linux/fs/quota/dquot.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Implementation of the diskquota system for the LINUX operating system. QUOTA
   4 * is implemented using the BSD system call interface as the means of
   5 * communication with the user level. This file contains the generic routines
   6 * called by the different filesystems on allocation of an inode or block.
   7 * These routines take care of the administration needed to have a consistent
   8 * diskquota tracking system. The ideas of both user and group quotas are based
   9 * on the Melbourne quota system as used on BSD derived systems. The internal
  10 * implementation is based on one of the several variants of the LINUX
  11 * inode-subsystem with added complexity of the diskquota system.
  12 *
  13 * Author:      Marco van Wieringen <mvw@planets.elm.net>
  14 *
  15 * Fixes:   Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
  16 *
  17 *              Revised list management to avoid races
  18 *              -- Bill Hawes, <whawes@star.net>, 9/98
  19 *
  20 *              Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
  21 *              As the consequence the locking was moved from dquot_decr_...(),
  22 *              dquot_incr_...() to calling functions.
  23 *              invalidate_dquots() now writes modified dquots.
  24 *              Serialized quota_off() and quota_on() for mount point.
  25 *              Fixed a few bugs in grow_dquots().
  26 *              Fixed deadlock in write_dquot() - we no longer account quotas on
  27 *              quota files
  28 *              remove_dquot_ref() moved to inode.c - it now traverses through inodes
  29 *              add_dquot_ref() restarts after blocking
  30 *              Added check for bogus uid and fixed check for group in quotactl.
  31 *              Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
  32 *
  33 *              Used struct list_head instead of own list struct
  34 *              Invalidation of referenced dquots is no longer possible
  35 *              Improved free_dquots list management
  36 *              Quota and i_blocks are now updated in one place to avoid races
  37 *              Warnings are now delayed so we won't block in critical section
  38 *              Write updated not to require dquot lock
  39 *              Jan Kara, <jack@suse.cz>, 9/2000
  40 *
  41 *              Added dynamic quota structure allocation
  42 *              Jan Kara <jack@suse.cz> 12/2000
  43 *
  44 *              Rewritten quota interface. Implemented new quota format and
  45 *              formats registering.
  46 *              Jan Kara, <jack@suse.cz>, 2001,2002
  47 *
  48 *              New SMP locking.
  49 *              Jan Kara, <jack@suse.cz>, 10/2002
  50 *
  51 *              Added journalled quota support, fix lock inversion problems
  52 *              Jan Kara, <jack@suse.cz>, 2003,2004
  53 *
  54 * (C) Copyright 1994 - 1997 Marco van Wieringen
  55 */
  56
  57#include <linux/errno.h>
  58#include <linux/kernel.h>
  59#include <linux/fs.h>
  60#include <linux/mount.h>
  61#include <linux/mm.h>
  62#include <linux/time.h>
  63#include <linux/types.h>
  64#include <linux/string.h>
  65#include <linux/fcntl.h>
  66#include <linux/stat.h>
  67#include <linux/tty.h>
  68#include <linux/file.h>
  69#include <linux/slab.h>
  70#include <linux/sysctl.h>
  71#include <linux/init.h>
  72#include <linux/module.h>
  73#include <linux/proc_fs.h>
  74#include <linux/security.h>
  75#include <linux/sched.h>
  76#include <linux/cred.h>
  77#include <linux/kmod.h>
  78#include <linux/namei.h>
  79#include <linux/capability.h>
  80#include <linux/quotaops.h>
  81#include <linux/blkdev.h>
  82#include "../internal.h" /* ugh */
  83
  84#include <linux/uaccess.h>
  85
  86/*
  87 * There are five quota SMP locks:
  88 * * dq_list_lock protects all lists with quotas and quota formats.
  89 * * dquot->dq_dqb_lock protects data from dq_dqb
  90 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
  91 *   consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
  92 *   dquot_transfer() can stabilize amount it transfers
  93 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
  94 *   pointers in the inode
  95 * * dq_state_lock protects modifications of quota state (on quotaon and
  96 *   quotaoff) and readers who care about latest values take it as well.
  97 *
  98 * The spinlock ordering is hence:
  99 *   dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
 100 *   dq_list_lock > dq_state_lock
 101 *
 102 * Note that some things (eg. sb pointer, type, id) doesn't change during
 103 * the life of the dquot structure and so needn't to be protected by a lock
 104 *
 105 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
 106 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
 107 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
 108 * inode and before dropping dquot references to avoid use of dquots after
 109 * they are freed. dq_data_lock is used to serialize the pointer setting and
 110 * clearing operations.
 111 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
 112 * inode is a quota file). Functions adding pointers from inode to dquots have
 113 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
 114 * have to do all pointer modifications before dropping dq_data_lock. This makes
 115 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
 116 * then drops all pointers to dquots from an inode.
 117 *
 118 * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
 119 * memory (or space for it is being allocated) on the first dqget(), when it is
 120 * being written out, and when it is being released on the last dqput(). The
 121 * allocation and release operations are serialized by the dq_lock and by
 122 * checking the use count in dquot_release().
 123 *
 124 * Lock ordering (including related VFS locks) is the following:
 125 *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
 126 */
 127
 128static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
 129static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
 130__cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
 131EXPORT_SYMBOL(dq_data_lock);
 132DEFINE_STATIC_SRCU(dquot_srcu);
 133
 134static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
 135
 136void __quota_error(struct super_block *sb, const char *func,
 137                   const char *fmt, ...)
 138{
 139        if (printk_ratelimit()) {
 140                va_list args;
 141                struct va_format vaf;
 142
 143                va_start(args, fmt);
 144
 145                vaf.fmt = fmt;
 146                vaf.va = &args;
 147
 148                printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
 149                       sb->s_id, func, &vaf);
 150
 151                va_end(args);
 152        }
 153}
 154EXPORT_SYMBOL(__quota_error);
 155
 156#if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
 157static char *quotatypes[] = INITQFNAMES;
 158#endif
 159static struct quota_format_type *quota_formats; /* List of registered formats */
 160static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 161
 162/* SLAB cache for dquot structures */
 163static struct kmem_cache *dquot_cachep;
 164
 165int register_quota_format(struct quota_format_type *fmt)
 166{
 167        spin_lock(&dq_list_lock);
 168        fmt->qf_next = quota_formats;
 169        quota_formats = fmt;
 170        spin_unlock(&dq_list_lock);
 171        return 0;
 172}
 173EXPORT_SYMBOL(register_quota_format);
 174
 175void unregister_quota_format(struct quota_format_type *fmt)
 176{
 177        struct quota_format_type **actqf;
 178
 179        spin_lock(&dq_list_lock);
 180        for (actqf = &quota_formats; *actqf && *actqf != fmt;
 181             actqf = &(*actqf)->qf_next)
 182                ;
 183        if (*actqf)
 184                *actqf = (*actqf)->qf_next;
 185        spin_unlock(&dq_list_lock);
 186}
 187EXPORT_SYMBOL(unregister_quota_format);
 188
 189static struct quota_format_type *find_quota_format(int id)
 190{
 191        struct quota_format_type *actqf;
 192
 193        spin_lock(&dq_list_lock);
 194        for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
 195             actqf = actqf->qf_next)
 196                ;
 197        if (!actqf || !try_module_get(actqf->qf_owner)) {
 198                int qm;
 199
 200                spin_unlock(&dq_list_lock);
 201
 202                for (qm = 0; module_names[qm].qm_fmt_id &&
 203                             module_names[qm].qm_fmt_id != id; qm++)
 204                        ;
 205                if (!module_names[qm].qm_fmt_id ||
 206                    request_module(module_names[qm].qm_mod_name))
 207                        return NULL;
 208
 209                spin_lock(&dq_list_lock);
 210                for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
 211                     actqf = actqf->qf_next)
 212                        ;
 213                if (actqf && !try_module_get(actqf->qf_owner))
 214                        actqf = NULL;
 215        }
 216        spin_unlock(&dq_list_lock);
 217        return actqf;
 218}
 219
 220static void put_quota_format(struct quota_format_type *fmt)
 221{
 222        module_put(fmt->qf_owner);
 223}
 224
 225/*
 226 * Dquot List Management:
 227 * The quota code uses four lists for dquot management: the inuse_list,
 228 * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
 229 * structure may be on some of those lists, depending on its current state.
 230 *
 231 * All dquots are placed to the end of inuse_list when first created, and this
 232 * list is used for invalidate operation, which must look at every dquot.
 233 *
 234 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 235 * and this list is searched whenever we need an available dquot.  Dquots are
 236 * removed from the list as soon as they are used again, and
 237 * dqstats.free_dquots gives the number of dquots on the list. When
 238 * dquot is invalidated it's completely released from memory.
 239 *
 240 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
 241 * dirtied, and this list is searched when writing dirty dquots back to
 242 * quota file. Note that some filesystems do dirty dquot tracking on their
 243 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
 244 *
 245 * Dquots with a specific identity (device, type and id) are placed on
 246 * one of the dquot_hash[] hash chains. The provides an efficient search
 247 * mechanism to locate a specific dquot.
 248 */
 249
 250static LIST_HEAD(inuse_list);
 251static LIST_HEAD(free_dquots);
 252static unsigned int dq_hash_bits, dq_hash_mask;
 253static struct hlist_head *dquot_hash;
 254
 255struct dqstats dqstats;
 256EXPORT_SYMBOL(dqstats);
 257
 258static qsize_t inode_get_rsv_space(struct inode *inode);
 259static qsize_t __inode_get_rsv_space(struct inode *inode);
 260static int __dquot_initialize(struct inode *inode, int type);
 261
 262static inline unsigned int
 263hashfn(const struct super_block *sb, struct kqid qid)
 264{
 265        unsigned int id = from_kqid(&init_user_ns, qid);
 266        int type = qid.type;
 267        unsigned long tmp;
 268
 269        tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 270        return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 271}
 272
 273/*
 274 * Following list functions expect dq_list_lock to be held
 275 */
 276static inline void insert_dquot_hash(struct dquot *dquot)
 277{
 278        struct hlist_head *head;
 279        head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
 280        hlist_add_head(&dquot->dq_hash, head);
 281}
 282
 283static inline void remove_dquot_hash(struct dquot *dquot)
 284{
 285        hlist_del_init(&dquot->dq_hash);
 286}
 287
 288static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
 289                                struct kqid qid)
 290{
 291        struct dquot *dquot;
 292
 293        hlist_for_each_entry(dquot, dquot_hash+hashent, dq_hash)
 294                if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
 295                        return dquot;
 296
 297        return NULL;
 298}
 299
 300/* Add a dquot to the tail of the free list */
 301static inline void put_dquot_last(struct dquot *dquot)
 302{
 303        list_add_tail(&dquot->dq_free, &free_dquots);
 304        dqstats_inc(DQST_FREE_DQUOTS);
 305}
 306
 307static inline void remove_free_dquot(struct dquot *dquot)
 308{
 309        if (list_empty(&dquot->dq_free))
 310                return;
 311        list_del_init(&dquot->dq_free);
 312        dqstats_dec(DQST_FREE_DQUOTS);
 313}
 314
 315static inline void put_inuse(struct dquot *dquot)
 316{
 317        /* We add to the back of inuse list so we don't have to restart
 318         * when traversing this list and we block */
 319        list_add_tail(&dquot->dq_inuse, &inuse_list);
 320        dqstats_inc(DQST_ALLOC_DQUOTS);
 321}
 322
 323static inline void remove_inuse(struct dquot *dquot)
 324{
 325        dqstats_dec(DQST_ALLOC_DQUOTS);
 326        list_del(&dquot->dq_inuse);
 327}
 328/*
 329 * End of list functions needing dq_list_lock
 330 */
 331
 332static void wait_on_dquot(struct dquot *dquot)
 333{
 334        mutex_lock(&dquot->dq_lock);
 335        mutex_unlock(&dquot->dq_lock);
 336}
 337
 338static inline int dquot_dirty(struct dquot *dquot)
 339{
 340        return test_bit(DQ_MOD_B, &dquot->dq_flags);
 341}
 342
 343static inline int mark_dquot_dirty(struct dquot *dquot)
 344{
 345        return dquot->dq_sb->dq_op->mark_dirty(dquot);
 346}
 347
 348/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
 349int dquot_mark_dquot_dirty(struct dquot *dquot)
 350{
 351        int ret = 1;
 352
 353        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 354                return 0;
 355
 356        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
 357                return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
 358
 359        /* If quota is dirty already, we don't have to acquire dq_list_lock */
 360        if (test_bit(DQ_MOD_B, &dquot->dq_flags))
 361                return 1;
 362
 363        spin_lock(&dq_list_lock);
 364        if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
 365                list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 366                                info[dquot->dq_id.type].dqi_dirty_list);
 367                ret = 0;
 368        }
 369        spin_unlock(&dq_list_lock);
 370        return ret;
 371}
 372EXPORT_SYMBOL(dquot_mark_dquot_dirty);
 373
 374/* Dirtify all the dquots - this can block when journalling */
 375static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
 376{
 377        int ret, err, cnt;
 378
 379        ret = err = 0;
 380        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 381                if (dquot[cnt])
 382                        /* Even in case of error we have to continue */
 383                        ret = mark_dquot_dirty(dquot[cnt]);
 384                if (!err)
 385                        err = ret;
 386        }
 387        return err;
 388}
 389
 390static inline void dqput_all(struct dquot **dquot)
 391{
 392        unsigned int cnt;
 393
 394        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 395                dqput(dquot[cnt]);
 396}
 397
 398static inline int clear_dquot_dirty(struct dquot *dquot)
 399{
 400        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
 401                return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
 402
 403        spin_lock(&dq_list_lock);
 404        if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
 405                spin_unlock(&dq_list_lock);
 406                return 0;
 407        }
 408        list_del_init(&dquot->dq_dirty);
 409        spin_unlock(&dq_list_lock);
 410        return 1;
 411}
 412
 413void mark_info_dirty(struct super_block *sb, int type)
 414{
 415        spin_lock(&dq_data_lock);
 416        sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
 417        spin_unlock(&dq_data_lock);
 418}
 419EXPORT_SYMBOL(mark_info_dirty);
 420
 421/*
 422 *      Read dquot from disk and alloc space for it
 423 */
 424
 425int dquot_acquire(struct dquot *dquot)
 426{
 427        int ret = 0, ret2 = 0;
 428        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 429
 430        mutex_lock(&dquot->dq_lock);
 431        if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
 432                ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
 433                if (ret < 0)
 434                        goto out_iolock;
 435        }
 436        /* Make sure flags update is visible after dquot has been filled */
 437        smp_mb__before_atomic();
 438        set_bit(DQ_READ_B, &dquot->dq_flags);
 439        /* Instantiate dquot if needed */
 440        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 441                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 442                /* Write the info if needed */
 443                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 444                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 445                                        dquot->dq_sb, dquot->dq_id.type);
 446                }
 447                if (ret < 0)
 448                        goto out_iolock;
 449                if (ret2 < 0) {
 450                        ret = ret2;
 451                        goto out_iolock;
 452                }
 453        }
 454        /*
 455         * Make sure flags update is visible after on-disk struct has been
 456         * allocated. Paired with smp_rmb() in dqget().
