lustre/lvfs/lustre_quota_fmt.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  */
  30 /*
  31  * This file is part of Lustre, http://www.lustre.org/
  32  * Lustre is a trademark of Sun Microsystems, Inc.
  33  *
  34  * lustre/lvfs/lustre_quota_fmt.c
  35  *
  36  * Lustre administrative quota format.
  37  * from linux/fs/quota_v2.c
  38  */
  39
  40 #ifndef EXPORT_SYMTAB
  41 # define EXPORT_SYMTAB
  42 #endif
  43
  44 #include <linux/errno.h>
  45 #include <linux/fs.h>
  46 #include <linux/mount.h>
  47 #include <linux/kernel.h>
  48 #include <linux/init.h>
  49 #include <linux/module.h>
  50 #include <linux/slab.h>
  51 #ifdef HAVE_QUOTAIO_V1_H
  52 # include <linux/quotaio_v1.h>
  53 #endif
  54
  55 #include <asm/byteorder.h>
  56 #include <asm/uaccess.h>
  57
  58 #include <lustre_quota.h>
  59 #include <obd_support.h>
  60 #include "lustre_quota_fmt.h"
  61
  62 #ifdef HAVE_QUOTA_SUPPORT
  63
  64 static const uint lustre_initqversions[][MAXQUOTAS] = {
  65         [LUSTRE_QUOTA_V2] = LUSTRE_INITQVERSIONS_V2
  66 };
  67
  68 static const int lustre_dqstrinblk[] = {
  69         [LUSTRE_QUOTA_V2] = LUSTRE_DQSTRINBLK_V2
  70 };
  71
  72 static const int lustre_disk_dqblk_sz[] = {
  73         [LUSTRE_QUOTA_V2] = sizeof(struct lustre_disk_dqblk_v2)
  74 };
  75
  76 static const union
  77 {
  78         struct lustre_disk_dqblk_v2 r1;
  79 } fakedquot[] = {
  80         [LUSTRE_QUOTA_V2] = {.r1 = {.dqb_itime = __constant_cpu_to_le64(1LLU)}}
  81 };
  82
  83 static const union
  84 {
  85         struct lustre_disk_dqblk_v2 r1;
  86 } emptydquot[] = {
  87         [LUSTRE_QUOTA_V2] = {.r1 = { 0 }}
  88 };
  89
  90 extern void *lustre_quota_journal_start(struct inode *inode, int delete);
  91 extern void lustre_quota_journal_stop(void *handle);
  92 extern ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
  93                                  char *buf, int count, loff_t pos);
  94 extern ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos);
  95
  96 int check_quota_file(struct file *f, struct inode *inode, int type,
  97                      lustre_quota_version_t version)
  98 {
  99         struct lustre_disk_dqheader dqhead;
 100         ssize_t size;
 101         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 102         const uint *quota_versions = lustre_initqversions[version];
 103
 104         size = lustre_read_quota(f, inode, type, (char *)&dqhead,
 105                                  sizeof(struct lustre_disk_dqheader), 0);
 106         if (size != sizeof(struct lustre_disk_dqheader))
 107                 return -EINVAL;
 108         if (le32_to_cpu(dqhead.dqh_magic) != quota_magics[type] ||
 109             le32_to_cpu(dqhead.dqh_version) != quota_versions[type])
 110                 return -EINVAL;
 111         return 0;
 112 }
 113
 114 /**
 115  * Check whether given file is really lustre admin quotafile
 116  */
 117 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type)
 118 {
 119         struct file *f = lqi->qi_files[type];
 120         return check_quota_file(f, NULL, type, lqi->qi_version);
 121 }
 122 EXPORT_SYMBOL(lustre_check_quota_file);
 123
 124 int lustre_read_quota_file_info(struct file* f, struct lustre_mem_dqinfo* info)
 125 {
 126         struct lustre_disk_dqinfo dinfo;
 127         ssize_t size;
 128
 129         size = lustre_read_quota(f, NULL, 0, (char *)&dinfo,
 130                                  sizeof(struct lustre_disk_dqinfo),
 131                                  LUSTRE_DQINFOOFF);
 132
 133         if (size != sizeof(struct lustre_disk_dqinfo)) {
 134                 CDEBUG(D_ERROR, "Can't read info structure on device %s.\n",
 135                        f->f_vfsmnt->mnt_sb->s_id);
 136                 return -EINVAL;
 137         }
