linux/fs/quota/dquot.c
<<
>>
Prefs
   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 <mvw@planets.elm.net>
  13 *
  14 * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
  15 *
  16 *              Revised list management to avoid races
  17 *              -- Bill Hawes, <whawes@star.net>, 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, <jack@suse.cz>, 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, <jack@suse.cz>, 9/2000
  39 *
  40 *              Added dynamic quota structure allocation
  41 *              Jan Kara <jack@suse.cz> 12/2000
  42 *
  43 *              Rewritten quota interface. Implemented new quota format and
  44 *              formats registering.
  45 *              Jan Kara, <jack@suse.cz>, 2001,2002
  46 *
  47 *              New SMP locking.
  48 *              Jan Kara, <jack@suse.cz>, 10/2002
  49 *
  50 *              Added journalled quota support, fix lock inversion problems
  51 *              Jan Kara, <jack@suse.cz>, 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/sched.h>
  75#include <linux/cred.h>
  76#include <linux/kmod.h>
  77#include <linux/namei.h>
  78#include <linux/capability.h>
  79#include <linux/quotaops.h>
  80#include "../internal.h" /* ugh */
  81
  82#include <linux/uaccess.h>
  83
  84/*
  85 * There are three quota SMP locks. dq_list_lock protects all lists with quotas
  86 * and quota formats.
  87 * dq_data_lock protects data from dq_dqb and also mem_dqinfo structures and
  88 * also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
  89 * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
  90 * in inode_add_bytes() and inode_sub_bytes(). dq_state_lock protects
  91 * modifications of quota state (on quotaon and quotaoff) and readers who care
  92 * about latest values take it as well.
  93 *
  94 * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock,
  95 *   dq_list_lock > dq_state_lock
  96 *
  97 * Note that some things (eg. sb pointer, type, id) doesn't change during
  98 * the life of the dquot structure and so needn't to be protected by a lock
  99 *
 100 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
 101 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
 102 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
 103 * inode and before dropping dquot references to avoid use of dquots after
 104 * they are freed. dq_data_lock is used to serialize the pointer setting and
 105 * clearing operations.
 106 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
 107 * inode is a quota file). Functions adding pointers from inode to dquots have
 108 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
 109 * have to do all pointer modifications before dropping dq_data_lock. This makes
 110 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
 111 * then drops all pointers to dquots from an inode.
 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 *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_mutex
 124 */
 125
 126static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
 127static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
 128__cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
 129EXPORT_SYMBOL(dq_data_lock);
 130DEFINE_STATIC_SRCU(dquot_srcu);
 131
 132void __quota_error(struct super_block *sb, const char *func,
 133                   const char *fmt, ...)
 134{
 135        if (printk_ratelimit()) {
 136                va_list args;
 137                struct va_format vaf;
 138
 139                va_start(args, fmt);
 140
 141                vaf.fmt = fmt;
 142                vaf.va = &args;
 143
 144                printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
 145                       sb->s_id, func, &vaf);
 146
 147                va_end(args);
 148        }
 149}
 150EXPORT_SYMBOL(__quota_error);
 151
 152#if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
 153static char *quotatypes[] = INITQFNAMES;
 154#endif
 155static struct quota_format_type *quota_formats; /* List of registered formats */
 156static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 157
 158/* SLAB cache for dquot structures */
 159static struct kmem_cache *dquot_cachep;
 160
 161int register_quota_format(struct quota_format_type *fmt)
 162{
 163        spin_lock(&dq_list_lock);
 164        fmt->qf_next = quota_formats;
 165        quota_formats = fmt;
 166        spin_unlock(&dq_list_lock);
 167        return 0;
 168}
 169EXPORT_SYMBOL(register_quota_format);
 170
 171void unregister_quota_format(struct quota_format_type *fmt)
 172{
 173        struct quota_format_type **actqf;
 174
 175        spin_lock(&dq_list_lock);
 176        for (actqf = &quota_formats; *actqf && *actqf != fmt;
 177             actqf = &(*actqf)->qf_next)
 178                ;
 179        if (*actqf)
 180                *actqf = (*actqf)->qf_next;
 181        spin_unlock(&dq_list_lock);
 182}
 183EXPORT_SYMBOL(unregister_quota_format);
 184
 185static struct quota_format_type *find_quota_format(int id)
 186{
 187        struct quota_format_type *actqf;
 188
 189        spin_lock(&dq_list_lock);
 190        for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
 191             actqf = actqf->qf_next)
 192                ;
 193        if (!actqf || !try_module_get(actqf->qf_owner)) {
 194                int qm;
 195
 196                spin_unlock(&dq_list_lock);
 197                
 198                for (qm = 0; module_names[qm].qm_fmt_id &&
 199                             module_names[qm].qm_fmt_id != id; qm++)
 200                        ;
 201                if (!module_names[qm].qm_fmt_id ||
 202                    request_module(module_names[qm].qm_mod_name))
 203                        return NULL;
 204
 205                spin_lock(&dq_list_lock);
 206                for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
 207                     actqf = actqf->qf_next)
 208                        ;
 209                if (actqf && !try_module_get(actqf->qf_owner))
 210                        actqf = NULL;
 211        }
 212        spin_unlock(&dq_list_lock);
 213        return actqf;
 214}
 215
 216static void put_quota_format(struct quota_format_type *fmt)
 217{
 218        module_put(fmt->qf_owner);
 219}
 220
 221/*
 222 * Dquot List Management:
 223 * The quota code uses three lists for dquot management: the inuse_list,
 224 * free_dquots, and dquot_hash[] array. A single dquot structure may be
 225 * on all three lists, depending on its current state.
 226 *
 227 * All dquots are placed to the end of inuse_list when first created, and this
 228 * list is used for invalidate operation, which must look at every dquot.
 229 *
 230 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 231 * and this list is searched whenever we need an available dquot.  Dquots are
 232 * removed from the list as soon as they are used again, and
 233 * dqstats.free_dquots gives the number of dquots on the list. When
 234 * dquot is invalidated it's completely released from memory.
 235 *
 236 * Dquots with a specific identity (device, type and id) are placed on
 237 * one of the dquot_hash[] hash chains. The provides an efficient search
 238 * mechanism to locate a specific dquot.
 239 */
 240
 241static LIST_HEAD(inuse_list);
 242static LIST_HEAD(free_dquots);
 243static unsigned int dq_hash_bits, dq_hash_mask;
 244static struct hlist_head *dquot_hash;
 245
 246struct dqstats dqstats;
 247EXPORT_SYMBOL(dqstats);
 248
 249static qsize_t inode_get_rsv_space(struct inode *inode);
 250static int __dquot_initialize(struct inode *inode, int type);
 251
 252static inline unsigned int
 253hashfn(const struct super_block *sb, struct kqid qid)
 254{
 255        unsigned int id = from_kqid(&init_user_ns, qid);
 256        int type = qid.type;
 257        unsigned long tmp;
 258
 259        tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 260        return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 261}
 262
 263/*
 264 * Following list functions expect dq_list_lock to be held
 265 */
 266static inline void insert_dquot_hash(struct dquot *dquot)
 267{
 268        struct hlist_head *head;
 269        head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
 270        hlist_add_head(&dquot->dq_hash, head);
 271}
 272
 273static inline void remove_dquot_hash(struct dquot *dquot)
 274{
 275        hlist_del_init(&dquot->dq_hash);
 276}
 277
 278static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
 279                                struct kqid qid)
 280{
 281        struct hlist_node *node;
 282        struct dquot *dquot;
 283
 284        hlist_for_each (node, dquot_hash+hashent) {
 285                dquot = hlist_entry(node, struct dquot, dq_hash);
 286                if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
 287                        return dquot;
 288        }
 289        return NULL;
 290}
 291
 292/* Add a dquot to the tail of the free list */
 293static inline void put_dquot_last(struct dquot *dquot)
 294{
 295        list_add_tail(&dquot->dq_free, &free_dquots);
 296        dqstats_inc(DQST_FREE_DQUOTS);
 297}
 298
 299static inline void remove_free_dquot(struct dquot *dquot)
 300{
 301        if (list_empty(&dquot->dq_free))
 302                return;
 303        list_del_init(&dquot->dq_free);
 304        dqstats_dec(DQST_FREE_DQUOTS);
 305}
 306
 307static inline void put_inuse(struct dquot *dquot)
 308{
 309        /* We add to the back of inuse list so we don't have to restart
 310         * when traversing this list and we block */
 311        list_add_tail(&dquot->dq_inuse, &inuse_list);
 312        dqstats_inc(DQST_ALLOC_DQUOTS);
 313}
 314
 315static inline void remove_inuse(struct dquot *dquot)
 316{
 317        dqstats_dec(DQST_ALLOC_DQUOTS);
 318        list_del(&dquot->dq_inuse);
 319}
 320/*
 321 * End of list functions needing dq_list_lock
 322 */
 323
 324static void wait_on_dquot(struct dquot *dquot)
 325{
 326        mutex_lock(&dquot->dq_lock);
 327        mutex_unlock(&dquot->dq_lock);
 328}
 329
 330static inline int dquot_dirty(struct dquot *dquot)
 331{
 332        return test_bit(DQ_MOD_B, &dquot->dq_flags);
 333}
 334
 335static inline int mark_dquot_dirty(struct dquot *dquot)
 336{
 337        return dquot->dq_sb->dq_op->mark_dirty(dquot);
 338}
 339
 340/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
 341int dquot_mark_dquot_dirty(struct dquot *dquot)
 342{
 343        int ret = 1;
 344
 345        /* If quota is dirty already, we don't have to acquire dq_list_lock */
 346        if (test_bit(DQ_MOD_B, &dquot->dq_flags))
 347                return 1;
 348
 349        spin_lock(&dq_list_lock);
 350        if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
 351                list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 352                                info[dquot->dq_id.type].dqi_dirty_list);
 353                ret = 0;
 354        }
 355        spin_unlock(&dq_list_lock);
 356        return ret;
 357}
 358EXPORT_SYMBOL(dquot_mark_dquot_dirty);
 359
 360/* Dirtify all the dquots - this can block when journalling */
 361static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
 362{
 363        int ret, err, cnt;
 364
 365        ret = err = 0;
 366        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 367                if (dquot[cnt])
 368                        /* Even in case of error we have to continue */
 369                        ret = mark_dquot_dirty(dquot[cnt]);
 370                if (!err)
 371                        err = ret;
 372        }
 373        return err;
 374}
 375
 376static inline void dqput_all(struct dquot **dquot)
 377{
 378        unsigned int cnt;
 379
 380        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 381                dqput(dquot[cnt]);
 382}
 383
 384/* This function needs dq_list_lock */
 385static inline int clear_dquot_dirty(struct dquot *dquot)
 386{
 387        if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
 388                return 0;
 389        list_del_init(&dquot->dq_dirty);
 390        return 1;
 391}
 392
 393void mark_info_dirty(struct super_block *sb, int type)
 394{
 395        set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
 396}
 397EXPORT_SYMBOL(mark_info_dirty);
 398
 399/*
 400 *      Read dquot from disk and alloc space for it
 401 */
 402
 403int dquot_acquire(struct dquot *dquot)
 404{
 405        int ret = 0, ret2 = 0;
 406        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 407
 408        mutex_lock(&dquot->dq_lock);
 409        mutex_lock(&dqopt->dqio_mutex);
 410        if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 411                ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
 412        if (ret < 0)
 413                goto out_iolock;
 414        /* Make sure flags update is visible after dquot has been filled */
 415        smp_mb__before_atomic();
 416        set_bit(DQ_READ_B, &dquot->dq_flags);
 417        /* Instantiate dquot if needed */
 418        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 419                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 420                /* Write the info if needed */
 421                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 422                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 423                                        dquot->dq_sb, dquot->dq_id.type);
 424                }
 425                if (ret < 0)
 426                        goto out_iolock;
 427                if (ret2 < 0) {
 428                        ret = ret2;
 429                        goto out_iolock;
 430                }
 431        }
 432        /*
 433         * Make sure flags update is visible after on-disk struct has been
 434         * allocated. Paired with smp_rmb() in dqget().
