqopt->dqonoff_mutex);
1569         if (sb_has_quota_enabled(sb, type)) {
1570                 error = -EBUSY;
1571                 goto out_lock;
1572         }
1573         /* We don't want quota and atime on quota files (deadlocks possible)
1574          * Also nobody should write to the file - we use special IO operations
1575          * which ignore the immutable bit. */
1576         down_write(&dqopt->dqptr_sem);
1577         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1578         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1579         up_write(&dqopt->dqptr_sem);
1580         sb->dq_op->drop(inode);
1581
1582         error = -EIO;
1583         dqopt->files[type] = igrab(inode);
1584         if (!dqopt->files[type])
1585                 goto out_lock;
1586         error = -EINVAL;
1587         if (!fmt->qf_ops->check_quota_file(sb, type))
1588                 goto out_file_init;
1589
1590         dqopt->ops[type] = fmt->qf_ops;
1591         dqopt->info[type].dqi_format = fmt;
1592         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1593         mutex_lock(&dqopt->dqio_mutex);
1594         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1595                 mutex_unlock(&dqopt->dqio_mutex);
1596                 goto out_file_init;
1597         }
1598         mutex_unlock(&dqopt->dqio_mutex);
1599         mutex_unlock(&inode->i_mutex);
1600         set_enable_flags(dqopt, type);
1601
1602         add_dquot_ref(sb, type);
1603         mutex_unlock(&dqopt->dqonoff_mutex);
1604
1605         return 0;
1606
1607 out_file_init:
1608         dqopt->files[type] = NULL;
1609         iput(inode);
1610 out_lock:
1611         mutex_unlock(&dqopt->dqonoff_mutex);
1612         if (oldflags != -1) {
1613                 down_write(&dqopt->dqptr_sem);
1614                 /* Set the flags back (in the case of accidental quotaon()
1615                  * on a wrong file we don't want to mess up the flags) */
1616                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1617                 inode->i_flags |= oldflags;
1618                 up_write(&dqopt->dqptr_sem);
1619         }
1620         mutex_unlock(&inode->i_mutex);
1621 out_fmt:
1622         put_quota_format(fmt);
1623
1624         return error;
1625 }
1626
1627 /* Actual function called from quotactl() */
1628 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
1629 {
1630         struct nameidata nd;
1631         int error;
1632
1633         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1634         if (error < 0)
1635                 return error;
1636         error = security_quota_on(nd.dentry);
1637         if (error)
1638                 goto out_path;
1639         /* Quota file not on the same filesystem? */
1640         if (nd.mnt->mnt_sb != sb)
1641                 error = -EXDEV;
1642         else
1643                 error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
1644 out_path:
1645         path_release(&nd);
1646         return error;
1647 }
1648
1649 /*
1650  * This function is used when filesystem needs to initialize quotas
1651  * during mount time.
1652  */
1653 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1654                 int format_id, int type)
1655 {
1656         struct dentry *dentry;
1657         int error;
1658
1659         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1660         if (IS_ERR(dentry))
1661                 return PTR_ERR(dentry);
1662
1663         if (!dentry->d_inode) {
1664                 error = -ENOENT;
1665                 goto out;
1666         }
1667
1668         error = security_quota_on(dentry);
1669         if (!error)
1670                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1671
1672 out:
1673         dput(dentry);
1674         return error;
1675 }
1676
1677 /* Generic routine for getting common part of quota structure */
1678 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1679 {
1680         struct mem_dqblk *dm = &dquot->dq_dqb;
1681
1682         spin_lock(&dq_data_lock);
1683         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1684         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1685         di->dqb_curspace = dm->dqb_curspace;
1686         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1687         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1688         di->dqb_curinodes = dm->dqb_curinodes;
1689         di->dqb_btime = dm->dqb_btime;
1690         di->dqb_itime = dm->dqb_itime;
1691         di->dqb_valid = QIF_ALL;
1692         spin_unlock(&dq_data_lock);
1693 }
1694
1695 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1696 {
1697         struct dquot *dquot;
1698
1699         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1700         if (!(dquot = dqget(sb, id, type))) {
1701                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1702                 return -ESRCH;
1703         }
1704         do_get_dqblk(dquot, di);
1705         dqput(dquot);
1706         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1707         return 0;
1708 }
1709
1710 /* Generic routine for setting common part of quota structure */
1711 static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1712 {
1713         struct mem_dqblk *dm = &dquot->dq_dqb;
1714         int check_blim = 0, check_ilim = 0;
1715
1716         spin_lock(&dq_data_lock);
1717         if (di->dqb_valid & QIF_SPACE) {
1718                 dm->dqb_curspace = di->dqb_curspace;
1719                 check_blim = 1;
1720         }
1721         if (di->dqb_valid & QIF_BLIMITS) {
