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  * Author:      Marco van Wieringen <[email protected]>
  13  *
  14  * Fixes:   Dmitry Gorodchanin <[email protected]>, 11 Feb 96
  15  *
  16  *              Revised list management to avoid races
  17  *              -- Bill Hawes, <[email protected]>, 9/98
  18  *
  19  *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
  20  *              As the consequence the locking was moved from dquot_decr_...(),
  21  *              dquot_incr_...() to calling functions.
  22  *              invalidate_dquots() now writes modified dquots.
  23  *              Serialized quota_off() and quota_on() for mount point.
  24  *              Fixed a few bugs in grow_dquots().
  25  *              Fixed deadlock in write_dquot() - we no longer account quotas on
  26  *              quota files
  27  *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
  28  *              add_dquot_ref() restarts after blocking
  29  *              Added check for bogus uid and fixed check for group in quotactl.
  30  *              Jan Kara, <[email protected]>, sponsored by SuSE CR, 10-11/99
  31  *
  32  *              Used struct list_head instead of own list struct
  33  *              Invalidation of referenced dquots is no longer possible
  34  *              Improved free_dquots list management
  35  *              Quota and i_blocks are now updated in one place to avoid races
  36  *              Warnings are now delayed so we won't block in critical section
  37  *              Write updated not to require dquot lock
  38  *              Jan Kara, <[email protected]>, 9/2000
  39  *
  40  *              Added dynamic quota structure allocation
  41  *              Jan Kara <[email protected]> 12/2000
  42  *
  43  *              Rewritten quota interface. Implemented new quota format and
  44  *              formats registering.
  45  *              Jan Kara, <[email protected]>, 2001,2002
  46  *
  47  *              New SMP locking.
  48  *              Jan Kara, <[email protected]>, 10/2002
  49  *
  50  *              Added journalled quota support, fix lock inversion problems
  51  *              Jan Kara, <[email protected]>, 2003,2004
  52  *
  53  * (C) Copyright 1994 - 1997 Marco van Wieringen
  54  */
  55
  56 #include <linux/errno.h>
  57 #include <linux/kernel.h>
  58 #include <linux/fs.h>
  59 #include <linux/mount.h>
  60 #include <linux/mm.h>
  61 #include <linux/time.h>
  62 #include <linux/types.h>
  63 #include <linux/string.h>
  64 #include <linux/fcntl.h>
  65 #include <linux/stat.h>
  66 #include <linux/tty.h>
  67 #include <linux/file.h>
  68 #include <linux/slab.h>
  69 #include <linux/sysctl.h>
  70 #include <linux/init.h>
  71 #include <linux/module.h>
  72 #include <linux/proc_fs.h>
  73 #include <linux/security.h>
  74 #include <linux/kmod.h>
  75 #include <linux/namei.h>
  76 #include <linux/buffer_head.h>
  77 #include <linux/capability.h>
  78 #include <linux/quotaops.h>
  79 #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
  80 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
  81 #include <net/netlink.h>
  82 #include <net/genetlink.h>
  83 #endif
  84
  85 #include <asm/uaccess.h>
  86
  87 #define __DQUOT_PARANOIA
  88
  89 /*
  90  * There are two quota SMP locks. dq_list_lock protects all lists with quotas
  91  * and quota formats and also dqstats structure containing statistics about the
  92  * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
  93  * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
  94  * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
  95  * in inode_add_bytes() and inode_sub_bytes().
  96  *
  97  * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
  98  *
  99  * Note that some things (eg. sb pointer, type, id) doesn't change during
 100  * the life of the dquot structure and so needn't to be protected by a lock
 101  *
 102  * Any operation working on dquots via inode pointers must hold dqptr_sem.  If
 103  * operation is just reading pointers from inode (or not using them at all) the
 104  * read lock is enough. If pointers are altered function must hold write lock
 105  * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
 106  * for altering the flag i_mutex is also needed).  If operation is holding
 107  * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
 108  * dqonoff_mutex.
 109  * This locking assures that:
 110  *   a) update/access to dquot pointers in inode is serialized
 111  *   b) everyone is guarded against invalidate_dquots()
 112  *
 113  * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
 114  * from inodes (dquot_alloc_space() and such don't check the dq_lock).
 115  * Currently dquot is locked only when it is being read to memory (or space for
 116  * it is being allocated) on the first dqget() and when it is being released on
 117  * the last dqput(). The allocation and release oparations are serialized by
 118  * the dq_lock and by checking the use count in dquot_release().  Write
 119  * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
 120  * spinlock to internal buffers before writing.
 121  *
 122  * Lock ordering (including related VFS locks) is the following:
 123  *   i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
 124  *   dqio_mutex
 125  * i_mutex on quota files is special (it's below dqio_mutex)
 126  */
 127
 128 static DEFINE_SPINLOCK(dq_list_lock);
 129 DEFINE_SPINLOCK(dq_data_lock);
 130
 131 static char *quotatypes[] = INITQFNAMES;
 132 static struct quota_format_type *quota_formats; /* List of registered formats */
 133 static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 134
 135 /* SLAB cache for dquot structures */
 136 static struct kmem_cache *dquot_cachep;
 137
 138 int register_quota_format(struct quota_format_type *fmt)
 139 {
 140         spin_lock(&dq_list_lock);
 141         fmt->qf_next = quota_formats;
 142         quota_formats = fmt;
 143         spin_unlock(&dq_list_lock);
 144         return 0;
 145 }
 146
 147 void unregister_quota_format(struct quota_format_type *fmt)
 148 {
 149         struct quota_format_type **actqf;
 150
 151         spin_lock(&dq_list_lock);
 152         for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
 153         if (*actqf)
 154                 *actqf = (*actqf)->qf_next;
 155         spin_unlock(&dq_list_lock);
 156 }
 157
 158 static struct quota_format_type *find_quota_format(int id)
 159 {
 160         struct quota_format_type *actqf;
 161
 162         spin_lock(&dq_list_lock);
 163         for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 164         if (!actqf || !try_module_get(actqf->qf_owner)) {
 165                 int qm;
 166
 167                 spin_unlock(&dq_list_lock);
 168
 169                 for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
 170                 if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
 171                         return NULL;
 172
 173                 spin_lock(&dq_list_lock);
 174                 for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
 175                 if (actqf && !try_module_get(actqf->qf_owner))
 176                         actqf = NULL;
 177         }
 178         spin_unlock(&dq_list_lock);
 179         return actqf;
 180 }
 181
 182 static void put_quota_format(struct quota_format_type *fmt)
 183 {
 184         module_put(fmt->qf_owner);
 185 }
 186
 187 /*
 188  * Dquot List Management:
 189  * The quota code uses three lists for dquot management: the inuse_list,
 190  * free_dquots, and dquot_hash[] array. A single dquot structure may be
 191  * on all three lists, depending on its current state.
 192  *
 193  * All dquots are placed to the end of inuse_list when first created, and this
 194  * list is used for invalidate operation, which must look at every dquot.
 195  *
 196  * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 197  * and this list is searched whenever we need an available dquot.  Dquots are
 198  * removed from the list as soon as they are used again, and
 199  * dqstats.free_dquots gives the number of dquots on the list. When
 200  * dquot is invalidated it's completely released from memory.
 201  *
 202  * Dquots with a specific identity (device, type and id) are placed on
 203  * one of the dquot_hash[] hash chains. The provides an efficient search
 204  * mechanism to locate a specific dquot.
