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 four lists for dquot management: the inuse_list,
 227 * free_dquots, dqi_dirty_list, and dquot_hash[] array. A single dquot
 228 * structure may be on some of those 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 * Dirty dquots are added to the dqi_dirty_list of quota_info when mark
 240 * dirtied, and this list is searched when writing dirty dquots back to
 241 * quota file. Note that some filesystems do dirty dquot tracking on their
 242 * own (e.g. in a journal) and thus don't use dqi_dirty_list.
 243 *
 244 * Dquots with a specific identity (device, type and id) are placed on
 245 * one of the dquot_hash[] hash chains. The provides an efficient search
 246 * mechanism to locate a specific dquot.
 247 */
 248
 249static LIST_HEAD(inuse_list);
 250static LIST_HEAD(free_dquots);
 251static unsigned int dq_hash_bits, dq_hash_mask;
 252static struct hlist_head *dquot_hash;
 253
 254struct dqstats dqstats;
 255EXPORT_SYMBOL(dqstats);
 256
 257static qsize_t inode_get_rsv_space(struct inode *inode);
 258static qsize_t __inode_get_rsv_space(struct inode *inode);
 259static int __dquot_initialize(struct inode *inode, int type);
 260
 261static inline unsigned int
 262hashfn(const struct super_block *sb, struct kqid qid)
 263{
 264        unsigned int id = from_kqid(&init_user_ns, qid);
 265        int type = qid.type;
 266        unsigned long tmp;
 267
 268        tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
 269        return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
 270}
 271
 272/*
 273 * Following list functions expect dq_list_lock to be held
 274 */
 275static inline void insert_dquot_hash(struct dquot *dquot)
 276{
 277        struct hlist_head *head;
 278        head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id);
 279        hlist_add_head(&dquot->dq_hash, head);
 280}
 281
 282static inline void remove_dquot_hash(struct dquot *dquot)
 283{
 284        hlist_del_init(&dquot->dq_hash);
 285}
 286
 287static struct dquot *find_dquot(unsigned int hashent, struct super_block *sb,
 288                                struct kqid qid)
 289{
 290        struct hlist_node *node;
 291        struct dquot *dquot;
 292
 293        hlist_for_each (node, dquot_hash+hashent) {
 294                dquot = hlist_entry(node, struct dquot, dq_hash);
 295                if (dquot->dq_sb == sb && qid_eq(dquot->dq_id, qid))
 296                        return dquot;
 297        }
 298        return NULL;
 299}
 300
 301/* Add a dquot to the tail of the free list */
 302static inline void put_dquot_last(struct dquot *dquot)
 303{
 304        list_add_tail(&dquot->dq_free, &free_dquots);
 305        dqstats_inc(DQST_FREE_DQUOTS);
 306}
 307
 308static inline void remove_free_dquot(struct dquot *dquot)
 309{
 310        if (list_empty(&dquot->dq_free))
 311                return;
 312        list_del_init(&dquot->dq_free);
 313        dqstats_dec(DQST_FREE_DQUOTS);
 314}
 315
 316static inline void put_inuse(struct dquot *dquot)
 317{
 318        /* We add to the back of inuse list so we don't have to restart
 319         * when traversing this list and we block */
 320        list_add_tail(&dquot->dq_inuse, &inuse_list);
 321        dqstats_inc(DQST_ALLOC_DQUOTS);
 322}
 323
 324static inline void remove_inuse(struct dquot *dquot)
 325{
 326        dqstats_dec(DQST_ALLOC_DQUOTS);
 327        list_del(&dquot->dq_inuse);
 328}
 329/*
 330 * End of list functions needing dq_list_lock
 331 */
 332
 333static void wait_on_dquot(struct dquot *dquot)
 334{
 335        mutex_lock(&dquot->dq_lock);
 336        mutex_unlock(&dquot->dq_lock);
 337}
 338
 339static inline int dquot_dirty(struct dquot *dquot)
 340{
 341        return test_bit(DQ_MOD_B, &dquot->dq_flags);
 342}
 343
 344static inline int mark_dquot_dirty(struct dquot *dquot)
 345{
 346        return dquot->dq_sb->dq_op->mark_dirty(dquot);
 347}
 348
 349/* Mark dquot dirty in atomic manner, and return it's old dirty flag state */
 350int dquot_mark_dquot_dirty(struct dquot *dquot)
 351{
 352        int ret = 1;
 353
 354        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 355                return 0;
 356
 357        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
 358                return test_and_set_bit(DQ_MOD_B, &dquot->dq_flags);
 359
 360        /* If quota is dirty already, we don't have to acquire dq_list_lock */
 361        if (test_bit(DQ_MOD_B, &dquot->dq_flags))
 362                return 1;
 363
 364        spin_lock(&dq_list_lock);
 365        if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags)) {
 366                list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
 367                                info[dquot->dq_id.type].dqi_dirty_list);
 368                ret = 0;
 369        }
 370        spin_unlock(&dq_list_lock);
 371        return ret;
 372}
 373EXPORT_SYMBOL(dquot_mark_dquot_dirty);
 374
 375/* Dirtify all the dquots - this can block when journalling */
 376static inline int mark_all_dquot_dirty(struct dquot * const *dquot)
 377{
 378        int ret, err, cnt;
 379
 380        ret = err = 0;
 381        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 382                if (dquot[cnt])
 383                        /* Even in case of error we have to continue */
 384                        ret = mark_dquot_dirty(dquot[cnt]);
 385                if (!err)
 386                        err = ret;
 387        }
 388        return err;
 389}
 390
 391static inline void dqput_all(struct dquot **dquot)
 392{
 393        unsigned int cnt;
 394
 395        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 396                dqput(dquot[cnt]);
 397}
 398
 399static inline int clear_dquot_dirty(struct dquot *dquot)
 400{
 401        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NOLIST_DIRTY)
 402                return test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags);
 403
 404        spin_lock(&dq_list_lock);
 405        if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags)) {
 406                spin_unlock(&dq_list_lock);
 407                return 0;
 408        }
 409        list_del_init(&dquot->dq_dirty);
 410        spin_unlock(&dq_list_lock);
 411        return 1;
 412}
 413
 414void mark_info_dirty(struct super_block *sb, int type)
 415{
 416        spin_lock(&dq_data_lock);
 417        sb_dqopt(sb)->info[type].dqi_flags |= DQF_INFO_DIRTY;
 418        spin_unlock(&dq_data_lock);
 419}
 420EXPORT_SYMBOL(mark_info_dirty);
 421
 422/*
 423 *      Read dquot from disk and alloc space for it
 424 */
 425
 426int dquot_acquire(struct dquot *dquot)
 427{
 428        int ret = 0, ret2 = 0;
 429        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 430
 431        mutex_lock(&dquot->dq_lock);
 432        if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
 433                ret = dqopt->ops[dquot->dq_id.type]->read_dqblk(dquot);
 434                if (ret < 0)
 435                        goto out_iolock;
 436        }
 437        /* Make sure flags update is visible after dquot has been filled */
 438        smp_mb__before_atomic();
 439        set_bit(DQ_READ_B, &dquot->dq_flags);
 440        /* Instantiate dquot if needed */
 441        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
 442                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 443                /* Write the info if needed */
 444                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 445                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 446                                        dquot->dq_sb, dquot->dq_id.type);
 447                }
 448                if (ret < 0)
 449                        goto out_iolock;
 450                if (ret2 < 0) {
 451                        ret = ret2;
 452                        goto out_iolock;
 453                }
 454        }
 455        /*
 456         * Make sure flags update is visible after on-disk struct has been
 457         * allocated. Paired with smp_rmb() in dqget().
 458         */
 459        smp_mb__before_atomic();
 460        set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 461out_iolock:
 462        mutex_unlock(&dquot->dq_lock);
 463        return ret;
 464}
 465EXPORT_SYMBOL(dquot_acquire);
 466
 467/*
 468 *      Write dquot to disk
 469 */
 470int dquot_commit(struct dquot *dquot)
 471{
 472        int ret = 0;
 473        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 474
 475        mutex_lock(&dquot->dq_lock);
 476        if (!clear_dquot_dirty(dquot))
 477                goto out_lock;
 478        /* Inactive dquot can be only if there was error during read/init
 479         * => we have better not writing it */
 480        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 481                ret = dqopt->ops[dquot->dq_id.type]->commit_dqblk(dquot);
 482        else
 483                ret = -EIO;
 484out_lock:
 485        mutex_unlock(&dquot->dq_lock);
 486        return ret;
 487}
 488EXPORT_SYMBOL(dquot_commit);
 489
 490/*
 491 *      Release dquot
 492 */
 493int dquot_release(struct dquot *dquot)
 494{
 495        int ret = 0, ret2 = 0;
 496        struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
 497
 498        mutex_lock(&dquot->dq_lock);
 499        /* Check whether we are not racing with some other dqget() */
 500        if (atomic_read(&dquot->dq_count) > 1)
 501                goto out_dqlock;
 502        if (dqopt->ops[dquot->dq_id.type]->release_dqblk) {
 503                ret = dqopt->ops[dquot->dq_id.type]->release_dqblk(dquot);
 504                /* Write the info */
 505                if (info_dirty(&dqopt->info[dquot->dq_id.type])) {
 506                        ret2 = dqopt->ops[dquot->dq_id.type]->write_file_info(
 507                                                dquot->dq_sb, dquot->dq_id.type);
 508                }
 509                if (ret >= 0)
 510                        ret = ret2;
 511        }
 512        clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 513out_dqlock:
 514        mutex_unlock(&dquot->dq_lock);
 515        return ret;
 516}
 517EXPORT_SYMBOL(dquot_release);
 518
 519void dquot_destroy(struct dquot *dquot)
 520{
 521        kmem_cache_free(dquot_cachep, dquot);
 522}
 523EXPORT_SYMBOL(dquot_destroy);
 524
 525static inline void do_destroy_dquot(struct dquot *dquot)
 526{
 527        dquot->dq_sb->dq_op->destroy_dquot(dquot);
 528}
 529
 530/* Invalidate all dquots on the list. Note that this function is called after
 531 * quota is disabled and pointers from inodes removed so there cannot be new
 532 * quota users. There can still be some users of quotas due to inodes being
 533 * just deleted or pruned by prune_icache() (those are not attached to any
 534 * list) or parallel quotactl call. We have to wait for such users.
 535 */
 536static void invalidate_dquots(struct super_block *sb, int type)
 537{
 538        struct dquot *dquot, *tmp;
 539
 540restart:
 541        spin_lock(&dq_list_lock);
 542        list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
 543                if (dquot->dq_sb != sb)
 544                        continue;
 545                if (dquot->dq_id.type != type)
 546                        continue;
 547                /* Wait for dquot users */
 548                if (atomic_read(&dquot->dq_count)) {
 549                        dqgrab(dquot);
 550                        spin_unlock(&dq_list_lock);
 551                        /*
 552                         * Once dqput() wakes us up, we know it's time to free
 553                         * the dquot.
 554                         * IMPORTANT: we rely on the fact that there is always
 555                         * at most one process waiting for dquot to free.
 556                         * Otherwise dq_count would be > 1 and we would never
 557                         * wake up.
 558                         */
 559                        wait_event(dquot_ref_wq,
 560                                   atomic_read(&dquot->dq_count) == 1);
 561                        dqput(dquot);
 562                        /* At this moment dquot() need not exist (it could be
 563                         * reclaimed by prune_dqcache(). Hence we must
 564                         * restart. */
 565                        goto restart;
 566                }
 567                /*
 568                 * Quota now has no users and it has been written on last
 569                 * dqput()
 570                 */
 571                remove_dquot_hash(dquot);
 572                remove_free_dquot(dquot);
 573                remove_inuse(dquot);
 574                do_destroy_dquot(dquot);
 575        }
 576        spin_unlock(&dq_list_lock);
 577}
 578
 579/* Call callback for every active dquot on given filesystem */
 580int dquot_scan_active(struct super_block *sb,
 581                      int (*fn)(struct dquot *dquot, unsigned long priv),
 582                      unsigned long priv)
 583{
 584        struct dquot *dquot, *old_dquot = NULL;
 585        int ret = 0;
 586
 587        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 588
 589        spin_lock(&dq_list_lock);
 590        list_for_each_entry(dquot, &inuse_list, dq_inuse) {
 591                if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags))
 592                        continue;
 593                if (dquot->dq_sb != sb)
 594                        continue;
 595                /* Now we have active dquot so we can just increase use count */
 596                atomic_inc(&dquot->dq_count);
 597                spin_unlock(&dq_list_lock);
 598                dqstats_inc(DQST_LOOKUPS);
 599                dqput(old_dquot);
 600                old_dquot = dquot;
 601                /*
 602                 * ->release_dquot() can be racing with us. Our reference
 603                 * protects us from new calls to it so just wait for any
 604                 * outstanding call and recheck the DQ_ACTIVE_B after that.
