SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/gfs2/quota.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License v.2.
   8  */
   9
  10 /*
  11  * Quota change tags are associated with each transaction that allocates or
  12  * deallocates space.  Those changes are accumulated locally to each node (in a
  13  * per-node file) and then are periodically synced to the quota file.  This
  14  * avoids the bottleneck of constantly touching the quota file, but introduces
  15  * fuzziness in the current usage value of IDs that are being used on different
  16  * nodes in the cluster simultaneously.  So, it is possible for a user on
  17  * multiple nodes to overrun their quota, but that overrun is controlable.
  18  * Since quota tags are part of transactions, there is no need to a quota check
  19  * program to be run on node crashes or anything like that.
  20  *
  21  * There are couple of knobs that let the administrator manage the quota
  22  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
  23  * sitting on one node before being synced to the quota file.  (The default is
  24  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
  25  * of quota file syncs increases as the user moves closer to their limit.  The
  26  * more frequent the syncs, the more accurate the quota enforcement, but that
  27  * means that there is more contention between the nodes for the quota file.
  28  * The default value is one.  This sets the maximum theoretical quota overrun
  29  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
  30  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
  31  * number greater than one makes quota syncs more frequent and reduces the
  32  * maximum overrun.  Numbers less than one (but greater than zero) make quota
  33  * syncs less frequent.
  34  *
  35  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
  36  * the quota file, so it is not being constantly read.
  37  */
  38
  39 #include <linux/sched.h>
  40 #include <linux/slab.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/completion.h>
  43 #include <linux/buffer_head.h>
  44 #include <linux/tty.h>
  45 #include <linux/sort.h>
  46 #include <linux/fs.h>
  47 #include <asm/semaphore.h>
  48
  49 #include "gfs2.h"
  50 #include "bmap.h"
  51 #include "glock.h"
  52 #include "glops.h"
  53 #include "log.h"
  54 #include "meta_io.h"
  55 #include "quota.h"
  56 #include "rgrp.h"
  57 #include "super.h"
  58 #include "trans.h"
  59 #include "inode.h"
  60 #include "ops_file.h"
  61 #include "ops_address.h"
  62
  63 #define QUOTA_USER 1
  64 #define QUOTA_GROUP 0
  65
  66 static uint64_t qd2offset(struct gfs2_quota_data *qd)
  67 {
  68         uint64_t offset;
  69
  70         offset = 2 * (uint64_t)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
  71         offset *= sizeof(struct gfs2_quota);
  72
  73         return offset;
  74 }
  75
  76 static int qd_alloc(struct gfs2_sbd *sdp, int user, uint32_t id,
  77                     struct gfs2_quota_data **qdp)
  78 {
  79         struct gfs2_quota_data *qd;
  80         int error;
  81
  82         qd = kzalloc(sizeof(struct gfs2_quota_data), GFP_KERNEL);
  83         if (!qd)
  84                 return -ENOMEM;
  85
  86         qd->qd_count = 1;
  87         qd->qd_id = id;
  88         if (user)
  89                 set_bit(QDF_USER, &qd->qd_flags);
  90         qd->qd_slot = -1;
  91
  92         error = gfs2_glock_get(sdp, 2 * (uint64_t)id + !user,
  93                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
  94         if (error)
  95                 goto fail;
  96
  97         error = gfs2_lvb_hold(qd->qd_gl);
  98         gfs2_glock_put(qd->qd_gl);
  99         if (error)
 100                 goto fail;
 101
 102         *qdp = qd;
 103
 104         return 0;
 105
 106  fail:
 107         kfree(qd);
 108         return error;
 109 }
 110
 111 static int qd_get(struct gfs2_sbd *sdp, int user, uint32_t id, int create,
 112                   struct gfs2_quota_data **qdp)
 113 {
 114         struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
 115         int error, found;
 116
 117         *qdp = NULL;
 118
 119         for (;;) {
 120                 found = 0;
 121                 spin_lock(&sdp->sd_quota_spin);
 122                 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 123                         if (qd->qd_id == id &&
 124                             !test_bit(QDF_USER, &qd->qd_flags) == !user) {
 125                                 qd->qd_count++;
 126                                 found = 1;
 127                                 break;
 128                         }
 129                 }
 130
 131                 if (!found)
 132                         qd = NULL;
 133
 134                 if (!qd && new_qd) {
 135                         qd = new_qd;
 136                         list_add(&qd->qd_list, &sdp->sd_quota_list);
 137                         atomic_inc(&sdp->sd_quota_count);
 138                         new_qd = NULL;
 139                 }
 140
 141                 spin_unlock(&sdp->sd_quota_spin);
 142
 143                 if (qd || !create) {
 144                         if (new_qd) {
 145                                 gfs2_lvb_unhold(new_qd->qd_gl);
 146                                 kfree(new_qd);
 147                         }
 148                         *qdp = qd;
 149                         return 0;
 150                 }
 151
 152                 error = qd_alloc(sdp, user, id, &new_qd);
 153                 if (error)
 154                         return error;
 155         }
 156 }
 157
 158 static void qd_hold(struct gfs2_quota_data *qd)