 138         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 139         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 140         info->dqi_flags = le32_to_cpu(dinfo.dqi_flags);
 141         info->dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 142         info->dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 143         info->dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 144         return 0;
 145 }
 146
 147 /**
 148  * Read information header from quota file
 149  */
 150 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type)
 151 {
 152         return lustre_read_quota_file_info(lqi->qi_files[type],
 153                                            &lqi->qi_info[type]);
 154 }
 155 EXPORT_SYMBOL(lustre_read_quota_info);
 156
 157 /**
 158  * Write information header to quota file
 159  */
 160 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type)
 161 {
 162         struct lustre_disk_dqinfo dinfo;
 163         struct lustre_mem_dqinfo *info = &lqi->qi_info[type];
 164         struct file *f = lqi->qi_files[type];
 165         ssize_t size;
 166
 167         info->dqi_flags &= ~DQF_INFO_DIRTY;
 168         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 169         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 170         dinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK);
 171         dinfo.dqi_blocks = cpu_to_le32(info->dqi_blocks);
 172         dinfo.dqi_free_blk = cpu_to_le32(info->dqi_free_blk);
 173         dinfo.dqi_free_entry = cpu_to_le32(info->dqi_free_entry);
 174
 175         size = lustre_write_quota(f, (char *)&dinfo,
 176                                   sizeof(struct lustre_disk_dqinfo),
 177                                   LUSTRE_DQINFOOFF);
 178
 179         if (size != sizeof(struct lustre_disk_dqinfo)) {
 180                 CDEBUG(D_WARNING,
 181                        "Can't write info structure on device %s.\n",
 182                        f->f_vfsmnt->mnt_sb->s_id);
 183                 return -1;
 184         }
 185         return 0;
 186 }
 187 EXPORT_SYMBOL(lustre_write_quota_info);
 188
 189 void disk2memdqb(struct lustre_mem_dqblk *m, void *d,
 190                  lustre_quota_version_t version)
 191 {
 192         struct lustre_disk_dqblk_v2 *dqblk = (struct lustre_disk_dqblk_v2 *)d;
 193
 194         LASSERT(version == LUSTRE_QUOTA_V2);
 195
 196         m->dqb_ihardlimit = le64_to_cpu(dqblk->dqb_ihardlimit);
 197         m->dqb_isoftlimit = le64_to_cpu(dqblk->dqb_isoftlimit);
 198         m->dqb_curinodes = le64_to_cpu(dqblk->dqb_curinodes);
 199         m->dqb_itime = le64_to_cpu(dqblk->dqb_itime);
 200         m->dqb_bhardlimit = le64_to_cpu(dqblk->dqb_bhardlimit);
 201         m->dqb_bsoftlimit = le64_to_cpu(dqblk->dqb_bsoftlimit);
 202         m->dqb_curspace = le64_to_cpu(dqblk->dqb_curspace);
 203         m->dqb_btime = le64_to_cpu(dqblk->dqb_btime);
 204 }
 205
 206 static int mem2diskdqb(void *d, struct lustre_mem_dqblk *m,
 207                        qid_t id, lustre_quota_version_t version)
 208 {
 209         struct lustre_disk_dqblk_v2 *dqblk = (struct lustre_disk_dqblk_v2 *)d;
 210
 211         LASSERT(version == LUSTRE_QUOTA_V2);
 212
 213         dqblk->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
 214         dqblk->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
 215         dqblk->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
 216         dqblk->dqb_itime = cpu_to_le64(m->dqb_itime);
 217         dqblk->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
 218         dqblk->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
 219         dqblk->dqb_curspace = cpu_to_le64(m->dqb_curspace);
 220         dqblk->dqb_btime = cpu_to_le64(m->dqb_btime);
 221         dqblk->dqb_id = cpu_to_le32(id);
 222
 223         return 0;
 224 }
 225
 226 dqbuf_t getdqbuf(void)
 227 {
 228         dqbuf_t buf = kmalloc(LUSTRE_DQBLKSIZE, GFP_NOFS);
