linux/fs/xfs/xfs_dquot.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2003 Silicon Graphics, Inc.
   4 * All Rights Reserved.
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_format.h"
   9#include "xfs_log_format.h"
  10#include "xfs_shared.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_bit.h"
  13#include "xfs_mount.h"
  14#include "xfs_defer.h"
  15#include "xfs_inode.h"
  16#include "xfs_bmap.h"
  17#include "xfs_quota.h"
  18#include "xfs_trans.h"
  19#include "xfs_buf_item.h"
  20#include "xfs_trans_space.h"
  21#include "xfs_trans_priv.h"
  22#include "xfs_qm.h"
  23#include "xfs_trace.h"
  24#include "xfs_log.h"
  25#include "xfs_bmap_btree.h"
  26
  27/*
  28 * Lock order:
  29 *
  30 * ip->i_lock
  31 *   qi->qi_tree_lock
  32 *     dquot->q_qlock (xfs_dqlock() and friends)
  33 *       dquot->q_flush (xfs_dqflock() and friends)
  34 *       qi->qi_lru_lock
  35 *
  36 * If two dquots need to be locked the order is user before group/project,
  37 * otherwise by the lowest id first, see xfs_dqlock2.
  38 */
  39
  40struct kmem_zone                *xfs_qm_dqtrxzone;
  41static struct kmem_zone         *xfs_qm_dqzone;
  42
  43static struct lock_class_key xfs_dquot_group_class;
  44static struct lock_class_key xfs_dquot_project_class;
  45
  46/*
  47 * This is called to free all the memory associated with a dquot
  48 */
  49void
  50xfs_qm_dqdestroy(
  51        xfs_dquot_t     *dqp)
  52{
  53        ASSERT(list_empty(&dqp->q_lru));
  54
  55        kmem_free(dqp->q_logitem.qli_item.li_lv_shadow);
  56        mutex_destroy(&dqp->q_qlock);
  57
  58        XFS_STATS_DEC(dqp->q_mount, xs_qm_dquot);
  59        kmem_zone_free(xfs_qm_dqzone, dqp);
  60}
  61
  62/*
  63 * If default limits are in force, push them into the dquot now.
  64 * We overwrite the dquot limits only if they are zero and this
  65 * is not the root dquot.
  66 */
  67void
  68xfs_qm_adjust_dqlimits(
  69        struct xfs_mount        *mp,
  70        struct xfs_dquot        *dq)
  71{
  72        struct xfs_quotainfo    *q = mp->m_quotainfo;
  73        struct xfs_disk_dquot   *d = &dq->q_core;
  74        struct xfs_def_quota    *defq;
  75        int                     prealloc = 0;
  76
  77        ASSERT(d->d_id);
  78        defq = xfs_get_defquota(dq, q);
  79
  80        if (defq->bsoftlimit && !d->d_blk_softlimit) {
  81                d->d_blk_softlimit = cpu_to_be64(defq->bsoftlimit);
  82                prealloc = 1;
  83        }
  84        if (defq->bhardlimit && !d->d_blk_hardlimit) {
  85                d->d_blk_hardlimit = cpu_to_be64(defq->bhardlimit);
  86                prealloc = 1;
  87        }
  88        if (defq->isoftlimit && !d->d_ino_softlimit)
  89                d->d_ino_softlimit = cpu_to_be64(defq->isoftlimit);
  90        if (defq->ihardlimit && !d->d_ino_hardlimit)
  91                d->d_ino_hardlimit = cpu_to_be64(defq->ihardlimit);
  92        if (defq->rtbsoftlimit && !d->d_rtb_softlimit)
  93                d->d_rtb_softlimit = cpu_to_be64(defq->rtbsoftlimit);
  94        if (defq->rtbhardlimit && !d->d_rtb_hardlimit)
  95                d->d_rtb_hardlimit = cpu_to_be64(defq->rtbhardlimit);
  96
  97        if (prealloc)
  98                xfs_dquot_set_prealloc_limits(dq);
  99}
 100
 101/*
 102 * Check the limits and timers of a dquot and start or reset timers
 103 * if necessary.
 104 * This gets called even when quota enforcement is OFF, which makes our
 105 * life a little less complicated. (We just don't reject any quota
 106 * reservations in that case, when enforcement is off).
 107 * We also return 0 as the values of the timers in Q_GETQUOTA calls, when
 108 * enforcement's off.
 109 * In contrast, warnings are a little different in that they don't
 110 * 'automatically' get started when limits get exceeded.  They do
 111 * get reset to zero, however, when we find the count to be under
 112 * the soft limit (they are only ever set non-zero via userspace).
