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) 2004-2006 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 version 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/sort.h>
  45 #include <linux/fs.h>
  46 #include <linux/gfs2_ondisk.h>
  47
  48 #include "gfs2.h"
  49 #include "lm_interface.h"
  50 #include "incore.h"
  51 #include "bmap.h"
  52 #include "glock.h"
  53 #include "glops.h"
  54 #include "log.h"
  55 #include "meta_io.h"
  56 #include "quota.h"
  57 #include "rgrp.h"
  58 #include "super.h"
  59 #include "trans.h"
  60 #include "inode.h"
  61 #include "ops_file.h"
  62 #include "ops_address.h"
  63 #include "util.h"
  64
  65 #define QUOTA_USER 1
  66 #define QUOTA_GROUP 0
  67
  68 static uint64_t qd2offset(struct gfs2_quota_data *qd)
  69 {
  70         uint64_t offset;
  71
  72         offset = 2 * (uint64_t)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
  73         offset *= sizeof(struct gfs2_quota);
  74
  75         return offset;
  76 }
  77
  78 static int qd_alloc(struct gfs2_sbd *sdp, int user, uint32_t id,
  79                     struct gfs2_quota_data **qdp)
  80 {
  81         struct gfs2_quota_data *qd;
  82         int error;
  83
  84         qd = kzalloc(sizeof(struct gfs2_quota_data), GFP_KERNEL);
  85         if (!qd)
  86                 return -ENOMEM;
  87
  88         qd->qd_count = 1;
  89         qd->qd_id = id;
  90         if (user)
  91                 set_bit(QDF_USER, &qd->qd_flags);
  92         qd->qd_slot = -1;
  93
  94         error = gfs2_glock_get(sdp, 2 * (uint64_t)id + !user,
  95                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
  96         if (error)
  97                 goto fail;
  98
  99         error = gfs2_lvb_hold(qd->qd_gl);
 100         gfs2_glock_put(qd->qd_gl);
 101         if (error)
 102                 goto fail;
 103
 104         *qdp = qd;
 105
 106         return 0;
 107
 108  fail:
 109         kfree(qd);
 110         return error;
 111 }
 112
 113 static int qd_get(struct gfs2_sbd *sdp, int user, uint32_t id, int create,
 114                   struct gfs2_quota_data **qdp)
 115 {
 116         struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
 117         int error, found;
 118
 119         *qdp = NULL;
 120
 121         for (;;) {
 122                 found = 0;
 123                 spin_lock(&sdp->sd_quota_spin);
 124                 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 125                         if (qd->qd_id == id &&
 126                             !test_bit(QDF_USER, &qd->qd_flags) == !user) {
 127                                 qd->qd_count++;
 128                                 found = 1;
 129                                 break;
 130                         }
 131                 }
 132
 133                 if (!found)
 134                         qd = NULL;
 135
 136                 if (!qd && new_qd) {
 137                         qd = new_qd;
 138                         list_add(&qd->qd_list, &sdp->sd_quota_list);
 139                         atomic_inc(&sdp->sd_quota_count);
 140                         new_qd = NULL;
 141                 }
 142
 143                 spin_unlock(&sdp->sd_quota_spin);
 144
 145                 if (qd || !create) {
 146                         if (new_qd) {
 147                                 gfs2_lvb_unhold(new_qd->qd_gl);
 148                                 kfree(new_qd);
 149                         }
 150                         *qdp = qd;
 151                         return 0;
 152                 }
 153
 154                 error = qd_alloc(sdp, user, id, &new_qd);
 155                 if (error)
 156                         return error;
 157         }
 158 }
 159
 160 static void qd_hold(struct gfs2_quota_data *qd)
 161 {
 162         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 163
 164         spin_lock(&sdp->sd_quota_spin);
 165         gfs2_assert(sdp, qd->qd_count);
 166         qd->qd_count++;
 167         spin_unlock(&sdp->sd_quota_spin);
 168 }
 169
 170 static void qd_put(struct gfs2_quota_data *qd)
 171 {
 172         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 173         spin_lock(&sdp->sd_quota_spin);
 174         gfs2_assert(sdp, qd->qd_count);
 175         if (!--qd->qd_count)
 176                 qd->qd_last_touched = jiffies;
 177         spin_unlock(&sdp->sd_quota_spin);
 178 }
 179
 180 static int slot_get(struct gfs2_quota_data *qd)
 181 {
 182         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 183         unsigned int c, o = 0, b;
 184         unsigned char byte = 0;
 185
 186         spin_lock(&sdp->sd_quota_spin);
 187
 188         if (qd->qd_slot_count++) {
 189                 spin_unlock(&sdp->sd_quota_spin);
 190                 return 0;
 191         }
 192
 193         for (c = 0; c < sdp->sd_quota_chunks; c++)
 194                 for (o = 0; o < PAGE_SIZE; o++) {
 195                         byte = sdp->sd_quota_bitmap[c][o];
 196                         if (byte != 0xFF)
 197                                 goto found;
 198                 }
 199
 200         goto fail;
 201
 202  found:
 203         for (b = 0; b < 8; b++)
 204                 if (!(byte & (1 << b)))
 205                         break;
 206         qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
 207
 208         if (qd->qd_slot >= sdp->sd_quota_slots)
 209                 goto fail;
 210
 211         sdp->sd_quota_bitmap[c][o] |= 1 << b;
 212
 213         spin_unlock(&sdp->sd_quota_spin);
 214
 215         return 0;
 216
 217  fail:
 218         qd->qd_slot_count--;
 219         spin_unlock(&sdp->sd_quota_spin);
 220         return -ENOSPC;
 221 }
 222
 223 static void slot_hold(struct gfs2_quota_data *qd)
 224 {
 225         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 226
 227         spin_lock(&sdp->sd_quota_spin);
 228         gfs2_assert(sdp, qd->qd_slot_count);
 229         qd->qd_slot_count++;
 230         spin_unlock(&sdp->sd_quota_spin);
 231 }
 232
 233 static void slot_put(struct gfs2_quota_data *qd)
 234 {
 235         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 236
 237         spin_lock(&sdp->sd_quota_spin);
 238         gfs2_assert(sdp, qd->qd_slot_count);
 239         if (!--qd->qd_slot_count) {
 240                 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
 241                 qd->qd_slot = -1;
 242         }
 243         spin_unlock(&sdp->sd_quota_spin);
 244 }
 245
 246 static int bh_get(struct gfs2_quota_data *qd)
 247 {
 248         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 249         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 250         unsigned int block, offset;
 251         uint64_t dblock;
 252         int new = 0;
 253         struct buffer_head *bh;
 254         int error;
 255         int boundary;
 256
 257         mutex_lock(&sdp->sd_quota_mutex);
 258
 259         if (qd->qd_bh_count++) {
 260                 mutex_unlock(&sdp->sd_quota_mutex);
 261                 return 0;
 262         }
 263
 264         block = qd->qd_slot / sdp->sd_qc_per_block;
 265         offset = qd->qd_slot % sdp->sd_qc_per_block;;
 266
 267         error = gfs2_block_map(&ip->i_inode, block, &new, &dblock, &boundary);
 268         if (error)
 269                 goto fail;
 270         error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT, &bh);
 271         if (error)
 272                 goto fail;
 273         error = -EIO;
 274         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
 275                 goto fail_brelse;
 276
 277         qd->qd_bh = bh;
 278         qd->qd_bh_qc = (struct gfs2_quota_change *)
 279                 (bh->b_data + sizeof(struct gfs2_meta_header) +
 280                  offset * sizeof(struct gfs2_quota_change));
 281
 282         mutex_lock(&sdp->sd_quota_mutex);
 283
 284         return 0;
 285
 286  fail_brelse:
 287         brelse(bh);
 288
 289  fail:
 290         qd->qd_bh_count--;
 291         mutex_unlock(&sdp->sd_quota_mutex);
 292         return error;
 293 }
 294
 295 static void bh_put(struct gfs2_quota_data *qd)
 296 {
 297         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 298
 299         mutex_lock(&sdp->sd_quota_mutex);
 300         gfs2_assert(sdp, qd->qd_bh_count);
 301         if (!--qd->qd_bh_count) {
 302                 brelse(qd->qd_bh);
 303                 qd->qd_bh = NULL;
 304                 qd->qd_bh_qc = NULL;
 305         }
 306         mutex_unlock(&sdp->sd_quota_mutex);
 307 }
 308
 309 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
 310 {
 311         struct gfs2_quota_data *qd = NULL;
 312         int error;
 313         int found = 0;
 314
 315         *qdp = NULL;
 316
 317         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 318                 return 0;
 319
 320         spin_lock(&sdp->sd_quota_spin);
 321
 322         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 323                 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 324                     !test_bit(QDF_CHANGE, &qd->qd_flags) ||
 325                     qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
 326                         continue;
 327
 328                 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 329
 330                 set_bit(QDF_LOCKED, &qd->qd_flags);
 331                 gfs2_assert_warn(sdp, qd->qd_count);
 332                 qd->qd_count++;
 333                 qd->qd_change_sync = qd->qd_change;
 334                 gfs2_assert_warn(sdp, qd->qd_slot_count);
 335                 qd->qd_slot_count++;
 336                 found = 1;
 337
 338                 break;
 339         }
 340
 341         if (!found)
 342                 qd = NULL;
 343
 344         spin_unlock(&sdp->sd_quota_spin);
 345
 346         if (qd) {
 347                 gfs2_assert_warn(sdp, qd->qd_change_sync);
 348                 error = bh_get(qd);
 349                 if (error) {
 350                         clear_bit(QDF_LOCKED, &qd->qd_flags);
 351                         slot_put(qd);
 352                         qd_put(qd);
 353                         return error;
 354                 }
 355         }
 356
 357         *qdp = qd;
 358
 359         return 0;
 360 }
 361
 362 static int qd_trylock(struct gfs2_quota_data *qd)
