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 u64 qd2offset(struct gfs2_quota_data *qd)
  69 {
  70         u64 offset;
  71
  72         offset = 2 * (u64)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, u32 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 * (u64)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, u32 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         u64 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 fail:
 289         qd->qd_bh_count--;
 290         mutex_unlock(&sdp->sd_quota_mutex);
 291         return error;
 292 }
 293
 294 static void bh_put(struct gfs2_quota_data *qd)
 295 {
 296         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 297
 298         mutex_lock(&sdp->sd_quota_mutex);
 299         gfs2_assert(sdp, qd->qd_bh_count);
 300         if (!--qd->qd_bh_count) {
 301                 brelse(qd->qd_bh);
 302                 qd->qd_bh = NULL;
 303                 qd->qd_bh_qc = NULL;
 304         }
 305         mutex_unlock(&sdp->sd_quota_mutex);
 306 }
 307
 308 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
 309 {
 310         struct gfs2_quota_data *qd = NULL;
 311         int error;
 312         int found = 0;
 313
 314         *qdp = NULL;
 315
 316         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 317                 return 0;
 318
 319         spin_lock(&sdp->sd_quota_spin);
 320
 321         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
 322                 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 323                     !test_bit(QDF_CHANGE, &qd->qd_flags) ||
 324                     qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
 325                         continue;
 326
 327                 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 328
 329                 set_bit(QDF_LOCKED, &qd->qd_flags);
 330                 gfs2_assert_warn(sdp, qd->qd_count);
 331                 qd->qd_count++;
 332                 qd->qd_change_sync = qd->qd_change;
 333                 gfs2_assert_warn(sdp, qd->qd_slot_count);
 334                 qd->qd_slot_count++;
 335                 found = 1;
 336
 337                 break;
 338         }
 339
 340         if (!found)
 341                 qd = NULL;
 342
 343         spin_unlock(&sdp->sd_quota_spin);
 344
 345         if (qd) {
 346                 gfs2_assert_warn(sdp, qd->qd_change_sync);
 347                 error = bh_get(qd);
 348                 if (error) {
 349                         clear_bit(QDF_LOCKED, &qd->qd_flags);
 350                         slot_put(qd);
 351                         qd_put(qd);
 352                         return error;
 353                 }
 354         }
 355
 356         *qdp = qd;
 357
 358         return 0;
 359 }
 360
 361 static int qd_trylock(struct gfs2_quota_data *qd)
 362 {
 363         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 364
 365         if (sdp->sd_vfs->s_flags & MS_RDONLY)
 366                 return 0;
 367
 368         spin_lock(&sdp->sd_quota_spin);
 369
 370         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
 371             !test_bit(QDF_CHANGE, &qd->qd_flags)) {
 372                 spin_unlock(&sdp->sd_quota_spin);
 373                 return 0;
 374         }
 375
 376         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 377
 378         set_bit(QDF_LOCKED, &qd->qd_flags);
 379         gfs2_assert_warn(sdp, qd->qd_count);
 380         qd->qd_count++;
 381         qd->qd_change_sync = qd->qd_change;
 382         gfs2_assert_warn(sdp, qd->qd_slot_count);
 383         qd->qd_slot_count++;
 384
 385         spin_unlock(&sdp->sd_quota_spin);
 386
 387         gfs2_assert_warn(sdp, qd->qd_change_sync);
 388         if (bh_get(qd)) {
 389                 clear_bit(QDF_LOCKED, &qd->qd_flags);
 390                 slot_put(qd);
 391                 qd_put(qd);
 392                 return 0;
 393         }
 394
 395         return 1;
 396 }
 397
 398 static void qd_unlock(struct gfs2_quota_data *qd)
 399 {
 400         gfs2_assert_warn(qd->qd_gl->gl_sbd,
 401                          test_bit(QDF_LOCKED, &qd->qd_flags));
 402         clear_bit(QDF_LOCKED, &qd->qd_flags);
 403         bh_put(qd);
 404         slot_put(qd);
 405         qd_put(qd);
 406 }
 407
 408 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
 409                     struct gfs2_quota_data **qdp)
 410 {
 411         int error;
 412
 413         error = qd_get(sdp, user, id, create, qdp);
 414         if (error)
 415                 return error;
 416
 417         error = slot_get(*qdp);
 418         if (error)
 419                 goto fail;
 420
 421         error = bh_get(*qdp);
 422         if (error)
 423                 goto fail_slot;
 424
 425         return 0;
 426