 435         */
 436        smp_mb__before_atomic();
 437        set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 438out_iolock:
 439        mutex_unlock(&dqopt->dqio_mutex);
 440        mutex_unlock(&dquot->dq_lock);
 441        return ret;
 442}
 443EXPORT_SYMBOL(dquot_acquire);
 444
 445/*
 446 *      Write dquot to disk
 447 */
 448int dquot_commit(struct dquot *dquot)
 449{
 450        int ret = 0;
 451        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 452
 453        mutex_lock(&dqopt->dqio_mutex);
 454        spin_lock(&dq_list_lock);
 455        if (!clear_dquot_dirty(dquot)) {
 456                spin_unlock(&dq_list_lock);
 457                goto out_sem;
 458        }
 459        spin_unlock(&dq_list_lock);
 460        /* Inactive dquot can be only if there was error during read/init
 461         * => we have better not writing it */
 462        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 463                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 464        else
 465                ret = -EIO;
 466out_sem:
 467        mutex_unlock(&dqopt->dqio_mutex);
 468        return ret;
 469}
 470EXPORT_SYMBOL(dquot_commit);
 471
 472/*
 473 *      Release dquot
 474 */
 475int dquot_release(struct dquot *dquot)
 476{
 477        int ret = 0, ret2 = 0;
 478        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 479
 480        mutex_lock(&dquot->dq_lock);
 481        /* Check whether we are not racing with some other dqget() */
 482        if (atomic_read(&dquot->dq_count) > 1)
 483                goto out_dqlock;
 484        mutex_lock(&dqopt->dqio_mutex);
 485        if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
 486                ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
 487                /* Write the info */
 488                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 489                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 490                                                dquot->dq_sb, dquot->dq_id.type);
 491                }
 492                if (ret >= 0)
 493                        ret = ret2;
 494        }
 495        clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 496        mutex_unlock(&dqopt->dqio_mutex);
 497out_dqlock:
 498        mutex_unlock(&dquot->dq_lock);
 499        return ret;
 500}
 501EXPORT_SYMBOL(dquot_release);
 502
 503void dquot_destroy(struct dquot *dquot)
 504{
 505        kmem_cache_free(dquot_cachep, dquot);
 506}
 507EXPORT_SYMBOL(dquot_destroy);
 508
 509static inline void do_destroy_dquot(struct dquot *dquot)
 510{
 511        dquot->dq_sb->dq_op->destroy_dquot(dquot);
 512}
 513
 514/* Invalidate all dquots on the list. Note that this function is called after
 515 * quota is disabled and pointers from inodes removed so there cannot be new
 516 * quota users. There can still be some users of quotas due to inodes being
 517 * just deleted or pruned by prune_icache() (those are not attached to any
 518 * list) or parallel quotactl call. We have to wait for such users.
 519 */
 520static void invalidate_dquots(struct super_block *sb, int type)
 521{
 522        struct dquot *dquot, *tmp;
 523
 524restart:
 525        spin_lock(&dq_list_lock);
 526        list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 527                if (dquot->dq_sb != sb)
 528                        continue;
 529                if (dquot->dq_id.type != type)
 530                        continue;
 531                /* Wait for dquot users */
 532                if (atomic_read(&dquot->dq_count)) {
 533                        DEFINE_WAIT(wait);
 534
 535                        dqgrab(dquot);
 536                        prepare_to_wait(&dquot->dq_wait_unused, &wait,
 537                                        TASK_UNINTERRUPTIBLE);
 538                        spin_unlock(&dq_list_lock);
 539                        /* Once dqput() wakes us up, we know it's time to free
 540                         * the dquot.
 541                         * IMPORTANT: we rely on the fact that there is always
 542                         * at most one process waiting for dquot to free.
 543                         * Otherwise dq_count would be > 1 and we would never
 544                         * wake up.
 545                         */
 546                        if (atomic_read(&dquot->dq_count) > 1)
 547                                schedule();
 548                        finish_wait(&dquot->dq_wait_unused, &wait);
 549                        dqput(dquot);
 550                        /* At this moment dquot() need not exist (it could be
 551                         * reclaimed by prune_dqcache(). Hence we must
 552                         * restart. */
 553                        goto restart;
 554                }
 555                /*
 556                 * Quota now has no users and it has been written on last
 557                 * dqput()
 558                 */
 559                remove_dquot_hash(dquot);
 560                remove_free_dquot(dquot);
 561                remove_inuse(dquot);
 562                do_destroy_dquot(dquot);
 563        }
 564        spin_unlock(&dq_list_lock);
 565}
 566
 567/* Call callback for every active dquot on given filesystem */
 568int dquot_scan_active(struct super_block *sb,
 569                      int (*fn)(struct dquot *dquot, unsigned long priv),
 570                      unsigned long priv)
 571{
 572        struct dquot *dquot, *old_dquot = NULL;
 573        int ret = 0;
 574
 575        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 576
 577        spin_lock(&dq_list_lock);
 578        list_for_each_entry(dquot, &inuse_list, dq_inuse) {
 579                if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 580                        continue;
 581                if (dquot->dq_sb != sb)
 582                        continue;
 583                /* Now we have active dquot so we can just increase use count */
 584                atomic_inc(&dquot->dq_count);
 585                spin_unlock(&dq_list_lock);
 586                dqstats_inc(DQST_LOOKUPS);
 587                dqput(old_dquot);
 588                old_dquot = dquot;
 589                /*
 590                 * ->release_dquot() can be racing with us. Our reference
 591                 * protects us from new calls to it so just wait for any
 592                 * outstanding call and recheck the DQ_ACTIVE_B after that.
 593                 */
 594                wait_on_dquot(dquot);
 595                if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 596                        ret = fn(dquot, priv);
 597                        if (ret < 0)
 598                                goto out;
 599                }
 600                spin_lock(&dq_list_lock);
 601                /* We are safe to continue now because our dquot could not
 602                 * be moved out of the inuse list while we hold the reference */
 603        }
 604        spin_unlock(&dq_list_lock);
 605out:
 606        dqput(old_dquot);
 607        return ret;
 608}
 609EXPORT_SYMBOL(dquot_scan_active);
 610
 611/* Write all dquot structures to quota files */
 612int dquot_writeback_dquots(struct super_block *sb, int type)
 613{
 614        struct list_head *dirty;
 615        struct dquot *dquot;
 616        struct quota_info *dqopt = sb_dqopt(sb);
 617        int cnt;
 618        int err, ret = 0;
 619
 620        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 621
 622        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 623                if (type != -1 && cnt != type)
 624                        continue;
 625                if (!sb_has_quota_active(sb, cnt))
 626                        continue;
 627                spin_lock(&dq_list_lock);
 628                dirty = &dqopt->info[cnt].dqi_dirty_list;
 629                while (!list_empty(dirty)) {
 630                        dquot = list_first_entry(dirty, struct dquot,
 631                                                 dq_dirty);
 632                        /* Dirty and inactive can be only bad dquot... */
 633                        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 634                                clear_dquot_dirty(dquot);
 635                                continue;
 636                        }
 637                        /* Now we have active dquot from which someone is
 638                         * holding reference so we can safely just increase
 639                         * use count */
 640                        dqgrab(dquot);
 641                        spin_unlock(&dq_list_lock);
 642                        dqstats_inc(DQST_LOOKUPS);
 643                        err = sb->dq_op->write_dquot(dquot);
 644                        if (!ret && err)
 645                                ret = err;
 646                        dqput(dquot);
 647                        spin_lock(&dq_list_lock);
 648                }
 649                spin_unlock(&dq_list_lock);
 650        }
 651
 652        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 653                if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
 654                    && info_dirty(&dqopt->info[cnt]))
 655                        sb->dq_op->write_info(sb, cnt);
 656        dqstats_inc(DQST_SYNCS);
 657
 658        return ret;
 659}
 660EXPORT_SYMBOL(dquot_writeback_dquots);
 661
 662/* Write all dquot structures to disk and make them visible from userspace */
 663int dquot_quota_sync(struct super_block *sb, int type)
 664{
 665        struct quota_info *dqopt = sb_dqopt(sb);
 666        int cnt;
 667        int ret;
 668
 669        ret = dquot_writeback_dquots(sb, type);
 670        if (ret)
 671                return ret;
 672        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
 673                return 0;
 674
 675        /* This is not very clever (and fast) but currently I don't know about
 676         * any other simple way of getting quota data to disk and we must get
 677         * them there for userspace to be visible... */
 678        if (sb->s_op->sync_fs)
 679                sb->s_op->sync_fs(sb, 1);
 680        sync_blockdev(sb->s_bdev);
 681
 682        /*
 683         * Now when everything is written we can discard the pagecache so
 684         * that userspace sees the changes.