 113 */
 114void
 115xfs_qm_adjust_dqtimers(
 116        xfs_mount_t             *mp,
 117        xfs_disk_dquot_t        *d)
 118{
 119        ASSERT(d->d_id);
 120
 121#ifdef DEBUG
 122        if (d->d_blk_hardlimit)
 123                ASSERT(be64_to_cpu(d->d_blk_softlimit) <=
 124                       be64_to_cpu(d->d_blk_hardlimit));
 125        if (d->d_ino_hardlimit)
 126                ASSERT(be64_to_cpu(d->d_ino_softlimit) <=
 127                       be64_to_cpu(d->d_ino_hardlimit));
 128        if (d->d_rtb_hardlimit)
 129                ASSERT(be64_to_cpu(d->d_rtb_softlimit) <=
 130                       be64_to_cpu(d->d_rtb_hardlimit));
 131#endif
 132
 133        if (!d->d_btimer) {
 134                if ((d->d_blk_softlimit &&
 135                     (be64_to_cpu(d->d_bcount) >
 136                      be64_to_cpu(d->d_blk_softlimit))) ||
 137                    (d->d_blk_hardlimit &&
 138                     (be64_to_cpu(d->d_bcount) >
 139                      be64_to_cpu(d->d_blk_hardlimit)))) {
 140                        d->d_btimer = cpu_to_be32(get_seconds() +
 141                                        mp->m_quotainfo->qi_btimelimit);
 142                } else {
 143                        d->d_bwarns = 0;
 144                }
 145        } else {
 146                if ((!d->d_blk_softlimit ||
 147                     (be64_to_cpu(d->d_bcount) <=
 148                      be64_to_cpu(d->d_blk_softlimit))) &&
 149                    (!d->d_blk_hardlimit ||
 150                    (be64_to_cpu(d->d_bcount) <=
 151                     be64_to_cpu(d->d_blk_hardlimit)))) {
 152                        d->d_btimer = 0;
 153                }
 154        }
 155
 156        if (!d->d_itimer) {
 157                if ((d->d_ino_softlimit &&
 158                     (be64_to_cpu(d->d_icount) >
 159                      be64_to_cpu(d->d_ino_softlimit))) ||
 160                    (d->d_ino_hardlimit &&
 161                     (be64_to_cpu(d->d_icount) >
 162                      be64_to_cpu(d->d_ino_hardlimit)))) {
 163                        d->d_itimer = cpu_to_be32(get_seconds() +
 164                                        mp->m_quotainfo->qi_itimelimit);
 165                } else {
 166                        d->d_iwarns = 0;
 167                }
 168        } else {
 169                if ((!d->d_ino_softlimit ||
 170                     (be64_to_cpu(d->d_icount) <=
 171                      be64_to_cpu(d->d_ino_softlimit)))  &&
 172                    (!d->d_ino_hardlimit ||
 173                     (be64_to_cpu(d->d_icount) <=
 174                      be64_to_cpu(d->d_ino_hardlimit)))) {
 175                        d->d_itimer = 0;
 176                }
 177        }
 178
 179        if (!d->d_rtbtimer) {
 180                if ((d->d_rtb_softlimit &&
 181                     (be64_to_cpu(d->d_rtbcount) >
 182                      be64_to_cpu(d->d_rtb_softlimit))) ||
 183                    (d->d_rtb_hardlimit &&
 184                     (be64_to_cpu(d->d_rtbcount) >
 185                      be64_to_cpu(d->d_rtb_hardlimit)))) {
 186                        d->d_rtbtimer = cpu_to_be32(get_seconds() +
 187                                        mp->m_quotainfo->qi_rtbtimelimit);
 188                } else {
 189                        d->d_rtbwarns = 0;
 190                }
 191        } else {
 192                if ((!d->d_rtb_softlimit ||
 193                     (be64_to_cpu(d->d_rtbcount) <=
 194                      be64_to_cpu(d->d_rtb_softlimit))) &&
 195                    (!d->d_rtb_hardlimit ||
 196                     (be64_to_cpu(d->d_rtbcount) <=
 197                      be64_to_cpu(d->d_rtb_hardlimit)))) {
 198                        d->d_rtbtimer = 0;
 199                }
 200        }
 201}
 202
 203/*
 204 * initialize a buffer full of dquots and log the whole thing
 205 */
 206STATIC void
 207xfs_qm_init_dquot_blk(
 208        xfs_trans_t     *tp,
 209        xfs_mount_t     *mp,
 210        xfs_dqid_t      id,
 211        uint            type,
 212        xfs_buf_t       *bp)
 213{
 214        struct xfs_quotainfo    *q = mp->m_quotainfo;
 215        xfs_dqblk_t     *d;
 216        xfs_dqid_t      curid;
 217        int             i;
 218
 219        ASSERT(tp);
 220        ASSERT(xfs_buf_islocked(bp));
 221
 222        d = bp->b_addr;
 223
 224        /*
 225         * ID of the first dquot in the block - id's are zero based.
 226         */
 227        curid = id - (id % q->qi_dqperchunk);
 228        memset(d, 0, BBTOB(q->qi_dqchunklen));
 229        for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) {
 230                d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
 231                d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
 232                d->dd_diskdq.d_id = cpu_to_be32(curid);
 233                d->dd_diskdq.d_flags = type;
 234                if (xfs_sb_version_hascrc(&mp->m_sb)) {
 235                        uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
 236                        xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
 237                                         XFS_DQUOT_CRC_OFF);
 238                }
 239        }
 240
 241        xfs_trans_dquot_buf(tp, bp,
 242                            (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
 243                            ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
 244                             XFS_BLF_GDQUOT_BUF)));
 245        xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
 246}
 247
 248/*
 249 * Initialize the dynamic speculative preallocation thresholds. The lo/hi
 250 * watermarks correspond to the soft and hard limits by default. If a soft limit
 251 * is not specified, we use 95% of the hard limit.
 252 */
 253void
 254xfs_dquot_set_prealloc_limits(struct xfs_dquot *dqp)
 255{
 256        uint64_t space;
 257
 258        dqp->q_prealloc_hi_wmark = be64_to_cpu(dqp->q_core.d_blk_hardlimit);
 259        dqp->q_prealloc_lo_wmark = be64_to_cpu(dqp->q_core.d_blk_softlimit);
 260        if (!dqp->q_prealloc_lo_wmark) {
 261                dqp->q_prealloc_lo_wmark = dqp->q_prealloc_hi_wmark;
 262                do_div(dqp->q_prealloc_lo_wmark, 100);
 263                dqp->q_prealloc_lo_wmark *= 95;
 264        }
 265
 266        space = dqp->q_prealloc_hi_wmark;
 267
 268        do_div(space, 100);
 269        dqp->q_low_space[XFS_QLOWSP_1_PCNT] = space;
 270        dqp->q_low_space[XFS_QLOWSP_3_PCNT] = space * 3;
 271        dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
 272}
 273
 274/*
 275 * Ensure that the given in-core dquot has a buffer on disk backing it, and
 276 * return the buffer locked and held. This is called when the bmapi finds a
 277 * hole.