 205  */
 206
 207 static LIST_HEAD(inuse_list);
 208 static LIST_HEAD(free_dquots);
 209 static unsigned int dq_hash_bits, dq_hash_mask;
 210 static struct hlist_head *dquot_hash;
 211
 212 struct dqstats dqstats;
 213
 214 static void dqput(struct dquot *dquot);
 215
 216 static inline unsigned int
 217 hashfn(const struct super_block *sb, unsigned int id, int type)
 218 {
 219         unsigned long tmp;
 220
 221         tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 222         return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 223 }
 224
 225 /*
 226  * Following list functions expect dq_list_lock to be held
 227  */
 228 static inline void insert_dquot_hash(struct dquot *dquot)
 229 {
 230         struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
 231         hlist_add_head(&dquot->dq_hash, head);
 232 }
 233
 234 static inline void remove_dquot_hash(struct dquot *dquot)
 235 {
 236         hlist_del_init(&dquot->dq_hash);
 237 }
 238
 239 static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
 240 {
 241         struct hlist_node *node;
 242         struct dquot *dquot;
 243
 244         hlist_for_each (node, dquot_hash+hashent) {
 245                 dquot = hlist_entry(node, struct dquot, dq_hash);
 246                 if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
 247                         return dquot;
 248         }
 249         return NODQUOT;
 250 }
 251
 252 /* Add a dquot to the tail of the free list */
 253 static inline void put_dquot_last(struct dquot *dquot)
 254 {
 255         list_add_tail(&dquot->dq_free, &free_dquots);
 256         dqstats.free_dquots++;
 257 }
 258
 259 static inline void remove_free_dquot(struct dquot *dquot)
 260 {
 261         if (list_empty(&dquot->dq_free))
 262                 return;
 263         list_del_init(&dquot->dq_free);
 264         dqstats.free_dquots--;
 265 }
 266
 267 static inline void put_inuse(struct dquot *dquot)
 268 {
 269         /* We add to the back of inuse list so we don't have to restart
 270          * when traversing this list and we block */
 271         list_add_tail(&dquot->dq_inuse, &inuse_list);
 272         dqstats.allocated_dquots++;
 273 }
 274
 275 static inline void remove_inuse(struct dquot *dquot)
 276 {
 277         dqstats.allocated_dquots--;
 278         list_del(&dquot->dq_inuse);
 279 }
 280 /*
 281  * End of list functions needing dq_list_lock
 282  */
 283
 284 static void wait_on_dquot(struct dquot *dquot)
 285 {
 286         mutex_lock(&dquot->dq_lock);
 287         mutex_unlock(&dquot->dq_lock);
 288 }
 289
 290 static inline int dquot_dirty(struct dquot *dquot)
 291 {
 292         return test_bit(DQ_MOD_B, &dquot->dq_flags);
 293 }
 294
 295 static inline int mark_dquot_dirty(struct dquot *dquot)
 296 {
 297         return dquot->dq_sb->dq_op->mark_dirty(dquot);
 298 }
 299
 300 int dquot_mark_dquot_dirty(struct dquot *dquot)
 301 {
 302         spin_lock(&dq_list_lock);
 303         if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
 304                 list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 305                                 info[dquot->dq_type].dqi_dirty_list);
 306         spin_unlock(&dq_list_lock);
 307         return 0;
 308 }
 309
 310 /* This function needs dq_list_lock */
 311 static inline int clear_dquot_dirty(struct dquot *dquot)
 312 {
 313         if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
 314                 return 0;
 315         list_del_init(&dquot->dq_dirty);
 316         return 1;
 317 }
 318
 319 void mark_info_dirty(struct super_block *sb, int type)
 320 {
 321         set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
 322 }
 323 EXPORT_SYMBOL(mark_info_dirty);
 324
 325 /*
 326  *      Read dquot from disk and alloc space for it
 327  */
 328
 329 int dquot_acquire(struct dquot *dquot)
 330 {
 331         int ret = 0, ret2 = 0;
 332         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 333
 334         mutex_lock(&dquot->dq_lock);
 335         mutex_lock(&dqopt->dqio_mutex);
 336         if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 337                 ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
 338         if (ret < 0)
 339                 goto out_iolock;
 340         set_bit(DQ_READ_B, &dquot->dq_flags);
 341         /* Instantiate dquot if needed */
 342         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 343                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 344                 /* Write the info if needed */
 345                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 346                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 347                 if (ret < 0)
 348                         goto out_iolock;
 349                 if (ret2 < 0) {
 350                         ret = ret2;
 351                         goto out_iolock;
 352                 }
 353         }
 354         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 355 out_iolock:
 356         mutex_unlock(&dqopt->dqio_mutex);
 357         mutex_unlock(&dquot->dq_lock);
 358         return ret;
 359 }
 360
 361 /*
 362  *      Write dquot to disk
 363  */
 364 int dquot_commit(struct dquot *dquot)
 365 {
 366         int ret = 0, ret2 = 0;
 367         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 368
 369         mutex_lock(&dqopt->dqio_mutex);
 370         spin_lock(&dq_list_lock);
 371         if (!clear_dquot_dirty(dquot)) {
 372                 spin_unlock(&dq_list_lock);
 373                 goto out_sem;
 374         }
 375         spin_unlock(&dq_list_lock);
 376         /* Inactive dquot can be only if there was error during read/init
 377          * => we have better not writing it */
 378         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 379                 ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
 380                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 381                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 382                 if (ret >= 0)
 383                         ret = ret2;
 384         }
 385 out_sem:
 386         mutex_unlock(&dqopt->dqio_mutex);
 387         return ret;
 388 }
 389
 390 /*
 391  *      Release dquot
 392  */
 393 int dquot_release(struct dquot *dquot)
 394 {
 395         int ret = 0, ret2 = 0;
 396         struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 397
 398         mutex_lock(&dquot->dq_lock);
 399         /* Check whether we are not racing with some other dqget() */
 400         if (atomic_read(&dquot->dq_count) > 1)
 401                 goto out_dqlock;
 402         mutex_lock(&dqopt->dqio_mutex);
 403         if (dqopt->ops[dquot->dq_type]->release_dqblk) {
 404                 ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
 405                 /* Write the info */
 406                 if (info_dirty(&dqopt->info[dquot->dq_type]))
 407                         ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
 408                 if (ret >= 0)
 409                         ret = ret2;
 410         }
 411         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 412         mutex_unlock(&dqopt->dqio_mutex);
 413 out_dqlock:
 414         mutex_unlock(&dquot->dq_lock);
 415         return ret;
 416 }
 417
 418 static void dquot_destroy(struct dquot *dquot)
 419 {
 420         kmem_cache_free(dquot_cachep, dquot);
 421 }
 422
 423 static inline void do_destroy_dquot(struct dquot *dquot)
 424 {
 425         dquot->dq_sb->dq_op->destroy_dquot(dquot);
 426 }
 427
 428 /* Invalidate all dquots on the list. Note that this function is called after
 429  * quota is disabled and pointers from inodes removed so there cannot be new
 430  * quota users. There can still be some users of quotas due to inodes being
 431  * just deleted or pruned by prune_icache() (those are not attached to any
 432  * list). We have to wait for such users.
 433  */
 434 static void invalidate_dquots(struct super_block *sb, int type)
 435 {
 436         struct dquot *dquot, *tmp;
 437
 438 restart:
 439         spin_lock(&dq_list_lock);
 440         list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 441                 if (dquot->dq_sb != sb)
 442                         continue;
 443                 if (dquot->dq_type != type)
 444                         continue;
 445                 /* Wait for dquot users */
 446                 if (atomic_read(&dquot->dq_count)) {
 447                         DEFINE_WAIT(wait);
 448
 449                         atomic_inc(&dquot->dq_count);
 450                         prepare_to_wait(&dquot->dq_wait_unused, &wait,
 451                                         TASK_UNINTERRUPTIBLE);
 452                         spin_unlock(&dq_list_lock);
 453                         /* Once dqput() wakes us up, we know it's time to free
 454                          * the dquot.
 455                          * IMPORTANT: we rely on the fact that there is always
 456                          * at most one process waiting for dquot to free.
 457                          * Otherwise dq_count would be > 1 and we would never
 458                          * wake up.
 459                          */
 460                         if (atomic_read(&dquot->dq_count) > 1)
 461                                 schedule();
 462                         finish_wait(&dquot->dq_wait_unused, &wait);
 463                         dqput(dquot);
 464                         /* At this moment dquot() need not exist (it could be
 465                          * reclaimed by prune_dqcache(). Hence we must
 466                          * restart. */
 467                         goto restart;
 468                 }
 469                 /*
 470                  * Quota now has no users and it has been written on last
 471                  * dqput()
 472                  */
 473                 remove_dquot_hash(dquot);
 474                 remove_free_dquot(dquot);
 475                 remove_inuse(dquot);
 476                 do_destroy_dquot(dquot);
 477         }
 478         spin_unlock(&dq_list_lock);
 479 }
 480
 481 int vfs_quota_sync(struct super_block *sb, int type)
 482 {
 483         struct list_head *dirty;
 484         struct dquot *dquot;
 485         struct quota_info *dqopt = sb_dqopt(sb);
 486         int cnt;
 487
 488         mutex_lock(&dqopt->dqonoff_mutex);
 489         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 490                 if (type != -1 && cnt != type)
 491                         continue;
 492                 if (!sb_has_quota_enabled(sb, cnt))
 493                         continue;
 494                 spin_lock(&dq_list_lock);
 495                 dirty = &dqopt->info[cnt].dqi_dirty_list;
 496                 while (!list_empty(dirty)) {
 497                         dquot = list_first_entry(dirty, struct dquot, dq_dirty);
 498                         /* Dirty and inactive can be only bad dquot... */
 499                         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 500                                 clear_dquot_dirty(dquot);
 501                                 continue;
 502                         }
 503                         /* Now we have active dquot from which someone is
 504                          * holding reference so we can safely just increase
 505                          * use count */
 506                         atomic_inc(&dquot->dq_count);
 507                         dqstats.lookups++;
 508                         spin_unlock(&dq_list_lock);
 509                         sb->dq_op->write_dquot(dquot);
 510                         dqput(dquot);
 511                         spin_lock(&dq_list_lock);
 512                 }
 513                 spin_unlock(&dq_list_lock);
 514         }
 515
 516         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 517                 if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
 518                         && info_dirty(&dqopt->info[cnt]))
 519                         sb->dq_op->write_info(sb, cnt);
 520         spin_lock(&dq_list_lock);
 521         dqstats.syncs++;
 522         spin_unlock(&dq_list_lock);
 523         mutex_unlock(&dqopt->dqonoff_mutex);
 524
 525         return 0;
 526 }
 527
 528 /* Free unused dquots from cache */
 529 static void prune_dqcache(int count)
 530 {
 531         struct list_head *head;
 532         struct dquot *dquot;
 533
 534         head = free_dquots.prev;
 535         while (head != &free_dquots && count) {
 536                 dquot = list_entry(head, struct dquot, dq_free);
 537                 remove_dquot_hash(dquot);
 538                 remove_free_dquot(dquot);
 539                 remove_inuse(dquot);
 540                 do_destroy_dquot(dquot);
 541                 count--;
 542                 head = free_dquots.prev;
 543         }
 544 }
 545
 546 /*
 547  * This is called from kswapd when we think we need some
 548  * more memory
 549  */
 550
 551 static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
 552 {
 553         if (nr) {
 554                 spin_lock(&dq_list_lock);
 555                 prune_dqcache(nr);
 556                 spin_unlock(&dq_list_lock);
 557         }
 558         return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
 559 }
 560
 561 static struct shrinker dqcache_shrinker = {
 562         .shrink = shrink_dqcache_memory,
 563         .seeks = DEFAULT_SEEKS,
 564 };
 565
 566 /*
 567  * Put reference to dquot
 568  * NOTE: If you change this function please check whether dqput_blocks() works right...