 605                 */
 606                wait_on_dquot(dquot);
 607                if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 608                        ret = fn(dquot, priv);
 609                        if (ret < 0)
 610                                goto out;
 611                }
 612                spin_lock(&dq_list_lock);
 613                /* We are safe to continue now because our dquot could not
 614                 * be moved out of the inuse list while we hold the reference */
 615        }
 616        spin_unlock(&dq_list_lock);
 617out:
 618        dqput(old_dquot);
 619        return ret;
 620}
 621EXPORT_SYMBOL(dquot_scan_active);
 622
 623/* Write all dquot structures to quota files */
 624int dquot_writeback_dquots(struct super_block *sb, int type)
 625{
 626        struct list_head *dirty;
 627        struct dquot *dquot;
 628        struct quota_info *dqopt = sb_dqopt(sb);
 629        int cnt;
 630        int err, ret = 0;
 631
 632        WARN_ON_ONCE(!rwsem_is_locked(&sb->s_umount));
 633
 634        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 635                if (type != -1 && cnt != type)
 636                        continue;
 637                if (!sb_has_quota_active(sb, cnt))
 638                        continue;
 639                spin_lock(&dq_list_lock);
 640                dirty = &dqopt->info[cnt].dqi_dirty_list;
 641                while (!list_empty(dirty)) {
 642                        dquot = list_first_entry(dirty, struct dquot,
 643                                                 dq_dirty);
 644
 645                        WARN_ON(!test_bit(DQ_ACTIVE_B, &dquot->dq_flags));
 646
 647                        /* Now we have active dquot from which someone is
 648                         * holding reference so we can safely just increase
 649                         * use count */
 650                        dqgrab(dquot);
 651                        spin_unlock(&dq_list_lock);
 652                        dqstats_inc(DQST_LOOKUPS);
 653                        err = sb->dq_op->write_dquot(dquot);
 654                        if (err) {
 655                                /*
 656                                 * Clear dirty bit anyway to avoid infinite
 657                                 * loop here.
 658                                 */
 659                                clear_dquot_dirty(dquot);
 660                                if (!ret)
 661                                        ret = err;
 662                        }
 663                        dqput(dquot);
 664                        spin_lock(&dq_list_lock);
 665                }
 666                spin_unlock(&dq_list_lock);
 667        }
 668
 669        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 670                if ((cnt == type || type == -1) && sb_has_quota_active(sb, cnt)
 671                    && info_dirty(&dqopt->info[cnt]))
 672                        sb->dq_op->write_info(sb, cnt);
 673        dqstats_inc(DQST_SYNCS);
 674
 675        return ret;
 676}
 677EXPORT_SYMBOL(dquot_writeback_dquots);
 678
 679/* Write all dquot structures to disk and make them visible from userspace */
 680int dquot_quota_sync(struct super_block *sb, int type)
 681{
 682        struct quota_info *dqopt = sb_dqopt(sb);
 683        int cnt;
 684        int ret;
 685
 686        ret = dquot_writeback_dquots(sb, type);
 687        if (ret)
 688                return ret;
 689        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
 690                return 0;
 691
 692        /* This is not very clever (and fast) but currently I don't know about
 693         * any other simple way of getting quota data to disk and we must get
 694         * them there for userspace to be visible... */
 695        if (sb->s_op->sync_fs)
 696                sb->s_op->sync_fs(sb, 1);
 697        sync_blockdev(sb->s_bdev);
 698
 699        /*
 700         * Now when everything is written we can discard the pagecache so
 701         * that userspace sees the changes.
 702         */
 703        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 704                if (type != -1 && cnt != type)
 705                        continue;
 706                if (!sb_has_quota_active(sb, cnt))
 707                        continue;
 708                inode_lock(dqopt->files[cnt]);
 709                truncate_inode_pages(&dqopt->files[cnt]->i_data, 0);
 710                inode_unlock(dqopt->files[cnt]);
 711        }
 712
 713        return 0;
 714}
 715EXPORT_SYMBOL(dquot_quota_sync);
 716
 717static unsigned long
 718dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 719{
 720        struct dquot *dquot;
 721        unsigned long freed = 0;
 722
 723        spin_lock(&dq_list_lock);
 724        while (!list_empty(&free_dquots) && sc->nr_to_scan) {
 725                dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
 726                remove_dquot_hash(dquot);
 727                remove_free_dquot(dquot);
 728                remove_inuse(dquot);
 729                do_destroy_dquot(dquot);
 730                sc->nr_to_scan--;
 731                freed++;
 732        }
 733        spin_unlock(&dq_list_lock);
 734        return freed;
 735}
 736
 737static unsigned long
 738dqcache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
 739{
 740        return vfs_pressure_ratio(
 741        percpu_counter_read_positive(&dqstats.counter[DQST_FREE_DQUOTS]));
 742}
 743
 744static struct shrinker dqcache_shrinker = {
 745        .count_objects = dqcache_shrink_count,
 746        .scan_objects = dqcache_shrink_scan,
 747        .seeks = DEFAULT_SEEKS,
 748};
 749
 750/*
 751 * Put reference to dquot
 752 */
 753void dqput(struct dquot *dquot)
 754{
 755        int ret;
 756
 757        if (!dquot)
 758                return;
 759#ifdef CONFIG_QUOTA_DEBUG
 760        if (!atomic_read(&dquot->dq_count)) {
 761                quota_error(dquot->dq_sb, "trying to free free dquot of %s %d",
 762                            quotatypes[dquot->dq_id.type],
 763                            from_kqid(&init_user_ns, dquot->dq_id));
 764                BUG();
 765        }
 766#endif
 767        dqstats_inc(DQST_DROPS);
 768we_slept:
 769        spin_lock(&dq_list_lock);
 770        if (atomic_read(&dquot->dq_count) > 1) {
 771                /* We have more than one user... nothing to do */
 772                atomic_dec(&dquot->dq_count);
 773                /* Releasing dquot during quotaoff phase? */
 774                if (!sb_has_quota_active(dquot->dq_sb, dquot->dq_id.type) &&
 775                    atomic_read(&dquot->dq_count) == 1)
 776                        wake_up(&dquot_ref_wq);
 777                spin_unlock(&dq_list_lock);
 778                return;
 779        }
 780        /* Need to release dquot? */
 781        if (dquot_dirty(dquot)) {
 782                spin_unlock(&dq_list_lock);
 783                /* Commit dquot before releasing */
 784                ret = dquot->dq_sb->dq_op->write_dquot(dquot);
 785                if (ret < 0) {
 786                        quota_error(dquot->dq_sb, "Can't write quota structure"
 787                                    " (error %d). Quota may get out of sync!",
 788                                    ret);
 789                        /*
 790                         * We clear dirty bit anyway, so that we avoid
 791                         * infinite loop here
 792                         */
 793                        clear_dquot_dirty(dquot);
 794                }
 795                goto we_slept;
 796        }
 797        if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 798                spin_unlock(&dq_list_lock);
 799                dquot->dq_sb->dq_op->release_dquot(dquot);
 800                goto we_slept;
 801        }
 802        atomic_dec(&dquot->dq_count);
 803#ifdef CONFIG_QUOTA_DEBUG
 804        /* sanity check */
 805        BUG_ON(!list_empty(&dquot->dq_free));
 806#endif
 807        put_dquot_last(dquot);
 808        spin_unlock(&dq_list_lock);
 809}
 810EXPORT_SYMBOL(dqput);
 811
 812struct dquot *dquot_alloc(struct super_block *sb, int type)
 813{
 814        return kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
 815}
 816EXPORT_SYMBOL(dquot_alloc);
 817
 818static struct dquot *get_empty_dquot(struct super_block *sb, int type)
 819{
 820        struct dquot *dquot;
 821
 822        dquot = sb->dq_op->alloc_dquot(sb, type);
 823        if(!dquot)
 824                return NULL;
 825
 826        mutex_init(&dquot->dq_lock);
 827        INIT_LIST_HEAD(&dquot->dq_free);
 828        INIT_LIST_HEAD(&dquot->dq_inuse);
 829        INIT_HLIST_NODE(&dquot->dq_hash);
 830        INIT_LIST_HEAD(&dquot->dq_dirty);
 831        dquot->dq_sb = sb;
 832        dquot->dq_id = make_kqid_invalid(type);
 833        atomic_set(&dquot->dq_count, 1);
 834        spin_lock_init(&dquot->dq_dqb_lock);
 835
 836        return dquot;
 837}
 838
 839/*
 840 * Get reference to dquot
 841 *
 842 * Locking is slightly tricky here. We are guarded from parallel quotaoff()
 843 * destroying our dquot by:
 844 *   a) checking for quota flags under dq_list_lock and
 845 *   b) getting a reference to dquot before we release dq_list_lock
 846 */
 847struct dquot *dqget(struct super_block *sb, struct kqid qid)
 848{
 849        unsigned int hashent = hashfn(sb, qid);
 850        struct dquot *dquot, *empty = NULL;
 851
 852        if (!qid_has_mapping(sb->s_user_ns, qid))
 853                return ERR_PTR(-EINVAL);
 854
 855        if (!sb_has_quota_active(sb, qid.type))
 856                return ERR_PTR(-ESRCH);
 857we_slept:
 858        spin_lock(&dq_list_lock);
 859        spin_lock(&dq_state_lock);
 860        if (!sb_has_quota_active(sb, qid.type)) {
 861                spin_unlock(&dq_state_lock);
 862                spin_unlock(&dq_list_lock);
 863                dquot = ERR_PTR(-ESRCH);
 864                goto out;
 865        }
 866        spin_unlock(&dq_state_lock);
 867
 868        dquot = find_dquot(hashent, sb, qid);
 869        if (!dquot) {
 870                if (!empty) {
 871                        spin_unlock(&dq_list_lock);
 872                        empty = get_empty_dquot(sb, qid.type);
 873                        if (!empty)
 874                                schedule();     /* Try to wait for a moment... */
 875                        goto we_slept;
 876                }
 877                dquot = empty;
 878                empty = NULL;
 879                dquot->dq_id = qid;
 880                /* all dquots go on the inuse_list */
 881                put_inuse(dquot);
 882                /* hash it first so it can be found */
 883                insert_dquot_hash(dquot);
 884                spin_unlock(&dq_list_lock);
 885                dqstats_inc(DQST_LOOKUPS);
 886        } else {
 887                if (!atomic_read(&dquot->dq_count))
 888                        remove_free_dquot(dquot);
 889                atomic_inc(&dquot->dq_count);
 890                spin_unlock(&dq_list_lock);
 891                dqstats_inc(DQST_CACHE_HITS);
 892                dqstats_inc(DQST_LOOKUPS);
 893        }
 894        /* Wait for dq_lock - after this we know that either dquot_release() is
 895         * already finished or it will be canceled due to dq_count > 1 test */
 896        wait_on_dquot(dquot);
 897        /* Read the dquot / allocate space in quota file */
 898        if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
 899                int err;
 900
 901                err = sb->dq_op->acquire_dquot(dquot);
 902                if (err < 0) {
 903                        dqput(dquot);
 904                        dquot = ERR_PTR(err);
 905                        goto out;
 906                }
 907        }
 908        /*
 909         * Make sure following reads see filled structure - paired with
 910         * smp_mb__before_atomic() in dquot_acquire().
 911         */
 912        smp_rmb();
 913#ifdef CONFIG_QUOTA_DEBUG
 914        BUG_ON(!dquot->dq_sb);  /* Has somebody invalidated entry under us? */
 915#endif
 916out:
 917        if (empty)
 918                do_destroy_dquot(empty);
 919
 920        return dquot;
 921}
 922EXPORT_SYMBOL(dqget);
 923
 924static inline struct dquot **i_dquot(struct inode *inode)
 925{
 926        return inode->i_sb->s_op->get_dquots(inode);
 927}
 928
 929static int dqinit_needed(struct inode *inode, int type)
 930{
 931        struct dquot * const *dquots;
 932        int cnt;
 933
 934        if (IS_NOQUOTA(inode))
 935                return 0;
 936
 937        dquots = i_dquot(inode);
 938        if (type != -1)
 939                return !dquots[type];
 940        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
 941                if (!dquots[cnt])
 942                        return 1;
 943        return 0;
 944}
 945
 946/* This routine is guarded by s_umount semaphore */
 947static int add_dquot_ref(struct super_block *sb, int type)
 948{
 949        struct inode *inode, *old_inode = NULL;
 950#ifdef CONFIG_QUOTA_DEBUG
 951        int reserved = 0;
 952#endif
 953        int err = 0;
 954
 955        spin_lock(&sb->s_inode_list_lock);
 956        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
 957                spin_lock(&inode->i_lock);
 958                if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
 959                    !atomic_read(&inode->i_writecount) ||
 960                    !dqinit_needed(inode, type)) {
 961                        spin_unlock(&inode->i_lock);
 962                        continue;
 963                }
 964                __iget(inode);
 965                spin_unlock(&inode->i_lock);
 966                spin_unlock(&sb->s_inode_list_lock);
 967
 968#ifdef CONFIG_QUOTA_DEBUG
 969                if (unlikely(inode_get_rsv_space(inode) > 0))
 970                        reserved = 1;
 971#endif
 972                iput(old_inode);
 973                err = __dquot_initialize(inode, type);
 974                if (err) {
 975                        iput(inode);
 976                        goto out;
 977                }
 978
 979                /*
 980                 * We hold a reference to 'inode' so it couldn't have been
 981                 * removed from s_inodes list while we dropped the
 982                 * s_inode_list_lock. We cannot iput the inode now as we can be
 983                 * holding the last reference and we cannot iput it under
 984                 * s_inode_list_lock. So we keep the reference and iput it
 985                 * later.
