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 static inline int dquot_dirty(struct dquot *dquot)
 293 {
 294         return test_bit(DQ_MOD_B, &dquot->dq_flags);
 295 }
 296
 297 static inline int mark_dquot_dirty(struct dquot *dquot)
 298 {
 299         return dquot->dq_sb->dq_op->mark_dirty(dquot);
 300 }
 301
 302 int dquot_mark_dquot_dirty(struct dquot *dquot)
 303 {
 304         spin_lock(&dq_list_lock);
 305         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
 306                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 307                                 info[dquot->dq_type].dqi_dirty_list);
 308         spin_unlock(&dq_list_lock);
 309         return 0;
 310 }
 311
 312 /* This function needs dq_list_lock */
 313 static inline int clear_dquot_dirty(struct dquot *dquot)
 314 {
 315         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
 316                 return 0;
 317         list_del_init(&dquot->dq_dirty);
 318         return 1;
 319 }
 320
 321 void mark_info_dirty(struct super_block *sb, int type)
 322 {
 323         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
 324 }
 325 EXPORT_SYMBOL(mark_info_dirty);
 326
 327 /*
 328  *      Read dquot from disk and alloc space for it
 329  */
 330
 331 int dquot_acquire(struct dquot *dquot)
 332 {
 333         int ret = 0, ret2 = 0;
 334         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 335
 336         mutex_lock(&dquot->dq_lock);
 337         mutex_lock(&dqopt->dqio_mutex);
 338         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 339                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
 340         if (ret < 0)
 341                 goto out_iolock;
 342         set_bit(DQ_READ_B, &dquot->dq_flags);
 343         /* Instantiate dquot if needed */
 344         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 345                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 346                 /* Write the info if needed */
 347                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 348                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 349                 if (ret < 0)
 350                         goto out_iolock;
 351                 if (ret2 < 0) {
 352                         ret = ret2;
 353                         goto out_iolock;
 354                 }
 355         }
 356         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 357 out_iolock:
 358         mutex_unlock(&dqopt->dqio_mutex);
 359         mutex_unlock(&dquot->dq_lock);
 360         return ret;
 361 }
 362
 363 /*
 364  *      Write dquot to disk
 365  */
 366 int dquot_commit(struct dquot *dquot)
 367 {
 368         int ret = 0, ret2 = 0;
 369         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 370
 371         mutex_lock(&dqopt->dqio_mutex);
 372         spin_lock(&dq_list_lock);
 373         if (!clear_dquot_dirty(dquot)) {
 374                 spin_unlock(&dq_list_lock);
 375                 goto out_sem;
 376         }
 377         spin_unlock(&dq_list_lock);
 378         /* Inactive dquot can be only if there was error during read/init
 379          * => we have better not writing it */
 380         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 381                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 382                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 383                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 384                 if (ret >= 0)
 385                         ret = ret2;
 386         }
 387 out_sem:
 388         mutex_unlock(&dqopt->dqio_mutex);
 389         return ret;
 390 }
 391
 392 /*
 393  *      Release dquot
 394  */
 395 int dquot_release(struct dquot *dquot)
 396 {
 397         int ret = 0, ret2 = 0;
 398         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 399
 400         mutex_lock(&dquot->dq_lock);
 401         /* Check whether we are not racing with some other dqget() */
 402         if (atomic_read(&dquot->dq_count) > 1)
 403                 goto out_dqlock;
 404         mutex_lock(&dqopt->dqio_mutex);
 405         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
 406                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
 407                 /* Write the info */
 408                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 409                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 410                 if (ret >= 0)
 411                         ret = ret2;
 412         }
 413         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 414         mutex_unlock(&dqopt->dqio_mutex);
 415 out_dqlock:
 416         mutex_unlock(&dquot->dq_lock);
 417         return ret;
 418 }
 419
 420 /* Invalidate all dquots on the list. Note that this function is called after
 421  * quota is disabled and pointers from inodes removed so there cannot be new
 422  * quota users. There can still be some users of quotas due to inodes being
 423  * just deleted or pruned by prune_icache() (those are not attached to any
 424  * list). We have to wait for such users.
 425  */
 426 static void invalidate_dquots(struct super_block *sb, int type)
 427 {
 428         struct dquot *dquot, *tmp;
 429
 430 restart:
 431         spin_lock(&dq_list_lock);
 432         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 433                 if (dquot->dq_sb != sb)
 434                         continue;
 435                 if (dquot->dq_type != type)
 436                         continue;
 437                 /* Wait for dquot users */
 438                 if (atomic_read(&dquot->dq_count)) {
 439                         DEFINE_WAIT(wait);
 440
 441                         atomic_inc(&dquot->dq_count);
 442                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
 443                                         TASK_UNINTERRUPTIBLE);
 444                         spin_unlock(&dq_list_lock);
 445                         /* Once dqput() wakes us up, we know it's time to free
 446                          * the dquot.
 447                          * IMPORTANT: we rely on the fact that there is always
 448                          * at most one process waiting for dquot to free.
 449                          * Otherwise dq_count would be > 1 and we would never
 450                          * wake up.
 451                          */
 452                         if (atomic_read(&dquot->dq_count) > 1)
 453                                 schedule();
 454                         finish_wait(&dquot->dq_wait_unused, &wait);
 455                         dqput(dquot);
 456                         /* At this moment dquot() need not exist (it could be
 457                          * reclaimed by prune_dqcache(). Hence we must
 458                          * restart. */
 459                         goto restart;
 460                 }
 461                 /*
 462                  * Quota now has no users and it has been written on last
 463                  * dqput()
 464                  */
 465                 remove_dquot_hash(dquot);
 466                 remove_free_dquot(dquot);
 467                 remove_inuse(dquot);
 468                 kmem_cache_free(dquot_cachep, dquot);
 469         }
 470         spin_unlock(&dq_list_lock);
 471 }
 472
 473 int vfs_quota_sync(struct super_block *sb, int type)
 474 {
 475         struct list_head *dirty;
 476         struct dquot *dquot;
 477         struct quota_info *dqopt = sb_dqopt(sb);
 478         int cnt;
 479
 480         mutex_lock(&dqopt->dqonoff_mutex);
 481         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 482                 if (type != -1 && cnt != type)
 483                         continue;
 484                 if (!sb_has_quota_enabled(sb, cnt))
 485                         continue;
 486                 spin_lock(&dq_list_lock);
 487                 dirty = &dqopt->info[cnt].dqi_dirty_list;
 488                 while (!list_empty(dirty)) {
 489                         dquot = list_first_entry(dirty, struct dquot, dq_dirty);
 490                         /* Dirty and inactive can be only bad dquot... */
 491                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 492                                 clear_dquot_dirty(dquot);
 493                                 continue;
 494                         }
 495                         /* Now we have active dquot from which someone is
 496                          * holding reference so we can safely just increase
 497                          * use count */
 498                         atomic_inc(&dquot->dq_count);
 499                         dqstats.lookups++;
 500                         spin_unlock(&dq_list_lock);
 501                         sb->dq_op->write_dquot(dquot);
 502                         dqput(dquot);
 503                         spin_lock(&dq_list_lock);
 504                 }
 505                 spin_unlock(&dq_list_lock);
 506         }
 507
 508         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 509                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
 510                         && info_dirty(&dqopt->info[cnt]))
 511                         sb->dq_op->write_info(sb, cnt);
 512         spin_lock(&dq_list_lock);
 513         dqstats.syncs++;
 514         spin_unlock(&dq_list_lock);
 515         mutex_unlock(&dqopt->dqonoff_mutex);
 516
 517         return 0;
 518 }
 519
 520 /* Free unused dquots from cache */
 521 static void prune_dqcache(int count)
 522 {
 523         struct list_head *head;
 524         struct dquot *dquot;
 525
 526         head = free_dquots.prev;
 527         while (head != &free_dquots && count) {
 528                 dquot = list_entry(head, struct dquot, dq_free);
 529                 remove_dquot_hash(dquot);
 530                 remove_free_dquot(dquot);
 531                 remove_inuse(dquot);
 532                 kmem_cache_free(dquot_cachep, dquot);
 533                 count--;
 534                 head = free_dquots.prev;
 535         }
 536 }
 537
 538 /*
 539  * This is called from kswapd when we think we need some
 540  * more memory
 541  */
 542
 543 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
 544 {
 545         if (nr) {
 546                 spin_lock(&dq_list_lock);
 547                 prune_dqcache(nr);
 548                 spin_unlock(&dq_list_lock);
 549         }
 550         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
 551 }
 552
 553 static struct shrinker dqcache_shrinker = {
 554         .shrink = shrink_dqcache_memory,
 555         .seeks = DEFAULT_SEEKS,
 556 };
 557
 558 /*
 559  * Put reference to dquot
 560  * NOTE: If you change this function please check whether dqput_blocks() works right...