 685         */
 686        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 687                if (type != -1 && cnt != type)
 688                        continue;
 689                if (!sb_has_quota_active(sb, cnt))
 690                        continue;
 691                inode_lock(dqopt->files[cnt]);
 692                truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
 693                inode_unlock(dqopt->files[cnt]);
 694        }
 695
 696        return 0;
 697}
 698EXPORT_SYMBOL(dquot_quota_sync);
 699
 700static unsigned long
 701dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 702{
 703        struct list_head *head;
 704        struct dquot *dquot;
 705        unsigned long freed = 0;
 706
 707        spin_lock(&dq_list_lock);
 708        head = free_dquots.prev;
 709        while (head != &free_dquots && sc->nr_to_scan) {
 710                dquot = list_entry(head, struct dquot, dq_free);
 711                remove_dquot_hash(dquot);
 712                remove_free_dquot(dquot);
 713                remove_inuse(dquot);
 714                do_destroy_dquot(dquot);
 715                sc->nr_to_scan--;
 716                freed++;
 717                head = free_dquots.prev;
 718        }
 719        spin_unlock(&dq_list_lock);
 720        return freed;
 721}
 722
 723static unsigned long
 724dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
 725{
 726        return vfs_pressure_ratio(
 727        percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
 728}
 729
 730static struct shrinker dqcache_shrinker = {
 731        .count_objects = dqcache_shrink_count,
 732        .scan_objects = dqcache_shrink_scan,
 733        .seeks = DEFAULT_SEEKS,
 734};
 735
 736/*
 737 * Put reference to dquot
 738 */
 739void dqput(struct dquot *dquot)
 740{
 741        int ret;
 742
 743        if (!dquot)
 744                return;
 745#ifdef CONFIG_QUOTA_DEBUG
 746        if (!atomic_read(&dquot->dq_count)) {
 747                quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
 748                            quotatypes[dquot->dq_id.type],
 749                            from_kqid(&init_user_ns, dquot->dq_id));
 750                BUG();
 751        }
 752#endif
 753        dqstats_inc(DQST_DROPS);
 754we_slept:
 755        spin_lock(&dq_list_lock);
 756        if (atomic_read(&dquot->dq_count) > 1) {
 757                /* We have more than one user... nothing to do */
 758                atomic_dec(&dquot->dq_count);
 759                /* Releasing dquot during quotaoff phase? */
 760                if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
 761                    atomic_read(&dquot->dq_count) == 1)
 762                        wake_up(&dquot->dq_wait_unused);
 763                spin_unlock(&dq_list_lock);
 764                return;
 765        }
 766        /* Need to release dquot? */
 767        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
 768                spin_unlock(&dq_list_lock);
 769                /* Commit dquot before releasing */
 770                ret = dquot->dq_sb->dq_op->write_dquot(dquot);
 771                if (ret < 0) {
 772                        quota_error(dquot->dq_sb, "Can't write quota structure"
 773                                    " (error %d). Quota may get out of sync!",
 774                                    ret);
 775                        /*
 776                         * We clear dirty bit anyway, so that we avoid
 777                         * infinite loop here
 778                         */
 779                        spin_lock(&dq_list_lock);
 780                        clear_dquot_dirty(dquot);
 781                        spin_unlock(&dq_list_lock);
 782                }
 783                goto we_slept;
 784        }
 785        /* Clear flag in case dquot was inactive (something bad happened) */
 786        clear_dquot_dirty(dquot);
 787        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 788                spin_unlock(&dq_list_lock);
 789                dquot->dq_sb->dq_op->release_dquot(dquot);
 790                goto we_slept;
 791        }
 792        atomic_dec(&dquot->dq_count);
 793#ifdef CONFIG_QUOTA_DEBUG
 794        /* sanity check */
 795        BUG_ON(!list_empty(&dquot->dq_free));
 796#endif
 797        put_dquot_last(dquot);
 798        spin_unlock(&dq_list_lock);
 799}
 800EXPORT_SYMBOL(dqput);
 801
 802struct dquot *dquot_alloc(struct super_block *sb, int type)
 803{
 804        return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
 805}
 806EXPORT_SYMBOL(dquot_alloc);
 807
 808static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 809{
 810        struct dquot *dquot;
 811
 812        dquot = sb->dq_op->alloc_dquot(sb, type);
 813        if(!dquot)
 814                return NULL;
 815
 816        mutex_init(&dquot->dq_lock);
 817        INIT_LIST_HEAD(&dquot->dq_free);
 818        INIT_LIST_HEAD(&dquot->dq_inuse);
 819        INIT_HLIST_NODE(&dquot->dq_hash);
 820        INIT_LIST_HEAD(&dquot->dq_dirty);
 821        init_waitqueue_head(&dquot->dq_wait_unused);
 822        dquot->dq_sb = sb;
 823        dquot->dq_id = make_kqid_invalid(type);
 824        atomic_set(&dquot->dq_count, 1);
 825
 826        return dquot;
 827}
 828
 829/*
 830 * Get reference to dquot
 831 *
 832 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
 833 * destroying our dquot by:
 834 *   a) checking for quota flags under dq_list_lock and
 835 *   b) getting a reference to dquot before we release dq_list_lock
 836 */
 837struct dquot *dqget(struct super_block *sb, struct kqid qid)
 838{
 839        unsigned int hashent = hashfn(sb, qid);
 840        struct dquot *dquot, *empty = NULL;
 841
 842        if (!qid_has_mapping(sb->s_user_ns, qid))
 843                return ERR_PTR(-EINVAL);
 844
 845        if (!sb_has_quota_active(sb, qid.type))
 846                return ERR_PTR(-ESRCH);
 847we_slept:
 848        spin_lock(&dq_list_lock);
 849        spin_lock(&dq_state_lock);
 850        if (!sb_has_quota_active(sb, qid.type)) {
 851                spin_unlock(&dq_state_lock);
 852                spin_unlock(&dq_list_lock);
 853                dquot = ERR_PTR(-ESRCH);
 854                goto out;
 855        }
 856        spin_unlock(&dq_state_lock);
 857
 858        dquot = find_dquot(hashent, sb, qid);
 859        if (!dquot) {
 860                if (!empty) {
 861                        spin_unlock(&dq_list_lock);
 862                        empty = get_empty_dquot(sb, qid.type);
 863                        if (!empty)
 864                                schedule();     /* Try to wait for a moment... */
 865                        goto we_slept;
 866                }
 867                dquot = empty;
 868                empty = NULL;
 869                dquot->dq_id = qid;
 870                /* all dquots go on the inuse_list */
 871                put_inuse(dquot);
 872                /* hash it first so it can be found */
 873                insert_dquot_hash(dquot);
 874                spin_unlock(&dq_list_lock);
 875                dqstats_inc(DQST_LOOKUPS);
 876        } else {
 877                if (!atomic_read(&dquot->dq_count))
 878                        remove_free_dquot(dquot);
 879                atomic_inc(&dquot->dq_count);
 880                spin_unlock(&dq_list_lock);
 881                dqstats_inc(DQST_CACHE_HITS);
 882                dqstats_inc(DQST_LOOKUPS);
 883        }
 884        /* Wait for dq_lock - after this we know that either dquot_release() is
 885         * already finished or it will be canceled due to dq_count > 1 test */
 886        wait_on_dquot(dquot);
 887        /* Read the dquot / allocate space in quota file */
 888        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 889                int err;
 890
 891                err = sb->dq_op->acquire_dquot(dquot);
 892                if (err < 0) {
 893                        dqput(dquot);
 894                        dquot = ERR_PTR(err);
 895                        goto out;
 896                }
 897        }
 898        /*
 899         * Make sure following reads see filled structure - paired with
 900         * smp_mb__before_atomic() in dquot_acquire().
 901         */
 902        smp_rmb();
 903#ifdef CONFIG_QUOTA_DEBUG
 904        BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 905#endif
 906out:
 907        if (empty)
 908                do_destroy_dquot(empty);
 909
 910        return dquot;
 911}
 912EXPORT_SYMBOL(dqget);
 913
 914static inline struct dquot **i_dquot(struct inode *inode)
 915{
 916        return inode->i_sb->s_op->get_dquots(inode);
 917}
 918
 919static int dqinit_needed(struct inode *inode, int type)
 920{
 921        struct dquot * const *dquots;
 922        int cnt;
 923
 924        if (IS_NOQUOTA(inode))
 925                return 0;
 926
 927        dquots = i_dquot(inode);
 928        if (type != -1)
 929                return !dquots[type];
 930        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 931                if (!dquots[cnt])
 932                        return 1;
 933        return 0;
 934}
 935
 936/* This routine is guarded by s_umount semaphore */
 937static void add_dquot_ref(struct super_block *sb, int type)
 938{
 939        struct inode *inode, *old_inode = NULL;
 940#ifdef CONFIG_QUOTA_DEBUG
 941        int reserved = 0;
 942#endif
 943
 944        spin_lock(&sb->s_inode_list_lock);
 945        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 946                spin_lock(&inode->i_lock);
 947                if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
 948                    !atomic_read(&inode->i_writecount) ||
 949                    !dqinit_needed(inode, type)) {
 950                        spin_unlock(&inode->i_lock);
 951                        continue;
 952                }
 953                __iget(inode);
 954                spin_unlock(&inode->i_lock);
 955                spin_unlock(&sb->s_inode_list_lock);
 956
 957#ifdef CONFIG_QUOTA_DEBUG
 958                if (unlikely(inode_get_rsv_space(inode) > 0))
 959                        reserved = 1;
 960#endif
 961                iput(old_inode);
 962                __dquot_initialize(inode, type);
 963
 964                /*
 965                 * We hold a reference to 'inode' so it couldn't have been
 966                 * removed from s_inodes list while we dropped the
 967                 * s_inode_list_lock. We cannot iput the inode now as we can be
 968                 * holding the last reference and we cannot iput it under
 969                 * s_inode_list_lock. So we keep the reference and iput it
 970                 * later.
 971                 */
 972                old_inode = inode;
 973                spin_lock(&sb->s_inode_list_lock);
 974        }
 975        spin_unlock(&sb->s_inode_list_lock);
 976        iput(old_inode);
 977
 978#ifdef CONFIG_QUOTA_DEBUG
 979        if (reserved) {
 980                quota_error(sb, "Writes happened before quota was turned on "
 981                        "thus quota information is probably inconsistent. "
 982                        "Please run quotacheck(8)");
 983        }
 984#endif
 985}
 986
 987/*
 988 * Remove references to dquots from inode and add dquot to list for freeing
 989 * if we have the last reference to dquot
 990 */
 991static void remove_inode_dquot_ref(struct inode *inode, int type,
 992                                   struct list_head *tofree_head)
 993{
 994        struct dquot **dquots = i_dquot(inode);
 995        struct dquot *dquot = dquots[type];
 996
 997        if (!dquot)
 998                return;
 999
1000        dquots[type] = NULL;
1001        if (list_empty(&dquot->dq_free)) {
1002                /*
1003                 * The inode still has reference to dquot so it can't be in the
1004                 * free list
1005                 */
1006                spin_lock(&dq_list_lock);
1007                list_add(&dquot->dq_free, tofree_head);
1008                spin_unlock(&dq_list_lock);
1009        } else {
1010                /*
1011                 * Dquot is already in a list to put so we won't drop the last
1012                 * reference here.
1013                 */
1014                dqput(dquot);
1015        }
1016}
1017
1018/*
1019 * Free list of dquots
1020 * Dquots are removed from inodes and no new references can be got so we are
1021 * the only ones holding reference
1022 */
1023static void put_dquot_list(struct list_head *tofree_head)
1024{
1025        struct list_head *act_head;
1026        struct dquot *dquot;
1027
1028        act_head = tofree_head->next;
1029        while (act_head != tofree_head) {
1030                dquot = list_entry(act_head, struct dquot, dq_free);
1031                act_head = act_head->next;
1032                /* Remove dquot from the list so we won't have problems... */
1033                list_del_init(&dquot->dq_free);
1034                dqput(dquot);
1035        }
1036}
1037
1038static void remove_dquot_ref(struct super_block *sb, int type,
1039                struct list_head *tofree_head)
1040{
1041        struct inode *inode;
1042        int reserved = 0;
1043
1044        spin_lock(&sb->s_inode_list_lock);
1045        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1046                /*
1047                 *  We have to scan also I_NEW inodes because they can already
1048                 *  have quota pointer initialized. Luckily, we need to touch
1049                 *  only quota pointers and these have separate locking
1050                 *  (dq_data_lock).