 457         */
 458        smp_mb__before_atomic();
 459        set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 460out_iolock:
 461        mutex_unlock(&dquot->dq_lock);
 462        return ret;
 463}
 464EXPORT_SYMBOL(dquot_acquire);
 465
 466/*
 467 *      Write dquot to disk
 468 */
 469int dquot_commit(struct dquot *dquot)
 470{
 471        int ret = 0;
 472        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 473
 474        mutex_lock(&dquot->dq_lock);
 475        if (!clear_dquot_dirty(dquot))
 476                goto out_lock;
 477        /* Inactive dquot can be only if there was error during read/init
 478         * => we have better not writing it */
 479        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 480                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 481        else
 482                ret = -EIO;
 483out_lock:
 484        mutex_unlock(&dquot->dq_lock);
 485        return ret;
 486}
 487EXPORT_SYMBOL(dquot_commit);
 488
 489/*
 490 *      Release dquot
 491 */
 492int dquot_release(struct dquot *dquot)
 493{
 494        int ret = 0, ret2 = 0;
 495        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 496
 497        mutex_lock(&dquot->dq_lock);
 498        /* Check whether we are not racing with some other dqget() */
 499        if (dquot_is_busy(dquot))
 500                goto out_dqlock;
 501        if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
 502                ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
 503                /* Write the info */
 504                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 505                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 506                                                dquot->dq_sb, dquot->dq_id.type);
 507                }
 508                if (ret >= 0)
 509                        ret = ret2;
 510        }
 511        clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 512out_dqlock:
 513        mutex_unlock(&dquot->dq_lock);
 514        return ret;
 515}
 516EXPORT_SYMBOL(dquot_release);
 517
 518void dquot_destroy(struct dquot *dquot)
 519{
 520        kmem_cache_free(dquot_cachep, dquot);
 521}
 522EXPORT_SYMBOL(dquot_destroy);
 523
 524static inline void do_destroy_dquot(struct dquot *dquot)
 525{
 526        dquot->dq_sb->dq_op->destroy_dquot(dquot);
 527}
 528
 529/* Invalidate all dquots on the list. Note that this function is called after
 530 * quota is disabled and pointers from inodes removed so there cannot be new
 531 * quota users. There can still be some users of quotas due to inodes being
 532 * just deleted or pruned by prune_icache() (those are not attached to any
 533 * list) or parallel quotactl call. We have to wait for such users.
 534 */
 535static void invalidate_dquots(struct super_block *sb, int type)
 536{
 537        struct dquot *dquot, *tmp;
 538
 539restart:
 540        spin_lock(&dq_list_lock);
 541        list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 542                if (dquot->dq_sb != sb)
 543                        continue;
 544                if (dquot->dq_id.type != type)
 545                        continue;
 546                /* Wait for dquot users */
 547                if (atomic_read(&dquot->dq_count)) {
 548                        dqgrab(dquot);
 549                        spin_unlock(&dq_list_lock);
 550                        /*
 551                         * Once dqput() wakes us up, we know it's time to free
 552                         * the dquot.
 553                         * IMPORTANT: we rely on the fact that there is always
 554                         * at most one process waiting for dquot to free.
 555                         * Otherwise dq_count would be > 1 and we would never
 556                         * wake up.
 557                         */
 558                        wait_event(dquot_ref_wq,
 559                                   atomic_read(&dquot->dq_count) == 1);
 560                        dqput(dquot);
 561                        /* At this moment dquot() need not exist (it could be
 562                         * reclaimed by prune_dqcache(). Hence we must
 563                         * restart. */
 564                        goto restart;
 565                }
 566                /*
 567                 * Quota now has no users and it has been written on last
 568                 * dqput()
 569                 */
 570                remove_dquot_hash(dquot);
 571                remove_free_dquot(dquot);
 572                remove_inuse(dquot);
 573                do_destroy_dquot(dquot);
 574        }
 575        spin_unlock(&dq_list_lock);
 576}
 577
 578/* Call callback for every active dquot on given filesystem */
 579int dquot_scan_active(struct super_block *sb,
 580                      int (*fn)(struct dquot *dquot, unsigned long priv),
 581                      unsigned long priv)
 582{
 583        struct dquot *dquot, *old_dquot = NULL;
 584        int ret = 0;
 585
 586        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 587
 588        spin_lock(&dq_list_lock);
 589        list_for_each_entry(dquot, &inuse_list, dq_inuse) {
 590                if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 591                        continue;
 592                if (dquot->dq_sb != sb)
 593                        continue;
 594                /* Now we have active dquot so we can just increase use count */
 595                atomic_inc(&dquot->dq_count);
 596                spin_unlock(&dq_list_lock);
 597                dqput(old_dquot);
 598                old_dquot = dquot;
 599                /*
 600                 * ->release_dquot() can be racing with us. Our reference
 601                 * protects us from new calls to it so just wait for any
 602                 * outstanding call and recheck the DQ_ACTIVE_B after that.
 603                 */
 604                wait_on_dquot(dquot);
 605                if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 606                        ret = fn(dquot, priv);
 607                        if (ret < 0)
 608                                goto out;
 609                }
 610                spin_lock(&dq_list_lock);
 611                /* We are safe to continue now because our dquot could not
 612                 * be moved out of the inuse list while we hold the reference */
 613        }
 614        spin_unlock(&dq_list_lock);
 615out:
 616        dqput(old_dquot);
 617        return ret;
 618}
 619EXPORT_SYMBOL(dquot_scan_active);
 620
 621/* Write all dquot structures to quota files */
 622int dquot_writeback_dquots(struct super_block *sb, int type)
 623{
 624        struct list_head dirty;
 625        struct dquot *dquot;
 626        struct quota_info *dqopt = sb_dqopt(sb);
 627        int cnt;
 628        int err, ret = 0;
 629
 630        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 631
 632        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 633                if (type != -1 && cnt != type)
 634                        continue;
 635                if (!sb_has_quota_active(sb, cnt))
 636                        continue;
 637                spin_lock(&dq_list_lock);
 638                /* Move list away to avoid livelock. */
 639                list_replace_init(&dqopt->info[cnt].dqi_dirty_list, &dirty);
 640                while (!list_empty(&dirty)) {
 641                        dquot = list_first_entry(&dirty, struct dquot,
 642                                                 dq_dirty);
 643
 644                        WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
 645
 646                        /* Now we have active dquot from which someone is
 647                         * holding reference so we can safely just increase
 648                         * use count */
 649                        dqgrab(dquot);
 650                        spin_unlock(&dq_list_lock);
 651                        err = sb->dq_op->write_dquot(dquot);
 652                        if (err) {
 653                                /*
 654                                 * Clear dirty bit anyway to avoid infinite
 655                                 * loop here.
 656                                 */
 657                                clear_dquot_dirty(dquot);
 658                                if (!ret)
 659                                        ret = err;
 660                        }
 661                        dqput(dquot);
 662                        spin_lock(&dq_list_lock);
 663                }
 664                spin_unlock(&dq_list_lock);
 665        }
 666
 667        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 668                if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
 669                    && info_dirty(&dqopt->info[cnt]))
 670                        sb->dq_op->write_info(sb, cnt);
 671        dqstats_inc(DQST_SYNCS);
 672
 673        return ret;
 674}
 675EXPORT_SYMBOL(dquot_writeback_dquots);
 676
 677/* Write all dquot structures to disk and make them visible from userspace */
 678int dquot_quota_sync(struct super_block *sb, int type)
 679{
 680        struct quota_info *dqopt = sb_dqopt(sb);
 681        int cnt;
 682        int ret;
 683
 684        ret = dquot_writeback_dquots(sb, type);
 685        if (ret)
 686                return ret;
 687        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
 688                return 0;
 689
 690        /* This is not very clever (and fast) but currently I don't know about
 691         * any other simple way of getting quota data to disk and we must get
 692         * them there for userspace to be visible... */
 693        if (sb->s_op->sync_fs)
 694                sb->s_op->sync_fs(sb, 1);
 695        sync_blockdev(sb->s_bdev);
 696
 697        /*
 698         * Now when everything is written we can discard the pagecache so
 699         * that userspace sees the changes.