 569  * MUST be called with either dqptr_sem or dqonoff_mutex held
 570  */
 571 static void dqput(struct dquot *dquot)
 572 {
 573         int ret;
 574
 575         if (!dquot)
 576                 return;
 577 #ifdef __DQUOT_PARANOIA
 578         if (!atomic_read(&dquot->dq_count)) {
 579                 printk("VFS: dqput: trying to free free dquot\n");
 580                 printk("VFS: device %s, dquot of %s %d\n",
 581                         dquot->dq_sb->s_id,
 582                         quotatypes[dquot->dq_type],
 583                         dquot->dq_id);
 584                 BUG();
 585         }
 586 #endif
 587
 588         spin_lock(&dq_list_lock);
 589         dqstats.drops++;
 590         spin_unlock(&dq_list_lock);
 591 we_slept:
 592         spin_lock(&dq_list_lock);
 593         if (atomic_read(&dquot->dq_count) > 1) {
 594                 /* We have more than one user... nothing to do */
 595                 atomic_dec(&dquot->dq_count);
 596                 /* Releasing dquot during quotaoff phase? */
 597                 if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
 598                     atomic_read(&dquot->dq_count) == 1)
 599                         wake_up(&dquot->dq_wait_unused);
 600                 spin_unlock(&dq_list_lock);
 601                 return;
 602         }
 603         /* Need to release dquot? */
 604         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 605                 spin_unlock(&dq_list_lock);
 606                 /* Commit dquot before releasing */
 607                 ret = dquot->dq_sb->dq_op->write_dquot(dquot);
 608                 if (ret < 0) {
 609                         printk(KERN_ERR "VFS: cannot write quota structure on "
 610                                 "device %s (error %d). Quota may get out of "
 611                                 "sync!\n", dquot->dq_sb->s_id, ret);
 612                         /*
 613                          * We clear dirty bit anyway, so that we avoid
 614                          * infinite loop here
 615                          */
 616                         spin_lock(&dq_list_lock);
 617                         clear_dquot_dirty(dquot);
 618                         spin_unlock(&dq_list_lock);
 619                 }
 620                 goto we_slept;
 621         }
 622         /* Clear flag in case dquot was inactive (something bad happened) */
 623         clear_dquot_dirty(dquot);
 624         if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 625                 spin_unlock(&dq_list_lock);
 626                 dquot->dq_sb->dq_op->release_dquot(dquot);
 627                 goto we_slept;
 628         }
 629         atomic_dec(&dquot->dq_count);
 630 #ifdef __DQUOT_PARANOIA
 631         /* sanity check */
 632         BUG_ON(!list_empty(&dquot->dq_free));
 633 #endif
 634         put_dquot_last(dquot);
 635         spin_unlock(&dq_list_lock);
 636 }
 637
 638 static struct dquot *dquot_alloc(struct super_block *sb, int type)
 639 {
 640         return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
 641 }
 642
 643 static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 644 {
 645         struct dquot *dquot;
 646
 647         dquot = sb->dq_op->alloc_dquot(sb, type);
 648         if(!dquot)
 649                 return NODQUOT;
 650
 651         mutex_init(&dquot->dq_lock);
 652         INIT_LIST_HEAD(&dquot->dq_free);
 653         INIT_LIST_HEAD(&dquot->dq_inuse);
 654         INIT_HLIST_NODE(&dquot->dq_hash);
 655         INIT_LIST_HEAD(&dquot->dq_dirty);
 656         init_waitqueue_head(&dquot->dq_wait_unused);
 657         dquot->dq_sb = sb;
 658         dquot->dq_type = type;
 659         atomic_set(&dquot->dq_count, 1);
 660
 661         return dquot;
 662 }
 663
 664 /*
 665  * Get reference to dquot
 666  * MUST be called with either dqptr_sem or dqonoff_mutex held
 667  */
 668 static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
 669 {
 670         unsigned int hashent = hashfn(sb, id, type);
 671         struct dquot *dquot, *empty = NODQUOT;
 672
 673         if (!sb_has_quota_enabled(sb, type))
 674                 return NODQUOT;
 675 we_slept:
 676         spin_lock(&dq_list_lock);
 677         if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
 678                 if (empty == NODQUOT) {
 679                         spin_unlock(&dq_list_lock);
 680                         if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
 681                                 schedule();     /* Try to wait for a moment... */
 682                         goto we_slept;
 683                 }
 684                 dquot = empty;
 685                 dquot->dq_id = id;
 686                 /* all dquots go on the inuse_list */
 687                 put_inuse(dquot);
 688                 /* hash it first so it can be found */
 689                 insert_dquot_hash(dquot);
 690                 dqstats.lookups++;
 691                 spin_unlock(&dq_list_lock);
 692         } else {
 693                 if (!atomic_read(&dquot->dq_count))
 694                         remove_free_dquot(dquot);
 695                 atomic_inc(&dquot->dq_count);
 696                 dqstats.cache_hits++;
 697                 dqstats.lookups++;
 698                 spin_unlock(&dq_list_lock);
 699                 if (empty)
 700                         do_destroy_dquot(empty);
 701         }
 702         /* Wait for dq_lock - after this we know that either dquot_release() is already
 703          * finished or it will be canceled due to dq_count > 1 test */
 704         wait_on_dquot(dquot);
 705         /* Read the dquot and instantiate it (everything done only if needed) */
 706         if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
 707                 dqput(dquot);
 708                 return NODQUOT;
 709         }
 710 #ifdef __DQUOT_PARANOIA
 711         BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 712 #endif
 713
 714         return dquot;
 715 }
 716
 717 static int dqinit_needed(struct inode *inode, int type)
 718 {
 719         int cnt;
 720
 721         if (IS_NOQUOTA(inode))
 722                 return 0;
 723         if (type != -1)
 724                 return inode->i_dquot[type] == NODQUOT;
 725         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 726                 if (inode->i_dquot[cnt] == NODQUOT)
 727                         return 1;
 728         return 0;
 729 }
 730
 731 /* This routine is guarded by dqonoff_mutex mutex */
 732 static void add_dquot_ref(struct super_block *sb, int type)
 733 {
 734         struct inode *inode, *old_inode = NULL;
 735
 736         spin_lock(&inode_lock);
 737         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 738                 if (!atomic_read(&inode->i_writecount))
 739                         continue;
 740                 if (!dqinit_needed(inode, type))
 741                         continue;
 742                 if (inode->i_state & (I_FREEING|I_WILL_FREE))
 743                         continue;
 744
 745                 __iget(inode);
 746                 spin_unlock(&inode_lock);
 747
 748                 iput(old_inode);
 749                 sb->dq_op->initialize(inode, type);
 750                 /* We hold a reference to 'inode' so it couldn't have been
 751                  * removed from s_inodes list while we dropped the inode_lock.
 752                  * We cannot iput the inode now as we can be holding the last
 753                  * reference and we cannot iput it under inode_lock. So we
 754                  * keep the reference and iput it later. */
 755                 old_inode = inode;
 756                 spin_lock(&inode_lock);
 757         }
 758         spin_unlock(&inode_lock);
 759         iput(old_inode);
 760 }
 761
 762 /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
 763 static inline int dqput_blocks(struct dquot *dquot)
 764 {
 765         if (atomic_read(&dquot->dq_count) <= 1)
 766                 return 1;
 767         return 0;
 768 }
 769
 770 /* Remove references to dquots from inode - add dquot to list for freeing if needed */
 771 /* We can't race with anybody because we hold dqptr_sem for writing... */
 772 static int remove_inode_dquot_ref(struct inode *inode, int type,
 773                                   struct list_head *tofree_head)
 774 {
 775         struct dquot *dquot = inode->i_dquot[type];
 776
 777         inode->i_dquot[type] = NODQUOT;
 778         if (dquot != NODQUOT) {
 779                 if (dqput_blocks(dquot)) {
 780 #ifdef __DQUOT_PARANOIA
 781                         if (atomic_read(&dquot->dq_count) != 1)
 782                                 printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
 783 #endif
 784                         spin_lock(&dq_list_lock);
 785                         list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
 786                         spin_unlock(&dq_list_lock);
 787                         return 1;
 788                 }
 789                 else
 790                         dqput(dquot);   /* We have guaranteed we won't block */
 791         }
 792         return 0;
 793 }
 794
 795 /* Free list of dquots - called from inode.c */
 796 /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
 797 static void put_dquot_list(struct list_head *tofree_head)
 798 {
 799         struct list_head *act_head;
 800         struct dquot *dquot;
 801
 802         act_head = tofree_head->next;
 803         /* So now we have dquots on the list... Just free them */
 804         while (act_head != tofree_head) {
 805                 dquot = list_entry(act_head, struct dquot, dq_free);
 806                 act_head = act_head->next;
 807                 list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
 808                 dqput(dquot);
 809         }
 810 }
 811
 812 static void remove_dquot_ref(struct super_block *sb, int type,
 813                 struct list_head *tofree_head)
 814 {
 815         struct inode *inode;
 816
 817         spin_lock(&inode_lock);
 818         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 819                 if (!IS_NOQUOTA(inode))
 820                         remove_inode_dquot_ref(inode, type, tofree_head);
 821         }
 822         spin_unlock(&inode_lock);
 823 }
 824
 825 /* Gather all references from inodes and drop them */
 826 static void drop_dquot_ref(struct super_block *sb, int type)
 827 {
 828         LIST_HEAD(tofree_head);
 829
 830         if (sb->dq_op) {
 831                 down_write(&sb_dqopt(sb)->dqptr_sem);
 832                 remove_dquot_ref(sb, type, &tofree_head);
 833                 up_write(&sb_dqopt(sb)->dqptr_sem);
 834                 put_dquot_list(&tofree_head);
 835         }
 836 }
 837
 838 static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
 839 {
 840         dquot->dq_dqb.dqb_curinodes += number;
 841 }
 842
 843 static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
 844 {
 845         dquot->dq_dqb.dqb_curspace += number;
 846 }
 847
 848 static inline void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
 849 {
 850         if (dquot->dq_dqb.dqb_curinodes > number)
 851                 dquot->dq_dqb.dqb_curinodes -= number;
 852         else
 853                 dquot->dq_dqb.dqb_curinodes = 0;
 854         if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
 855                 dquot->dq_dqb.dqb_itime = (time_t) 0;
 856         clear_bit(DQ_INODES_B, &dquot->dq_flags);
 857 }
 858
 859 static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
 860 {
 861         if (dquot->dq_dqb.dqb_curspace > number)
 862                 dquot->dq_dqb.dqb_curspace -= number;
 863         else
 864                 dquot->dq_dqb.dqb_curspace = 0;
 865         if (dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
 866                 dquot->dq_dqb.dqb_btime = (time_t) 0;
 867         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 868 }
 869
 870 static int warning_issued(struct dquot *dquot, const int warntype)
 871 {
 872         int flag = (warntype == QUOTA_NL_BHARDWARN ||
 873                 warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
 874                 ((warntype == QUOTA_NL_IHARDWARN ||
 875                 warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
 876
 877         if (!flag)
 878                 return 0;
 879         return test_and_set_bit(flag, &dquot->dq_flags);
 880 }
 881
 882 #ifdef CONFIG_PRINT_QUOTA_WARNING
 883 static int flag_print_warnings = 1;
 884
 885 static inline int need_print_warning(struct dquot *dquot)
 886 {
 887         if (!flag_print_warnings)
 888                 return 0;
 889
 890         switch (dquot->dq_type) {
 891                 case USRQUOTA:
 892                         return current_fsuid() == dquot->dq_id;
 893                 case GRPQUOTA:
 894                         return in_group_p(dquot->dq_id);
 895         }
 896         return 0;
 897 }
 898
 899 /* Print warning to user which exceeded quota */
 900 static void print_warning(struct dquot *dquot, const int warntype)
 901 {
 902         char *msg = NULL;
 903         struct tty_struct *tty;
 904
 905         if (warntype == QUOTA_NL_IHARDBELOW ||
 906             warntype == QUOTA_NL_ISOFTBELOW ||
 907             warntype == QUOTA_NL_BHARDBELOW ||