 278 */
 279STATIC int
 280xfs_dquot_disk_alloc(
 281        struct xfs_trans        **tpp,
 282        struct xfs_dquot        *dqp,
 283        struct xfs_buf          **bpp)
 284{
 285        struct xfs_bmbt_irec    map;
 286        struct xfs_trans        *tp = *tpp;
 287        struct xfs_mount        *mp = tp->t_mountp;
 288        struct xfs_buf          *bp;
 289        struct xfs_inode        *quotip = xfs_quota_inode(mp, dqp->dq_flags);
 290        int                     nmaps = 1;
 291        int                     error;
 292
 293        trace_xfs_dqalloc(dqp);
 294
 295        xfs_ilock(quotip, XFS_ILOCK_EXCL);
 296        if (!xfs_this_quota_on(dqp->q_mount, dqp->dq_flags)) {
 297                /*
 298                 * Return if this type of quotas is turned off while we didn't
 299                 * have an inode lock
 300                 */
 301                xfs_iunlock(quotip, XFS_ILOCK_EXCL);
 302                return -ESRCH;
 303        }
 304
 305        /* Create the block mapping. */
 306        xfs_trans_ijoin(tp, quotip, XFS_ILOCK_EXCL);
 307        error = xfs_bmapi_write(tp, quotip, dqp->q_fileoffset,
 308                        XFS_DQUOT_CLUSTER_SIZE_FSB, XFS_BMAPI_METADATA,
 309                        XFS_QM_DQALLOC_SPACE_RES(mp), &map, &nmaps);
 310        if (error)
 311                return error;
 312        ASSERT(map.br_blockcount == XFS_DQUOT_CLUSTER_SIZE_FSB);
 313        ASSERT(nmaps == 1);
 314        ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
 315               (map.br_startblock != HOLESTARTBLOCK));
 316
 317        /*
 318         * Keep track of the blkno to save a lookup later
 319         */
 320        dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 321
 322        /* now we can just get the buffer (there's nothing to read yet) */
 323        bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, dqp->q_blkno,
 324                        mp->m_quotainfo->qi_dqchunklen, 0);
 325        if (!bp)
 326                return -ENOMEM;
 327        bp->b_ops = &xfs_dquot_buf_ops;
 328
 329        /*
 330         * Make a chunk of dquots out of this buffer and log
 331         * the entire thing.
 332         */
 333        xfs_qm_init_dquot_blk(tp, mp, be32_to_cpu(dqp->q_core.d_id),
 334                              dqp->dq_flags & XFS_DQ_ALLTYPES, bp);
 335        xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 336
 337        /*
 338         * Hold the buffer and join it to the dfops so that we'll still own
 339         * the buffer when we return to the caller.  The buffer disposal on
 340         * error must be paid attention to very carefully, as it has been
 341         * broken since commit efa092f3d4c6 "[XFS] Fixes a bug in the quota
 342         * code when allocating a new dquot record" in 2005, and the later
 343         * conversion to xfs_defer_ops in commit 310a75a3c6c747 failed to keep
 344         * the buffer locked across the _defer_finish call.  We can now do
 345         * this correctly with xfs_defer_bjoin.
 346         *
 347         * Above, we allocated a disk block for the dquot information and used
 348         * get_buf to initialize the dquot. If the _defer_finish fails, the old
 349         * transaction is gone but the new buffer is not joined or held to any
 350         * transaction, so we must _buf_relse it.
 351         *
 352         * If everything succeeds, the caller of this function is returned a
 353         * buffer that is locked and held to the transaction.  The caller
 354         * is responsible for unlocking any buffer passed back, either
 355         * manually or by committing the transaction.  On error, the buffer is
 356         * released and not passed back.
 357         */
 358        xfs_trans_bhold(tp, bp);
 359        error = xfs_defer_finish(tpp);
 360        if (error) {
 361                xfs_trans_bhold_release(*tpp, bp);
 362                xfs_trans_brelse(*tpp, bp);
 363                return error;
 364        }
 365        *bpp = bp;
 366        return 0;
 367}
 368
 369/*
 370 * Read in the in-core dquot's on-disk metadata and return the buffer.
 371 * Returns ENOENT to signal a hole.
 372 */
 373STATIC int
 374xfs_dquot_disk_read(
 375        struct xfs_mount        *mp,
 376        struct xfs_dquot        *dqp,
 377        struct xfs_buf          **bpp)
 378{
 379        struct xfs_bmbt_irec    map;
 380        struct xfs_buf          *bp;
 381        struct xfs_inode        *quotip = xfs_quota_inode(mp, dqp->dq_flags);
 382        uint                    lock_mode;
 383        int                     nmaps = 1;
 384        int                     error;
 385
 386        lock_mode = xfs_ilock_data_map_shared(quotip);
 387        if (!xfs_this_quota_on(mp, dqp->dq_flags)) {
 388                /*
 389                 * Return if this type of quotas is turned off while we
 390                 * didn't have the quota inode lock.
 391                 */
 392                xfs_iunlock(quotip, lock_mode);
 393                return -ESRCH;
 394        }
 395
 396        /*
 397         * Find the block map; no allocations yet
 398         */
 399        error = xfs_bmapi_read(quotip, dqp->q_fileoffset,
 400                        XFS_DQUOT_CLUSTER_SIZE_FSB, &map, &nmaps, 0);
 401        xfs_iunlock(quotip, lock_mode);
 402        if (error)
 403                return error;
 404
 405        ASSERT(nmaps == 1);
 406        ASSERT(map.br_blockcount >= 1);
 407        ASSERT(map.br_startblock != DELAYSTARTBLOCK);
 408        if (map.br_startblock == HOLESTARTBLOCK)
 409                return -ENOENT;
 410
 411        trace_xfs_dqtobp_read(dqp);
 412
 413        /*
 414         * store the blkno etc so that we don't have to do the
 415         * mapping all the time
 416         */
 417        dqp->q_blkno = XFS_FSB_TO_DADDR(mp, map.br_startblock);
 418
 419        error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
 420                        mp->m_quotainfo->qi_dqchunklen, 0, &bp,
 421                        &xfs_dquot_buf_ops);
 422        if (error) {
 423                ASSERT(bp == NULL);
 424                return error;
 425        }
 426
 427        ASSERT(xfs_buf_islocked(bp));
 428        xfs_buf_set_ref(bp, XFS_DQUOT_REF);
 429        *bpp = bp;
 430
 431        return 0;
 432}
 433
 434/* Allocate and initialize everything we need for an incore dquot. */
 435STATIC struct xfs_dquot *
 436xfs_dquot_alloc(
 437        struct xfs_mount        *mp,
 438        xfs_dqid_t              id,
 439        uint                    type)
 440{
 441        struct xfs_dquot        *dqp;
 442
 443        dqp = kmem_zone_zalloc(xfs_qm_dqzone, 0);
 444
 445        dqp->dq_flags = type;
 446        dqp->q_core.d_id = cpu_to_be32(id);
 447        dqp->q_mount = mp;
 448        INIT_LIST_HEAD(&dqp->q_lru);
 449        mutex_init(&dqp->q_qlock);
 450        init_waitqueue_head(&dqp->q_pinwait);
 451        dqp->q_fileoffset = (xfs_fileoff_t)id / mp->m_quotainfo->qi_dqperchunk;
 452        /*
 453         * Offset of dquot in the (fixed sized) dquot chunk.