 159 {
 160         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 161
 162         spin_lock(&sdp->sd_quota_spin);
 163         gfs2_assert(sdp, qd->qd_count);
 164         qd->qd_count++;
 165         spin_unlock(&sdp->sd_quota_spin);
 166 }
 167
 168 static void qd_put(struct gfs2_quota_data *qd)
 169 {
 170         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 171         spin_lock(&sdp->sd_quota_spin);
 172         gfs2_assert(sdp, qd->qd_count);
 173         if (!--qd->qd_count)
 174                 qd->qd_last_touched = jiffies;
 175         spin_unlock(&sdp->sd_quota_spin);
 176 }
 177
 178 static int slot_get(struct gfs2_quota_data *qd)
 179 {
 180         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 181         unsigned int c, o = 0, b;
 182         unsigned char byte = 0;
 183
 184         spin_lock(&sdp->sd_quota_spin);
 185
 186         if (qd->qd_slot_count++) {
 187                 spin_unlock(&sdp->sd_quota_spin);
 188                 return 0;
 189         }
 190
 191         for (c = 0; c < sdp->sd_quota_chunks; c++)
 192                 for (o = 0; o < PAGE_SIZE; o++) {
 193                         byte = sdp->sd_quota_bitmap[c][o];
 194                         if (byte != 0xFF)
 195                                 goto found;
 196                 }
 197
 198         goto fail;
 199
 200  found:
 201         for (b = 0; b < 8; b++)
 202                 if (!(byte & (1 << b)))
 203                         break;
 204         qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
 205
 206         if (qd->qd_slot >= sdp->sd_quota_slots)
 207                 goto fail;
 208
 209         sdp->sd_quota_bitmap[c][o] |= 1 << b;
 210
 211         spin_unlock(&sdp->sd_quota_spin);
 212
 213         return 0;
 214
 215  fail:
 216         qd->qd_slot_count--;
 217         spin_unlock(&sdp->sd_quota_spin);
 218         return -ENOSPC;
 219 }
 220
 221 static void slot_hold(struct gfs2_quota_data *qd)
 222 {
 223         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 224
 225         spin_lock(&sdp->sd_quota_spin);
 226         gfs2_assert(sdp, qd->qd_slot_count);
 227         qd->qd_slot_count++;
 228         spin_unlock(&sdp->sd_quota_spin);
 229 }
 230
 231 static void slot_put(struct gfs2_quota_data *qd)
 232 {
 233         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 234
 235         spin_lock(&sdp->sd_quota_spin);
 236         gfs2_assert(sdp, qd->qd_slot_count);
 237         if (!--qd->qd_slot_count) {
 238                 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
 239                 qd->qd_slot = -1;
 240         }
 241         spin_unlock(&sdp->sd_quota_spin);
 242 }
 243
 244 static int bh_get(struct gfs2_quota_data *qd)
 245 {
 246         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 247         struct gfs2_inode *ip = get_v2ip(sdp->sd_qc_inode);
 248         unsigned int block, offset;
 249         uint64_t dblock;
 250         int new = 0;
 251         struct buffer_head *bh;
 252         int error;
 253
 254         mutex_lock(&sdp->sd_quota_mutex);
 255
 256         if (qd->qd_bh_count++) {
 257                 mutex_unlock(&sdp->sd_quota_mutex);
 258                 return 0;
 259         }
 260
 261         block = qd->qd_slot / sdp->sd_qc_per_block;
 262         offset = qd->qd_slot % sdp->sd_qc_per_block;;
 263
 264         error = gfs2_block_map(ip, block, &new, &dblock, NULL);
 265         if (error)
 266                 goto fail;
 267         error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT, &bh);
 268         if (error)
 269                 goto fail;
 270         error = -EIO;
 271         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
 272                 goto fail_brelse;
 273
 274         qd->qd_bh = bh;
 275         qd->qd_bh_qc = (struct gfs2_quota_change *)
 276                 (bh->b_data + sizeof(struct gfs2_meta_header) +
 277                  offset * sizeof(struct gfs2_quota_change));
 278
 279         mutex_lock(&sdp->sd_quota_mutex);
 280
 281         return 0;
 282
 283  fail_brelse:
 284         brelse(bh);
 285
 286  fail:
 287         qd->qd_bh_count--;
 288         mutex_unlock(&sdp->sd_quota_mutex);
 289         return error;
 290 }
 291
 292 static void bh_put(struct gfs2_quota_data *qd)
 293 {
 294         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 295
 296         mutex_lock(&sdp->sd_quota_mutex);
 297         gfs2_assert(sdp, qd->qd_bh_count);
 298         if (!--qd->qd_bh_count) {
 299                 brelse(qd->qd_bh);
 300                 qd->qd_bh = NULL;
 301                 qd->qd_bh_qc = NULL;
 302         }
 303         mutex_unlock(&sdp->sd_quota_mutex);
 304 }
 305
 306 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
 307 {
 308         struct gfs2_quota_data *qd = NULL;
 309         int error;
 310         int found = 0;
 311
 312         *qdp = NULL;
 313
 314         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 315                 return 0;
 316
 317         spin_lock(&sdp->sd_quota_spin);
 318
 319         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 320                 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 321                     !test_bit(QDF_CHANGE, &qd->qd_flags) ||
 322                     qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
 323                         continue;
 324
 325                 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 326
 327                 set_bit(QDF_LOCKED, &qd->qd_flags);
 328                 gfs2_assert_warn(sdp, qd->qd_count);
 329                 qd->qd_count++;
 330                 qd->qd_change_sync = qd->qd_change;