 908             warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(dquot))
 909                 return;
 910
 911         tty = get_current_tty();
 912         if (!tty)
 913                 return;
 914         tty_write_message(tty, dquot->dq_sb->s_id);
 915         if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
 916                 tty_write_message(tty, ": warning, ");
 917         else
 918                 tty_write_message(tty, ": write failed, ");
 919         tty_write_message(tty, quotatypes[dquot->dq_type]);
 920         switch (warntype) {
 921                 case QUOTA_NL_IHARDWARN:
 922                         msg = " file limit reached.\r\n";
 923                         break;
 924                 case QUOTA_NL_ISOFTLONGWARN:
 925                         msg = " file quota exceeded too long.\r\n";
 926                         break;
 927                 case QUOTA_NL_ISOFTWARN:
 928                         msg = " file quota exceeded.\r\n";
 929                         break;
 930                 case QUOTA_NL_BHARDWARN:
 931                         msg = " block limit reached.\r\n";
 932                         break;
 933                 case QUOTA_NL_BSOFTLONGWARN:
 934                         msg = " block quota exceeded too long.\r\n";
 935                         break;
 936                 case QUOTA_NL_BSOFTWARN:
 937                         msg = " block quota exceeded.\r\n";
 938                         break;
 939         }
 940         tty_write_message(tty, msg);
 941         tty_kref_put(tty);
 942 }
 943 #endif
 944
 945 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
 946
 947 /* Netlink family structure for quota */
 948 static struct genl_family quota_genl_family = {
 949         .id = GENL_ID_GENERATE,
 950         .hdrsize = 0,
 951         .name = "VFS_DQUOT",
 952         .version = 1,
 953         .maxattr = QUOTA_NL_A_MAX,
 954 };
 955
 956 /* Send warning to userspace about user which exceeded quota */
 957 static void send_warning(const struct dquot *dquot, const char warntype)
 958 {
 959         static atomic_t seq;
 960         struct sk_buff *skb;
 961         void *msg_head;
 962         int ret;
 963         int msg_size = 4 * nla_total_size(sizeof(u32)) +
 964                        2 * nla_total_size(sizeof(u64));
 965
 966         /* We have to allocate using GFP_NOFS as we are called from a
 967          * filesystem performing write and thus further recursion into
 968          * the fs to free some data could cause deadlocks. */
 969         skb = genlmsg_new(msg_size, GFP_NOFS);
 970         if (!skb) {
 971                 printk(KERN_ERR
 972                   "VFS: Not enough memory to send quota warning.\n");
 973                 return;
 974         }
 975         msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
 976                         &quota_genl_family, 0, QUOTA_NL_C_WARNING);
 977         if (!msg_head) {
 978                 printk(KERN_ERR
 979                   "VFS: Cannot store netlink header in quota warning.\n");
 980                 goto err_out;
 981         }
 982         ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
 983         if (ret)
 984                 goto attr_err_out;
 985         ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
 986         if (ret)
 987                 goto attr_err_out;
 988         ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
 989         if (ret)
 990                 goto attr_err_out;
 991         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
 992                 MAJOR(dquot->dq_sb->s_dev));
 993         if (ret)
 994                 goto attr_err_out;
 995         ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
 996                 MINOR(dquot->dq_sb->s_dev));
 997         if (ret)
 998                 goto attr_err_out;
 999         ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current_uid());
1000         if (ret)
1001                 goto attr_err_out;
1002         genlmsg_end(skb, msg_head);
1003
1004         ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
1005         if (ret < 0 && ret != -ESRCH)
1006                 printk(KERN_ERR
1007                         "VFS: Failed to send notification message: %d\n", ret);
1008         return;
1009 attr_err_out:
1010         printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
1011 err_out:
1012         kfree_skb(skb);
1013 }
1014 #endif
1015
1016 static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
1017 {
1018         int i;
1019
1020         for (i = 0; i < MAXQUOTAS; i++)
1021                 if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
1022                     !warning_issued(dquots[i], warntype[i])) {
1023 #ifdef CONFIG_PRINT_QUOTA_WARNING
1024                         print_warning(dquots[i], warntype[i]);
1025 #endif
1026 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
1027                         send_warning(dquots[i], warntype[i]);
1028 #endif
1029                 }
1030 }
1031
1032 static inline char ignore_hardlimit(struct dquot *dquot)
1033 {
1034         struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1035
1036         return capable(CAP_SYS_RESOURCE) &&
1037             (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
1038 }
1039
1040 /* needs dq_data_lock */
1041 static int check_idq(struct dquot *dquot, qsize_t inodes, char *warntype)
1042 {
1043         *warntype = QUOTA_NL_NOWARN;
1044         if (test_bit(DQ_FAKE_B, &dquot->dq_flags))
1045                 return QUOTA_OK;
1046
1047         if (dquot->dq_dqb.dqb_ihardlimit &&
1048            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
1049             !ignore_hardlimit(dquot)) {
1050                 *warntype = QUOTA_NL_IHARDWARN;
1051                 return NO_QUOTA;
1052         }
1053
1054         if (dquot->dq_dqb.dqb_isoftlimit &&
1055            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1056             dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
1057             !ignore_hardlimit(dquot)) {
1058                 *warntype = QUOTA_NL_ISOFTLONGWARN;
1059                 return NO_QUOTA;
1060         }
1061
1062         if (dquot->dq_dqb.dqb_isoftlimit &&
1063            (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
1064             dquot->dq_dqb.dqb_itime == 0) {
1065                 *warntype = QUOTA_NL_ISOFTWARN;
1066                 dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
1067         }
1068
1069         return QUOTA_OK;
1070 }
1071
1072 /* needs dq_data_lock */
1073 static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
1074 {
1075         *warntype = QUOTA_NL_NOWARN;
1076         if (test_bit(DQ_FAKE_B, &dquot->dq_flags))
1077                 return QUOTA_OK;
1078
1079         if (dquot->dq_dqb.dqb_bhardlimit &&
1080             dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bhardlimit &&
1081             !ignore_hardlimit(dquot)) {
1082                 if (!prealloc)
1083                         *warntype = QUOTA_NL_BHARDWARN;
1084                 return NO_QUOTA;
1085         }
1086
1087         if (dquot->dq_dqb.dqb_bsoftlimit &&
1088             dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1089             dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
1090             !ignore_hardlimit(dquot)) {
1091                 if (!prealloc)
1092                         *warntype = QUOTA_NL_BSOFTLONGWARN;
1093                 return NO_QUOTA;
1094         }
1095
1096         if (dquot->dq_dqb.dqb_bsoftlimit &&
1097             dquot->dq_dqb.dqb_curspace + space > dquot->dq_dqb.dqb_bsoftlimit &&
1098             dquot->dq_dqb.dqb_btime == 0) {
1099                 if (!prealloc) {
1100                         *warntype = QUOTA_NL_BSOFTWARN;
1101                         dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
1102                 }
1103                 else
1104                         /*
1105                          * We don't allow preallocation to exceed softlimit so exceeding will
1106                          * be always printed
1107                          */
1108                         return NO_QUOTA;
1109         }
1110
1111         return QUOTA_OK;
1112 }
1113
1114 static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1115 {
1116         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1117             dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1118                 return QUOTA_NL_NOWARN;
1119
1120         if (dquot->dq_dqb.dqb_curinodes - inodes <= dquot->dq_dqb.dqb_isoftlimit)
1121                 return QUOTA_NL_ISOFTBELOW;
1122         if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1123             dquot->dq_dqb.dqb_curinodes - inodes < dquot->dq_dqb.dqb_ihardlimit)
1124                 return QUOTA_NL_IHARDBELOW;
1125         return QUOTA_NL_NOWARN;
1126 }
1127
1128 static int info_bdq_free(struct dquot *dquot, qsize_t space)
1129 {
1130         if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1131             dquot->dq_dqb.dqb_curspace <= dquot->dq_dqb.dqb_bsoftlimit)
1132                 return QUOTA_NL_NOWARN;
1133
1134         if (dquot->dq_dqb.dqb_curspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1135                 return QUOTA_NL_BSOFTBELOW;
1136         if (dquot->dq_dqb.dqb_curspace >= dquot->dq_dqb.dqb_bhardlimit &&
1137             dquot->dq_dqb.dqb_curspace - space < dquot->dq_dqb.dqb_bhardlimit)
1138                 return QUOTA_NL_BHARDBELOW;
1139         return QUOTA_NL_NOWARN;
1140 }
1141 /*
1142  *      Initialize quota pointers in inode
1143  *      Transaction must be started at entry
1144  */
1145 int dquot_initialize(struct inode *inode, int type)
1146 {
1147         unsigned int id = 0;
1148         int cnt, ret = 0;
1149
1150         /* First test before acquiring mutex - solves deadlocks when we
1151          * re-enter the quota code and are already holding the mutex */
1152         if (IS_NOQUOTA(inode))
1153                 return 0;
1154         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1155         /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
1156         if (IS_NOQUOTA(inode))
1157                 goto out_err;
1158         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1159                 if (type != -1 && cnt != type)
1160                         continue;
1161                 if (inode->i_dquot[cnt] == NODQUOT) {
1162                         switch (cnt) {
1163                                 case USRQUOTA:
1164                                         id = inode->i_uid;
1165                                         break;
1166                                 case GRPQUOTA:
1167                                         id = inode->i_gid;
1168                                         break;
1169                         }
1170                         inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
1171                 }
1172         }
1173 out_err:
1174         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1175         return ret;
1176 }
1177
1178 /*
1179  *      Release all quotas referenced by inode
1180  *      Transaction must be started at an entry
1181  */
1182 int dquot_drop(struct inode *inode)
1183 {
1184         int cnt;
1185
1186         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1187         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1188                 if (inode->i_dquot[cnt] != NODQUOT) {
1189                         dqput(inode->i_dquot[cnt]);
1190                         inode->i_dquot[cnt] = NODQUOT;
1191                 }
1192         }
1193         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1194         return 0;
1195 }
1196
1197 /* Wrapper to remove references to quota structures from inode */
1198 void vfs_dq_drop(struct inode *inode)
1199 {
1200         /* Here we can get arbitrary inode from clear_inode() so we have
1201          * to be careful. OTOH we don't need locking as quota operations
1202          * are allowed to change only at mount time */
1203         if (!IS_NOQUOTA(inode) && inode->i_sb && inode->i_sb->dq_op
1204             && inode->i_sb->dq_op->drop) {
1205                 int cnt;
1206                 /* Test before calling to rule out calls from proc and such
1207                  * where we are not allowed to block. Note that this is
1208                  * actually reliable test even without the lock - the caller
1209                  * must assure that nobody can come after the DQUOT_DROP and
1210                  * add quota pointers back anyway */
1211                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1212                         if (inode->i_dquot[cnt] != NODQUOT)
1213                                 break;
1214                 if (cnt < MAXQUOTAS)
1215                         inode->i_sb->dq_op->drop(inode);
1216         }
1217 }
1218
1219 /*
1220  * Following four functions update i_blocks+i_bytes fields and
1221  * quota information (together with appropriate checks)
1222  * NOTE: We absolutely rely on the fact that caller dirties
1223  * the inode (usually macros in quotaops.h care about this) and
1224  * holds a handle for the current transaction so that dquot write and
1225  * inode write go into the same transaction.
