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 "../internal.h" /* ugh */
  82
  83#include <linux/uaccess.h>
  84
  85/*
  86 * There are five quota SMP locks:
  87 * * dq_list_lock protects all lists with quotas and quota formats.
  88 * * dquot->dq_dqb_lock protects data from dq_dqb
  89 * * inode->i_lock protects inode->i_blocks, i_bytes and also guards
  90 *   consistency of dquot->dq_dqb with inode->i_blocks, i_bytes so that
  91 *   dquot_transfer() can stabilize amount it transfers
  92 * * dq_data_lock protects mem_dqinfo structures and modifications of dquot
  93 *   pointers in the inode
  94 * * dq_state_lock protects modifications of quota state (on quotaon and
  95 *   quotaoff) and readers who care about latest values take it as well.
  96 *
  97 * The spinlock ordering is hence:
  98 *   dq_data_lock > dq_list_lock > i_lock > dquot->dq_dqb_lock,
  99 *   dq_list_lock > dq_state_lock
 100 *
 101 * Note that some things (eg. sb pointer, type, id) doesn't change during
 102 * the life of the dquot structure and so needn't to be protected by a lock
 103 *
 104 * Operation accessing dquots via inode pointers are protected by dquot_srcu.
 105 * Operation of reading pointer needs srcu_read_lock(&dquot_srcu), and
 106 * synchronize_srcu(&dquot_srcu) is called after clearing pointers from
 107 * inode and before dropping dquot references to avoid use of dquots after
 108 * they are freed. dq_data_lock is used to serialize the pointer setting and
 109 * clearing operations.
 110 * Special care needs to be taken about S_NOQUOTA inode flag (marking that
 111 * inode is a quota file). Functions adding pointers from inode to dquots have
 112 * to check this flag under dq_data_lock and then (if S_NOQUOTA is not set) they
 113 * have to do all pointer modifications before dropping dq_data_lock. This makes
 114 * sure they cannot race with quotaon which first sets S_NOQUOTA flag and
 115 * then drops all pointers to dquots from an inode.
 116 *
 117 * Each dquot has its dq_lock mutex.  Dquot is locked when it is being read to
 118 * memory (or space for it is being allocated) on the first dqget(), when it is
 119 * being written out, and when it is being released on the last dqput(). The
 120 * allocation and release operations are serialized by the dq_lock and by
 121 * checking the use count in dquot_release().
 122 *
 123 * Lock ordering (including related VFS locks) is the following:
 124 *   s_umount > i_mutex > journal_lock > dquot->dq_lock > dqio_sem
 125 */
 126
 127static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_list_lock);
 128static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_state_lock);
 129__cacheline_aligned_in_smp DEFINE_SPINLOCK(dq_data_lock);
 130EXPORT_SYMBOL(dq_data_lock);
 131DEFINE_STATIC_SRCU(dquot_srcu);
 132
 133static DECLARE_WAIT_QUEUE_HEAD(dquot_ref_wq);
 134
 135void __quota_error(struct super_block *sb, const char *func,
 136                   const char *fmt, ...)
 137{
 138        if (printk_ratelimit()) {
 139                va_list args;
 140                struct va_format vaf;
 141
 142                va_start(args, fmt);
 143
 144                vaf.fmt = fmt;
 145                vaf.va = &args;
 146
 147                printk(KERN_ERR "Quota error (device %s): %s: %pV\n",
 148                       sb->s_id, func, &vaf);
 149
 150                va_end(args);
 151        }
 152}
 153EXPORT_SYMBOL(__quota_error);
 154
 155#if defined(CONFIG_QUOTA_DEBUG) || defined(CONFIG_PRINT_QUOTA_WARNING)
 156static char *quotatypes[] = INITQFNAMES;
 157#endif
 158static struct quota_format_type *quota_formats; /* List of registered formats */
 159static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
 160
 161/* SLAB cache for dquot structures */
 162static struct kmem_cache *dquot_cachep;
 163
 164int register_quota_format(struct quota_format_type *fmt)
 165{
 166        spin_lock(&dq_list_lock);
 167        fmt->qf_next = quota_formats;
 168        quota_formats = fmt;
 169        spin_unlock(&dq_list_lock);
 170        return 0;
 171}
 172EXPORT_SYMBOL(register_quota_format);
 173
 174void unregister_quota_format(struct quota_format_type *fmt)
 175{
 176        struct quota_format_type **actqf;
 177
 178        spin_lock(&dq_list_lock);
 179        for (actqf = &quota_formats; *actqf && *actqf != fmt;
 180             actqf = &(*actqf)->qf_next)
 181                ;
 182        if (*actqf)
 183                *actqf = (*actqf)->qf_next;
 184        spin_unlock(&dq_list_lock);
 185}
 186EXPORT_SYMBOL(unregister_quota_format);
 187
 188static struct quota_format_type *find_quota_format(int id)
 189{
 190        struct quota_format_type *actqf;
 191
 192        spin_lock(&dq_list_lock);
 193        for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
 194             actqf = actqf->qf_next)
 195                ;
 196        if (!actqf || !try_module_get(actqf->qf_owner)) {
 197                int qm;
 198
 199                spin_unlock(&dq_list_lock);
 200                
 201                for (qm = 0; module_names[qm].qm_fmt_id &&
 202                             module_names[qm].qm_fmt_id != id; qm++)
 203                        ;
 204                if (!module_names[qm].qm_fmt_id ||
 205                    request_module(module_names[qm].qm_mod_name))
 206                        return NULL;
 207
 208                spin_lock(&dq_list_lock);
 209                for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id;
 210                     actqf = actqf->qf_next)
 211                        ;
 212                if (actqf && !try_module_get(actqf->qf_owner))
 213                        actqf = NULL;
 214        }
 215        spin_unlock(&dq_list_lock);
 216        return actqf;
 217}
 218
 219static void put_quota_format(struct quota_format_type *fmt)
 220{
 221        module_put(fmt->qf_owner);
 222}
 223
 224/*
 225 * Dquot List Management:
 226 * The quota code uses three lists for dquot management: the inuse_list,
 227 * free_dquots, and dquot_hash[] array. A single dquot structure may be
 228 * on all three lists, depending on its current state.
 229 *
 230 * All dquots are placed to the end of inuse_list when first created, and this
 231 * list is used for invalidate operation, which must look at every dquot.
 232 *
 233 * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
 234 * and this list is searched whenever we need an available dquot.  Dquots are
 235 * removed from the list as soon as they are used again, and
 236 * dqstats.free_dquots gives the number of dquots on the list. When
 237 * dquot is invalidated it's completely released from memory.
 238 *
 239 * Dquots with a specific identity (device, type and id) are placed on
 240 * one of the dquot_hash[] hash chains. The provides an efficient search
 241 * mechanism to locate a specific dquot.
 242 */
 243
 244static LIST_HEAD(inuse_list);
 245static LIST_HEAD(free_dquots);
 246static unsigned int dq_hash_bits, dq_hash_mask;
 247static struct hlist_head *dquot_hash;
 248
 249struct dqstats dqstats;
 250EXPORT_SYMBOL(dqstats);
 251
 252static qsize_t inode_get_rsv_space(struct inode *inode);
 253static qsize_t __inode_get_rsv_space(struct inode *inode);
 254static int __dquot_initialize(struct inode *inode, int type);
 255
 256static inline unsigned int
 257hashfn(const struct super_block *sb, struct kqid qid)
 258{
 259        unsigned int id = from_kqid(&init_user_ns, qid);
 260        int type = qid.type;
 261        unsigned long tmp;
 262
 263        tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 264        return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 265}
 266
 267/*
 268 * Following list functions expect dq_list_lock to be held
 269 */
 270static inline void insert_dquot_hash(struct dquot *dquot)
 271{
 272        struct hlist_head *head;
 273        head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
 274        hlist_add_head(&dquot->dq_hash, head);
 275}
 276
 277static inline void remove_dquot_hash(struct dquot *dquot)
 278{
 279        hlist_del_init(&dquot->dq_hash);
 280}
 281
 282static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
 283                                struct kqid qid)
 284{
 285        struct hlist_node *node;
 286        struct dquot *dquot;
 287
 288        hlist_for_each (node, dquot_hash+hashent) {
 289                dquot = hlist_entry(node, struct dquot, dq_hash);
 290                if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
 291                        return dquot;
 292        }
 293        return NULL;
 294}
 295
 296/* Add a dquot to the tail of the free list */
 297static inline void put_dquot_last(struct dquot *dquot)
 298{
 299        list_add_tail(&dquot->dq_free, &free_dquots);
 300        dqstats_inc(DQST_FREE_DQUOTS);
 301}
 302
 303static inline void remove_free_dquot(struct dquot *dquot)
 304{
 305        if (list_empty(&dquot->dq_free))
 306                return;
 307        list_del_init(&dquot->dq_free);
 308        dqstats_dec(DQST_FREE_DQUOTS);
 309}
 310
 311static inline void put_inuse(struct dquot *dquot)
 312{
 313        /* We add to the back of inuse list so we don't have to restart
 314         * when traversing this list and we block */
 315        list_add_tail(&dquot->dq_inuse, &inuse_list);
 316        dqstats_inc(DQST_ALLOC_DQUOTS);
 317}
 318
 319static inline void remove_inuse(struct dquot *dquot)
 320{
 321        dqstats_dec(DQST_ALLOC_DQUOTS);
 322        list_del(&dquot->dq_inuse);
 323}
 324/*
 325 * End of list functions needing dq_list_lock
 326 */
 327
 328static void wait_on_dquot(struct dquot *dquot)
 329{
 330        mutex_lock(&dquot->dq_lock);
 331        mutex_unlock(&dquot->dq_lock);
 332}
 333
 334static inline int dquot_dirty(struct dquot *dquot)
 335{
 336        return test_bit(DQ_MOD_B, &dquot->dq_flags);
 337}
 338
 339static inline int mark_dquot_dirty(struct dquot *dquot)
 340{
 341        return dquot->dq_sb->dq_op->mark_dirty(dquot);
 342}
 343
 344/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
 345int dquot_mark_dquot_dirty(struct dquot *dquot)
 346{
 347        int ret = 1;
 348
 349        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 350                return 0;
 351
 352        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
 353                return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
 354
 355        /* If quota is dirty already, we don't have to acquire dq_list_lock */
 356        if (test_bit(DQ_MOD_B, &dquot->dq_flags))
 357                return 1;
 358
 359        spin_lock(&dq_list_lock);
 360        if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
 361                list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 362                                info[dquot->dq_id.type].dqi_dirty_list);
 363                ret = 0;
 364        }
 365        spin_unlock(&dq_list_lock);
 366        return ret;
 367}
 368EXPORT_SYMBOL(dquot_mark_dquot_dirty);
 369
 370/* Dirtify all the dquots - this can block when journalling */
 371static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
 372{
 373        int ret, err, cnt;
 374
 375        ret = err = 0;
 376        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 377                if (dquot[cnt])
 378                        /* Even in case of error we have to continue */
 379                        ret = mark_dquot_dirty(dquot[cnt]);
 380                if (!err)
 381                        err = ret;
 382        }
 383        return err;
 384}
 385
 386static inline void dqput_all(struct dquot **dquot)
 387{
 388        unsigned int cnt;
 389
 390        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 391                dqput(dquot[cnt]);
 392}
 393
 394static inline int clear_dquot_dirty(struct dquot *dquot)
 395{
 396        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
 397                return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
 398
 399        spin_lock(&dq_list_lock);
 400        if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
 401                spin_unlock(&dq_list_lock);
 402                return 0;
 403        }
 404        list_del_init(&dquot->dq_dirty);
 405        spin_unlock(&dq_list_lock);
 406        return 1;
 407}
 408
 409void mark_info_dirty(struct super_block *sb, int type)
 410{
 411        spin_lock(&dq_data_lock);
 412        sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
 413        spin_unlock(&dq_data_lock);
 414}
 415EXPORT_SYMBOL(mark_info_dirty);
 416
 417/*
 418 *      Read dquot from disk and alloc space for it
 419 */
 420
 421int dquot_acquire(struct dquot *dquot)
 422{
 423        int ret = 0, ret2 = 0;
 424        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 425
 426        mutex_lock(&dquot->dq_lock);
 427        if (!test_bit(DQ_READ_B, &dquot->dq_flags))
 428                ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
 429        if (ret < 0)
 430                goto out_iolock;
 431        /* Make sure flags update is visible after dquot has been filled */
 432        smp_mb__before_atomic();
 433        set_bit(DQ_READ_B, &dquot->dq_flags);
 434        /* Instantiate dquot if needed */
 435        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 436                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 437                /* Write the info if needed */
 438                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 439                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 440                                        dquot->dq_sb, dquot->dq_id.type);
 441                }
 442                if (ret < 0)
 443                        goto out_iolock;
 444                if (ret2 < 0) {
 445                        ret = ret2;
 446                        goto out_iolock;
 447                }
 448        }
 449        /*
 450         * Make sure flags update is visible after on-disk struct has been
 451         * allocated. Paired with smp_rmb() in dqget().