 229         if (!buf)
 230                 CDEBUG(D_WARNING,
 231                        "VFS: Not enough memory for quota buffers.\n");
 232         return buf;
 233 }
 234
 235 void freedqbuf(dqbuf_t buf)
 236 {
 237         kfree(buf);
 238 }
 239
 240 ssize_t read_blk(struct file *filp, struct inode *inode, int type,
 241                  uint blk, dqbuf_t buf)
 242 {
 243         ssize_t ret;
 244
 245         memset(buf, 0, LUSTRE_DQBLKSIZE);
 246         ret = lustre_read_quota(filp, inode, type, (char *)buf, LUSTRE_DQBLKSIZE,
 247                                 blk << LUSTRE_DQBLKSIZE_BITS);
 248
 249         /* Reading past EOF just returns a block of zeros */
 250         if (ret == -EBADR)
 251                 ret = 0;
 252
 253         return ret;
 254 }
 255
 256 ssize_t write_blk(struct file *filp, uint blk, dqbuf_t buf)
 257 {
 258         ssize_t ret;
 259
 260         ret = lustre_write_quota(filp, (char *)buf, LUSTRE_DQBLKSIZE,
 261                                  blk << LUSTRE_DQBLKSIZE_BITS);
 262
 263         return ret;
 264 }
 265
 266 void lustre_mark_info_dirty(struct lustre_mem_dqinfo *info)
 267 {
 268         cfs_set_bit(DQF_INFO_DIRTY_B, &info->dqi_flags);
 269 }
 270
 271 /**
 272  * Remove empty block from list and return it
 273  */
 274 int get_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info)
 275 {
 276         dqbuf_t buf = getdqbuf();
 277         struct lustre_disk_dqdbheader *dh =
 278             (struct lustre_disk_dqdbheader *)buf;
 279         int ret, blk;
 280
 281         if (!buf)
 282                 return -ENOMEM;
 283         if (info->dqi_free_blk) {
 284                 blk = info->dqi_free_blk;
 285                 if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0)
 286                         goto out_buf;
 287                 info->dqi_free_blk = le32_to_cpu(dh->dqdh_next_free);
 288         } else {
 289                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 290                 /* Assure block allocation... */
 291                 if ((ret = write_blk(filp, info->dqi_blocks, buf)) < 0)
 292                         goto out_buf;
 293                 blk = info->dqi_blocks++;
 294         }
 295         lustre_mark_info_dirty(info);
 296         ret = blk;
 297 out_buf:
 298         freedqbuf(buf);
 299         return ret;
 300 }
 301
 302 /**
 303  * Insert empty block to the list
 304  */
 305 int put_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info,
 306                    dqbuf_t buf, uint blk)
 307 {
 308         struct lustre_disk_dqdbheader *dh =
 309             (struct lustre_disk_dqdbheader *)buf;
 310         int err;
 311
 312         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_blk);
 313         dh->dqdh_prev_free = cpu_to_le32(0);
 314         dh->dqdh_entries = cpu_to_le16(0);
 315         info->dqi_free_blk = blk;
 316         lustre_mark_info_dirty(info);
 317         if ((err = write_blk(filp, blk, buf)) < 0)
 318                 /* Some strange block. We had better leave it... */
 319                 return err;
 320         return 0;
 321 }
 322
 323 /**
 324  * Remove given block from the list of blocks with free entries
 325  */
 326 int remove_free_dqentry(struct file *filp,
 327                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 328                         uint blk)
 329 {
 330         dqbuf_t tmpbuf = getdqbuf();
 331         struct lustre_disk_dqdbheader *dh =
 332             (struct lustre_disk_dqdbheader *)buf;
 333         uint nextblk = le32_to_cpu(dh->dqdh_next_free), prevblk =
 334             le32_to_cpu(dh->dqdh_prev_free);
 335         int err;
 336
 337         if (!tmpbuf)
 338                 return -ENOMEM;
 339         if (nextblk) {
 340                 if ((err = read_blk(filp, NULL, 0, nextblk, tmpbuf)) < 0)