1226  */
1227
1228 /*
1229  * This operation can block, but only after everything is updated
1230  */
1231 int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
1232 {
1233         int cnt, ret = NO_QUOTA;
1234         char warntype[MAXQUOTAS];
1235
1236         /* First test before acquiring mutex - solves deadlocks when we
1237          * re-enter the quota code and are already holding the mutex */
1238         if (IS_NOQUOTA(inode)) {
1239 out_add:
1240                 inode_add_bytes(inode, number);
1241                 return QUOTA_OK;
1242         }
1243         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1244                 warntype[cnt] = QUOTA_NL_NOWARN;
1245
1246         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1247         if (IS_NOQUOTA(inode)) {        /* Now we can do reliable test... */
1248                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1249                 goto out_add;
1250         }
1251         spin_lock(&dq_data_lock);
1252         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1253                 if (inode->i_dquot[cnt] == NODQUOT)
1254                         continue;
1255                 if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
1256                         goto warn_put_all;
1257         }
1258         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1259                 if (inode->i_dquot[cnt] == NODQUOT)
1260                         continue;
1261                 dquot_incr_space(inode->i_dquot[cnt], number);
1262         }
1263         inode_add_bytes(inode, number);
1264         ret = QUOTA_OK;
1265 warn_put_all:
1266         spin_unlock(&dq_data_lock);
1267         if (ret == QUOTA_OK)
1268                 /* Dirtify all the dquots - this can block when journalling */
1269                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1270                         if (inode->i_dquot[cnt])
1271                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1272         flush_warnings(inode->i_dquot, warntype);
1273         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1274         return ret;
1275 }
1276
1277 /*
1278  * This operation can block, but only after everything is updated
1279  */
1280 int dquot_alloc_inode(const struct inode *inode, qsize_t number)
1281 {
1282         int cnt, ret = NO_QUOTA;
1283         char warntype[MAXQUOTAS];
1284
1285         /* First test before acquiring mutex - solves deadlocks when we
1286          * re-enter the quota code and are already holding the mutex */
1287         if (IS_NOQUOTA(inode))
1288                 return QUOTA_OK;
1289         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1290                 warntype[cnt] = QUOTA_NL_NOWARN;
1291         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1292         if (IS_NOQUOTA(inode)) {
1293                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1294                 return QUOTA_OK;
1295         }
1296         spin_lock(&dq_data_lock);
1297         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1298                 if (inode->i_dquot[cnt] == NODQUOT)
1299                         continue;
1300                 if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
1301                         goto warn_put_all;
1302         }
1303
1304         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1305                 if (inode->i_dquot[cnt] == NODQUOT)
1306                         continue;
1307                 dquot_incr_inodes(inode->i_dquot[cnt], number);
1308         }
1309         ret = QUOTA_OK;
1310 warn_put_all:
1311         spin_unlock(&dq_data_lock);
1312         if (ret == QUOTA_OK)
1313                 /* Dirtify all the dquots - this can block when journalling */
1314                 for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1315                         if (inode->i_dquot[cnt])
1316                                 mark_dquot_dirty(inode->i_dquot[cnt]);
1317         flush_warnings(inode->i_dquot, warntype);
1318         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1319         return ret;
1320 }
1321
1322 /*
1323  * This operation can block, but only after everything is updated
1324  */
1325 int dquot_free_space(struct inode *inode, qsize_t number)
1326 {
1327         unsigned int cnt;
1328         char warntype[MAXQUOTAS];
1329
1330         /* First test before acquiring mutex - solves deadlocks when we
1331          * re-enter the quota code and are already holding the mutex */
1332         if (IS_NOQUOTA(inode)) {
1333 out_sub:
1334                 inode_sub_bytes(inode, number);
1335                 return QUOTA_OK;
1336         }
1337
1338         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1339         /* Now recheck reliably when holding dqptr_sem */
1340         if (IS_NOQUOTA(inode)) {
1341                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1342                 goto out_sub;
1343         }
1344         spin_lock(&dq_data_lock);
1345         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1346                 if (inode->i_dquot[cnt] == NODQUOT)
1347                         continue;
1348                 warntype[cnt] = info_bdq_free(inode->i_dquot[cnt], number);
1349                 dquot_decr_space(inode->i_dquot[cnt], number);
1350         }
1351         inode_sub_bytes(inode, number);
1352         spin_unlock(&dq_data_lock);
1353         /* Dirtify all the dquots - this can block when journalling */
1354         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1355                 if (inode->i_dquot[cnt])
1356                         mark_dquot_dirty(inode->i_dquot[cnt]);
1357         flush_warnings(inode->i_dquot, warntype);
1358         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1359         return QUOTA_OK;
1360 }
1361
1362 /*
1363  * This operation can block, but only after everything is updated
1364  */
1365 int dquot_free_inode(const struct inode *inode, qsize_t number)
1366 {
1367         unsigned int cnt;
1368         char warntype[MAXQUOTAS];
1369
1370         /* First test before acquiring mutex - solves deadlocks when we
1371          * re-enter the quota code and are already holding the mutex */
1372         if (IS_NOQUOTA(inode))
1373                 return QUOTA_OK;
1374
1375         down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1376         /* Now recheck reliably when holding dqptr_sem */
1377         if (IS_NOQUOTA(inode)) {
1378                 up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1379                 return QUOTA_OK;
1380         }
1381         spin_lock(&dq_data_lock);
1382         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1383                 if (inode->i_dquot[cnt] == NODQUOT)
1384                         continue;
1385                 warntype[cnt] = info_idq_free(inode->i_dquot[cnt], number);
1386                 dquot_decr_inodes(inode->i_dquot[cnt], number);
1387         }
1388         spin_unlock(&dq_data_lock);
1389         /* Dirtify all the dquots - this can block when journalling */
1390         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1391                 if (inode->i_dquot[cnt])
1392                         mark_dquot_dirty(inode->i_dquot[cnt]);
1393         flush_warnings(inode->i_dquot, warntype);
1394         up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
1395         return QUOTA_OK;
1396 }
1397
1398 /*
1399  * Transfer the number of inode and blocks from one diskquota to an other.
1400  *
1401  * This operation can block, but only after everything is updated
1402  * A transaction must be started when entering this function.
1403  */
1404 int dquot_transfer(struct inode *inode, struct iattr *iattr)
1405 {
1406         qsize_t space;
1407         struct dquot *transfer_from[MAXQUOTAS];
1408         struct dquot *transfer_to[MAXQUOTAS];
1409         int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
1410             chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
1411         char warntype_to[MAXQUOTAS];
1412         char warntype_from_inodes[MAXQUOTAS], warntype_from_space[MAXQUOTAS];
1413
1414         /* First test before acquiring mutex - solves deadlocks when we
1415          * re-enter the quota code and are already holding the mutex */
1416         if (IS_NOQUOTA(inode))
1417                 return QUOTA_OK;
1418         /* Clear the arrays */
1419         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1420                 transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
1421                 warntype_to[cnt] = QUOTA_NL_NOWARN;
1422         }
1423         down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1424         /* Now recheck reliably when holding dqptr_sem */
1425         if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1426                 up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1427                 return QUOTA_OK;
1428         }
1429         /* First build the transfer_to list - here we can block on
1430          * reading/instantiating of dquots.  We know that the transaction for
1431          * us was already started so we don't violate lock ranking here */
1432         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1433                 switch (cnt) {
1434                         case USRQUOTA:
1435                                 if (!chuid)
1436                                         continue;
1437                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
1438                                 break;
1439                         case GRPQUOTA:
1440                                 if (!chgid)
1441                                         continue;
1442                                 transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
1443                                 break;
1444                 }
1445         }
1446         spin_lock(&dq_data_lock);
1447         space = inode_get_bytes(inode);
1448         /* Build the transfer_from list and check the limits */
1449         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1450                 if (transfer_to[cnt] == NODQUOT)
1451                         continue;
1452                 transfer_from[cnt] = inode->i_dquot[cnt];
1453                 if (check_idq(transfer_to[cnt], 1, warntype_to + cnt) ==
1454                     NO_QUOTA || check_bdq(transfer_to[cnt], space, 0,
1455                     warntype_to + cnt) == NO_QUOTA)
1456                         goto warn_put_all;
1457         }
1458
1459         /*
1460          * Finally perform the needed transfer from transfer_from to transfer_to
1461          */
1462         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1463                 /*
1464                  * Skip changes for same uid or gid or for turned off quota-type.