 363 {
 364         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 365
 366         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 367                 return 0;
 368
 369         spin_lock(&sdp->sd_quota_spin);
 370
 371         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 372             !test_bit(QDF_CHANGE, &qd->qd_flags)) {
 373                 spin_unlock(&sdp->sd_quota_spin);
 374                 return 0;
 375         }
 376
 377         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 378
 379         set_bit(QDF_LOCKED, &qd->qd_flags);
 380         gfs2_assert_warn(sdp, qd->qd_count);
 381         qd->qd_count++;
 382         qd->qd_change_sync = qd->qd_change;
 383         gfs2_assert_warn(sdp, qd->qd_slot_count);
 384         qd->qd_slot_count++;
 385
 386         spin_unlock(&sdp->sd_quota_spin);
 387
 388         gfs2_assert_warn(sdp, qd->qd_change_sync);
 389         if (bh_get(qd)) {
 390                 clear_bit(QDF_LOCKED, &qd->qd_flags);
 391                 slot_put(qd);
 392                 qd_put(qd);
 393                 return 0;
 394         }
 395
 396         return 1;
 397 }
 398
 399 static void qd_unlock(struct gfs2_quota_data *qd)
 400 {
 401         gfs2_assert_warn(qd->qd_gl->gl_sbd,
 402                          test_bit(QDF_LOCKED, &qd->qd_flags));
 403         clear_bit(QDF_LOCKED, &qd->qd_flags);
 404         bh_put(qd);
 405         slot_put(qd);
 406         qd_put(qd);
 407 }
 408
 409 static int qdsb_get(struct gfs2_sbd *sdp, int user, uint32_t id, int create,
 410                     struct gfs2_quota_data **qdp)
 411 {
 412         int error;
 413
 414         error = qd_get(sdp, user, id, create, qdp);
 415         if (error)
 416                 return error;
 417
 418         error = slot_get(*qdp);
 419         if (error)
 420                 goto fail;
 421
 422         error = bh_get(*qdp);
 423         if (error)
 424                 goto fail_slot;
 425
 426         return 0;
 427
 428  fail_slot:
 429         slot_put(*qdp);
 430
 431  fail:
 432         qd_put(*qdp);
 433         return error;
 434 }
 435
 436 static void qdsb_put(struct gfs2_quota_data *qd)
 437 {
 438         bh_put(qd);
 439         slot_put(qd);
 440         qd_put(qd);
 441 }
 442
 443 int gfs2_quota_hold(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
 444 {
 445         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 446         struct gfs2_alloc *al = &ip->i_alloc;
 447         struct gfs2_quota_data **qd = al->al_qd;
 448         int error;
 449
 450         if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
 451             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
 452                 return -EIO;
 453
 454         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 455                 return 0;
 456
 457         error = qdsb_get(sdp, QUOTA_USER, ip->i_di.di_uid, CREATE, qd);
 458         if (error)
 459                 goto out;
 460         al->al_qd_num++;
 461         qd++;
 462
 463         error = qdsb_get(sdp, QUOTA_GROUP, ip->i_di.di_gid, CREATE, qd);
 464         if (error)
 465                 goto out;
 466         al->al_qd_num++;
 467         qd++;
 468
 469         if (uid != NO_QUOTA_CHANGE && uid != ip->i_di.di_uid) {
 470                 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
 471                 if (error)
 472                         goto out;
 473                 al->al_qd_num++;
 474                 qd++;
 475         }
 476
 477         if (gid != NO_QUOTA_CHANGE && gid != ip->i_di.di_gid) {
 478                 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
 479                 if (error)
 480                         goto out;
 481                 al->al_qd_num++;
 482                 qd++;
 483         }
 484
 485  out:
 486         if (error)
 487                 gfs2_quota_unhold(ip);
 488
 489         return error;
 490 }
 491
 492 void gfs2_quota_unhold(struct gfs2_inode *ip)
 493 {
 494         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 495         struct gfs2_alloc *al = &ip->i_alloc;
 496         unsigned int x;
 497
 498         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
 499
 500         for (x = 0; x < al->al_qd_num; x++) {
 501                 qdsb_put(al->al_qd[x]);
 502                 al->al_qd[x] = NULL;
 503         }
 504         al->al_qd_num = 0;
 505 }
 506
 507 static int sort_qd(const void *a, const void *b)
 508 {
 509         struct gfs2_quota_data *qd_a = *(struct gfs2_quota_data **)a;
 510         struct gfs2_quota_data *qd_b = *(struct gfs2_quota_data **)b;
 511         int ret = 0;
 512
 513         if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
 514             !test_bit(QDF_USER, &qd_b->qd_flags)) {
 515                 if (test_bit(QDF_USER, &qd_a->qd_flags))
 516                         ret = -1;
 517                 else
 518                         ret = 1;
 519         } else {
 520                 if (qd_a->qd_id < qd_b->qd_id)
 521                         ret = -1;
 522                 else if (qd_a->qd_id > qd_b->qd_id)