 561  * MUST be called with either dqptr_sem or dqonoff_mutex held
 562  */
 563 static void dqput(struct dquot *dquot)
 564 {
 565         int ret;
 566
 567         if (!dquot)
 568                 return;
 569 #ifdef __DQUOT_PARANOIA
 570         if (!atomic_read(&dquot->dq_count)) {
 571                 printk("VFS: dqput: trying to free free dquot\n");
 572                 printk("VFS: device %s, dquot of %s %d\n",
 573                         dquot->dq_sb->s_id,
 574                         quotatypes[dquot->dq_type],
 575                         dquot->dq_id);
 576                 BUG();
 577         }
 578 #endif
 579
 580         spin_lock(&dq_list_lock);
 581         dqstats.drops++;
 582         spin_unlock(&dq_list_lock);
 583 we_slept:
 584         spin_lock(&dq_list_lock);
 585         if (atomic_read(&dquot->dq_count) > 1) {
 586                 /* We have more than one user... nothing to do */
 587                 atomic_dec(&dquot->dq_count);
 588                 /* Releasing dquot during quotaoff phase? */
 589                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
 590                     atomic_read(&dquot->dq_count) == 1)
 591                         wake_up(&dquot->dq_wait_unused);
 592                 spin_unlock(&dq_list_lock);
 593                 return;
 594         }
 595         /* Need to release dquot? */
 596         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 597                 spin_unlock(&dq_list_lock);
 598                 /* Commit dquot before releasing */
 599                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
 600                 if (ret < 0) {
 601                         printk(KERN_ERR "VFS: cannot write quota structure on "
 602                                 "device %s (error %d). Quota may get out of "
 603                                 "sync!\n", dquot->dq_sb->s_id, ret);
 604                         /*
 605                          * We clear dirty bit anyway, so that we avoid
 606                          * infinite loop here
 607                          */
 608                         spin_lock(&dq_list_lock);
 609                         clear_dquot_dirty(dquot);
 610                         spin_unlock(&dq_list_lock);
 611                 }
 612                 goto we_slept;
 613         }
 614         /* Clear flag in case dquot was inactive (something bad happened) */
 615         clear_dquot_dirty(dquot);
 616         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 617                 spin_unlock(&dq_list_lock);
 618                 dquot->dq_sb->dq_op->release_dquot(dquot);
 619                 goto we_slept;
 620         }
 621         atomic_dec(&dquot->dq_count);
 622 #ifdef __DQUOT_PARANOIA
 623         /* sanity check */
 624         BUG_ON(!list_empty(&dquot->dq_free));
 625 #endif
 626         put_dquot_last(dquot);
 627         spin_unlock(&dq_list_lock);
 628 }
 629
 630 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 631 {
 632         struct dquot *dquot;
 633
 634         dquot = kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
 635         if(!dquot)
 636                 return NODQUOT;
 637
 638         mutex_init(&dquot->dq_lock);
 639         INIT_LIST_HEAD(&dquot->dq_free);
 640         INIT_LIST_HEAD(&dquot->dq_inuse);
 641         INIT_HLIST_NODE(&dquot->dq_hash);
 642         INIT_LIST_HEAD(&dquot->dq_dirty);
 643         init_waitqueue_head(&dquot->dq_wait_unused);
 644         dquot->dq_sb = sb;
 645         dquot->dq_type = type;
 646         atomic_set(&dquot->dq_count, 1);
 647
 648         return dquot;
 649 }
 650
 651 /*
 652  * Get reference to dquot
 653  * MUST be called with either dqptr_sem or dqonoff_mutex held
 654  */
 655 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 656 {
 657         unsigned int hashent = hashfn(sb, id, type);
 658         struct dquot *dquot, *empty = NODQUOT;
 659
 660         if (!sb_has_quota_enabled(sb, type))
 661                 return NODQUOT;
 662 we_slept:
 663         spin_lock(&dq_list_lock);
 664         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 665                 if (empty == NODQUOT) {
 666                         spin_unlock(&dq_list_lock);
 667                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 668                                 schedule();     /* Try to wait for a moment... */
 669                         goto we_slept;
 670                 }
 671                 dquot = empty;
 672                 dquot->dq_id = id;
 673                 /* all dquots go on the inuse_list */
 674                 put_inuse(dquot);
 675                 /* hash it first so it can be found */
 676                 insert_dquot_hash(dquot);
 677                 dqstats.lookups++;
 678                 spin_unlock(&dq_list_lock);
 679         } else {
 680                 if (!atomic_read(&dquot->dq_count))
 681                         remove_free_dquot(dquot);
 682                 atomic_inc(&dquot->dq_count);
 683                 dqstats.cache_hits++;
 684                 dqstats.lookups++;
 685                 spin_unlock(&dq_list_lock);
 686                 if (empty)
 687                         kmem_cache_free(dquot_cachep, empty);
 688         }
 689         /* Wait for dq_lock - after this we know that either dquot_release() is already
 690          * finished or it will be canceled due to dq_count > 1 test */
 691         wait_on_dquot(dquot);
 692         /* Read the dquot and instantiate it (everything done only if needed) */
 693         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 694                 dqput(dquot);
 695                 return NODQUOT;
 696         }
 697 #ifdef __DQUOT_PARANOIA
 698         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 699 #endif
 700
 701         return dquot;
 702 }
 703
 704 static int dqinit_needed(struct inode *inode, int type)
 705 {
 706         int cnt;
 707
 708         if (IS_NOQUOTA(inode))
 709                 return 0;
 710         if (type != -1)
 711                 return inode->i_dquot[type] == NODQUOT;
 712         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 713                 if (inode->i_dquot[cnt] == NODQUOT)
 714                         return 1;
 715         return 0;
 716 }
 717
 718 /* This routine is guarded by dqonoff_mutex mutex */
 719 static void add_dquot_ref(struct super_block *sb, int type)
 720 {
 721         struct inode *inode, *old_inode = NULL;
 722
 723         spin_lock(&inode_lock);
 724         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 725                 if (!atomic_read(&inode->i_writecount))
 726                         continue;
 727                 if (!dqinit_needed(inode, type))
 728                         continue;
 729                 if (inode->i_state & (I_FREEING|I_WILL_FREE))
 730                         continue;
 731
 732                 __iget(inode);
 733                 spin_unlock(&inode_lock);
 734
 735                 iput(old_inode);
 736                 sb->dq_op->initialize(inode, type);
 737                 /* We hold a reference to 'inode' so it couldn't have been
 738                  * removed from s_inodes list while we dropped the inode_lock.