 454         */
 455        dqp->q_bufoffset = (id % mp->m_quotainfo->qi_dqperchunk) *
 456                        sizeof(xfs_dqblk_t);
 457
 458        /*
 459         * Because we want to use a counting completion, complete
 460         * the flush completion once to allow a single access to
 461         * the flush completion without blocking.
 462         */
 463        init_completion(&dqp->q_flush);
 464        complete(&dqp->q_flush);
 465
 466        /*
 467         * Make sure group quotas have a different lock class than user
 468         * quotas.
 469         */
 470        switch (type) {
 471        case XFS_DQ_USER:
 472                /* uses the default lock class */
 473                break;
 474        case XFS_DQ_GROUP:
 475                lockdep_set_class(&dqp->q_qlock, &xfs_dquot_group_class);
 476                break;
 477        case XFS_DQ_PROJ:
 478                lockdep_set_class(&dqp->q_qlock, &xfs_dquot_project_class);
 479                break;
 480        default:
 481                ASSERT(0);
 482                break;
 483        }
 484
 485        xfs_qm_dquot_logitem_init(dqp);
 486
 487        XFS_STATS_INC(mp, xs_qm_dquot);
 488        return dqp;
 489}
 490
 491/* Copy the in-core quota fields in from the on-disk buffer. */
 492STATIC void
 493xfs_dquot_from_disk(
 494        struct xfs_dquot        *dqp,
 495        struct xfs_buf          *bp)
 496{
 497        struct xfs_disk_dquot   *ddqp = bp->b_addr + dqp->q_bufoffset;
 498
 499        /* copy everything from disk dquot to the incore dquot */
 500        memcpy(&dqp->q_core, ddqp, sizeof(xfs_disk_dquot_t));
 501
 502        /*
 503         * Reservation counters are defined as reservation plus current usage
 504         * to avoid having to add every time.
 505         */
 506        dqp->q_res_bcount = be64_to_cpu(ddqp->d_bcount);
 507        dqp->q_res_icount = be64_to_cpu(ddqp->d_icount);
 508        dqp->q_res_rtbcount = be64_to_cpu(ddqp->d_rtbcount);
 509
 510        /* initialize the dquot speculative prealloc thresholds */
 511        xfs_dquot_set_prealloc_limits(dqp);
 512}
 513
 514/* Allocate and initialize the dquot buffer for this in-core dquot. */
 515static int
 516xfs_qm_dqread_alloc(
 517        struct xfs_mount        *mp,
 518        struct xfs_dquot        *dqp,
 519        struct xfs_buf          **bpp)
 520{
 521        struct xfs_trans        *tp;
 522        int                     error;
 523
 524        error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_dqalloc,
 525                        XFS_QM_DQALLOC_SPACE_RES(mp), 0, 0, &tp);
 526        if (error)
 527                goto err;
 528
 529        error = xfs_dquot_disk_alloc(&tp, dqp, bpp);
 530        if (error)
 531                goto err_cancel;
 532
 533        error = xfs_trans_commit(tp);
 534        if (error) {
 535                /*
 536                 * Buffer was held to the transaction, so we have to unlock it
 537                 * manually here because we're not passing it back.
 538                 */
 539                xfs_buf_relse(*bpp);
 540                *bpp = NULL;
 541                goto err;
 542        }
 543        return 0;
 544
 545err_cancel:
 546        xfs_trans_cancel(tp);
 547err:
 548        return error;
 549}
 550
 551/*
 552 * Read in the ondisk dquot using dqtobp() then copy it to an incore version,
 553 * and release the buffer immediately.  If @can_alloc is true, fill any
 554 * holes in the on-disk metadata.
 555 */
 556static int
 557xfs_qm_dqread(
 558        struct xfs_mount        *mp,
 559        xfs_dqid_t              id,
 560        uint                    type,
 561        bool                    can_alloc,
 562        struct xfs_dquot        **dqpp)
 563{
 564        struct xfs_dquot        *dqp;
 565        struct xfs_buf          *bp;
 566        int                     error;
 567
 568        dqp = xfs_dquot_alloc(mp, id, type);
 569        trace_xfs_dqread(dqp);
 570
 571        /* Try to read the buffer, allocating if necessary. */
 572        error = xfs_dquot_disk_read(mp, dqp, &bp);
 573        if (error == -ENOENT && can_alloc)
 574                error = xfs_qm_dqread_alloc(mp, dqp, &bp);
 575        if (error)
 576                goto err;
 577
 578        /*
 579         * At this point we should have a clean locked buffer.  Copy the data
 580         * to the incore dquot and release the buffer since the incore dquot
 581         * has its own locking protocol so we needn't tie up the buffer any
 582         * further.
 583         */
 584        ASSERT(xfs_buf_islocked(bp));
 585        xfs_dquot_from_disk(dqp, bp);
 586
 587        xfs_buf_relse(bp);
 588        *dqpp = dqp;
 589        return error;
 590
 591err:
 592        trace_xfs_dqread_fail(dqp);
 593        xfs_qm_dqdestroy(dqp);
 594        *dqpp = NULL;
 595        return error;
 596}
 597
 598/*
 599 * Advance to the next id in the current chunk, or if at the
 600 * end of the chunk, skip ahead to first id in next allocated chunk
 601 * using the SEEK_DATA interface.