 523                         ret = 1;
 524         }
 525
 526         return ret;
 527 }
 528
 529 static void do_qc(struct gfs2_quota_data *qd, int64_t change)
 530 {
 531         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 532         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 533         struct gfs2_quota_change *qc = qd->qd_bh_qc;
 534         int64_t x;
 535
 536         mutex_lock(&sdp->sd_quota_mutex);
 537         gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
 538
 539         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
 540                 qc->qc_change = 0;
 541                 qc->qc_flags = 0;
 542                 if (test_bit(QDF_USER, &qd->qd_flags))
 543                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
 544                 qc->qc_id = cpu_to_be32(qd->qd_id);
 545         }
 546
 547         x = qc->qc_change;
 548         x = be64_to_cpu(x) + change;
 549         qc->qc_change = cpu_to_be64(x);
 550
 551         spin_lock(&sdp->sd_quota_spin);
 552         qd->qd_change = x;
 553         spin_unlock(&sdp->sd_quota_spin);
 554
 555         if (!x) {
 556                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
 557                 clear_bit(QDF_CHANGE, &qd->qd_flags);
 558                 qc->qc_flags = 0;
 559                 qc->qc_id = 0;
 560                 slot_put(qd);
 561                 qd_put(qd);
 562         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
 563                 qd_hold(qd);
 564                 slot_hold(qd);
 565         }
 566
 567         mutex_unlock(&sdp->sd_quota_mutex);
 568 }
 569
 570 /**
 571  * gfs2_adjust_quota
 572  *
 573  * This function was mostly borrowed from gfs2_block_truncate_page which was
 574  * in turn mostly borrowed from ext3
 575  */
 576 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
 577                              int64_t change, struct gfs2_quota_data *qd)
 578 {
 579         struct inode *inode = &ip->i_inode;
 580         struct address_space *mapping = inode->i_mapping;
 581         unsigned long index = loc >> PAGE_CACHE_SHIFT;
 582         unsigned offset = loc & (PAGE_CACHE_SHIFT - 1);
 583         unsigned blocksize, iblock, pos;
 584         struct buffer_head *bh;
 585         struct page *page;
 586         void *kaddr;
 587         __be64 *ptr;
 588         s64 value;
 589         int err = -EIO;
 590
 591         page = grab_cache_page(mapping, index);
 592         if (!page)
 593                 return -ENOMEM;
 594
 595         blocksize = inode->i_sb->s_blocksize;
 596         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
 597
 598         if (!page_has_buffers(page))
 599                 create_empty_buffers(page, blocksize, 0);
 600
 601         bh = page_buffers(page);
 602         pos = blocksize;
 603         while (offset >= pos) {
 604                 bh = bh->b_this_page;
 605                 iblock++;
 606                 pos += blocksize;
 607         }
 608
 609         if (!buffer_mapped(bh)) {
 610                 gfs2_get_block(inode, iblock, bh, 1);
 611                 if (!buffer_mapped(bh))
 612                         goto unlock;
 613         }
 614
 615         if (PageUptodate(page))
 616                 set_buffer_uptodate(bh);
 617
 618         if (!buffer_uptodate(bh)) {
 619                 ll_rw_block(READ, 1, &bh);
 620                 wait_on_buffer(bh);
 621                 if (!buffer_uptodate(bh))
 622                         goto unlock;
 623         }
 624
 625         gfs2_trans_add_bh(ip->i_gl, bh, 0);
 626
 627         kaddr = kmap_atomic(page, KM_USER0);
 628         ptr = (__be64 *)(kaddr + offset);
 629         value = (s64)be64_to_cpu(*ptr) + change;
 630         *ptr = cpu_to_be64(value);
 631         flush_dcache_page(page);
 632         kunmap_atomic(kaddr, KM_USER0);
 633         err = 0;
 634         qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
 635 #if 0
 636         qd->qd_qb.qb_limit = cpu_to_be64(q.qu_limit);
 637         qd->qd_qb.qb_warn = cpu_to_be64(q.qu_warn);
 638 #endif
 639         qd->qd_qb.qb_value = cpu_to_be64(value);
 640 unlock:
 641         unlock_page(page);
 642         page_cache_release(page);
 643         return err;
 644 }
 645
 646 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 647 {
 648         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
 649         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 650         unsigned int data_blocks, ind_blocks;
 651         struct gfs2_holder *ghs, i_gh;
 652         unsigned int qx, x;
 653         struct gfs2_quota_data *qd;
 654         loff_t offset;
 655         unsigned int nalloc = 0;
 656         struct gfs2_alloc *al = NULL;
 657         int error;
 658
 659         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
 660                               &data_blocks, &ind_blocks);
 661
 662         ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
 663         if (!ghs)
 664                 return -ENOMEM;
 665
 666         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
 667         for (qx = 0; qx < num_qd; qx++) {