 452         */
 453        smp_mb__before_atomic();
 454        set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 455out_iolock:
 456        mutex_unlock(&dquot->dq_lock);
 457        return ret;
 458}
 459EXPORT_SYMBOL(dquot_acquire);
 460
 461/*
 462 *      Write dquot to disk
 463 */
 464int dquot_commit(struct dquot *dquot)
 465{
 466        int ret = 0;
 467        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 468
 469        mutex_lock(&dquot->dq_lock);
 470        if (!clear_dquot_dirty(dquot))
 471                goto out_lock;
 472        /* Inactive dquot can be only if there was error during read/init
 473         * => we have better not writing it */
 474        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 475                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 476        else
 477                ret = -EIO;
 478out_lock:
 479        mutex_unlock(&dquot->dq_lock);
 480        return ret;
 481}
 482EXPORT_SYMBOL(dquot_commit);
 483
 484/*
 485 *      Release dquot
 486 */
 487int dquot_release(struct dquot *dquot)
 488{
 489        int ret = 0, ret2 = 0;
 490        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 491
 492        mutex_lock(&dquot->dq_lock);
 493        /* Check whether we are not racing with some other dqget() */
 494        if (atomic_read(&dquot->dq_count) > 1)
 495                goto out_dqlock;
 496        if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
 497                ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
 498                /* Write the info */
 499                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 500                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 501                                                dquot->dq_sb, dquot->dq_id.type);
 502                }
 503                if (ret >= 0)
 504                        ret = ret2;
 505        }
 506        clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 507out_dqlock:
 508        mutex_unlock(&dquot->dq_lock);
 509        return ret;
 510}
 511EXPORT_SYMBOL(dquot_release);
 512
 513void dquot_destroy(struct dquot *dquot)
 514{
 515        kmem_cache_free(dquot_cachep, dquot);
 516}
 517EXPORT_SYMBOL(dquot_destroy);
 518
 519static inline void do_destroy_dquot(struct dquot *dquot)
 520{
 521        dquot->dq_sb->dq_op->destroy_dquot(dquot);
 522}
 523
 524/* Invalidate all dquots on the list. Note that this function is called after
 525 * quota is disabled and pointers from inodes removed so there cannot be new
 526 * quota users. There can still be some users of quotas due to inodes being
 527 * just deleted or pruned by prune_icache() (those are not attached to any
 528 * list) or parallel quotactl call. We have to wait for such users.
 529 */
 530static void invalidate_dquots(struct super_block *sb, int type)
 531{
 532        struct dquot *dquot, *tmp;
 533
 534restart:
 535        spin_lock(&dq_list_lock);
 536        list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 537                if (dquot->dq_sb != sb)
 538                        continue;
 539                if (dquot->dq_id.type != type)
 540                        continue;
 541                /* Wait for dquot users */
 542                if (atomic_read(&dquot->dq_count)) {
 543                        dqgrab(dquot);
 544                        spin_unlock(&dq_list_lock);
 545                        /*
 546                         * Once dqput() wakes us up, we know it's time to free
 547                         * the dquot.
 548                         * IMPORTANT: we rely on the fact that there is always
 549                         * at most one process waiting for dquot to free.
 550                         * Otherwise dq_count would be > 1 and we would never
 551                         * wake up.
 552                         */
 553                        wait_event(dquot_ref_wq,
 554                                   atomic_read(&dquot->dq_count) == 1);
 555                        dqput(dquot);
 556                        /* At this moment dquot() need not exist (it could be
 557                         * reclaimed by prune_dqcache(). Hence we must
 558                         * restart. */
 559                        goto restart;
 560                }
 561                /*
 562                 * Quota now has no users and it has been written on last
 563                 * dqput()
 564                 */
 565                remove_dquot_hash(dquot);
 566                remove_free_dquot(dquot);
 567                remove_inuse(dquot);
 568                do_destroy_dquot(dquot);
 569        }
 570        spin_unlock(&dq_list_lock);
 571}
 572
 573/* Call callback for every active dquot on given filesystem */
 574int dquot_scan_active(struct super_block *sb,
 575                      int (*fn)(struct dquot *dquot, unsigned long priv),
 576                      unsigned long priv)
 577{
 578        struct dquot *dquot, *old_dquot = NULL;
 579        int ret = 0;
 580
 581        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 582
 583        spin_lock(&dq_list_lock);
 584        list_for_each_entry(dquot, &inuse_list, dq_inuse) {
 585                if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 586                        continue;
 587                if (dquot->dq_sb != sb)
 588                        continue;
 589                /* Now we have active dquot so we can just increase use count */
 590                atomic_inc(&dquot->dq_count);
 591                spin_unlock(&dq_list_lock);
 592                dqstats_inc(DQST_LOOKUPS);
 593                dqput(old_dquot);
 594                old_dquot = dquot;
 595                /*
 596                 * ->release_dquot() can be racing with us. Our reference
 597                 * protects us from new calls to it so just wait for any
 598                 * outstanding call and recheck the DQ_ACTIVE_B after that.
 599                 */
 600                wait_on_dquot(dquot);
 601                if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 602                        ret = fn(dquot, priv);
 603                        if (ret < 0)
 604                                goto out;
 605                }
 606                spin_lock(&dq_list_lock);
 607                /* We are safe to continue now because our dquot could not
 608                 * be moved out of the inuse list while we hold the reference */
 609        }
 610        spin_unlock(&dq_list_lock);
 611out:
 612        dqput(old_dquot);
 613        return ret;
 614}
 615EXPORT_SYMBOL(dquot_scan_active);
 616
 617/* Write all dquot structures to quota files */
 618int dquot_writeback_dquots(struct super_block *sb, int type)
 619{
 620        struct list_head *dirty;
 621        struct dquot *dquot;
 622        struct quota_info *dqopt = sb_dqopt(sb);
 623        int cnt;
 624        int err, ret = 0;
 625
 626        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 627
 628        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 629                if (type != -1 && cnt != type)
 630                        continue;
 631                if (!sb_has_quota_active(sb, cnt))
 632                        continue;
 633                spin_lock(&dq_list_lock);
 634                dirty = &dqopt->info[cnt].dqi_dirty_list;
 635                while (!list_empty(dirty)) {
 636                        dquot = list_first_entry(dirty, struct dquot,
 637                                                 dq_dirty);
 638
 639                        WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
 640
 641                        /* Now we have active dquot from which someone is
 642                         * holding reference so we can safely just increase
 643                         * use count */
 644                        dqgrab(dquot);
 645                        spin_unlock(&dq_list_lock);
 646                        dqstats_inc(DQST_LOOKUPS);
 647                        err = sb->dq_op->write_dquot(dquot);
 648                        if (err) {
 649                                /*
 650                                 * Clear dirty bit anyway to avoid infinite
 651                                 * loop here.
 652                                 */
 653                                clear_dquot_dirty(dquot);
 654                                if (!ret)
 655                                        ret = err;
 656                        }
 657                        dqput(dquot);
 658                        spin_lock(&dq_list_lock);
 659                }
 660                spin_unlock(&dq_list_lock);
 661        }
 662
 663        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 664                if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
 665                    && info_dirty(&dqopt->info[cnt]))
 666                        sb->dq_op->write_info(sb, cnt);
 667        dqstats_inc(DQST_SYNCS);
 668
 669        return ret;
 670}
 671EXPORT_SYMBOL(dquot_writeback_dquots);
 672
 673/* Write all dquot structures to disk and make them visible from userspace */
 674int dquot_quota_sync(struct super_block *sb, int type)
 675{
 676        struct quota_info *dqopt = sb_dqopt(sb);
 677        int cnt;
 678        int ret;
 679
 680        ret = dquot_writeback_dquots(sb, type);
 681        if (ret)
 682                return ret;
 683        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
 684                return 0;
 685
 686        /* This is not very clever (and fast) but currently I don't know about
 687         * any other simple way of getting quota data to disk and we must get
 688         * them there for userspace to be visible... */
 689        if (sb->s_op->sync_fs)
 690                sb->s_op->sync_fs(sb, 1);
 691        sync_blockdev(sb->s_bdev);
 692
 693        /*
 694         * Now when everything is written we can discard the pagecache so
 695         * that userspace sees the changes.