 602 */
 603static int
 604xfs_dq_get_next_id(
 605        struct xfs_mount        *mp,
 606        uint                    type,
 607        xfs_dqid_t              *id)
 608{
 609        struct xfs_inode        *quotip = xfs_quota_inode(mp, type);
 610        xfs_dqid_t              next_id = *id + 1; /* simple advance */
 611        uint                    lock_flags;
 612        struct xfs_bmbt_irec    got;
 613        struct xfs_iext_cursor  cur;
 614        xfs_fsblock_t           start;
 615        int                     error = 0;
 616
 617        /* If we'd wrap past the max ID, stop */
 618        if (next_id < *id)
 619                return -ENOENT;
 620
 621        /* If new ID is within the current chunk, advancing it sufficed */
 622        if (next_id % mp->m_quotainfo->qi_dqperchunk) {
 623                *id = next_id;
 624                return 0;
 625        }
 626
 627        /* Nope, next_id is now past the current chunk, so find the next one */
 628        start = (xfs_fsblock_t)next_id / mp->m_quotainfo->qi_dqperchunk;
 629
 630        lock_flags = xfs_ilock_data_map_shared(quotip);
 631        if (!(quotip->i_df.if_flags & XFS_IFEXTENTS)) {
 632                error = xfs_iread_extents(NULL, quotip, XFS_DATA_FORK);
 633                if (error)
 634                        return error;
 635        }
 636
 637        if (xfs_iext_lookup_extent(quotip, &quotip->i_df, start, &cur, &got)) {
 638                /* contiguous chunk, bump startoff for the id calculation */
 639                if (got.br_startoff < start)
 640                        got.br_startoff = start;
 641                *id = got.br_startoff * mp->m_quotainfo->qi_dqperchunk;
 642        } else {
 643                error = -ENOENT;
 644        }
 645
 646        xfs_iunlock(quotip, lock_flags);
 647
 648        return error;
 649}
 650
 651/*
 652 * Look up the dquot in the in-core cache.  If found, the dquot is returned
 653 * locked and ready to go.
 654 */
 655static struct xfs_dquot *
 656xfs_qm_dqget_cache_lookup(
 657        struct xfs_mount        *mp,
 658        struct xfs_quotainfo    *qi,
 659        struct radix_tree_root  *tree,
 660        xfs_dqid_t              id)
 661{
 662        struct xfs_dquot        *dqp;
 663
 664restart:
 665        mutex_lock(&qi->qi_tree_lock);
 666        dqp = radix_tree_lookup(tree, id);
 667        if (!dqp) {
 668                mutex_unlock(&qi->qi_tree_lock);
 669                XFS_STATS_INC(mp, xs_qm_dqcachemisses);
 670                return NULL;
 671        }
 672
 673        xfs_dqlock(dqp);
 674        if (dqp->dq_flags & XFS_DQ_FREEING) {
 675                xfs_dqunlock(dqp);
 676                mutex_unlock(&qi->qi_tree_lock);
 677                trace_xfs_dqget_freeing(dqp);
 678                delay(1);
 679                goto restart;
 680        }
 681
 682        dqp->q_nrefs++;
 683        mutex_unlock(&qi->qi_tree_lock);
 684
 685        trace_xfs_dqget_hit(dqp);
 686        XFS_STATS_INC(mp, xs_qm_dqcachehits);
 687        return dqp;
 688}
 689
 690/*
 691 * Try to insert a new dquot into the in-core cache.  If an error occurs the
 692 * caller should throw away the dquot and start over.  Otherwise, the dquot
 693 * is returned locked (and held by the cache) as if there had been a cache
 694 * hit.
 695 */
 696static int
 697xfs_qm_dqget_cache_insert(
 698        struct xfs_mount        *mp,
 699        struct xfs_quotainfo    *qi,
 700        struct radix_tree_root  *tree,
 701        xfs_dqid_t              id,
 702        struct xfs_dquot        *dqp)
 703{
 704        int                     error;
 705
 706        mutex_lock(&qi->qi_tree_lock);
 707        error = radix_tree_insert(tree, id, dqp);
 708        if (unlikely(error)) {
 709                /* Duplicate found!  Caller must try again. */
 710                WARN_ON(error != -EEXIST);
 711                mutex_unlock(&qi->qi_tree_lock);
 712                trace_xfs_dqget_dup(dqp);
 713                return error;
 714        }
 715
 716        /* Return a locked dquot to the caller, with a reference taken. */
 717        xfs_dqlock(dqp);
 718        dqp->q_nrefs = 1;
 719
 720        qi->qi_dquots++;
 721        mutex_unlock(&qi->qi_tree_lock);
 722
 723        return 0;
 724}
 725
 726/* Check our input parameters. */
 727static int
 728xfs_qm_dqget_checks(
 729        struct xfs_mount        *mp,
 730        uint                    type)
 731{
 732        if (WARN_ON_ONCE(!XFS_IS_QUOTA_RUNNING(mp)))
 733                return -ESRCH;
 734
 735        switch (type) {
 736        case XFS_DQ_USER:
 737                if (!XFS_IS_UQUOTA_ON(mp))
 738                        return -ESRCH;
 739                return 0;
 740        case XFS_DQ_GROUP:
 741                if (!XFS_IS_GQUOTA_ON(mp))
 742                        return -ESRCH;
 743                return 0;
 744        case XFS_DQ_PROJ:
 745                if (!XFS_IS_PQUOTA_ON(mp))
 746                        return -ESRCH;
 747                return 0;
 748        default:
 749                WARN_ON_ONCE(0);
 750                return -EINVAL;
 751        }
 752}
 753
 754/*
 755 * Given the file system, id, and type (UDQUOT/GDQUOT), return a a locked
 756 * dquot, doing an allocation (if requested) as needed.