 700         */
 701        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 702                if (type != -1 && cnt != type)
 703                        continue;
 704                if (!sb_has_quota_active(sb, cnt))
 705                        continue;
 706                inode_lock(dqopt->files[cnt]);
 707                truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
 708                inode_unlock(dqopt->files[cnt]);
 709        }
 710
 711        return 0;
 712}
 713EXPORT_SYMBOL(dquot_quota_sync);
 714
 715static unsigned long
 716dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 717{
 718        struct dquot *dquot;
 719        unsigned long freed = 0;
 720
 721        spin_lock(&dq_list_lock);
 722        while (!list_empty(&free_dquots) && sc->nr_to_scan) {
 723                dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
 724                remove_dquot_hash(dquot);
 725                remove_free_dquot(dquot);
 726                remove_inuse(dquot);
 727                do_destroy_dquot(dquot);
 728                sc->nr_to_scan--;
 729                freed++;
 730        }
 731        spin_unlock(&dq_list_lock);
 732        return freed;
 733}
 734
 735static unsigned long
 736dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
 737{
 738        return vfs_pressure_ratio(
 739        percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
 740}
 741
 742static struct shrinker dqcache_shrinker = {
 743        .count_objects = dqcache_shrink_count,
 744        .scan_objects = dqcache_shrink_scan,
 745        .seeks = DEFAULT_SEEKS,
 746};
 747
 748/*
 749 * Put reference to dquot
 750 */
 751void dqput(struct dquot *dquot)
 752{
 753        int ret;
 754
 755        if (!dquot)
 756                return;
 757#ifdef CONFIG_QUOTA_DEBUG
 758        if (!atomic_read(&dquot->dq_count)) {
 759                quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
 760                            quotatypes[dquot->dq_id.type],
 761                            from_kqid(&init_user_ns, dquot->dq_id));
 762                BUG();
 763        }
 764#endif
 765        dqstats_inc(DQST_DROPS);
 766we_slept:
 767        spin_lock(&dq_list_lock);
 768        if (atomic_read(&dquot->dq_count) > 1) {
 769                /* We have more than one user... nothing to do */
 770                atomic_dec(&dquot->dq_count);
 771                /* Releasing dquot during quotaoff phase? */
 772                if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
 773                    atomic_read(&dquot->dq_count) == 1)
 774                        wake_up(&dquot_ref_wq);
 775                spin_unlock(&dq_list_lock);
 776                return;
 777        }
 778        /* Need to release dquot? */
 779        if (dquot_dirty(dquot)) {
 780                spin_unlock(&dq_list_lock);
 781                /* Commit dquot before releasing */
 782                ret = dquot->dq_sb->dq_op->write_dquot(dquot);
 783                if (ret < 0) {
 784                        quota_error(dquot->dq_sb, "Can't write quota structure"
 785                                    " (error %d). Quota may get out of sync!",
 786                                    ret);
 787                        /*
 788                         * We clear dirty bit anyway, so that we avoid
 789                         * infinite loop here
 790                         */
 791                        clear_dquot_dirty(dquot);
 792                }
 793                goto we_slept;
 794        }
 795        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 796                spin_unlock(&dq_list_lock);
 797                dquot->dq_sb->dq_op->release_dquot(dquot);
 798                goto we_slept;
 799        }
 800        atomic_dec(&dquot->dq_count);
 801#ifdef CONFIG_QUOTA_DEBUG
 802        /* sanity check */
 803        BUG_ON(!list_empty(&dquot->dq_free));
 804#endif
 805        put_dquot_last(dquot);
 806        spin_unlock(&dq_list_lock);
 807}
 808EXPORT_SYMBOL(dqput);
 809
 810struct dquot *dquot_alloc(struct super_block *sb, int type)
 811{
 812        return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
 813}
 814EXPORT_SYMBOL(dquot_alloc);
 815
 816static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 817{
 818        struct dquot *dquot;
 819
 820        dquot = sb->dq_op->alloc_dquot(sb, type);
 821        if(!dquot)
 822                return NULL;
 823
 824        mutex_init(&dquot->dq_lock);
 825        INIT_LIST_HEAD(&dquot->dq_free);
 826        INIT_LIST_HEAD(&dquot->dq_inuse);
 827        INIT_HLIST_NODE(&dquot->dq_hash);
 828        INIT_LIST_HEAD(&dquot->dq_dirty);
 829        dquot->dq_sb = sb;
 830        dquot->dq_id = make_kqid_invalid(type);
 831        atomic_set(&dquot->dq_count, 1);
 832        spin_lock_init(&dquot->dq_dqb_lock);
 833
 834        return dquot;
 835}
 836
 837/*
 838 * Get reference to dquot
 839 *
 840 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
 841 * destroying our dquot by:
 842 *   a) checking for quota flags under dq_list_lock and
 843 *   b) getting a reference to dquot before we release dq_list_lock
 844 */
 845struct dquot *dqget(struct super_block *sb, struct kqid qid)
 846{
 847        unsigned int hashent = hashfn(sb, qid);
 848        struct dquot *dquot, *empty = NULL;
 849
 850        if (!qid_has_mapping(sb->s_user_ns, qid))
 851                return ERR_PTR(-EINVAL);
 852
 853        if (!sb_has_quota_active(sb, qid.type))
 854                return ERR_PTR(-ESRCH);
 855we_slept:
 856        spin_lock(&dq_list_lock);
 857        spin_lock(&dq_state_lock);
 858        if (!sb_has_quota_active(sb, qid.type)) {
 859                spin_unlock(&dq_state_lock);
 860                spin_unlock(&dq_list_lock);
 861                dquot = ERR_PTR(-ESRCH);
 862                goto out;
 863        }
 864        spin_unlock(&dq_state_lock);
 865
 866        dquot = find_dquot(hashent, sb, qid);
 867        if (!dquot) {
 868                if (!empty) {
 869                        spin_unlock(&dq_list_lock);
 870                        empty = get_empty_dquot(sb, qid.type);
 871                        if (!empty)
 872                                schedule();     /* Try to wait for a moment... */
 873                        goto we_slept;
 874                }
 875                dquot = empty;
 876                empty = NULL;
 877                dquot->dq_id = qid;
 878                /* all dquots go on the inuse_list */
 879                put_inuse(dquot);
 880                /* hash it first so it can be found */
 881                insert_dquot_hash(dquot);
 882                spin_unlock(&dq_list_lock);
 883                dqstats_inc(DQST_LOOKUPS);
 884        } else {
 885                if (!atomic_read(&dquot->dq_count))
 886                        remove_free_dquot(dquot);
 887                atomic_inc(&dquot->dq_count);
 888                spin_unlock(&dq_list_lock);
 889                dqstats_inc(DQST_CACHE_HITS);
 890                dqstats_inc(DQST_LOOKUPS);
 891        }
 892        /* Wait for dq_lock - after this we know that either dquot_release() is
 893         * already finished or it will be canceled due to dq_count > 1 test */
 894        wait_on_dquot(dquot);
 895        /* Read the dquot / allocate space in quota file */
 896        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 897                int err;
 898
 899                err = sb->dq_op->acquire_dquot(dquot);
 900                if (err < 0) {
 901                        dqput(dquot);
 902                        dquot = ERR_PTR(err);
 903                        goto out;
 904                }
 905        }
 906        /*
 907         * Make sure following reads see filled structure - paired with
 908         * smp_mb__before_atomic() in dquot_acquire().
 909         */
 910        smp_rmb();
 911#ifdef CONFIG_QUOTA_DEBUG
 912        BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 913#endif
 914out:
 915        if (empty)
 916                do_destroy_dquot(empty);
 917
 918        return dquot;
 919}
 920EXPORT_SYMBOL(dqget);
 921
 922static inline struct dquot **i_dquot(struct inode *inode)
 923{
 924        return inode->i_sb->s_op->get_dquots(inode);
 925}
 926
 927static int dqinit_needed(struct inode *inode, int type)
 928{
 929        struct dquot * const *dquots;
 930        int cnt;
 931
 932        if (IS_NOQUOTA(inode))
 933                return 0;
 934
 935        dquots = i_dquot(inode);
 936        if (type != -1)
 937                return !dquots[type];
 938        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 939                if (!dquots[cnt])
 940                        return 1;
 941        return 0;
 942}
 943
 944/* This routine is guarded by s_umount semaphore */
 945static int add_dquot_ref(struct super_block *sb, int type)
 946{
 947        struct inode *inode, *old_inode = NULL;
 948#ifdef CONFIG_QUOTA_DEBUG
 949        int reserved = 0;
 950#endif
 951        int err = 0;
 952
 953        spin_lock(&sb->s_inode_list_lock);
 954        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 955                spin_lock(&inode->i_lock);
 956                if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
 957                    !atomic_read(&inode->i_writecount) ||
 958                    !dqinit_needed(inode, type)) {
 959                        spin_unlock(&inode->i_lock);
 960                        continue;
 961                }
 962                __iget(inode);
 963                spin_unlock(&inode->i_lock);
 964                spin_unlock(&sb->s_inode_list_lock);
 965
 966#ifdef CONFIG_QUOTA_DEBUG
 967                if (unlikely(inode_get_rsv_space(inode) > 0))
 968                        reserved = 1;
 969#endif
 970                iput(old_inode);
 971                err = __dquot_initialize(inode, type);
 972                if (err) {
 973                        iput(inode);
 974                        goto out;
 975                }
 976
 977                /*
 978                 * We hold a reference to 'inode' so it couldn't have been
 979                 * removed from s_inodes list while we dropped the
 980                 * s_inode_list_lock. We cannot iput the inode now as we can be
 981                 * holding the last reference and we cannot iput it under
 982                 * s_inode_list_lock. So we keep the reference and iput it
 983                 * later.
 984                 */
 985                old_inode = inode;
 986                cond_resched();
 987                spin_lock(&sb->s_inode_list_lock);
 988        }
 989        spin_unlock(&sb->s_inode_list_lock);
 990        iput(old_inode);
 991out:
 992#ifdef CONFIG_QUOTA_DEBUG
 993        if (reserved) {
 994                quota_error(sb, "Writes happened before quota was turned on "
 995                        "thus quota information is probably inconsistent. "
 996                        "Please run quotacheck(8)");
 997        }
 998#endif
 999        return err;
1000}
1001
1002/*
1003 * Remove references to dquots from inode and add dquot to list for freeing
1004 * if we have the last reference to dquot
1005 */
1006static void remove_inode_dquot_ref(struct inode *inode, int type,
1007                                   struct list_head *tofree_head)
1008{
1009        struct dquot **dquots = i_dquot(inode);
1010        struct dquot *dquot = dquots[type];
1011
1012        if (!dquot)
1013                return;
1014
1015        dquots[type] = NULL;
1016        if (list_empty(&dquot->dq_free)) {
1017                /*
1018                 * The inode still has reference to dquot so it can't be in the
1019                 * free list
1020                 */
1021                spin_lock(&dq_list_lock);
1022                list_add(&dquot->dq_free, tofree_head);
1023                spin_unlock(&dq_list_lock);
1024        } else {
1025                /*
1026                 * Dquot is already in a list to put so we won't drop the last
1027                 * reference here.
1028                 */
1029                dqput(dquot);
1030        }
1031}
1032
1033/*
1034 * Free list of dquots
1035 * Dquots are removed from inodes and no new references can be got so we are
1036 * the only ones holding reference
1037 */
1038static void put_dquot_list(struct list_head *tofree_head)
1039{
1040        struct list_head *act_head;
1041        struct dquot *dquot;
1042
1043        act_head = tofree_head->next;
1044        while (act_head != tofree_head) {
1045                dquot = list_entry(act_head, struct dquot, dq_free);
1046                act_head = act_head->next;
1047                /* Remove dquot from the list so we won't have problems... */
1048                list_del_init(&dquot->dq_free);
1049                dqput(dquot);
1050        }
1051}
1052
1053static void remove_dquot_ref(struct super_block *sb, int type,
1054                struct list_head *tofree_head)
1055{
1056        struct inode *inode;
1057#ifdef CONFIG_QUOTA_DEBUG
1058        int reserved = 0;
1059#endif
1060
1061        spin_lock(&sb->s_inode_list_lock);
1062        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1063                /*
1064                 *  We have to scan also I_NEW inodes because they can already
1065                 *  have quota pointer initialized. Luckily, we need to touch
1066                 *  only quota pointers and these have separate locking
1067                 *  (dq_data_lock).