 696         */
 697        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 698                if (type != -1 && cnt != type)
 699                        continue;
 700                if (!sb_has_quota_active(sb, cnt))
 701                        continue;
 702                inode_lock(dqopt->files[cnt]);
 703                truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
 704                inode_unlock(dqopt->files[cnt]);
 705        }
 706
 707        return 0;
 708}
 709EXPORT_SYMBOL(dquot_quota_sync);
 710
 711static unsigned long
 712dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 713{
 714        struct list_head *head;
 715        struct dquot *dquot;
 716        unsigned long freed = 0;
 717
 718        spin_lock(&dq_list_lock);
 719        head = free_dquots.prev;
 720        while (head != &free_dquots && sc->nr_to_scan) {
 721                dquot = list_entry(head, struct dquot, dq_free);
 722                remove_dquot_hash(dquot);
 723                remove_free_dquot(dquot);
 724                remove_inuse(dquot);
 725                do_destroy_dquot(dquot);
 726                sc->nr_to_scan--;
 727                freed++;
 728                head = free_dquots.prev;
 729        }
 730        spin_unlock(&dq_list_lock);
 731        return freed;
 732}
 733
 734static unsigned long
 735dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
 736{
 737        return vfs_pressure_ratio(
 738        percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
 739}
 740
 741static struct shrinker dqcache_shrinker = {
 742        .count_objects = dqcache_shrink_count,
 743        .scan_objects = dqcache_shrink_scan,
 744        .seeks = DEFAULT_SEEKS,
 745};
 746
 747/*
 748 * Put reference to dquot
 749 */
 750void dqput(struct dquot *dquot)
 751{
 752        int ret;
 753
 754        if (!dquot)
 755                return;
 756#ifdef CONFIG_QUOTA_DEBUG
 757        if (!atomic_read(&dquot->dq_count)) {
 758                quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
 759                            quotatypes[dquot->dq_id.type],
 760                            from_kqid(&init_user_ns, dquot->dq_id));
 761                BUG();
 762        }
 763#endif
 764        dqstats_inc(DQST_DROPS);
 765we_slept:
 766        spin_lock(&dq_list_lock);
 767        if (atomic_read(&dquot->dq_count) > 1) {
 768                /* We have more than one user... nothing to do */
 769                atomic_dec(&dquot->dq_count);
 770                /* Releasing dquot during quotaoff phase? */
 771                if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
 772                    atomic_read(&dquot->dq_count) == 1)
 773                        wake_up(&dquot_ref_wq);
 774                spin_unlock(&dq_list_lock);
 775                return;
 776        }
 777        /* Need to release dquot? */
 778        if (dquot_dirty(dquot)) {
 779                spin_unlock(&dq_list_lock);
 780                /* Commit dquot before releasing */
 781                ret = dquot->dq_sb->dq_op->write_dquot(dquot);
 782                if (ret < 0) {
 783                        quota_error(dquot->dq_sb, "Can't write quota structure"
 784                                    " (error %d). Quota may get out of sync!",
 785                                    ret);
 786                        /*
 787                         * We clear dirty bit anyway, so that we avoid
 788                         * infinite loop here
 789                         */
 790                        clear_dquot_dirty(dquot);
 791                }
 792                goto we_slept;
 793        }
 794        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 795                spin_unlock(&dq_list_lock);
 796                dquot->dq_sb->dq_op->release_dquot(dquot);
 797                goto we_slept;
 798        }
 799        atomic_dec(&dquot->dq_count);
 800#ifdef CONFIG_QUOTA_DEBUG
 801        /* sanity check */
 802        BUG_ON(!list_empty(&dquot->dq_free));
 803#endif
 804        put_dquot_last(dquot);
 805        spin_unlock(&dq_list_lock);
 806}
 807EXPORT_SYMBOL(dqput);
 808
 809struct dquot *dquot_alloc(struct super_block *sb, int type)
 810{
 811        return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
 812}
 813EXPORT_SYMBOL(dquot_alloc);
 814
 815static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 816{
 817        struct dquot *dquot;
 818
 819        dquot = sb->dq_op->alloc_dquot(sb, type);
 820        if(!dquot)
 821                return NULL;
 822
 823        mutex_init(&dquot->dq_lock);
 824        INIT_LIST_HEAD(&dquot->dq_free);
 825        INIT_LIST_HEAD(&dquot->dq_inuse);
 826        INIT_HLIST_NODE(&dquot->dq_hash);
 827        INIT_LIST_HEAD(&dquot->dq_dirty);
 828        dquot->dq_sb = sb;
 829        dquot->dq_id = make_kqid_invalid(type);
 830        atomic_set(&dquot->dq_count, 1);
 831        spin_lock_init(&dquot->dq_dqb_lock);
 832
 833        return dquot;
 834}
 835
 836/*
 837 * Get reference to dquot
 838 *
 839 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
 840 * destroying our dquot by:
 841 *   a) checking for quota flags under dq_list_lock and
 842 *   b) getting a reference to dquot before we release dq_list_lock
 843 */
 844struct dquot *dqget(struct super_block *sb, struct kqid qid)
 845{
 846        unsigned int hashent = hashfn(sb, qid);
 847        struct dquot *dquot, *empty = NULL;
 848
 849        if (!qid_has_mapping(sb->s_user_ns, qid))
 850                return ERR_PTR(-EINVAL);
 851
 852        if (!sb_has_quota_active(sb, qid.type))
 853                return ERR_PTR(-ESRCH);
 854we_slept:
 855        spin_lock(&dq_list_lock);
 856        spin_lock(&dq_state_lock);
 857        if (!sb_has_quota_active(sb, qid.type)) {
 858                spin_unlock(&dq_state_lock);
 859                spin_unlock(&dq_list_lock);
 860                dquot = ERR_PTR(-ESRCH);
 861                goto out;
 862        }
 863        spin_unlock(&dq_state_lock);
 864
 865        dquot = find_dquot(hashent, sb, qid);
 866        if (!dquot) {
 867                if (!empty) {
 868                        spin_unlock(&dq_list_lock);
 869                        empty = get_empty_dquot(sb, qid.type);
 870                        if (!empty)
 871                                schedule();     /* Try to wait for a moment... */
 872                        goto we_slept;
 873                }
 874                dquot = empty;
 875                empty = NULL;
 876                dquot->dq_id = qid;
 877                /* all dquots go on the inuse_list */
 878                put_inuse(dquot);
 879                /* hash it first so it can be found */
 880                insert_dquot_hash(dquot);
 881                spin_unlock(&dq_list_lock);
 882                dqstats_inc(DQST_LOOKUPS);
 883        } else {
 884                if (!atomic_read(&dquot->dq_count))
 885                        remove_free_dquot(dquot);
 886                atomic_inc(&dquot->dq_count);
 887                spin_unlock(&dq_list_lock);
 888                dqstats_inc(DQST_CACHE_HITS);
 889                dqstats_inc(DQST_LOOKUPS);
 890        }
 891        /* Wait for dq_lock - after this we know that either dquot_release() is
 892         * already finished or it will be canceled due to dq_count > 1 test */
 893        wait_on_dquot(dquot);
 894        /* Read the dquot / allocate space in quota file */
 895        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 896                int err;
 897
 898                err = sb->dq_op->acquire_dquot(dquot);
 899                if (err < 0) {
 900                        dqput(dquot);
 901                        dquot = ERR_PTR(err);
 902                        goto out;
 903                }
 904        }
 905        /*
 906         * Make sure following reads see filled structure - paired with
 907         * smp_mb__before_atomic() in dquot_acquire().
 908         */
 909        smp_rmb();
 910#ifdef CONFIG_QUOTA_DEBUG
 911        BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 912#endif
 913out:
 914        if (empty)
 915                do_destroy_dquot(empty);
 916
 917        return dquot;
 918}
 919EXPORT_SYMBOL(dqget);
 920
 921static inline struct dquot **i_dquot(struct inode *inode)
 922{
 923        return inode->i_sb->s_op->get_dquots(inode);
 924}
 925
 926static int dqinit_needed(struct inode *inode, int type)
 927{
 928        struct dquot * const *dquots;
 929        int cnt;
 930
 931        if (IS_NOQUOTA(inode))
 932                return 0;
 933
 934        dquots = i_dquot(inode);
 935        if (type != -1)
 936                return !dquots[type];
 937        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 938                if (!dquots[cnt])
 939                        return 1;
 940        return 0;
 941}
 942
 943/* This routine is guarded by s_umount semaphore */
 944static int add_dquot_ref(struct super_block *sb, int type)
 945{
 946        struct inode *inode, *old_inode = NULL;
 947#ifdef CONFIG_QUOTA_DEBUG
 948        int reserved = 0;
 949#endif
 950        int err = 0;
 951
 952        spin_lock(&sb->s_inode_list_lock);
 953        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 954                spin_lock(&inode->i_lock);
 955                if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
 956                    !atomic_read(&inode->i_writecount) ||
 957                    !dqinit_needed(inode, type)) {
 958                        spin_unlock(&inode->i_lock);
 959                        continue;
 960                }
 961                __iget(inode);
 962                spin_unlock(&inode->i_lock);
 963                spin_unlock(&sb->s_inode_list_lock);
 964
 965#ifdef CONFIG_QUOTA_DEBUG
 966                if (unlikely(inode_get_rsv_space(inode) > 0))
 967                        reserved = 1;
 968#endif
 969                iput(old_inode);
 970                err = __dquot_initialize(inode, type);
 971                if (err) {
 972                        iput(inode);
 973                        goto out;
 974                }
 975
 976                /*
 977                 * We hold a reference to 'inode' so it couldn't have been
 978                 * removed from s_inodes list while we dropped the
 979                 * s_inode_list_lock. We cannot iput the inode now as we can be
 980                 * holding the last reference and we cannot iput it under
 981                 * s_inode_list_lock. So we keep the reference and iput it
 982                 * later.
 983                 */
 984                old_inode = inode;
 985                spin_lock(&sb->s_inode_list_lock);
 986        }
 987        spin_unlock(&sb->s_inode_list_lock);
 988        iput(old_inode);
 989out:
 990#ifdef CONFIG_QUOTA_DEBUG
 991        if (reserved) {
 992                quota_error(sb, "Writes happened before quota was turned on "
 993                        "thus quota information is probably inconsistent. "
 994                        "Please run quotacheck(8)");
 995        }
 996#endif
 997        return err;
 998}
 999
1000/*
1001 * Remove references to dquots from inode and add dquot to list for freeing
1002 * if we have the last reference to dquot
1003 */
1004static void remove_inode_dquot_ref(struct inode *inode, int type,
1005                                   struct list_head *tofree_head)
1006{
1007        struct dquot **dquots = i_dquot(inode);
1008        struct dquot *dquot = dquots[type];
1009
1010        if (!dquot)
1011                return;
1012
1013        dquots[type] = NULL;
1014        if (list_empty(&dquot->dq_free)) {
1015                /*
1016                 * The inode still has reference to dquot so it can't be in the
1017                 * free list
1018                 */
1019                spin_lock(&dq_list_lock);
1020                list_add(&dquot->dq_free, tofree_head);
1021                spin_unlock(&dq_list_lock);
1022        } else {
1023                /*
1024                 * Dquot is already in a list to put so we won't drop the last
1025                 * reference here.
1026                 */
1027                dqput(dquot);
1028        }
1029}
1030
1031/*
1032 * Free list of dquots
1033 * Dquots are removed from inodes and no new references can be got so we are
1034 * the only ones holding reference
1035 */
1036static void put_dquot_list(struct list_head *tofree_head)
1037{
1038        struct list_head *act_head;
1039        struct dquot *dquot;
1040
1041        act_head = tofree_head->next;
1042        while (act_head != tofree_head) {
1043                dquot = list_entry(act_head, struct dquot, dq_free);
1044                act_head = act_head->next;
1045                /* Remove dquot from the list so we won't have problems... */
1046                list_del_init(&dquot->dq_free);
1047                dqput(dquot);
1048        }
1049}
1050
1051static void remove_dquot_ref(struct super_block *sb, int type,
1052                struct list_head *tofree_head)
1053{
1054        struct inode *inode;
1055        int reserved = 0;
1056
1057        spin_lock(&sb->s_inode_list_lock);
1058        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1059                /*
1060                 *  We have to scan also I_NEW inodes because they can already
1061                 *  have quota pointer initialized. Luckily, we need to touch
1062                 *  only quota pointers and these have separate locking
1063                 *  (dq_data_lock).
1064                 */
1065                spin_lock(&dq_data_lock);
1066                if (!IS_NOQUOTA(inode)) {
1067                        if (unlikely(inode_get_rsv_space(inode) > 0))
1068                                reserved = 1;
1069                        remove_inode_dquot_ref(inode, type, tofree_head);
1070                }
1071                spin_unlock(&dq_data_lock);
1072        }
1073        spin_unlock(&sb->s_inode_list_lock);
1074#ifdef CONFIG_QUOTA_DEBUG
1075        if (reserved) {
1076                printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1077                        " was disabled thus quota information is probably "
1078                        "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1079        }
1080#endif
1081}
1082
1083/* Gather all references from inodes and drop them */
1084static void drop_dquot_ref(struct super_block *sb, int type)
1085{
1086        LIST_HEAD(tofree_head);
1087
1088        if (sb->dq_op) {
1089                remove_dquot_ref(sb, type, &tofree_head);
1090                synchronize_srcu(&dquot_srcu);
1091                put_dquot_list(&tofree_head);
1092        }
1093}
1094
1095static inline
1096void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1097{
1098        if (dquot->dq_dqb.dqb_rsvspace >= number)
1099                dquot->dq_dqb.dqb_rsvspace -= number;
1100        else {
1101                WARN_ON_ONCE(1);
1102                dquot->dq_dqb.dqb_rsvspace = 0;
1103        }
1104        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1105            dquot->dq_dqb.dqb_bsoftlimit)
1106                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1107        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1108}
1109
1110static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1111{
1112        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1113            dquot->dq_dqb.dqb_curinodes >= number)
1114                dquot->dq_dqb.dqb_curinodes -= number;
1115        else
1116                dquot->dq_dqb.dqb_curinodes = 0;
1117        if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1118                dquot->dq_dqb.dqb_itime = (time64_t) 0;
1119        clear_bit(DQ_INODES_B, &dquot->dq_flags);
1120}
1121
1122static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1123{
1124        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1125            dquot->dq_dqb.dqb_curspace >= number)
1126                dquot->dq_dqb.dqb_curspace -= number;
1127        else
1128                dquot->dq_dqb.dqb_curspace = 0;
1129        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1130            dquot->dq_dqb.dqb_bsoftlimit)
1131                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1132        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1133}
1134
1135struct dquot_warn {
1136        struct super_block *w_sb;
1137        struct kqid w_dq_id;
1138        short w_type;
1139};
1140
1141static int warning_issued(struct dquot *dquot, const int warntype)
1142{
1143        int flag = (warntype == QUOTA_NL_BHARDWARN ||
1144                warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1145                ((warntype == QUOTA_NL_IHARDWARN ||
1146                warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1147
1148        if (!flag)
1149                return 0;
1150        return test_and_set_bit(flag, &dquot->dq_flags);
1151}
1152
1153#ifdef CONFIG_PRINT_QUOTA_WARNING
1154static int flag_print_warnings = 1;
1155
1156static int need_print_warning(struct dquot_warn *warn)
1157{
1158        if (!flag_print_warnings)
1159                return 0;
1160
1161        switch (warn->w_dq_id.type) {
1162                case USRQUOTA:
1163                        return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1164                case GRPQUOTA:
1165                        return in_group_p(warn->w_dq_id.gid);
1166                case PRJQUOTA:
1167                        return 1;
1168        }
1169        return 0;
1170}
1171
1172/* Print warning to user which exceeded quota */
1173static void print_warning(struct dquot_warn *warn)
1174{
1175        char *msg = NULL;
1176        struct tty_struct *tty;
1177        int warntype = warn->w_type;
1178
1179        if (warntype == QUOTA_NL_IHARDBELOW ||
1180            warntype == QUOTA_NL_ISOFTBELOW ||
1181            warntype == QUOTA_NL_BHARDBELOW ||
1182            warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1183                return;
1184
1185        tty = get_current_tty();
1186        if (!tty)
1187                return;
1188        tty_write_message(tty, warn->w_sb->s_id);
1189        if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1190                tty_write_message(tty, ": warning, ");
1191        else
1192                tty_write_message(tty, ": write failed, ");
1193        tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1194        switch (warntype) {
1195                case QUOTA_NL_IHARDWARN:
1196                        msg = " file limit reached.\r\n";
1197                        break;
1198                case QUOTA_NL_ISOFTLONGWARN:
1199                        msg = " file quota exceeded too long.\r\n";
1200                        break;
1201                case QUOTA_NL_ISOFTWARN:
1202                        msg = " file quota exceeded.\r\n";
1203                        break;
1204                case QUOTA_NL_BHARDWARN:
1205                        msg = " block limit reached.\r\n";
1206                        break;
1207                case QUOTA_NL_BSOFTLONGWARN:
1208                        msg = " block quota exceeded too long.\r\n";
1209                        break;
1210                case QUOTA_NL_BSOFTWARN:
1211                        msg = " block quota exceeded.\r\n";
1212                        break;
1213        }
1214        tty_write_message(tty, msg);
1215        tty_kref_put(tty);
1216}
1217#endif
1218
1219static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1220                            int warntype)
1221{
1222        if (warning_issued(dquot, warntype))
1223                return;
1224        warn->w_type = warntype;
1225        warn->w_sb = dquot->dq_sb;
1226        warn->w_dq_id = dquot->dq_id;
1227}
1228
1229/*
1230 * Write warnings to the console and send warning messages over netlink.