1051                 */
1052                spin_lock(&dq_data_lock);
1053                if (!IS_NOQUOTA(inode)) {
1054                        if (unlikely(inode_get_rsv_space(inode) > 0))
1055                                reserved = 1;
1056                        remove_inode_dquot_ref(inode, type, tofree_head);
1057                }
1058                spin_unlock(&dq_data_lock);
1059        }
1060        spin_unlock(&sb->s_inode_list_lock);
1061#ifdef CONFIG_QUOTA_DEBUG
1062        if (reserved) {
1063                printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1064                        " was disabled thus quota information is probably "
1065                        "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1066        }
1067#endif
1068}
1069
1070/* Gather all references from inodes and drop them */
1071static void drop_dquot_ref(struct super_block *sb, int type)
1072{
1073        LIST_HEAD(tofree_head);
1074
1075        if (sb->dq_op) {
1076                remove_dquot_ref(sb, type, &tofree_head);
1077                synchronize_srcu(&dquot_srcu);
1078                put_dquot_list(&tofree_head);
1079        }
1080}
1081
1082static inline void dquot_incr_inodes(struct dquot *dquot, qsize_t number)
1083{
1084        dquot->dq_dqb.dqb_curinodes += number;
1085}
1086
1087static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
1088{
1089        dquot->dq_dqb.dqb_curspace += number;
1090}
1091
1092static inline void dquot_resv_space(struct dquot *dquot, qsize_t number)
1093{
1094        dquot->dq_dqb.dqb_rsvspace += number;
1095}
1096
1097/*
1098 * Claim reserved quota space
1099 */
1100static void dquot_claim_reserved_space(struct dquot *dquot, qsize_t number)
1101{
1102        if (dquot->dq_dqb.dqb_rsvspace < number) {
1103                WARN_ON_ONCE(1);
1104                number = dquot->dq_dqb.dqb_rsvspace;
1105        }
1106        dquot->dq_dqb.dqb_curspace += number;
1107        dquot->dq_dqb.dqb_rsvspace -= number;
1108}
1109
1110static void dquot_reclaim_reserved_space(struct dquot *dquot, qsize_t number)
1111{
1112        if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1113                number = dquot->dq_dqb.dqb_curspace;
1114        dquot->dq_dqb.dqb_rsvspace += number;
1115        dquot->dq_dqb.dqb_curspace -= number;
1116}
1117
1118static inline
1119void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1120{
1121        if (dquot->dq_dqb.dqb_rsvspace >= number)
1122                dquot->dq_dqb.dqb_rsvspace -= number;
1123        else {
1124                WARN_ON_ONCE(1);
1125                dquot->dq_dqb.dqb_rsvspace = 0;
1126        }
1127        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1128            dquot->dq_dqb.dqb_bsoftlimit)
1129                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1130        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1131}
1132
1133static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1134{
1135        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1136            dquot->dq_dqb.dqb_curinodes >= number)
1137                dquot->dq_dqb.dqb_curinodes -= number;
1138        else
1139                dquot->dq_dqb.dqb_curinodes = 0;
1140        if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1141                dquot->dq_dqb.dqb_itime = (time64_t) 0;
1142        clear_bit(DQ_INODES_B, &dquot->dq_flags);
1143}
1144
1145static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1146{
1147        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1148            dquot->dq_dqb.dqb_curspace >= number)
1149                dquot->dq_dqb.dqb_curspace -= number;
1150        else
1151                dquot->dq_dqb.dqb_curspace = 0;
1152        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1153            dquot->dq_dqb.dqb_bsoftlimit)
1154                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1155        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1156}
1157
1158struct dquot_warn {
1159        struct super_block *w_sb;
1160        struct kqid w_dq_id;
1161        short w_type;
1162};
1163
1164static int warning_issued(struct dquot *dquot, const int warntype)
1165{
1166        int flag = (warntype == QUOTA_NL_BHARDWARN ||
1167                warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1168                ((warntype == QUOTA_NL_IHARDWARN ||
1169                warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1170
1171        if (!flag)
1172                return 0;
1173        return test_and_set_bit(flag, &dquot->dq_flags);
1174}
1175
1176#ifdef CONFIG_PRINT_QUOTA_WARNING
1177static int flag_print_warnings = 1;
1178
1179static int need_print_warning(struct dquot_warn *warn)
1180{
1181        if (!flag_print_warnings)
1182                return 0;
1183
1184        switch (warn->w_dq_id.type) {
1185                case USRQUOTA:
1186                        return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1187                case GRPQUOTA:
1188                        return in_group_p(warn->w_dq_id.gid);
1189                case PRJQUOTA:
1190                        return 1;
1191        }
1192        return 0;
1193}
1194
1195/* Print warning to user which exceeded quota */
1196static void print_warning(struct dquot_warn *warn)
1197{
1198        char *msg = NULL;
1199        struct tty_struct *tty;
1200        int warntype = warn->w_type;
1201
1202        if (warntype == QUOTA_NL_IHARDBELOW ||
1203            warntype == QUOTA_NL_ISOFTBELOW ||
1204            warntype == QUOTA_NL_BHARDBELOW ||
1205            warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1206                return;
1207
1208        tty = get_current_tty();
1209        if (!tty)
1210                return;
1211        tty_write_message(tty, warn->w_sb->s_id);
1212        if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1213                tty_write_message(tty, ": warning, ");
1214        else
1215                tty_write_message(tty, ": write failed, ");
1216        tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1217        switch (warntype) {
1218                case QUOTA_NL_IHARDWARN:
1219                        msg = " file limit reached.\r\n";
1220                        break;
1221                case QUOTA_NL_ISOFTLONGWARN:
1222                        msg = " file quota exceeded too long.\r\n";
1223                        break;
1224                case QUOTA_NL_ISOFTWARN:
1225                        msg = " file quota exceeded.\r\n";
1226                        break;
1227                case QUOTA_NL_BHARDWARN:
1228                        msg = " block limit reached.\r\n";
1229                        break;
1230                case QUOTA_NL_BSOFTLONGWARN:
1231                        msg = " block quota exceeded too long.\r\n";
1232                        break;
1233                case QUOTA_NL_BSOFTWARN:
1234                        msg = " block quota exceeded.\r\n";
1235                        break;
1236        }
1237        tty_write_message(tty, msg);
1238        tty_kref_put(tty);
1239}
1240#endif
1241
1242static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1243                            int warntype)
1244{
1245        if (warning_issued(dquot, warntype))
1246                return;
1247        warn->w_type = warntype;
1248        warn->w_sb = dquot->dq_sb;
1249        warn->w_dq_id = dquot->dq_id;
1250}
1251
1252/*
1253 * Write warnings to the console and send warning messages over netlink.
1254 *
1255 * Note that this function can call into tty and networking code.
1256 */
1257static void flush_warnings(struct dquot_warn *warn)
1258{
1259        int i;
1260
1261        for (i = 0; i < MAXQUOTAS; i++) {
1262                if (warn[i].w_type == QUOTA_NL_NOWARN)
1263                        continue;
1264#ifdef CONFIG_PRINT_QUOTA_WARNING
1265                print_warning(&warn[i]);
1266#endif
1267                quota_send_warning(warn[i].w_dq_id,
1268                                   warn[i].w_sb->s_dev, warn[i].w_type);
1269        }
1270}
1271
1272static int ignore_hardlimit(struct dquot *dquot)
1273{
1274        struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1275
1276        return capable(CAP_SYS_RESOURCE) &&
1277               (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1278                !(info->dqi_flags & DQF_ROOT_SQUASH));
1279}
1280
1281/* needs dq_data_lock */
1282static int check_idq(struct dquot *dquot, qsize_t inodes,
1283                     struct dquot_warn *warn)
1284{
1285        qsize_t newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1286
1287        if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1288            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1289                return 0;
1290
1291        if (dquot->dq_dqb.dqb_ihardlimit &&
1292            newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1293            !ignore_hardlimit(dquot)) {
1294                prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1295                return -EDQUOT;
1296        }
1297
1298        if (dquot->dq_dqb.dqb_isoftlimit &&
1299            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1300            dquot->dq_dqb.dqb_itime &&
1301            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1302            !ignore_hardlimit(dquot)) {
1303                prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1304                return -EDQUOT;
1305        }
1306
1307        if (dquot->dq_dqb.dqb_isoftlimit &&
1308            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1309            dquot->dq_dqb.dqb_itime == 0) {
1310                prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1311                dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1312                    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1313        }
1314
1315        return 0;
1316}
1317
1318/* needs dq_data_lock */
1319static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc,
1320                     struct dquot_warn *warn)
1321{
1322        qsize_t tspace;
1323        struct super_block *sb = dquot->dq_sb;
1324
1325        if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1326            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1327                return 0;
1328
1329        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1330                + space;
1331
1332        if (dquot->dq_dqb.dqb_bhardlimit &&
1333            tspace > dquot->dq_dqb.dqb_bhardlimit &&
1334            !ignore_hardlimit(dquot)) {
1335                if (!prealloc)
1336                        prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1337                return -EDQUOT;
1338        }
1339
1340        if (dquot->dq_dqb.dqb_bsoftlimit &&
1341            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1342            dquot->dq_dqb.dqb_btime &&
1343            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1344            !ignore_hardlimit(dquot)) {
1345                if (!prealloc)
1346                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1347                return -EDQUOT;
1348        }
1349
1350        if (dquot->dq_dqb.dqb_bsoftlimit &&
1351            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1352            dquot->dq_dqb.dqb_btime == 0) {
1353                if (!prealloc) {
1354                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1355                        dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1356                            sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1357                }
1358                else
1359                        /*
1360                         * We don't allow preallocation to exceed softlimit so exceeding will
1361                         * be always printed
1362                         */
1363                        return -EDQUOT;
1364        }
1365
1366        return 0;
1367}
1368
1369static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1370{
1371        qsize_t newinodes;
1372
1373        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1374            dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1375            !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1376                return QUOTA_NL_NOWARN;
1377
1378        newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1379        if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1380                return QUOTA_NL_ISOFTBELOW;
1381        if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1382            newinodes < dquot->dq_dqb.dqb_ihardlimit)
1383                return QUOTA_NL_IHARDBELOW;
1384        return QUOTA_NL_NOWARN;
1385}
1386
1387static int info_bdq_free(struct dquot *dquot, qsize_t space)
1388{
1389        qsize_t tspace;
1390
1391        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1392
1393        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1394            tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1395                return QUOTA_NL_NOWARN;
1396
1397        if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1398                return QUOTA_NL_BSOFTBELOW;
1399        if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1400            tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1401                return QUOTA_NL_BHARDBELOW;
1402        return QUOTA_NL_NOWARN;
1403}
1404
1405static int dquot_active(const struct inode *inode)
1406{
1407        struct super_block *sb = inode->i_sb;
1408
1409        if (IS_NOQUOTA(inode))
1410                return 0;
1411        return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1412}
1413
1414/*
1415 * Initialize quota pointers in inode
1416 *
1417 * It is better to call this function outside of any transaction as it
1418 * might need a lot of space in journal for dquot structure allocation.
1419 */
1420static int __dquot_initialize(struct inode *inode, int type)
1421{
1422        int cnt, init_needed = 0;
1423        struct dquot **dquots, *got[MAXQUOTAS] = {};
1424        struct super_block *sb = inode->i_sb;
1425        qsize_t rsv;
1426        int ret = 0;
1427
1428        if (!dquot_active(inode))
1429                return 0;
1430
1431        dquots = i_dquot(inode);
1432
1433        /* First get references to structures we might need. */
1434        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1435                struct kqid qid;
1436                kprojid_t projid;
1437                int rc;
1438                struct dquot *dquot;
1439
1440                if (type != -1 && cnt != type)
1441                        continue;
1442                /*
1443                 * The i_dquot should have been initialized in most cases,
1444                 * we check it without locking here to avoid unnecessary
1445                 * dqget()/dqput() calls.
1446                 */
1447                if (dquots[cnt])
1448                        continue;
1449
1450                if (!sb_has_quota_active(sb, cnt))
1451                        continue;
1452
1453                init_needed = 1;
1454
1455                switch (cnt) {
1456                case USRQUOTA:
1457                        qid = make_kqid_uid(inode->i_uid);
1458                        break;
1459                case GRPQUOTA:
1460                        qid = make_kqid_gid(inode->i_gid);
1461                        break;
1462                case PRJQUOTA:
1463                        rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1464                        if (rc)
1465                                continue;
1466                        qid = make_kqid_projid(projid);
1467                        break;
1468                }
1469                dquot = dqget(sb, qid);
1470                if (IS_ERR(dquot)) {
1471                        /* We raced with somebody turning quotas off... */
1472                        if (PTR_ERR(dquot) != -ESRCH) {
1473                                ret = PTR_ERR(dquot);
1474                                goto out_put;
1475                        }
1476                        dquot = NULL;
1477                }
1478                got[cnt] = dquot;
1479        }
1480
1481        /* All required i_dquot has been initialized */
1482        if (!init_needed)
1483                return 0;
1484
1485        spin_lock(&dq_data_lock);
1486        if (IS_NOQUOTA(inode))
1487                goto out_lock;
1488        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1489                if (type != -1 && cnt != type)
1490                        continue;
1491                /* Avoid races with quotaoff() */
1492                if (!sb_has_quota_active(sb, cnt))
1493                        continue;
1494                /* We could race with quotaon or dqget() could have failed */
1495                if (!got[cnt])
1496                        continue;
1497                if (!dquots[cnt]) {
1498                        dquots[cnt] = got[cnt];
1499                        got[cnt] = NULL;
1500                        /*
1501                         * Make quota reservation system happy if someone
1502                         * did a write before quota was turned on
1503                         */
1504                        rsv = inode_get_rsv_space(inode);
1505                        if (unlikely(rsv))
1506                                dquot_resv_space(dquots[cnt], rsv);
1507                }
1508        }
1509out_lock:
1510        spin_unlock(&dq_data_lock);
1511out_put:
1512        /* Drop unused references */
1513        dqput_all(got);
1514
1515        return ret;
1516}
1517
1518int dquot_initialize(struct inode *inode)
1519{
1520        return __dquot_initialize(inode, -1);
1521}
1522EXPORT_SYMBOL(dquot_initialize);
1523
1524bool dquot_initialize_needed(struct inode *inode)
1525{
1526        struct dquot **dquots;
1527        int i;
1528
1529        if (!dquot_active(inode))
1530                return false;
1531
1532        dquots = i_dquot(inode);
1533        for (i = 0; i < MAXQUOTAS; i++)
1534                if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1535                        return true;
1536        return false;
1537}
1538EXPORT_SYMBOL(dquot_initialize_needed);
1539
1540/*
1541 * Release all quotas referenced by inode.
1542 *
1543 * This function only be called on inode free or converting
1544 * a file to quota file, no other users for the i_dquot in
1545 * both cases, so we needn't call synchronize_srcu() after
1546 * clearing i_dquot.