 427 fail_slot:
 428         slot_put(*qdp);
 429 fail:
 430         qd_put(*qdp);
 431         return error;
 432 }
 433
 434 static void qdsb_put(struct gfs2_quota_data *qd)
 435 {
 436         bh_put(qd);
 437         slot_put(qd);
 438         qd_put(qd);
 439 }
 440
 441 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
 442 {
 443         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 444         struct gfs2_alloc *al = &ip->i_alloc;
 445         struct gfs2_quota_data **qd = al->al_qd;
 446         int error;
 447
 448         if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
 449             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
 450                 return -EIO;
 451
 452         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 453                 return 0;
 454
 455         error = qdsb_get(sdp, QUOTA_USER, ip->i_di.di_uid, CREATE, qd);
 456         if (error)
 457                 goto out;
 458         al->al_qd_num++;
 459         qd++;
 460
 461         error = qdsb_get(sdp, QUOTA_GROUP, ip->i_di.di_gid, CREATE, qd);
 462         if (error)
 463                 goto out;
 464         al->al_qd_num++;
 465         qd++;
 466
 467         if (uid != NO_QUOTA_CHANGE && uid != ip->i_di.di_uid) {
 468                 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
 469                 if (error)
 470                         goto out;
 471                 al->al_qd_num++;
 472                 qd++;
 473         }
 474
 475         if (gid != NO_QUOTA_CHANGE && gid != ip->i_di.di_gid) {
 476                 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
 477                 if (error)
 478                         goto out;
 479                 al->al_qd_num++;
 480                 qd++;
 481         }
 482
 483 out:
 484         if (error)
 485                 gfs2_quota_unhold(ip);
 486         return error;
 487 }
 488
 489 void gfs2_quota_unhold(struct gfs2_inode *ip)
 490 {
 491         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 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, s64 change)
 527 {
 528         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 529         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
 530         struct gfs2_quota_change *qc = qd->qd_bh_qc;
 531         s64 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                              s64 change, struct gfs2_quota_data *qd)
 575 {
 576         struct inode *inode = &ip->i_inode;
 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         s64 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 = (s64)be64_to_cpu(*ptr) + change;
 627         *ptr = cpu_to_be64(value);
 628         flush_dcache_page(page);
 629         kunmap_atomic(kaddr, KM_USER0);
 630         err = 0;
 631         qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
 632 #if 0
 633         qd->qd_qb.qb_limit = cpu_to_be64(q.qu_limit);
 634         qd->qd_qb.qb_warn = cpu_to_be64(q.qu_warn);
 635 #endif
 636         qd->qd_qb.qb_value = cpu_to_be64(value);
 637 unlock:
 638         unlock_page(page);
 639         page_cache_release(page);
 640         return err;
 641 }
 642
 643 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 644 {
 645         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
 646         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 647         unsigned int data_blocks, ind_blocks;
 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         for (x = 0; x < num_qd; x++) {
 715                 qd = qda[x];
 716                 offset = qd2offset(qd);
 717                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
 718                                           (struct gfs2_quota_data *)
 719                                           qd->qd_gl->gl_lvb);
 720                 if (error)
 721                         goto out_end_trans;
 722
 723                 do_qc(qd, -qd->qd_change_sync);
 724         }
 725
 726         error = 0;
 727
 728 out_end_trans:
 729         gfs2_trans_end(sdp);
 730 out_ipres:
 731         if (nalloc)
 732                 gfs2_inplace_release(ip);
 733 out_alloc:
 734         if (nalloc)
 735                 gfs2_alloc_put(ip);
 736 out_gunlock:
 737         gfs2_glock_dq_uninit(&i_gh);
 738 out:
 739         while (qx--)
 740                 gfs2_glock_dq_uninit(&ghs[qx]);
 741         kfree(ghs);
 742         gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
 743         return error;
 744 }
 745
 746 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
 747                     struct gfs2_holder *q_gh)
 748 {
 749         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 750         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
 751         struct gfs2_holder i_gh;