 668                 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
 669                                            LM_ST_EXCLUSIVE,
 670                                            GL_NOCACHE, &ghs[qx]);
 671                 if (error)
 672                         goto out;
 673         }
 674
 675         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
 676         if (error)
 677                 goto out;
 678
 679         for (x = 0; x < num_qd; x++) {
 680                 int alloc_required;
 681
 682                 offset = qd2offset(qda[x]);
 683                 error = gfs2_write_alloc_required(ip, offset,
 684                                                   sizeof(struct gfs2_quota),
 685                                                   &alloc_required);
 686                 if (error)
 687                         goto out_gunlock;
 688                 if (alloc_required)
 689                         nalloc++;
 690         }
 691
 692         if (nalloc) {
 693                 al = gfs2_alloc_get(ip);
 694
 695                 al->al_requested = nalloc * (data_blocks + ind_blocks);
 696
 697                 error = gfs2_inplace_reserve(ip);
 698                 if (error)
 699                         goto out_alloc;
 700
 701                 error = gfs2_trans_begin(sdp,
 702                                          al->al_rgd->rd_ri.ri_length +
 703                                          num_qd * data_blocks +
 704                                          nalloc * ind_blocks +
 705                                          RES_DINODE + num_qd +
 706                                          RES_STATFS, 0);
 707                 if (error)
 708                         goto out_ipres;
 709         } else {
 710                 error = gfs2_trans_begin(sdp,
 711                                          num_qd * data_blocks +
 712                                          RES_DINODE + num_qd, 0);
 713                 if (error)
 714                         goto out_gunlock;
 715         }
 716
 717         for (x = 0; x < num_qd; x++) {
 718                 qd = qda[x];
 719                 offset = qd2offset(qd);
 720                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
 721                                           (struct gfs2_quota_data *)
 722                                           qd->qd_gl->gl_lvb);
 723                 if (error)
 724                         goto out_end_trans;
 725
 726                 do_qc(qd, -qd->qd_change_sync);
 727         }
 728
 729         error = 0;
 730
 731  out_end_trans:
 732         gfs2_trans_end(sdp);
 733
 734  out_ipres:
 735         if (nalloc)
 736                 gfs2_inplace_release(ip);
 737
 738  out_alloc:
 739         if (nalloc)
 740                 gfs2_alloc_put(ip);
 741
 742  out_gunlock:
 743         gfs2_glock_dq_uninit(&i_gh);
 744
 745  out:
 746         while (qx--)
 747                 gfs2_glock_dq_uninit(&ghs[qx]);
 748         kfree(ghs);
 749         gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
 750
 751         return error;
 752 }
 753
 754 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
 755                     struct gfs2_holder *q_gh)
 756 {
 757         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 758         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 759         struct gfs2_holder i_gh;
 760         struct gfs2_quota q;
 761         char buf[sizeof(struct gfs2_quota)];
 762         struct file_ra_state ra_state;
 763         int error;
 764         struct gfs2_quota_lvb *qlvb;
 765
 766         file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
 767  restart:
 768         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
 769         if (error)
 770                 return error;
 771
 772         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
 773
 774         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
 775                 loff_t pos;
 776                 gfs2_glock_dq_uninit(q_gh);
 777                 error = gfs2_glock_nq_init(qd->qd_gl,
 778                                           LM_ST_EXCLUSIVE, GL_NOCACHE,
 779                                           q_gh);
 780                 if (error)
 781                         return error;
 782
 783                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
 784                 if (error)
 785                         goto fail;
 786
 787                 memset(buf, 0, sizeof(struct gfs2_quota));
 788                 pos = qd2offset(qd);
 789                 error = gfs2_internal_read(ip, &ra_state, buf,
 790                                            &pos, sizeof(struct gfs2_quota));
 791                 if (error < 0)
 792                         goto fail_gunlock;
 793
 794                 gfs2_glock_dq_uninit(&i_gh);
 795
 796
 797                 gfs2_quota_in(&q, buf);
 798                 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
 799                 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
 800                 qlvb->__pad = 0;
 801                 qlvb->qb_limit = cpu_to_be64(q.qu_limit);