1231 *
1232 * Note that this function can call into tty and networking code.
1233 */
1234static void flush_warnings(struct dquot_warn *warn)
1235{
1236        int i;
1237
1238        for (i = 0; i < MAXQUOTAS; i++) {
1239                if (warn[i].w_type == QUOTA_NL_NOWARN)
1240                        continue;
1241#ifdef CONFIG_PRINT_QUOTA_WARNING
1242                print_warning(&warn[i]);
1243#endif
1244                quota_send_warning(warn[i].w_dq_id,
1245                                   warn[i].w_sb->s_dev, warn[i].w_type);
1246        }
1247}
1248
1249static int ignore_hardlimit(struct dquot *dquot)
1250{
1251        struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1252
1253        return capable(CAP_SYS_RESOURCE) &&
1254               (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1255                !(info->dqi_flags & DQF_ROOT_SQUASH));
1256}
1257
1258static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1259                            struct dquot_warn *warn)
1260{
1261        qsize_t newinodes;
1262        int ret = 0;
1263
1264        spin_lock(&dquot->dq_dqb_lock);
1265        newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1266        if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1267            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1268                goto add;
1269
1270        if (dquot->dq_dqb.dqb_ihardlimit &&
1271            newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1272            !ignore_hardlimit(dquot)) {
1273                prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1274                ret = -EDQUOT;
1275                goto out;
1276        }
1277
1278        if (dquot->dq_dqb.dqb_isoftlimit &&
1279            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1280            dquot->dq_dqb.dqb_itime &&
1281            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1282            !ignore_hardlimit(dquot)) {
1283                prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1284                ret = -EDQUOT;
1285                goto out;
1286        }
1287
1288        if (dquot->dq_dqb.dqb_isoftlimit &&
1289            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1290            dquot->dq_dqb.dqb_itime == 0) {
1291                prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1292                dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1293                    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1294        }
1295add:
1296        dquot->dq_dqb.dqb_curinodes = newinodes;
1297
1298out:
1299        spin_unlock(&dquot->dq_dqb_lock);
1300        return ret;
1301}
1302
1303static int dquot_add_space(struct dquot *dquot, qsize_t space,
1304                           qsize_t rsv_space, unsigned int flags,
1305                           struct dquot_warn *warn)
1306{
1307        qsize_t tspace;
1308        struct super_block *sb = dquot->dq_sb;
1309        int ret = 0;
1310
1311        spin_lock(&dquot->dq_dqb_lock);
1312        if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1313            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1314                goto finish;
1315
1316        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1317                + space + rsv_space;
1318
1319        if (dquot->dq_dqb.dqb_bhardlimit &&
1320            tspace > dquot->dq_dqb.dqb_bhardlimit &&
1321            !ignore_hardlimit(dquot)) {
1322                if (flags & DQUOT_SPACE_WARN)
1323                        prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1324                ret = -EDQUOT;
1325                goto finish;
1326        }
1327
1328        if (dquot->dq_dqb.dqb_bsoftlimit &&
1329            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1330            dquot->dq_dqb.dqb_btime &&
1331            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1332            !ignore_hardlimit(dquot)) {
1333                if (flags & DQUOT_SPACE_WARN)
1334                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1335                ret = -EDQUOT;
1336                goto finish;
1337        }
1338
1339        if (dquot->dq_dqb.dqb_bsoftlimit &&
1340            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1341            dquot->dq_dqb.dqb_btime == 0) {
1342                if (flags & DQUOT_SPACE_WARN) {
1343                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1344                        dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1345                            sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1346                } else {
1347                        /*
1348                         * We don't allow preallocation to exceed softlimit so exceeding will
1349                         * be always printed
1350                         */
1351                        ret = -EDQUOT;
1352                        goto finish;
1353                }
1354        }
1355finish:
1356        /*
1357         * We have to be careful and go through warning generation & grace time
1358         * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1359         * only here...
1360         */
1361        if (flags & DQUOT_SPACE_NOFAIL)
1362                ret = 0;
1363        if (!ret) {
1364                dquot->dq_dqb.dqb_rsvspace += rsv_space;
1365                dquot->dq_dqb.dqb_curspace += space;
1366        }
1367        spin_unlock(&dquot->dq_dqb_lock);
1368        return ret;
1369}
1370
1371static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1372{
1373        qsize_t newinodes;
1374
1375        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1376            dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1377            !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1378                return QUOTA_NL_NOWARN;
1379
1380        newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1381        if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1382                return QUOTA_NL_ISOFTBELOW;
1383        if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1384            newinodes < dquot->dq_dqb.dqb_ihardlimit)
1385                return QUOTA_NL_IHARDBELOW;
1386        return QUOTA_NL_NOWARN;
1387}
1388
1389static int info_bdq_free(struct dquot *dquot, qsize_t space)
1390{
1391        qsize_t tspace;
1392
1393        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1394
1395        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1396            tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1397                return QUOTA_NL_NOWARN;
1398
1399        if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1400                return QUOTA_NL_BSOFTBELOW;
1401        if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1402            tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1403                return QUOTA_NL_BHARDBELOW;
1404        return QUOTA_NL_NOWARN;
1405}
1406
1407static int dquot_active(const struct inode *inode)
1408{
1409        struct super_block *sb = inode->i_sb;
1410
1411        if (IS_NOQUOTA(inode))
1412                return 0;
1413        return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1414}
1415
1416/*
1417 * Initialize quota pointers in inode
1418 *
1419 * It is better to call this function outside of any transaction as it
1420 * might need a lot of space in journal for dquot structure allocation.
1421 */
1422static int __dquot_initialize(struct inode *inode, int type)
1423{
1424        int cnt, init_needed = 0;
1425        struct dquot **dquots, *got[MAXQUOTAS] = {};
1426        struct super_block *sb = inode->i_sb;
1427        qsize_t rsv;
1428        int ret = 0;
1429
1430        if (!dquot_active(inode))
1431                return 0;
1432
1433        dquots = i_dquot(inode);
1434
1435        /* First get references to structures we might need. */
1436        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1437                struct kqid qid;
1438                kprojid_t projid;
1439                int rc;
1440                struct dquot *dquot;
1441
1442                if (type != -1 && cnt != type)
1443                        continue;
1444                /*
1445                 * The i_dquot should have been initialized in most cases,
1446                 * we check it without locking here to avoid unnecessary
1447                 * dqget()/dqput() calls.
1448                 */
1449                if (dquots[cnt])
1450                        continue;
1451
1452                if (!sb_has_quota_active(sb, cnt))
1453                        continue;
1454
1455                init_needed = 1;
1456
1457                switch (cnt) {
1458                case USRQUOTA:
1459                        qid = make_kqid_uid(inode->i_uid);
1460                        break;
1461                case GRPQUOTA:
1462                        qid = make_kqid_gid(inode->i_gid);
1463                        break;
1464                case PRJQUOTA:
1465                        rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1466                        if (rc)
1467                                continue;
1468                        qid = make_kqid_projid(projid);
1469                        break;
1470                }
1471                dquot = dqget(sb, qid);
1472                if (IS_ERR(dquot)) {
1473                        /* We raced with somebody turning quotas off... */
1474                        if (PTR_ERR(dquot) != -ESRCH) {
1475                                ret = PTR_ERR(dquot);
1476                                goto out_put;
1477                        }
1478                        dquot = NULL;
1479                }
1480                got[cnt] = dquot;
1481        }
1482
1483        /* All required i_dquot has been initialized */
1484        if (!init_needed)
1485                return 0;
1486
1487        spin_lock(&dq_data_lock);
1488        if (IS_NOQUOTA(inode))
1489                goto out_lock;
1490        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1491                if (type != -1 && cnt != type)
1492                        continue;
1493                /* Avoid races with quotaoff() */
1494                if (!sb_has_quota_active(sb, cnt))
1495                        continue;
1496                /* We could race with quotaon or dqget() could have failed */
1497                if (!got[cnt])
1498                        continue;
1499                if (!dquots[cnt]) {
1500                        dquots[cnt] = got[cnt];
1501                        got[cnt] = NULL;
1502                        /*
1503                         * Make quota reservation system happy if someone
1504                         * did a write before quota was turned on
1505                         */
1506                        rsv = inode_get_rsv_space(inode);
1507                        if (unlikely(rsv)) {
1508                                spin_lock(&inode->i_lock);
1509                                /* Get reservation again under proper lock */
1510                                rsv = __inode_get_rsv_space(inode);
1511                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1512                                dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1513                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1514                                spin_unlock(&inode->i_lock);
1515                        }
1516                }
1517        }
1518out_lock:
1519        spin_unlock(&dq_data_lock);
1520out_put:
1521        /* Drop unused references */
1522        dqput_all(got);
1523
1524        return ret;
1525}
1526
1527int dquot_initialize(struct inode *inode)
1528{
1529        return __dquot_initialize(inode, -1);
1530}
1531EXPORT_SYMBOL(dquot_initialize);
1532
1533bool dquot_initialize_needed(struct inode *inode)
1534{
1535        struct dquot **dquots;
1536        int i;
1537
1538        if (!dquot_active(inode))
1539                return false;
1540
1541        dquots = i_dquot(inode);
1542        for (i = 0; i < MAXQUOTAS; i++)
1543                if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1544                        return true;
1545        return false;
1546}
1547EXPORT_SYMBOL(dquot_initialize_needed);
1548
1549/*
1550 * Release all quotas referenced by inode.
1551 *
1552 * This function only be called on inode free or converting
1553 * a file to quota file, no other users for the i_dquot in
1554 * both cases, so we needn't call synchronize_srcu() after
1555 * clearing i_dquot.
1556 */
1557static void __dquot_drop(struct inode *inode)
1558{
1559        int cnt;
1560        struct dquot **dquots = i_dquot(inode);
1561        struct dquot *put[MAXQUOTAS];
1562
1563        spin_lock(&dq_data_lock);
1564        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1565                put[cnt] = dquots[cnt];
1566                dquots[cnt] = NULL;
1567        }
1568        spin_unlock(&dq_data_lock);
1569        dqput_all(put);
1570}
1571
1572void dquot_drop(struct inode *inode)
1573{
1574        struct dquot * const *dquots;
1575        int cnt;
1576
1577        if (IS_NOQUOTA(inode))
1578                return;
1579
1580        /*
1581         * Test before calling to rule out calls from proc and such
1582         * where we are not allowed to block. Note that this is
1583         * actually reliable test even without the lock - the caller
1584         * must assure that nobody can come after the DQUOT_DROP and
1585         * add quota pointers back anyway.