 341                         goto out_buf;
 342                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 343                     dh->dqdh_prev_free;
 344                 if ((err = write_blk(filp, nextblk, tmpbuf)) < 0)
 345                         goto out_buf;
 346         }
 347         if (prevblk) {
 348                 if ((err = read_blk(filp, NULL, 0, prevblk, tmpbuf)) < 0)
 349                         goto out_buf;
 350                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_next_free =
 351                     dh->dqdh_next_free;
 352                 if ((err = write_blk(filp, prevblk, tmpbuf)) < 0)
 353                         goto out_buf;
 354         } else {
 355                 info->dqi_free_entry = nextblk;
 356                 lustre_mark_info_dirty(info);
 357         }
 358         freedqbuf(tmpbuf);
 359         dh->dqdh_next_free = dh->dqdh_prev_free = cpu_to_le32(0);
 360         if (write_blk(filp, blk, buf) < 0)
 361                 /* No matter whether write succeeds block is out of list */
 362                 CDEBUG(D_ERROR,
 363                        "VFS: Can't write block (%u) with free entries.\n", blk);
 364         return 0;
 365 out_buf:
 366         freedqbuf(tmpbuf);
 367         return err;
 368 }
 369
 370 /**
 371  * Insert given block to the beginning of list with free entries
 372  */
 373 int insert_free_dqentry(struct file *filp,
 374                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 375                         uint blk)
 376 {
 377         dqbuf_t tmpbuf = getdqbuf();
 378         struct lustre_disk_dqdbheader *dh =
 379             (struct lustre_disk_dqdbheader *)buf;
 380         int err;
 381
 382         if (!tmpbuf)
 383                 return -ENOMEM;
 384         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_entry);
 385         dh->dqdh_prev_free = cpu_to_le32(0);
 386         if ((err = write_blk(filp, blk, buf)) < 0)
 387                 goto out_buf;
 388         if (info->dqi_free_entry) {
 389                 if ((err = read_blk(filp, NULL, 0, info->dqi_free_entry, tmpbuf)) < 0)
 390                         goto out_buf;
 391                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 392                     cpu_to_le32(blk);
 393                 if ((err = write_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 394                         goto out_buf;
 395         }
 396         freedqbuf(tmpbuf);
 397         info->dqi_free_entry = blk;
 398         lustre_mark_info_dirty(info);
 399         return 0;
 400 out_buf:
 401         freedqbuf(tmpbuf);
 402         return err;
 403 }
 404
 405
 406
 407 /**
 408  * Find space for dquot
 409  */
 410 static uint find_free_dqentry(struct lustre_dquot *dquot, int *err,
 411                               lustre_quota_version_t version)
 412 {
 413         struct lustre_quota_info *lqi = dquot->dq_info;
 414         struct file *filp = lqi->qi_files[dquot->dq_type];
 415         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 416         uint blk, i;
 417         struct lustre_disk_dqdbheader *dh;
 418         void *ddquot;
 419         int dqblk_sz = lustre_disk_dqblk_sz[version];
 420         int dqstrinblk = lustre_dqstrinblk[version];
 421         dqbuf_t buf;
 422
 423         *err = 0;
 424         if (!(buf = getdqbuf())) {
 425                 *err = -ENOMEM;
 426                 return 0;
 427         }
 428         dh = (struct lustre_disk_dqdbheader *)buf;
 429         ddquot = GETENTRIES(buf, version);
 430         if (info->dqi_free_entry) {
 431                 blk = info->dqi_free_entry;
 432                 if ((*err = read_blk(filp, NULL, 0, blk, buf)) < 0)
 433                         goto out_buf;
 434         } else {
 435                 blk = get_free_dqblk(filp, info);