 331                 gfs2_assert_warn(sdp, qd->qd_slot_count);
 332                 qd->qd_slot_count++;
 333                 found = 1;
 334
 335                 break;
 336         }
 337
 338         if (!found)
 339                 qd = NULL;
 340
 341         spin_unlock(&sdp->sd_quota_spin);
 342
 343         if (qd) {
 344                 gfs2_assert_warn(sdp, qd->qd_change_sync);
 345                 error = bh_get(qd);
 346                 if (error) {
 347                         clear_bit(QDF_LOCKED, &qd->qd_flags);
 348                         slot_put(qd);
 349                         qd_put(qd);
 350                         return error;
 351                 }
 352         }
 353
 354         *qdp = qd;
 355
 356         return 0;
 357 }
 358
 359 static int qd_trylock(struct gfs2_quota_data *qd)
 360 {
 361         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 362
 363         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 364                 return 0;
 365
 366         spin_lock(&sdp->sd_quota_spin);
 367
 368         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 369             !test_bit(QDF_CHANGE, &qd->qd_flags)) {
 370                 spin_unlock(&sdp->sd_quota_spin);
 371                 return 0;
 372         }
 373
 374         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 375
 376         set_bit(QDF_LOCKED, &qd->qd_flags);
 377         gfs2_assert_warn(sdp, qd->qd_count);
 378         qd->qd_count++;
 379         qd->qd_change_sync = qd->qd_change;
 380         gfs2_assert_warn(sdp, qd->qd_slot_count);
 381         qd->qd_slot_count++;
 382
 383         spin_unlock(&sdp->sd_quota_spin);
 384
 385         gfs2_assert_warn(sdp, qd->qd_change_sync);
 386         if (bh_get(qd)) {
 387                 clear_bit(QDF_LOCKED, &qd->qd_flags);
 388                 slot_put(qd);
 389                 qd_put(qd);
 390                 return 0;
 391         }
 392
 393         return 1;
 394 }
 395
 396 static void qd_unlock(struct gfs2_quota_data *qd)
 397 {
 398         gfs2_assert_warn(qd->qd_gl->gl_sbd,
 399                          test_bit(QDF_LOCKED, &qd->qd_flags));
 400         clear_bit(QDF_LOCKED, &qd->qd_flags);
 401         bh_put(qd);
 402         slot_put(qd);
 403         qd_put(qd);
 404 }
 405
 406 static int qdsb_get(struct gfs2_sbd *sdp, int user, uint32_t id, int create,
 407                     struct gfs2_quota_data **qdp)
 408 {
 409         int error;
 410
 411         error = qd_get(sdp, user, id, create, qdp);
 412         if (error)
 413                 return error;
 414
 415         error = slot_get(*qdp);
 416         if (error)
 417                 goto fail;
 418
 419         error = bh_get(*qdp);
 420         if (error)
 421                 goto fail_slot;
 422
 423         return 0;
 424
 425  fail_slot:
 426         slot_put(*qdp);
 427
 428  fail:
 429         qd_put(*qdp);
 430         return error;
 431 }
 432
 433 static void qdsb_put(struct gfs2_quota_data *qd)
 434 {
 435         bh_put(qd);
 436         slot_put(qd);
 437         qd_put(qd);
 438 }
 439
 440 int gfs2_quota_hold(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
 441 {
 442         struct gfs2_sbd *sdp = ip->i_sbd;
 443         struct gfs2_alloc *al = &ip->i_alloc;
 444         struct gfs2_quota_data **qd = al->al_qd;
 445         int error;
 446
 447         if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
 448             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
 449                 return -EIO;
 450
 451         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 452                 return 0;
 453
 454         error = qdsb_get(sdp, QUOTA_USER, ip->i_di.di_uid, CREATE, qd);
 455         if (error)
 456                 goto out;
 457         al->al_qd_num++;
 458         qd++;
 459
 460         error = qdsb_get(sdp, QUOTA_GROUP, ip->i_di.di_gid, CREATE, qd);
 461         if (error)
 462                 goto out;
 463         al->al_qd_num++;
 464         qd++;
 465
 466         if (uid != NO_QUOTA_CHANGE && uid != ip->i_di.di_uid) {
 467                 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
 468                 if (error)
 469                         goto out;
 470                 al->al_qd_num++;
 471                 qd++;
 472         }
 473
 474         if (gid != NO_QUOTA_CHANGE && gid != ip->i_di.di_gid) {
 475                 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
 476                 if (error)
 477                         goto out;
 478                 al->al_qd_num++;
 479                 qd++;
 480         }
 481
 482  out:
 483         if (error)
 484                 gfs2_quota_unhold(ip);
 485
 486         return error;
 487 }
 488
 489 void gfs2_quota_unhold(struct gfs2_inode *ip)
 490 {
 491         struct gfs2_sbd *sdp = ip->i_sbd;
 492         struct gfs2_alloc *al = &ip->i_alloc;
 493         unsigned int x;
 494
 495         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
 496
 497         for (x = 0; x < al->al_qd_num; x++) {
 498                 qdsb_put(al->al_qd[x]);
 499                 al->al_qd[x] = NULL;
 500         }
 501         al->al_qd_num = 0;
 502 }
 503
 504 static int sort_qd(const void *a, const void *b)
 505 {
 506         struct gfs2_quota_data *qd_a = *(struct gfs2_quota_data **)a;
 507         struct gfs2_quota_data *qd_b = *(struct gfs2_quota_data **)b;
 508         int ret = 0;
 509
 510         if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
 511             !test_bit(QDF_USER, &qd_b->qd_flags)) {
 512                 if (test_bit(QDF_USER, &qd_a->qd_flags))
 513                         ret = -1;
 514                 else