1586         */
1587        dquots = i_dquot(inode);
1588        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1589                if (dquots[cnt])
1590                        break;
1591        }
1592
1593        if (cnt < MAXQUOTAS)
1594                __dquot_drop(inode);
1595}
1596EXPORT_SYMBOL(dquot_drop);
1597
1598/*
1599 * inode_reserved_space is managed internally by quota, and protected by
1600 * i_lock similar to i_blocks+i_bytes.
1601 */
1602static qsize_t *inode_reserved_space(struct inode * inode)
1603{
1604        /* Filesystem must explicitly define it's own method in order to use
1605         * quota reservation interface */
1606        BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1607        return inode->i_sb->dq_op->get_reserved_space(inode);
1608}
1609
1610static qsize_t __inode_get_rsv_space(struct inode *inode)
1611{
1612        if (!inode->i_sb->dq_op->get_reserved_space)
1613                return 0;
1614        return *inode_reserved_space(inode);
1615}
1616
1617static qsize_t inode_get_rsv_space(struct inode *inode)
1618{
1619        qsize_t ret;
1620
1621        if (!inode->i_sb->dq_op->get_reserved_space)
1622                return 0;
1623        spin_lock(&inode->i_lock);
1624        ret = __inode_get_rsv_space(inode);
1625        spin_unlock(&inode->i_lock);
1626        return ret;
1627}
1628
1629/*
1630 * This functions updates i_blocks+i_bytes fields and quota information
1631 * (together with appropriate checks).
1632 *
1633 * NOTE: We absolutely rely on the fact that caller dirties the inode
1634 * (usually helpers in quotaops.h care about this) and holds a handle for
1635 * the current transaction so that dquot write and inode write go into the
1636 * same transaction.
1637 */
1638
1639/*
1640 * This operation can block, but only after everything is updated
1641 */
1642int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1643{
1644        int cnt, ret = 0, index;
1645        struct dquot_warn warn[MAXQUOTAS];
1646        int reserve = flags & DQUOT_SPACE_RESERVE;
1647        struct dquot **dquots;
1648
1649        if (!dquot_active(inode)) {
1650                if (reserve) {
1651                        spin_lock(&inode->i_lock);
1652                        *inode_reserved_space(inode) += number;
1653                        spin_unlock(&inode->i_lock);
1654                } else {
1655                        inode_add_bytes(inode, number);
1656                }
1657                goto out;
1658        }
1659
1660        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1661                warn[cnt].w_type = QUOTA_NL_NOWARN;
1662
1663        dquots = i_dquot(inode);
1664        index = srcu_read_lock(&dquot_srcu);
1665        spin_lock(&inode->i_lock);
1666        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1667                if (!dquots[cnt])
1668                        continue;
1669                if (flags & DQUOT_SPACE_RESERVE) {
1670                        ret = dquot_add_space(dquots[cnt], 0, number, flags,
1671                                              &warn[cnt]);
1672                } else {
1673                        ret = dquot_add_space(dquots[cnt], number, 0, flags,
1674                                              &warn[cnt]);
1675                }
1676                if (ret) {
1677                        /* Back out changes we already did */
1678                        for (cnt--; cnt >= 0; cnt--) {
1679                                if (!dquots[cnt])
1680                                        continue;
1681                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1682                                if (flags & DQUOT_SPACE_RESERVE) {
1683                                        dquots[cnt]->dq_dqb.dqb_rsvspace -=
1684                                                                        number;
1685                                } else {
1686                                        dquots[cnt]->dq_dqb.dqb_curspace -=
1687                                                                        number;
1688                                }
1689                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1690                        }
1691                        spin_unlock(&inode->i_lock);
1692                        goto out_flush_warn;
1693                }
1694        }
1695        if (reserve)
1696                *inode_reserved_space(inode) += number;
1697        else
1698                __inode_add_bytes(inode, number);
1699        spin_unlock(&inode->i_lock);
1700
1701        if (reserve)
1702                goto out_flush_warn;
1703        mark_all_dquot_dirty(dquots);
1704out_flush_warn:
1705        srcu_read_unlock(&dquot_srcu, index);
1706        flush_warnings(warn);
1707out:
1708        return ret;
1709}
1710EXPORT_SYMBOL(__dquot_alloc_space);
1711
1712/*
1713 * This operation can block, but only after everything is updated
1714 */
1715int dquot_alloc_inode(struct inode *inode)
1716{
1717        int cnt, ret = 0, index;
1718        struct dquot_warn warn[MAXQUOTAS];
1719        struct dquot * const *dquots;
1720
1721        if (!dquot_active(inode))
1722                return 0;
1723        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1724                warn[cnt].w_type = QUOTA_NL_NOWARN;
1725
1726        dquots = i_dquot(inode);
1727        index = srcu_read_lock(&dquot_srcu);
1728        spin_lock(&inode->i_lock);
1729        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1730                if (!dquots[cnt])
1731                        continue;
1732                ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1733                if (ret) {
1734                        for (cnt--; cnt >= 0; cnt--) {
1735                                if (!dquots[cnt])
1736                                        continue;
1737                                /* Back out changes we already did */
1738                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1739                                dquots[cnt]->dq_dqb.dqb_curinodes--;
1740                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1741                        }
1742                        goto warn_put_all;
1743                }
1744        }
1745
1746warn_put_all:
1747        spin_unlock(&inode->i_lock);
1748        if (ret == 0)
1749                mark_all_dquot_dirty(dquots);
1750        srcu_read_unlock(&dquot_srcu, index);
1751        flush_warnings(warn);
1752        return ret;
1753}
1754EXPORT_SYMBOL(dquot_alloc_inode);
1755
1756/*
1757 * Convert in-memory reserved quotas to real consumed quotas
1758 */
1759int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1760{
1761        struct dquot **dquots;
1762        int cnt, index;
1763
1764        if (!dquot_active(inode)) {
1765                spin_lock(&inode->i_lock);
1766                *inode_reserved_space(inode) -= number;
1767                __inode_add_bytes(inode, number);
1768                spin_unlock(&inode->i_lock);
1769                return 0;
1770        }
1771
1772        dquots = i_dquot(inode);
1773        index = srcu_read_lock(&dquot_srcu);
1774        spin_lock(&inode->i_lock);
1775        /* Claim reserved quotas to allocated quotas */
1776        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1777                if (dquots[cnt]) {
1778                        struct dquot *dquot = dquots[cnt];
1779
1780                        spin_lock(&dquot->dq_dqb_lock);
1781                        if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1782                                number = dquot->dq_dqb.dqb_rsvspace;
1783                        dquot->dq_dqb.dqb_curspace += number;
1784                        dquot->dq_dqb.dqb_rsvspace -= number;
1785                        spin_unlock(&dquot->dq_dqb_lock);
1786                }
1787        }
1788        /* Update inode bytes */
1789        *inode_reserved_space(inode) -= number;
1790        __inode_add_bytes(inode, number);
1791        spin_unlock(&inode->i_lock);
1792        mark_all_dquot_dirty(dquots);
1793        srcu_read_unlock(&dquot_srcu, index);
1794        return 0;
1795}
1796EXPORT_SYMBOL(dquot_claim_space_nodirty);
1797
1798/*
1799 * Convert allocated space back to in-memory reserved quotas
1800 */
1801void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1802{
1803        struct dquot **dquots;
1804        int cnt, index;
1805
1806        if (!dquot_active(inode)) {
1807                spin_lock(&inode->i_lock);
1808                *inode_reserved_space(inode) += number;
1809                __inode_sub_bytes(inode, number);
1810                spin_unlock(&inode->i_lock);
1811                return;
1812        }
1813
1814        dquots = i_dquot(inode);
1815        index = srcu_read_lock(&dquot_srcu);
1816        spin_lock(&inode->i_lock);
1817        /* Claim reserved quotas to allocated quotas */
1818        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1819                if (dquots[cnt]) {
1820                        struct dquot *dquot = dquots[cnt];
1821
1822                        spin_lock(&dquot->dq_dqb_lock);
1823                        if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1824                                number = dquot->dq_dqb.dqb_curspace;
1825                        dquot->dq_dqb.dqb_rsvspace += number;
1826                        dquot->dq_dqb.dqb_curspace -= number;
1827                        spin_unlock(&dquot->dq_dqb_lock);
1828                }
1829        }
1830        /* Update inode bytes */
1831        *inode_reserved_space(inode) += number;
1832        __inode_sub_bytes(inode, number);
1833        spin_unlock(&inode->i_lock);
1834        mark_all_dquot_dirty(dquots);
1835        srcu_read_unlock(&dquot_srcu, index);
1836        return;
1837}
1838EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1839
1840/*
1841 * This operation can block, but only after everything is updated
1842 */
1843void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1844{
1845        unsigned int cnt;
1846        struct dquot_warn warn[MAXQUOTAS];
1847        struct dquot **dquots;
1848        int reserve = flags & DQUOT_SPACE_RESERVE, index;
1849
1850        if (!dquot_active(inode)) {
1851                if (reserve) {
1852                        spin_lock(&inode->i_lock);
1853                        *inode_reserved_space(inode) -= number;
1854                        spin_unlock(&inode->i_lock);
1855                } else {
1856                        inode_sub_bytes(inode, number);
1857                }
1858                return;
1859        }
1860
1861        dquots = i_dquot(inode);
1862        index = srcu_read_lock(&dquot_srcu);
1863        spin_lock(&inode->i_lock);
1864        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1865                int wtype;
1866
1867                warn[cnt].w_type = QUOTA_NL_NOWARN;
1868                if (!dquots[cnt])
1869                        continue;
1870                spin_lock(&dquots[cnt]->dq_dqb_lock);
1871                wtype = info_bdq_free(dquots[cnt], number);
1872                if (wtype != QUOTA_NL_NOWARN)
1873                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1874                if (reserve)
1875                        dquot_free_reserved_space(dquots[cnt], number);
1876                else
1877                        dquot_decr_space(dquots[cnt], number);
1878                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1879        }
1880        if (reserve)
1881                *inode_reserved_space(inode) -= number;
1882        else
1883                __inode_sub_bytes(inode, number);
1884        spin_unlock(&inode->i_lock);
1885
1886        if (reserve)
1887                goto out_unlock;
1888        mark_all_dquot_dirty(dquots);
1889out_unlock:
1890        srcu_read_unlock(&dquot_srcu, index);
1891        flush_warnings(warn);
1892}
1893EXPORT_SYMBOL(__dquot_free_space);
1894
1895/*
1896 * This operation can block, but only after everything is updated
1897 */
1898void dquot_free_inode(struct inode *inode)
1899{
1900        unsigned int cnt;
1901        struct dquot_warn warn[MAXQUOTAS];
1902        struct dquot * const *dquots;
1903        int index;
1904
1905        if (!dquot_active(inode))
1906                return;
1907
1908        dquots = i_dquot(inode);
1909        index = srcu_read_lock(&dquot_srcu);
1910        spin_lock(&inode->i_lock);
1911        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1912                int wtype;
1913
1914                warn[cnt].w_type = QUOTA_NL_NOWARN;
1915                if (!dquots[cnt])
1916                        continue;
1917                spin_lock(&dquots[cnt]->dq_dqb_lock);
1918                wtype = info_idq_free(dquots[cnt], 1);
1919                if (wtype != QUOTA_NL_NOWARN)
1920                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1921                dquot_decr_inodes(dquots[cnt], 1);
1922                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1923        }
1924        spin_unlock(&inode->i_lock);
1925        mark_all_dquot_dirty(dquots);
1926        srcu_read_unlock(&dquot_srcu, index);
1927        flush_warnings(warn);
1928}
1929EXPORT_SYMBOL(dquot_free_inode);
1930
1931/*
1932 * Transfer the number of inode and blocks from one diskquota to an other.
1933 * On success, dquot references in transfer_to are consumed and references
1934 * to original dquots that need to be released are placed there. On failure,
1935 * references are kept untouched.
1936 *
1937 * This operation can block, but only after everything is updated
1938 * A transaction must be started when entering this function.
1939 *
1940 * We are holding reference on transfer_from & transfer_to, no need to
1941 * protect them by srcu_read_lock().