1722                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1723                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1724                 check_blim = 1;
1725         }
1726         if (di->dqb_valid & QIF_INODES) {
1727                 dm->dqb_curinodes = di->dqb_curinodes;
1728                 check_ilim = 1;
1729         }
1730         if (di->dqb_valid & QIF_ILIMITS) {
1731                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1732                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1733                 check_ilim = 1;
1734         }
1735         if (di->dqb_valid & QIF_BTIME)
1736                 dm->dqb_btime = di->dqb_btime;
1737         if (di->dqb_valid & QIF_ITIME)
1738                 dm->dqb_itime = di->dqb_itime;
1739
1740         if (check_blim) {
1741                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1742                         dm->dqb_btime = 0;
1743                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1744                 }
1745                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1746                         dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1747         }
1748         if (check_ilim) {
1749                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1750                         dm->dqb_itime = 0;
1751                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1752                 }
1753                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1754                         dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1755         }
1756         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1757                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1758         else
1759                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1760         spin_unlock(&dq_data_lock);
1761         mark_dquot_dirty(dquot);
1762 }
1763
1764 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1765 {
1766         struct dquot *dquot;
1767
1768         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1769         if (!(dquot = dqget(sb, id, type))) {
1770                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1771                 return -ESRCH;
1772         }
1773         do_set_dqblk(dquot, di);
1774         dqput(dquot);
1775         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1776         return 0;
1777 }
1778
1779 /* Generic routine for getting common part of quota file information */
1780 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1781 {
1782         struct mem_dqinfo *mi;
1783
1784         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1785         if (!sb_has_quota_enabled(sb, type)) {
1786                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1787                 return -ESRCH;
1788         }
1789         mi = sb_dqopt(sb)->info + type;
1790         spin_lock(&dq_data_lock);
1791         ii->dqi_bgrace = mi->dqi_bgrace;
1792         ii->dqi_igrace = mi->dqi_igrace;
1793         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1794         ii->dqi_valid = IIF_ALL;
1795         spin_unlock(&dq_data_lock);
1796         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1797         return 0;
1798 }
1799
1800 /* Generic routine for setting common part of quota file information */
1801 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1802 {
1803         struct mem_dqinfo *mi;
1804
1805         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1806         if (!sb_has_quota_enabled(sb, type)) {
1807                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1808                 return -ESRCH;
1809         }
1810         mi = sb_dqopt(sb)->info + type;
1811         spin_lock(&dq_data_lock);
1812         if (ii->dqi_valid & IIF_BGRACE)
1813                 mi->dqi_bgrace = ii->dqi_bgrace;
1814         if (ii->dqi_valid & IIF_IGRACE)
1815                 mi->dqi_igrace = ii->dqi_igrace;
1816         if (ii->dqi_valid & IIF_FLAGS)
1817                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
1818         spin_unlock(&dq_data_lock);
1819         mark_info_dirty(sb, type);
1820         /* Force write to disk */
1821         sb->dq_op->write_info(sb, type);
1822         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1823         return 0;
1824 }
1825
1826 struct quotactl_ops vfs_quotactl_ops = {
1827         .quota_on       = vfs_quota_on,
1828         .quota_off      = vfs_quota_off,
1829         .quota_sync     = vfs_quota_sync,
1830         .get_info       = vfs_get_dqinfo,
1831         .set_info       = vfs_set_dqinfo,
1832         .get_dqblk      = vfs_get_dqblk,
1833         .set_dqblk      = vfs_set_dqblk
1834 };
1835
1836 static ctl_table fs_dqstats_table[] = {
1837         {
1838                 .ctl_name       = FS_DQ_LOOKUPS,
1839                 .procname       = "lookups",
1840                 .data           = &dqstats.lookups,
1841                 .maxlen         = sizeof(int),
1842                 .mode           = 0444,
1843                 .proc_handler   = &proc_dointvec,
1844         },
1845         {
1846                 .ctl_name       = FS_DQ_DROPS,
1847                 .procname       = "drops",
1848                 .data           = &dqstats.drops,
1849                 .maxlen         = sizeof(int),
1850                 .mode           = 0444,
1851                 .proc_handler   = &proc_dointvec,
1852         },
1853         {
1854                 .ctl_name       = FS_DQ_READS,
1855                 .procname       = "reads",