 515                         ret = 1;
 516         } else {
 517                 if (qd_a->qd_id < qd_b->qd_id)
 518                         ret = -1;
 519                 else if (qd_a->qd_id > qd_b->qd_id)
 520                         ret = 1;
 521         }
 522
 523         return ret;
 524 }
 525
 526 static void do_qc(struct gfs2_quota_data *qd, int64_t change)
 527 {
 528         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 529         struct gfs2_inode *ip = get_v2ip(sdp->sd_qc_inode);
 530         struct gfs2_quota_change *qc = qd->qd_bh_qc;
 531         int64_t x;
 532
 533         mutex_lock(&sdp->sd_quota_mutex);
 534         gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
 535
 536         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
 537                 qc->qc_change = 0;
 538                 qc->qc_flags = 0;
 539                 if (test_bit(QDF_USER, &qd->qd_flags))
 540                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
 541                 qc->qc_id = cpu_to_be32(qd->qd_id);
 542         }
 543
 544         x = qc->qc_change;
 545         x = be64_to_cpu(x) + change;
 546         qc->qc_change = cpu_to_be64(x);
 547
 548         spin_lock(&sdp->sd_quota_spin);
 549         qd->qd_change = x;
 550         spin_unlock(&sdp->sd_quota_spin);
 551
 552         if (!x) {
 553                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
 554                 clear_bit(QDF_CHANGE, &qd->qd_flags);
 555                 qc->qc_flags = 0;
 556                 qc->qc_id = 0;
 557                 slot_put(qd);
 558                 qd_put(qd);
 559         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
 560                 qd_hold(qd);
 561                 slot_hold(qd);
 562         }
 563
 564         mutex_unlock(&sdp->sd_quota_mutex);
 565 }
 566
 567 /**
 568  * gfs2_adjust_quota
 569  *
 570  * This function was mostly borrowed from gfs2_block_truncate_page which was
 571  * in turn mostly borrowed from ext3
 572  */
 573 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
 574                              int64_t change, struct gfs2_quota_data *qd)
 575 {
 576         struct inode *inode = ip->i_vnode;
 577         struct address_space *mapping = inode->i_mapping;
 578         unsigned long index = loc >> PAGE_CACHE_SHIFT;
 579         unsigned offset = loc & (PAGE_CACHE_SHIFT - 1);
 580         unsigned blocksize, iblock, pos;
 581         struct buffer_head *bh;
 582         struct page *page;
 583         void *kaddr;
 584         __be64 *ptr;
 585         u64 value;
 586         int err = -EIO;
 587
 588         page = grab_cache_page(mapping, index);
 589         if (!page)
 590                 return -ENOMEM;
 591
 592         blocksize = inode->i_sb->s_blocksize;
 593         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 594
 595         if (!page_has_buffers(page))
 596                 create_empty_buffers(page, blocksize, 0);
 597
 598         bh = page_buffers(page);
 599         pos = blocksize;
 600         while (offset >= pos) {
 601                 bh = bh->b_this_page;
 602                 iblock++;
 603                 pos += blocksize;
 604         }
 605
 606         if (!buffer_mapped(bh)) {
 607                 gfs2_get_block(inode, iblock, bh, 1);
 608                 if (!buffer_mapped(bh))
 609                         goto unlock;
 610         }
 611
 612         if (PageUptodate(page))
 613                 set_buffer_uptodate(bh);
 614
 615         if (!buffer_uptodate(bh)) {
 616                 ll_rw_block(READ, 1, &bh);
 617                 wait_on_buffer(bh);
 618                 if (!buffer_uptodate(bh))
 619                         goto unlock;
 620         }
 621
 622         gfs2_trans_add_bh(ip->i_gl, bh, 0);
 623
 624         kaddr = kmap_atomic(page, KM_USER0);
 625         ptr = (__be64 *)(kaddr + offset);
 626         value = *ptr = cpu_to_be64(be64_to_cpu(*ptr) + change);
 627         flush_dcache_page(page);
 628         kunmap_atomic(kaddr, KM_USER0);
 629         err = 0;
 630         qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
 631 #if 0
 632         qd->qd_qb.qb_limit = cpu_to_be64(q.qu_limit);
 633         qd->qd_qb.qb_warn = cpu_to_be64(q.qu_warn);
 634 #endif
 635         qd->qd_qb.qb_value = cpu_to_be64(value);
 636 unlock:
 637         unlock_page(page);
 638         page_cache_release(page);
 639         return err;
 640 }
 641
 642 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 643 {
 644         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
 645         struct gfs2_inode *ip = get_v2ip(sdp->sd_quota_inode);
 646         unsigned int data_blocks, ind_blocks;
 647         struct file_ra_state ra_state;
 648         struct gfs2_holder *ghs, i_gh;
 649         unsigned int qx, x;
 650         struct gfs2_quota_data *qd;
 651         loff_t offset;
 652         unsigned int nalloc = 0;
 653         struct gfs2_alloc *al = NULL;
 654         int error;
 655
 656         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
 657                               &data_blocks, &ind_blocks);
 658
 659         ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
 660         if (!ghs)
 661                 return -ENOMEM;
 662
 663         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
 664         for (qx = 0; qx < num_qd; qx++) {
 665                 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
 666                                            LM_ST_EXCLUSIVE,
 667                                            GL_NOCACHE, &ghs[qx]);
 668                 if (error)
 669                         goto out;
 670         }
 671