1068                 */
1069                spin_lock(&dq_data_lock);
1070                if (!IS_NOQUOTA(inode)) {
1071#ifdef CONFIG_QUOTA_DEBUG
1072                        if (unlikely(inode_get_rsv_space(inode) > 0))
1073                                reserved = 1;
1074#endif
1075                        remove_inode_dquot_ref(inode, type, tofree_head);
1076                }
1077                spin_unlock(&dq_data_lock);
1078        }
1079        spin_unlock(&sb->s_inode_list_lock);
1080#ifdef CONFIG_QUOTA_DEBUG
1081        if (reserved) {
1082                printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1083                        " was disabled thus quota information is probably "
1084                        "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1085        }
1086#endif
1087}
1088
1089/* Gather all references from inodes and drop them */
1090static void drop_dquot_ref(struct super_block *sb, int type)
1091{
1092        LIST_HEAD(tofree_head);
1093
1094        if (sb->dq_op) {
1095                remove_dquot_ref(sb, type, &tofree_head);
1096                synchronize_srcu(&dquot_srcu);
1097                put_dquot_list(&tofree_head);
1098        }
1099}
1100
1101static inline
1102void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1103{
1104        if (dquot->dq_dqb.dqb_rsvspace >= number)
1105                dquot->dq_dqb.dqb_rsvspace -= number;
1106        else {
1107                WARN_ON_ONCE(1);
1108                dquot->dq_dqb.dqb_rsvspace = 0;
1109        }
1110        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1111            dquot->dq_dqb.dqb_bsoftlimit)
1112                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1113        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1114}
1115
1116static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1117{
1118        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1119            dquot->dq_dqb.dqb_curinodes >= number)
1120                dquot->dq_dqb.dqb_curinodes -= number;
1121        else
1122                dquot->dq_dqb.dqb_curinodes = 0;
1123        if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1124                dquot->dq_dqb.dqb_itime = (time64_t) 0;
1125        clear_bit(DQ_INODES_B, &dquot->dq_flags);
1126}
1127
1128static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1129{
1130        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1131            dquot->dq_dqb.dqb_curspace >= number)
1132                dquot->dq_dqb.dqb_curspace -= number;
1133        else
1134                dquot->dq_dqb.dqb_curspace = 0;
1135        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1136            dquot->dq_dqb.dqb_bsoftlimit)
1137                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1138        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1139}
1140
1141struct dquot_warn {
1142        struct super_block *w_sb;
1143        struct kqid w_dq_id;
1144        short w_type;
1145};
1146
1147static int warning_issued(struct dquot *dquot, const int warntype)
1148{
1149        int flag = (warntype == QUOTA_NL_BHARDWARN ||
1150                warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1151                ((warntype == QUOTA_NL_IHARDWARN ||
1152                warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1153
1154        if (!flag)
1155                return 0;
1156        return test_and_set_bit(flag, &dquot->dq_flags);
1157}
1158
1159#ifdef CONFIG_PRINT_QUOTA_WARNING
1160static int flag_print_warnings = 1;
1161
1162static int need_print_warning(struct dquot_warn *warn)
1163{
1164        if (!flag_print_warnings)
1165                return 0;
1166
1167        switch (warn->w_dq_id.type) {
1168                case USRQUOTA:
1169                        return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1170                case GRPQUOTA:
1171                        return in_group_p(warn->w_dq_id.gid);
1172                case PRJQUOTA:
1173                        return 1;
1174        }
1175        return 0;
1176}
1177
1178/* Print warning to user which exceeded quota */
1179static void print_warning(struct dquot_warn *warn)
1180{
1181        char *msg = NULL;
1182        struct tty_struct *tty;
1183        int warntype = warn->w_type;
1184
1185        if (warntype == QUOTA_NL_IHARDBELOW ||
1186            warntype == QUOTA_NL_ISOFTBELOW ||
1187            warntype == QUOTA_NL_BHARDBELOW ||
1188            warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1189                return;
1190
1191        tty = get_current_tty();
1192        if (!tty)
1193                return;
1194        tty_write_message(tty, warn->w_sb->s_id);
1195        if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1196                tty_write_message(tty, ": warning, ");
1197        else
1198                tty_write_message(tty, ": write failed, ");
1199        tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1200        switch (warntype) {
1201                case QUOTA_NL_IHARDWARN:
1202                        msg = " file limit reached.\r\n";
1203                        break;
1204                case QUOTA_NL_ISOFTLONGWARN:
1205                        msg = " file quota exceeded too long.\r\n";
1206                        break;
1207                case QUOTA_NL_ISOFTWARN:
1208                        msg = " file quota exceeded.\r\n";
1209                        break;
1210                case QUOTA_NL_BHARDWARN:
1211                        msg = " block limit reached.\r\n";
1212                        break;
1213                case QUOTA_NL_BSOFTLONGWARN:
1214                        msg = " block quota exceeded too long.\r\n";
1215                        break;
1216                case QUOTA_NL_BSOFTWARN:
1217                        msg = " block quota exceeded.\r\n";
1218                        break;
1219        }
1220        tty_write_message(tty, msg);
1221        tty_kref_put(tty);
1222}
1223#endif
1224
1225static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1226                            int warntype)
1227{
1228        if (warning_issued(dquot, warntype))
1229                return;
1230        warn->w_type = warntype;
1231        warn->w_sb = dquot->dq_sb;
1232        warn->w_dq_id = dquot->dq_id;
1233}
1234
1235/*
1236 * Write warnings to the console and send warning messages over netlink.
1237 *
1238 * Note that this function can call into tty and networking code.
1239 */
1240static void flush_warnings(struct dquot_warn *warn)
1241{
1242        int i;
1243
1244        for (i = 0; i < MAXQUOTAS; i++) {
1245                if (warn[i].w_type == QUOTA_NL_NOWARN)
1246                        continue;
1247#ifdef CONFIG_PRINT_QUOTA_WARNING
1248                print_warning(&warn[i]);
1249#endif
1250                quota_send_warning(warn[i].w_dq_id,
1251                                   warn[i].w_sb->s_dev, warn[i].w_type);
1252        }
1253}
1254
1255static int ignore_hardlimit(struct dquot *dquot)
1256{
1257        struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1258
1259        return capable(CAP_SYS_RESOURCE) &&
1260               (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1261                !(info->dqi_flags & DQF_ROOT_SQUASH));
1262}
1263
1264static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1265                            struct dquot_warn *warn)
1266{
1267        qsize_t newinodes;
1268        int ret = 0;
1269
1270        spin_lock(&dquot->dq_dqb_lock);
1271        newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1272        if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1273            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1274                goto add;
1275
1276        if (dquot->dq_dqb.dqb_ihardlimit &&
1277            newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1278            !ignore_hardlimit(dquot)) {
1279                prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1280                ret = -EDQUOT;
1281                goto out;
1282        }
1283
1284        if (dquot->dq_dqb.dqb_isoftlimit &&
1285            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1286            dquot->dq_dqb.dqb_itime &&
1287            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1288            !ignore_hardlimit(dquot)) {
1289                prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1290                ret = -EDQUOT;
1291                goto out;
1292        }
1293
1294        if (dquot->dq_dqb.dqb_isoftlimit &&
1295            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1296            dquot->dq_dqb.dqb_itime == 0) {
1297                prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1298                dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1299                    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1300        }
1301add:
1302        dquot->dq_dqb.dqb_curinodes = newinodes;
1303
1304out:
1305        spin_unlock(&dquot->dq_dqb_lock);
1306        return ret;
1307}
1308
1309static int dquot_add_space(struct dquot *dquot, qsize_t space,
1310                           qsize_t rsv_space, unsigned int flags,
1311                           struct dquot_warn *warn)
1312{
1313        qsize_t tspace;
1314        struct super_block *sb = dquot->dq_sb;
1315        int ret = 0;
1316
1317        spin_lock(&dquot->dq_dqb_lock);
1318        if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1319            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1320                goto finish;
1321
1322        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1323                + space + rsv_space;
1324
1325        if (dquot->dq_dqb.dqb_bhardlimit &&
1326            tspace > dquot->dq_dqb.dqb_bhardlimit &&
1327            !ignore_hardlimit(dquot)) {
1328                if (flags & DQUOT_SPACE_WARN)
1329                        prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1330                ret = -EDQUOT;
1331                goto finish;
1332        }
1333
1334        if (dquot->dq_dqb.dqb_bsoftlimit &&
1335            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1336            dquot->dq_dqb.dqb_btime &&
1337            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1338            !ignore_hardlimit(dquot)) {
1339                if (flags & DQUOT_SPACE_WARN)
1340                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1341                ret = -EDQUOT;
1342                goto finish;
1343        }
1344
1345        if (dquot->dq_dqb.dqb_bsoftlimit &&
1346            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1347            dquot->dq_dqb.dqb_btime == 0) {
1348                if (flags & DQUOT_SPACE_WARN) {
1349                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1350                        dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1351                            sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1352                } else {
1353                        /*
1354                         * We don't allow preallocation to exceed softlimit so exceeding will
1355                         * be always printed
1356                         */
1357                        ret = -EDQUOT;
1358                        goto finish;
1359                }
1360        }
1361finish:
1362        /*
1363         * We have to be careful and go through warning generation & grace time
1364         * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1365         * only here...
1366         */
1367        if (flags & DQUOT_SPACE_NOFAIL)
1368                ret = 0;
1369        if (!ret) {
1370                dquot->dq_dqb.dqb_rsvspace += rsv_space;
1371                dquot->dq_dqb.dqb_curspace += space;
1372        }
1373        spin_unlock(&dquot->dq_dqb_lock);
1374        return ret;
1375}
1376
1377static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1378{
1379        qsize_t newinodes;
1380
1381        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1382            dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1383            !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1384                return QUOTA_NL_NOWARN;
1385
1386        newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1387        if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1388                return QUOTA_NL_ISOFTBELOW;
1389        if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1390            newinodes < dquot->dq_dqb.dqb_ihardlimit)
1391                return QUOTA_NL_IHARDBELOW;
1392        return QUOTA_NL_NOWARN;
1393}
1394
1395static int info_bdq_free(struct dquot *dquot, qsize_t space)
1396{
1397        qsize_t tspace;
1398
1399        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1400
1401        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1402            tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1403                return QUOTA_NL_NOWARN;
1404
1405        if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1406                return QUOTA_NL_BSOFTBELOW;
1407        if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1408            tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1409                return QUOTA_NL_BHARDBELOW;
1410        return QUOTA_NL_NOWARN;
1411}
1412
1413static int dquot_active(const struct inode *inode)
1414{
1415        struct super_block *sb = inode->i_sb;
1416
1417        if (IS_NOQUOTA(inode))
1418                return 0;
1419        return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1420}
1421
1422/*
1423 * Initialize quota pointers in inode
1424 *
1425 * It is better to call this function outside of any transaction as it
1426 * might need a lot of space in journal for dquot structure allocation.
1427 */
1428static int __dquot_initialize(struct inode *inode, int type)
1429{
1430        int cnt, init_needed = 0;
1431        struct dquot **dquots, *got[MAXQUOTAS] = {};
1432        struct super_block *sb = inode->i_sb;
1433        qsize_t rsv;
1434        int ret = 0;
1435
1436        if (!dquot_active(inode))
1437                return 0;
1438
1439        dquots = i_dquot(inode);
1440
1441        /* First get references to structures we might need. */
1442        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1443                struct kqid qid;
1444                kprojid_t projid;
1445                int rc;
1446                struct dquot *dquot;
1447
1448                if (type != -1 && cnt != type)
1449                        continue;
1450                /*
1451                 * The i_dquot should have been initialized in most cases,
1452                 * we check it without locking here to avoid unnecessary
1453                 * dqget()/dqput() calls.
1454                 */
1455                if (dquots[cnt])
1456                        continue;
1457
1458                if (!sb_has_quota_active(sb, cnt))
1459                        continue;
1460
1461                init_needed = 1;
1462
1463                switch (cnt) {
1464                case USRQUOTA:
1465                        qid = make_kqid_uid(inode->i_uid);
1466                        break;
1467                case GRPQUOTA:
1468                        qid = make_kqid_gid(inode->i_gid);
1469                        break;
1470                case PRJQUOTA:
1471                        rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1472                        if (rc)
1473                                continue;
1474                        qid = make_kqid_projid(projid);
1475                        break;
1476                }
1477                dquot = dqget(sb, qid);
1478                if (IS_ERR(dquot)) {
1479                        /* We raced with somebody turning quotas off... */
1480                        if (PTR_ERR(dquot) != -ESRCH) {
1481                                ret = PTR_ERR(dquot);
1482                                goto out_put;
1483                        }
1484                        dquot = NULL;
1485                }
1486                got[cnt] = dquot;
1487        }
1488
1489        /* All required i_dquot has been initialized */
1490        if (!init_needed)
1491                return 0;
1492
1493        spin_lock(&dq_data_lock);
1494        if (IS_NOQUOTA(inode))
1495                goto out_lock;
1496        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1497                if (type != -1 && cnt != type)
1498                        continue;
1499                /* Avoid races with quotaoff() */
1500                if (!sb_has_quota_active(sb, cnt))
1501                        continue;
1502                /* We could race with quotaon or dqget() could have failed */
1503                if (!got[cnt])
1504                        continue;
1505                if (!dquots[cnt]) {
1506                        dquots[cnt] = got[cnt];
1507                        got[cnt] = NULL;
1508                        /*
1509                         * Make quota reservation system happy if someone
1510                         * did a write before quota was turned on
1511                         */
1512                        rsv = inode_get_rsv_space(inode);
1513                        if (unlikely(rsv)) {
1514                                spin_lock(&inode->i_lock);
1515                                /* Get reservation again under proper lock */
1516                                rsv = __inode_get_rsv_space(inode);
1517                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1518                                dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1519                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1520                                spin_unlock(&inode->i_lock);
1521                        }
1522                }
1523        }
1524out_lock:
1525        spin_unlock(&dq_data_lock);
1526out_put:
1527        /* Drop unused references */
1528        dqput_all(got);
1529
1530        return ret;
1531}
1532
1533int dquot_initialize(struct inode *inode)
1534{
1535        return __dquot_initialize(inode, -1);
1536}
1537EXPORT_SYMBOL(dquot_initialize);
1538
1539bool dquot_initialize_needed(struct inode *inode)
1540{
1541        struct dquot **dquots;
1542        int i;
1543
1544        if (!dquot_active(inode))
1545                return false;
1546
1547        dquots = i_dquot(inode);
1548        for (i = 0; i < MAXQUOTAS; i++)
1549                if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1550                        return true;
1551        return false;
1552}
1553EXPORT_SYMBOL(dquot_initialize_needed);
1554
1555/*
1556 * Release all quotas referenced by inode.