 986                 */
 987                old_inode = inode;
 988                spin_lock(&sb->s_inode_list_lock);
 989        }
 990        spin_unlock(&sb->s_inode_list_lock);
 991        iput(old_inode);
 992out:
 993#ifdef CONFIG_QUOTA_DEBUG
 994        if (reserved) {
 995                quota_error(sb, "Writes happened before quota was turned on "
 996                        "thus quota information is probably inconsistent. "
 997                        "Please run quotacheck(8)");
 998        }
 999#endif
1000        return err;
1001}
1002
1003/*
1004 * Remove references to dquots from inode and add dquot to list for freeing
1005 * if we have the last reference to dquot
1006 */
1007static void remove_inode_dquot_ref(struct inode *inode, int type,
1008                                   struct list_head *tofree_head)
1009{
1010        struct dquot **dquots = i_dquot(inode);
1011        struct dquot *dquot = dquots[type];
1012
1013        if (!dquot)
1014                return;
1015
1016        dquots[type] = NULL;
1017        if (list_empty(&dquot->dq_free)) {
1018                /*
1019                 * The inode still has reference to dquot so it can't be in the
1020                 * free list
1021                 */
1022                spin_lock(&dq_list_lock);
1023                list_add(&dquot->dq_free, tofree_head);
1024                spin_unlock(&dq_list_lock);
1025        } else {
1026                /*
1027                 * Dquot is already in a list to put so we won't drop the last
1028                 * reference here.
1029                 */
1030                dqput(dquot);
1031        }
1032}
1033
1034/*
1035 * Free list of dquots
1036 * Dquots are removed from inodes and no new references can be got so we are
1037 * the only ones holding reference
1038 */
1039static void put_dquot_list(struct list_head *tofree_head)
1040{
1041        struct list_head *act_head;
1042        struct dquot *dquot;
1043
1044        act_head = tofree_head->next;
1045        while (act_head != tofree_head) {
1046                dquot = list_entry(act_head, struct dquot, dq_free);
1047                act_head = act_head->next;
1048                /* Remove dquot from the list so we won't have problems... */
1049                list_del_init(&dquot->dq_free);
1050                dqput(dquot);
1051        }
1052}
1053
1054static void remove_dquot_ref(struct super_block *sb, int type,
1055                struct list_head *tofree_head)
1056{
1057        struct inode *inode;
1058        int reserved = 0;
1059
1060        spin_lock(&sb->s_inode_list_lock);
1061        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1062                /*
1063                 *  We have to scan also I_NEW inodes because they can already
1064                 *  have quota pointer initialized. Luckily, we need to touch
1065                 *  only quota pointers and these have separate locking
1066                 *  (dq_data_lock).
1067                 */
1068                spin_lock(&dq_data_lock);
1069                if (!IS_NOQUOTA(inode)) {
1070                        if (unlikely(inode_get_rsv_space(inode) > 0))
1071                                reserved = 1;
1072                        remove_inode_dquot_ref(inode, type, tofree_head);
1073                }
1074                spin_unlock(&dq_data_lock);
1075        }
1076        spin_unlock(&sb->s_inode_list_lock);
1077#ifdef CONFIG_QUOTA_DEBUG
1078        if (reserved) {
1079                printk(KERN_WARNING "VFS (%s): Writes happened after quota"
1080                        " was disabled thus quota information is probably "
1081                        "inconsistent. Please run quotacheck(8).\n", sb->s_id);
1082        }
1083#endif
1084}
1085
1086/* Gather all references from inodes and drop them */
1087static void drop_dquot_ref(struct super_block *sb, int type)
1088{
1089        LIST_HEAD(tofree_head);
1090
1091        if (sb->dq_op) {
1092                remove_dquot_ref(sb, type, &tofree_head);
1093                synchronize_srcu(&dquot_srcu);
1094                put_dquot_list(&tofree_head);
1095        }
1096}
1097
1098static inline
1099void dquot_free_reserved_space(struct dquot *dquot, qsize_t number)
1100{
1101        if (dquot->dq_dqb.dqb_rsvspace >= number)
1102                dquot->dq_dqb.dqb_rsvspace -= number;
1103        else {
1104                WARN_ON_ONCE(1);
1105                dquot->dq_dqb.dqb_rsvspace = 0;
1106        }
1107        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1108            dquot->dq_dqb.dqb_bsoftlimit)
1109                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1110        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1111}
1112
1113static void dquot_decr_inodes(struct dquot *dquot, qsize_t number)
1114{
1115        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1116            dquot->dq_dqb.dqb_curinodes >= number)
1117                dquot->dq_dqb.dqb_curinodes -= number;
1118        else
1119                dquot->dq_dqb.dqb_curinodes = 0;
1120        if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
1121                dquot->dq_dqb.dqb_itime = (time64_t) 0;
1122        clear_bit(DQ_INODES_B, &dquot->dq_flags);
1123}
1124
1125static void dquot_decr_space(struct dquot *dquot, qsize_t number)
1126{
1127        if (sb_dqopt(dquot->dq_sb)->flags & DQUOT_NEGATIVE_USAGE ||
1128            dquot->dq_dqb.dqb_curspace >= number)
1129                dquot->dq_dqb.dqb_curspace -= number;
1130        else
1131                dquot->dq_dqb.dqb_curspace = 0;
1132        if (dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace <=
1133            dquot->dq_dqb.dqb_bsoftlimit)
1134                dquot->dq_dqb.dqb_btime = (time64_t) 0;
1135        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
1136}
1137
1138struct dquot_warn {
1139        struct super_block *w_sb;
1140        struct kqid w_dq_id;
1141        short w_type;
1142};
1143
1144static int warning_issued(struct dquot *dquot, const int warntype)
1145{
1146        int flag = (warntype == QUOTA_NL_BHARDWARN ||
1147                warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
1148                ((warntype == QUOTA_NL_IHARDWARN ||
1149                warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
1150
1151        if (!flag)
1152                return 0;
1153        return test_and_set_bit(flag, &dquot->dq_flags);
1154}
1155
1156#ifdef CONFIG_PRINT_QUOTA_WARNING
1157static int flag_print_warnings = 1;
1158
1159static int need_print_warning(struct dquot_warn *warn)
1160{
1161        if (!flag_print_warnings)
1162                return 0;
1163
1164        switch (warn->w_dq_id.type) {
1165                case USRQUOTA:
1166                        return uid_eq(current_fsuid(), warn->w_dq_id.uid);
1167                case GRPQUOTA:
1168                        return in_group_p(warn->w_dq_id.gid);
1169                case PRJQUOTA:
1170                        return 1;
1171        }
1172        return 0;
1173}
1174
1175/* Print warning to user which exceeded quota */
1176static void print_warning(struct dquot_warn *warn)
1177{
1178        char *msg = NULL;
1179        struct tty_struct *tty;
1180        int warntype = warn->w_type;
1181
1182        if (warntype == QUOTA_NL_IHARDBELOW ||
1183            warntype == QUOTA_NL_ISOFTBELOW ||
1184            warntype == QUOTA_NL_BHARDBELOW ||
1185            warntype == QUOTA_NL_BSOFTBELOW || !need_print_warning(warn))
1186                return;
1187
1188        tty = get_current_tty();
1189        if (!tty)
1190                return;
1191        tty_write_message(tty, warn->w_sb->s_id);
1192        if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
1193                tty_write_message(tty, ": warning, ");
1194        else
1195                tty_write_message(tty, ": write failed, ");
1196        tty_write_message(tty, quotatypes[warn->w_dq_id.type]);
1197        switch (warntype) {
1198                case QUOTA_NL_IHARDWARN:
1199                        msg = " file limit reached.\r\n";
1200                        break;
1201                case QUOTA_NL_ISOFTLONGWARN:
1202                        msg = " file quota exceeded too long.\r\n";
1203                        break;
1204                case QUOTA_NL_ISOFTWARN:
1205                        msg = " file quota exceeded.\r\n";
1206                        break;
1207                case QUOTA_NL_BHARDWARN:
1208                        msg = " block limit reached.\r\n";
1209                        break;
1210                case QUOTA_NL_BSOFTLONGWARN:
1211                        msg = " block quota exceeded too long.\r\n";
1212                        break;
1213                case QUOTA_NL_BSOFTWARN:
1214                        msg = " block quota exceeded.\r\n";
1215                        break;
1216        }
1217        tty_write_message(tty, msg);
1218        tty_kref_put(tty);
1219}
1220#endif
1221
1222static void prepare_warning(struct dquot_warn *warn, struct dquot *dquot,
1223                            int warntype)
1224{
1225        if (warning_issued(dquot, warntype))
1226                return;
1227        warn->w_type = warntype;
1228        warn->w_sb = dquot->dq_sb;
1229        warn->w_dq_id = dquot->dq_id;
1230}
1231
1232/*
1233 * Write warnings to the console and send warning messages over netlink.
1234 *
1235 * Note that this function can call into tty and networking code.
1236 */
1237static void flush_warnings(struct dquot_warn *warn)
1238{
1239        int i;
1240
1241        for (i = 0; i < MAXQUOTAS; i++) {
1242                if (warn[i].w_type == QUOTA_NL_NOWARN)
1243                        continue;
1244#ifdef CONFIG_PRINT_QUOTA_WARNING
1245                print_warning(&warn[i]);
1246#endif
1247                quota_send_warning(warn[i].w_dq_id,
1248                                   warn[i].w_sb->s_dev, warn[i].w_type);
1249        }
1250}
1251
1252static int ignore_hardlimit(struct dquot *dquot)
1253{
1254        struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
1255
1256        return capable(CAP_SYS_RESOURCE) &&
1257               (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD ||
1258                !(info->dqi_flags & DQF_ROOT_SQUASH));
1259}
1260
1261static int dquot_add_inodes(struct dquot *dquot, qsize_t inodes,
1262                            struct dquot_warn *warn)
1263{
1264        qsize_t newinodes;
1265        int ret = 0;
1266
1267        spin_lock(&dquot->dq_dqb_lock);
1268        newinodes = dquot->dq_dqb.dqb_curinodes + inodes;
1269        if (!sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type) ||
1270            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1271                goto add;
1272
1273        if (dquot->dq_dqb.dqb_ihardlimit &&
1274            newinodes > dquot->dq_dqb.dqb_ihardlimit &&
1275            !ignore_hardlimit(dquot)) {
1276                prepare_warning(warn, dquot, QUOTA_NL_IHARDWARN);
1277                ret = -EDQUOT;
1278                goto out;
1279        }
1280
1281        if (dquot->dq_dqb.dqb_isoftlimit &&
1282            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1283            dquot->dq_dqb.dqb_itime &&
1284            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_itime &&
1285            !ignore_hardlimit(dquot)) {
1286                prepare_warning(warn, dquot, QUOTA_NL_ISOFTLONGWARN);
1287                ret = -EDQUOT;
1288                goto out;
1289        }
1290
1291        if (dquot->dq_dqb.dqb_isoftlimit &&
1292            newinodes > dquot->dq_dqb.dqb_isoftlimit &&
1293            dquot->dq_dqb.dqb_itime == 0) {
1294                prepare_warning(warn, dquot, QUOTA_NL_ISOFTWARN);
1295                dquot->dq_dqb.dqb_itime = ktime_get_real_seconds() +
1296                    sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type].dqi_igrace;
1297        }
1298add:
1299        dquot->dq_dqb.dqb_curinodes = newinodes;
1300
1301out:
1302        spin_unlock(&dquot->dq_dqb_lock);
1303        return ret;
1304}
1305
1306static int dquot_add_space(struct dquot *dquot, qsize_t space,
1307                           qsize_t rsv_space, unsigned int flags,
1308                           struct dquot_warn *warn)
1309{
1310        qsize_t tspace;
1311        struct super_block *sb = dquot->dq_sb;
1312        int ret = 0;
1313
1314        spin_lock(&dquot->dq_dqb_lock);
1315        if (!sb_has_quota_limits_enabled(sb, dquot->dq_id.type) ||
1316            test_bit(DQ_FAKE_B, &dquot->dq_flags))
1317                goto finish;
1318
1319        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace
1320                + space + rsv_space;
1321
1322        if (dquot->dq_dqb.dqb_bhardlimit &&
1323            tspace > dquot->dq_dqb.dqb_bhardlimit &&
1324            !ignore_hardlimit(dquot)) {
1325                if (flags & DQUOT_SPACE_WARN)
1326                        prepare_warning(warn, dquot, QUOTA_NL_BHARDWARN);
1327                ret = -EDQUOT;
1328                goto finish;
1329        }
1330
1331        if (dquot->dq_dqb.dqb_bsoftlimit &&
1332            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1333            dquot->dq_dqb.dqb_btime &&
1334            ktime_get_real_seconds() >= dquot->dq_dqb.dqb_btime &&
1335            !ignore_hardlimit(dquot)) {
1336                if (flags & DQUOT_SPACE_WARN)
1337                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTLONGWARN);
1338                ret = -EDQUOT;
1339                goto finish;
1340        }
1341
1342        if (dquot->dq_dqb.dqb_bsoftlimit &&
1343            tspace > dquot->dq_dqb.dqb_bsoftlimit &&
1344            dquot->dq_dqb.dqb_btime == 0) {
1345                if (flags & DQUOT_SPACE_WARN) {
1346                        prepare_warning(warn, dquot, QUOTA_NL_BSOFTWARN);
1347                        dquot->dq_dqb.dqb_btime = ktime_get_real_seconds() +
1348                            sb_dqopt(sb)->info[dquot->dq_id.type].dqi_bgrace;
1349                } else {
1350                        /*
1351                         * We don't allow preallocation to exceed softlimit so exceeding will
1352                         * be always printed
1353                         */
1354                        ret = -EDQUOT;
1355                        goto finish;
1356                }
1357        }
1358finish:
1359        /*
1360         * We have to be careful and go through warning generation & grace time
1361         * setting even if DQUOT_SPACE_NOFAIL is set. That's why we check it
1362         * only here...