1465                  */
1466                 if (transfer_to[cnt] == NODQUOT)
1467                         continue;
1468
1469                 /* Due to IO error we might not have transfer_from[] structure */
1470                 if (transfer_from[cnt]) {
1471                         warntype_from_inodes[cnt] =
1472                                 info_idq_free(transfer_from[cnt], 1);
1473                         warntype_from_space[cnt] =
1474                                 info_bdq_free(transfer_from[cnt], space);
1475                         dquot_decr_inodes(transfer_from[cnt], 1);
1476                         dquot_decr_space(transfer_from[cnt], space);
1477                 }
1478
1479                 dquot_incr_inodes(transfer_to[cnt], 1);
1480                 dquot_incr_space(transfer_to[cnt], space);
1481
1482                 inode->i_dquot[cnt] = transfer_to[cnt];
1483         }
1484         ret = QUOTA_OK;
1485 warn_put_all:
1486         spin_unlock(&dq_data_lock);
1487         /* Dirtify all the dquots - this can block when journalling */
1488         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1489                 if (transfer_from[cnt])
1490                         mark_dquot_dirty(transfer_from[cnt]);
1491                 if (transfer_to[cnt])
1492                         mark_dquot_dirty(transfer_to[cnt]);
1493         }
1494         flush_warnings(transfer_to, warntype_to);
1495         flush_warnings(transfer_from, warntype_from_inodes);
1496         flush_warnings(transfer_from, warntype_from_space);
1497
1498         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1499                 if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
1500                         dqput(transfer_from[cnt]);
1501                 if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
1502                         dqput(transfer_to[cnt]);
1503         }
1504         up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
1505         return ret;
1506 }
1507
1508 /* Wrapper for transferring ownership of an inode */
1509 int vfs_dq_transfer(struct inode *inode, struct iattr *iattr)
1510 {
1511         if (sb_any_quota_enabled(inode->i_sb) && !IS_NOQUOTA(inode)) {
1512                 vfs_dq_init(inode);
1513                 if (inode->i_sb->dq_op->transfer(inode, iattr) == NO_QUOTA)
1514                         return 1;
1515         }
1516         return 0;
1517 }
1518
1519
1520 /*
1521  * Write info of quota file to disk
1522  */
1523 int dquot_commit_info(struct super_block *sb, int type)
1524 {
1525         int ret;
1526         struct quota_info *dqopt = sb_dqopt(sb);
1527
1528         mutex_lock(&dqopt->dqio_mutex);
1529         ret = dqopt->ops[type]->write_file_info(sb, type);
1530         mutex_unlock(&dqopt->dqio_mutex);
1531         return ret;
1532 }
1533
1534 /*
1535  * Definitions of diskquota operations.
1536  */
1537 struct dquot_operations dquot_operations = {
1538         .initialize     = dquot_initialize,
1539         .drop           = dquot_drop,
1540         .alloc_space    = dquot_alloc_space,
1541         .alloc_inode    = dquot_alloc_inode,
1542         .free_space     = dquot_free_space,
1543         .free_inode     = dquot_free_inode,
1544         .transfer       = dquot_transfer,
1545         .write_dquot    = dquot_commit,
1546         .acquire_dquot  = dquot_acquire,
1547         .release_dquot  = dquot_release,
1548         .mark_dirty     = dquot_mark_dquot_dirty,
1549         .write_info     = dquot_commit_info,
1550         .alloc_dquot    = dquot_alloc,
1551         .destroy_dquot  = dquot_destroy,
1552 };
1553
1554 static inline void set_enable_flags(struct quota_info *dqopt, int type)
1555 {
1556         switch (type) {
1557                 case USRQUOTA:
1558                         dqopt->flags |= DQUOT_USR_ENABLED;
1559                         dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1560                         break;
1561                 case GRPQUOTA:
1562                         dqopt->flags |= DQUOT_GRP_ENABLED;
1563                         dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1564                         break;
1565         }
1566 }
1567
1568 static inline void reset_enable_flags(struct quota_info *dqopt, int type,
1569                                       int remount)
1570 {
1571         switch (type) {
1572                 case USRQUOTA:
1573                         dqopt->flags &= ~DQUOT_USR_ENABLED;
1574                         if (remount)
1575                                 dqopt->flags |= DQUOT_USR_SUSPENDED;
1576                         else
1577                                 dqopt->flags &= ~DQUOT_USR_SUSPENDED;
1578                         break;
1579                 case GRPQUOTA:
1580                         dqopt->flags &= ~DQUOT_GRP_ENABLED;
1581                         if (remount)
1582                                 dqopt->flags |= DQUOT_GRP_SUSPENDED;
1583                         else
1584                                 dqopt->flags &= ~DQUOT_GRP_SUSPENDED;
1585                         break;
1586         }
1587 }
1588
1589
1590 /*
1591  * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
1592  */
1593 int vfs_quota_off(struct super_block *sb, int type, int remount)
1594 {
1595         int cnt, ret = 0;
1596         struct quota_info *dqopt = sb_dqopt(sb);
1597         struct inode *toputinode[MAXQUOTAS];
1598
1599         /* We need to serialize quota_off() for device */
1600         mutex_lock(&dqopt->dqonoff_mutex);
1601
1602         /*
1603          * Skip everything if there's nothing to do. We have to do this because
1604          * sometimes we are called when fill_super() failed and calling
1605          * sync_fs() in such cases does no good.
1606          */
1607         if (!sb_any_quota_enabled(sb) && !sb_any_quota_suspended(sb)) {
1608                 mutex_unlock(&dqopt->dqonoff_mutex);
1609                 return 0;
1610         }
1611         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1612                 toputinode[cnt] = NULL;
1613                 if (type != -1 && cnt != type)
1614                         continue;
1615                 /* If we keep inodes of quota files after remount and quotaoff
1616                  * is called, drop kept inodes. */
1617                 if (!remount && sb_has_quota_suspended(sb, cnt)) {
1618                         iput(dqopt->files[cnt]);
1619                         dqopt->files[cnt] = NULL;
1620                         reset_enable_flags(dqopt, cnt, 0);
1621                         continue;
1622                 }
1623                 if (!sb_has_quota_enabled(sb, cnt))
1624                         continue;
1625                 reset_enable_flags(dqopt, cnt, remount);
1626
1627                 /* Note: these are blocking operations */
1628                 drop_dquot_ref(sb, cnt);
1629                 invalidate_dquots(sb, cnt);
1630                 /*
1631                  * Now all dquots should be invalidated, all writes done so we should be only
1632                  * users of the info. No locks needed.
1633                  */
1634                 if (info_dirty(&dqopt->info[cnt]))
1635                         sb->dq_op->write_info(sb, cnt);
1636                 if (dqopt->ops[cnt]->free_file_info)
1637                         dqopt->ops[cnt]->free_file_info(sb, cnt);
1638                 put_quota_format(dqopt->info[cnt].dqi_format);
1639
1640                 toputinode[cnt] = dqopt->files[cnt];
1641                 if (!remount)
1642                         dqopt->files[cnt] = NULL;
1643                 dqopt->info[cnt].dqi_flags = 0;
1644                 dqopt->info[cnt].dqi_igrace = 0;
1645                 dqopt->info[cnt].dqi_bgrace = 0;
1646                 dqopt->ops[cnt] = NULL;
1647         }
1648         mutex_unlock(&dqopt->dqonoff_mutex);
1649         /* Sync the superblock so that buffers with quota data are written to
1650          * disk (and so userspace sees correct data afterwards). */
1651         if (sb->s_op->sync_fs)
1652                 sb->s_op->sync_fs(sb, 1);
1653         sync_blockdev(sb->s_bdev);
1654         /* Now the quota files are just ordinary files and we can set the
1655          * inode flags back. Moreover we discard the pagecache so that
1656          * userspace sees the writes we did bypassing the pagecache. We
1657          * must also discard the blockdev buffers so that we see the
1658          * changes done by userspace on the next quotaon() */
1659         for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1660                 if (toputinode[cnt]) {
1661                         mutex_lock(&dqopt->dqonoff_mutex);
1662                         /* If quota was reenabled in the meantime, we have
1663                          * nothing to do */
1664                         if (!sb_has_quota_enabled(sb, cnt)) {
1665                                 mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
1666                                 toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
1667                                   S_NOATIME | S_NOQUOTA);
1668                                 truncate_inode_pages(&toputinode[cnt]->i_data, 0);
1669                                 mutex_unlock(&toputinode[cnt]->i_mutex);
1670                                 mark_inode_dirty(toputinode[cnt]);
1671                         }
1672                         mutex_unlock(&dqopt->dqonoff_mutex);
1673                         /* On remount RO, we keep the inode pointer so that we
1674                          * can reenable quota on the subsequent remount RW.