 739                  * We cannot iput the inode now as we can be holding the last
 740                  * reference and we cannot iput it under inode_lock. So we
 741                  * keep the reference and iput it later. */
 742                 old_inode = inode;
 743                 spin_lock(&inode_lock);
 744         }
 745         spin_unlock(&inode_lock);
 746         iput(old_inode);
 747 }
 748
 749 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 750 static inline int dqput_blocks(struct dquot *dquot)
 751 {
 752         if (atomic_read(&dquot->dq_count) <= 1)
 753                 return 1;
 754         return 0;
 755 }
 756
 757 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 758 /* We can't race with anybody because we hold dqptr_sem for writing... */
 759 static int remove_inode_dquot_ref(struct inode *inode, int type,
 760                                   struct list_head *tofree_head)
 761 {
 762         struct dquot *dquot = inode->i_dquot[type];
 763
 764         inode->i_dquot[type] = NODQUOT;
 765         if (dquot != NODQUOT) {
 766                 if (dqput_blocks(dquot)) {
 767 #ifdef __DQUOT_PARANOIA
 768                         if (atomic_read(&dquot->dq_count) != 1)
 769                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 770 #endif
 771                         spin_lock(&dq_list_lock);
 772                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 773                         spin_unlock(&dq_list_lock);
 774                         return 1;
 775                 }
 776                 else
 777                         dqput(dquot);   /* We have guaranteed we won't block */
 778         }
 779         return 0;
 780 }
 781
 782 /* Free list of dquots - called from inode.c */
 783 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 784 static void put_dquot_list(struct list_head *tofree_head)
 785 {
 786         struct list_head *act_head;
 787         struct dquot *dquot;
 788
 789         act_head = tofree_head->next;
 790         /* So now we have dquots on the list... Just free them */
 791         while (act_head != tofree_head) {
 792                 dquot = list_entry(act_head, struct dquot, dq_free);
 793                 act_head = act_head->next;
 794                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 795                 dqput(dquot);
 796         }
 797 }
 798
 799 static void remove_dquot_ref(struct super_block *sb, int type,
 800                 struct list_head *tofree_head)
 801 {
 802         struct inode *inode;
 803
 804         spin_lock(&inode_lock);
 805         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 806                 if (!IS_NOQUOTA(inode))
 807                         remove_inode_dquot_ref(inode, type, tofree_head);
 808         }
 809         spin_unlock(&inode_lock);
 810 }
 811
 812 /* Gather all references from inodes and drop them */
 813 static void drop_dquot_ref(struct super_block *sb, int type)
 814 {
 815         LIST_HEAD(tofree_head);
 816
 817         if (sb->dq_op) {
 818                 down_write(&sb_dqopt(sb)->dqptr_sem);
 819                 remove_dquot_ref(sb, type, &tofree_head);
 820                 up_write(&sb_dqopt(sb)->dqptr_sem);
 821                 put_dquot_list(&tofree_head);
 822         }
 823 }
 824
 825 static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
 826 {
 827         dquot->dq_dqb.dqb_curinodes += number;
 828 }
 829
 830 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 831 {
 832         dquot->dq_dqb.dqb_curspace += number;
 833 }
 834
 835 static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
 836 {
 837         if (dquot->dq_dqb.dqb_curinodes > number)
 838                 dquot->dq_dqb.dqb_curinodes -= number;
 839         else
 840                 dquot->dq_dqb.dqb_curinodes = 0;
 841         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
 842                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 843         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 844 }
 845
 846 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 847 {
 848         if (dquot->dq_dqb.dqb_curspace > number)
 849                 dquot->dq_dqb.dqb_curspace -= number;
 850         else
 851                 dquot->dq_dqb.dqb_curspace = 0;
 852         if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
 853                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 854         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 855 }
 856
 857 static int warning_issued(struct dquot *dquot, const int warntype)
 858 {
 859         int flag = (warntype == QUOTA_NL_BHARDWARN ||
 860                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
 861                 ((warntype == QUOTA_NL_IHARDWARN ||
 862                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
 863
 864         if (!flag)
 865                 return 0;
 866         return test_and_set_bit(flag, &dquot->dq_flags);
 867 }
 868
 869 #ifdef CONFIG_PRINT_QUOTA_WARNING
 870 static int flag_print_warnings = 1;
 871
 872 static inline int need_print_warning(struct dquot *dquot)
 873 {
 874         if (!flag_print_warnings)
 875                 return 0;
 876
 877         switch (dquot->dq_type) {
 878                 case USRQUOTA:
 879                         return current->fsuid == dquot->dq_id;
 880                 case GRPQUOTA:
 881                         return in_group_p(dquot->dq_id);
 882         }
 883         return 0;
 884 }
 885
 886 /* Print warning to user which exceeded quota */
 887 static void print_warning(struct dquot *dquot, const int warntype)
 888 {
 889         char *msg = NULL;
 890         struct tty_struct *tty;
 891
 892         if (warntype == QUOTA_NL_IHARDBELOW ||
 893             warntype == QUOTA_NL_ISOFTBELOW ||
 894             warntype == QUOTA_NL_BHARDBELOW ||
 895             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
 896                 return;
 897
 898         mutex_lock(&tty_mutex);
 899         tty = get_current_tty();
 900         if (!tty)
 901                 goto out_lock;
 902         tty_write_message(tty, dquot->dq_sb->s_id);
 903         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
 904                 tty_write_message(tty, ": warning, ");
 905         else
 906                 tty_write_message(tty, ": write failed, ");
 907         tty_write_message(tty, quotatypes[dquot->dq_type]);
 908         switch (warntype) {
 909                 case QUOTA_NL_IHARDWARN:
 910                         msg = " file limit reached.\r\n";
 911                         break;
 912                 case QUOTA_NL_ISOFTLONGWARN:
 913                         msg = " file quota exceeded too long.\r\n";
 914                         break;
 915                 case QUOTA_NL_ISOFTWARN:
 916                         msg = " file quota exceeded.\r\n";
 917                         break;
 918                 case QUOTA_NL_BHARDWARN:
 919                         msg = " block limit reached.\r\n";
 920                         break;
 921                 case QUOTA_NL_BSOFTLONGWARN:
 922                         msg = " block quota exceeded too long.\r\n";
 923                         break;
 924                 case QUOTA_NL_BSOFTWARN:
 925                         msg = " block quota exceeded.\r\n";
 926                         break;
 927         }
 928         tty_write_message(tty, msg);
 929 out_lock:
 930         mutex_unlock(&tty_mutex);
 931 }
 932 #endif
 933
 934 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
 935
 936 /* Netlink family structure for quota */
 937 static struct genl_family quota_genl_family = {
 938         .id = GENL_ID_GENERATE,
 939         .hdrsize = 0,
 940         .name = "VFS_DQUOT",
 941         .version = 1,
 942         .maxattr = QUOTA_NL_A_MAX,
 943 };
 944
 945 /* Send warning to userspace about user which exceeded quota */
 946 static void send_warning(const struct dquot *dquot, const char warntype)
 947 {
 948         static atomic_t seq;
 949         struct sk_buff *skb;
 950         void *msg_head;
 951         int ret;
 952         int msg_size = 4 * nla_total_size(sizeof(u32)) +
 953                        2 * nla_total_size(sizeof(u64));
 954
 955         /* We have to allocate using GFP_NOFS as we are called from a
 956          * filesystem performing write and thus further recursion into
 957          * the fs to free some data could cause deadlocks. */
 958         skb = genlmsg_new(msg_size, GFP_NOFS);
 959         if (!skb) {
 960                 printk(KERN_ERR
 961                   "VFS: Not enough memory to send quota warning.\n");
 962                 return;
 963         }
 964         msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
 965                         &quota_genl_family, 0, QUOTA_NL_C_WARNING);
 966         if (!msg_head) {
 967                 printk(KERN_ERR
 968                   "VFS: Cannot store netlink header in quota warning.\n");
 969                 goto err_out;
 970         }
 971         ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
 972         if (ret)
 973                 goto attr_err_out;
 974         ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
 975         if (ret)
 976                 goto attr_err_out;
 977         ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
 978         if (ret)
 979                 goto attr_err_out;
 980         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
 981                 MAJOR(dquot->dq_sb->s_dev));
 982         if (ret)
 983                 goto attr_err_out;