1363         */
1364        if (flags & DQUOT_SPACE_NOFAIL)
1365                ret = 0;
1366        if (!ret) {
1367                dquot->dq_dqb.dqb_rsvspace += rsv_space;
1368                dquot->dq_dqb.dqb_curspace += space;
1369        }
1370        spin_unlock(&dquot->dq_dqb_lock);
1371        return ret;
1372}
1373
1374static int info_idq_free(struct dquot *dquot, qsize_t inodes)
1375{
1376        qsize_t newinodes;
1377
1378        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1379            dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit ||
1380            !sb_has_quota_limits_enabled(dquot->dq_sb, dquot->dq_id.type))
1381                return QUOTA_NL_NOWARN;
1382
1383        newinodes = dquot->dq_dqb.dqb_curinodes - inodes;
1384        if (newinodes <= dquot->dq_dqb.dqb_isoftlimit)
1385                return QUOTA_NL_ISOFTBELOW;
1386        if (dquot->dq_dqb.dqb_curinodes >= dquot->dq_dqb.dqb_ihardlimit &&
1387            newinodes < dquot->dq_dqb.dqb_ihardlimit)
1388                return QUOTA_NL_IHARDBELOW;
1389        return QUOTA_NL_NOWARN;
1390}
1391
1392static int info_bdq_free(struct dquot *dquot, qsize_t space)
1393{
1394        qsize_t tspace;
1395
1396        tspace = dquot->dq_dqb.dqb_curspace + dquot->dq_dqb.dqb_rsvspace;
1397
1398        if (test_bit(DQ_FAKE_B, &dquot->dq_flags) ||
1399            tspace <= dquot->dq_dqb.dqb_bsoftlimit)
1400                return QUOTA_NL_NOWARN;
1401
1402        if (tspace - space <= dquot->dq_dqb.dqb_bsoftlimit)
1403                return QUOTA_NL_BSOFTBELOW;
1404        if (tspace >= dquot->dq_dqb.dqb_bhardlimit &&
1405            tspace - space < dquot->dq_dqb.dqb_bhardlimit)
1406                return QUOTA_NL_BHARDBELOW;
1407        return QUOTA_NL_NOWARN;
1408}
1409
1410static int dquot_active(const struct inode *inode)
1411{
1412        struct super_block *sb = inode->i_sb;
1413
1414        if (IS_NOQUOTA(inode))
1415                return 0;
1416        return sb_any_quota_loaded(sb) & ~sb_any_quota_suspended(sb);
1417}
1418
1419/*
1420 * Initialize quota pointers in inode
1421 *
1422 * It is better to call this function outside of any transaction as it
1423 * might need a lot of space in journal for dquot structure allocation.
1424 */
1425static int __dquot_initialize(struct inode *inode, int type)
1426{
1427        int cnt, init_needed = 0;
1428        struct dquot **dquots, *got[MAXQUOTAS] = {};
1429        struct super_block *sb = inode->i_sb;
1430        qsize_t rsv;
1431        int ret = 0;
1432
1433        if (!dquot_active(inode))
1434                return 0;
1435
1436        dquots = i_dquot(inode);
1437
1438        /* First get references to structures we might need. */
1439        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1440                struct kqid qid;
1441                kprojid_t projid;
1442                int rc;
1443                struct dquot *dquot;
1444
1445                if (type != -1 && cnt != type)
1446                        continue;
1447                /*
1448                 * The i_dquot should have been initialized in most cases,
1449                 * we check it without locking here to avoid unnecessary
1450                 * dqget()/dqput() calls.
1451                 */
1452                if (dquots[cnt])
1453                        continue;
1454
1455                if (!sb_has_quota_active(sb, cnt))
1456                        continue;
1457
1458                init_needed = 1;
1459
1460                switch (cnt) {
1461                case USRQUOTA:
1462                        qid = make_kqid_uid(inode->i_uid);
1463                        break;
1464                case GRPQUOTA:
1465                        qid = make_kqid_gid(inode->i_gid);
1466                        break;
1467                case PRJQUOTA:
1468                        rc = inode->i_sb->dq_op->get_projid(inode, &projid);
1469                        if (rc)
1470                                continue;
1471                        qid = make_kqid_projid(projid);
1472                        break;
1473                }
1474                dquot = dqget(sb, qid);
1475                if (IS_ERR(dquot)) {
1476                        /* We raced with somebody turning quotas off... */
1477                        if (PTR_ERR(dquot) != -ESRCH) {
1478                                ret = PTR_ERR(dquot);
1479                                goto out_put;
1480                        }
1481                        dquot = NULL;
1482                }
1483                got[cnt] = dquot;
1484        }
1485
1486        /* All required i_dquot has been initialized */
1487        if (!init_needed)
1488                return 0;
1489
1490        spin_lock(&dq_data_lock);
1491        if (IS_NOQUOTA(inode))
1492                goto out_lock;
1493        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1494                if (type != -1 && cnt != type)
1495                        continue;
1496                /* Avoid races with quotaoff() */
1497                if (!sb_has_quota_active(sb, cnt))
1498                        continue;
1499                /* We could race with quotaon or dqget() could have failed */
1500                if (!got[cnt])
1501                        continue;
1502                if (!dquots[cnt]) {
1503                        dquots[cnt] = got[cnt];
1504                        got[cnt] = NULL;
1505                        /*
1506                         * Make quota reservation system happy if someone
1507                         * did a write before quota was turned on
1508                         */
1509                        rsv = inode_get_rsv_space(inode);
1510                        if (unlikely(rsv)) {
1511                                spin_lock(&inode->i_lock);
1512                                /* Get reservation again under proper lock */
1513                                rsv = __inode_get_rsv_space(inode);
1514                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1515                                dquots[cnt]->dq_dqb.dqb_rsvspace += rsv;
1516                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1517                                spin_unlock(&inode->i_lock);
1518                        }
1519                }
1520        }
1521out_lock:
1522        spin_unlock(&dq_data_lock);
1523out_put:
1524        /* Drop unused references */
1525        dqput_all(got);
1526
1527        return ret;
1528}
1529
1530int dquot_initialize(struct inode *inode)
1531{
1532        return __dquot_initialize(inode, -1);
1533}
1534EXPORT_SYMBOL(dquot_initialize);
1535
1536bool dquot_initialize_needed(struct inode *inode)
1537{
1538        struct dquot **dquots;
1539        int i;
1540
1541        if (!dquot_active(inode))
1542                return false;
1543
1544        dquots = i_dquot(inode);
1545        for (i = 0; i < MAXQUOTAS; i++)
1546                if (!dquots[i] && sb_has_quota_active(inode->i_sb, i))
1547                        return true;
1548        return false;
1549}
1550EXPORT_SYMBOL(dquot_initialize_needed);
1551
1552/*
1553 * Release all quotas referenced by inode.
1554 *
1555 * This function only be called on inode free or converting
1556 * a file to quota file, no other users for the i_dquot in
1557 * both cases, so we needn't call synchronize_srcu() after
1558 * clearing i_dquot.
1559 */
1560static void __dquot_drop(struct inode *inode)
1561{
1562        int cnt;
1563        struct dquot **dquots = i_dquot(inode);
1564        struct dquot *put[MAXQUOTAS];
1565
1566        spin_lock(&dq_data_lock);
1567        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1568                put[cnt] = dquots[cnt];
1569                dquots[cnt] = NULL;
1570        }
1571        spin_unlock(&dq_data_lock);
1572        dqput_all(put);
1573}
1574
1575void dquot_drop(struct inode *inode)
1576{
1577        struct dquot * const *dquots;
1578        int cnt;
1579
1580        if (IS_NOQUOTA(inode))
1581                return;
1582
1583        /*
1584         * Test before calling to rule out calls from proc and such
1585         * where we are not allowed to block. Note that this is
1586         * actually reliable test even without the lock - the caller
1587         * must assure that nobody can come after the DQUOT_DROP and
1588         * add quota pointers back anyway.
1589         */
1590        dquots = i_dquot(inode);
1591        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1592                if (dquots[cnt])
1593                        break;
1594        }
1595
1596        if (cnt < MAXQUOTAS)
1597                __dquot_drop(inode);
1598}
1599EXPORT_SYMBOL(dquot_drop);
1600
1601/*
1602 * inode_reserved_space is managed internally by quota, and protected by
1603 * i_lock similar to i_blocks+i_bytes.
1604 */
1605static qsize_t *inode_reserved_space(struct inode * inode)
1606{
1607        /* Filesystem must explicitly define it's own method in order to use
1608         * quota reservation interface */
1609        BUG_ON(!inode->i_sb->dq_op->get_reserved_space);
1610        return inode->i_sb->dq_op->get_reserved_space(inode);
1611}
1612
1613static qsize_t __inode_get_rsv_space(struct inode *inode)
1614{
1615        if (!inode->i_sb->dq_op->get_reserved_space)
1616                return 0;
1617        return *inode_reserved_space(inode);
1618}
1619
1620static qsize_t inode_get_rsv_space(struct inode *inode)
1621{
1622        qsize_t ret;
1623
1624        if (!inode->i_sb->dq_op->get_reserved_space)
1625                return 0;
1626        spin_lock(&inode->i_lock);
1627        ret = __inode_get_rsv_space(inode);
1628        spin_unlock(&inode->i_lock);
1629        return ret;
1630}
1631
1632/*
1633 * This functions updates i_blocks+i_bytes fields and quota information
1634 * (together with appropriate checks).
1635 *
1636 * NOTE: We absolutely rely on the fact that caller dirties the inode
1637 * (usually helpers in quotaops.h care about this) and holds a handle for
1638 * the current transaction so that dquot write and inode write go into the
1639 * same transaction.