1557 *
1558 * This function only be called on inode free or converting
1559 * a file to quota file, no other users for the i_dquot in
1560 * both cases, so we needn't call synchronize_srcu() after
1561 * clearing i_dquot.
1562 */
1563static void __dquot_drop(struct inode *inode)
1564{
1565        int cnt;
1566        struct dquot **dquots = i_dquot(inode);
1567        struct dquot *put[MAXQUOTAS];
1568
1569        spin_lock(&dq_data_lock);
1570        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1571                put[cnt] = dquots[cnt];
1572                dquots[cnt] = NULL;
1573        }
1574        spin_unlock(&dq_data_lock);
1575        dqput_all(put);
1576}
1577
1578void dquot_drop(struct inode *inode)
1579{
1580        struct dquot * const *dquots;
1581        int cnt;
1582
1583        if (IS_NOQUOTA(inode))
1584                return;
1585
1586        /*
1587         * Test before calling to rule out calls from proc and such
1588         * where we are not allowed to block. Note that this is
1589         * actually reliable test even without the lock - the caller
1590         * must assure that nobody can come after the DQUOT_DROP and
1591         * add quota pointers back anyway.
1592         */
1593        dquots = i_dquot(inode);
1594        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1595                if (dquots[cnt])
1596                        break;
1597        }
1598
1599        if (cnt < MAXQUOTAS)
1600                __dquot_drop(inode);
1601}
1602EXPORT_SYMBOL(dquot_drop);
1603
1604/*
1605 * inode_reserved_space is managed internally by quota, and protected by
1606 * i_lock similar to i_blocks+i_bytes.
1607 */
1608static qsize_t *inode_reserved_space(struct inode * inode)
1609{
1610        /* Filesystem must explicitly define it's own method in order to use
1611         * quota reservation interface */
1612        BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1613        return inode->i_sb->dq_op->get_reserved_space(inode);
1614}
1615
1616static qsize_t __inode_get_rsv_space(struct inode *inode)
1617{
1618        if (!inode->i_sb->dq_op->get_reserved_space)
1619                return 0;
1620        return *inode_reserved_space(inode);
1621}
1622
1623static qsize_t inode_get_rsv_space(struct inode *inode)
1624{
1625        qsize_t ret;
1626
1627        if (!inode->i_sb->dq_op->get_reserved_space)
1628                return 0;
1629        spin_lock(&inode->i_lock);
1630        ret = __inode_get_rsv_space(inode);
1631        spin_unlock(&inode->i_lock);
1632        return ret;
1633}
1634
1635/*
1636 * This functions updates i_blocks+i_bytes fields and quota information
1637 * (together with appropriate checks).
1638 *
1639 * NOTE: We absolutely rely on the fact that caller dirties the inode
1640 * (usually helpers in quotaops.h care about this) and holds a handle for
1641 * the current transaction so that dquot write and inode write go into the
1642 * same transaction.
1643 */
1644
1645/*
1646 * This operation can block, but only after everything is updated
1647 */
1648int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1649{
1650        int cnt, ret = 0, index;
1651        struct dquot_warn warn[MAXQUOTAS];
1652        int reserve = flags & DQUOT_SPACE_RESERVE;
1653        struct dquot **dquots;
1654
1655        if (!dquot_active(inode)) {
1656                if (reserve) {
1657                        spin_lock(&inode->i_lock);
1658                        *inode_reserved_space(inode) += number;
1659                        spin_unlock(&inode->i_lock);
1660                } else {
1661                        inode_add_bytes(inode, number);
1662                }
1663                goto out;
1664        }
1665
1666        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1667                warn[cnt].w_type = QUOTA_NL_NOWARN;
1668
1669        dquots = i_dquot(inode);
1670        index = srcu_read_lock(&dquot_srcu);
1671        spin_lock(&inode->i_lock);
1672        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1673                if (!dquots[cnt])
1674                        continue;
1675                if (reserve) {
1676                        ret = dquot_add_space(dquots[cnt], 0, number, flags,
1677                                              &warn[cnt]);
1678                } else {
1679                        ret = dquot_add_space(dquots[cnt], number, 0, flags,
1680                                              &warn[cnt]);
1681                }
1682                if (ret) {
1683                        /* Back out changes we already did */
1684                        for (cnt--; cnt >= 0; cnt--) {
1685                                if (!dquots[cnt])
1686                                        continue;
1687                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1688                                if (reserve)
1689                                        dquot_free_reserved_space(dquots[cnt],
1690                                                                  number);
1691                                else
1692                                        dquot_decr_space(dquots[cnt], number);
1693                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1694                        }
1695                        spin_unlock(&inode->i_lock);
1696                        goto out_flush_warn;
1697                }
1698        }
1699        if (reserve)
1700                *inode_reserved_space(inode) += number;
1701        else
1702                __inode_add_bytes(inode, number);
1703        spin_unlock(&inode->i_lock);
1704
1705        if (reserve)
1706                goto out_flush_warn;
1707        mark_all_dquot_dirty(dquots);
1708out_flush_warn:
1709        srcu_read_unlock(&dquot_srcu, index);
1710        flush_warnings(warn);
1711out:
1712        return ret;
1713}
1714EXPORT_SYMBOL(__dquot_alloc_space);
1715
1716/*
1717 * This operation can block, but only after everything is updated
1718 */
1719int dquot_alloc_inode(struct inode *inode)
1720{
1721        int cnt, ret = 0, index;
1722        struct dquot_warn warn[MAXQUOTAS];
1723        struct dquot * const *dquots;
1724
1725        if (!dquot_active(inode))
1726                return 0;
1727        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1728                warn[cnt].w_type = QUOTA_NL_NOWARN;
1729
1730        dquots = i_dquot(inode);
1731        index = srcu_read_lock(&dquot_srcu);
1732        spin_lock(&inode->i_lock);
1733        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1734                if (!dquots[cnt])
1735                        continue;
1736                ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1737                if (ret) {
1738                        for (cnt--; cnt >= 0; cnt--) {
1739                                if (!dquots[cnt])
1740                                        continue;
1741                                /* Back out changes we already did */
1742                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1743                                dquot_decr_inodes(dquots[cnt], 1);
1744                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1745                        }
1746                        goto warn_put_all;
1747                }
1748        }
1749
1750warn_put_all:
1751        spin_unlock(&inode->i_lock);
1752        if (ret == 0)
1753                mark_all_dquot_dirty(dquots);
1754        srcu_read_unlock(&dquot_srcu, index);
1755        flush_warnings(warn);
1756        return ret;
1757}
1758EXPORT_SYMBOL(dquot_alloc_inode);
1759
1760/*
1761 * Convert in-memory reserved quotas to real consumed quotas
1762 */
1763int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1764{
1765        struct dquot **dquots;
1766        int cnt, index;
1767
1768        if (!dquot_active(inode)) {
1769                spin_lock(&inode->i_lock);
1770                *inode_reserved_space(inode) -= number;
1771                __inode_add_bytes(inode, number);
1772                spin_unlock(&inode->i_lock);
1773                return 0;
1774        }
1775
1776        dquots = i_dquot(inode);
1777        index = srcu_read_lock(&dquot_srcu);
1778        spin_lock(&inode->i_lock);
1779        /* Claim reserved quotas to allocated quotas */
1780        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1781                if (dquots[cnt]) {
1782                        struct dquot *dquot = dquots[cnt];
1783
1784                        spin_lock(&dquot->dq_dqb_lock);
1785                        if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1786                                number = dquot->dq_dqb.dqb_rsvspace;
1787                        dquot->dq_dqb.dqb_curspace += number;
1788                        dquot->dq_dqb.dqb_rsvspace -= number;
1789                        spin_unlock(&dquot->dq_dqb_lock);
1790                }
1791        }
1792        /* Update inode bytes */
1793        *inode_reserved_space(inode) -= number;
1794        __inode_add_bytes(inode, number);
1795        spin_unlock(&inode->i_lock);
1796        mark_all_dquot_dirty(dquots);
1797        srcu_read_unlock(&dquot_srcu, index);
1798        return 0;
1799}
1800EXPORT_SYMBOL(dquot_claim_space_nodirty);
1801
1802/*
1803 * Convert allocated space back to in-memory reserved quotas
1804 */
1805void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1806{
1807        struct dquot **dquots;
1808        int cnt, index;
1809
1810        if (!dquot_active(inode)) {
1811                spin_lock(&inode->i_lock);
1812                *inode_reserved_space(inode) += number;
1813                __inode_sub_bytes(inode, number);
1814                spin_unlock(&inode->i_lock);
1815                return;
1816        }
1817
1818        dquots = i_dquot(inode);
1819        index = srcu_read_lock(&dquot_srcu);
1820        spin_lock(&inode->i_lock);
1821        /* Claim reserved quotas to allocated quotas */
1822        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1823                if (dquots[cnt]) {
1824                        struct dquot *dquot = dquots[cnt];
1825
1826                        spin_lock(&dquot->dq_dqb_lock);
1827                        if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1828                                number = dquot->dq_dqb.dqb_curspace;
1829                        dquot->dq_dqb.dqb_rsvspace += number;
1830                        dquot->dq_dqb.dqb_curspace -= number;
1831                        spin_unlock(&dquot->dq_dqb_lock);
1832                }
1833        }
1834        /* Update inode bytes */
1835        *inode_reserved_space(inode) += number;
1836        __inode_sub_bytes(inode, number);
1837        spin_unlock(&inode->i_lock);
1838        mark_all_dquot_dirty(dquots);
1839        srcu_read_unlock(&dquot_srcu, index);
1840        return;
1841}
1842EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1843
1844/*
1845 * This operation can block, but only after everything is updated
1846 */
1847void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1848{
1849        unsigned int cnt;
1850        struct dquot_warn warn[MAXQUOTAS];
1851        struct dquot **dquots;
1852        int reserve = flags & DQUOT_SPACE_RESERVE, index;
1853
1854        if (!dquot_active(inode)) {
1855                if (reserve) {
1856                        spin_lock(&inode->i_lock);
1857                        *inode_reserved_space(inode) -= number;
1858                        spin_unlock(&inode->i_lock);
1859                } else {
1860                        inode_sub_bytes(inode, number);
1861                }
1862                return;
1863        }
1864
1865        dquots = i_dquot(inode);
1866        index = srcu_read_lock(&dquot_srcu);
1867        spin_lock(&inode->i_lock);
1868        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1869                int wtype;
1870
1871                warn[cnt].w_type = QUOTA_NL_NOWARN;
1872                if (!dquots[cnt])
1873                        continue;
1874                spin_lock(&dquots[cnt]->dq_dqb_lock);
1875                wtype = info_bdq_free(dquots[cnt], number);
1876                if (wtype != QUOTA_NL_NOWARN)
1877                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1878                if (reserve)
1879                        dquot_free_reserved_space(dquots[cnt], number);
1880                else
1881                        dquot_decr_space(dquots[cnt], number);
1882                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1883        }
1884        if (reserve)
1885                *inode_reserved_space(inode) -= number;
1886        else
1887                __inode_sub_bytes(inode, number);
1888        spin_unlock(&inode->i_lock);
1889
1890        if (reserve)
1891                goto out_unlock;
1892        mark_all_dquot_dirty(dquots);
1893out_unlock:
1894        srcu_read_unlock(&dquot_srcu, index);
1895        flush_warnings(warn);
1896}
1897EXPORT_SYMBOL(__dquot_free_space);
1898
1899/*
1900 * This operation can block, but only after everything is updated
1901 */
1902void dquot_free_inode(struct inode *inode)
1903{
1904        unsigned int cnt;
1905        struct dquot_warn warn[MAXQUOTAS];
1906        struct dquot * const *dquots;
1907        int index;
1908
1909        if (!dquot_active(inode))
1910                return;
1911
1912        dquots = i_dquot(inode);
1913        index = srcu_read_lock(&dquot_srcu);
1914        spin_lock(&inode->i_lock);
1915        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1916                int wtype;
1917
1918                warn[cnt].w_type = QUOTA_NL_NOWARN;
1919                if (!dquots[cnt])
1920                        continue;
1921                spin_lock(&dquots[cnt]->dq_dqb_lock);
1922                wtype = info_idq_free(dquots[cnt], 1);
1923                if (wtype != QUOTA_NL_NOWARN)
1924                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1925                dquot_decr_inodes(dquots[cnt], 1);
1926                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1927        }
1928        spin_unlock(&inode->i_lock);
1929        mark_all_dquot_dirty(dquots);
1930        srcu_read_unlock(&dquot_srcu, index);
1931        flush_warnings(warn);
1932}
1933EXPORT_SYMBOL(dquot_free_inode);
1934
1935/*
1936 * Transfer the number of inode and blocks from one diskquota to an other.