 984         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
 985                 MINOR(dquot->dq_sb->s_dev));
 986         if (ret)
 987                 goto attr_err_out;
 988         ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
 989         if (ret)
 990                 goto attr_err_out;
 991         genlmsg_end(skb, msg_head);
 992
 993         ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
 994         if (ret < 0 && ret != -ESRCH)
 995                 printk(KERN_ERR
 996                         "VFS: Failed to send notification message: %d\n", ret);
 997         return;
 998 attr_err_out:
 999         printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
1000 err_out:
1001         kfree_skb(skb);
1002 }
1003 #endif
1004
1005 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
1006 {
1007         int i;
1008
1009         for (i = 0; i < MAXQUOTAS; i++)
1010                 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
1011                     !warning_issued(dquots[i], warntype[i])) {
1012 #ifdef CONFIG_PRINT_QUOTA_WARNING
1013                         print_warning(dquots[i], warntype[i]);
1014 #endif
1015 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1016                         send_warning(dquots[i], warntype[i]);
1017 #endif
1018                 }
1019 }
1020
1021 static inline char ignore_hardlimit(struct dquot *dquot)
1022 {
1023         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1024
1025         return capable(CAP_SYS_RESOURCE) &&
1026             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1027 }
1028
1029 /* needs dq_data_lock */
1030 static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
1031 {
1032         *warntype = QUOTA_NL_NOWARN;
1033         if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1034                 return QUOTA_OK;
1035
1036         if (dquot->dq_dqb.dqb_ihardlimit &&
1037            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1038             !ignore_hardlimit(dquot)) {
1039                 *warntype = QUOTA_NL_IHARDWARN;
1040                 return NO_QUOTA;
1041         }
1042
1043         if (dquot->dq_dqb.dqb_isoftlimit &&
1044            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1045             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1046             !ignore_hardlimit(dquot)) {
1047                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1048                 return NO_QUOTA;
1049         }
1050
1051         if (dquot->dq_dqb.dqb_isoftlimit &&
1052            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1053             dquot->dq_dqb.dqb_itime == 0) {
1054                 *warntype = QUOTA_NL_ISOFTWARN;
1055                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1056         }
1057
1058         return QUOTA_OK;
1059 }
1060
1061 /* needs dq_data_lock */
1062 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1063 {
1064         *warntype = QUOTA_NL_NOWARN;
1065         if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
1066                 return QUOTA_OK;
1067
1068         if (dquot->dq_dqb.dqb_bhardlimit &&
1069            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
1070             !ignore_hardlimit(dquot)) {
1071                 if (!prealloc)
1072                         *warntype = QUOTA_NL_BHARDWARN;
1073                 return NO_QUOTA;
1074         }
1075
1076         if (dquot->dq_dqb.dqb_bsoftlimit &&
1077            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1078             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1079             !ignore_hardlimit(dquot)) {
1080                 if (!prealloc)
1081                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1082                 return NO_QUOTA;
1083         }
1084
1085         if (dquot->dq_dqb.dqb_bsoftlimit &&
1086            toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
1087             dquot->dq_dqb.dqb_btime == 0) {
1088                 if (!prealloc) {
1089                         *warntype = QUOTA_NL_BSOFTWARN;
1090                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1091                 }
1092                 else
1093                         /*
1094                          * We don't allow preallocation to exceed softlimit so exceeding will
1095                          * be always printed
1096                          */
1097                         return NO_QUOTA;
1098         }
1099
1100         return QUOTA_OK;
1101 }
1102
1103 static int info_idq_free(struct dquot *dquot, ulong inodes)
1104 {
1105         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1106             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1107                 return QUOTA_NL_NOWARN;
1108
1109         if (dquot->dq_dqb.dqb_curinodes - inodes <= dquot->dq_dqb.dqb_isoftlimit)
1110                 return QUOTA_NL_ISOFTBELOW;
1111         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1112             dquot->dq_dqb.dqb_curinodes - inodes < dquot->dq_dqb.dqb_ihardlimit)
1113                 return QUOTA_NL_IHARDBELOW;
1114         return QUOTA_NL_NOWARN;
1115 }
1116
1117 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1118 {
1119         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1120             toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
1121                 return QUOTA_NL_NOWARN;
1122
1123         if (toqb(dquot->dq_dqb.dqb_curspace - space) <=
1124             dquot->dq_dqb.dqb_bsoftlimit)
1125                 return QUOTA_NL_BSOFTBELOW;
1126         if (toqb(dquot->dq_dqb.dqb_curspace) >= dquot->dq_dqb.dqb_bhardlimit &&
1127             toqb(dquot->dq_dqb.dqb_curspace - space) <
1128                                                 dquot->dq_dqb.dqb_bhardlimit)
1129                 return QUOTA_NL_BHARDBELOW;
1130         return QUOTA_NL_NOWARN;
1131 }
1132 /*
1133  *      Initialize quota pointers in inode
1134  *      Transaction must be started at entry
1135  */
1136 int dquot_initialize(struct inode *inode, int type)
1137 {
1138         unsigned int id = 0;
1139         int cnt, ret = 0;
1140
1141         /* First test before acquiring mutex - solves deadlocks when we
1142          * re-enter the quota code and are already holding the mutex */
1143         if (IS_NOQUOTA(inode))
1144                 return 0;
1145         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1146         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1147         if (IS_NOQUOTA(inode))
1148                 goto out_err;
1149         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1150                 if (type != -1 && cnt != type)
1151                         continue;
1152                 if (inode->i_dquot[cnt] == NODQUOT) {
1153                         switch (cnt) {
1154                                 case USRQUOTA:
1155                                         id = inode->i_uid;
1156                                         break;
1157                                 case GRPQUOTA:
1158                                         id = inode->i_gid;
1159                                         break;
1160                         }
1161                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1162                 }
1163         }
1164 out_err:
1165         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1166         return ret;
1167 }
1168
1169 /*
1170  *      Release all quotas referenced by inode
1171  *      Transaction must be started at an entry
1172  */
1173 int dquot_drop(struct inode *inode)
1174 {
1175         int cnt;
1176
1177         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1178         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1179                 if (inode->i_dquot[cnt] != NODQUOT) {
1180                         dqput(inode->i_dquot[cnt]);
1181                         inode->i_dquot[cnt] = NODQUOT;
1182                 }
1183         }
1184         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1185         return 0;
1186 }
1187
1188 /* Wrapper to remove references to quota structures from inode */
1189 void vfs_dq_drop(struct inode *inode)
1190 {
1191         /* Here we can get arbitrary inode from clear_inode() so we have
1192          * to be careful. OTOH we don't need locking as quota operations
1193          * are allowed to change only at mount time */
1194         if (!IS_NOQUOTA(inode) && inode->i_sb && inode->i_sb->dq_op
1195             && inode->i_sb->dq_op->drop) {
1196                 int cnt;
1197                 /* Test before calling to rule out calls from proc and such
1198                  * where we are not allowed to block. Note that this is
1199                  * actually reliable test even without the lock - the caller
1200                  * must assure that nobody can come after the DQUOT_DROP and
1201                  * add quota pointers back anyway */
1202                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1203                         if (inode->i_dquot[cnt] != NODQUOT)
1204                                 break;
1205                 if (cnt < MAXQUOTAS)
1206                         inode->i_sb->dq_op->drop(inode);
1207         }
1208 }
1209
1210 /*
1211  * Following four functions update i_blocks+i_bytes fields and
1212  * quota information (together with appropriate checks)
1213  * NOTE: We absolutely rely on the fact that caller dirties
1214  * the inode (usually macros in quotaops.h care about this) and
1215  * holds a handle for the current transaction so that dquot write and
1216  * inode write go into the same transaction.