1547 */
1548static void __dquot_drop(struct inode *inode)
1549{
1550        int cnt;
1551        struct dquot **dquots = i_dquot(inode);
1552        struct dquot *put[MAXQUOTAS];
1553
1554        spin_lock(&dq_data_lock);
1555        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1556                put[cnt] = dquots[cnt];
1557                dquots[cnt] = NULL;
1558        }
1559        spin_unlock(&dq_data_lock);
1560        dqput_all(put);
1561}
1562
1563void dquot_drop(struct inode *inode)
1564{
1565        struct dquot * const *dquots;
1566        int cnt;
1567
1568        if (IS_NOQUOTA(inode))
1569                return;
1570
1571        /*
1572         * Test before calling to rule out calls from proc and such
1573         * where we are not allowed to block. Note that this is
1574         * actually reliable test even without the lock - the caller
1575         * must assure that nobody can come after the DQUOT_DROP and
1576         * add quota pointers back anyway.
1577         */
1578        dquots = i_dquot(inode);
1579        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1580                if (dquots[cnt])
1581                        break;
1582        }
1583
1584        if (cnt < MAXQUOTAS)
1585                __dquot_drop(inode);
1586}
1587EXPORT_SYMBOL(dquot_drop);
1588
1589/*
1590 * inode_reserved_space is managed internally by quota, and protected by
1591 * i_lock similar to i_blocks+i_bytes.
1592 */
1593static qsize_t *inode_reserved_space(struct inode * inode)
1594{
1595        /* Filesystem must explicitly define it's own method in order to use
1596         * quota reservation interface */
1597        BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1598        return inode->i_sb->dq_op->get_reserved_space(inode);
1599}
1600
1601void inode_add_rsv_space(struct inode *inode, qsize_t number)
1602{
1603        spin_lock(&inode->i_lock);
1604        *inode_reserved_space(inode) += number;
1605        spin_unlock(&inode->i_lock);
1606}
1607EXPORT_SYMBOL(inode_add_rsv_space);
1608
1609void inode_claim_rsv_space(struct inode *inode, qsize_t number)
1610{
1611        spin_lock(&inode->i_lock);
1612        *inode_reserved_space(inode) -= number;
1613        __inode_add_bytes(inode, number);
1614        spin_unlock(&inode->i_lock);
1615}
1616EXPORT_SYMBOL(inode_claim_rsv_space);
1617
1618void inode_reclaim_rsv_space(struct inode *inode, qsize_t number)
1619{
1620        spin_lock(&inode->i_lock);
1621        *inode_reserved_space(inode) += number;
1622        __inode_sub_bytes(inode, number);
1623        spin_unlock(&inode->i_lock);
1624}
1625EXPORT_SYMBOL(inode_reclaim_rsv_space);
1626
1627void inode_sub_rsv_space(struct inode *inode, qsize_t number)
1628{
1629        spin_lock(&inode->i_lock);
1630        *inode_reserved_space(inode) -= number;
1631        spin_unlock(&inode->i_lock);
1632}
1633EXPORT_SYMBOL(inode_sub_rsv_space);
1634
1635static qsize_t inode_get_rsv_space(struct inode *inode)
1636{
1637        qsize_t ret;
1638
1639        if (!inode->i_sb->dq_op->get_reserved_space)
1640                return 0;
1641        spin_lock(&inode->i_lock);
1642        ret = *inode_reserved_space(inode);
1643        spin_unlock(&inode->i_lock);
1644        return ret;
1645}
1646
1647static void inode_incr_space(struct inode *inode, qsize_t number,
1648                                int reserve)
1649{
1650        if (reserve)
1651                inode_add_rsv_space(inode, number);
1652        else
1653                inode_add_bytes(inode, number);
1654}
1655
1656static void inode_decr_space(struct inode *inode, qsize_t number, int reserve)
1657{
1658        if (reserve)
1659                inode_sub_rsv_space(inode, number);
1660        else
1661                inode_sub_bytes(inode, number);
1662}
1663
1664/*
1665 * This functions updates i_blocks+i_bytes fields and quota information
1666 * (together with appropriate checks).
1667 *
1668 * NOTE: We absolutely rely on the fact that caller dirties the inode
1669 * (usually helpers in quotaops.h care about this) and holds a handle for
1670 * the current transaction so that dquot write and inode write go into the
1671 * same transaction.
1672 */
1673
1674/*
1675 * This operation can block, but only after everything is updated
1676 */
1677int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1678{
1679        int cnt, ret = 0, index;
1680        struct dquot_warn warn[MAXQUOTAS];
1681        int reserve = flags & DQUOT_SPACE_RESERVE;
1682        struct dquot **dquots;
1683
1684        if (!dquot_active(inode)) {
1685                inode_incr_space(inode, number, reserve);
1686                goto out;
1687        }
1688
1689        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1690                warn[cnt].w_type = QUOTA_NL_NOWARN;
1691
1692        dquots = i_dquot(inode);
1693        index = srcu_read_lock(&dquot_srcu);
1694        spin_lock(&dq_data_lock);
1695        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1696                if (!dquots[cnt])
1697                        continue;
1698                ret = check_bdq(dquots[cnt], number,
1699                                !(flags & DQUOT_SPACE_WARN), &warn[cnt]);
1700                if (ret && !(flags & DQUOT_SPACE_NOFAIL)) {
1701                        spin_unlock(&dq_data_lock);
1702                        goto out_flush_warn;
1703                }
1704        }
1705        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1706                if (!dquots[cnt])
1707                        continue;
1708                if (reserve)
1709                        dquot_resv_space(dquots[cnt], number);
1710                else
1711                        dquot_incr_space(dquots[cnt], number);
1712        }
1713        inode_incr_space(inode, number, reserve);
1714        spin_unlock(&dq_data_lock);
1715
1716        if (reserve)
1717                goto out_flush_warn;
1718        mark_all_dquot_dirty(dquots);
1719out_flush_warn:
1720        srcu_read_unlock(&dquot_srcu, index);
1721        flush_warnings(warn);
1722out:
1723        return ret;
1724}
1725EXPORT_SYMBOL(__dquot_alloc_space);
1726
1727/*
1728 * This operation can block, but only after everything is updated
1729 */
1730int dquot_alloc_inode(struct inode *inode)
1731{
1732        int cnt, ret = 0, index;
1733        struct dquot_warn warn[MAXQUOTAS];
1734        struct dquot * const *dquots;
1735
1736        if (!dquot_active(inode))
1737                return 0;
1738        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1739                warn[cnt].w_type = QUOTA_NL_NOWARN;
1740
1741        dquots = i_dquot(inode);
1742        index = srcu_read_lock(&dquot_srcu);
1743        spin_lock(&dq_data_lock);
1744        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1745                if (!dquots[cnt])
1746                        continue;
1747                ret = check_idq(dquots[cnt], 1, &warn[cnt]);
1748                if (ret)
1749                        goto warn_put_all;
1750        }
1751
1752        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1753                if (!dquots[cnt])
1754                        continue;
1755                dquot_incr_inodes(dquots[cnt], 1);
1756        }
1757
1758warn_put_all:
1759        spin_unlock(&dq_data_lock);
1760        if (ret == 0)
1761                mark_all_dquot_dirty(dquots);
1762        srcu_read_unlock(&dquot_srcu, index);
1763        flush_warnings(warn);
1764        return ret;
1765}
1766EXPORT_SYMBOL(dquot_alloc_inode);
1767
1768/*
1769 * Convert in-memory reserved quotas to real consumed quotas
1770 */
1771int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1772{
1773        struct dquot **dquots;
1774        int cnt, index;
1775
1776        if (!dquot_active(inode)) {
1777                inode_claim_rsv_space(inode, number);
1778                return 0;
1779        }
1780
1781        dquots = i_dquot(inode);
1782        index = srcu_read_lock(&dquot_srcu);
1783        spin_lock(&dq_data_lock);
1784        /* Claim reserved quotas to allocated quotas */
1785        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1786                if (dquots[cnt])
1787                        dquot_claim_reserved_space(dquots[cnt], number);
1788        }
1789        /* Update inode bytes */
1790        inode_claim_rsv_space(inode, number);
1791        spin_unlock(&dq_data_lock);
1792        mark_all_dquot_dirty(dquots);
1793        srcu_read_unlock(&dquot_srcu, index);
1794        return 0;
1795}
1796EXPORT_SYMBOL(dquot_claim_space_nodirty);
1797
1798/*
1799 * Convert allocated space back to in-memory reserved quotas
1800 */
1801void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1802{
1803        struct dquot **dquots;
1804        int cnt, index;
1805
1806        if (!dquot_active(inode)) {
1807                inode_reclaim_rsv_space(inode, number);
1808                return;
1809        }
1810
1811        dquots = i_dquot(inode);
1812        index = srcu_read_lock(&dquot_srcu);
1813        spin_lock(&dq_data_lock);
1814        /* Claim reserved quotas to allocated quotas */
1815        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1816                if (dquots[cnt])
1817                        dquot_reclaim_reserved_space(dquots[cnt], number);
1818        }
1819        /* Update inode bytes */
1820        inode_reclaim_rsv_space(inode, number);
1821        spin_unlock(&dq_data_lock);
1822        mark_all_dquot_dirty(dquots);
1823        srcu_read_unlock(&dquot_srcu, index);
1824        return;
1825}
1826EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1827
1828/*
1829 * This operation can block, but only after everything is updated
1830 */
1831void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1832{
1833        unsigned int cnt;
1834        struct dquot_warn warn[MAXQUOTAS];
1835        struct dquot **dquots;
1836        int reserve = flags & DQUOT_SPACE_RESERVE, index;
1837
1838        if (!dquot_active(inode)) {
1839                inode_decr_space(inode, number, reserve);
1840                return;
1841        }
1842
1843        dquots = i_dquot(inode);
1844        index = srcu_read_lock(&dquot_srcu);
1845        spin_lock(&dq_data_lock);
1846        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1847                int wtype;
1848
1849                warn[cnt].w_type = QUOTA_NL_NOWARN;
1850                if (!dquots[cnt])
1851                        continue;
1852                wtype = info_bdq_free(dquots[cnt], number);
1853                if (wtype != QUOTA_NL_NOWARN)
1854                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1855                if (reserve)
1856                        dquot_free_reserved_space(dquots[cnt], number);
1857                else
1858                        dquot_decr_space(dquots[cnt], number);
1859        }
1860        inode_decr_space(inode, number, reserve);
1861        spin_unlock(&dq_data_lock);
1862
1863        if (reserve)
1864                goto out_unlock;
1865        mark_all_dquot_dirty(dquots);
1866out_unlock:
1867        srcu_read_unlock(&dquot_srcu, index);
1868        flush_warnings(warn);
1869}
1870EXPORT_SYMBOL(__dquot_free_space);
1871
1872/*
1873 * This operation can block, but only after everything is updated
1874 */
1875void dquot_free_inode(struct inode *inode)
1876{
1877        unsigned int cnt;
1878        struct dquot_warn warn[MAXQUOTAS];
1879        struct dquot * const *dquots;
1880        int index;
1881
1882        if (!dquot_active(inode))
1883                return;
1884
1885        dquots = i_dquot(inode);
1886        index = srcu_read_lock(&dquot_srcu);
1887        spin_lock(&dq_data_lock);
1888        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1889                int wtype;
1890
1891                warn[cnt].w_type = QUOTA_NL_NOWARN;
1892                if (!dquots[cnt])
1893                        continue;
1894                wtype = info_idq_free(dquots[cnt], 1);
1895                if (wtype != QUOTA_NL_NOWARN)
1896                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1897                dquot_decr_inodes(dquots[cnt], 1);
1898        }
1899        spin_unlock(&dq_data_lock);
1900        mark_all_dquot_dirty(dquots);
1901        srcu_read_unlock(&dquot_srcu, index);
1902        flush_warnings(warn);
1903}
1904EXPORT_SYMBOL(dquot_free_inode);
1905
1906/*
1907 * Transfer the number of inode and blocks from one diskquota to an other.
1908 * On success, dquot references in transfer_to are consumed and references
1909 * to original dquots that need to be released are placed there. On failure,
1910 * references are kept untouched.
1911 *
1912 * This operation can block, but only after everything is updated
1913 * A transaction must be started when entering this function.
1914 *
1915 * We are holding reference on transfer_from & transfer_to, no need to
1916 * protect them by srcu_read_lock().