 802                 qlvb->qb_warn = cpu_to_be64(q.qu_warn);
 803                 qlvb->qb_value = cpu_to_be64(q.qu_value);
 804                 qd->qd_qb = *qlvb;
 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 = GFS2_SB(&ip->i_inode);
 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 ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
 880                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
 881                 do_sync = 0;
 882         else {
 883                 value *= gfs2_jindex_size(sdp) * num;
 884                 do_div(value, den);
 885                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
 886                 if (value < (int64_t)be64_to_cpu(qd->qd_qb.qb_limit))
 887                         do_sync = 0;
 888         }
 889
 890         return do_sync;
 891 }
 892
 893 void gfs2_quota_unlock(struct gfs2_inode *ip)
 894 {
 895         struct gfs2_alloc *al = &ip->i_alloc;
 896         struct gfs2_quota_data *qda[4];
 897         unsigned int count = 0;
 898         unsigned int x;
 899
 900         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
 901                 goto out;
 902
 903         for (x = 0; x < al->al_qd_num; x++) {
 904                 struct gfs2_quota_data *qd;
 905                 int sync;
 906
 907                 qd = al->al_qd[x];
 908                 sync = need_sync(qd);
 909
 910                 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
 911
 912                 if (sync && qd_trylock(qd))
 913                         qda[count++] = qd;
 914         }
 915
 916         if (count) {
 917                 do_sync(count, qda);
 918                 for (x = 0; x < count; x++)
 919                         qd_unlock(qda[x]);
 920         }
 921
 922  out:
 923         gfs2_quota_unhold(ip);
 924 }
 925
 926 #define MAX_LINE 256
 927
 928 static int print_message(struct gfs2_quota_data *qd, char *type)
 929 {
 930         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 931
 932         printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
 933                sdp->sd_fsname, type,
 934                (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
 935                qd->qd_id);
 936
 937         return 0;
 938 }
 939
 940 int gfs2_quota_check(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
 941 {
 942         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 943         struct gfs2_alloc *al = &ip->i_alloc;
 944         struct gfs2_quota_data *qd;
 945         int64_t value;
 946         unsigned int x;
 947         int error = 0;
 948
 949         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
 950                 return 0;
 951
 952         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
 953                 return 0;
 954
 955         for (x = 0; x < al->al_qd_num; x++) {
 956                 qd = al->al_qd[x];
 957
 958                 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
 959                       (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
 960                         continue;
 961
 962                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
 963                 spin_lock(&sdp->sd_quota_spin);
 964                 value += qd->qd_change;
 965                 spin_unlock(&sdp->sd_quota_spin);
 966
 967                 if (be64_to_cpu(qd->qd_qb.qb_limit) && (int64_t)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
 968                         print_message(qd, "exceeded");
 969                         error = -EDQUOT;
 970                         break;
 971                 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
 972                            (int64_t)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
 973                            time_after_eq(jiffies, qd->qd_last_warn +
 974                                          gfs2_tune_get(sdp,
 975                                                 gt_quota_warn_period) * HZ)) {
 976                         error = print_message(qd, "warning");
 977                         qd->qd_last_warn = jiffies;
 978                 }
 979         }
 980
 981         return error;
 982 }
 983
 984 void gfs2_quota_change(struct gfs2_inode *ip, int64_t change,
 985                        uint32_t uid, uint32_t gid)
 986 {
 987         struct gfs2_alloc *al = &ip->i_alloc;
 988         struct gfs2_quota_data *qd;
 989         unsigned int x;
 990         unsigned int found = 0;
 991
 992         if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
 993                 return;
 994         if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
 995                 return;
 996
 997         for (x = 0; x < al->al_qd_num; x++) {
 998                 qd = al->al_qd[x];
 999
1000                 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1001                     (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1002                         do_qc(qd, change);
1003                         found++;
1004                 }
1005         }
1006 }
1007
1008 int gfs2_quota_sync(struct gfs2_sbd *sdp)
1009 {