 672         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
 673         if (error)
 674                 goto out;
 675
 676         for (x = 0; x < num_qd; x++) {
 677                 int alloc_required;
 678
 679                 offset = qd2offset(qda[x]);
 680                 error = gfs2_write_alloc_required(ip, offset,
 681                                                   sizeof(struct gfs2_quota),
 682                                                   &alloc_required);
 683                 if (error)
 684                         goto out_gunlock;
 685                 if (alloc_required)
 686                         nalloc++;
 687         }
 688
 689         if (nalloc) {
 690                 al = gfs2_alloc_get(ip);
 691
 692                 al->al_requested = nalloc * (data_blocks + ind_blocks);
 693
 694                 error = gfs2_inplace_reserve(ip);
 695                 if (error)
 696                         goto out_alloc;
 697
 698                 error = gfs2_trans_begin(sdp,
 699                                          al->al_rgd->rd_ri.ri_length +
 700                                          num_qd * data_blocks +
 701                                          nalloc * ind_blocks +
 702                                          RES_DINODE + num_qd +
 703                                          RES_STATFS, 0);
 704                 if (error)
 705                         goto out_ipres;
 706         } else {
 707                 error = gfs2_trans_begin(sdp,
 708                                          num_qd * data_blocks +
 709                                          RES_DINODE + num_qd, 0);
 710                 if (error)
 711                         goto out_gunlock;
 712         }
 713
 714         file_ra_state_init(&ra_state, ip->i_vnode->i_mapping);
 715         for (x = 0; x < num_qd; x++) {
 716                 qd = qda[x];
 717                 offset = qd2offset(qd);
 718                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
 719                                           (struct gfs2_quota_data *)
 720                                           qd->qd_gl->gl_lvb);
 721                 if (error)
 722                         goto out_end_trans;
 723
 724                 do_qc(qd, -qd->qd_change_sync);
 725         }
 726
 727         error = 0;
 728
 729  out_end_trans:
 730         gfs2_trans_end(sdp);
 731
 732  out_ipres:
 733         if (nalloc)
 734                 gfs2_inplace_release(ip);
 735
 736  out_alloc:
 737         if (nalloc)
 738                 gfs2_alloc_put(ip);
 739
 740  out_gunlock:
 741         gfs2_glock_dq_uninit(&i_gh);
 742
 743  out:
 744         while (qx--)
 745                 gfs2_glock_dq_uninit(&ghs[qx]);
 746         kfree(ghs);
 747         gfs2_log_flush_glock(ip->i_gl);
 748
 749         return error;
 750 }
 751
 752 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
 753                     struct gfs2_holder *q_gh)
 754 {
 755         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 756         struct gfs2_holder i_gh;
 757         struct gfs2_quota q;
 758         char buf[sizeof(struct gfs2_quota)];
 759         struct file_ra_state ra_state;
 760         int error;
 761
 762         file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
 763  restart:
 764         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
 765         if (error)
 766                 return error;
 767
 768         gfs2_quota_lvb_in(&qd->qd_qb, qd->qd_gl->gl_lvb);
 769
 770         if (force_refresh || qd->qd_qb.qb_magic != GFS2_MAGIC) {
 771                 loff_t pos;
 772                 gfs2_glock_dq_uninit(q_gh);
 773                 error = gfs2_glock_nq_init(qd->qd_gl,
 774                                           LM_ST_EXCLUSIVE, GL_NOCACHE,
 775                                           q_gh);
 776                 if (error)
 777                         return error;
 778
 779                 error = gfs2_glock_nq_init(get_v2ip(sdp->sd_quota_inode)->i_gl,
 780                                           LM_ST_SHARED, 0,
 781                                           &i_gh);
 782                 if (error)
 783                         goto fail;
 784
 785                 memset(buf, 0, sizeof(struct gfs2_quota));
 786                 pos = qd2offset(qd);
 787                 error = gfs2_internal_read(get_v2ip(sdp->sd_quota_inode),
 788                                             &ra_state, buf,
 789                                             &pos,
 790                                             sizeof(struct gfs2_quota));
 791                 if (error < 0)
 792                         goto fail_gunlock;
 793
 794                 gfs2_glock_dq_uninit(&i_gh);
 795
 796                 gfs2_quota_in(&q, buf);
 797
 798                 memset(&qd->qd_qb, 0, sizeof(struct gfs2_quota_lvb));
 799                 qd->qd_qb.qb_magic = GFS2_MAGIC;
 800                 qd->qd_qb.qb_limit = q.qu_limit;
 801                 qd->qd_qb.qb_warn = q.qu_warn;
 802                 qd->qd_qb.qb_value = q.qu_value;
 803
 804                 gfs2_quota_lvb_out(&qd->qd_qb, qd->qd_gl->gl_lvb);
 805
 806                 if (gfs2_glock_is_blocking(qd->qd_gl)) {
 807                         gfs2_glock_dq_uninit(q_gh);
 808                         force_refresh = 0;
 809                         goto restart;
 810                 }
 811         }
 812
 813         return 0;
 814
 815  fail_gunlock:
 816         gfs2_glock_dq_uninit(&i_gh);
 817
 818  fail:
 819         gfs2_glock_dq_uninit(q_gh);
 820
 821         return error;
 822 }
 823
 824 int gfs2_quota_lock(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
 825 {