 752         struct gfs2_quota q;
 753         char buf[sizeof(struct gfs2_quota)];
 754         struct file_ra_state ra_state;
 755         int error;
 756         struct gfs2_quota_lvb *qlvb;
 757
 758         file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
 759 restart:
 760         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
 761         if (error)
 762                 return error;
 763
 764         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
 765
 766         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
 767                 loff_t pos;
 768                 gfs2_glock_dq_uninit(q_gh);
 769                 error = gfs2_glock_nq_init(qd->qd_gl,
 770                                           LM_ST_EXCLUSIVE, GL_NOCACHE,
 771                                           q_gh);
 772                 if (error)
 773                         return error;
 774
 775                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
 776                 if (error)
 777                         goto fail;
 778
 779                 memset(buf, 0, sizeof(struct gfs2_quota));
 780                 pos = qd2offset(qd);
 781                 error = gfs2_internal_read(ip, &ra_state, buf,
 782                                            &pos, sizeof(struct gfs2_quota));
 783                 if (error < 0)
 784                         goto fail_gunlock;
 785
 786                 gfs2_glock_dq_uninit(&i_gh);
 787
 788
 789                 gfs2_quota_in(&q, buf);
 790                 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
 791                 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
 792                 qlvb->__pad = 0;
 793                 qlvb->qb_limit = cpu_to_be64(q.qu_limit);
 794                 qlvb->qb_warn = cpu_to_be64(q.qu_warn);
 795                 qlvb->qb_value = cpu_to_be64(q.qu_value);
 796                 qd->qd_qb = *qlvb;
 797
 798                 if (gfs2_glock_is_blocking(qd->qd_gl)) {
 799                         gfs2_glock_dq_uninit(q_gh);
 800                         force_refresh = 0;
 801                         goto restart;
 802                 }
 803         }
 804
 805         return 0;
 806
 807 fail_gunlock:
 808         gfs2_glock_dq_uninit(&i_gh);
 809 fail:
 810         gfs2_glock_dq_uninit(q_gh);
 811         return error;
 812 }
 813
 814 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
 815 {
 816         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 817         struct gfs2_alloc *al = &ip->i_alloc;
 818         unsigned int x;
 819         int error = 0;
 820
 821         gfs2_quota_hold(ip, uid, gid);
 822
 823         if (capable(CAP_SYS_RESOURCE) ||
 824             sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
 825                 return 0;
 826
 827         sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
 828              sort_qd, NULL);
 829
 830         for (x = 0; x < al->al_qd_num; x++) {
 831                 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
 832                 if (error)
 833                         break;
 834         }
 835
 836         if (!error)
 837                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
 838         else {
 839                 while (x--)
 840                         gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
 841                 gfs2_quota_unhold(ip);
 842         }
 843
 844         return error;
 845 }
 846
 847 static int need_sync(struct gfs2_quota_data *qd)
 848 {
 849         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 850         struct gfs2_tune *gt = &sdp->sd_tune;
 851         s64 value;
 852         unsigned int num, den;
 853         int do_sync = 1;
 854
 855         if (!qd->qd_qb.qb_limit)
 856                 return 0;
 857
 858         spin_lock(&sdp->sd_quota_spin);
 859         value = qd->qd_change;
 860         spin_unlock(&sdp->sd_quota_spin);
 861
 862         spin_lock(&gt->gt_spin);
 863         num = gt->gt_quota_scale_num;
 864         den = gt->gt_quota_scale_den;
 865         spin_unlock(&gt->gt_spin);
 866
 867         if (value < 0)
 868                 do_sync = 0;
 869         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
 870                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
 871                 do_sync = 0;
 872         else {
 873                 value *= gfs2_jindex_size(sdp) * num;
 874                 do_div(value, den);
 875                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
 876                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
 877                         do_sync = 0;
 878         }
 879
 880         return do_sync;
 881 }
 882
 883 void gfs2_quota_unlock(struct gfs2_inode *ip)
 884 {