1217  */
1218
1219 /*
1220  * This operation can block, but only after everything is updated
1221  */
1222 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1223 {
1224         int cnt, ret = NO_QUOTA;
1225         char warntype[MAXQUOTAS];
1226
1227         /* First test before acquiring mutex - solves deadlocks when we
1228          * re-enter the quota code and are already holding the mutex */
1229         if (IS_NOQUOTA(inode)) {
1230 out_add:
1231                 inode_add_bytes(inode, number);
1232                 return QUOTA_OK;
1233         }
1234         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1235                 warntype[cnt] = QUOTA_NL_NOWARN;
1236
1237         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1238         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1239                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1240                 goto out_add;
1241         }
1242         spin_lock(&dq_data_lock);
1243         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1244                 if (inode->i_dquot[cnt] == NODQUOT)
1245                         continue;
1246                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1247                         goto warn_put_all;
1248         }
1249         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1250                 if (inode->i_dquot[cnt] == NODQUOT)
1251                         continue;
1252                 dquot_incr_space(inode->i_dquot[cnt], number);
1253         }
1254         inode_add_bytes(inode, number);
1255         ret = QUOTA_OK;
1256 warn_put_all:
1257         spin_unlock(&dq_data_lock);
1258         if (ret == QUOTA_OK)
1259                 /* Dirtify all the dquots - this can block when journalling */
1260                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1261                         if (inode->i_dquot[cnt])
1262                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1263         flush_warnings(inode->i_dquot, warntype);
1264         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1265         return ret;
1266 }
1267
1268 /*
1269  * This operation can block, but only after everything is updated
1270  */
1271 int dquot_alloc_inode(const struct inode *inode, unsigned long number)
1272 {
1273         int cnt, ret = NO_QUOTA;
1274         char warntype[MAXQUOTAS];
1275
1276         /* First test before acquiring mutex - solves deadlocks when we
1277          * re-enter the quota code and are already holding the mutex */
1278         if (IS_NOQUOTA(inode))
1279                 return QUOTA_OK;
1280         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1281                 warntype[cnt] = QUOTA_NL_NOWARN;
1282         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1283         if (IS_NOQUOTA(inode)) {
1284                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1285                 return QUOTA_OK;
1286         }
1287         spin_lock(&dq_data_lock);
1288         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1289                 if (inode->i_dquot[cnt] == NODQUOT)
1290                         continue;
1291                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1292                         goto warn_put_all;
1293         }
1294
1295         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1296                 if (inode->i_dquot[cnt] == NODQUOT)
1297                         continue;
1298                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1299         }
1300         ret = QUOTA_OK;
1301 warn_put_all:
1302         spin_unlock(&dq_data_lock);
1303         if (ret == QUOTA_OK)
1304                 /* Dirtify all the dquots - this can block when journalling */
1305                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1306                         if (inode->i_dquot[cnt])
1307                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1308         flush_warnings(inode->i_dquot, warntype);
1309         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1310         return ret;
1311 }
1312
1313 /*
1314  * This operation can block, but only after everything is updated
1315  */
1316 int dquot_free_space(struct inode *inode, qsize_t number)
1317 {
1318         unsigned int cnt;
1319         char warntype[MAXQUOTAS];
1320
1321         /* First test before acquiring mutex - solves deadlocks when we
1322          * re-enter the quota code and are already holding the mutex */
1323         if (IS_NOQUOTA(inode)) {
1324 out_sub:
1325                 inode_sub_bytes(inode, number);
1326                 return QUOTA_OK;
1327         }
1328
1329         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1330         /* Now recheck reliably when holding dqptr_sem */
1331         if (IS_NOQUOTA(inode)) {
1332                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1333                 goto out_sub;
1334         }
1335         spin_lock(&dq_data_lock);
1336         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1337                 if (inode->i_dquot[cnt] == NODQUOT)
1338                         continue;
1339                 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1340                 dquot_decr_space(inode->i_dquot[cnt], number);
1341         }
1342         inode_sub_bytes(inode, number);
1343         spin_unlock(&dq_data_lock);
1344         /* Dirtify all the dquots - this can block when journalling */
1345         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1346                 if (inode->i_dquot[cnt])
1347                         mark_dquot_dirty(inode->i_dquot[cnt]);
1348         flush_warnings(inode->i_dquot, warntype);
1349         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1350         return QUOTA_OK;
1351 }
1352
1353 /*
1354  * This operation can block, but only after everything is updated
1355  */
1356 int dquot_free_inode(const struct inode *inode, unsigned long number)
1357 {
1358         unsigned int cnt;
1359         char warntype[MAXQUOTAS];
1360
1361         /* First test before acquiring mutex - solves deadlocks when we
1362          * re-enter the quota code and are already holding the mutex */
1363         if (IS_NOQUOTA(inode))
1364                 return QUOTA_OK;
1365
1366         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1367         /* Now recheck reliably when holding dqptr_sem */
1368         if (IS_NOQUOTA(inode)) {
1369                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1370                 return QUOTA_OK;
1371         }
1372         spin_lock(&dq_data_lock);
1373         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1374                 if (inode->i_dquot[cnt] == NODQUOT)
1375                         continue;
1376                 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], number);
1377                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1378         }
1379         spin_unlock(&dq_data_lock);
1380         /* Dirtify all the dquots - this can block when journalling */
1381         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1382                 if (inode->i_dquot[cnt])
1383                         mark_dquot_dirty(inode->i_dquot[cnt]);
1384         flush_warnings(inode->i_dquot, warntype);
1385         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1386         return QUOTA_OK;
1387 }
1388
1389 /*
1390  * Transfer the number of inode and blocks from one diskquota to an other.
1391  *
1392  * This operation can block, but only after everything is updated
1393  * A transaction must be started when entering this function.
1394  */
1395 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1396 {
1397         qsize_t space;
1398         struct dquot *transfer_from[MAXQUOTAS];
1399         struct dquot *transfer_to[MAXQUOTAS];
1400         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1401             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1402         char warntype_to[MAXQUOTAS];
1403         char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1404
1405         /* First test before acquiring mutex - solves deadlocks when we
1406          * re-enter the quota code and are already holding the mutex */
1407         if (IS_NOQUOTA(inode))
1408                 return QUOTA_OK;
1409         /* Clear the arrays */
1410         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1411                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1412                 warntype_to[cnt] = QUOTA_NL_NOWARN;
1413         }
1414         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1415         /* Now recheck reliably when holding dqptr_sem */
1416         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1417                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1418                 return QUOTA_OK;
1419         }
1420         /* First build the transfer_to list - here we can block on
1421          * reading/instantiating of dquots.  We know that the transaction for
1422          * us was already started so we don't violate lock ranking here */
1423         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1424                 switch (cnt) {
1425                         case USRQUOTA:
1426                                 if (!chuid)
1427                                         continue;
1428                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1429                                 break;
1430                         case GRPQUOTA:
1431                                 if (!chgid)
1432                                         continue;
1433                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1434                                 break;
1435                 }
1436         }
1437         spin_lock(&dq_data_lock);
1438         space = inode_get_bytes(inode);
1439         /* Build the transfer_from list and check the limits */
1440         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1441                 if (transfer_to[cnt] == NODQUOT)
1442                         continue;
1443                 transfer_from[cnt] = inode->i_dquot[cnt];
1444                 if (check_idq(transfer_to[cnt], 1, warntype_to + cnt) ==
1445                     NO_QUOTA || check_bdq(transfer_to[cnt], space, 0,
1446                     warntype_to + cnt) == NO_QUOTA)
1447                         goto warn_put_all;
1448         }
1449
1450         /*
1451          * Finally perform the needed transfer from transfer_from to transfer_to
1452          */
1453         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1454                 /*
1455                  * Skip changes for same uid or gid or for turned off quota-type.
1456                  */
1457                 if (transfer_to[cnt] == NODQUOT)
1458                         continue;
1459
1460                 /* Due to IO error we might not have transfer_from[] structure */
1461                 if (transfer_from[cnt]) {
1462                         warntype_from_inodes[cnt] =
1463                                 info_idq_free(transfer_from[cnt], 1);
1464                         warntype_from_space[cnt] =
1465                                 info_bdq_free(transfer_from[cnt], space);
1466                         dquot_decr_inodes(transfer_from[cnt], 1);
1467                         dquot_decr_space(transfer_from[cnt], space);
1468                 }
1469
1470                 dquot_incr_inodes(transfer_to[cnt], 1);
1471                 dquot_incr_space(transfer_to[cnt], space);
1472
1473                 inode->i_dquot[cnt] = transfer_to[cnt];
1474         }
1475         ret = QUOTA_OK;
1476 warn_put_all:
1477         spin_unlock(&dq_data_lock);
1478         /* Dirtify all the dquots - this can block when journalling */
1479         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1480                 if (transfer_from[cnt])
1481                         mark_dquot_dirty(transfer_from[cnt]);
1482                 if (transfer_to[cnt])
1483                         mark_dquot_dirty(transfer_to[cnt]);
1484         }
1485         flush_warnings(transfer_to, warntype_to);
1486         flush_warnings(transfer_from, warntype_from_inodes);
1487         flush_warnings(transfer_from, warntype_from_space);
1488
1489         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1490                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1491                         dqput(transfer_from[cnt]);
1492                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1493                         dqput(transfer_to[cnt]);
1494         }
1495         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1496         return ret;
1497 }
1498
1499 /* Wrapper for transferring ownership of an inode */
1500 int vfs_dq_transfer(struct inode *inode, struct iattr *iattr)
1501 {
1502         if (sb_any_quota_enabled(inode->i_sb) && !IS_NOQUOTA(inode)) {
1503                 vfs_dq_init(inode);
1504                 if (inode->i_sb->dq_op->transfer(inode, iattr) == NO_QUOTA)
1505                         return 1;
1506         }
1507         return 0;
1508 }
1509
1510
1511 /*
1512  * Write info of quota file to disk
1513  */
1514 int dquot_commit_info(struct super_block *sb, int type)
1515 {
1516         int ret;
1517         struct quota_info *dqopt = sb_dqopt(sb);
1518
1519         mutex_lock(&dqopt->dqio_mutex);
1520         ret = dqopt->ops[type]->write_file_info(sb, type);
1521         mutex_unlock(&dqopt->dqio_mutex);
1522         return ret;
1523 }
1524
1525 /*
1526  * Definitions of diskquota operations.