1010         struct gfs2_quota_data **qda;
1011         unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1012         unsigned int num_qd;
1013         unsigned int x;
1014         int error = 0;
1015
1016         sdp->sd_quota_sync_gen++;
1017
1018         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1019         if (!qda)
1020                 return -ENOMEM;
1021
1022         do {
1023                 num_qd = 0;
1024
1025                 for (;;) {
1026                         error = qd_fish(sdp, qda + num_qd);
1027                         if (error || !qda[num_qd])
1028                                 break;
1029                         if (++num_qd == max_qd)
1030                                 break;
1031                 }
1032
1033                 if (num_qd) {
1034                         if (!error)
1035                                 error = do_sync(num_qd, qda);
1036                         if (!error)
1037                                 for (x = 0; x < num_qd; x++)
1038                                         qda[x]->qd_sync_gen =
1039                                                 sdp->sd_quota_sync_gen;
1040
1041                         for (x = 0; x < num_qd; x++)
1042                                 qd_unlock(qda[x]);
1043                 }
1044         } while (!error && num_qd == max_qd);
1045
1046         kfree(qda);
1047
1048         return error;
1049 }
1050
1051 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, uint32_t id)
1052 {
1053         struct gfs2_quota_data *qd;
1054         struct gfs2_holder q_gh;
1055         int error;
1056
1057         error = qd_get(sdp, user, id, CREATE, &qd);
1058         if (error)
1059                 return error;
1060
1061         error = do_glock(qd, FORCE, &q_gh);
1062         if (!error)
1063                 gfs2_glock_dq_uninit(&q_gh);
1064
1065         qd_put(qd);
1066
1067         return error;
1068 }
1069
1070 #if 0
1071 int gfs2_quota_read(struct gfs2_sbd *sdp, int user, uint32_t id,
1072                     struct gfs2_quota *q)
1073 {
1074         struct gfs2_quota_data *qd;
1075         struct gfs2_holder q_gh;
1076         int error;
1077
1078         if (((user) ? (id != current->fsuid) : (!in_group_p(id))) &&
1079             !capable(CAP_SYS_ADMIN))
1080                 return -EACCES;
1081
1082         error = qd_get(sdp, user, id, CREATE, &qd);
1083         if (error)
1084                 return error;
1085
1086         error = do_glock(qd, NO_FORCE, &q_gh);
1087         if (error)
1088                 goto out;
1089
1090         memset(q, 0, sizeof(struct gfs2_quota));
1091         q->qu_limit = be64_to_cpu(qd->qd_qb.qb_limit);
1092         q->qu_warn = be64_to_cpu(qd->qd_qb.qb_warn);
1093         q->qu_value = be64_to_cpu(qd->qd_qb.qb_value);
1094
1095         spin_lock(&sdp->sd_quota_spin);
1096         q->qu_value += qd->qd_change;
1097         spin_unlock(&sdp->sd_quota_spin);
1098
1099         gfs2_glock_dq_uninit(&q_gh);
1100
1101  out:
1102         qd_put(qd);
1103
1104         return error;
1105 }
1106 #endif  /*  0  */
1107
1108 int gfs2_quota_init(struct gfs2_sbd *sdp)
1109 {
1110         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1111         unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1112         unsigned int x, slot = 0;
1113         unsigned int found = 0;
1114         uint64_t dblock;
1115         uint32_t extlen = 0;
1116         int error;
1117
1118         if (!ip->i_di.di_size ||
1119             ip->i_di.di_size > (64 << 20) ||
1120             ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1121                 gfs2_consist_inode(ip);
1122                 return -EIO;
1123         }
1124         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1125         sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1126
1127         error = -ENOMEM;
1128
1129         sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1130                                        sizeof(unsigned char *), GFP_KERNEL);
1131         if (!sdp->sd_quota_bitmap)
1132                 return error;
1133
1134         for (x = 0; x < sdp->sd_quota_chunks; x++) {
1135                 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1136                 if (!sdp->sd_quota_bitmap[x])
1137                         goto fail;
1138         }
1139
1140         for (x = 0; x < blocks; x++) {
1141                 struct buffer_head *bh;
1142                 unsigned int y;
1143
1144                 if (!extlen) {
1145                         int new = 0;
1146                         error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1147                         if (error)
1148                                 goto fail;
1149                 }
1150                 gfs2_meta_ra(ip->i_gl,  dblock, extlen);
1151                 error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT,
1152                                        &bh);
1153                 if (error)
1154                         goto fail;
1155                 error = -EIO;
1156                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1157                         brelse(bh);