1937 * On success, dquot references in transfer_to are consumed and references
1938 * to original dquots that need to be released are placed there. On failure,
1939 * references are kept untouched.
1940 *
1941 * This operation can block, but only after everything is updated
1942 * A transaction must be started when entering this function.
1943 *
1944 * We are holding reference on transfer_from & transfer_to, no need to
1945 * protect them by srcu_read_lock().
1946 */
1947int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1948{
1949        qsize_t cur_space;
1950        qsize_t rsv_space = 0;
1951        qsize_t inode_usage = 1;
1952        struct dquot *transfer_from[MAXQUOTAS] = {};
1953        int cnt, ret = 0;
1954        char is_valid[MAXQUOTAS] = {};
1955        struct dquot_warn warn_to[MAXQUOTAS];
1956        struct dquot_warn warn_from_inodes[MAXQUOTAS];
1957        struct dquot_warn warn_from_space[MAXQUOTAS];
1958
1959        if (IS_NOQUOTA(inode))
1960                return 0;
1961
1962        if (inode->i_sb->dq_op->get_inode_usage) {
1963                ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1964                if (ret)
1965                        return ret;
1966        }
1967
1968        /* Initialize the arrays */
1969        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1970                warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1971                warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1972                warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1973        }
1974
1975        spin_lock(&dq_data_lock);
1976        spin_lock(&inode->i_lock);
1977        if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1978                spin_unlock(&inode->i_lock);
1979                spin_unlock(&dq_data_lock);
1980                return 0;
1981        }
1982        cur_space = __inode_get_bytes(inode);
1983        rsv_space = __inode_get_rsv_space(inode);
1984        /*
1985         * Build the transfer_from list, check limits, and update usage in
1986         * the target structures.
1987         */
1988        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1989                /*
1990                 * Skip changes for same uid or gid or for turned off quota-type.
1991                 */
1992                if (!transfer_to[cnt])
1993                        continue;
1994                /* Avoid races with quotaoff() */
1995                if (!sb_has_quota_active(inode->i_sb, cnt))
1996                        continue;
1997                is_valid[cnt] = 1;
1998                transfer_from[cnt] = i_dquot(inode)[cnt];
1999                ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
2000                                       &warn_to[cnt]);
2001                if (ret)
2002                        goto over_quota;
2003                ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2004                                      DQUOT_SPACE_WARN, &warn_to[cnt]);
2005                if (ret) {
2006                        spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2007                        dquot_decr_inodes(transfer_to[cnt], inode_usage);
2008                        spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2009                        goto over_quota;
2010                }
2011        }
2012
2013        /* Decrease usage for source structures and update quota pointers */
2014        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2015                if (!is_valid[cnt])
2016                        continue;
2017                /* Due to IO error we might not have transfer_from[] structure */
2018                if (transfer_from[cnt]) {
2019                        int wtype;
2020
2021                        spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2022                        wtype = info_idq_free(transfer_from[cnt], inode_usage);
2023                        if (wtype != QUOTA_NL_NOWARN)
2024                                prepare_warning(&warn_from_inodes[cnt],
2025                                                transfer_from[cnt], wtype);
2026                        wtype = info_bdq_free(transfer_from[cnt],
2027                                              cur_space + rsv_space);
2028                        if (wtype != QUOTA_NL_NOWARN)
2029                                prepare_warning(&warn_from_space[cnt],
2030                                                transfer_from[cnt], wtype);
2031                        dquot_decr_inodes(transfer_from[cnt], inode_usage);
2032                        dquot_decr_space(transfer_from[cnt], cur_space);
2033                        dquot_free_reserved_space(transfer_from[cnt],
2034                                                  rsv_space);
2035                        spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2036                }
2037                i_dquot(inode)[cnt] = transfer_to[cnt];
2038        }
2039        spin_unlock(&inode->i_lock);
2040        spin_unlock(&dq_data_lock);
2041
2042        mark_all_dquot_dirty(transfer_from);
2043        mark_all_dquot_dirty(transfer_to);
2044        flush_warnings(warn_to);
2045        flush_warnings(warn_from_inodes);
2046        flush_warnings(warn_from_space);
2047        /* Pass back references to put */
2048        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2049                if (is_valid[cnt])
2050                        transfer_to[cnt] = transfer_from[cnt];
2051        return 0;
2052over_quota:
2053        /* Back out changes we already did */
2054        for (cnt--; cnt >= 0; cnt--) {
2055                if (!is_valid[cnt])
2056                        continue;
2057                spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2058                dquot_decr_inodes(transfer_to[cnt], inode_usage);
2059                dquot_decr_space(transfer_to[cnt], cur_space);
2060                dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2061                spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2062        }
2063        spin_unlock(&inode->i_lock);
2064        spin_unlock(&dq_data_lock);
2065        flush_warnings(warn_to);
2066        return ret;
2067}
2068EXPORT_SYMBOL(__dquot_transfer);
2069
2070/* Wrapper for transferring ownership of an inode for uid/gid only
2071 * Called from FSXXX_setattr()
2072 */
2073int dquot_transfer(struct inode *inode, struct iattr *iattr)
2074{
2075        struct dquot *transfer_to[MAXQUOTAS] = {};
2076        struct dquot *dquot;
2077        struct super_block *sb = inode->i_sb;
2078        int ret;
2079
2080        if (!dquot_active(inode))
2081                return 0;
2082
2083        if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2084                dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2085                if (IS_ERR(dquot)) {
2086                        if (PTR_ERR(dquot) != -ESRCH) {
2087                                ret = PTR_ERR(dquot);
2088                                goto out_put;
2089                        }
2090                        dquot = NULL;
2091                }
2092                transfer_to[USRQUOTA] = dquot;
2093        }
2094        if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2095                dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2096                if (IS_ERR(dquot)) {
2097                        if (PTR_ERR(dquot) != -ESRCH) {
2098                                ret = PTR_ERR(dquot);
2099                                goto out_put;
2100                        }
2101                        dquot = NULL;
2102                }
2103                transfer_to[GRPQUOTA] = dquot;
2104        }
2105        ret = __dquot_transfer(inode, transfer_to);
2106out_put:
2107        dqput_all(transfer_to);
2108        return ret;
2109}
2110EXPORT_SYMBOL(dquot_transfer);
2111
2112/*
2113 * Write info of quota file to disk
2114 */
2115int dquot_commit_info(struct super_block *sb, int type)
2116{
2117        struct quota_info *dqopt = sb_dqopt(sb);
2118
2119        return dqopt->ops[type]->write_file_info(sb, type);
2120}
2121EXPORT_SYMBOL(dquot_commit_info);
2122
2123int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2124{
2125        struct quota_info *dqopt = sb_dqopt(sb);
2126
2127        if (!sb_has_quota_active(sb, qid->type))
2128                return -ESRCH;
2129        if (!dqopt->ops[qid->type]->get_next_id)
2130                return -ENOSYS;
2131        return dqopt->ops[qid->type]->get_next_id(sb, qid);
2132}
2133EXPORT_SYMBOL(dquot_get_next_id);
2134
2135/*
2136 * Definitions of diskquota operations.
2137 */
2138const struct dquot_operations dquot_operations = {
2139        .write_dquot    = dquot_commit,
2140        .acquire_dquot  = dquot_acquire,
2141        .release_dquot  = dquot_release,
2142        .mark_dirty     = dquot_mark_dquot_dirty,
2143        .write_info     = dquot_commit_info,
2144        .alloc_dquot    = dquot_alloc,
2145        .destroy_dquot  = dquot_destroy,
2146        .get_next_id    = dquot_get_next_id,
2147};
2148EXPORT_SYMBOL(dquot_operations);
2149
2150/*
2151 * Generic helper for ->open on filesystems supporting disk quotas.
2152 */
2153int dquot_file_open(struct inode *inode, struct file *file)
2154{
2155        int error;
2156
2157        error = generic_file_open(inode, file);
2158        if (!error && (file->f_mode & FMODE_WRITE))
2159                error = dquot_initialize(inode);
2160        return error;
2161}
2162EXPORT_SYMBOL(dquot_file_open);
2163
2164static void vfs_cleanup_quota_inode(struct super_block *sb, int type)
2165{
2166        struct quota_info *dqopt = sb_dqopt(sb);
2167        struct inode *inode = dqopt->files[type];
2168
2169        if (!inode)
2170                return;
2171        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2172                inode_lock(inode);
2173                inode->i_flags &= ~S_NOQUOTA;
2174                inode_unlock(inode);
2175        }
2176        dqopt->files[type] = NULL;
2177        iput(inode);
2178}
2179
2180/*
2181 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2182 */
2183int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2184{
2185        int cnt;
2186        struct quota_info *dqopt = sb_dqopt(sb);
2187
2188        /* s_umount should be held in exclusive mode */
2189        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2190                up_read(&sb->s_umount);
2191
2192        /* Cannot turn off usage accounting without turning off limits, or
2193         * suspend quotas and simultaneously turn quotas off. */
2194        if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2195            || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2196            DQUOT_USAGE_ENABLED)))
2197                return -EINVAL;
2198
2199        /*
2200         * Skip everything if there's nothing to do. We have to do this because
2201         * sometimes we are called when fill_super() failed and calling
2202         * sync_fs() in such cases does no good.
2203         */
2204        if (!sb_any_quota_loaded(sb))
2205                return 0;
2206
2207        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2208                if (type != -1 && cnt != type)
2209                        continue;
2210                if (!sb_has_quota_loaded(sb, cnt))
2211                        continue;
2212
2213                if (flags & DQUOT_SUSPENDED) {
2214                        spin_lock(&dq_state_lock);
2215                        dqopt->flags |=
2216                                dquot_state_flag(DQUOT_SUSPENDED, cnt);
2217                        spin_unlock(&dq_state_lock);
2218                } else {
2219                        spin_lock(&dq_state_lock);
2220                        dqopt->flags &= ~dquot_state_flag(flags, cnt);
2221                        /* Turning off suspended quotas? */
2222                        if (!sb_has_quota_loaded(sb, cnt) &&
2223                            sb_has_quota_suspended(sb, cnt)) {
2224                                dqopt->flags &= ~dquot_state_flag(
2225                                                        DQUOT_SUSPENDED, cnt);
2226                                spin_unlock(&dq_state_lock);
2227                                vfs_cleanup_quota_inode(sb, cnt);
2228                                continue;
2229                        }
2230                        spin_unlock(&dq_state_lock);
2231                }
2232
2233                /* We still have to keep quota loaded? */
2234                if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2235                        continue;
2236
2237                /* Note: these are blocking operations */
2238                drop_dquot_ref(sb, cnt);
2239                invalidate_dquots(sb, cnt);
2240                /*
2241                 * Now all dquots should be invalidated, all writes done so we
2242                 * should be only users of the info. No locks needed.