1527  */
1528 struct dquot_operations dquot_operations = {
1529         .initialize     = dquot_initialize,
1530         .drop           = dquot_drop,
1531         .alloc_space    = dquot_alloc_space,
1532         .alloc_inode    = dquot_alloc_inode,
1533         .free_space     = dquot_free_space,
1534         .free_inode     = dquot_free_inode,
1535         .transfer       = dquot_transfer,
1536         .write_dquot    = dquot_commit,
1537         .acquire_dquot  = dquot_acquire,
1538         .release_dquot  = dquot_release,
1539         .mark_dirty     = dquot_mark_dquot_dirty,
1540         .write_info     = dquot_commit_info
1541 };
1542
1543 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1544 {
1545         switch (type) {
1546                 case USRQUOTA:
1547                         dqopt->flags |= DQUOT_USR_ENABLED;
1548                         dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1549                         break;
1550                 case GRPQUOTA:
1551                         dqopt->flags |= DQUOT_GRP_ENABLED;
1552                         dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1553                         break;
1554         }
1555 }
1556
1557 static inline void reset_enable_flags(struct quota_info *dqopt, int type,
1558                                       int remount)
1559 {
1560         switch (type) {
1561                 case USRQUOTA:
1562                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1563                         if (remount)
1564                                 dqopt->flags |= DQUOT_USR_SUSPENDED;
1565                         else
1566                                 dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1567                         break;
1568                 case GRPQUOTA:
1569                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1570                         if (remount)
1571                                 dqopt->flags |= DQUOT_GRP_SUSPENDED;
1572                         else
1573                                 dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1574                         break;
1575         }
1576 }
1577
1578
1579 /*
1580  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1581  */
1582 int vfs_quota_off(struct super_block *sb, int type, int remount)
1583 {
1584         int cnt, ret = 0;
1585         struct quota_info *dqopt = sb_dqopt(sb);
1586         struct inode *toputinode[MAXQUOTAS];
1587
1588         /* We need to serialize quota_off() for device */
1589         mutex_lock(&dqopt->dqonoff_mutex);
1590
1591         /*
1592          * Skip everything if there's nothing to do. We have to do this because
1593          * sometimes we are called when fill_super() failed and calling
1594          * sync_fs() in such cases does no good.
1595          */
1596         if (!sb_any_quota_enabled(sb) && !sb_any_quota_suspended(sb)) {
1597                 mutex_unlock(&dqopt->dqonoff_mutex);
1598                 return 0;
1599         }
1600         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1601                 toputinode[cnt] = NULL;
1602                 if (type != -1 && cnt != type)
1603                         continue;
1604                 /* If we keep inodes of quota files after remount and quotaoff
1605                  * is called, drop kept inodes. */
1606                 if (!remount && sb_has_quota_suspended(sb, cnt)) {
1607                         iput(dqopt->files[cnt]);
1608                         dqopt->files[cnt] = NULL;
1609                         reset_enable_flags(dqopt, cnt, 0);
1610                         continue;
1611                 }
1612                 if (!sb_has_quota_enabled(sb, cnt))
1613                         continue;
1614                 reset_enable_flags(dqopt, cnt, remount);
1615
1616                 /* Note: these are blocking operations */
1617                 drop_dquot_ref(sb, cnt);
1618                 invalidate_dquots(sb, cnt);
1619                 /*
1620                  * Now all dquots should be invalidated, all writes done so we should be only
1621                  * users of the info. No locks needed.
1622                  */
1623                 if (info_dirty(&dqopt->info[cnt]))
1624                         sb->dq_op->write_info(sb, cnt);
1625                 if (dqopt->ops[cnt]->free_file_info)
1626                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1627                 put_quota_format(dqopt->info[cnt].dqi_format);
1628
1629                 toputinode[cnt] = dqopt->files[cnt];
1630                 if (!remount)
1631                         dqopt->files[cnt] = NULL;
1632                 dqopt->info[cnt].dqi_flags = 0;
1633                 dqopt->info[cnt].dqi_igrace = 0;
1634                 dqopt->info[cnt].dqi_bgrace = 0;
1635                 dqopt->ops[cnt] = NULL;
1636         }
1637         mutex_unlock(&dqopt->dqonoff_mutex);
1638         /* Sync the superblock so that buffers with quota data are written to
1639          * disk (and so userspace sees correct data afterwards). */
1640         if (sb->s_op->sync_fs)
1641                 sb->s_op->sync_fs(sb, 1);
1642         sync_blockdev(sb->s_bdev);
1643         /* Now the quota files are just ordinary files and we can set the
1644          * inode flags back. Moreover we discard the pagecache so that
1645          * userspace sees the writes we did bypassing the pagecache. We
1646          * must also discard the blockdev buffers so that we see the
1647          * changes done by userspace on the next quotaon() */
1648         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1649                 if (toputinode[cnt]) {
1650                         mutex_lock(&dqopt->dqonoff_mutex);
1651                         /* If quota was reenabled in the meantime, we have
1652                          * nothing to do */
1653                         if (!sb_has_quota_enabled(sb, cnt)) {
1654                                 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1655                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1656                                   S_NOATIME | S_NOQUOTA);
1657                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1658                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1659                                 mark_inode_dirty(toputinode[cnt]);
1660                         }
1661                         mutex_unlock(&dqopt->dqonoff_mutex);
1662                         /* On remount RO, we keep the inode pointer so that we
1663                          * can reenable quota on the subsequent remount RW.