 757 */
 758int
 759xfs_qm_dqget(
 760        struct xfs_mount        *mp,
 761        xfs_dqid_t              id,
 762        uint                    type,
 763        bool                    can_alloc,
 764        struct xfs_dquot        **O_dqpp)
 765{
 766        struct xfs_quotainfo    *qi = mp->m_quotainfo;
 767        struct radix_tree_root  *tree = xfs_dquot_tree(qi, type);
 768        struct xfs_dquot        *dqp;
 769        int                     error;
 770
 771        error = xfs_qm_dqget_checks(mp, type);
 772        if (error)
 773                return error;
 774
 775restart:
 776        dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
 777        if (dqp) {
 778                *O_dqpp = dqp;
 779                return 0;
 780        }
 781
 782        error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
 783        if (error)
 784                return error;
 785
 786        error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
 787        if (error) {
 788                /*
 789                 * Duplicate found. Just throw away the new dquot and start
 790                 * over.
 791                 */
 792                xfs_qm_dqdestroy(dqp);
 793                XFS_STATS_INC(mp, xs_qm_dquot_dups);
 794                goto restart;
 795        }
 796
 797        trace_xfs_dqget_miss(dqp);
 798        *O_dqpp = dqp;
 799        return 0;
 800}
 801
 802/*
 803 * Given a dquot id and type, read and initialize a dquot from the on-disk
 804 * metadata.  This function is only for use during quota initialization so
 805 * it ignores the dquot cache assuming that the dquot shrinker isn't set up.
 806 * The caller is responsible for _qm_dqdestroy'ing the returned dquot.
 807 */
 808int
 809xfs_qm_dqget_uncached(
 810        struct xfs_mount        *mp,
 811        xfs_dqid_t              id,
 812        uint                    type,
 813        struct xfs_dquot        **dqpp)
 814{
 815        int                     error;
 816
 817        error = xfs_qm_dqget_checks(mp, type);
 818        if (error)
 819                return error;
 820
 821        return xfs_qm_dqread(mp, id, type, 0, dqpp);
 822}
 823
 824/* Return the quota id for a given inode and type. */
 825xfs_dqid_t
 826xfs_qm_id_for_quotatype(
 827        struct xfs_inode        *ip,
 828        uint                    type)
 829{
 830        switch (type) {
 831        case XFS_DQ_USER:
 832                return ip->i_d.di_uid;
 833        case XFS_DQ_GROUP:
 834                return ip->i_d.di_gid;
 835        case XFS_DQ_PROJ:
 836                return xfs_get_projid(ip);
 837        }
 838        ASSERT(0);
 839        return 0;
 840}
 841
 842/*
 843 * Return the dquot for a given inode and type.  If @can_alloc is true, then
 844 * allocate blocks if needed.  The inode's ILOCK must be held and it must not
 845 * have already had an inode attached.
 846 */
 847int
 848xfs_qm_dqget_inode(
 849        struct xfs_inode        *ip,
 850        uint                    type,
 851        bool                    can_alloc,
 852        struct xfs_dquot        **O_dqpp)
 853{
 854        struct xfs_mount        *mp = ip->i_mount;
 855        struct xfs_quotainfo    *qi = mp->m_quotainfo;
 856        struct radix_tree_root  *tree = xfs_dquot_tree(qi, type);
 857        struct xfs_dquot        *dqp;
 858        xfs_dqid_t              id;
 859        int                     error;
 860
 861        error = xfs_qm_dqget_checks(mp, type);
 862        if (error)
 863                return error;
 864
 865        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 866        ASSERT(xfs_inode_dquot(ip, type) == NULL);
 867
 868        id = xfs_qm_id_for_quotatype(ip, type);
 869
 870restart:
 871        dqp = xfs_qm_dqget_cache_lookup(mp, qi, tree, id);
 872        if (dqp) {
 873                *O_dqpp = dqp;
 874                return 0;
 875        }
 876
 877        /*
 878         * Dquot cache miss. We don't want to keep the inode lock across
 879         * a (potential) disk read. Also we don't want to deal with the lock
 880         * ordering between quotainode and this inode. OTOH, dropping the inode
 881         * lock here means dealing with a chown that can happen before
 882         * we re-acquire the lock.
 883         */
 884        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 885        error = xfs_qm_dqread(mp, id, type, can_alloc, &dqp);
 886        xfs_ilock(ip, XFS_ILOCK_EXCL);
 887        if (error)
 888                return error;
 889
 890        /*
 891         * A dquot could be attached to this inode by now, since we had
 892         * dropped the ilock.
 893         */
 894        if (xfs_this_quota_on(mp, type)) {
 895                struct xfs_dquot        *dqp1;
 896
 897                dqp1 = xfs_inode_dquot(ip, type);
 898                if (dqp1) {
 899                        xfs_qm_dqdestroy(dqp);
 900                        dqp = dqp1;
 901                        xfs_dqlock(dqp);
 902                        goto dqret;
 903                }
 904        } else {
 905                /* inode stays locked on return */
 906                xfs_qm_dqdestroy(dqp);
 907                return -ESRCH;
 908        }
 909
 910        error = xfs_qm_dqget_cache_insert(mp, qi, tree, id, dqp);
 911        if (error) {
 912                /*
 913                 * Duplicate found. Just throw away the new dquot and start
 914                 * over.
 915                 */
 916                xfs_qm_dqdestroy(dqp);
 917                XFS_STATS_INC(mp, xs_qm_dquot_dups);
 918                goto restart;
 919        }
 920
 921dqret:
 922        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 923        trace_xfs_dqget_miss(dqp);
 924        *O_dqpp = dqp;
 925        return 0;
 926}
 927
 928/*
 929 * Starting at @id and progressing upwards, look for an initialized incore
 930 * dquot, lock it, and return it.