1942 */
1943int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1944{
1945        qsize_t cur_space;
1946        qsize_t rsv_space = 0;
1947        qsize_t inode_usage = 1;
1948        struct dquot *transfer_from[MAXQUOTAS] = {};
1949        int cnt, ret = 0;
1950        char is_valid[MAXQUOTAS] = {};
1951        struct dquot_warn warn_to[MAXQUOTAS];
1952        struct dquot_warn warn_from_inodes[MAXQUOTAS];
1953        struct dquot_warn warn_from_space[MAXQUOTAS];
1954
1955        if (IS_NOQUOTA(inode))
1956                return 0;
1957
1958        if (inode->i_sb->dq_op->get_inode_usage) {
1959                ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1960                if (ret)
1961                        return ret;
1962        }
1963
1964        /* Initialize the arrays */
1965        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1966                warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1967                warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1968                warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1969        }
1970
1971        spin_lock(&dq_data_lock);
1972        spin_lock(&inode->i_lock);
1973        if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1974                spin_unlock(&inode->i_lock);
1975                spin_unlock(&dq_data_lock);
1976                return 0;
1977        }
1978        cur_space = __inode_get_bytes(inode);
1979        rsv_space = __inode_get_rsv_space(inode);
1980        /*
1981         * Build the transfer_from list, check limits, and update usage in
1982         * the target structures.
1983         */
1984        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1985                /*
1986                 * Skip changes for same uid or gid or for turned off quota-type.
1987                 */
1988                if (!transfer_to[cnt])
1989                        continue;
1990                /* Avoid races with quotaoff() */
1991                if (!sb_has_quota_active(inode->i_sb, cnt))
1992                        continue;
1993                is_valid[cnt] = 1;
1994                transfer_from[cnt] = i_dquot(inode)[cnt];
1995                ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1996                                       &warn_to[cnt]);
1997                if (ret)
1998                        goto over_quota;
1999                ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space, 0,
2000                                      &warn_to[cnt]);
2001                if (ret) {
2002                        spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2003                        dquot_decr_inodes(transfer_to[cnt], inode_usage);
2004                        spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2005                        goto over_quota;
2006                }
2007        }
2008
2009        /* Decrease usage for source structures and update quota pointers */
2010        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2011                if (!is_valid[cnt])
2012                        continue;
2013                /* Due to IO error we might not have transfer_from[] structure */
2014                if (transfer_from[cnt]) {
2015                        int wtype;
2016
2017                        spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2018                        wtype = info_idq_free(transfer_from[cnt], inode_usage);
2019                        if (wtype != QUOTA_NL_NOWARN)
2020                                prepare_warning(&warn_from_inodes[cnt],
2021                                                transfer_from[cnt], wtype);
2022                        wtype = info_bdq_free(transfer_from[cnt],
2023                                              cur_space + rsv_space);
2024                        if (wtype != QUOTA_NL_NOWARN)
2025                                prepare_warning(&warn_from_space[cnt],
2026                                                transfer_from[cnt], wtype);
2027                        dquot_decr_inodes(transfer_from[cnt], inode_usage);
2028                        dquot_decr_space(transfer_from[cnt], cur_space);
2029                        dquot_free_reserved_space(transfer_from[cnt],
2030                                                  rsv_space);
2031                        spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2032                }
2033                i_dquot(inode)[cnt] = transfer_to[cnt];
2034        }
2035        spin_unlock(&inode->i_lock);
2036        spin_unlock(&dq_data_lock);
2037
2038        mark_all_dquot_dirty(transfer_from);
2039        mark_all_dquot_dirty(transfer_to);
2040        flush_warnings(warn_to);
2041        flush_warnings(warn_from_inodes);
2042        flush_warnings(warn_from_space);
2043        /* Pass back references to put */
2044        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2045                if (is_valid[cnt])
2046                        transfer_to[cnt] = transfer_from[cnt];
2047        return 0;
2048over_quota:
2049        /* Back out changes we already did */
2050        for (cnt--; cnt >= 0; cnt--) {
2051                if (!is_valid[cnt])
2052                        continue;
2053                spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2054                dquot_decr_inodes(transfer_to[cnt], inode_usage);
2055                dquot_decr_space(transfer_to[cnt], cur_space);
2056                dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2057                spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2058        }
2059        spin_unlock(&inode->i_lock);
2060        spin_unlock(&dq_data_lock);
2061        flush_warnings(warn_to);
2062        return ret;
2063}
2064EXPORT_SYMBOL(__dquot_transfer);
2065
2066/* Wrapper for transferring ownership of an inode for uid/gid only
2067 * Called from FSXXX_setattr()
2068 */
2069int dquot_transfer(struct inode *inode, struct iattr *iattr)
2070{
2071        struct dquot *transfer_to[MAXQUOTAS] = {};
2072        struct dquot *dquot;
2073        struct super_block *sb = inode->i_sb;
2074        int ret;
2075
2076        if (!dquot_active(inode))
2077                return 0;
2078
2079        if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2080                dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2081                if (IS_ERR(dquot)) {
2082                        if (PTR_ERR(dquot) != -ESRCH) {
2083                                ret = PTR_ERR(dquot);
2084                                goto out_put;
2085                        }
2086                        dquot = NULL;
2087                }
2088                transfer_to[USRQUOTA] = dquot;
2089        }
2090        if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2091                dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2092                if (IS_ERR(dquot)) {
2093                        if (PTR_ERR(dquot) != -ESRCH) {
2094                                ret = PTR_ERR(dquot);
2095                                goto out_put;
2096                        }
2097                        dquot = NULL;
2098                }
2099                transfer_to[GRPQUOTA] = dquot;
2100        }
2101        ret = __dquot_transfer(inode, transfer_to);
2102out_put:
2103        dqput_all(transfer_to);
2104        return ret;
2105}
2106EXPORT_SYMBOL(dquot_transfer);
2107
2108/*
2109 * Write info of quota file to disk
2110 */
2111int dquot_commit_info(struct super_block *sb, int type)
2112{
2113        struct quota_info *dqopt = sb_dqopt(sb);
2114
2115        return dqopt->ops[type]->write_file_info(sb, type);
2116}
2117EXPORT_SYMBOL(dquot_commit_info);
2118
2119int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2120{
2121        struct quota_info *dqopt = sb_dqopt(sb);
2122
2123        if (!sb_has_quota_active(sb, qid->type))
2124                return -ESRCH;
2125        if (!dqopt->ops[qid->type]->get_next_id)
2126                return -ENOSYS;
2127        return dqopt->ops[qid->type]->get_next_id(sb, qid);
2128}
2129EXPORT_SYMBOL(dquot_get_next_id);
2130
2131/*
2132 * Definitions of diskquota operations.
2133 */
2134const struct dquot_operations dquot_operations = {
2135        .write_dquot    = dquot_commit,
2136        .acquire_dquot  = dquot_acquire,
2137        .release_dquot  = dquot_release,
2138        .mark_dirty     = dquot_mark_dquot_dirty,
2139        .write_info     = dquot_commit_info,
2140        .alloc_dquot    = dquot_alloc,
2141        .destroy_dquot  = dquot_destroy,
2142        .get_next_id    = dquot_get_next_id,
2143};
2144EXPORT_SYMBOL(dquot_operations);
2145
2146/*
2147 * Generic helper for ->open on filesystems supporting disk quotas.
2148 */
2149int dquot_file_open(struct inode *inode, struct file *file)
2150{
2151        int error;
2152
2153        error = generic_file_open(inode, file);
2154        if (!error && (file->f_mode & FMODE_WRITE))
2155                error = dquot_initialize(inode);
2156        return error;
2157}
2158EXPORT_SYMBOL(dquot_file_open);
2159
2160/*
2161 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2162 */
2163int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2164{
2165        int cnt, ret = 0;
2166        struct quota_info *dqopt = sb_dqopt(sb);
2167        struct inode *toputinode[MAXQUOTAS];
2168
2169        /* s_umount should be held in exclusive mode */
2170        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2171                up_read(&sb->s_umount);
2172
2173        /* Cannot turn off usage accounting without turning off limits, or
2174         * suspend quotas and simultaneously turn quotas off. */
2175        if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2176            || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2177            DQUOT_USAGE_ENABLED)))
2178                return -EINVAL;
2179
2180        /*
2181         * Skip everything if there's nothing to do. We have to do this because
2182         * sometimes we are called when fill_super() failed and calling
2183         * sync_fs() in such cases does no good.
2184         */
2185        if (!sb_any_quota_loaded(sb))
2186                return 0;
2187
2188        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2189                toputinode[cnt] = NULL;
2190                if (type != -1 && cnt != type)
2191                        continue;
2192                if (!sb_has_quota_loaded(sb, cnt))
2193                        continue;
2194
2195                if (flags & DQUOT_SUSPENDED) {
2196                        spin_lock(&dq_state_lock);
2197                        dqopt->flags |=
2198                                dquot_state_flag(DQUOT_SUSPENDED, cnt);
2199                        spin_unlock(&dq_state_lock);
2200                } else {
2201                        spin_lock(&dq_state_lock);
2202                        dqopt->flags &= ~dquot_state_flag(flags, cnt);
2203                        /* Turning off suspended quotas? */
2204                        if (!sb_has_quota_loaded(sb, cnt) &&
2205                            sb_has_quota_suspended(sb, cnt)) {
2206                                dqopt->flags &= ~dquot_state_flag(
2207                                                        DQUOT_SUSPENDED, cnt);
2208                                spin_unlock(&dq_state_lock);
2209                                iput(dqopt->files[cnt]);
2210                                dqopt->files[cnt] = NULL;
2211                                continue;
2212                        }
2213                        spin_unlock(&dq_state_lock);
2214                }
2215
2216                /* We still have to keep quota loaded? */
2217                if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2218                        continue;
2219
2220                /* Note: these are blocking operations */
2221                drop_dquot_ref(sb, cnt);
2222                invalidate_dquots(sb, cnt);
2223                /*
2224                 * Now all dquots should be invalidated, all writes done so we
2225                 * should be only users of the info. No locks needed.
2226                 */
2227                if (info_dirty(&dqopt->info[cnt]))
2228                        sb->dq_op->write_info(sb, cnt);
2229                if (dqopt->ops[cnt]->free_file_info)
2230                        dqopt->ops[cnt]->free_file_info(sb, cnt);
2231                put_quota_format(dqopt->info[cnt].dqi_format);
2232
2233                toputinode[cnt] = dqopt->files[cnt];
2234                if (!sb_has_quota_loaded(sb, cnt))
2235                        dqopt->files[cnt] = NULL;
2236                dqopt->info[cnt].dqi_flags = 0;
2237                dqopt->info[cnt].dqi_igrace = 0;
2238                dqopt->info[cnt].dqi_bgrace = 0;
2239                dqopt->ops[cnt] = NULL;
2240        }
2241
2242        /* Skip syncing and setting flags if quota files are hidden */
2243        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2244                goto put_inodes;
2245
2246        /* Sync the superblock so that buffers with quota data are written to
2247         * disk (and so userspace sees correct data afterwards). */
2248        if (sb->s_op->sync_fs)
2249                sb->s_op->sync_fs(sb, 1);
2250        sync_blockdev(sb->s_bdev);
2251        /* Now the quota files are just ordinary files and we can set the
2252         * inode flags back. Moreover we discard the pagecache so that
2253         * userspace sees the writes we did bypassing the pagecache. We
2254         * must also discard the blockdev buffers so that we see the
2255         * changes done by userspace on the next quotaon() */
2256        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2257                /* This can happen when suspending quotas on remount-ro... */
2258                if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2259                        inode_lock(toputinode[cnt]);
2260                        toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2261                        truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2262                        inode_unlock(toputinode[cnt]);
2263                        mark_inode_dirty_sync(toputinode[cnt]);
2264                }
2265        if (sb->s_bdev)
2266                invalidate_bdev(sb->s_bdev);
2267put_inodes:
2268        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2269                if (toputinode[cnt]) {
2270                        /* On remount RO, we keep the inode pointer so that we
2271                         * can reenable quota on the subsequent remount RW. We
2272                         * have to check 'flags' variable and not use sb_has_
2273                         * function because another quotaon / quotaoff could
2274                         * change global state before we got here. We refuse
2275                         * to suspend quotas when there is pending delete on
2276                         * the quota file... */
2277                        if (!(flags & DQUOT_SUSPENDED))
2278                                iput(toputinode[cnt]);
2279                        else if (!toputinode[cnt]->i_nlink)
2280                                ret = -EBUSY;
2281                }
2282        return ret;
2283}
2284EXPORT_SYMBOL(dquot_disable);
2285
2286int dquot_quota_off(struct super_block *sb, int type)
2287{
2288        return dquot_disable(sb, type,
2289                             DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2290}
2291EXPORT_SYMBOL(dquot_quota_off);
2292
2293/*
2294 *      Turn quotas on on a device
2295 */
2296
2297/*
2298 * Helper function to turn quotas on when we already have the inode of
2299 * quota file and no quota information is loaded.