1856                 .data           = &dqstats.reads,
1857                 .maxlen         = sizeof(int),
1858                 .mode           = 0444,
1859                 .proc_handler   = &proc_dointvec,
1860         },
1861         {
1862                 .ctl_name       = FS_DQ_WRITES,
1863                 .procname       = "writes",
1864                 .data           = &dqstats.writes,
1865                 .maxlen         = sizeof(int),
1866                 .mode           = 0444,
1867                 .proc_handler   = &proc_dointvec,
1868         },
1869         {
1870                 .ctl_name       = FS_DQ_CACHE_HITS,
1871                 .procname       = "cache_hits",
1872                 .data           = &dqstats.cache_hits,
1873                 .maxlen         = sizeof(int),
1874                 .mode           = 0444,
1875                 .proc_handler   = &proc_dointvec,
1876         },
1877         {
1878                 .ctl_name       = FS_DQ_ALLOCATED,
1879                 .procname       = "allocated_dquots",
1880                 .data           = &dqstats.allocated_dquots,
1881                 .maxlen         = sizeof(int),
1882                 .mode           = 0444,
1883                 .proc_handler   = &proc_dointvec,
1884         },
1885         {
1886                 .ctl_name       = FS_DQ_FREE,
1887                 .procname       = "free_dquots",
1888                 .data           = &dqstats.free_dquots,
1889                 .maxlen         = sizeof(int),
1890                 .mode           = 0444,
1891                 .proc_handler   = &proc_dointvec,
1892         },
1893         {
1894                 .ctl_name       = FS_DQ_SYNCS,
1895                 .procname       = "syncs",
1896                 .data           = &dqstats.syncs,
1897                 .maxlen         = sizeof(int),
1898                 .mode           = 0444,
1899                 .proc_handler   = &proc_dointvec,
1900         },
1901 #ifdef CONFIG_PRINT_QUOTA_WARNING
1902         {
1903                 .ctl_name       = FS_DQ_WARNINGS,
1904                 .procname       = "warnings",
1905                 .data           = &flag_print_warnings,
1906                 .maxlen         = sizeof(int),
1907                 .mode           = 0644,
1908                 .proc_handler   = &proc_dointvec,
1909         },
1910 #endif
1911         { .ctl_name = 0 },
1912 };
1913
1914 static ctl_table fs_table[] = {
1915         {
1916                 .ctl_name       = FS_DQSTATS,
1917                 .procname       = "quota",
1918                 .mode           = 0555,
1919                 .child          = fs_dqstats_table,
1920         },
1921         { .ctl_name = 0 },
1922 };
1923
1924 static ctl_table sys_table[] = {
1925         {
1926                 .ctl_name       = CTL_FS,
1927                 .procname       = "fs",
1928                 .mode           = 0555,
1929                 .child          = fs_table,
1930         },
1931         { .ctl_name = 0 },
1932 };
1933
1934 static int __init dquot_init(void)
1935 {
1936         int i;
1937         unsigned long nr_hash, order;
1938
1939         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
1940
1941         register_sysctl_table(sys_table);
1942
1943         dquot_cachep = kmem_cache_create("dquot",
1944                         sizeof(struct dquot), sizeof(unsigned long) * 4,
1945                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1946                                 SLAB_MEM_SPREAD|SLAB_PANIC),
1947                         NULL);
1948
1949         order = 0;
1950         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
1951         if (!dquot_hash)
1952                 panic("Cannot create dquot hash table");
1953
1954         /* Find power-of-two hlist_heads which can fit into allocation */
1955         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
1956         dq_hash_bits = 0;
1957         do {
1958                 dq_hash_bits++;
1959         } while (nr_hash >> dq_hash_bits);
1960         dq_hash_bits--;
1961
1962         nr_hash = 1UL << dq_hash_bits;
1963         dq_hash_mask = nr_hash - 1;
1964         for (i = 0; i < nr_hash; i++)
1965                 INIT_HLIST_HEAD(dquot_hash + i);
1966
1967         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
1968                         nr_hash, order, (PAGE_SIZE << order));
1969
1970         register_shrinker(&dqcache_shrinker);
1971
1972 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1973         if (genl_register_family(&quota_genl_family) != 0)
1974                 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
1975 #endif
1976
1977         return 0;
1978 }
1979 module_init(dquot_init);
1980
1981 EXPORT_SYMBOL(register_quota_format);
1982 EXPORT_SYMBOL(unregister_quota_format);
1983 EXPORT_SYMBOL(dqstats);
1984 EXPORT_SYMBOL(dq_data_lock);
1985 EXPORT_SYMBOL(vfs_quota_on);
1986 EXPORT_SYMBOL(vfs_quota_on_mount);
1987 EXPORT_SYMBOL(vfs_quota_off);
1988 EXPORT_SYMBOL(vfs_quota_sync);
1989 EXPORT_SYMBOL(vfs_get_dqinfo);
1990 EXPORT_SYMBOL(vfs_set_dqinfo);
1991 EXPORT_SYMBOL(vfs_get_dqblk);
1992 EXPORT_SYMBOL(vfs_set_dqblk);
1993 EXPORT_SYMBOL(dquot_commit);
1994 EXPORT_SYMBOL(dquot_commit_info);
1995 EXPORT_SYMBOL(dquot_acquire);
1996 EXPORT_SYMBOL(dquot_release);
1997 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
1998 EXPORT_SYMBOL(dquot_initialize);
1999 EXPORT_SYMBOL(dquot_drop);
2000 EXPORT_SYMBOL(dquot_alloc_space);
2001 EXPORT_SYMBOL(dquot_alloc_inode);
2002 EXPORT_SYMBOL(dquot_free_space);
2003 EXPORT_SYMBOL(dquot_free_inode);
2004 EXPORT_SYMBOL(dquot_transfer);