1675                          * But we have better not keep inode pointer when there
1676                          * is pending delete on the quota file... */
1677                         if (!remount)
1678                                 iput(toputinode[cnt]);
1679                         else if (!toputinode[cnt]->i_nlink)
1680                                 ret = -EBUSY;
1681                 }
1682         if (sb->s_bdev)
1683                 invalidate_bdev(sb->s_bdev);
1684         return ret;
1685 }
1686
1687 /*
1688  *      Turn quotas on on a device
1689  */
1690
1691 /* Helper function when we already have the inode */
1692 static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
1693 {
1694         struct quota_format_type *fmt = find_quota_format(format_id);
1695         struct super_block *sb = inode->i_sb;
1696         struct quota_info *dqopt = sb_dqopt(sb);
1697         int error;
1698         int oldflags = -1;
1699
1700         if (!fmt)
1701                 return -ESRCH;
1702         if (!S_ISREG(inode->i_mode)) {
1703                 error = -EACCES;
1704                 goto out_fmt;
1705         }
1706         if (IS_RDONLY(inode)) {
1707                 error = -EROFS;
1708                 goto out_fmt;
1709         }
1710         if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
1711                 error = -EINVAL;
1712                 goto out_fmt;
1713         }
1714
1715         /* As we bypass the pagecache we must now flush the inode so that
1716          * we see all the changes from userspace... */
1717         write_inode_now(inode, 1);
1718         /* And now flush the block cache so that kernel sees the changes */
1719         invalidate_bdev(sb->s_bdev);
1720         mutex_lock(&inode->i_mutex);
1721         mutex_lock(&dqopt->dqonoff_mutex);
1722         if (sb_has_quota_enabled(sb, type) ||
1723                         sb_has_quota_suspended(sb, type)) {
1724                 error = -EBUSY;
1725                 goto out_lock;
1726         }
1727         /* We don't want quota and atime on quota files (deadlocks possible)
1728          * Also nobody should write to the file - we use special IO operations
1729          * which ignore the immutable bit. */
1730         down_write(&dqopt->dqptr_sem);
1731         oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
1732         inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
1733         up_write(&dqopt->dqptr_sem);
1734         sb->dq_op->drop(inode);
1735
1736         error = -EIO;
1737         dqopt->files[type] = igrab(inode);
1738         if (!dqopt->files[type])
1739                 goto out_lock;
1740         error = -EINVAL;
1741         if (!fmt->qf_ops->check_quota_file(sb, type))
1742                 goto out_file_init;
1743
1744         dqopt->ops[type] = fmt->qf_ops;
1745         dqopt->info[type].dqi_format = fmt;
1746         dqopt->info[type].dqi_fmt_id = format_id;
1747         INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
1748         mutex_lock(&dqopt->dqio_mutex);
1749         if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
1750                 mutex_unlock(&dqopt->dqio_mutex);
1751                 goto out_file_init;
1752         }
1753         mutex_unlock(&dqopt->dqio_mutex);
1754         mutex_unlock(&inode->i_mutex);
1755         set_enable_flags(dqopt, type);
1756
1757         add_dquot_ref(sb, type);
1758         mutex_unlock(&dqopt->dqonoff_mutex);
1759
1760         return 0;
1761
1762 out_file_init:
1763         dqopt->files[type] = NULL;
1764         iput(inode);
1765 out_lock:
1766         mutex_unlock(&dqopt->dqonoff_mutex);
1767         if (oldflags != -1) {
1768                 down_write(&dqopt->dqptr_sem);
1769                 /* Set the flags back (in the case of accidental quotaon()
1770                  * on a wrong file we don't want to mess up the flags) */
1771                 inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
1772                 inode->i_flags |= oldflags;
1773                 up_write(&dqopt->dqptr_sem);
1774         }
1775         mutex_unlock(&inode->i_mutex);
1776 out_fmt:
1777         put_quota_format(fmt);
1778
1779         return error;
1780 }
1781
1782 /* Reenable quotas on remount RW */
1783 static int vfs_quota_on_remount(struct super_block *sb, int type)
1784 {
1785         struct quota_info *dqopt = sb_dqopt(sb);
1786         struct inode *inode;
1787         int ret;
1788
1789         mutex_lock(&dqopt->dqonoff_mutex);
1790         if (!sb_has_quota_suspended(sb, type)) {
1791                 mutex_unlock(&dqopt->dqonoff_mutex);
1792                 return 0;
1793         }
1794         BUG_ON(sb_has_quota_enabled(sb, type));
1795
1796         inode = dqopt->files[type];
1797         dqopt->files[type] = NULL;
1798         reset_enable_flags(dqopt, type, 0);
1799         mutex_unlock(&dqopt->dqonoff_mutex);
1800
1801         ret = vfs_quota_on_inode(inode, type, dqopt->info[type].dqi_fmt_id);
1802         iput(inode);
1803
1804         return ret;
1805 }
1806
1807 int vfs_quota_on_path(struct super_block *sb, int type, int format_id,
1808                       struct path *path)
1809 {
1810         int error = security_quota_on(path->dentry);
1811         if (error)
1812                 return error;
1813         /* Quota file not on the same filesystem? */
1814         if (path->mnt->mnt_sb != sb)
1815                 error = -EXDEV;
1816         else
1817                 error = vfs_quota_on_inode(path->dentry->d_inode, type,
1818                                            format_id);
1819         return error;
1820 }
1821
1822 /* Actual function called from quotactl() */
1823 int vfs_quota_on(struct super_block *sb, int type, int format_id, char *name,
1824                  int remount)
1825 {
1826         struct path path;
1827         int error;
1828
1829         if (remount)
1830                 return vfs_quota_on_remount(sb, type);
1831
1832         error = kern_path(name, LOOKUP_FOLLOW, &path);
1833         if (!error) {
1834                 error = vfs_quota_on_path(sb, type, format_id, &path);
1835                 path_put(&path);
1836         }
1837         return error;
1838 }
1839
1840 /*
1841  * This function is used when filesystem needs to initialize quotas
1842  * during mount time.
1843  */
1844 int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
1845                 int format_id, int type)
1846 {
1847         struct dentry *dentry;
1848         int error;
1849
1850         dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
1851         if (IS_ERR(dentry))
1852                 return PTR_ERR(dentry);
1853
1854         if (!dentry->d_inode) {
1855                 error = -ENOENT;
1856                 goto out;
1857         }
1858
1859         error = security_quota_on(dentry);
1860         if (!error)
1861                 error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
1862
1863 out:
1864         dput(dentry);
1865         return error;
1866 }
1867
1868 /* Wrapper to turn on quotas when remounting rw */
1869 int vfs_dq_quota_on_remount(struct super_block *sb)
1870 {
1871         int cnt;
1872         int ret = 0, err;
1873
1874         if (!sb->s_qcop || !sb->s_qcop->quota_on)
1875                 return -ENOSYS;
1876         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1877                 err = sb->s_qcop->quota_on(sb, cnt, 0, NULL, 1);
1878                 if (err < 0 && !ret)
1879                         ret = err;
1880         }
1881         return ret;
1882 }
1883
1884 static inline qsize_t qbtos(qsize_t blocks)
1885 {
1886         return blocks << QIF_DQBLKSIZE_BITS;
1887 }
1888
1889 static inline qsize_t stoqb(qsize_t space)
1890 {
1891         return (space + QIF_DQBLKSIZE - 1) >> QIF_DQBLKSIZE_BITS;
1892 }
1893
1894 /* Generic routine for getting common part of quota structure */
1895 static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
1896 {
1897         struct mem_dqblk *dm = &dquot->dq_dqb;
1898
1899         spin_lock(&dq_data_lock);
1900         di->dqb_bhardlimit = stoqb(dm->dqb_bhardlimit);
1901         di->dqb_bsoftlimit = stoqb(dm->dqb_bsoftlimit);
1902         di->dqb_curspace = dm->dqb_curspace;
1903         di->dqb_ihardlimit = dm->dqb_ihardlimit;
1904         di->dqb_isoftlimit = dm->dqb_isoftlimit;
1905         di->dqb_curinodes = dm->dqb_curinodes;
1906         di->dqb_btime = dm->dqb_btime;
1907         di->dqb_itime = dm->dqb_itime;
1908         di->dqb_valid = QIF_ALL;
1909         spin_unlock(&dq_data_lock);
1910 }
1911
1912 int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1913 {
1914         struct dquot *dquot;
1915
1916         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1917         if (!(dquot = dqget(sb, id, type))) {
1918                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1919                 return -ESRCH;
1920         }
1921         do_get_dqblk(dquot, di);
1922         dqput(dquot);
1923         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
1924         return 0;
1925 }
1926
1927 /* Generic routine for setting common part of quota structure */
1928 static int do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
1929 {
1930         struct mem_dqblk *dm = &dquot->dq_dqb;
1931         int check_blim = 0, check_ilim = 0;
1932         struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
1933
1934         if ((di->dqb_valid & QIF_BLIMITS &&
1935              (di->dqb_bhardlimit > dqi->dqi_maxblimit ||
1936               di->dqb_bsoftlimit > dqi->dqi_maxblimit)) ||
1937             (di->dqb_valid & QIF_ILIMITS &&
1938              (di->dqb_ihardlimit > dqi->dqi_maxilimit ||
1939               di->dqb_isoftlimit > dqi->dqi_maxilimit)))
1940                 return -ERANGE;
1941
1942         spin_lock(&dq_data_lock);
1943         if (di->dqb_valid & QIF_SPACE) {
1944                 dm->dqb_curspace = di->dqb_curspace;
1945                 check_blim = 1;
1946         }
1947         if (di->dqb_valid & QIF_BLIMITS) {
1948                 dm->dqb_bsoftlimit = qbtos(di->dqb_bsoftlimit);
1949                 dm->dqb_bhardlimit = qbtos(di->dqb_bhardlimit);
1950                 check_blim = 1;
1951         }
1952         if (di->dqb_valid & QIF_INODES) {
1953                 dm->dqb_curinodes = di->dqb_curinodes;
1954                 check_ilim = 1;
1955         }
1956         if (di->dqb_valid & QIF_ILIMITS) {
1957                 dm->dqb_isoftlimit = di->dqb_isoftlimit;
1958                 dm->dqb_ihardlimit = di->dqb_ihardlimit;
1959                 check_ilim = 1;
1960         }
1961         if (di->dqb_valid & QIF_BTIME)
1962                 dm->dqb_btime = di->dqb_btime;
1963         if (di->dqb_valid & QIF_ITIME)
1964                 dm->dqb_itime = di->dqb_itime;
1965