2300 */
2301static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2302        unsigned int flags)
2303{
2304        struct quota_format_type *fmt = find_quota_format(format_id);
2305        struct super_block *sb = inode->i_sb;
2306        struct quota_info *dqopt = sb_dqopt(sb);
2307        int error;
2308
2309        if (!fmt)
2310                return -ESRCH;
2311        if (!S_ISREG(inode->i_mode)) {
2312                error = -EACCES;
2313                goto out_fmt;
2314        }
2315        if (IS_RDONLY(inode)) {
2316                error = -EROFS;
2317                goto out_fmt;
2318        }
2319        if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2320            (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2321                error = -EINVAL;
2322                goto out_fmt;
2323        }
2324        /* Filesystems outside of init_user_ns not yet supported */
2325        if (sb->s_user_ns != &init_user_ns) {
2326                error = -EINVAL;
2327                goto out_fmt;
2328        }
2329        /* Usage always has to be set... */
2330        if (!(flags & DQUOT_USAGE_ENABLED)) {
2331                error = -EINVAL;
2332                goto out_fmt;
2333        }
2334        if (sb_has_quota_loaded(sb, type)) {
2335                error = -EBUSY;
2336                goto out_fmt;
2337        }
2338
2339        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2340                /* As we bypass the pagecache we must now flush all the
2341                 * dirty data and invalidate caches so that kernel sees
2342                 * changes from userspace. It is not enough to just flush
2343                 * the quota file since if blocksize < pagesize, invalidation
2344                 * of the cache could fail because of other unrelated dirty
2345                 * data */
2346                sync_filesystem(sb);
2347                invalidate_bdev(sb->s_bdev);
2348        }
2349
2350        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2351                /* We don't want quota and atime on quota files (deadlocks
2352                 * possible) Also nobody should write to the file - we use
2353                 * special IO operations which ignore the immutable bit. */
2354                inode_lock(inode);
2355                inode->i_flags |= S_NOQUOTA;
2356                inode_unlock(inode);
2357                /*
2358                 * When S_NOQUOTA is set, remove dquot references as no more
2359                 * references can be added
2360                 */
2361                __dquot_drop(inode);
2362        }
2363
2364        error = -EIO;
2365        dqopt->files[type] = igrab(inode);
2366        if (!dqopt->files[type])
2367                goto out_file_flags;
2368        error = -EINVAL;
2369        if (!fmt->qf_ops->check_quota_file(sb, type))
2370                goto out_file_init;
2371
2372        dqopt->ops[type] = fmt->qf_ops;
2373        dqopt->info[type].dqi_format = fmt;
2374        dqopt->info[type].dqi_fmt_id = format_id;
2375        INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2376        error = dqopt->ops[type]->read_file_info(sb, type);
2377        if (error < 0)
2378                goto out_file_init;
2379        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2380                spin_lock(&dq_data_lock);
2381                dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2382                spin_unlock(&dq_data_lock);
2383        }
2384        spin_lock(&dq_state_lock);
2385        dqopt->flags |= dquot_state_flag(flags, type);
2386        spin_unlock(&dq_state_lock);
2387
2388        error = add_dquot_ref(sb, type);
2389        if (error)
2390                dquot_disable(sb, type, flags);
2391
2392        return error;
2393out_file_init:
2394        dqopt->files[type] = NULL;
2395        iput(inode);
2396out_file_flags:
2397        inode_lock(inode);
2398        inode->i_flags &= ~S_NOQUOTA;
2399        inode_unlock(inode);
2400out_fmt:
2401        put_quota_format(fmt);
2402
2403        return error; 
2404}
2405
2406/* Reenable quotas on remount RW */
2407int dquot_resume(struct super_block *sb, int type)
2408{
2409        struct quota_info *dqopt = sb_dqopt(sb);
2410        struct inode *inode;
2411        int ret = 0, cnt;
2412        unsigned int flags;
2413
2414        /* s_umount should be held in exclusive mode */
2415        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2416                up_read(&sb->s_umount);
2417
2418        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2419                if (type != -1 && cnt != type)
2420                        continue;
2421                if (!sb_has_quota_suspended(sb, cnt))
2422                        continue;
2423
2424                inode = dqopt->files[cnt];
2425                dqopt->files[cnt] = NULL;
2426                spin_lock(&dq_state_lock);
2427                flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2428                                                        DQUOT_LIMITS_ENABLED,
2429                                                        cnt);
2430                dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2431                spin_unlock(&dq_state_lock);
2432
2433                flags = dquot_generic_flag(flags, cnt);
2434                ret = vfs_load_quota_inode(inode, cnt,
2435                                dqopt->info[cnt].dqi_fmt_id, flags);
2436                iput(inode);
2437        }
2438
2439        return ret;
2440}
2441EXPORT_SYMBOL(dquot_resume);
2442
2443int dquot_quota_on(struct super_block *sb, int type, int format_id,
2444                   const struct path *path)
2445{
2446        int error = security_quota_on(path->dentry);
2447        if (error)
2448                return error;
2449        /* Quota file not on the same filesystem? */
2450        if (path->dentry->d_sb != sb)
2451                error = -EXDEV;
2452        else
2453                error = vfs_load_quota_inode(d_inode(path->dentry), type,
2454                                             format_id, DQUOT_USAGE_ENABLED |
2455                                             DQUOT_LIMITS_ENABLED);
2456        return error;
2457}
2458EXPORT_SYMBOL(dquot_quota_on);
2459
2460/*
2461 * More powerful function for turning on quotas allowing setting
2462 * of individual quota flags
2463 */
2464int dquot_enable(struct inode *inode, int type, int format_id,
2465                 unsigned int flags)
2466{
2467        struct super_block *sb = inode->i_sb;
2468
2469        /* Just unsuspend quotas? */
2470        BUG_ON(flags & DQUOT_SUSPENDED);
2471        /* s_umount should be held in exclusive mode */
2472        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2473                up_read(&sb->s_umount);
2474
2475        if (!flags)
2476                return 0;
2477        /* Just updating flags needed? */
2478        if (sb_has_quota_loaded(sb, type)) {
2479                if (flags & DQUOT_USAGE_ENABLED &&
2480                    sb_has_quota_usage_enabled(sb, type))
2481                        return -EBUSY;
2482                if (flags & DQUOT_LIMITS_ENABLED &&
2483                    sb_has_quota_limits_enabled(sb, type))
2484                        return -EBUSY;
2485                spin_lock(&dq_state_lock);
2486                sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2487                spin_unlock(&dq_state_lock);
2488                return 0;
2489        }
2490
2491        return vfs_load_quota_inode(inode, type, format_id, flags);
2492}
2493EXPORT_SYMBOL(dquot_enable);
2494
2495/*
2496 * This function is used when filesystem needs to initialize quotas
2497 * during mount time.
2498 */
2499int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2500                int format_id, int type)
2501{
2502        struct dentry *dentry;
2503        int error;
2504
2505        dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2506        if (IS_ERR(dentry))
2507                return PTR_ERR(dentry);
2508
2509        if (d_really_is_negative(dentry)) {
2510                error = -ENOENT;
2511                goto out;
2512        }
2513
2514        error = security_quota_on(dentry);
2515        if (!error)
2516                error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2517                                DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2518
2519out:
2520        dput(dentry);
2521        return error;
2522}
2523EXPORT_SYMBOL(dquot_quota_on_mount);
2524
2525static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2526{
2527        int ret;
2528        int type;
2529        struct quota_info *dqopt = sb_dqopt(sb);
2530
2531        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2532                return -ENOSYS;
2533        /* Accounting cannot be turned on while fs is mounted */
2534        flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2535        if (!flags)
2536                return -EINVAL;
2537        for (type = 0; type < MAXQUOTAS; type++) {
2538                if (!(flags & qtype_enforce_flag(type)))
2539                        continue;
2540                /* Can't enforce without accounting */
2541                if (!sb_has_quota_usage_enabled(sb, type))
2542                        return -EINVAL;
2543                ret = dquot_enable(dqopt->files[type], type,
2544                                   dqopt->info[type].dqi_fmt_id,
2545                                   DQUOT_LIMITS_ENABLED);
2546                if (ret < 0)
2547                        goto out_err;
2548        }
2549        return 0;
2550out_err:
2551        /* Backout enforcement enablement we already did */
2552        for (type--; type >= 0; type--)  {
2553                if (flags & qtype_enforce_flag(type))
2554                        dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2555        }
2556        /* Error code translation for better compatibility with XFS */
2557        if (ret == -EBUSY)
2558                ret = -EEXIST;
2559        return ret;
2560}
2561
2562static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2563{
2564        int ret;
2565        int type;
2566        struct quota_info *dqopt = sb_dqopt(sb);
2567
2568        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2569                return -ENOSYS;
2570        /*
2571         * We don't support turning off accounting via quotactl. In principle
2572         * quota infrastructure can do this but filesystems don't expect
2573         * userspace to be able to do it.