 826         struct gfs2_sbd *sdp = ip->i_sbd;
 827         struct gfs2_alloc *al = &ip->i_alloc;
 828         unsigned int x;
 829         int error = 0;
 830
 831         gfs2_quota_hold(ip, uid, gid);
 832
 833         if (capable(CAP_SYS_RESOURCE) ||
 834             sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
 835                 return 0;
 836
 837         sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
 838              sort_qd, NULL);
 839
 840         for (x = 0; x < al->al_qd_num; x++) {
 841                 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
 842                 if (error)
 843                         break;
 844         }
 845
 846         if (!error)
 847                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
 848         else {
 849                 while (x--)
 850                         gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
 851                 gfs2_quota_unhold(ip);
 852         }
 853
 854         return error;
 855 }
 856
 857 static int need_sync(struct gfs2_quota_data *qd)
 858 {
 859         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 860         struct gfs2_tune *gt = &sdp->sd_tune;
 861         int64_t value;
 862         unsigned int num, den;
 863         int do_sync = 1;
 864
 865         if (!qd->qd_qb.qb_limit)
 866                 return 0;
 867
 868         spin_lock(&sdp->sd_quota_spin);
 869         value = qd->qd_change;
 870         spin_unlock(&sdp->sd_quota_spin);
 871
 872         spin_lock(&gt->gt_spin);
 873         num = gt->gt_quota_scale_num;
 874         den = gt->gt_quota_scale_den;
 875         spin_unlock(&gt->gt_spin);
 876
 877         if (value < 0)
 878                 do_sync = 0;
 879         else if (qd->qd_qb.qb_value >= (int64_t)qd->qd_qb.qb_limit)
 880                 do_sync = 0;
 881         else {
 882                 value *= gfs2_jindex_size(sdp) * num;
 883                 do_div(value, den);
 884                 value += qd->qd_qb.qb_value;
 885                 if (value < (int64_t)qd->qd_qb.qb_limit)
 886                         do_sync = 0;
 887         }
 888
 889         return do_sync;
 890 }
 891
 892 void gfs2_quota_unlock(struct gfs2_inode *ip)
 893 {
 894         struct gfs2_alloc *al = &ip->i_alloc;
 895         struct gfs2_quota_data *qda[4];
 896         unsigned int count = 0;
 897         unsigned int x;
 898
 899         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
 900                 goto out;
 901
 902         for (x = 0; x < al->al_qd_num; x++) {
 903                 struct gfs2_quota_data *qd;
 904                 int sync;
 905
 906                 qd = al->al_qd[x];
 907                 sync = need_sync(qd);
 908
 909                 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
 910
 911                 if (sync && qd_trylock(qd))
 912                         qda[count++] = qd;
 913         }
 914
 915         if (count) {
 916                 do_sync(count, qda);
 917                 for (x = 0; x < count; x++)
 918                         qd_unlock(qda[x]);
 919         }
 920
 921  out:
 922         gfs2_quota_unhold(ip);
 923 }
 924
 925 #define MAX_LINE 256
 926
 927 static int print_message(struct gfs2_quota_data *qd, char *type)
 928 {
 929         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 930         char *line;
 931         int len;
 932
 933         line = kmalloc(MAX_LINE, GFP_KERNEL);
 934         if (!line)
 935                 return -ENOMEM;
 936
 937         len = snprintf(line, MAX_LINE-1,
 938                        "GFS2: fsid=%s: quota %s for %s %u\r\n",
 939                        sdp->sd_fsname, type,
 940                        (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
 941                        qd->qd_id);
 942         line[MAX_LINE-1] = 0;
 943
 944         if (current->signal) { /* Is this test still required? */
 945                 tty_write_message(current->signal->tty, line);
 946         }
 947
 948         kfree(line);
 949
 950         return 0;
 951 }
 952
 953 int gfs2_quota_check(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
 954 {
 955         struct gfs2_sbd *sdp = ip->i_sbd;
 956         struct gfs2_alloc *al = &ip->i_alloc;
 957         struct gfs2_quota_data *qd;
 958         int64_t value;
 959         unsigned int x;
 960         int error = 0;
 961
 962         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
 963                 return 0;
 964
 965         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
 966                 return 0;
 967
 968         for (x = 0; x < al->al_qd_num; x++) {
 969                 qd = al->al_qd[x];
 970
 971                 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
 972                       (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
 973                         continue;
 974
 975                 value = qd->qd_qb.qb_value;
 976                 spin_lock(&sdp->sd_quota_spin);
 977                 value += qd->qd_change;
 978                 spin_unlock(&sdp->sd_quota_spin);
 979
 980                 if (qd->qd_qb.qb_limit && (int64_t)qd->qd_qb.qb_limit < value) {
 981                         print_message(qd, "exceeded");
 982                         error = -EDQUOT;
 983                         break;
 984                 } else if (qd->qd_qb.qb_warn &&
 985                            (int64_t)qd->qd_qb.qb_warn < value &&
 986                            time_after_eq(jiffies, qd->qd_last_warn +
 987                                          gfs2_tune_get(sdp,
 988                                                 gt_quota_warn_period) * HZ)) {