2243                 */
2244                if (info_dirty(&dqopt->info[cnt]))
2245                        sb->dq_op->write_info(sb, cnt);
2246                if (dqopt->ops[cnt]->free_file_info)
2247                        dqopt->ops[cnt]->free_file_info(sb, cnt);
2248                put_quota_format(dqopt->info[cnt].dqi_format);
2249                dqopt->info[cnt].dqi_flags = 0;
2250                dqopt->info[cnt].dqi_igrace = 0;
2251                dqopt->info[cnt].dqi_bgrace = 0;
2252                dqopt->ops[cnt] = NULL;
2253        }
2254
2255        /* Skip syncing and setting flags if quota files are hidden */
2256        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2257                goto put_inodes;
2258
2259        /* Sync the superblock so that buffers with quota data are written to
2260         * disk (and so userspace sees correct data afterwards). */
2261        if (sb->s_op->sync_fs)
2262                sb->s_op->sync_fs(sb, 1);
2263        sync_blockdev(sb->s_bdev);
2264        /* Now the quota files are just ordinary files and we can set the
2265         * inode flags back. Moreover we discard the pagecache so that
2266         * userspace sees the writes we did bypassing the pagecache. We
2267         * must also discard the blockdev buffers so that we see the
2268         * changes done by userspace on the next quotaon() */
2269        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2270                if (!sb_has_quota_loaded(sb, cnt) && dqopt->files[cnt]) {
2271                        inode_lock(dqopt->files[cnt]);
2272                        truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
2273                        inode_unlock(dqopt->files[cnt]);
2274                }
2275        if (sb->s_bdev)
2276                invalidate_bdev(sb->s_bdev);
2277put_inodes:
2278        /* We are done when suspending quotas */
2279        if (flags & DQUOT_SUSPENDED)
2280                return 0;
2281
2282        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2283                if (!sb_has_quota_loaded(sb, cnt))
2284                        vfs_cleanup_quota_inode(sb, cnt);
2285        return 0;
2286}
2287EXPORT_SYMBOL(dquot_disable);
2288
2289int dquot_quota_off(struct super_block *sb, int type)
2290{
2291        return dquot_disable(sb, type,
2292                             DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2293}
2294EXPORT_SYMBOL(dquot_quota_off);
2295
2296/*
2297 *      Turn quotas on on a device
2298 */
2299
2300static int vfs_setup_quota_inode(struct inode *inode, int type)
2301{
2302        struct super_block *sb = inode->i_sb;
2303        struct quota_info *dqopt = sb_dqopt(sb);
2304
2305        if (!S_ISREG(inode->i_mode))
2306                return -EACCES;
2307        if (IS_RDONLY(inode))
2308                return -EROFS;
2309        if (sb_has_quota_loaded(sb, type))
2310                return -EBUSY;
2311
2312        dqopt->files[type] = igrab(inode);
2313        if (!dqopt->files[type])
2314                return -EIO;
2315        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2316                /* We don't want quota and atime on quota files (deadlocks
2317                 * possible) Also nobody should write to the file - we use
2318                 * special IO operations which ignore the immutable bit. */
2319                inode_lock(inode);
2320                inode->i_flags |= S_NOQUOTA;
2321                inode_unlock(inode);
2322                /*
2323                 * When S_NOQUOTA is set, remove dquot references as no more
2324                 * references can be added
2325                 */
2326                __dquot_drop(inode);
2327        }
2328        return 0;
2329}
2330
2331int dquot_load_quota_sb(struct super_block *sb, int type, int format_id,
2332        unsigned int flags)
2333{
2334        struct quota_format_type *fmt = find_quota_format(format_id);
2335        struct quota_info *dqopt = sb_dqopt(sb);
2336        int error;
2337
2338        /* Just unsuspend quotas? */
2339        BUG_ON(flags & DQUOT_SUSPENDED);
2340        /* s_umount should be held in exclusive mode */
2341        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2342                up_read(&sb->s_umount);
2343
2344        if (!fmt)
2345                return -ESRCH;
2346        if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2347            (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2348                error = -EINVAL;
2349                goto out_fmt;
2350        }
2351        /* Filesystems outside of init_user_ns not yet supported */
2352        if (sb->s_user_ns != &init_user_ns) {
2353                error = -EINVAL;
2354                goto out_fmt;
2355        }
2356        /* Usage always has to be set... */
2357        if (!(flags & DQUOT_USAGE_ENABLED)) {
2358                error = -EINVAL;
2359                goto out_fmt;
2360        }
2361        if (sb_has_quota_loaded(sb, type)) {
2362                error = -EBUSY;
2363                goto out_fmt;
2364        }
2365
2366        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2367                /* As we bypass the pagecache we must now flush all the
2368                 * dirty data and invalidate caches so that kernel sees
2369                 * changes from userspace. It is not enough to just flush
2370                 * the quota file since if blocksize < pagesize, invalidation
2371                 * of the cache could fail because of other unrelated dirty
2372                 * data */
2373                sync_filesystem(sb);
2374                invalidate_bdev(sb->s_bdev);
2375        }
2376
2377        error = -EINVAL;
2378        if (!fmt->qf_ops->check_quota_file(sb, type))
2379                goto out_fmt;
2380
2381        dqopt->ops[type] = fmt->qf_ops;
2382        dqopt->info[type].dqi_format = fmt;
2383        dqopt->info[type].dqi_fmt_id = format_id;
2384        INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2385        error = dqopt->ops[type]->read_file_info(sb, type);
2386        if (error < 0)
2387                goto out_fmt;
2388        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2389                spin_lock(&dq_data_lock);
2390                dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2391                spin_unlock(&dq_data_lock);
2392        }
2393        spin_lock(&dq_state_lock);
2394        dqopt->flags |= dquot_state_flag(flags, type);
2395        spin_unlock(&dq_state_lock);
2396
2397        error = add_dquot_ref(sb, type);
2398        if (error)
2399                dquot_disable(sb, type, flags);
2400
2401        return error;
2402out_fmt:
2403        put_quota_format(fmt);
2404
2405        return error;
2406}
2407EXPORT_SYMBOL(dquot_load_quota_sb);
2408
2409/*
2410 * More powerful function for turning on quotas on given quota inode allowing
2411 * setting of individual quota flags
2412 */
2413int dquot_load_quota_inode(struct inode *inode, int type, int format_id,
2414        unsigned int flags)
2415{
2416        int err;
2417
2418        err = vfs_setup_quota_inode(inode, type);
2419        if (err < 0)
2420                return err;
2421        err = dquot_load_quota_sb(inode->i_sb, type, format_id, flags);
2422        if (err < 0)
2423                vfs_cleanup_quota_inode(inode->i_sb, type);
2424        return err;
2425}
2426EXPORT_SYMBOL(dquot_load_quota_inode);
2427
2428/* Reenable quotas on remount RW */
2429int dquot_resume(struct super_block *sb, int type)
2430{
2431        struct quota_info *dqopt = sb_dqopt(sb);
2432        int ret = 0, cnt;
2433        unsigned int flags;
2434
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        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2440                if (type != -1 && cnt != type)
2441                        continue;
2442                if (!sb_has_quota_suspended(sb, cnt))
2443                        continue;
2444
2445                spin_lock(&dq_state_lock);
2446                flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2447                                                        DQUOT_LIMITS_ENABLED,
2448                                                        cnt);
2449                dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2450                spin_unlock(&dq_state_lock);
2451
2452                flags = dquot_generic_flag(flags, cnt);
2453                ret = dquot_load_quota_sb(sb, cnt, dqopt->info[cnt].dqi_fmt_id,
2454                                          flags);
2455                if (ret < 0)
2456                        vfs_cleanup_quota_inode(sb, cnt);
2457        }
2458
2459        return ret;
2460}
2461EXPORT_SYMBOL(dquot_resume);
2462
2463int dquot_quota_on(struct super_block *sb, int type, int format_id,
2464                   const struct path *path)
2465{
2466        int error = security_quota_on(path->dentry);
2467        if (error)
2468                return error;
2469        /* Quota file not on the same filesystem? */
2470        if (path->dentry->d_sb != sb)
2471                error = -EXDEV;
2472        else
2473                error = dquot_load_quota_inode(d_inode(path->dentry), type,
2474                                             format_id, DQUOT_USAGE_ENABLED |
2475                                             DQUOT_LIMITS_ENABLED);
2476        return error;
2477}
2478EXPORT_SYMBOL(dquot_quota_on);
2479
2480/*
2481 * This function is used when filesystem needs to initialize quotas
2482 * during mount time.
2483 */
2484int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2485                int format_id, int type)
2486{
2487        struct dentry *dentry;
2488        int error;
2489
2490        dentry = lookup_positive_unlocked(qf_name, sb->s_root, strlen(qf_name));
2491        if (IS_ERR(dentry))
2492                return PTR_ERR(dentry);
2493
2494        error = security_quota_on(dentry);
2495        if (!error)
2496                error = dquot_load_quota_inode(d_inode(dentry), type, format_id,
2497                                DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2498
2499        dput(dentry);
2500        return error;
2501}
2502EXPORT_SYMBOL(dquot_quota_on_mount);
2503
2504static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2505{
2506        int ret;
2507        int type;
2508        struct quota_info *dqopt = sb_dqopt(sb);
2509
2510        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2511                return -ENOSYS;
2512        /* Accounting cannot be turned on while fs is mounted */
2513        flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2514        if (!flags)
2515                return -EINVAL;
2516        for (type = 0; type < MAXQUOTAS; type++) {
2517                if (!(flags & qtype_enforce_flag(type)))
2518                        continue;
2519                /* Can't enforce without accounting */
2520                if (!sb_has_quota_usage_enabled(sb, type)) {
2521                        ret = -EINVAL;
2522                        goto out_err;
2523                }
2524                if (sb_has_quota_limits_enabled(sb, type)) {
2525                        ret = -EBUSY;
2526                        goto out_err;
2527                }
2528                spin_lock(&dq_state_lock);
2529                dqopt->flags |= dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2530                spin_unlock(&dq_state_lock);
2531        }
2532        return 0;
2533out_err:
2534        /* Backout enforcement enablement we already did */
2535        for (type--; type >= 0; type--)  {
2536                if (flags & qtype_enforce_flag(type))
2537                        dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2538        }
2539        /* Error code translation for better compatibility with XFS */
2540        if (ret == -EBUSY)
2541                ret = -EEXIST;
2542        return ret;
2543}
2544
2545static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2546{
2547        int ret;
2548        int type;
2549        struct quota_info *dqopt = sb_dqopt(sb);
2550
2551        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2552                return -ENOSYS;
2553        /*
2554         * We don't support turning off accounting via quotactl. In principle
2555         * quota infrastructure can do this but filesystems don't expect
2556         * userspace to be able to do it.