1664                          * But we have better not keep inode pointer when there
1665                          * is pending delete on the quota file... */
1666                         if (!remount)
1667                                 iput(toputinode[cnt]);
1668                         else if (!toputinode[cnt]->i_nlink)
1669                                 ret = -EBUSY;
1670                 }
1671         if (sb->s_bdev)
1672                 invalidate_bdev(sb->s_bdev);
1673         return ret;
1674 }
1675
1676 /*
1677  *      Turn quotas on on a device
1678  */
1679
1680 /* Helper function when we already have the inode */
1681 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1682 {
1683         struct quota_format_type *fmt = find_quota_format(format_id);
1684         struct super_block *sb = inode->i_sb;
1685         struct quota_info *dqopt = sb_dqopt(sb);
1686         int error;
1687         int oldflags = -1;
1688
1689         if (!fmt)
1690                 return -ESRCH;
1691         if (!S_ISREG(inode->i_mode)) {
1692                 error = -EACCES;
1693                 goto out_fmt;
1694         }
1695         if (IS_RDONLY(inode)) {
1696                 error = -EROFS;
1697                 goto out_fmt;
1698         }
1699         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1700                 error = -EINVAL;
1701                 goto out_fmt;
1702         }
1703
1704         /* As we bypass the pagecache we must now flush the inode so that
1705          * we see all the changes from userspace... */
1706         write_inode_now(inode, 1);
1707         /* And now flush the block cache so that kernel sees the changes */
1708         invalidate_bdev(sb->s_bdev);
1709         mutex_lock(&inode->i_mutex);
1710         mutex_lock(&dqopt->dqonoff_mutex);
1711         if (sb_has_quota_enabled(sb, type) ||
1712                         sb_has_quota_suspended(sb, type)) {
1713                 error = -EBUSY;
1714                 goto out_lock;
1715         }
1716         /* We don't want quota and atime on quota files (deadlocks possible)
1717          * Also nobody should write to the file - we use special IO operations
1718          * which ignore the immutable bit. */
1719         down_write(&dqopt->dqptr_sem);
1720         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1721         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1722         up_write(&dqopt->dqptr_sem);
1723         sb->dq_op->drop(inode);
1724
1725         error = -EIO;
1726         dqopt->files[type] = igrab(inode);
1727         if (!dqopt->files[type])
1728                 goto out_lock;
1729         error = -EINVAL;
1730         if (!fmt->qf_ops->check_quota_file(sb, type))
1731                 goto out_file_init;
1732
1733         dqopt->ops[type] = fmt->qf_ops;
1734         dqopt->info[type].dqi_format = fmt;
1735         dqopt->info[type].dqi_fmt_id = format_id;
1736         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1737         mutex_lock(&dqopt->dqio_mutex);
1738         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1739                 mutex_unlock(&dqopt->dqio_mutex);
1740                 goto out_file_init;
1741         }
1742         mutex_unlock(&dqopt->dqio_mutex);
1743         mutex_unlock(&inode->i_mutex);
1744         set_enable_flags(dqopt, type);
1745
1746         add_dquot_ref(sb, type);
1747         mutex_unlock(&dqopt->dqonoff_mutex);
1748
1749         return 0;
1750
1751 out_file_init:
1752         dqopt->files[type] = NULL;
1753         iput(inode);
1754 out_lock:
1755         mutex_unlock(&dqopt->dqonoff_mutex);
1756         if (oldflags != -1) {
1757                 down_write(&dqopt->dqptr_sem);
1758                 /* Set the flags back (in the case of accidental quotaon()
1759                  * on a wrong file we don't want to mess up the flags) */
1760                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1761                 inode->i_flags |= oldflags;
1762                 up_write(&dqopt->dqptr_sem);
1763         }
1764         mutex_unlock(&inode->i_mutex);
1765 out_fmt:
1766         put_quota_format(fmt);
1767
1768         return error;
1769 }
1770
1771 /* Reenable quotas on remount RW */
1772 static int vfs_quota_on_remount(struct super_block *sb, int type)
1773 {
1774         struct quota_info *dqopt = sb_dqopt(sb);
1775         struct inode *inode;
1776         int ret;
1777
1778         mutex_lock(&dqopt->dqonoff_mutex);
1779         if (!sb_has_quota_suspended(sb, type)) {
1780                 mutex_unlock(&dqopt->dqonoff_mutex);
1781                 return 0;
1782         }
1783         BUG_ON(sb_has_quota_enabled(sb, type));
1784
1785         inode = dqopt->files[type];
1786         dqopt->files[type] = NULL;
1787         reset_enable_flags(dqopt, type, 0);
1788         mutex_unlock(&dqopt->dqonoff_mutex);
1789
1790         ret = vfs_quota_on_inode(inode, type, dqopt->info[type].dqi_fmt_id);
1791         iput(inode);
1792
1793         return ret;
1794 }
1795
1796 /* Actual function called from quotactl() */
1797 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path,
1798                  int remount)
1799 {
1800         struct nameidata nd;
1801         int error;
1802
1803         if (remount)
1804                 return vfs_quota_on_remount(sb, type);
1805
1806         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1807         if (error < 0)
1808                 return error;
1809         error = security_quota_on(nd.path.dentry);
1810         if (error)
1811                 goto out_path;
1812         /* Quota file not on the same filesystem? */
1813         if (nd.path.mnt->mnt_sb != sb)
1814                 error = -EXDEV;
1815         else
1816                 error = vfs_quota_on_inode(nd.path.dentry->d_inode, type,
1817                                            format_id);
1818 out_path:
1819         path_put(&nd.path);
1820         return error;
1821 }
1822
1823 /*
1824  * This function is used when filesystem needs to initialize quotas
1825  * during mount time.
1826  */
1827 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1828                 int format_id, int type)
1829 {
1830         struct dentry *dentry;
1831         int error;
1832
1833         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1834         if (IS_ERR(dentry))
1835                 return PTR_ERR(dentry);
1836
1837         if (!dentry->d_inode) {
1838                 error = -ENOENT;
1839                 goto out;
1840         }
1841
1842         error = security_quota_on(dentry);
1843         if (!error)
1844                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1845
1846 out:
1847         dput(dentry);
1848         return error;
1849 }
1850
1851 /* Wrapper to turn on quotas when remounting rw */
1852 int vfs_dq_quota_on_remount(struct super_block *sb)
1853 {
1854         int cnt;
1855         int ret = 0, err;
1856
1857         if (!sb->s_qcop || !sb->s_qcop->quota_on)
1858                 return -ENOSYS;
1859         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1860                 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
1861                 if (err < 0 && !ret)
1862                         ret = err;
1863         }
1864         return ret;
1865 }
1866
1867 /* Generic routine for getting common part of quota structure */
1868 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1869 {
1870         struct mem_dqblk *dm = &dquot->dq_dqb;
1871
1872         spin_lock(&dq_data_lock);
1873         di->dqb_bhardlimit = dm->dqb_bhardlimit;
1874         di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
1875         di->dqb_curspace = dm->dqb_curspace;
1876         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1877         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1878         di->dqb_curinodes = dm->dqb_curinodes;
1879         di->dqb_btime = dm->dqb_btime;
1880         di->dqb_itime = dm->dqb_itime;
1881         di->dqb_valid = QIF_ALL;
1882         spin_unlock(&dq_data_lock);
1883 }
1884
1885 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1886 {
1887         struct dquot *dquot;
1888
1889         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1890         if (!(dquot = dqget(sb, id, type))) {
1891                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1892                 return -ESRCH;
1893         }
1894         do_get_dqblk(dquot, di);
1895         dqput(dquot);
1896         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1897         return 0;
1898 }
1899
1900 /* Generic routine for setting common part of quota structure */
1901 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1902 {
1903         struct mem_dqblk *dm = &dquot->dq_dqb;
1904         int check_blim = 0, check_ilim = 0;
1905         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1906
1907         if ((di->dqb_valid & QIF_BLIMITS &&
1908              (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
1909               di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
1910             (di->dqb_valid & QIF_ILIMITS &&
1911              (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
1912               di->dqb_isoftlimit > dqi->dqi_maxilimit)))
1913                 return -ERANGE;
1914
1915         spin_lock(&dq_data_lock);
1916         if (di->dqb_valid & QIF_SPACE) {
1917                 dm->dqb_curspace = di->dqb_curspace;
1918                 check_blim = 1;
1919         }
1920         if (di->dqb_valid & QIF_BLIMITS) {
1921                 dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
1922                 dm->dqb_bhardlimit = di->dqb_bhardlimit;
1923                 check_blim = 1;
1924         }
1925         if (di->dqb_valid & QIF_INODES) {
1926                 dm->dqb_curinodes = di->dqb_curinodes;
1927                 check_ilim = 1;
1928         }
1929         if (di->dqb_valid & QIF_ILIMITS) {
1930                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1931                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1932                 check_ilim = 1;
1933         }
1934         if (di->dqb_valid & QIF_BTIME)
1935                 dm->dqb_btime = di->dqb_btime;
1936         if (di->dqb_valid & QIF_ITIME)
1937                 dm->dqb_itime = di->dqb_itime;
1938
1939         if (check_blim) {
1940                 if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
1941                         dm->dqb_btime = 0;