1917 */
1918int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1919{
1920        qsize_t space, cur_space;
1921        qsize_t rsv_space = 0;
1922        qsize_t inode_usage = 1;
1923        struct dquot *transfer_from[MAXQUOTAS] = {};
1924        int cnt, ret = 0;
1925        char is_valid[MAXQUOTAS] = {};
1926        struct dquot_warn warn_to[MAXQUOTAS];
1927        struct dquot_warn warn_from_inodes[MAXQUOTAS];
1928        struct dquot_warn warn_from_space[MAXQUOTAS];
1929
1930        if (IS_NOQUOTA(inode))
1931                return 0;
1932
1933        if (inode->i_sb->dq_op->get_inode_usage) {
1934                ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1935                if (ret)
1936                        return ret;
1937        }
1938
1939        /* Initialize the arrays */
1940        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1941                warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1942                warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1943                warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1944        }
1945
1946        spin_lock(&dq_data_lock);
1947        if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1948                spin_unlock(&dq_data_lock);
1949                return 0;
1950        }
1951        cur_space = inode_get_bytes(inode);
1952        rsv_space = inode_get_rsv_space(inode);
1953        space = cur_space + rsv_space;
1954        /* Build the transfer_from list and check the limits */
1955        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1956                /*
1957                 * Skip changes for same uid or gid or for turned off quota-type.
1958                 */
1959                if (!transfer_to[cnt])
1960                        continue;
1961                /* Avoid races with quotaoff() */
1962                if (!sb_has_quota_active(inode->i_sb, cnt))
1963                        continue;
1964                is_valid[cnt] = 1;
1965                transfer_from[cnt] = i_dquot(inode)[cnt];
1966                ret = check_idq(transfer_to[cnt], inode_usage, &warn_to[cnt]);
1967                if (ret)
1968                        goto over_quota;
1969                ret = check_bdq(transfer_to[cnt], space, 0, &warn_to[cnt]);
1970                if (ret)
1971                        goto over_quota;
1972        }
1973
1974        /*
1975         * Finally perform the needed transfer from transfer_from to transfer_to
1976         */
1977        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1978                if (!is_valid[cnt])
1979                        continue;
1980                /* Due to IO error we might not have transfer_from[] structure */
1981                if (transfer_from[cnt]) {
1982                        int wtype;
1983                        wtype = info_idq_free(transfer_from[cnt], inode_usage);
1984                        if (wtype != QUOTA_NL_NOWARN)
1985                                prepare_warning(&warn_from_inodes[cnt],
1986                                                transfer_from[cnt], wtype);
1987                        wtype = info_bdq_free(transfer_from[cnt], space);
1988                        if (wtype != QUOTA_NL_NOWARN)
1989                                prepare_warning(&warn_from_space[cnt],
1990                                                transfer_from[cnt], wtype);
1991                        dquot_decr_inodes(transfer_from[cnt], inode_usage);
1992                        dquot_decr_space(transfer_from[cnt], cur_space);
1993                        dquot_free_reserved_space(transfer_from[cnt],
1994                                                  rsv_space);
1995                }
1996
1997                dquot_incr_inodes(transfer_to[cnt], inode_usage);
1998                dquot_incr_space(transfer_to[cnt], cur_space);
1999                dquot_resv_space(transfer_to[cnt], rsv_space);
2000
2001                i_dquot(inode)[cnt] = transfer_to[cnt];
2002        }
2003        spin_unlock(&dq_data_lock);
2004
2005        mark_all_dquot_dirty(transfer_from);
2006        mark_all_dquot_dirty(transfer_to);
2007        flush_warnings(warn_to);
2008        flush_warnings(warn_from_inodes);
2009        flush_warnings(warn_from_space);
2010        /* Pass back references to put */
2011        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2012                if (is_valid[cnt])
2013                        transfer_to[cnt] = transfer_from[cnt];
2014        return 0;
2015over_quota:
2016        spin_unlock(&dq_data_lock);
2017        flush_warnings(warn_to);
2018        return ret;
2019}
2020EXPORT_SYMBOL(__dquot_transfer);
2021
2022/* Wrapper for transferring ownership of an inode for uid/gid only
2023 * Called from FSXXX_setattr()
2024 */
2025int dquot_transfer(struct inode *inode, struct iattr *iattr)
2026{
2027        struct dquot *transfer_to[MAXQUOTAS] = {};
2028        struct dquot *dquot;
2029        struct super_block *sb = inode->i_sb;
2030        int ret;
2031
2032        if (!dquot_active(inode))
2033                return 0;
2034
2035        if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2036                dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2037                if (IS_ERR(dquot)) {
2038                        if (PTR_ERR(dquot) != -ESRCH) {
2039                                ret = PTR_ERR(dquot);
2040                                goto out_put;
2041                        }
2042                        dquot = NULL;
2043                }
2044                transfer_to[USRQUOTA] = dquot;
2045        }
2046        if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2047                dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2048                if (IS_ERR(dquot)) {
2049                        if (PTR_ERR(dquot) != -ESRCH) {
2050                                ret = PTR_ERR(dquot);
2051                                goto out_put;
2052                        }
2053                        dquot = NULL;
2054                }
2055                transfer_to[GRPQUOTA] = dquot;
2056        }
2057        ret = __dquot_transfer(inode, transfer_to);
2058out_put:
2059        dqput_all(transfer_to);
2060        return ret;
2061}
2062EXPORT_SYMBOL(dquot_transfer);
2063
2064/*
2065 * Write info of quota file to disk
2066 */
2067int dquot_commit_info(struct super_block *sb, int type)
2068{
2069        int ret;
2070        struct quota_info *dqopt = sb_dqopt(sb);
2071
2072        mutex_lock(&dqopt->dqio_mutex);
2073        ret = dqopt->ops[type]->write_file_info(sb, type);
2074        mutex_unlock(&dqopt->dqio_mutex);
2075        return ret;
2076}
2077EXPORT_SYMBOL(dquot_commit_info);
2078
2079int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2080{
2081        struct quota_info *dqopt = sb_dqopt(sb);
2082        int err;
2083
2084        if (!sb_has_quota_active(sb, qid->type))
2085                return -ESRCH;
2086        if (!dqopt->ops[qid->type]->get_next_id)
2087                return -ENOSYS;
2088        mutex_lock(&dqopt->dqio_mutex);
2089        err = dqopt->ops[qid->type]->get_next_id(sb, qid);
2090        mutex_unlock(&dqopt->dqio_mutex);
2091        return err;
2092}
2093EXPORT_SYMBOL(dquot_get_next_id);
2094
2095/*
2096 * Definitions of diskquota operations.
2097 */
2098const struct dquot_operations dquot_operations = {
2099        .write_dquot    = dquot_commit,
2100        .acquire_dquot  = dquot_acquire,
2101        .release_dquot  = dquot_release,
2102        .mark_dirty     = dquot_mark_dquot_dirty,
2103        .write_info     = dquot_commit_info,
2104        .alloc_dquot    = dquot_alloc,
2105        .destroy_dquot  = dquot_destroy,
2106        .get_next_id    = dquot_get_next_id,
2107};
2108EXPORT_SYMBOL(dquot_operations);
2109
2110/*
2111 * Generic helper for ->open on filesystems supporting disk quotas.
2112 */
2113int dquot_file_open(struct inode *inode, struct file *file)
2114{
2115        int error;
2116
2117        error = generic_file_open(inode, file);
2118        if (!error && (file->f_mode & FMODE_WRITE))
2119                dquot_initialize(inode);
2120        return error;
2121}
2122EXPORT_SYMBOL(dquot_file_open);
2123
2124/*
2125 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2126 */
2127int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2128{
2129        int cnt, ret = 0;
2130        struct quota_info *dqopt = sb_dqopt(sb);
2131        struct inode *toputinode[MAXQUOTAS];
2132
2133        /* s_umount should be held in exclusive mode */
2134        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2135                up_read(&sb->s_umount);
2136
2137        /* Cannot turn off usage accounting without turning off limits, or
2138         * suspend quotas and simultaneously turn quotas off. */
2139        if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2140            || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2141            DQUOT_USAGE_ENABLED)))
2142                return -EINVAL;
2143
2144        /*
2145         * Skip everything if there's nothing to do. We have to do this because
2146         * sometimes we are called when fill_super() failed and calling
2147         * sync_fs() in such cases does no good.
2148         */
2149        if (!sb_any_quota_loaded(sb))
2150                return 0;
2151
2152        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2153                toputinode[cnt] = NULL;
2154                if (type != -1 && cnt != type)
2155                        continue;
2156                if (!sb_has_quota_loaded(sb, cnt))
2157                        continue;
2158
2159                if (flags & DQUOT_SUSPENDED) {
2160                        spin_lock(&dq_state_lock);
2161                        dqopt->flags |=
2162                                dquot_state_flag(DQUOT_SUSPENDED, cnt);
2163                        spin_unlock(&dq_state_lock);
2164                } else {
2165                        spin_lock(&dq_state_lock);
2166                        dqopt->flags &= ~dquot_state_flag(flags, cnt);
2167                        /* Turning off suspended quotas? */
2168                        if (!sb_has_quota_loaded(sb, cnt) &&
2169                            sb_has_quota_suspended(sb, cnt)) {
2170                                dqopt->flags &= ~dquot_state_flag(
2171                                                        DQUOT_SUSPENDED, cnt);
2172                                spin_unlock(&dq_state_lock);
2173                                iput(dqopt->files[cnt]);
2174                                dqopt->files[cnt] = NULL;
2175                                continue;
2176                        }
2177                        spin_unlock(&dq_state_lock);
2178                }
2179
2180                /* We still have to keep quota loaded? */
2181                if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2182                        continue;
2183
2184                /* Note: these are blocking operations */
2185                drop_dquot_ref(sb, cnt);
2186                invalidate_dquots(sb, cnt);
2187                /*
2188                 * Now all dquots should be invalidated, all writes done so we
2189                 * should be only users of the info. No locks needed.
2190                 */
2191                if (info_dirty(&dqopt->info[cnt]))
2192                        sb->dq_op->write_info(sb, cnt);
2193                if (dqopt->ops[cnt]->free_file_info)
2194                        dqopt->ops[cnt]->free_file_info(sb, cnt);
2195                put_quota_format(dqopt->info[cnt].dqi_format);
2196
2197                toputinode[cnt] = dqopt->files[cnt];
2198                if (!sb_has_quota_loaded(sb, cnt))
2199                        dqopt->files[cnt] = NULL;
2200                dqopt->info[cnt].dqi_flags = 0;
2201                dqopt->info[cnt].dqi_igrace = 0;
2202                dqopt->info[cnt].dqi_bgrace = 0;
2203                dqopt->ops[cnt] = NULL;
2204        }
2205
2206        /* Skip syncing and setting flags if quota files are hidden */
2207        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2208                goto put_inodes;
2209
2210        /* Sync the superblock so that buffers with quota data are written to
2211         * disk (and so userspace sees correct data afterwards). */
2212        if (sb->s_op->sync_fs)
2213                sb->s_op->sync_fs(sb, 1);
2214        sync_blockdev(sb->s_bdev);
2215        /* Now the quota files are just ordinary files and we can set the
2216         * inode flags back. Moreover we discard the pagecache so that
2217         * userspace sees the writes we did bypassing the pagecache. We
2218         * must also discard the blockdev buffers so that we see the
2219         * changes done by userspace on the next quotaon() */
2220        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2221                /* This can happen when suspending quotas on remount-ro... */
2222                if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2223                        inode_lock(toputinode[cnt]);
2224                        toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2225                        truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2226                        inode_unlock(toputinode[cnt]);
2227                        mark_inode_dirty_sync(toputinode[cnt]);
2228                }
2229        if (sb->s_bdev)
2230                invalidate_bdev(sb->s_bdev);
2231put_inodes:
2232        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2233                if (toputinode[cnt]) {
2234                        /* On remount RO, we keep the inode pointer so that we
2235                         * can reenable quota on the subsequent remount RW. We
2236                         * have to check 'flags' variable and not use sb_has_
2237                         * function because another quotaon / quotaoff could
2238                         * change global state before we got here. We refuse
2239                         * to suspend quotas when there is pending delete on
2240                         * the quota file... */
2241                        if (!(flags & DQUOT_SUSPENDED))
2242                                iput(toputinode[cnt]);
2243                        else if (!toputinode[cnt]->i_nlink)
2244                                ret = -EBUSY;
2245                }
2246        return ret;
2247}
2248EXPORT_SYMBOL(dquot_disable);
2249
2250int dquot_quota_off(struct super_block *sb, int type)
2251{
2252        return dquot_disable(sb, type,
2253                             DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2254}
2255EXPORT_SYMBOL(dquot_quota_off);
2256
2257/*
2258 *      Turn quotas on on a device
2259 */
2260
2261/*
2262 * Helper function to turn quotas on when we already have the inode of
2263 * quota file and no quota information is loaded.