1640 */
1641
1642/*
1643 * This operation can block, but only after everything is updated
1644 */
1645int __dquot_alloc_space(struct inode *inode, qsize_t number, int flags)
1646{
1647        int cnt, ret = 0, index;
1648        struct dquot_warn warn[MAXQUOTAS];
1649        int reserve = flags & DQUOT_SPACE_RESERVE;
1650        struct dquot **dquots;
1651
1652        if (!dquot_active(inode)) {
1653                if (reserve) {
1654                        spin_lock(&inode->i_lock);
1655                        *inode_reserved_space(inode) += number;
1656                        spin_unlock(&inode->i_lock);
1657                } else {
1658                        inode_add_bytes(inode, number);
1659                }
1660                goto out;
1661        }
1662
1663        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1664                warn[cnt].w_type = QUOTA_NL_NOWARN;
1665
1666        dquots = i_dquot(inode);
1667        index = srcu_read_lock(&dquot_srcu);
1668        spin_lock(&inode->i_lock);
1669        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1670                if (!dquots[cnt])
1671                        continue;
1672                if (reserve) {
1673                        ret = dquot_add_space(dquots[cnt], 0, number, flags,
1674                                              &warn[cnt]);
1675                } else {
1676                        ret = dquot_add_space(dquots[cnt], number, 0, flags,
1677                                              &warn[cnt]);
1678                }
1679                if (ret) {
1680                        /* Back out changes we already did */
1681                        for (cnt--; cnt >= 0; cnt--) {
1682                                if (!dquots[cnt])
1683                                        continue;
1684                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1685                                if (reserve)
1686                                        dquot_free_reserved_space(dquots[cnt],
1687                                                                  number);
1688                                else
1689                                        dquot_decr_space(dquots[cnt], number);
1690                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1691                        }
1692                        spin_unlock(&inode->i_lock);
1693                        goto out_flush_warn;
1694                }
1695        }
1696        if (reserve)
1697                *inode_reserved_space(inode) += number;
1698        else
1699                __inode_add_bytes(inode, number);
1700        spin_unlock(&inode->i_lock);
1701
1702        if (reserve)
1703                goto out_flush_warn;
1704        mark_all_dquot_dirty(dquots);
1705out_flush_warn:
1706        srcu_read_unlock(&dquot_srcu, index);
1707        flush_warnings(warn);
1708out:
1709        return ret;
1710}
1711EXPORT_SYMBOL(__dquot_alloc_space);
1712
1713/*
1714 * This operation can block, but only after everything is updated
1715 */
1716int dquot_alloc_inode(struct inode *inode)
1717{
1718        int cnt, ret = 0, index;
1719        struct dquot_warn warn[MAXQUOTAS];
1720        struct dquot * const *dquots;
1721
1722        if (!dquot_active(inode))
1723                return 0;
1724        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
1725                warn[cnt].w_type = QUOTA_NL_NOWARN;
1726
1727        dquots = i_dquot(inode);
1728        index = srcu_read_lock(&dquot_srcu);
1729        spin_lock(&inode->i_lock);
1730        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1731                if (!dquots[cnt])
1732                        continue;
1733                ret = dquot_add_inodes(dquots[cnt], 1, &warn[cnt]);
1734                if (ret) {
1735                        for (cnt--; cnt >= 0; cnt--) {
1736                                if (!dquots[cnt])
1737                                        continue;
1738                                /* Back out changes we already did */
1739                                spin_lock(&dquots[cnt]->dq_dqb_lock);
1740                                dquot_decr_inodes(dquots[cnt], 1);
1741                                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1742                        }
1743                        goto warn_put_all;
1744                }
1745        }
1746
1747warn_put_all:
1748        spin_unlock(&inode->i_lock);
1749        if (ret == 0)
1750                mark_all_dquot_dirty(dquots);
1751        srcu_read_unlock(&dquot_srcu, index);
1752        flush_warnings(warn);
1753        return ret;
1754}
1755EXPORT_SYMBOL(dquot_alloc_inode);
1756
1757/*
1758 * Convert in-memory reserved quotas to real consumed quotas
1759 */
1760int dquot_claim_space_nodirty(struct inode *inode, qsize_t number)
1761{
1762        struct dquot **dquots;
1763        int cnt, index;
1764
1765        if (!dquot_active(inode)) {
1766                spin_lock(&inode->i_lock);
1767                *inode_reserved_space(inode) -= number;
1768                __inode_add_bytes(inode, number);
1769                spin_unlock(&inode->i_lock);
1770                return 0;
1771        }
1772
1773        dquots = i_dquot(inode);
1774        index = srcu_read_lock(&dquot_srcu);
1775        spin_lock(&inode->i_lock);
1776        /* Claim reserved quotas to allocated quotas */
1777        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1778                if (dquots[cnt]) {
1779                        struct dquot *dquot = dquots[cnt];
1780
1781                        spin_lock(&dquot->dq_dqb_lock);
1782                        if (WARN_ON_ONCE(dquot->dq_dqb.dqb_rsvspace < number))
1783                                number = dquot->dq_dqb.dqb_rsvspace;
1784                        dquot->dq_dqb.dqb_curspace += number;
1785                        dquot->dq_dqb.dqb_rsvspace -= number;
1786                        spin_unlock(&dquot->dq_dqb_lock);
1787                }
1788        }
1789        /* Update inode bytes */
1790        *inode_reserved_space(inode) -= number;
1791        __inode_add_bytes(inode, number);
1792        spin_unlock(&inode->i_lock);
1793        mark_all_dquot_dirty(dquots);
1794        srcu_read_unlock(&dquot_srcu, index);
1795        return 0;
1796}
1797EXPORT_SYMBOL(dquot_claim_space_nodirty);
1798
1799/*
1800 * Convert allocated space back to in-memory reserved quotas
1801 */
1802void dquot_reclaim_space_nodirty(struct inode *inode, qsize_t number)
1803{
1804        struct dquot **dquots;
1805        int cnt, index;
1806
1807        if (!dquot_active(inode)) {
1808                spin_lock(&inode->i_lock);
1809                *inode_reserved_space(inode) += number;
1810                __inode_sub_bytes(inode, number);
1811                spin_unlock(&inode->i_lock);
1812                return;
1813        }
1814
1815        dquots = i_dquot(inode);
1816        index = srcu_read_lock(&dquot_srcu);
1817        spin_lock(&inode->i_lock);
1818        /* Claim reserved quotas to allocated quotas */
1819        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1820                if (dquots[cnt]) {
1821                        struct dquot *dquot = dquots[cnt];
1822
1823                        spin_lock(&dquot->dq_dqb_lock);
1824                        if (WARN_ON_ONCE(dquot->dq_dqb.dqb_curspace < number))
1825                                number = dquot->dq_dqb.dqb_curspace;
1826                        dquot->dq_dqb.dqb_rsvspace += number;
1827                        dquot->dq_dqb.dqb_curspace -= number;
1828                        spin_unlock(&dquot->dq_dqb_lock);
1829                }
1830        }
1831        /* Update inode bytes */
1832        *inode_reserved_space(inode) += number;
1833        __inode_sub_bytes(inode, number);
1834        spin_unlock(&inode->i_lock);
1835        mark_all_dquot_dirty(dquots);
1836        srcu_read_unlock(&dquot_srcu, index);
1837        return;
1838}
1839EXPORT_SYMBOL(dquot_reclaim_space_nodirty);
1840
1841/*
1842 * This operation can block, but only after everything is updated
1843 */
1844void __dquot_free_space(struct inode *inode, qsize_t number, int flags)
1845{
1846        unsigned int cnt;
1847        struct dquot_warn warn[MAXQUOTAS];
1848        struct dquot **dquots;
1849        int reserve = flags & DQUOT_SPACE_RESERVE, index;
1850
1851        if (!dquot_active(inode)) {
1852                if (reserve) {
1853                        spin_lock(&inode->i_lock);
1854                        *inode_reserved_space(inode) -= number;
1855                        spin_unlock(&inode->i_lock);
1856                } else {
1857                        inode_sub_bytes(inode, number);
1858                }
1859                return;
1860        }
1861
1862        dquots = i_dquot(inode);
1863        index = srcu_read_lock(&dquot_srcu);
1864        spin_lock(&inode->i_lock);
1865        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1866                int wtype;
1867
1868                warn[cnt].w_type = QUOTA_NL_NOWARN;
1869                if (!dquots[cnt])
1870                        continue;
1871                spin_lock(&dquots[cnt]->dq_dqb_lock);
1872                wtype = info_bdq_free(dquots[cnt], number);
1873                if (wtype != QUOTA_NL_NOWARN)
1874                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1875                if (reserve)
1876                        dquot_free_reserved_space(dquots[cnt], number);
1877                else
1878                        dquot_decr_space(dquots[cnt], number);
1879                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1880        }
1881        if (reserve)
1882                *inode_reserved_space(inode) -= number;
1883        else
1884                __inode_sub_bytes(inode, number);
1885        spin_unlock(&inode->i_lock);
1886
1887        if (reserve)
1888                goto out_unlock;
1889        mark_all_dquot_dirty(dquots);
1890out_unlock:
1891        srcu_read_unlock(&dquot_srcu, index);
1892        flush_warnings(warn);
1893}
1894EXPORT_SYMBOL(__dquot_free_space);
1895
1896/*
1897 * This operation can block, but only after everything is updated
1898 */
1899void dquot_free_inode(struct inode *inode)
1900{
1901        unsigned int cnt;
1902        struct dquot_warn warn[MAXQUOTAS];
1903        struct dquot * const *dquots;
1904        int index;
1905
1906        if (!dquot_active(inode))
1907                return;
1908
1909        dquots = i_dquot(inode);
1910        index = srcu_read_lock(&dquot_srcu);
1911        spin_lock(&inode->i_lock);
1912        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1913                int wtype;
1914
1915                warn[cnt].w_type = QUOTA_NL_NOWARN;
1916                if (!dquots[cnt])
1917                        continue;
1918                spin_lock(&dquots[cnt]->dq_dqb_lock);
1919                wtype = info_idq_free(dquots[cnt], 1);
1920                if (wtype != QUOTA_NL_NOWARN)
1921                        prepare_warning(&warn[cnt], dquots[cnt], wtype);
1922                dquot_decr_inodes(dquots[cnt], 1);
1923                spin_unlock(&dquots[cnt]->dq_dqb_lock);
1924        }
1925        spin_unlock(&inode->i_lock);
1926        mark_all_dquot_dirty(dquots);
1927        srcu_read_unlock(&dquot_srcu, index);
1928        flush_warnings(warn);
1929}
1930EXPORT_SYMBOL(dquot_free_inode);
1931
1932/*
1933 * Transfer the number of inode and blocks from one diskquota to an other.
1934 * On success, dquot references in transfer_to are consumed and references
1935 * to original dquots that need to be released are placed there. On failure,
1936 * references are kept untouched.
1937 *
1938 * This operation can block, but only after everything is updated
1939 * A transaction must be started when entering this function.
1940 *
1941 * We are holding reference on transfer_from & transfer_to, no need to
1942 * protect them by srcu_read_lock().
1943 */
1944int __dquot_transfer(struct inode *inode, struct dquot **transfer_to)
1945{
1946        qsize_t cur_space;
1947        qsize_t rsv_space = 0;
1948        qsize_t inode_usage = 1;
1949        struct dquot *transfer_from[MAXQUOTAS] = {};
1950        int cnt, ret = 0;
1951        char is_valid[MAXQUOTAS] = {};
1952        struct dquot_warn warn_to[MAXQUOTAS];
1953        struct dquot_warn warn_from_inodes[MAXQUOTAS];
1954        struct dquot_warn warn_from_space[MAXQUOTAS];
1955
1956        if (IS_NOQUOTA(inode))
1957                return 0;
1958
1959        if (inode->i_sb->dq_op->get_inode_usage) {
1960                ret = inode->i_sb->dq_op->get_inode_usage(inode, &inode_usage);
1961                if (ret)
1962                        return ret;
1963        }
1964
1965        /* Initialize the arrays */
1966        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1967                warn_to[cnt].w_type = QUOTA_NL_NOWARN;
1968                warn_from_inodes[cnt].w_type = QUOTA_NL_NOWARN;
1969                warn_from_space[cnt].w_type = QUOTA_NL_NOWARN;
1970        }
1971
1972        spin_lock(&dq_data_lock);
1973        spin_lock(&inode->i_lock);
1974        if (IS_NOQUOTA(inode)) {        /* File without quota accounting? */
1975                spin_unlock(&inode->i_lock);
1976                spin_unlock(&dq_data_lock);
1977                return 0;
1978        }
1979        cur_space = __inode_get_bytes(inode);
1980        rsv_space = __inode_get_rsv_space(inode);
1981        /*
1982         * Build the transfer_from list, check limits, and update usage in
1983         * the target structures.
1984         */
1985        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1986                /*
1987                 * Skip changes for same uid or gid or for turned off quota-type.