2557         */
2558        if (flags &
2559                  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2560                return -EOPNOTSUPP;
2561
2562        /* Filter out limits not enabled */
2563        for (type = 0; type < MAXQUOTAS; type++)
2564                if (!sb_has_quota_limits_enabled(sb, type))
2565                        flags &= ~qtype_enforce_flag(type);
2566        /* Nothing left? */
2567        if (!flags)
2568                return -EEXIST;
2569        for (type = 0; type < MAXQUOTAS; type++) {
2570                if (flags & qtype_enforce_flag(type)) {
2571                        ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2572                        if (ret < 0)
2573                                goto out_err;
2574                }
2575        }
2576        return 0;
2577out_err:
2578        /* Backout enforcement disabling we already did */
2579        for (type--; type >= 0; type--)  {
2580                if (flags & qtype_enforce_flag(type)) {
2581                        spin_lock(&dq_state_lock);
2582                        dqopt->flags |=
2583                                dquot_state_flag(DQUOT_LIMITS_ENABLED, type);
2584                        spin_unlock(&dq_state_lock);
2585                }
2586        }
2587        return ret;
2588}
2589
2590/* Generic routine for getting common part of quota structure */
2591static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2592{
2593        struct mem_dqblk *dm = &dquot->dq_dqb;
2594
2595        memset(di, 0, sizeof(*di));
2596        spin_lock(&dquot->dq_dqb_lock);
2597        di->d_spc_hardlimit = dm->dqb_bhardlimit;
2598        di->d_spc_softlimit = dm->dqb_bsoftlimit;
2599        di->d_ino_hardlimit = dm->dqb_ihardlimit;
2600        di->d_ino_softlimit = dm->dqb_isoftlimit;
2601        di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2602        di->d_ino_count = dm->dqb_curinodes;
2603        di->d_spc_timer = dm->dqb_btime;
2604        di->d_ino_timer = dm->dqb_itime;
2605        spin_unlock(&dquot->dq_dqb_lock);
2606}
2607
2608int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2609                    struct qc_dqblk *di)
2610{
2611        struct dquot *dquot;
2612
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_dqblk);
2622
2623int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2624                         struct qc_dqblk *di)
2625{
2626        struct dquot *dquot;
2627        int err;
2628
2629        if (!sb->dq_op->get_next_id)
2630                return -ENOSYS;
2631        err = sb->dq_op->get_next_id(sb, qid);
2632        if (err < 0)
2633                return err;
2634        dquot = dqget(sb, *qid);
2635        if (IS_ERR(dquot))
2636                return PTR_ERR(dquot);
2637        do_get_dqblk(dquot, di);
2638        dqput(dquot);
2639
2640        return 0;
2641}
2642EXPORT_SYMBOL(dquot_get_next_dqblk);
2643
2644#define VFS_QC_MASK \
2645        (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2646         QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2647         QC_SPC_TIMER | QC_INO_TIMER)
2648
2649/* Generic routine for setting common part of quota structure */
2650static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2651{
2652        struct mem_dqblk *dm = &dquot->dq_dqb;
2653        int check_blim = 0, check_ilim = 0;
2654        struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2655
2656        if (di->d_fieldmask & ~VFS_QC_MASK)
2657                return -EINVAL;
2658
2659        if (((di->d_fieldmask & QC_SPC_SOFT) &&
2660             di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2661            ((di->d_fieldmask & QC_SPC_HARD) &&
2662             di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2663            ((di->d_fieldmask & QC_INO_SOFT) &&
2664             (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2665            ((di->d_fieldmask & QC_INO_HARD) &&
2666             (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2667                return -ERANGE;
2668
2669        spin_lock(&dquot->dq_dqb_lock);
2670        if (di->d_fieldmask & QC_SPACE) {
2671                dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2672                check_blim = 1;
2673                set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2674        }
2675
2676        if (di->d_fieldmask & QC_SPC_SOFT)
2677                dm->dqb_bsoftlimit = di->d_spc_softlimit;
2678        if (di->d_fieldmask & QC_SPC_HARD)
2679                dm->dqb_bhardlimit = di->d_spc_hardlimit;
2680        if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2681                check_blim = 1;
2682                set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2683        }
2684
2685        if (di->d_fieldmask & QC_INO_COUNT) {
2686                dm->dqb_curinodes = di->d_ino_count;
2687                check_ilim = 1;
2688                set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2689        }
2690
2691        if (di->d_fieldmask & QC_INO_SOFT)
2692                dm->dqb_isoftlimit = di->d_ino_softlimit;
2693        if (di->d_fieldmask & QC_INO_HARD)
2694                dm->dqb_ihardlimit = di->d_ino_hardlimit;
2695        if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2696                check_ilim = 1;
2697                set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2698        }
2699
2700        if (di->d_fieldmask & QC_SPC_TIMER) {
2701                dm->dqb_btime = di->d_spc_timer;
2702                check_blim = 1;
2703                set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2704        }
2705
2706        if (di->d_fieldmask & QC_INO_TIMER) {
2707                dm->dqb_itime = di->d_ino_timer;
2708                check_ilim = 1;
2709                set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2710        }
2711
2712        if (check_blim) {
2713                if (!dm->dqb_bsoftlimit ||
2714                    dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2715                        dm->dqb_btime = 0;
2716                        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2717                } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2718                        /* Set grace only if user hasn't provided his own... */
2719                        dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2720        }
2721        if (check_ilim) {
2722                if (!dm->dqb_isoftlimit ||
2723                    dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2724                        dm->dqb_itime = 0;
2725                        clear_bit(DQ_INODES_B, &dquot->dq_flags);
2726                } else if (!(di->d_fieldmask & QC_INO_TIMER))
2727                        /* Set grace only if user hasn't provided his own... */
2728                        dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2729        }
2730        if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2731            dm->dqb_isoftlimit)
2732                clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2733        else
2734                set_bit(DQ_FAKE_B, &dquot->dq_flags);
2735        spin_unlock(&dquot->dq_dqb_lock);
2736        mark_dquot_dirty(dquot);
2737
2738        return 0;
2739}
2740
2741int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2742                  struct qc_dqblk *di)
2743{
2744        struct dquot *dquot;
2745        int rc;
2746
2747        dquot = dqget(sb, qid);
2748        if (IS_ERR(dquot)) {
2749                rc = PTR_ERR(dquot);
2750                goto out;
2751        }
2752        rc = do_set_dqblk(dquot, di);
2753        dqput(dquot);
2754out:
2755        return rc;
2756}
2757EXPORT_SYMBOL(dquot_set_dqblk);
2758
2759/* Generic routine for getting common part of quota file information */
2760int dquot_get_state(struct super_block *sb, struct qc_state *state)
2761{
2762        struct mem_dqinfo *mi;
2763        struct qc_type_state *tstate;
2764        struct quota_info *dqopt = sb_dqopt(sb);
2765        int type;
2766
2767        memset(state, 0, sizeof(*state));
2768        for (type = 0; type < MAXQUOTAS; type++) {
2769                if (!sb_has_quota_active(sb, type))
2770                        continue;
2771                tstate = state->s_state + type;
2772                mi = sb_dqopt(sb)->info + type;
2773                tstate->flags = QCI_ACCT_ENABLED;
2774                spin_lock(&dq_data_lock);
2775                if (mi->dqi_flags & DQF_SYS_FILE)
2776                        tstate->flags |= QCI_SYSFILE;
2777                if (mi->dqi_flags & DQF_ROOT_SQUASH)
2778                        tstate->flags |= QCI_ROOT_SQUASH;
2779                if (sb_has_quota_limits_enabled(sb, type))
2780                        tstate->flags |= QCI_LIMITS_ENFORCED;
2781                tstate->spc_timelimit = mi->dqi_bgrace;
2782                tstate->ino_timelimit = mi->dqi_igrace;
2783                if (dqopt->files[type]) {
2784                        tstate->ino = dqopt->files[type]->i_ino;
2785                        tstate->blocks = dqopt->files[type]->i_blocks;
2786                }
2787                tstate->nextents = 1;   /* We don't know... */
2788                spin_unlock(&dq_data_lock);
2789        }
2790        return 0;
2791}
2792EXPORT_SYMBOL(dquot_get_state);
2793
2794/* Generic routine for setting common part of quota file information */
2795int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2796{
2797        struct mem_dqinfo *mi;
2798        int err = 0;
2799
2800        if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2801            (ii->i_fieldmask & QC_RT_SPC_TIMER))
2802                return -EINVAL;
2803        if (!sb_has_quota_active(sb, type))
2804                return -ESRCH;
2805        mi = sb_dqopt(sb)->info + type;
2806        if (ii->i_fieldmask & QC_FLAGS) {
2807                if ((ii->i_flags & QCI_ROOT_SQUASH &&
2808                     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2809                        return -EINVAL;
2810        }
2811        spin_lock(&dq_data_lock);
2812        if (ii->i_fieldmask & QC_SPC_TIMER)
2813                mi->dqi_bgrace = ii->i_spc_timelimit;
2814        if (ii->i_fieldmask & QC_INO_TIMER)
2815                mi->dqi_igrace = ii->i_ino_timelimit;
2816        if (ii->i_fieldmask & QC_FLAGS) {
2817                if (ii->i_flags & QCI_ROOT_SQUASH)
2818                        mi->dqi_flags |= DQF_ROOT_SQUASH;
2819                else
2820                        mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2821        }
2822        spin_unlock(&dq_data_lock);
2823        mark_info_dirty(sb, type);
2824        /* Force write to disk */
2825        sb->dq_op->write_info(sb, type);
2826        return err;
2827}
2828EXPORT_SYMBOL(dquot_set_dqinfo);
2829
2830const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2831        .quota_enable   = dquot_quota_enable,
2832        .quota_disable  = dquot_quota_disable,
2833        .quota_sync     = dquot_quota_sync,
2834        .get_state      = dquot_get_state,
2835        .set_info       = dquot_set_dqinfo,
2836        .get_dqblk      = dquot_get_dqblk,
2837        .get_nextdqblk  = dquot_get_next_dqblk,
2838        .set_dqblk      = dquot_set_dqblk
2839};
2840EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2841
2842static int do_proc_dqstats(struct ctl_table *table, int write,
2843                     void *buffer, size_t *lenp, loff_t *ppos)
2844{
2845        unsigned int type = (unsigned long *)table->data - dqstats.stat;
2846        s64 value = percpu_counter_sum(&dqstats.counter[type]);
2847
2848        /* Filter negative values for non-monotonic counters */
2849        if (value < 0 && (type == DQST_ALLOC_DQUOTS ||
2850                          type == DQST_FREE_DQUOTS))
2851                value = 0;
2852
2853        /* Update global table */
2854        dqstats.stat[type] = value;
2855        return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
2856}
2857
2858static struct ctl_table fs_dqstats_table[] = {
2859        {
2860                .procname       = "lookups",
2861                .data           = &dqstats.stat[DQST_LOOKUPS],
2862                .maxlen         = sizeof(unsigned long),
2863                .mode           = 0444,
2864                .proc_handler   = do_proc_dqstats,
2865        },
2866        {
2867                .procname       = "drops",
2868                .data           = &dqstats.stat[DQST_DROPS],
2869                .maxlen         = sizeof(unsigned long),
2870                .mode           = 0444,
2871                .proc_handler   = do_proc_dqstats,
2872        },
2873        {
2874                .procname       = "reads",
2875                .data           = &dqstats.stat[DQST_READS],
2876                .maxlen         = sizeof(unsigned long),
2877                .mode           = 0444,
2878                .proc_handler   = do_proc_dqstats,
2879        },
2880        {
2881                .procname       = "writes",
2882                .data           = &dqstats.stat[DQST_WRITES],
2883                .maxlen         = sizeof(unsigned long),
2884                .mode           = 0444,
2885                .proc_handler   = do_proc_dqstats,
2886        },
2887        {
2888                .procname       = "cache_hits",
2889                .data           = &dqstats.stat[DQST_CACHE_HITS],
2890                .maxlen         = sizeof(unsigned long),
2891                .mode           = 0444,
2892                .proc_handler   = do_proc_dqstats,
2893        },
2894        {
2895                .procname       = "allocated_dquots",
2896                .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2897                .maxlen         = sizeof(unsigned long),
2898                .mode           = 0444,
2899                .proc_handler   = do_proc_dqstats,
2900        },
2901        {
2902                .procname       = "free_dquots",
2903                .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2904                .maxlen         = sizeof(unsigned long),
2905                .mode           = 0444,
2906                .proc_handler   = do_proc_dqstats,
2907        },
2908        {
2909                .procname       = "syncs",
2910                .data           = &dqstats.stat[DQST_SYNCS],
2911                .maxlen         = sizeof(unsigned long),
2912                .mode           = 0444,
2913                .proc_handler   = do_proc_dqstats,
2914        },
2915#ifdef CONFIG_PRINT_QUOTA_WARNING
2916        {
2917                .procname       = "warnings",
2918                .data           = &flag_print_warnings,
2919                .maxlen         = sizeof(int),
2920                .mode           = 0644,
2921                .proc_handler   = proc_dointvec,
2922        },
2923#endif
2924        { },
2925};
2926
2927static struct ctl_table fs_table[] = {
2928        {
2929                .procname       = "quota",
2930                .mode           = 0555,
2931                .child          = fs_dqstats_table,
2932        },
2933        { },
2934};
2935
2936static struct ctl_table sys_table[] = {
2937        {
2938                .procname       = "fs",
2939                .mode           = 0555,
2940                .child          = fs_table,
2941        },
2942        { },
2943};
2944
2945static int __init dquot_init(void)
2946{
2947        int i, ret;
2948        unsigned long nr_hash, order;
2949
2950        printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2951
2952        register_sysctl_table(sys_table);
2953
2954        dquot_cachep = kmem_cache_create("dquot",
2955                        sizeof(struct dquot), sizeof(unsigned long) * 4,
2956                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2957                                SLAB_MEM_SPREAD|SLAB_PANIC),
2958                        NULL);
2959
2960        order = 0;
2961        dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2962        if (!dquot_hash)
2963                panic("Cannot create dquot hash table");
2964
2965        for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2966                ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2967                if (ret)
2968                        panic("Cannot create dquot stat counters");
2969        }
2970
2971        /* Find power-of-two hlist_heads which can fit into allocation */
2972        nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2973        dq_hash_bits = ilog2(nr_hash);
2974
2975        nr_hash = 1UL << dq_hash_bits;
2976        dq_hash_mask = nr_hash - 1;
2977        for (i = 0; i < nr_hash; i++)
2978                INIT_HLIST_HEAD(dquot_hash + i);
2979
2980        pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2981                " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2982
2983        if (register_shrinker(&dqcache_shrinker))
2984                panic("Cannot register dquot shrinker");
2985
2986        return 0;
2987}
2988fs_initcall(dquot_init);
2989