2264 */
2265static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2266        unsigned int flags)
2267{
2268        struct quota_format_type *fmt = find_quota_format(format_id);
2269        struct super_block *sb = inode->i_sb;
2270        struct quota_info *dqopt = sb_dqopt(sb);
2271        int error;
2272
2273        if (!fmt)
2274                return -ESRCH;
2275        if (!S_ISREG(inode->i_mode)) {
2276                error = -EACCES;
2277                goto out_fmt;
2278        }
2279        if (IS_RDONLY(inode)) {
2280                error = -EROFS;
2281                goto out_fmt;
2282        }
2283        if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2284            (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2285                error = -EINVAL;
2286                goto out_fmt;
2287        }
2288        /* Filesystems outside of init_user_ns not yet supported */
2289        if (sb->s_user_ns != &init_user_ns) {
2290                error = -EINVAL;
2291                goto out_fmt;
2292        }
2293        /* Usage always has to be set... */
2294        if (!(flags & DQUOT_USAGE_ENABLED)) {
2295                error = -EINVAL;
2296                goto out_fmt;
2297        }
2298        if (sb_has_quota_loaded(sb, type)) {
2299                error = -EBUSY;
2300                goto out_fmt;
2301        }
2302
2303        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2304                /* As we bypass the pagecache we must now flush all the
2305                 * dirty data and invalidate caches so that kernel sees
2306                 * changes from userspace. It is not enough to just flush
2307                 * the quota file since if blocksize < pagesize, invalidation
2308                 * of the cache could fail because of other unrelated dirty
2309                 * data */
2310                sync_filesystem(sb);
2311                invalidate_bdev(sb->s_bdev);
2312        }
2313
2314        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2315                /* We don't want quota and atime on quota files (deadlocks
2316                 * possible) Also nobody should write to the file - we use
2317                 * special IO operations which ignore the immutable bit. */
2318                inode_lock(inode);
2319                inode->i_flags |= S_NOQUOTA;
2320                inode_unlock(inode);
2321                /*
2322                 * When S_NOQUOTA is set, remove dquot references as no more
2323                 * references can be added
2324                 */
2325                __dquot_drop(inode);
2326        }
2327
2328        error = -EIO;
2329        dqopt->files[type] = igrab(inode);
2330        if (!dqopt->files[type])
2331                goto out_file_flags;
2332        error = -EINVAL;
2333        if (!fmt->qf_ops->check_quota_file(sb, type))
2334                goto out_file_init;
2335
2336        dqopt->ops[type] = fmt->qf_ops;
2337        dqopt->info[type].dqi_format = fmt;
2338        dqopt->info[type].dqi_fmt_id = format_id;
2339        INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2340        mutex_lock(&dqopt->dqio_mutex);
2341        error = dqopt->ops[type]->read_file_info(sb, type);
2342        if (error < 0) {
2343                mutex_unlock(&dqopt->dqio_mutex);
2344                goto out_file_init;
2345        }
2346        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2347                dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2348        mutex_unlock(&dqopt->dqio_mutex);
2349        spin_lock(&dq_state_lock);
2350        dqopt->flags |= dquot_state_flag(flags, type);
2351        spin_unlock(&dq_state_lock);
2352
2353        add_dquot_ref(sb, type);
2354
2355        return 0;
2356
2357out_file_init:
2358        dqopt->files[type] = NULL;
2359        iput(inode);
2360out_file_flags:
2361        inode_lock(inode);
2362        inode->i_flags &= ~S_NOQUOTA;
2363        inode_unlock(inode);
2364out_fmt:
2365        put_quota_format(fmt);
2366
2367        return error; 
2368}
2369
2370/* Reenable quotas on remount RW */
2371int dquot_resume(struct super_block *sb, int type)
2372{
2373        struct quota_info *dqopt = sb_dqopt(sb);
2374        struct inode *inode;
2375        int ret = 0, cnt;
2376        unsigned int flags;
2377
2378        /* s_umount should be held in exclusive mode */
2379        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2380                up_read(&sb->s_umount);
2381
2382        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2383                if (type != -1 && cnt != type)
2384                        continue;
2385                if (!sb_has_quota_suspended(sb, cnt))
2386                        continue;
2387
2388                inode = dqopt->files[cnt];
2389                dqopt->files[cnt] = NULL;
2390                spin_lock(&dq_state_lock);
2391                flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2392                                                        DQUOT_LIMITS_ENABLED,
2393                                                        cnt);
2394                dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2395                spin_unlock(&dq_state_lock);
2396
2397                flags = dquot_generic_flag(flags, cnt);
2398                ret = vfs_load_quota_inode(inode, cnt,
2399                                dqopt->info[cnt].dqi_fmt_id, flags);
2400                iput(inode);
2401        }
2402
2403        return ret;
2404}
2405EXPORT_SYMBOL(dquot_resume);
2406
2407int dquot_quota_on(struct super_block *sb, int type, int format_id,
2408                   const struct path *path)
2409{
2410        int error = security_quota_on(path->dentry);
2411        if (error)
2412                return error;
2413        /* Quota file not on the same filesystem? */
2414        if (path->dentry->d_sb != sb)
2415                error = -EXDEV;
2416        else
2417                error = vfs_load_quota_inode(d_inode(path->dentry), type,
2418                                             format_id, DQUOT_USAGE_ENABLED |
2419                                             DQUOT_LIMITS_ENABLED);
2420        return error;
2421}
2422EXPORT_SYMBOL(dquot_quota_on);
2423
2424/*
2425 * More powerful function for turning on quotas allowing setting
2426 * of individual quota flags
2427 */
2428int dquot_enable(struct inode *inode, int type, int format_id,
2429                 unsigned int flags)
2430{
2431        struct super_block *sb = inode->i_sb;
2432
2433        /* Just unsuspend quotas? */
2434        BUG_ON(flags & DQUOT_SUSPENDED);
2435        /* s_umount should be held in exclusive mode */
2436        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2437                up_read(&sb->s_umount);
2438
2439        if (!flags)
2440                return 0;
2441        /* Just updating flags needed? */
2442        if (sb_has_quota_loaded(sb, type)) {
2443                if (flags & DQUOT_USAGE_ENABLED &&
2444                    sb_has_quota_usage_enabled(sb, type))
2445                        return -EBUSY;
2446                if (flags & DQUOT_LIMITS_ENABLED &&
2447                    sb_has_quota_limits_enabled(sb, type))
2448                        return -EBUSY;
2449                spin_lock(&dq_state_lock);
2450                sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2451                spin_unlock(&dq_state_lock);
2452                return 0;
2453        }
2454
2455        return vfs_load_quota_inode(inode, type, format_id, flags);
2456}
2457EXPORT_SYMBOL(dquot_enable);
2458
2459/*
2460 * This function is used when filesystem needs to initialize quotas
2461 * during mount time.
2462 */
2463int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2464                int format_id, int type)
2465{
2466        struct dentry *dentry;
2467        int error;
2468
2469        dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2470        if (IS_ERR(dentry))
2471                return PTR_ERR(dentry);
2472
2473        if (d_really_is_negative(dentry)) {
2474                error = -ENOENT;
2475                goto out;
2476        }
2477
2478        error = security_quota_on(dentry);
2479        if (!error)
2480                error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2481                                DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2482
2483out:
2484        dput(dentry);
2485        return error;
2486}
2487EXPORT_SYMBOL(dquot_quota_on_mount);
2488
2489static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2490{
2491        int ret;
2492        int type;
2493        struct quota_info *dqopt = sb_dqopt(sb);
2494
2495        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2496                return -ENOSYS;
2497        /* Accounting cannot be turned on while fs is mounted */
2498        flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2499        if (!flags)
2500                return -EINVAL;
2501        for (type = 0; type < MAXQUOTAS; type++) {
2502                if (!(flags & qtype_enforce_flag(type)))
2503                        continue;
2504                /* Can't enforce without accounting */
2505                if (!sb_has_quota_usage_enabled(sb, type))
2506                        return -EINVAL;
2507                ret = dquot_enable(dqopt->files[type], type,
2508                                   dqopt->info[type].dqi_fmt_id,
2509                                   DQUOT_LIMITS_ENABLED);
2510                if (ret < 0)
2511                        goto out_err;
2512        }
2513        return 0;
2514out_err:
2515        /* Backout enforcement enablement we already did */
2516        for (type--; type >= 0; type--)  {
2517                if (flags & qtype_enforce_flag(type))
2518                        dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2519        }
2520        /* Error code translation for better compatibility with XFS */
2521        if (ret == -EBUSY)
2522                ret = -EEXIST;
2523        return ret;
2524}
2525
2526static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2527{
2528        int ret;
2529        int type;
2530        struct quota_info *dqopt = sb_dqopt(sb);
2531
2532        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2533                return -ENOSYS;
2534        /*
2535         * We don't support turning off accounting via quotactl. In principle
2536         * quota infrastructure can do this but filesystems don't expect
2537         * userspace to be able to do it.