1988                 */
1989                if (!transfer_to[cnt])
1990                        continue;
1991                /* Avoid races with quotaoff() */
1992                if (!sb_has_quota_active(inode->i_sb, cnt))
1993                        continue;
1994                is_valid[cnt] = 1;
1995                transfer_from[cnt] = i_dquot(inode)[cnt];
1996                ret = dquot_add_inodes(transfer_to[cnt], inode_usage,
1997                                       &warn_to[cnt]);
1998                if (ret)
1999                        goto over_quota;
2000                ret = dquot_add_space(transfer_to[cnt], cur_space, rsv_space,
2001                                      DQUOT_SPACE_WARN, &warn_to[cnt]);
2002                if (ret) {
2003                        spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2004                        dquot_decr_inodes(transfer_to[cnt], inode_usage);
2005                        spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2006                        goto over_quota;
2007                }
2008        }
2009
2010        /* Decrease usage for source structures and update quota pointers */
2011        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2012                if (!is_valid[cnt])
2013                        continue;
2014                /* Due to IO error we might not have transfer_from[] structure */
2015                if (transfer_from[cnt]) {
2016                        int wtype;
2017
2018                        spin_lock(&transfer_from[cnt]->dq_dqb_lock);
2019                        wtype = info_idq_free(transfer_from[cnt], inode_usage);
2020                        if (wtype != QUOTA_NL_NOWARN)
2021                                prepare_warning(&warn_from_inodes[cnt],
2022                                                transfer_from[cnt], wtype);
2023                        wtype = info_bdq_free(transfer_from[cnt],
2024                                              cur_space + rsv_space);
2025                        if (wtype != QUOTA_NL_NOWARN)
2026                                prepare_warning(&warn_from_space[cnt],
2027                                                transfer_from[cnt], wtype);
2028                        dquot_decr_inodes(transfer_from[cnt], inode_usage);
2029                        dquot_decr_space(transfer_from[cnt], cur_space);
2030                        dquot_free_reserved_space(transfer_from[cnt],
2031                                                  rsv_space);
2032                        spin_unlock(&transfer_from[cnt]->dq_dqb_lock);
2033                }
2034                i_dquot(inode)[cnt] = transfer_to[cnt];
2035        }
2036        spin_unlock(&inode->i_lock);
2037        spin_unlock(&dq_data_lock);
2038
2039        mark_all_dquot_dirty(transfer_from);
2040        mark_all_dquot_dirty(transfer_to);
2041        flush_warnings(warn_to);
2042        flush_warnings(warn_from_inodes);
2043        flush_warnings(warn_from_space);
2044        /* Pass back references to put */
2045        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2046                if (is_valid[cnt])
2047                        transfer_to[cnt] = transfer_from[cnt];
2048        return 0;
2049over_quota:
2050        /* Back out changes we already did */
2051        for (cnt--; cnt >= 0; cnt--) {
2052                if (!is_valid[cnt])
2053                        continue;
2054                spin_lock(&transfer_to[cnt]->dq_dqb_lock);
2055                dquot_decr_inodes(transfer_to[cnt], inode_usage);
2056                dquot_decr_space(transfer_to[cnt], cur_space);
2057                dquot_free_reserved_space(transfer_to[cnt], rsv_space);
2058                spin_unlock(&transfer_to[cnt]->dq_dqb_lock);
2059        }
2060        spin_unlock(&inode->i_lock);
2061        spin_unlock(&dq_data_lock);
2062        flush_warnings(warn_to);
2063        return ret;
2064}
2065EXPORT_SYMBOL(__dquot_transfer);
2066
2067/* Wrapper for transferring ownership of an inode for uid/gid only
2068 * Called from FSXXX_setattr()
2069 */
2070int dquot_transfer(struct inode *inode, struct iattr *iattr)
2071{
2072        struct dquot *transfer_to[MAXQUOTAS] = {};
2073        struct dquot *dquot;
2074        struct super_block *sb = inode->i_sb;
2075        int ret;
2076
2077        if (!dquot_active(inode))
2078                return 0;
2079
2080        if (iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)){
2081                dquot = dqget(sb, make_kqid_uid(iattr->ia_uid));
2082                if (IS_ERR(dquot)) {
2083                        if (PTR_ERR(dquot) != -ESRCH) {
2084                                ret = PTR_ERR(dquot);
2085                                goto out_put;
2086                        }
2087                        dquot = NULL;
2088                }
2089                transfer_to[USRQUOTA] = dquot;
2090        }
2091        if (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid)){
2092                dquot = dqget(sb, make_kqid_gid(iattr->ia_gid));
2093                if (IS_ERR(dquot)) {
2094                        if (PTR_ERR(dquot) != -ESRCH) {
2095                                ret = PTR_ERR(dquot);
2096                                goto out_put;
2097                        }
2098                        dquot = NULL;
2099                }
2100                transfer_to[GRPQUOTA] = dquot;
2101        }
2102        ret = __dquot_transfer(inode, transfer_to);
2103out_put:
2104        dqput_all(transfer_to);
2105        return ret;
2106}
2107EXPORT_SYMBOL(dquot_transfer);
2108
2109/*
2110 * Write info of quota file to disk
2111 */
2112int dquot_commit_info(struct super_block *sb, int type)
2113{
2114        struct quota_info *dqopt = sb_dqopt(sb);
2115
2116        return dqopt->ops[type]->write_file_info(sb, type);
2117}
2118EXPORT_SYMBOL(dquot_commit_info);
2119
2120int dquot_get_next_id(struct super_block *sb, struct kqid *qid)
2121{
2122        struct quota_info *dqopt = sb_dqopt(sb);
2123
2124        if (!sb_has_quota_active(sb, qid->type))
2125                return -ESRCH;
2126        if (!dqopt->ops[qid->type]->get_next_id)
2127                return -ENOSYS;
2128        return dqopt->ops[qid->type]->get_next_id(sb, qid);
2129}
2130EXPORT_SYMBOL(dquot_get_next_id);
2131
2132/*
2133 * Definitions of diskquota operations.
2134 */
2135const struct dquot_operations dquot_operations = {
2136        .write_dquot    = dquot_commit,
2137        .acquire_dquot  = dquot_acquire,
2138        .release_dquot  = dquot_release,
2139        .mark_dirty     = dquot_mark_dquot_dirty,
2140        .write_info     = dquot_commit_info,
2141        .alloc_dquot    = dquot_alloc,
2142        .destroy_dquot  = dquot_destroy,
2143        .get_next_id    = dquot_get_next_id,
2144};
2145EXPORT_SYMBOL(dquot_operations);
2146
2147/*
2148 * Generic helper for ->open on filesystems supporting disk quotas.
2149 */
2150int dquot_file_open(struct inode *inode, struct file *file)
2151{
2152        int error;
2153
2154        error = generic_file_open(inode, file);
2155        if (!error && (file->f_mode & FMODE_WRITE))
2156                error = dquot_initialize(inode);
2157        return error;
2158}
2159EXPORT_SYMBOL(dquot_file_open);
2160
2161/*
2162 * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
2163 */
2164int dquot_disable(struct super_block *sb, int type, unsigned int flags)
2165{
2166        int cnt, ret = 0;
2167        struct quota_info *dqopt = sb_dqopt(sb);
2168        struct inode *toputinode[MAXQUOTAS];
2169
2170        /* s_umount should be held in exclusive mode */
2171        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2172                up_read(&sb->s_umount);
2173
2174        /* Cannot turn off usage accounting without turning off limits, or
2175         * suspend quotas and simultaneously turn quotas off. */
2176        if ((flags & DQUOT_USAGE_ENABLED && !(flags & DQUOT_LIMITS_ENABLED))
2177            || (flags & DQUOT_SUSPENDED && flags & (DQUOT_LIMITS_ENABLED |
2178            DQUOT_USAGE_ENABLED)))
2179                return -EINVAL;
2180
2181        /*
2182         * Skip everything if there's nothing to do. We have to do this because
2183         * sometimes we are called when fill_super() failed and calling
2184         * sync_fs() in such cases does no good.
2185         */
2186        if (!sb_any_quota_loaded(sb))
2187                return 0;
2188
2189        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2190                toputinode[cnt] = NULL;
2191                if (type != -1 && cnt != type)
2192                        continue;
2193                if (!sb_has_quota_loaded(sb, cnt))
2194                        continue;
2195
2196                if (flags & DQUOT_SUSPENDED) {
2197                        spin_lock(&dq_state_lock);
2198                        dqopt->flags |=
2199                                dquot_state_flag(DQUOT_SUSPENDED, cnt);
2200                        spin_unlock(&dq_state_lock);
2201                } else {
2202                        spin_lock(&dq_state_lock);
2203                        dqopt->flags &= ~dquot_state_flag(flags, cnt);
2204                        /* Turning off suspended quotas? */
2205                        if (!sb_has_quota_loaded(sb, cnt) &&
2206                            sb_has_quota_suspended(sb, cnt)) {
2207                                dqopt->flags &= ~dquot_state_flag(
2208                                                        DQUOT_SUSPENDED, cnt);
2209                                spin_unlock(&dq_state_lock);
2210                                iput(dqopt->files[cnt]);
2211                                dqopt->files[cnt] = NULL;
2212                                continue;
2213                        }
2214                        spin_unlock(&dq_state_lock);
2215                }
2216
2217                /* We still have to keep quota loaded? */
2218                if (sb_has_quota_loaded(sb, cnt) && !(flags & DQUOT_SUSPENDED))
2219                        continue;
2220
2221                /* Note: these are blocking operations */
2222                drop_dquot_ref(sb, cnt);
2223                invalidate_dquots(sb, cnt);
2224                /*
2225                 * Now all dquots should be invalidated, all writes done so we
2226                 * should be only users of the info. No locks needed.
2227                 */
2228                if (info_dirty(&dqopt->info[cnt]))
2229                        sb->dq_op->write_info(sb, cnt);
2230                if (dqopt->ops[cnt]->free_file_info)
2231                        dqopt->ops[cnt]->free_file_info(sb, cnt);
2232                put_quota_format(dqopt->info[cnt].dqi_format);
2233
2234                toputinode[cnt] = dqopt->files[cnt];
2235                if (!sb_has_quota_loaded(sb, cnt))
2236                        dqopt->files[cnt] = NULL;
2237                dqopt->info[cnt].dqi_flags = 0;
2238                dqopt->info[cnt].dqi_igrace = 0;
2239                dqopt->info[cnt].dqi_bgrace = 0;
2240                dqopt->ops[cnt] = NULL;
2241        }
2242
2243        /* Skip syncing and setting flags if quota files are hidden */
2244        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE)
2245                goto put_inodes;
2246
2247        /* Sync the superblock so that buffers with quota data are written to
2248         * disk (and so userspace sees correct data afterwards). */
2249        if (sb->s_op->sync_fs)
2250                sb->s_op->sync_fs(sb, 1);
2251        sync_blockdev(sb->s_bdev);
2252        /* Now the quota files are just ordinary files and we can set the
2253         * inode flags back. Moreover we discard the pagecache so that
2254         * userspace sees the writes we did bypassing the pagecache. We
2255         * must also discard the blockdev buffers so that we see the
2256         * changes done by userspace on the next quotaon() */
2257        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2258                /* This can happen when suspending quotas on remount-ro... */
2259                if (toputinode[cnt] && !sb_has_quota_loaded(sb, cnt)) {
2260                        inode_lock(toputinode[cnt]);
2261                        toputinode[cnt]->i_flags &= ~S_NOQUOTA;
2262                        truncate_inode_pages(&toputinode[cnt]->i_data, 0);
2263                        inode_unlock(toputinode[cnt]);
2264                        mark_inode_dirty_sync(toputinode[cnt]);
2265                }
2266        if (sb->s_bdev)
2267                invalidate_bdev(sb->s_bdev);
2268put_inodes:
2269        for (cnt = 0; cnt < MAXQUOTAS; cnt++)
2270                if (toputinode[cnt]) {
2271                        /* On remount RO, we keep the inode pointer so that we
2272                         * can reenable quota on the subsequent remount RW. We
2273                         * have to check 'flags' variable and not use sb_has_
2274                         * function because another quotaon / quotaoff could
2275                         * change global state before we got here. We refuse
2276                         * to suspend quotas when there is pending delete on
2277                         * the quota file... */
2278                        if (!(flags & DQUOT_SUSPENDED))
2279                                iput(toputinode[cnt]);
2280                        else if (!toputinode[cnt]->i_nlink)
2281                                ret = -EBUSY;
2282                }
2283        return ret;
2284}
2285EXPORT_SYMBOL(dquot_disable);
2286
2287int dquot_quota_off(struct super_block *sb, int type)
2288{
2289        return dquot_disable(sb, type,
2290                             DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2291}
2292EXPORT_SYMBOL(dquot_quota_off);
2293
2294/*
2295 *      Turn quotas on on a device
2296 */
2297
2298/*
2299 * Helper function to turn quotas on when we already have the inode of
2300 * quota file and no quota information is loaded.