 931 */
 932int
 933xfs_qm_dqget_next(
 934        struct xfs_mount        *mp,
 935        xfs_dqid_t              id,
 936        uint                    type,
 937        struct xfs_dquot        **dqpp)
 938{
 939        struct xfs_dquot        *dqp;
 940        int                     error = 0;
 941
 942        *dqpp = NULL;
 943        for (; !error; error = xfs_dq_get_next_id(mp, type, &id)) {
 944                error = xfs_qm_dqget(mp, id, type, false, &dqp);
 945                if (error == -ENOENT)
 946                        continue;
 947                else if (error != 0)
 948                        break;
 949
 950                if (!XFS_IS_DQUOT_UNINITIALIZED(dqp)) {
 951                        *dqpp = dqp;
 952                        return 0;
 953                }
 954
 955                xfs_qm_dqput(dqp);
 956        }
 957
 958        return error;
 959}
 960
 961/*
 962 * Release a reference to the dquot (decrement ref-count) and unlock it.
 963 *
 964 * If there is a group quota attached to this dquot, carefully release that
 965 * too without tripping over deadlocks'n'stuff.
 966 */
 967void
 968xfs_qm_dqput(
 969        struct xfs_dquot        *dqp)
 970{
 971        ASSERT(dqp->q_nrefs > 0);
 972        ASSERT(XFS_DQ_IS_LOCKED(dqp));
 973
 974        trace_xfs_dqput(dqp);
 975
 976        if (--dqp->q_nrefs == 0) {
 977                struct xfs_quotainfo    *qi = dqp->q_mount->m_quotainfo;
 978                trace_xfs_dqput_free(dqp);
 979
 980                if (list_lru_add(&qi->qi_lru, &dqp->q_lru))
 981                        XFS_STATS_INC(dqp->q_mount, xs_qm_dquot_unused);
 982        }
 983        xfs_dqunlock(dqp);
 984}
 985
 986/*
 987 * Release a dquot. Flush it if dirty, then dqput() it.
 988 * dquot must not be locked.
 989 */
 990void
 991xfs_qm_dqrele(
 992        xfs_dquot_t     *dqp)
 993{
 994        if (!dqp)
 995                return;
 996
 997        trace_xfs_dqrele(dqp);
 998
 999        xfs_dqlock(dqp);
1000        /*
1001         * We don't care to flush it if the dquot is dirty here.
1002         * That will create stutters that we want to avoid.
1003         * Instead we do a delayed write when we try to reclaim
1004         * a dirty dquot. Also xfs_sync will take part of the burden...
1005         */
1006        xfs_qm_dqput(dqp);
1007}
1008
1009/*
1010 * This is the dquot flushing I/O completion routine.  It is called
1011 * from interrupt level when the buffer containing the dquot is
1012 * flushed to disk.  It is responsible for removing the dquot logitem
1013 * from the AIL if it has not been re-logged, and unlocking the dquot's
1014 * flush lock. This behavior is very similar to that of inodes..
1015 */
1016STATIC void
1017xfs_qm_dqflush_done(
1018        struct xfs_buf          *bp,
1019        struct xfs_log_item     *lip)
1020{
1021        xfs_dq_logitem_t        *qip = (struct xfs_dq_logitem *)lip;
1022        xfs_dquot_t             *dqp = qip->qli_dquot;
1023        struct xfs_ail          *ailp = lip->li_ailp;
1024
1025        /*
1026         * We only want to pull the item from the AIL if its
1027         * location in the log has not changed since we started the flush.
1028         * Thus, we only bother if the dquot's lsn has
1029         * not changed. First we check the lsn outside the lock
1030         * since it's cheaper, and then we recheck while
1031         * holding the lock before removing the dquot from the AIL.
1032         */
1033        if (test_bit(XFS_LI_IN_AIL, &lip->li_flags) &&
1034            ((lip->li_lsn == qip->qli_flush_lsn) ||
1035             test_bit(XFS_LI_FAILED, &lip->li_flags))) {
1036
1037                /* xfs_trans_ail_delete() drops the AIL lock. */
1038                spin_lock(&ailp->ail_lock);
1039                if (lip->li_lsn == qip->qli_flush_lsn) {
1040                        xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1041                } else {
1042                        /*
1043                         * Clear the failed state since we are about to drop the
1044                         * flush lock
1045                         */
1046                        xfs_clear_li_failed(lip);
1047                        spin_unlock(&ailp->ail_lock);
1048                }
1049        }
1050
1051        /*
1052         * Release the dq's flush lock since we're done with it.
1053         */
1054        xfs_dqfunlock(dqp);
1055}
1056
1057/*
1058 * Write a modified dquot to disk.
1059 * The dquot must be locked and the flush lock too taken by caller.
1060 * The flush lock will not be unlocked until the dquot reaches the disk,
1061 * but the dquot is free to be unlocked and modified by the caller
1062 * in the interim. Dquot is still locked on return. This behavior is
1063 * identical to that of inodes.
1064 */
1065int
1066xfs_qm_dqflush(
1067        struct xfs_dquot        *dqp,
1068        struct xfs_buf          **bpp)
1069{
1070        struct xfs_mount        *mp = dqp->q_mount;
1071        struct xfs_buf          *bp;
1072        struct xfs_dqblk        *dqb;
1073        struct xfs_disk_dquot   *ddqp;
1074        xfs_failaddr_t          fa;
1075        int                     error;
1076
1077        ASSERT(XFS_DQ_IS_LOCKED(dqp));
1078        ASSERT(!completion_done(&dqp->q_flush));
1079
1080        trace_xfs_dqflush(dqp);
1081
1082        *bpp = NULL;
1083
1084        xfs_qm_dqunpin_wait(dqp);
1085
1086        /*
1087         * This may have been unpinned because the filesystem is shutting
1088         * down forcibly. If that's the case we must not write this dquot
1089         * to disk, because the log record didn't make it to disk.
1090         *
1091         * We also have to remove the log item from the AIL in this case,
1092         * as we wait for an emptry AIL as part of the unmount process.
1093         */
1094        if (XFS_FORCED_SHUTDOWN(mp)) {
1095                struct xfs_log_item     *lip = &dqp->q_logitem.qli_item;
1096                dqp->dq_flags &= ~XFS_DQ_DIRTY;
1097
1098                xfs_trans_ail_remove(lip, SHUTDOWN_CORRUPT_INCORE);
1099
1100                error = -EIO;
1101                goto out_unlock;
1102        }
1103
1104        /*
1105         * Get the buffer containing the on-disk dquot
1106         */
1107        error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dqp->q_blkno,
1108                                   mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1109                                   &xfs_dquot_buf_ops);
1110        if (error)
1111                goto out_unlock;
1112
1113        /*
1114         * Calculate the location of the dquot inside the buffer.