 885         struct gfs2_alloc *al = &ip->i_alloc;
 886         struct gfs2_quota_data *qda[4];
 887         unsigned int count = 0;
 888         unsigned int x;
 889
 890         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
 891                 goto out;
 892
 893         for (x = 0; x < al->al_qd_num; x++) {
 894                 struct gfs2_quota_data *qd;
 895                 int sync;
 896
 897                 qd = al->al_qd[x];
 898                 sync = need_sync(qd);
 899
 900                 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
 901
 902                 if (sync && qd_trylock(qd))
 903                         qda[count++] = qd;
 904         }
 905
 906         if (count) {
 907                 do_sync(count, qda);
 908                 for (x = 0; x < count; x++)
 909                         qd_unlock(qda[x]);
 910         }
 911
 912 out:
 913         gfs2_quota_unhold(ip);
 914 }
 915
 916 #define MAX_LINE 256
 917
 918 static int print_message(struct gfs2_quota_data *qd, char *type)
 919 {
 920         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
 921
 922         printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
 923                sdp->sd_fsname, type,
 924                (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
 925                qd->qd_id);
 926
 927         return 0;
 928 }
 929
 930 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
 931 {
 932         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 933         struct gfs2_alloc *al = &ip->i_alloc;
 934         struct gfs2_quota_data *qd;
 935         s64 value;
 936         unsigned int x;
 937         int error = 0;
 938
 939         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
 940                 return 0;
 941
 942         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
 943                 return 0;
 944
 945         for (x = 0; x < al->al_qd_num; x++) {
 946                 qd = al->al_qd[x];
 947
 948                 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
 949                       (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
 950                         continue;
 951
 952                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
 953                 spin_lock(&sdp->sd_quota_spin);
 954                 value += qd->qd_change;
 955                 spin_unlock(&sdp->sd_quota_spin);
 956
 957                 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
 958                         print_message(qd, "exceeded");
 959                         error = -EDQUOT;
 960                         break;
 961                 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
 962                            (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
 963                            time_after_eq(jiffies, qd->qd_last_warn +
 964                                          gfs2_tune_get(sdp,
 965                                                 gt_quota_warn_period) * HZ)) {
 966                         error = print_message(qd, "warning");
 967                         qd->qd_last_warn = jiffies;
 968                 }
 969         }
 970
 971         return error;
 972 }
 973
 974 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
 975                        u32 uid, u32 gid)
 976 {
 977         struct gfs2_alloc *al = &ip->i_alloc;
 978         struct gfs2_quota_data *qd;
 979         unsigned int x;
 980         unsigned int found = 0;
 981
 982         if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
 983                 return;
 984         if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
 985                 return;
 986
 987         for (x = 0; x < al->al_qd_num; x++) {
 988                 qd = al->al_qd[x];
 989
 990                 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
 991                     (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
 992                         do_qc(qd, change);
 993                         found++;
 994                 }
 995         }
 996 }
 997
 998 int gfs2_quota_sync(struct gfs2_sbd *sdp)
 999 {
1000         struct gfs2_quota_data **qda;
1001         unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1002         unsigned int num_qd;
1003         unsigned int x;
1004         int error = 0;
1005
1006         sdp->sd_quota_sync_gen++;
1007
1008         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1009         if (!qda)
1010                 return -ENOMEM;
1011
1012         do {
1013                 num_qd = 0;
1014
1015                 for (;;) {
1016                         error = qd_fish(sdp, qda + num_qd);
1017                         if (error || !qda[num_qd])
1018                                 break;
1019                         if (++num_qd == max_qd)