2538         */
2539        if (flags &
2540                  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2541                return -EOPNOTSUPP;
2542
2543        /* Filter out limits not enabled */
2544        for (type = 0; type < MAXQUOTAS; type++)
2545                if (!sb_has_quota_limits_enabled(sb, type))
2546                        flags &= ~qtype_enforce_flag(type);
2547        /* Nothing left? */
2548        if (!flags)
2549                return -EEXIST;
2550        for (type = 0; type < MAXQUOTAS; type++) {
2551                if (flags & qtype_enforce_flag(type)) {
2552                        ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2553                        if (ret < 0)
2554                                goto out_err;
2555                }
2556        }
2557        return 0;
2558out_err:
2559        /* Backout enforcement disabling we already did */
2560        for (type--; type >= 0; type--)  {
2561                if (flags & qtype_enforce_flag(type))
2562                        dquot_enable(dqopt->files[type], type,
2563                                     dqopt->info[type].dqi_fmt_id,
2564                                     DQUOT_LIMITS_ENABLED);
2565        }
2566        return ret;
2567}
2568
2569/* Generic routine for getting common part of quota structure */
2570static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2571{
2572        struct mem_dqblk *dm = &dquot->dq_dqb;
2573
2574        memset(di, 0, sizeof(*di));
2575        spin_lock(&dq_data_lock);
2576        di->d_spc_hardlimit = dm->dqb_bhardlimit;
2577        di->d_spc_softlimit = dm->dqb_bsoftlimit;
2578        di->d_ino_hardlimit = dm->dqb_ihardlimit;
2579        di->d_ino_softlimit = dm->dqb_isoftlimit;
2580        di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2581        di->d_ino_count = dm->dqb_curinodes;
2582        di->d_spc_timer = dm->dqb_btime;
2583        di->d_ino_timer = dm->dqb_itime;
2584        spin_unlock(&dq_data_lock);
2585}
2586
2587int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2588                    struct qc_dqblk *di)
2589{
2590        struct dquot *dquot;
2591
2592        dquot = dqget(sb, qid);
2593        if (IS_ERR(dquot))
2594                return PTR_ERR(dquot);
2595        do_get_dqblk(dquot, di);
2596        dqput(dquot);
2597
2598        return 0;
2599}
2600EXPORT_SYMBOL(dquot_get_dqblk);
2601
2602int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2603                         struct qc_dqblk *di)
2604{
2605        struct dquot *dquot;
2606        int err;
2607
2608        if (!sb->dq_op->get_next_id)
2609                return -ENOSYS;
2610        err = sb->dq_op->get_next_id(sb, qid);
2611        if (err < 0)
2612                return err;
2613        dquot = dqget(sb, *qid);
2614        if (IS_ERR(dquot))
2615                return PTR_ERR(dquot);
2616        do_get_dqblk(dquot, di);
2617        dqput(dquot);
2618
2619        return 0;
2620}
2621EXPORT_SYMBOL(dquot_get_next_dqblk);
2622
2623#define VFS_QC_MASK \
2624        (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2625         QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2626         QC_SPC_TIMER | QC_INO_TIMER)
2627
2628/* Generic routine for setting common part of quota structure */
2629static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2630{
2631        struct mem_dqblk *dm = &dquot->dq_dqb;
2632        int check_blim = 0, check_ilim = 0;
2633        struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2634
2635        if (di->d_fieldmask & ~VFS_QC_MASK)
2636                return -EINVAL;
2637
2638        if (((di->d_fieldmask & QC_SPC_SOFT) &&
2639             di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2640            ((di->d_fieldmask & QC_SPC_HARD) &&
2641             di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2642            ((di->d_fieldmask & QC_INO_SOFT) &&
2643             (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2644            ((di->d_fieldmask & QC_INO_HARD) &&
2645             (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2646                return -ERANGE;
2647
2648        spin_lock(&dq_data_lock);
2649        if (di->d_fieldmask & QC_SPACE) {
2650                dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2651                check_blim = 1;
2652                set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2653        }
2654
2655        if (di->d_fieldmask & QC_SPC_SOFT)
2656                dm->dqb_bsoftlimit = di->d_spc_softlimit;
2657        if (di->d_fieldmask & QC_SPC_HARD)
2658                dm->dqb_bhardlimit = di->d_spc_hardlimit;
2659        if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2660                check_blim = 1;
2661                set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2662        }
2663
2664        if (di->d_fieldmask & QC_INO_COUNT) {
2665                dm->dqb_curinodes = di->d_ino_count;
2666                check_ilim = 1;
2667                set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2668        }
2669
2670        if (di->d_fieldmask & QC_INO_SOFT)
2671                dm->dqb_isoftlimit = di->d_ino_softlimit;
2672        if (di->d_fieldmask & QC_INO_HARD)
2673                dm->dqb_ihardlimit = di->d_ino_hardlimit;
2674        if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2675                check_ilim = 1;
2676                set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2677        }
2678
2679        if (di->d_fieldmask & QC_SPC_TIMER) {
2680                dm->dqb_btime = di->d_spc_timer;
2681                check_blim = 1;
2682                set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2683        }
2684
2685        if (di->d_fieldmask & QC_INO_TIMER) {
2686                dm->dqb_itime = di->d_ino_timer;
2687                check_ilim = 1;
2688                set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2689        }
2690
2691        if (check_blim) {
2692                if (!dm->dqb_bsoftlimit ||
2693                    dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2694                        dm->dqb_btime = 0;
2695                        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2696                } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2697                        /* Set grace only if user hasn't provided his own... */
2698                        dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2699        }
2700        if (check_ilim) {
2701                if (!dm->dqb_isoftlimit ||
2702                    dm->dqb_curinodes < dm->dqb_isoftlimit) {
2703                        dm->dqb_itime = 0;
2704                        clear_bit(DQ_INODES_B, &dquot->dq_flags);
2705                } else if (!(di->d_fieldmask & QC_INO_TIMER))
2706                        /* Set grace only if user hasn't provided his own... */
2707                        dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2708        }
2709        if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2710            dm->dqb_isoftlimit)
2711                clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2712        else
2713                set_bit(DQ_FAKE_B, &dquot->dq_flags);
2714        spin_unlock(&dq_data_lock);
2715        mark_dquot_dirty(dquot);
2716
2717        return 0;
2718}
2719
2720int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2721                  struct qc_dqblk *di)
2722{
2723        struct dquot *dquot;
2724        int rc;
2725
2726        dquot = dqget(sb, qid);
2727        if (IS_ERR(dquot)) {
2728                rc = PTR_ERR(dquot);
2729                goto out;
2730        }
2731        rc = do_set_dqblk(dquot, di);
2732        dqput(dquot);
2733out:
2734        return rc;
2735}
2736EXPORT_SYMBOL(dquot_set_dqblk);
2737
2738/* Generic routine for getting common part of quota file information */
2739int dquot_get_state(struct super_block *sb, struct qc_state *state)
2740{
2741        struct mem_dqinfo *mi;
2742        struct qc_type_state *tstate;
2743        struct quota_info *dqopt = sb_dqopt(sb);
2744        int type;
2745  
2746        memset(state, 0, sizeof(*state));
2747        for (type = 0; type < MAXQUOTAS; type++) {
2748                if (!sb_has_quota_active(sb, type))
2749                        continue;
2750                tstate = state->s_state + type;
2751                mi = sb_dqopt(sb)->info + type;
2752                tstate->flags = QCI_ACCT_ENABLED;
2753                spin_lock(&dq_data_lock);
2754                if (mi->dqi_flags & DQF_SYS_FILE)
2755                        tstate->flags |= QCI_SYSFILE;
2756                if (mi->dqi_flags & DQF_ROOT_SQUASH)
2757                        tstate->flags |= QCI_ROOT_SQUASH;
2758                if (sb_has_quota_limits_enabled(sb, type))
2759                        tstate->flags |= QCI_LIMITS_ENFORCED;
2760                tstate->spc_timelimit = mi->dqi_bgrace;
2761                tstate->ino_timelimit = mi->dqi_igrace;
2762                tstate->ino = dqopt->files[type]->i_ino;
2763                tstate->blocks = dqopt->files[type]->i_blocks;
2764                tstate->nextents = 1;   /* We don't know... */
2765                spin_unlock(&dq_data_lock);
2766        }
2767        return 0;
2768}
2769EXPORT_SYMBOL(dquot_get_state);
2770
2771/* Generic routine for setting common part of quota file information */
2772int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2773{
2774        struct mem_dqinfo *mi;
2775        int err = 0;
2776
2777        if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2778            (ii->i_fieldmask & QC_RT_SPC_TIMER))
2779                return -EINVAL;
2780        if (!sb_has_quota_active(sb, type))
2781                return -ESRCH;
2782        mi = sb_dqopt(sb)->info + type;
2783        if (ii->i_fieldmask & QC_FLAGS) {
2784                if ((ii->i_flags & QCI_ROOT_SQUASH &&
2785                     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2786                        return -EINVAL;
2787        }
2788        spin_lock(&dq_data_lock);
2789        if (ii->i_fieldmask & QC_SPC_TIMER)
2790                mi->dqi_bgrace = ii->i_spc_timelimit;
2791        if (ii->i_fieldmask & QC_INO_TIMER)
2792                mi->dqi_igrace = ii->i_ino_timelimit;
2793        if (ii->i_fieldmask & QC_FLAGS) {
2794                if (ii->i_flags & QCI_ROOT_SQUASH)
2795                        mi->dqi_flags |= DQF_ROOT_SQUASH;
2796                else
2797                        mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2798        }
2799        spin_unlock(&dq_data_lock);
2800        mark_info_dirty(sb, type);
2801        /* Force write to disk */
2802        sb->dq_op->write_info(sb, type);
2803        return err;
2804}
2805EXPORT_SYMBOL(dquot_set_dqinfo);
2806
2807const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2808        .quota_enable   = dquot_quota_enable,
2809        .quota_disable  = dquot_quota_disable,
2810        .quota_sync     = dquot_quota_sync,
2811        .get_state      = dquot_get_state,
2812        .set_info       = dquot_set_dqinfo,
2813        .get_dqblk      = dquot_get_dqblk,
2814        .get_nextdqblk  = dquot_get_next_dqblk,
2815        .set_dqblk      = dquot_set_dqblk
2816};
2817EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2818
2819static int do_proc_dqstats(struct ctl_table *table, int write,
2820                     void __user *buffer, size_t *lenp, loff_t *ppos)
2821{
2822        unsigned int type = (int *)table->data - dqstats.stat;
2823
2824        /* Update global table */
2825        dqstats.stat[type] =
2826                        percpu_counter_sum_positive(&dqstats.counter[type]);
2827        return proc_dointvec(table, write, buffer, lenp, ppos);
2828}
2829
2830static struct ctl_table fs_dqstats_table[] = {
2831        {
2832                .procname       = "lookups",
2833                .data           = &dqstats.stat[DQST_LOOKUPS],
2834                .maxlen         = sizeof(int),
2835                .mode           = 0444,
2836                .proc_handler   = do_proc_dqstats,
2837        },
2838        {
2839                .procname       = "drops",
2840                .data           = &dqstats.stat[DQST_DROPS],
2841                .maxlen         = sizeof(int),
2842                .mode           = 0444,
2843                .proc_handler   = do_proc_dqstats,
2844        },
2845        {
2846                .procname       = "reads",
2847                .data           = &dqstats.stat[DQST_READS],
2848                .maxlen         = sizeof(int),
2849                .mode           = 0444,
2850                .proc_handler   = do_proc_dqstats,
2851        },
2852        {
2853                .procname       = "writes",
2854                .data           = &dqstats.stat[DQST_WRITES],
2855                .maxlen         = sizeof(int),
2856                .mode           = 0444,
2857                .proc_handler   = do_proc_dqstats,
2858        },
2859        {
2860                .procname       = "cache_hits",
2861                .data           = &dqstats.stat[DQST_CACHE_HITS],
2862                .maxlen         = sizeof(int),
2863                .mode           = 0444,
2864                .proc_handler   = do_proc_dqstats,
2865        },
2866        {
2867                .procname       = "allocated_dquots",
2868                .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2869                .maxlen         = sizeof(int),
2870                .mode           = 0444,
2871                .proc_handler   = do_proc_dqstats,
2872        },
2873        {
2874                .procname       = "free_dquots",
2875                .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2876                .maxlen         = sizeof(int),
2877                .mode           = 0444,
2878                .proc_handler   = do_proc_dqstats,
2879        },
2880        {
2881                .procname       = "syncs",
2882                .data           = &dqstats.stat[DQST_SYNCS],
2883                .maxlen         = sizeof(int),
2884                .mode           = 0444,
2885                .proc_handler   = do_proc_dqstats,
2886        },
2887#ifdef CONFIG_PRINT_QUOTA_WARNING
2888        {
2889                .procname       = "warnings",
2890                .data           = &flag_print_warnings,
2891                .maxlen         = sizeof(int),
2892                .mode           = 0644,
2893                .proc_handler   = proc_dointvec,
2894        },
2895#endif
2896        { },
2897};
2898
2899static struct ctl_table fs_table[] = {
2900        {
2901                .procname       = "quota",
2902                .mode           = 0555,
2903                .child          = fs_dqstats_table,
2904        },
2905        { },
2906};
2907
2908static struct ctl_table sys_table[] = {
2909        {
2910                .procname       = "fs",
2911                .mode           = 0555,
2912                .child          = fs_table,
2913        },
2914        { },
2915};
2916
2917static int __init dquot_init(void)
2918{
2919        int i, ret;
2920        unsigned long nr_hash, order;
2921
2922        printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2923
2924        register_sysctl_table(sys_table);
2925
2926        dquot_cachep = kmem_cache_create("dquot",
2927                        sizeof(struct dquot), sizeof(unsigned long) * 4,
2928                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2929                                SLAB_MEM_SPREAD|SLAB_PANIC),
2930                        NULL);
2931
2932        order = 0;
2933        dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2934        if (!dquot_hash)
2935                panic("Cannot create dquot hash table");
2936
2937        for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2938                ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2939                if (ret)
2940                        panic("Cannot create dquot stat counters");
2941        }
2942
2943        /* Find power-of-two hlist_heads which can fit into allocation */
2944        nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2945        dq_hash_bits = 0;
2946        do {
2947                dq_hash_bits++;
2948        } while (nr_hash >> dq_hash_bits);
2949        dq_hash_bits--;
2950
2951        nr_hash = 1UL << dq_hash_bits;
2952        dq_hash_mask = nr_hash - 1;
2953        for (i = 0; i < nr_hash; i++)
2954                INIT_HLIST_HEAD(dquot_hash + i);
2955
2956        pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2957                " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2958
2959        register_shrinker(&dqcache_shrinker);
2960
2961        return 0;
2962}
2963fs_initcall(dquot_init);
2964