1115         */
1116        dqb = bp->b_addr + dqp->q_bufoffset;
1117        ddqp = &dqb->dd_diskdq;
1118
1119        /*
1120         * A simple sanity check in case we got a corrupted dquot.
1121         */
1122        fa = xfs_dqblk_verify(mp, dqb, be32_to_cpu(ddqp->d_id), 0);
1123        if (fa) {
1124                xfs_alert(mp, "corrupt dquot ID 0x%x in memory at %pS",
1125                                be32_to_cpu(ddqp->d_id), fa);
1126                xfs_buf_relse(bp);
1127                xfs_dqfunlock(dqp);
1128                xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1129                return -EIO;
1130        }
1131
1132        /* This is the only portion of data that needs to persist */
1133        memcpy(ddqp, &dqp->q_core, sizeof(xfs_disk_dquot_t));
1134
1135        /*
1136         * Clear the dirty field and remember the flush lsn for later use.
1137         */
1138        dqp->dq_flags &= ~XFS_DQ_DIRTY;
1139
1140        xfs_trans_ail_copy_lsn(mp->m_ail, &dqp->q_logitem.qli_flush_lsn,
1141                                        &dqp->q_logitem.qli_item.li_lsn);
1142
1143        /*
1144         * copy the lsn into the on-disk dquot now while we have the in memory
1145         * dquot here. This can't be done later in the write verifier as we
1146         * can't get access to the log item at that point in time.
1147         *
1148         * We also calculate the CRC here so that the on-disk dquot in the
1149         * buffer always has a valid CRC. This ensures there is no possibility
1150         * of a dquot without an up-to-date CRC getting to disk.
1151         */
1152        if (xfs_sb_version_hascrc(&mp->m_sb)) {
1153                dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
1154                xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
1155                                 XFS_DQUOT_CRC_OFF);
1156        }
1157
1158        /*
1159         * Attach an iodone routine so that we can remove this dquot from the
1160         * AIL and release the flush lock once the dquot is synced to disk.
1161         */
1162        xfs_buf_attach_iodone(bp, xfs_qm_dqflush_done,
1163                                  &dqp->q_logitem.qli_item);
1164
1165        /*
1166         * If the buffer is pinned then push on the log so we won't
1167         * get stuck waiting in the write for too long.
1168         */
1169        if (xfs_buf_ispinned(bp)) {
1170                trace_xfs_dqflush_force(dqp);
1171                xfs_log_force(mp, 0);
1172        }
1173
1174        trace_xfs_dqflush_done(dqp);
1175        *bpp = bp;
1176        return 0;
1177
1178out_unlock:
1179        xfs_dqfunlock(dqp);
1180        return -EIO;
1181}
1182
1183/*
1184 * Lock two xfs_dquot structures.
1185 *
1186 * To avoid deadlocks we always lock the quota structure with
1187 * the lowerd id first.
1188 */
1189void
1190xfs_dqlock2(
1191        xfs_dquot_t     *d1,
1192        xfs_dquot_t     *d2)
1193{
1194        if (d1 && d2) {
1195                ASSERT(d1 != d2);
1196                if (be32_to_cpu(d1->q_core.d_id) >
1197                    be32_to_cpu(d2->q_core.d_id)) {
1198                        mutex_lock(&d2->q_qlock);
1199                        mutex_lock_nested(&d1->q_qlock, XFS_QLOCK_NESTED);
1200                } else {
1201                        mutex_lock(&d1->q_qlock);
1202                        mutex_lock_nested(&d2->q_qlock, XFS_QLOCK_NESTED);
1203                }
1204        } else if (d1) {
1205                mutex_lock(&d1->q_qlock);
1206        } else if (d2) {
1207                mutex_lock(&d2->q_qlock);
1208        }
1209}
1210
1211int __init
1212xfs_qm_init(void)
1213{
1214        xfs_qm_dqzone =
1215                kmem_zone_init(sizeof(struct xfs_dquot), "xfs_dquot");
1216        if (!xfs_qm_dqzone)
1217                goto out;
1218
1219        xfs_qm_dqtrxzone =
1220                kmem_zone_init(sizeof(struct xfs_dquot_acct), "xfs_dqtrx");
1221        if (!xfs_qm_dqtrxzone)
1222                goto out_free_dqzone;
1223
1224        return 0;
1225
1226out_free_dqzone:
1227        kmem_zone_destroy(xfs_qm_dqzone);
1228out:
1229        return -ENOMEM;
1230}
1231
1232void
1233xfs_qm_exit(void)
1234{
1235        kmem_zone_destroy(xfs_qm_dqtrxzone);
1236        kmem_zone_destroy(xfs_qm_dqzone);
1237}
1238
1239/*
1240 * Iterate every dquot of a particular type.  The caller must ensure that the
1241 * particular quota type is active.  iter_fn can return negative error codes,
1242 * or -ECANCELED to indicate that it wants to stop iterating.
1243 */
1244int
1245xfs_qm_dqiterate(
1246        struct xfs_mount        *mp,
1247        uint                    dqtype,
1248        xfs_qm_dqiterate_fn     iter_fn,
1249        void                    *priv)
1250{
1251        struct xfs_dquot        *dq;
1252        xfs_dqid_t              id = 0;
1253        int                     error;
1254
1255        do {
1256                error = xfs_qm_dqget_next(mp, id, dqtype, &dq);
1257                if (error == -ENOENT)
1258                        return 0;
1259                if (error)
1260                        return error;
1261
1262                error = iter_fn(dq, dqtype, priv);
1263                id = be32_to_cpu(dq->q_core.d_id);
1264                xfs_qm_dqput(dq);
1265                id++;
1266        } while (error == 0 && id != 0);
1267
1268        return error;
1269}
1270