1942                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1943                 }
1944                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1945                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
1946         }
1947         if (check_ilim) {
1948                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1949                         dm->dqb_itime = 0;
1950                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1951                 }
1952                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1953                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
1954         }
1955         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1956                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1957         else
1958                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1959         spin_unlock(&dq_data_lock);
1960         mark_dquot_dirty(dquot);
1961
1962         return 0;
1963 }
1964
1965 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1966 {
1967         struct dquot *dquot;
1968         int rc;
1969
1970         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1971         if (!(dquot = dqget(sb, id, type))) {
1972                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1973                 return -ESRCH;
1974         }
1975         rc = do_set_dqblk(dquot, di);
1976         dqput(dquot);
1977         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1978         return rc;
1979 }
1980
1981 /* Generic routine for getting common part of quota file information */
1982 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
1983 {
1984         struct mem_dqinfo *mi;
1985
1986         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1987         if (!sb_has_quota_enabled(sb, type)) {
1988                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1989                 return -ESRCH;
1990         }
1991         mi = sb_dqopt(sb)->info + type;
1992         spin_lock(&dq_data_lock);
1993         ii->dqi_bgrace = mi->dqi_bgrace;
1994         ii->dqi_igrace = mi->dqi_igrace;
1995         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
1996         ii->dqi_valid = IIF_ALL;
1997         spin_unlock(&dq_data_lock);
1998         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1999         return 0;
2000 }
2001
2002 /* Generic routine for setting common part of quota file information */
2003 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2004 {
2005         struct mem_dqinfo *mi;
2006
2007         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2008         if (!sb_has_quota_enabled(sb, type)) {
2009                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2010                 return -ESRCH;
2011         }
2012         mi = sb_dqopt(sb)->info + type;
2013         spin_lock(&dq_data_lock);
2014         if (ii->dqi_valid & IIF_BGRACE)
2015                 mi->dqi_bgrace = ii->dqi_bgrace;
2016         if (ii->dqi_valid & IIF_IGRACE)
2017                 mi->dqi_igrace = ii->dqi_igrace;
2018         if (ii->dqi_valid & IIF_FLAGS)
2019                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
2020         spin_unlock(&dq_data_lock);
2021         mark_info_dirty(sb, type);
2022         /* Force write to disk */
2023         sb->dq_op->write_info(sb, type);
2024         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2025         return 0;
2026 }
2027
2028 struct quotactl_ops vfs_quotactl_ops = {
2029         .quota_on       = vfs_quota_on,
2030         .quota_off      = vfs_quota_off,
2031         .quota_sync     = vfs_quota_sync,
2032         .get_info       = vfs_get_dqinfo,
2033         .set_info       = vfs_set_dqinfo,
2034         .get_dqblk      = vfs_get_dqblk,
2035         .set_dqblk      = vfs_set_dqblk
2036 };
2037
2038 static ctl_table fs_dqstats_table[] = {
2039         {
2040                 .ctl_name       = FS_DQ_LOOKUPS,
2041                 .procname       = "lookups",
2042                 .data           = &dqstats.lookups,
2043                 .maxlen         = sizeof(int),
2044                 .mode           = 0444,
2045                 .proc_handler   = &proc_dointvec,
2046         },
2047         {
2048                 .ctl_name       = FS_DQ_DROPS,
2049                 .procname       = "drops",
2050                 .data           = &dqstats.drops,
2051                 .maxlen         = sizeof(int),
2052                 .mode           = 0444,
2053                 .proc_handler   = &proc_dointvec,
2054         },
2055         {
2056                 .ctl_name       = FS_DQ_READS,
2057                 .procname       = "reads",
2058                 .data           = &dqstats.reads,
2059                 .maxlen         = sizeof(int),
2060                 .mode           = 0444,
2061                 .proc_handler   = &proc_dointvec,
2062         },
2063         {
2064                 .ctl_name       = FS_DQ_WRITES,
2065                 .procname       = "writes",
2066                 .data           = &dqstats.writes,
2067                 .maxlen         = sizeof(int),
2068                 .mode           = 0444,
2069                 .proc_handler   = &proc_dointvec,
2070         },
2071         {
2072                 .ctl_name       = FS_DQ_CACHE_HITS,
2073                 .procname       = "cache_hits",
2074                 .data           = &dqstats.cache_hits,
2075                 .maxlen         = sizeof(int),
2076                 .mode           = 0444,
2077                 .proc_handler   = &proc_dointvec,
2078         },
2079         {
2080                 .ctl_name       = FS_DQ_ALLOCATED,
2081                 .procname       = "allocated_dquots",
2082                 .data           = &dqstats.allocated_dquots,
2083                 .maxlen         = sizeof(int),
2084                 .mode           = 0444,
2085                 .proc_handler   = &proc_dointvec,
2086         },
2087         {
2088                 .ctl_name       = FS_DQ_FREE,
2089                 .procname       = "free_dquots",
2090                 .data           = &dqstats.free_dquots,
2091                 .maxlen         = sizeof(int),
2092                 .mode           = 0444,
2093                 .proc_handler   = &proc_dointvec,
2094         },
2095         {
2096                 .ctl_name       = FS_DQ_SYNCS,
2097                 .procname       = "syncs",
2098                 .data           = &dqstats.syncs,
2099                 .maxlen         = sizeof(int),
2100                 .mode           = 0444,
2101                 .proc_handler   = &proc_dointvec,
2102         },
2103 #ifdef CONFIG_PRINT_QUOTA_WARNING
2104         {
2105                 .ctl_name       = FS_DQ_WARNINGS,
2106                 .procname       = "warnings",
2107                 .data           = &flag_print_warnings,
2108                 .maxlen         = sizeof(int),
2109                 .mode           = 0644,
2110                 .proc_handler   = &proc_dointvec,
2111         },
2112 #endif
2113         { .ctl_name = 0 },
2114 };
2115
2116 static ctl_table fs_table[] = {
2117         {
2118                 .ctl_name       = FS_DQSTATS,
2119                 .procname       = "quota",
2120                 .mode           = 0555,
2121                 .child          = fs_dqstats_table,
2122         },
2123         { .ctl_name = 0 },
2124 };
2125
2126 static ctl_table sys_table[] = {
2127         {
2128                 .ctl_name       = CTL_FS,
2129                 .procname       = "fs",
2130                 .mode           = 0555,
2131                 .child          = fs_table,
2132         },
2133         { .ctl_name = 0 },
2134 };
2135
2136 static int __init dquot_init(void)
2137 {
2138         int i;
2139         unsigned long nr_hash, order;
2140
2141         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2142
2143         register_sysctl_table(sys_table);
2144
2145         dquot_cachep = kmem_cache_create("dquot",
2146                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2147                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2148                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2149                         NULL);
2150
2151         order = 0;
2152         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2153         if (!dquot_hash)
2154                 panic("Cannot create dquot hash table");
2155
2156         /* Find power-of-two hlist_heads which can fit into allocation */
2157         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2158         dq_hash_bits = 0;
2159         do {
2160                 dq_hash_bits++;
2161         } while (nr_hash >> dq_hash_bits);
2162         dq_hash_bits--;
2163
2164         nr_hash = 1UL << dq_hash_bits;
2165         dq_hash_mask = nr_hash - 1;
2166         for (i = 0; i < nr_hash; i++)
2167                 INIT_HLIST_HEAD(dquot_hash + i);
2168
2169         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2170                         nr_hash, order, (PAGE_SIZE << order));
2171
2172         register_shrinker(&dqcache_shrinker);
2173
2174 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2175         if (genl_register_family(&quota_genl_family) != 0)
2176                 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2177 #endif
2178
2179         return 0;
2180 }
2181 module_init(dquot_init);
2182
2183 EXPORT_SYMBOL(register_quota_format);
2184 EXPORT_SYMBOL(unregister_quota_format);
2185 EXPORT_SYMBOL(dqstats);
2186 EXPORT_SYMBOL(dq_data_lock);
2187 EXPORT_SYMBOL(vfs_quota_on);
2188 EXPORT_SYMBOL(vfs_quota_on_mount);
2189 EXPORT_SYMBOL(vfs_quota_off);
2190 EXPORT_SYMBOL(vfs_quota_sync);
2191 EXPORT_SYMBOL(vfs_get_dqinfo);
2192 EXPORT_SYMBOL(vfs_set_dqinfo);
2193 EXPORT_SYMBOL(vfs_get_dqblk);
2194 EXPORT_SYMBOL(vfs_set_dqblk);
2195 EXPORT_SYMBOL(dquot_commit);
2196 EXPORT_SYMBOL(dquot_commit_info);
2197 EXPORT_SYMBOL(dquot_acquire);
2198 EXPORT_SYMBOL(dquot_release);
2199 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2200 EXPORT_SYMBOL(dquot_initialize);
2201 EXPORT_SYMBOL(dquot_drop);
2202 EXPORT_SYMBOL(vfs_dq_drop);
2203 EXPORT_SYMBOL(dquot_alloc_space);
2204 EXPORT_SYMBOL(dquot_alloc_inode);
2205 EXPORT_SYMBOL(dquot_free_space);
2206 EXPORT_SYMBOL(dquot_free_inode);
2207 EXPORT_SYMBOL(dquot_transfer);
2208 EXPORT_SYMBOL(vfs_dq_transfer);
2209 EXPORT_SYMBOL(vfs_dq_quota_on_remount);