1020                                 break;
1021                 }
1022
1023                 if (num_qd) {
1024                         if (!error)
1025                                 error = do_sync(num_qd, qda);
1026                         if (!error)
1027                                 for (x = 0; x < num_qd; x++)
1028                                         qda[x]->qd_sync_gen =
1029                                                 sdp->sd_quota_sync_gen;
1030
1031                         for (x = 0; x < num_qd; x++)
1032                                 qd_unlock(qda[x]);
1033                 }
1034         } while (!error && num_qd == max_qd);
1035
1036         kfree(qda);
1037
1038         return error;
1039 }
1040
1041 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1042 {
1043         struct gfs2_quota_data *qd;
1044         struct gfs2_holder q_gh;
1045         int error;
1046
1047         error = qd_get(sdp, user, id, CREATE, &qd);
1048         if (error)
1049                 return error;
1050
1051         error = do_glock(qd, FORCE, &q_gh);
1052         if (!error)
1053                 gfs2_glock_dq_uninit(&q_gh);
1054
1055         qd_put(qd);
1056
1057         return error;
1058 }
1059
1060 #if 0
1061 int gfs2_quota_read(struct gfs2_sbd *sdp, int user, u32 id,
1062                     struct gfs2_quota *q)
1063 {
1064         struct gfs2_quota_data *qd;
1065         struct gfs2_holder q_gh;
1066         int error;
1067
1068         if (((user) ? (id != current->fsuid) : (!in_group_p(id))) &&
1069             !capable(CAP_SYS_ADMIN))
1070                 return -EACCES;
1071
1072         error = qd_get(sdp, user, id, CREATE, &qd);
1073         if (error)
1074                 return error;
1075
1076         error = do_glock(qd, NO_FORCE, &q_gh);
1077         if (error)
1078                 goto out;
1079
1080         memset(q, 0, sizeof(struct gfs2_quota));
1081         q->qu_limit = be64_to_cpu(qd->qd_qb.qb_limit);
1082         q->qu_warn = be64_to_cpu(qd->qd_qb.qb_warn);
1083         q->qu_value = be64_to_cpu(qd->qd_qb.qb_value);
1084
1085         spin_lock(&sdp->sd_quota_spin);
1086         q->qu_value += qd->qd_change;
1087         spin_unlock(&sdp->sd_quota_spin);
1088
1089         gfs2_glock_dq_uninit(&q_gh);
1090
1091 out:
1092         qd_put(qd);
1093         return error;
1094 }
1095 #endif  /*  0  */
1096
1097 int gfs2_quota_init(struct gfs2_sbd *sdp)
1098 {
1099         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1100         unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1101         unsigned int x, slot = 0;
1102         unsigned int found = 0;
1103         u64 dblock;
1104         u32 extlen = 0;
1105         int error;
1106
1107         if (!ip->i_di.di_size ||
1108             ip->i_di.di_size > (64 << 20) ||
1109             ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1110                 gfs2_consist_inode(ip);
1111                 return -EIO;
1112         }
1113         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1114         sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1115
1116         error = -ENOMEM;
1117
1118         sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1119                                        sizeof(unsigned char *), GFP_KERNEL);
1120         if (!sdp->sd_quota_bitmap)
1121                 return error;
1122
1123         for (x = 0; x < sdp->sd_quota_chunks; x++) {
1124                 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1125                 if (!sdp->sd_quota_bitmap[x])
1126                         goto fail;
1127         }
1128
1129         for (x = 0; x < blocks; x++) {
1130                 struct buffer_head *bh;
1131                 unsigned int y;
1132
1133                 if (!extlen) {
1134                         int new = 0;
1135                         error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1136                         if (error)
1137                                 goto fail;
1138                 }
1139                 gfs2_meta_ra(ip->i_gl,  dblock, extlen);
1140                 error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT,
1141                                        &bh);
1142                 if (error)
1143                         goto fail;
1144                 error = -EIO;
1145                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1146                         brelse(bh);
1147                         goto fail;
1148                 }
1149
1150                 for (y = 0;
1151                      y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1152                      y++, slot++) {
1153                         struct gfs2_quota_change qc;
1154                         struct gfs2_quota_data *qd;
1155