1158                         goto fail;
1159                 }
1160
1161                 for (y = 0;
1162                      y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1163                      y++, slot++) {
1164                         struct gfs2_quota_change qc;
1165                         struct gfs2_quota_data *qd;
1166
1167                         gfs2_quota_change_in(&qc, bh->b_data +
1168                                           sizeof(struct gfs2_meta_header) +
1169                                           y * sizeof(struct gfs2_quota_change));
1170                         if (!qc.qc_change)
1171                                 continue;
1172
1173                         error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1174                                          qc.qc_id, &qd);
1175                         if (error) {
1176                                 brelse(bh);
1177                                 goto fail;
1178                         }
1179
1180                         set_bit(QDF_CHANGE, &qd->qd_flags);
1181                         qd->qd_change = qc.qc_change;
1182                         qd->qd_slot = slot;
1183                         qd->qd_slot_count = 1;
1184                         qd->qd_last_touched = jiffies;
1185
1186                         spin_lock(&sdp->sd_quota_spin);
1187                         gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1188                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1189                         atomic_inc(&sdp->sd_quota_count);
1190                         spin_unlock(&sdp->sd_quota_spin);
1191
1192                         found++;
1193                 }
1194
1195                 brelse(bh);
1196                 dblock++;
1197                 extlen--;
1198         }
1199
1200         if (found)
1201                 fs_info(sdp, "found %u quota changes\n", found);
1202
1203         return 0;
1204
1205  fail:
1206         gfs2_quota_cleanup(sdp);
1207         return error;
1208 }
1209
1210 void gfs2_quota_scan(struct gfs2_sbd *sdp)
1211 {
1212         struct gfs2_quota_data *qd, *safe;
1213         LIST_HEAD(dead);
1214
1215         spin_lock(&sdp->sd_quota_spin);
1216         list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1217                 if (!qd->qd_count &&
1218                     time_after_eq(jiffies, qd->qd_last_touched +
1219                                 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1220                         list_move(&qd->qd_list, &dead);
1221                         gfs2_assert_warn(sdp,
1222                                          atomic_read(&sdp->sd_quota_count) > 0);
1223                         atomic_dec(&sdp->sd_quota_count);
1224                 }
1225         }
1226         spin_unlock(&sdp->sd_quota_spin);
1227
1228         while (!list_empty(&dead)) {
1229                 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1230                 list_del(&qd->qd_list);
1231
1232                 gfs2_assert_warn(sdp, !qd->qd_change);
1233                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1234                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1235
1236                 gfs2_lvb_unhold(qd->qd_gl);
1237                 kfree(qd);
1238         }
1239 }
1240
1241 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1242 {
1243         struct list_head *head = &sdp->sd_quota_list;
1244         struct gfs2_quota_data *qd;
1245         unsigned int x;
1246
1247         spin_lock(&sdp->sd_quota_spin);
1248         while (!list_empty(head)) {
1249                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1250
1251                 if (qd->qd_count > 1 ||
1252                     (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1253                         list_move(&qd->qd_list, head);
1254                         spin_unlock(&sdp->sd_quota_spin);
1255                         schedule();
1256                         spin_lock(&sdp->sd_quota_spin);
1257                         continue;
1258                 }
1259
1260                 list_del(&qd->qd_list);
1261                 atomic_dec(&sdp->sd_quota_count);
1262                 spin_unlock(&sdp->sd_quota_spin);
1263
1264                 if (!qd->qd_count) {
1265                         gfs2_assert_warn(sdp, !qd->qd_change);
1266                         gfs2_assert_warn(sdp, !qd->qd_slot_count);
1267                 } else
1268                         gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1269                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1270
1271                 gfs2_lvb_unhold(qd->qd_gl);
1272                 kfree(qd);
1273
1274                 spin_lock(&sdp->sd_quota_spin);
1275         }
1276         spin_unlock(&sdp->sd_quota_spin);
1277
1278         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1279
1280         if (sdp->sd_quota_bitmap) {
1281                 for (x = 0; x < sdp->sd_quota_chunks; x++)
1282                         kfree(sdp->sd_quota_bitmap[x]);
1283                 kfree(sdp->sd_quota_bitmap);
1284         }
1285 }
1286