 989                         error = print_message(qd, "warning");
 990                         qd->qd_last_warn = jiffies;
 991                 }
 992         }
 993
 994         return error;
 995 }
 996
 997 void gfs2_quota_change(struct gfs2_inode *ip, int64_t change,
 998                        uint32_t uid, uint32_t gid)
 999 {
1000         struct gfs2_alloc *al = &ip->i_alloc;
1001         struct gfs2_quota_data *qd;
1002         unsigned int x;
1003         unsigned int found = 0;
1004
1005         if (gfs2_assert_warn(ip->i_sbd, change))
1006                 return;
1007         if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
1008                 return;
1009
1010         for (x = 0; x < al->al_qd_num; x++) {
1011                 qd = al->al_qd[x];
1012
1013                 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1014                     (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1015                         do_qc(qd, change);
1016                         found++;
1017                 }
1018         }
1019 }
1020
1021 int gfs2_quota_sync(struct gfs2_sbd *sdp)
1022 {
1023         struct gfs2_quota_data **qda;
1024         unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1025         unsigned int num_qd;
1026         unsigned int x;
1027         int error = 0;
1028
1029         sdp->sd_quota_sync_gen++;
1030
1031         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1032         if (!qda)
1033                 return -ENOMEM;
1034
1035         do {
1036                 num_qd = 0;
1037
1038                 for (;;) {
1039                         error = qd_fish(sdp, qda + num_qd);
1040                         if (error || !qda[num_qd])
1041                                 break;
1042                         if (++num_qd == max_qd)
1043                                 break;
1044                 }
1045
1046                 if (num_qd) {
1047                         if (!error)
1048                                 error = do_sync(num_qd, qda);
1049                         if (!error)
1050                                 for (x = 0; x < num_qd; x++)
1051                                         qda[x]->qd_sync_gen =
1052                                                 sdp->sd_quota_sync_gen;
1053
1054                         for (x = 0; x < num_qd; x++)
1055                                 qd_unlock(qda[x]);
1056                 }
1057         } while (!error && num_qd == max_qd);
1058
1059         kfree(qda);
1060
1061         return error;
1062 }
1063
1064 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, uint32_t id)
1065 {
1066         struct gfs2_quota_data *qd;
1067         struct gfs2_holder q_gh;
1068         int error;
1069
1070         error = qd_get(sdp, user, id, CREATE, &qd);
1071         if (error)
1072                 return error;
1073
1074         error = do_glock(qd, FORCE, &q_gh);
1075         if (!error)
1076                 gfs2_glock_dq_uninit(&q_gh);
1077
1078         qd_put(qd);
1079
1080         return error;
1081 }
1082
1083 int gfs2_quota_read(struct gfs2_sbd *sdp, int user, uint32_t id,
1084                     struct gfs2_quota *q)
1085 {
1086         struct gfs2_quota_data *qd;
1087         struct gfs2_holder q_gh;
1088         int error;
1089
1090         if (((user) ? (id != current->fsuid) : (!in_group_p(id))) &&
1091             !capable(CAP_SYS_ADMIN))
1092                 return -EACCES;
1093
1094         error = qd_get(sdp, user, id, CREATE, &qd);
1095         if (error)
1096                 return error;
1097
1098         error = do_glock(qd, NO_FORCE, &q_gh);
1099         if (error)
1100                 goto out;
1101
1102         memset(q, 0, sizeof(struct gfs2_quota));
1103         q->qu_limit = qd->qd_qb.qb_limit;
1104         q->qu_warn = qd->qd_qb.qb_warn;
1105         q->qu_value = qd->qd_qb.qb_value;
1106
1107         spin_lock(&sdp->sd_quota_spin);
1108         q->qu_value += qd->qd_change;
1109         spin_unlock(&sdp->sd_quota_spin);
1110
1111         gfs2_glock_dq_uninit(&q_gh);
1112
1113  out:
1114         qd_put(qd);
1115
1116         return error;
1117 }
1118
1119 int gfs2_quota_init(struct gfs2_sbd *sdp)
1120 {
1121         struct gfs2_inode *ip = get_v2ip(sdp->sd_qc_inode);
1122         unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1123         unsigned int x, slot = 0;
1124         unsigned int found = 0;
1125         uint64_t dblock;
1126         uint32_t extlen = 0;
1127         int error;
1128
1129         if (!ip->i_di.di_size ||
1130             ip->i_di.di_size > (64 << 20) ||
1131             ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1132                 gfs2_consist_inode(ip);
1133                 return -EIO;
1134         }
1135         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1136         sdp->sd_quota_chunks = DIV_RU(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1137
1138         error = -ENOMEM;
1139
1140         sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1141                                        sizeof(unsigned char *), GFP_KERNEL);
1142         if (!sdp->sd_quota_bitmap)
1143                 return error;
1144
1145         for (x = 0; x < sdp->sd_quota_chunks; x++) {
1146                 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1147                 if (!sdp->sd_quota_bitmap[x])
1148                         goto fail;
1149         }
1150
1151         for (x = 0; x < blocks; x++) {
1152                 struct buffer_head *bh;
1153                 unsigned int y;
1154
1155                 if (!extlen) {
1156                         int new = 0;
1157                         error = gfs2_block_map(ip, x, &new, &dblock, &extlen);