2301 */
2302static int vfs_load_quota_inode(struct inode *inode, int type, int format_id,
2303        unsigned int flags)
2304{
2305        struct quota_format_type *fmt = find_quota_format(format_id);
2306        struct super_block *sb = inode->i_sb;
2307        struct quota_info *dqopt = sb_dqopt(sb);
2308        int error;
2309
2310        if (!fmt)
2311                return -ESRCH;
2312        if (!S_ISREG(inode->i_mode)) {
2313                error = -EACCES;
2314                goto out_fmt;
2315        }
2316        if (IS_RDONLY(inode)) {
2317                error = -EROFS;
2318                goto out_fmt;
2319        }
2320        if (!sb->s_op->quota_write || !sb->s_op->quota_read ||
2321            (type == PRJQUOTA && sb->dq_op->get_projid == NULL)) {
2322                error = -EINVAL;
2323                goto out_fmt;
2324        }
2325        /* Filesystems outside of init_user_ns not yet supported */
2326        if (sb->s_user_ns != &init_user_ns) {
2327                error = -EINVAL;
2328                goto out_fmt;
2329        }
2330        /* Usage always has to be set... */
2331        if (!(flags & DQUOT_USAGE_ENABLED)) {
2332                error = -EINVAL;
2333                goto out_fmt;
2334        }
2335        if (sb_has_quota_loaded(sb, type)) {
2336                error = -EBUSY;
2337                goto out_fmt;
2338        }
2339
2340        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2341                /* As we bypass the pagecache we must now flush all the
2342                 * dirty data and invalidate caches so that kernel sees
2343                 * changes from userspace. It is not enough to just flush
2344                 * the quota file since if blocksize < pagesize, invalidation
2345                 * of the cache could fail because of other unrelated dirty
2346                 * data */
2347                sync_filesystem(sb);
2348                invalidate_bdev(sb->s_bdev);
2349        }
2350
2351        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE)) {
2352                /* We don't want quota and atime on quota files (deadlocks
2353                 * possible) Also nobody should write to the file - we use
2354                 * special IO operations which ignore the immutable bit. */
2355                inode_lock(inode);
2356                inode->i_flags |= S_NOQUOTA;
2357                inode_unlock(inode);
2358                /*
2359                 * When S_NOQUOTA is set, remove dquot references as no more
2360                 * references can be added
2361                 */
2362                __dquot_drop(inode);
2363        }
2364
2365        error = -EIO;
2366        dqopt->files[type] = igrab(inode);
2367        if (!dqopt->files[type])
2368                goto out_file_flags;
2369        error = -EINVAL;
2370        if (!fmt->qf_ops->check_quota_file(sb, type))
2371                goto out_file_init;
2372
2373        dqopt->ops[type] = fmt->qf_ops;
2374        dqopt->info[type].dqi_format = fmt;
2375        dqopt->info[type].dqi_fmt_id = format_id;
2376        INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
2377        error = dqopt->ops[type]->read_file_info(sb, type);
2378        if (error < 0)
2379                goto out_file_init;
2380        if (dqopt->flags & DQUOT_QUOTA_SYS_FILE) {
2381                spin_lock(&dq_data_lock);
2382                dqopt->info[type].dqi_flags |= DQF_SYS_FILE;
2383                spin_unlock(&dq_data_lock);
2384        }
2385        spin_lock(&dq_state_lock);
2386        dqopt->flags |= dquot_state_flag(flags, type);
2387        spin_unlock(&dq_state_lock);
2388
2389        error = add_dquot_ref(sb, type);
2390        if (error)
2391                dquot_disable(sb, type, flags);
2392
2393        return error;
2394out_file_init:
2395        dqopt->files[type] = NULL;
2396        iput(inode);
2397out_file_flags:
2398        inode_lock(inode);
2399        inode->i_flags &= ~S_NOQUOTA;
2400        inode_unlock(inode);
2401out_fmt:
2402        put_quota_format(fmt);
2403
2404        return error;
2405}
2406
2407/* Reenable quotas on remount RW */
2408int dquot_resume(struct super_block *sb, int type)
2409{
2410        struct quota_info *dqopt = sb_dqopt(sb);
2411        struct inode *inode;
2412        int ret = 0, cnt;
2413        unsigned int flags;
2414
2415        /* s_umount should be held in exclusive mode */
2416        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2417                up_read(&sb->s_umount);
2418
2419        for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
2420                if (type != -1 && cnt != type)
2421                        continue;
2422                if (!sb_has_quota_suspended(sb, cnt))
2423                        continue;
2424
2425                inode = dqopt->files[cnt];
2426                dqopt->files[cnt] = NULL;
2427                spin_lock(&dq_state_lock);
2428                flags = dqopt->flags & dquot_state_flag(DQUOT_USAGE_ENABLED |
2429                                                        DQUOT_LIMITS_ENABLED,
2430                                                        cnt);
2431                dqopt->flags &= ~dquot_state_flag(DQUOT_STATE_FLAGS, cnt);
2432                spin_unlock(&dq_state_lock);
2433
2434                flags = dquot_generic_flag(flags, cnt);
2435                ret = vfs_load_quota_inode(inode, cnt,
2436                                dqopt->info[cnt].dqi_fmt_id, flags);
2437                iput(inode);
2438        }
2439
2440        return ret;
2441}
2442EXPORT_SYMBOL(dquot_resume);
2443
2444int dquot_quota_on(struct super_block *sb, int type, int format_id,
2445                   const struct path *path)
2446{
2447        int error = security_quota_on(path->dentry);
2448        if (error)
2449                return error;
2450        /* Quota file not on the same filesystem? */
2451        if (path->dentry->d_sb != sb)
2452                error = -EXDEV;
2453        else
2454                error = vfs_load_quota_inode(d_inode(path->dentry), type,
2455                                             format_id, DQUOT_USAGE_ENABLED |
2456                                             DQUOT_LIMITS_ENABLED);
2457        return error;
2458}
2459EXPORT_SYMBOL(dquot_quota_on);
2460
2461/*
2462 * More powerful function for turning on quotas allowing setting
2463 * of individual quota flags
2464 */
2465int dquot_enable(struct inode *inode, int type, int format_id,
2466                 unsigned int flags)
2467{
2468        struct super_block *sb = inode->i_sb;
2469
2470        /* Just unsuspend quotas? */
2471        BUG_ON(flags & DQUOT_SUSPENDED);
2472        /* s_umount should be held in exclusive mode */
2473        if (WARN_ON_ONCE(down_read_trylock(&sb->s_umount)))
2474                up_read(&sb->s_umount);
2475
2476        if (!flags)
2477                return 0;
2478        /* Just updating flags needed? */
2479        if (sb_has_quota_loaded(sb, type)) {
2480                if (flags & DQUOT_USAGE_ENABLED &&
2481                    sb_has_quota_usage_enabled(sb, type))
2482                        return -EBUSY;
2483                if (flags & DQUOT_LIMITS_ENABLED &&
2484                    sb_has_quota_limits_enabled(sb, type))
2485                        return -EBUSY;
2486                spin_lock(&dq_state_lock);
2487                sb_dqopt(sb)->flags |= dquot_state_flag(flags, type);
2488                spin_unlock(&dq_state_lock);
2489                return 0;
2490        }
2491
2492        return vfs_load_quota_inode(inode, type, format_id, flags);
2493}
2494EXPORT_SYMBOL(dquot_enable);
2495
2496/*
2497 * This function is used when filesystem needs to initialize quotas
2498 * during mount time.
2499 */
2500int dquot_quota_on_mount(struct super_block *sb, char *qf_name,
2501                int format_id, int type)
2502{
2503        struct dentry *dentry;
2504        int error;
2505
2506        dentry = lookup_one_len_unlocked(qf_name, sb->s_root, strlen(qf_name));
2507        if (IS_ERR(dentry))
2508                return PTR_ERR(dentry);
2509
2510        if (d_really_is_negative(dentry)) {
2511                error = -ENOENT;
2512                goto out;
2513        }
2514
2515        error = security_quota_on(dentry);
2516        if (!error)
2517                error = vfs_load_quota_inode(d_inode(dentry), type, format_id,
2518                                DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
2519
2520out:
2521        dput(dentry);
2522        return error;
2523}
2524EXPORT_SYMBOL(dquot_quota_on_mount);
2525
2526static int dquot_quota_enable(struct super_block *sb, unsigned int flags)
2527{
2528        int ret;
2529        int type;
2530        struct quota_info *dqopt = sb_dqopt(sb);
2531
2532        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2533                return -ENOSYS;
2534        /* Accounting cannot be turned on while fs is mounted */
2535        flags &= ~(FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT);
2536        if (!flags)
2537                return -EINVAL;
2538        for (type = 0; type < MAXQUOTAS; type++) {
2539                if (!(flags & qtype_enforce_flag(type)))
2540                        continue;
2541                /* Can't enforce without accounting */
2542                if (!sb_has_quota_usage_enabled(sb, type))
2543                        return -EINVAL;
2544                ret = dquot_enable(dqopt->files[type], type,
2545                                   dqopt->info[type].dqi_fmt_id,
2546                                   DQUOT_LIMITS_ENABLED);
2547                if (ret < 0)
2548                        goto out_err;
2549        }
2550        return 0;
2551out_err:
2552        /* Backout enforcement enablement we already did */
2553        for (type--; type >= 0; type--)  {
2554                if (flags & qtype_enforce_flag(type))
2555                        dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2556        }
2557        /* Error code translation for better compatibility with XFS */
2558        if (ret == -EBUSY)
2559                ret = -EEXIST;
2560        return ret;
2561}
2562
2563static int dquot_quota_disable(struct super_block *sb, unsigned int flags)
2564{
2565        int ret;
2566        int type;
2567        struct quota_info *dqopt = sb_dqopt(sb);
2568
2569        if (!(dqopt->flags & DQUOT_QUOTA_SYS_FILE))
2570                return -ENOSYS;
2571        /*
2572         * We don't support turning off accounting via quotactl. In principle
2573         * quota infrastructure can do this but filesystems don't expect
2574         * userspace to be able to do it.
2575         */
2576        if (flags &
2577                  (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT | FS_QUOTA_PDQ_ACCT))
2578                return -EOPNOTSUPP;
2579
2580        /* Filter out limits not enabled */
2581        for (type = 0; type < MAXQUOTAS; type++)
2582                if (!sb_has_quota_limits_enabled(sb, type))
2583                        flags &= ~qtype_enforce_flag(type);
2584        /* Nothing left? */
2585        if (!flags)
2586                return -EEXIST;
2587        for (type = 0; type < MAXQUOTAS; type++) {
2588                if (flags & qtype_enforce_flag(type)) {
2589                        ret = dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
2590                        if (ret < 0)
2591                                goto out_err;
2592                }
2593        }
2594        return 0;
2595out_err:
2596        /* Backout enforcement disabling we already did */
2597        for (type--; type >= 0; type--)  {
2598                if (flags & qtype_enforce_flag(type))
2599                        dquot_enable(dqopt->files[type], type,
2600                                     dqopt->info[type].dqi_fmt_id,
2601                                     DQUOT_LIMITS_ENABLED);
2602        }
2603        return ret;
2604}
2605
2606/* Generic routine for getting common part of quota structure */
2607static void do_get_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2608{
2609        struct mem_dqblk *dm = &dquot->dq_dqb;
2610
2611        memset(di, 0, sizeof(*di));
2612        spin_lock(&dquot->dq_dqb_lock);
2613        di->d_spc_hardlimit = dm->dqb_bhardlimit;
2614        di->d_spc_softlimit = dm->dqb_bsoftlimit;
2615        di->d_ino_hardlimit = dm->dqb_ihardlimit;
2616        di->d_ino_softlimit = dm->dqb_isoftlimit;
2617        di->d_space = dm->dqb_curspace + dm->dqb_rsvspace;
2618        di->d_ino_count = dm->dqb_curinodes;
2619        di->d_spc_timer = dm->dqb_btime;
2620        di->d_ino_timer = dm->dqb_itime;
2621        spin_unlock(&dquot->dq_dqb_lock);
2622}
2623
2624int dquot_get_dqblk(struct super_block *sb, struct kqid qid,
2625                    struct qc_dqblk *di)
2626{
2627        struct dquot *dquot;
2628
2629        dquot = dqget(sb, qid);
2630        if (IS_ERR(dquot))
2631                return PTR_ERR(dquot);
2632        do_get_dqblk(dquot, di);
2633        dqput(dquot);
2634
2635        return 0;
2636}
2637EXPORT_SYMBOL(dquot_get_dqblk);
2638
2639int dquot_get_next_dqblk(struct super_block *sb, struct kqid *qid,
2640                         struct qc_dqblk *di)
2641{
2642        struct dquot *dquot;
2643        int err;
2644
2645        if (!sb->dq_op->get_next_id)
2646                return -ENOSYS;
2647        err = sb->dq_op->get_next_id(sb, qid);
2648        if (err < 0)
2649                return err;
2650        dquot = dqget(sb, *qid);
2651        if (IS_ERR(dquot))
2652                return PTR_ERR(dquot);
2653        do_get_dqblk(dquot, di);
2654        dqput(dquot);
2655
2656        return 0;
2657}
2658EXPORT_SYMBOL(dquot_get_next_dqblk);
2659
2660#define VFS_QC_MASK \
2661        (QC_SPACE | QC_SPC_SOFT | QC_SPC_HARD | \