1156                         gfs2_quota_change_in(&qc, bh->b_data +
1157                                           sizeof(struct gfs2_meta_header) +
1158                                           y * sizeof(struct gfs2_quota_change));
1159                         if (!qc.qc_change)
1160                                 continue;
1161
1162                         error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1163                                          qc.qc_id, &qd);
1164                         if (error) {
1165                                 brelse(bh);
1166                                 goto fail;
1167                         }
1168
1169                         set_bit(QDF_CHANGE, &qd->qd_flags);
1170                         qd->qd_change = qc.qc_change;
1171                         qd->qd_slot = slot;
1172                         qd->qd_slot_count = 1;
1173                         qd->qd_last_touched = jiffies;
1174
1175                         spin_lock(&sdp->sd_quota_spin);
1176                         gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1177                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1178                         atomic_inc(&sdp->sd_quota_count);
1179                         spin_unlock(&sdp->sd_quota_spin);
1180
1181                         found++;
1182                 }
1183
1184                 brelse(bh);
1185                 dblock++;
1186                 extlen--;
1187         }
1188
1189         if (found)
1190                 fs_info(sdp, "found %u quota changes\n", found);
1191
1192         return 0;
1193
1194 fail:
1195         gfs2_quota_cleanup(sdp);
1196         return error;
1197 }
1198
1199 void gfs2_quota_scan(struct gfs2_sbd *sdp)
1200 {
1201         struct gfs2_quota_data *qd, *safe;
1202         LIST_HEAD(dead);
1203
1204         spin_lock(&sdp->sd_quota_spin);
1205         list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1206                 if (!qd->qd_count &&
1207                     time_after_eq(jiffies, qd->qd_last_touched +
1208                                 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1209                         list_move(&qd->qd_list, &dead);
1210                         gfs2_assert_warn(sdp,
1211                                          atomic_read(&sdp->sd_quota_count) > 0);
1212                         atomic_dec(&sdp->sd_quota_count);
1213                 }
1214         }
1215         spin_unlock(&sdp->sd_quota_spin);
1216
1217         while (!list_empty(&dead)) {
1218                 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1219                 list_del(&qd->qd_list);
1220
1221                 gfs2_assert_warn(sdp, !qd->qd_change);
1222                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1223                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1224
1225                 gfs2_lvb_unhold(qd->qd_gl);
1226                 kfree(qd);
1227         }
1228 }
1229
1230 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1231 {
1232         struct list_head *head = &sdp->sd_quota_list;
1233         struct gfs2_quota_data *qd;
1234         unsigned int x;
1235
1236         spin_lock(&sdp->sd_quota_spin);
1237         while (!list_empty(head)) {
1238                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1239
1240                 if (qd->qd_count > 1 ||
1241                     (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1242                         list_move(&qd->qd_list, head);
1243                         spin_unlock(&sdp->sd_quota_spin);
1244                         schedule();
1245                         spin_lock(&sdp->sd_quota_spin);
1246                         continue;
1247                 }
1248
1249                 list_del(&qd->qd_list);
1250                 atomic_dec(&sdp->sd_quota_count);
1251                 spin_unlock(&sdp->sd_quota_spin);
1252
1253                 if (!qd->qd_count) {
1254                         gfs2_assert_warn(sdp, !qd->qd_change);
1255                         gfs2_assert_warn(sdp, !qd->qd_slot_count);
1256                 } else
1257                         gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1258                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1259
1260                 gfs2_lvb_unhold(qd->qd_gl);
1261                 kfree(qd);
1262
1263                 spin_lock(&sdp->sd_quota_spin);
1264         }
1265         spin_unlock(&sdp->sd_quota_spin);
1266
1267         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1268
1269         if (sdp->sd_quota_bitmap) {
1270                 for (x = 0; x < sdp->sd_quota_chunks; x++)
1271                         kfree(sdp->sd_quota_bitmap[x]);
1272                 kfree(sdp->sd_quota_bitmap);
1273         }
1274 }
1275