1966         if (check_blim) {
1967                 if (!dm->dqb_bsoftlimit || dm->dqb_curspace < dm->dqb_bsoftlimit) {
1968                         dm->dqb_btime = 0;
1969                         clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1970                 }
1971                 else if (!(di->dqb_valid & QIF_BTIME))  /* Set grace only if user hasn't provided his own... */
1972                         dm->dqb_btime = get_seconds() + dqi->dqi_bgrace;
1973         }
1974         if (check_ilim) {
1975                 if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
1976                         dm->dqb_itime = 0;
1977                         clear_bit(DQ_INODES_B, &dquot->dq_flags);
1978                 }
1979                 else if (!(di->dqb_valid & QIF_ITIME))  /* Set grace only if user hasn't provided his own... */
1980                         dm->dqb_itime = get_seconds() + dqi->dqi_igrace;
1981         }
1982         if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
1983                 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1984         else
1985                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1986         spin_unlock(&dq_data_lock);
1987         mark_dquot_dirty(dquot);
1988
1989         return 0;
1990 }
1991
1992 int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
1993 {
1994         struct dquot *dquot;
1995         int rc;
1996
1997         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
1998         if (!(dquot = dqget(sb, id, type))) {
1999                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2000                 return -ESRCH;
2001         }
2002         rc = do_set_dqblk(dquot, di);
2003         dqput(dquot);
2004         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2005         return rc;
2006 }
2007
2008 /* Generic routine for getting common part of quota file information */
2009 int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2010 {
2011         struct mem_dqinfo *mi;
2012
2013         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2014         if (!sb_has_quota_enabled(sb, type)) {
2015                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2016                 return -ESRCH;
2017         }
2018         mi = sb_dqopt(sb)->info + type;
2019         spin_lock(&dq_data_lock);
2020         ii->dqi_bgrace = mi->dqi_bgrace;
2021         ii->dqi_igrace = mi->dqi_igrace;
2022         ii->dqi_flags = mi->dqi_flags & DQF_MASK;
2023         ii->dqi_valid = IIF_ALL;
2024         spin_unlock(&dq_data_lock);
2025         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2026         return 0;
2027 }
2028
2029 /* Generic routine for setting common part of quota file information */
2030 int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
2031 {
2032         struct mem_dqinfo *mi;
2033
2034         mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
2035         if (!sb_has_quota_enabled(sb, type)) {
2036                 mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2037                 return -ESRCH;
2038         }
2039         mi = sb_dqopt(sb)->info + type;
2040         spin_lock(&dq_data_lock);
2041         if (ii->dqi_valid & IIF_BGRACE)
2042                 mi->dqi_bgrace = ii->dqi_bgrace;
2043         if (ii->dqi_valid & IIF_IGRACE)
2044                 mi->dqi_igrace = ii->dqi_igrace;
2045         if (ii->dqi_valid & IIF_FLAGS)
2046                 mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
2047         spin_unlock(&dq_data_lock);
2048         mark_info_dirty(sb, type);
2049         /* Force write to disk */
2050         sb->dq_op->write_info(sb, type);
2051         mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
2052         return 0;
2053 }
2054
2055 struct quotactl_ops vfs_quotactl_ops = {
2056         .quota_on       = vfs_quota_on,
2057         .quota_off      = vfs_quota_off,
2058         .quota_sync     = vfs_quota_sync,
2059         .get_info       = vfs_get_dqinfo,
2060         .set_info       = vfs_set_dqinfo,
2061         .get_dqblk      = vfs_get_dqblk,
2062         .set_dqblk      = vfs_set_dqblk
2063 };
2064
2065 static ctl_table fs_dqstats_table[] = {
2066         {
2067                 .ctl_name       = FS_DQ_LOOKUPS,
2068                 .procname       = "lookups",
2069                 .data           = &dqstats.lookups,
2070                 .maxlen         = sizeof(int),
2071                 .mode           = 0444,
2072                 .proc_handler   = &proc_dointvec,
2073         },
2074         {
2075                 .ctl_name       = FS_DQ_DROPS,
2076                 .procname       = "drops",
2077                 .data           = &dqstats.drops,
2078                 .maxlen         = sizeof(int),
2079                 .mode           = 0444,
2080                 .proc_handler   = &proc_dointvec,
2081         },
2082         {
2083                 .ctl_name       = FS_DQ_READS,
2084                 .procname       = "reads",
2085                 .data           = &dqstats.reads,
2086                 .maxlen         = sizeof(int),
2087                 .mode           = 0444,
2088                 .proc_handler   = &proc_dointvec,
2089         },
2090         {
2091                 .ctl_name       = FS_DQ_WRITES,
2092                 .procname       = "writes",
2093                 .data           = &dqstats.writes,
2094                 .maxlen         = sizeof(int),
2095                 .mode           = 0444,
2096                 .proc_handler   = &proc_dointvec,
2097         },
2098         {
2099                 .ctl_name       = FS_DQ_CACHE_HITS,
2100                 .procname       = "cache_hits",
2101                 .data           = &dqstats.cache_hits,
2102                 .maxlen         = sizeof(int),
2103                 .mode           = 0444,
2104                 .proc_handler   = &proc_dointvec,
2105         },
2106         {
2107                 .ctl_name       = FS_DQ_ALLOCATED,
2108                 .procname       = "allocated_dquots",
2109                 .data           = &dqstats.allocated_dquots,
2110                 .maxlen         = sizeof(int),
2111                 .mode           = 0444,
2112                 .proc_handler   = &proc_dointvec,
2113         },
2114         {
2115                 .ctl_name       = FS_DQ_FREE,
2116                 .procname       = "free_dquots",
2117                 .data           = &dqstats.free_dquots,
2118                 .maxlen         = sizeof(int),
2119                 .mode           = 0444,
2120                 .proc_handler   = &proc_dointvec,
2121         },
2122         {
2123                 .ctl_name       = FS_DQ_SYNCS,
2124                 .procname       = "syncs",
2125                 .data           = &dqstats.syncs,
2126                 .maxlen         = sizeof(int),
2127                 .mode           = 0444,
2128                 .proc_handler   = &proc_dointvec,
2129         },
2130 #ifdef CONFIG_PRINT_QUOTA_WARNING
2131         {
2132                 .ctl_name       = FS_DQ_WARNINGS,
2133                 .procname       = "warnings",
2134                 .data           = &flag_print_warnings,
2135                 .maxlen         = sizeof(int),
2136                 .mode           = 0644,
2137                 .proc_handler   = &proc_dointvec,
2138         },
2139 #endif
2140         { .ctl_name = 0 },
2141 };
2142
2143 static ctl_table fs_table[] = {
2144         {
2145                 .ctl_name       = FS_DQSTATS,
2146                 .procname       = "quota",
2147                 .mode           = 0555,
2148                 .child          = fs_dqstats_table,
2149         },
2150         { .ctl_name = 0 },
2151 };
2152
2153 static ctl_table sys_table[] = {
2154         {
2155                 .ctl_name       = CTL_FS,
2156                 .procname       = "fs",
2157                 .mode           = 0555,
2158                 .child          = fs_table,
2159         },
2160         { .ctl_name = 0 },
2161 };
2162
2163 static int __init dquot_init(void)
2164 {
2165         int i;
2166         unsigned long nr_hash, order;
2167
2168         printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2169
2170         register_sysctl_table(sys_table);
2171
2172         dquot_cachep = kmem_cache_create("dquot",
2173                         sizeof(struct dquot), sizeof(unsigned long) * 4,
2174                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2175                                 SLAB_MEM_SPREAD|SLAB_PANIC),
2176                         NULL);
2177
2178         order = 0;
2179         dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2180         if (!dquot_hash)
2181                 panic("Cannot create dquot hash table");
2182
2183         /* Find power-of-two hlist_heads which can fit into allocation */
2184         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2185         dq_hash_bits = 0;
2186         do {
2187                 dq_hash_bits++;
2188         } while (nr_hash >> dq_hash_bits);
2189         dq_hash_bits--;
2190
2191         nr_hash = 1UL << dq_hash_bits;
2192         dq_hash_mask = nr_hash - 1;
2193         for (i = 0; i < nr_hash; i++)
2194                 INIT_HLIST_HEAD(dquot_hash + i);
2195
2196         printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
2197                         nr_hash, order, (PAGE_SIZE << order));
2198
2199         register_shrinker(&dqcache_shrinker);
2200
2201 #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
2202         if (genl_register_family(&quota_genl_family) != 0)
2203                 printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
2204 #endif
2205
2206         return 0;
2207 }
2208 module_init(dquot_init);
2209
2210 EXPORT_SYMBOL(register_quota_format);
2211 EXPORT_SYMBOL(unregister_quota_format);
2212 EXPORT_SYMBOL(dqstats);
2213 EXPORT_SYMBOL(dq_data_lock);
2214 EXPORT_SYMBOL(vfs_quota_on);
2215 EXPORT_SYMBOL(vfs_quota_on_path);
2216 EXPORT_SYMBOL(vfs_quota_on_mount);
2217 EXPORT_SYMBOL(vfs_quota_off);
2218 EXPORT_SYMBOL(vfs_quota_sync);
2219 EXPORT_SYMBOL(vfs_get_dqinfo);
2220 EXPORT_SYMBOL(vfs_set_dqinfo);
2221 EXPORT_SYMBOL(vfs_get_dqblk);
2222 EXPORT_SYMBOL(vfs_set_dqblk);
2223 EXPORT_SYMBOL(dquot_commit);
2224 EXPORT_SYMBOL(dquot_commit_info);
2225 EXPORT_SYMBOL(dquot_acquire);
2226 EXPORT_SYMBOL(dquot_release);
2227 EXPORT_SYMBOL(dquot_mark_dquot_dirty);
2228 EXPORT_SYMBOL(dquot_initialize);
2229 EXPORT_SYMBOL(dquot_drop);
2230 EXPORT_SYMBOL(vfs_dq_drop);
2231 EXPORT_SYMBOL(dquot_alloc_space);
2232 EXPORT_SYMBOL(dquot_alloc_inode);
2233 EXPORT_SYMBOL(dquot_free_space);
2234 EXPORT_SYMBOL(dquot_free_inode);
2235 EXPORT_SYMBOL(dquot_transfer);
2236 EXPORT_SYMBOL(vfs_dq_transfer);
2237 EXPORT_SYMBOL(vfs_dq_quota_on_remount);