2662         QC_INO_COUNT | QC_INO_SOFT | QC_INO_HARD | \
2663         QC_SPC_TIMER | QC_INO_TIMER)
2664
2665/* Generic routine for setting common part of quota structure */
2666static int do_set_dqblk(struct dquot *dquot, struct qc_dqblk *di)
2667{
2668        struct mem_dqblk *dm = &dquot->dq_dqb;
2669        int check_blim = 0, check_ilim = 0;
2670        struct mem_dqinfo *dqi = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_id.type];
2671
2672        if (di->d_fieldmask & ~VFS_QC_MASK)
2673                return -EINVAL;
2674
2675        if (((di->d_fieldmask & QC_SPC_SOFT) &&
2676             di->d_spc_softlimit > dqi->dqi_max_spc_limit) ||
2677            ((di->d_fieldmask & QC_SPC_HARD) &&
2678             di->d_spc_hardlimit > dqi->dqi_max_spc_limit) ||
2679            ((di->d_fieldmask & QC_INO_SOFT) &&
2680             (di->d_ino_softlimit > dqi->dqi_max_ino_limit)) ||
2681            ((di->d_fieldmask & QC_INO_HARD) &&
2682             (di->d_ino_hardlimit > dqi->dqi_max_ino_limit)))
2683                return -ERANGE;
2684
2685        spin_lock(&dquot->dq_dqb_lock);
2686        if (di->d_fieldmask & QC_SPACE) {
2687                dm->dqb_curspace = di->d_space - dm->dqb_rsvspace;
2688                check_blim = 1;
2689                set_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
2690        }
2691
2692        if (di->d_fieldmask & QC_SPC_SOFT)
2693                dm->dqb_bsoftlimit = di->d_spc_softlimit;
2694        if (di->d_fieldmask & QC_SPC_HARD)
2695                dm->dqb_bhardlimit = di->d_spc_hardlimit;
2696        if (di->d_fieldmask & (QC_SPC_SOFT | QC_SPC_HARD)) {
2697                check_blim = 1;
2698                set_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
2699        }
2700
2701        if (di->d_fieldmask & QC_INO_COUNT) {
2702                dm->dqb_curinodes = di->d_ino_count;
2703                check_ilim = 1;
2704                set_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
2705        }
2706
2707        if (di->d_fieldmask & QC_INO_SOFT)
2708                dm->dqb_isoftlimit = di->d_ino_softlimit;
2709        if (di->d_fieldmask & QC_INO_HARD)
2710                dm->dqb_ihardlimit = di->d_ino_hardlimit;
2711        if (di->d_fieldmask & (QC_INO_SOFT | QC_INO_HARD)) {
2712                check_ilim = 1;
2713                set_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
2714        }
2715
2716        if (di->d_fieldmask & QC_SPC_TIMER) {
2717                dm->dqb_btime = di->d_spc_timer;
2718                check_blim = 1;
2719                set_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
2720        }
2721
2722        if (di->d_fieldmask & QC_INO_TIMER) {
2723                dm->dqb_itime = di->d_ino_timer;
2724                check_ilim = 1;
2725                set_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
2726        }
2727
2728        if (check_blim) {
2729                if (!dm->dqb_bsoftlimit ||
2730                    dm->dqb_curspace + dm->dqb_rsvspace <= dm->dqb_bsoftlimit) {
2731                        dm->dqb_btime = 0;
2732                        clear_bit(DQ_BLKS_B, &dquot->dq_flags);
2733                } else if (!(di->d_fieldmask & QC_SPC_TIMER))
2734                        /* Set grace only if user hasn't provided his own... */
2735                        dm->dqb_btime = ktime_get_real_seconds() + dqi->dqi_bgrace;
2736        }
2737        if (check_ilim) {
2738                if (!dm->dqb_isoftlimit ||
2739                    dm->dqb_curinodes <= dm->dqb_isoftlimit) {
2740                        dm->dqb_itime = 0;
2741                        clear_bit(DQ_INODES_B, &dquot->dq_flags);
2742                } else if (!(di->d_fieldmask & QC_INO_TIMER))
2743                        /* Set grace only if user hasn't provided his own... */
2744                        dm->dqb_itime = ktime_get_real_seconds() + dqi->dqi_igrace;
2745        }
2746        if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit ||
2747            dm->dqb_isoftlimit)
2748                clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2749        else
2750                set_bit(DQ_FAKE_B, &dquot->dq_flags);
2751        spin_unlock(&dquot->dq_dqb_lock);
2752        mark_dquot_dirty(dquot);
2753
2754        return 0;
2755}
2756
2757int dquot_set_dqblk(struct super_block *sb, struct kqid qid,
2758                  struct qc_dqblk *di)
2759{
2760        struct dquot *dquot;
2761        int rc;
2762
2763        dquot = dqget(sb, qid);
2764        if (IS_ERR(dquot)) {
2765                rc = PTR_ERR(dquot);
2766                goto out;
2767        }
2768        rc = do_set_dqblk(dquot, di);
2769        dqput(dquot);
2770out:
2771        return rc;
2772}
2773EXPORT_SYMBOL(dquot_set_dqblk);
2774
2775/* Generic routine for getting common part of quota file information */
2776int dquot_get_state(struct super_block *sb, struct qc_state *state)
2777{
2778        struct mem_dqinfo *mi;
2779        struct qc_type_state *tstate;
2780        struct quota_info *dqopt = sb_dqopt(sb);
2781        int type;
2782
2783        memset(state, 0, sizeof(*state));
2784        for (type = 0; type < MAXQUOTAS; type++) {
2785                if (!sb_has_quota_active(sb, type))
2786                        continue;
2787                tstate = state->s_state + type;
2788                mi = sb_dqopt(sb)->info + type;
2789                tstate->flags = QCI_ACCT_ENABLED;
2790                spin_lock(&dq_data_lock);
2791                if (mi->dqi_flags & DQF_SYS_FILE)
2792                        tstate->flags |= QCI_SYSFILE;
2793                if (mi->dqi_flags & DQF_ROOT_SQUASH)
2794                        tstate->flags |= QCI_ROOT_SQUASH;
2795                if (sb_has_quota_limits_enabled(sb, type))
2796                        tstate->flags |= QCI_LIMITS_ENFORCED;
2797                tstate->spc_timelimit = mi->dqi_bgrace;
2798                tstate->ino_timelimit = mi->dqi_igrace;
2799                tstate->ino = dqopt->files[type]->i_ino;
2800                tstate->blocks = dqopt->files[type]->i_blocks;
2801                tstate->nextents = 1;   /* We don't know... */
2802                spin_unlock(&dq_data_lock);
2803        }
2804        return 0;
2805}
2806EXPORT_SYMBOL(dquot_get_state);
2807
2808/* Generic routine for setting common part of quota file information */
2809int dquot_set_dqinfo(struct super_block *sb, int type, struct qc_info *ii)
2810{
2811        struct mem_dqinfo *mi;
2812        int err = 0;
2813
2814        if ((ii->i_fieldmask & QC_WARNS_MASK) ||
2815            (ii->i_fieldmask & QC_RT_SPC_TIMER))
2816                return -EINVAL;
2817        if (!sb_has_quota_active(sb, type))
2818                return -ESRCH;
2819        mi = sb_dqopt(sb)->info + type;
2820        if (ii->i_fieldmask & QC_FLAGS) {
2821                if ((ii->i_flags & QCI_ROOT_SQUASH &&
2822                     mi->dqi_format->qf_fmt_id != QFMT_VFS_OLD))
2823                        return -EINVAL;
2824        }
2825        spin_lock(&dq_data_lock);
2826        if (ii->i_fieldmask & QC_SPC_TIMER)
2827                mi->dqi_bgrace = ii->i_spc_timelimit;
2828        if (ii->i_fieldmask & QC_INO_TIMER)
2829                mi->dqi_igrace = ii->i_ino_timelimit;
2830        if (ii->i_fieldmask & QC_FLAGS) {
2831                if (ii->i_flags & QCI_ROOT_SQUASH)
2832                        mi->dqi_flags |= DQF_ROOT_SQUASH;
2833                else
2834                        mi->dqi_flags &= ~DQF_ROOT_SQUASH;
2835        }
2836        spin_unlock(&dq_data_lock);
2837        mark_info_dirty(sb, type);
2838        /* Force write to disk */
2839        sb->dq_op->write_info(sb, type);
2840        return err;
2841}
2842EXPORT_SYMBOL(dquot_set_dqinfo);
2843
2844const struct quotactl_ops dquot_quotactl_sysfile_ops = {
2845        .quota_enable   = dquot_quota_enable,
2846        .quota_disable  = dquot_quota_disable,
2847        .quota_sync     = dquot_quota_sync,
2848        .get_state      = dquot_get_state,
2849        .set_info       = dquot_set_dqinfo,
2850        .get_dqblk      = dquot_get_dqblk,
2851        .get_nextdqblk  = dquot_get_next_dqblk,
2852        .set_dqblk      = dquot_set_dqblk
2853};
2854EXPORT_SYMBOL(dquot_quotactl_sysfile_ops);
2855
2856static int do_proc_dqstats(struct ctl_table *table, int write,
2857                     void __user *buffer, size_t *lenp, loff_t *ppos)
2858{
2859        unsigned int type = (int *)table->data - dqstats.stat;
2860
2861        /* Update global table */
2862        dqstats.stat[type] =
2863                        percpu_counter_sum_positive(&dqstats.counter[type]);
2864        return proc_dointvec(table, write, buffer, lenp, ppos);
2865}
2866
2867static struct ctl_table fs_dqstats_table[] = {
2868        {
2869                .procname       = "lookups",
2870                .data           = &dqstats.stat[DQST_LOOKUPS],
2871                .maxlen         = sizeof(int),
2872                .mode           = 0444,
2873                .proc_handler   = do_proc_dqstats,
2874        },
2875        {
2876                .procname       = "drops",
2877                .data           = &dqstats.stat[DQST_DROPS],
2878                .maxlen         = sizeof(int),
2879                .mode           = 0444,
2880                .proc_handler   = do_proc_dqstats,
2881        },
2882        {
2883                .procname       = "reads",
2884                .data           = &dqstats.stat[DQST_READS],
2885                .maxlen         = sizeof(int),
2886                .mode           = 0444,
2887                .proc_handler   = do_proc_dqstats,
2888        },
2889        {
2890                .procname       = "writes",
2891                .data           = &dqstats.stat[DQST_WRITES],
2892                .maxlen         = sizeof(int),
2893                .mode           = 0444,
2894                .proc_handler   = do_proc_dqstats,
2895        },
2896        {
2897                .procname       = "cache_hits",
2898                .data           = &dqstats.stat[DQST_CACHE_HITS],
2899                .maxlen         = sizeof(int),
2900                .mode           = 0444,
2901                .proc_handler   = do_proc_dqstats,
2902        },
2903        {
2904                .procname       = "allocated_dquots",
2905                .data           = &dqstats.stat[DQST_ALLOC_DQUOTS],
2906                .maxlen         = sizeof(int),
2907                .mode           = 0444,
2908                .proc_handler   = do_proc_dqstats,
2909        },
2910        {
2911                .procname       = "free_dquots",
2912                .data           = &dqstats.stat[DQST_FREE_DQUOTS],
2913                .maxlen         = sizeof(int),
2914                .mode           = 0444,
2915                .proc_handler   = do_proc_dqstats,
2916        },
2917        {
2918                .procname       = "syncs",
2919                .data           = &dqstats.stat[DQST_SYNCS],
2920                .maxlen         = sizeof(int),
2921                .mode           = 0444,
2922                .proc_handler   = do_proc_dqstats,
2923        },
2924#ifdef CONFIG_PRINT_QUOTA_WARNING
2925        {
2926                .procname       = "warnings",
2927                .data           = &flag_print_warnings,
2928                .maxlen         = sizeof(int),
2929                .mode           = 0644,
2930                .proc_handler   = proc_dointvec,
2931        },
2932#endif
2933        { },
2934};
2935
2936static struct ctl_table fs_table[] = {
2937        {
2938                .procname       = "quota",
2939                .mode           = 0555,
2940                .child          = fs_dqstats_table,
2941        },
2942        { },
2943};
2944
2945static struct ctl_table sys_table[] = {
2946        {
2947                .procname       = "fs",
2948                .mode           = 0555,
2949                .child          = fs_table,
2950        },
2951        { },
2952};
2953
2954static int __init dquot_init(void)
2955{
2956        int i, ret;
2957        unsigned long nr_hash, order;
2958
2959        printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
2960
2961        register_sysctl_table(sys_table);
2962
2963        dquot_cachep = kmem_cache_create("dquot",
2964                        sizeof(struct dquot), sizeof(unsigned long) * 4,
2965                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
2966                                SLAB_MEM_SPREAD|SLAB_PANIC),
2967                        NULL);
2968
2969        order = 0;
2970        dquot_hash = (struct hlist_head *)__get_free_pages(GFP_KERNEL, order);
2971        if (!dquot_hash)
2972                panic("Cannot create dquot hash table");
2973
2974        for (i = 0; i < _DQST_DQSTAT_LAST; i++) {
2975                ret = percpu_counter_init(&dqstats.counter[i], 0, GFP_KERNEL);
2976                if (ret)
2977                        panic("Cannot create dquot stat counters");
2978        }
2979
2980        /* Find power-of-two hlist_heads which can fit into allocation */
2981        nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
2982        dq_hash_bits = 0;
2983        do {
2984                dq_hash_bits++;
2985        } while (nr_hash >> dq_hash_bits);
2986        dq_hash_bits--;
2987
2988        nr_hash = 1UL << dq_hash_bits;
2989        dq_hash_mask = nr_hash - 1;
2990        for (i = 0; i < nr_hash; i++)
2991                INIT_HLIST_HEAD(dquot_hash + i);
2992
2993        pr_info("VFS: Dquot-cache hash table entries: %ld (order %ld,"
2994                " %ld bytes)\n", nr_hash, order, (PAGE_SIZE << order));
2995
2996        if (register_shrinker(&dqcache_shrinker))
2997                panic("Cannot register dquot shrinker");
2998
2999        return 0;
3000}
3001fs_initcall(dquot_init);
3002