2574         */
2575        if (flags &
2576                  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2577                return -EOPNOTSUPP;
2578
2579        /* Filter out limits not enabled */
2580        for (type = 0; type < MAXQUOTAS; type++)
2581                if (!sb_has_quota_limits_enabled(sb, type))
2582                        flags &= ~qtype_enforce_flag(type);
2583        /* Nothing left? */
2584        if (!flags)
2585                return -EEXIST;
2586        for (type = 0; type < MAXQUOTAS; type++) {
2587                if (flags & qtype_enforce_flag(type)) {
2588                        ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2589                        if (ret < 0)
2590                                goto out_err;
2591                }
2592        }
2593        return 0;
2594out_err:
2595        /* Backout enforcement disabling we already did */
2596        for (type--; type >= 0; type--)  {
2597                if (flags & qtype_enforce_flag(type))
2598                        dquot_enable(dqopt->files[type], type,
2599                                     dqopt->info[type].dqi_fmt_id,
2600                                     DQUOT_LIMITS_ENABLED);
2601        }
2602        return ret;
2603}
2604
2605/* Generic routine for getting common part of quota structure */
2606static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2607{
2608        struct mem_dqblk *dm = &dquot->dq_dqb;
2609
2610        memset(di, 0, sizeof(*di));
2611        spin_lock(&dquot->dq_dqb_lock);
2612        di->d_spc_hardlimit = dm->dqb_bhardlimit;
2613        di->d_spc_softlimit = dm->dqb_bsoftlimit;
2614        di->d_ino_hardlimit = dm->dqb_ihardlimit;
2615        di->d_ino_softlimit = dm->dqb_isoftlimit;
2616        di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2617        di->d_ino_count = dm->dqb_curinodes;
2618        di->d_spc_timer = dm->dqb_btime;
2619        di->d_ino_timer = dm->dqb_itime;
2620        spin_unlock(&dquot->dq_dqb_lock);
2621}
2622
2623int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2624                    struct qc_dqblk *di)
2625{
2626        struct dquot *dquot;
2627
2628        dquot = dqget(sb, qid);
2629        if (IS_ERR(dquot))
2630                return PTR_ERR(dquot);
2631        do_get_dqblk(dquot, di);
2632        dqput(dquot);
2633
2634        return 0;
2635}
2636EXPORT_SYMBOL(dquot_get_dqblk);
2637
2638int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2639                         struct qc_dqblk *di)
2640{
2641        struct dquot *dquot;
2642        int err;
2643
2644        if (!sb->dq_op->get_next_id)
2645                return -ENOSYS;
2646        err = sb->dq_op->get_next_id(sb, qid);
2647        if (err < 0)
2648                return err;
2649        dquot = dqget(sb, *qid);
2650        if (IS_ERR(dquot))
2651                return PTR_ERR(dquot);
2652        do_get_dqblk(dquot, di);
2653        dqput(dquot);
2654
2655        return 0;
2656}
2657EXPORT_SYMBOL(dquot_get_next_dqblk);
2658
2659#define VFS_QC_MASK \
2660        (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2661         QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2662         QC_SPC_TIMER | QC_INO_TIMER)
2663
2664/* Generic routine for setting common part of quota structure */
2665static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2666{
2667        struct mem_dqblk *dm = &dquot->dq_dqb;
2668        int check_blim = 0, check_ilim = 0;
2669        struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2670
2671        if (di->d_fieldmask & ~VFS_QC_MASK)
2672                return -EINVAL;
2673
2674        if (((di->d_fieldmask & QC_SPC_SOFT) &&
2675             di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2676            ((di->d_fieldmask & QC_SPC_HARD) &&
2677             di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2678            ((di->d_fieldmask & QC_INO_SOFT) &&
2679             (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2680            ((di->d_fieldmask & QC_INO_HARD) &&
2681             (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2682                return -ERANGE;
2683
2684        spin_lock(&dquot->dq_dqb_lock);
2685        if (di->d_fieldmask & QC_SPACE) {
2686                dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2687                check_blim = 1;
2688                set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2689        }
2690
2691        if (di->d_fieldmask & QC_SPC_SOFT)
2692                dm->dqb_bsoftlimit = di->d_spc_softlimit;
2693        if (di->d_fieldmask & QC_SPC_HARD)
2694                dm->dqb_bhardlimit = di->d_spc_hardlimit;
2695        if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2696                check_blim = 1;
2697                set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2698        }
2699
2700        if (di->d_fieldmask & QC_INO_COUNT) {
2701                dm->dqb_curinodes = di->d_ino_count;
2702                check_ilim = 1;
2703                set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2704        }
2705
2706        if (di->d_fieldmask & QC_INO_SOFT)
2707                dm->dqb_isoftlimit = di->d_ino_softlimit;
2708        if (di->d_fieldmask & QC_INO_HARD)
2709                dm->dqb_ihardlimit = di->d_ino_hardlimit;
2710        if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2711                check_ilim = 1;
2712                set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2713        }
2714
2715        if (di->d_fieldmask & QC_SPC_TIMER) {
2716                dm->dqb_btime = di->d_spc_timer;
2717                check_blim = 1;
2718                set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2719        }
2720
2721        if (di->d_fieldmask & QC_INO_TIMER) {
2722                dm->dqb_itime = di->d_ino_timer;
2723                check_ilim = 1;
2724                set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2725        }
2726
2727        if (check_blim) {
2728                if (!dm->dqb_bsoftlimit ||
2729                    dm->dqb_curspace + dm->dqb_rsvspace < dm->dqb_bsoftlimit) {
2730                        dm->dqb_btime = 0;
2731                        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2732                } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2733                        /* Set grace only if user hasn't provided his own... */
2734                        dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2735        }
2736        if (check_ilim) {
2737                if (!dm->dqb_isoftlimit ||
2738                    dm->dqb_curinodes < dm->dqb_isoftlimit) {
2739                        dm->dqb_itime = 0;
2740                        clear_bit(DQ_INODES_B, &dquot->dq_flags);
2741                } else if (!(di->d_fieldmask & QC_INO_TIMER))
2742                        /* Set grace only if user hasn't provided his own... */
2743                        dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2744        }
2745        if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2746            dm->dqb_isoftlimit)
2747                clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2748        else
2749                set_bit(DQ_FAKE_B, &dquot->dq_flags);
2750        spin_unlock(&dquot->dq_dqb_lock);
2751        mark_dquot_dirty(dquot);
2752
2753        return 0;
2754}
2755
2756int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2757                  struct qc_dqblk *di)
2758{
2759        struct dquot *dquot;
2760        int rc;
2761
2762        dquot = dqget(sb, qid);
2763        if (IS_ERR(dquot)) {
2764                rc = PTR_ERR(dquot);
2765                goto out;
2766        }
2767        rc = do_set_dqblk(dquot, di);
2768        dqput(dquot);
2769out:
2770        return rc;
2771}
2772EXPORT_SYMBOL(dquot_set_dqblk);
2773
2774/* Generic routine for getting common part of quota file information */
2775int dquot_get_state(struct super_block *sb, struct qc_state *state)
2776{
2777        struct mem_dqinfo *mi;
2778        struct qc_type_state *tstate;
2779        struct quota_info *dqopt = sb_dqopt(sb);
2780        int type;
2781  
2782        memset(state, 0, sizeof(*state));
2783        for (type = 0; type < MAXQUOTAS; type++) {
2784                if (!sb_has_quota_active(sb, type))
2785                        continue;
2786                tstate = state->s_state + type;
2787                mi = sb_dqopt(sb)->info + type;
2788                tstate->flags = QCI_ACCT_ENABLED;
2789                spin_lock(&dq_data_lock);
2790                if (mi->dqi_flags & DQF_SYS_FILE)
2791                        tstate->flags |= QCI_SYSFILE;
2792                if (mi->dqi_flags & DQF_ROOT_SQUASH)
2793                        tstate->flags |= QCI_ROOT_SQUASH;
2794                if (sb_has_quota_limits_enabled(sb, type))
2795                        tstate->flags |= QCI_LIMITS_ENFORCED;
2796                tstate->spc_timelimit = mi->dqi_bgrace;
2797                tstate->ino_timelimit = mi->dqi_igrace;
2798                tstate->ino = dqopt->files[type]->i_ino;
2799                tstate->blocks = dqopt->files[type]->i_blocks;
2800                tstate->nextents = 1;   /* We don't know... */
2801                spin_unlock(&dq_data_lock);
2802        }
2803        return 0;
2804}
2805EXPORT_SYMBOL(dquot_get_state);
2806
2807/* Generic routine for setting common part of quota file information */
2808int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2809{
2810        struct mem_dqinfo *mi;
2811        int err = 0;
2812
2813        if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2814            (ii->i_fieldmask & QC_RT_SPC_TIMER))
2815                return -EINVAL;
2816        if (!sb_has_quota_active(sb, type))
2817                return -ESRCH;
2818        mi = sb_dqopt(sb)->info + type;
2819        if (ii->i_fieldmask & QC_FLAGS) {
2820                if ((ii->i_flags & QCI_ROOT_SQUASH &&
2821                     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2822                        return -EINVAL;
2823        }
2824        spin_lock(&dq_data_lock);
2825        if (ii->i_fieldmask & QC_SPC_TIMER)
2826                mi->dqi_bgrace = ii->i_spc_timelimit;
2827        if (ii->i_fieldmask & QC_INO_TIMER)
2828                mi->dqi_igrace = ii->i_ino_timelimit;
2829        if (ii->i_fieldmask & QC_FLAGS) {
2830                if (ii->i_flags & QCI_ROOT_SQUASH)
2831                        mi->dqi_flags |= DQF_ROOT_SQUASH;
2832                else
2833                        mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2834        }
2835        spin_unlock(&dq_data_lock);
2836        mark_info_dirty(sb, type);
2837        /* Force write to disk */
2838        sb->dq_op->write_info(sb, type);
2839        return err;
2840}
2841EXPORT_SYMBOL(dquot_set_dqinfo);
2842
2843const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2844        .quota_enable   = dquot_quota_enable,
2845        .quota_disable  = dquot_quota_disable,
2846        .quota_sync     = dquot_quota_sync,
2847        .get_state      = dquot_get_state,
2848        .set_info       = dquot_set_dqinfo,
2849        .get_dqblk      = dquot_get_dqblk,
2850        .get_nextdqblk  = dquot_get_next_dqblk,
2851        .set_dqblk      = dquot_set_dqblk
2852};
2853EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2854
2855static int do_proc_dqstats(struct ctl_table *table, int write,
2856                     void __user *buffer, size_t *lenp, loff_t *ppos)
2857{
2858        unsigned int type = (int *)table->data - dqstats.stat;
2859
2860        /* Update global table */
2861        dqstats.stat[type] =
2862                        percpu_counter_sum_positive(&dqstats.counter[type]);
2863        return proc_dointvec(table, write, buffer, lenp, ppos);
2864}
2865
2866static struct ctl_table fs_dqstats_table[] = {
2867        {
2868                .procname       = "lookups",
2869                .data           = &dqstats.stat[DQST_LOOKUPS],
2870                .maxlen         = sizeof(int),
2871                .mode           = 0444,
2872                .proc_handler   = do_proc_dqstats,
2873        },
2874        {
2875                .procname       = "drops",
2876                .data           = &dqstats.stat[DQST_DROPS],
2877                .maxlen         = sizeof(int),
2878                .mode           = 0444,
2879                .proc_handler   = do_proc_dqstats,
2880        },
2881        {
2882                .procname       = "reads",
2883                .data           = &dqstats.stat[DQST_READS],
2884                .maxlen         = sizeof(int),
2885                .mode           = 0444,
2886                .proc_handler   = do_proc_dqstats,
2887        },
2888        {
2889                .procname       = "writes",
2890                .data           = &dqstats.stat[DQST_WRITES],
2891                .maxlen         = sizeof(int),
2892                .mode           = 0444,
2893                .proc_handler   = do_proc_dqstats,
2894        },
2895        {
2896                .procname       = "cache_hits",
2897                .data           = &dqstats.stat[DQST_CACHE_HITS],
2898                .maxlen         = sizeof(int),
2899                .mode           = 0444,
2900                .proc_handler   = do_proc_dqstats,
2901        },
2902        {
2903                .procname       = "allocated_dquots",
2904                .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2905                .maxlen         = sizeof(int),
2906                .mode           = 0444,
2907                .proc_handler   = do_proc_dqstats,
2908        },
2909        {
2910                .procname       = "free_dquots",
2911                .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2912                .maxlen         = sizeof(int),
2913                .mode           = 0444,
2914                .proc_handler   = do_proc_dqstats,
2915        },
2916        {
2917                .procname       = "syncs",
2918                .data           = &dqstats.stat[DQST_SYNCS],
2919                .maxlen         = sizeof(int),
2920                .mode           = 0444,
2921                .proc_handler   = do_proc_dqstats,
2922        },
2923#ifdef CONFIG_PRINT_QUOTA_WARNING
2924        {
2925                .procname       = "warnings",
2926                .data           = &flag_print_warnings,
2927                .maxlen         = sizeof(int),
2928                .mode           = 0644,
2929                .proc_handler   = proc_dointvec,
2930        },
2931#endif
2932        { },
2933};
2934
2935static struct ctl_table fs_table[] = {
2936        {
2937                .procname       = "quota",
2938                .mode           = 0555,
2939                .child          = fs_dqstats_table,
2940        },
2941        { },
2942};
2943
2944static struct ctl_table sys_table[] = {
2945        {
2946                .procname       = "fs",
2947                .mode           = 0555,
2948                .child          = fs_table,
2949        },
2950        { },
2951};
2952
2953static int __init dquot_init(void)
2954{
2955        int i, ret;
2956        unsigned long nr_hash, order;
2957
2958        printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2959
2960        register_sysctl_table(sys_table);
2961
2962        dquot_cachep = kmem_cache_create("dquot",
2963                        sizeof(struct dquot), sizeof(unsigned long) * 4,
2964                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2965                                SLAB_MEM_SPREAD|SLAB_PANIC),
2966                        NULL);
2967
2968        order = 0;
2969        dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
2970        if (!dquot_hash)
2971                panic("Cannot create dquot hash table");
2972
2973        for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2974                ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2975                if (ret)
2976                        panic("Cannot create dquot stat counters");
2977        }
2978
2979        /* Find power-of-two hlist_heads which can fit into allocation */
2980        nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2981        dq_hash_bits = 0;
2982        do {
2983                dq_hash_bits++;
2984        } while (nr_hash >> dq_hash_bits);
2985        dq_hash_bits--;
2986
2987        nr_hash = 1UL << dq_hash_bits;
2988        dq_hash_mask = nr_hash - 1;
2989        for (i = 0; i < nr_hash; i++)
2990                INIT_HLIST_HEAD(dquot_hash + i);
2991
2992        pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2993                " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2994
2995        if (register_shrinker(&dqcache_shrinker))
2996                panic("Cannot register dquot shrinker");
2997
2998        return 0;
2999}
3000fs_initcall(dquot_init);
3001