1158                         if (error)
1159                                 goto fail;
1160                 }
1161                 gfs2_meta_ra(ip->i_gl,  dblock, extlen);
1162                 error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT,
1163                                        &bh);
1164                 if (error)
1165                         goto fail;
1166                 error = -EIO;
1167                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1168                         brelse(bh);
1169                         goto fail;
1170                 }
1171
1172                 for (y = 0;
1173                      y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1174                      y++, slot++) {
1175                         struct gfs2_quota_change qc;
1176                         struct gfs2_quota_data *qd;
1177
1178                         gfs2_quota_change_in(&qc, bh->b_data +
1179                                           sizeof(struct gfs2_meta_header) +
1180                                           y * sizeof(struct gfs2_quota_change));
1181                         if (!qc.qc_change)
1182                                 continue;
1183
1184                         error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1185                                          qc.qc_id, &qd);
1186                         if (error) {
1187                                 brelse(bh);
1188                                 goto fail;
1189                         }
1190
1191                         set_bit(QDF_CHANGE, &qd->qd_flags);
1192                         qd->qd_change = qc.qc_change;
1193                         qd->qd_slot = slot;
1194                         qd->qd_slot_count = 1;
1195                         qd->qd_last_touched = jiffies;
1196
1197                         spin_lock(&sdp->sd_quota_spin);
1198                         gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1199                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1200                         atomic_inc(&sdp->sd_quota_count);
1201                         spin_unlock(&sdp->sd_quota_spin);
1202
1203                         found++;
1204                 }
1205
1206                 brelse(bh);
1207                 dblock++;
1208                 extlen--;
1209         }
1210
1211         if (found)
1212                 fs_info(sdp, "found %u quota changes\n", found);
1213
1214         return 0;
1215
1216  fail:
1217         gfs2_quota_cleanup(sdp);
1218         return error;
1219 }
1220
1221 void gfs2_quota_scan(struct gfs2_sbd *sdp)
1222 {
1223         struct gfs2_quota_data *qd, *safe;
1224         LIST_HEAD(dead);
1225
1226         spin_lock(&sdp->sd_quota_spin);
1227         list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1228                 if (!qd->qd_count &&
1229                     time_after_eq(jiffies, qd->qd_last_touched +
1230                                 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1231                         list_move(&qd->qd_list, &dead);
1232                         gfs2_assert_warn(sdp,
1233                                          atomic_read(&sdp->sd_quota_count) > 0);
1234                         atomic_dec(&sdp->sd_quota_count);
1235                 }
1236         }
1237         spin_unlock(&sdp->sd_quota_spin);
1238
1239         while (!list_empty(&dead)) {
1240                 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1241                 list_del(&qd->qd_list);
1242
1243                 gfs2_assert_warn(sdp, !qd->qd_change);
1244                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1245                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1246
1247                 gfs2_lvb_unhold(qd->qd_gl);
1248                 kfree(qd);
1249         }
1250 }
1251
1252 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1253 {
1254         struct list_head *head = &sdp->sd_quota_list;
1255         struct gfs2_quota_data *qd;
1256         unsigned int x;
1257
1258         spin_lock(&sdp->sd_quota_spin);
1259         while (!list_empty(head)) {
1260                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1261
1262                 if (qd->qd_count > 1 ||
1263                     (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1264                         list_move(&qd->qd_list, head);
1265                         spin_unlock(&sdp->sd_quota_spin);
1266                         schedule();
1267                         spin_lock(&sdp->sd_quota_spin);
1268                         continue;
1269                 }
1270
1271                 list_del(&qd->qd_list);
1272                 atomic_dec(&sdp->sd_quota_count);
1273                 spin_unlock(&sdp->sd_quota_spin);
1274
1275                 if (!qd->qd_count) {
1276                         gfs2_assert_warn(sdp, !qd->qd_change);
1277                         gfs2_assert_warn(sdp, !qd->qd_slot_count);
1278                 } else
1279                         gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1280                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1281
1282                 gfs2_lvb_unhold(qd->qd_gl);
1283                 kfree(qd);
1284
1285                 spin_lock(&sdp->sd_quota_spin);
1286         }
1287         spin_unlock(&sdp->sd_quota_spin);
1288
1289         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1290
1291         if (sdp->sd_quota_bitmap) {
1292                 for (x = 0; x < sdp->sd_quota_chunks; x++)
1293                         kfree(sdp->sd_quota_bitmap[x]);
1294                 kfree(sdp->sd_quota_bitmap);
1295         }
1296 }
1297