 436                 if ((int)blk < 0) {
 437                         *err = blk;
 438                         freedqbuf(buf);
 439                         return 0;
 440                 }
 441                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 442                 info->dqi_free_entry = blk; /* This is enough as block is
 443                                                already zeroed and entry list
 444                                                is empty... */
 445                 lustre_mark_info_dirty(info);
 446         }
 447
 448         /* Will block be full */
 449         if (le16_to_cpu(dh->dqdh_entries) + 1 >= dqstrinblk)
 450                 if ((*err = remove_free_dqentry(filp, info, buf, blk)) < 0) {
 451                         CDEBUG(D_ERROR,
 452                                "VFS: find_free_dqentry(): Can't remove block "
 453                                "(%u) from entry free list.\n", blk);
 454                         goto out_buf;
 455                 }
 456         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) + 1);
 457         /* Find free structure in block */
 458         for (i = 0; i < dqstrinblk &&
 459              memcmp((char *)&emptydquot[version],
 460                     (char *)ddquot + i * dqblk_sz, dqblk_sz);
 461              i++);
 462
 463         if (i == dqstrinblk) {
 464                 CDEBUG(D_ERROR,
 465                        "VFS: find_free_dqentry(): Data block full but it "
 466                        "shouldn't.\n");
 467                 *err = -EIO;
 468                 goto out_buf;
 469         }
 470
 471         if ((*err = write_blk(filp, blk, buf)) < 0) {
 472                 CDEBUG(D_ERROR,
 473                        "VFS: find_free_dqentry(): Can't write quota data "
 474                        "block %u.\n", blk);
 475                 goto out_buf;
 476         }
 477         dquot->dq_off =
 478             (blk << LUSTRE_DQBLKSIZE_BITS) +
 479             sizeof(struct lustre_disk_dqdbheader) +
 480             i * dqblk_sz;
 481         freedqbuf(buf);
 482         return blk;
 483 out_buf:
 484         freedqbuf(buf);
 485         return 0;
 486 }
 487
 488 /**
 489  * Insert reference to structure into the trie
 490  */
 491 static int do_insert_tree(struct lustre_dquot *dquot, uint * treeblk, int depth,
 492                           lustre_quota_version_t version)
 493 {
 494         struct lustre_quota_info *lqi = dquot->dq_info;
 495         struct file *filp = lqi->qi_files[dquot->dq_type];
 496         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 497         dqbuf_t buf;
 498         int ret = 0, newson = 0, newact = 0;
 499         u32 *ref;
 500         uint newblk;
 501
 502         if (!(buf = getdqbuf()))
 503                 return -ENOMEM;
 504         if (!*treeblk) {
 505                 ret = get_free_dqblk(filp, info);
 506                 if (ret < 0)
 507                         goto out_buf;
 508                 *treeblk = ret;
 509                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 510                 newact = 1;
 511         } else {
 512                 if ((ret = read_blk(filp, NULL, 0, *treeblk, buf)) < 0) {
 513                         CERROR("VFS: Can't read tree quota block %u.\n",
 514                                *treeblk);
 515                         goto out_buf;
 516                 }
 517         }
 518         ref = (u32 *) buf;
 519         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 520         if (!newblk)
 521                 newson = 1;
 522         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 523
 524                 if (newblk) {
 525                         CDEBUG(D_ERROR,
 526                                "VFS: Inserting already present quota entry "
 527                                "(block %u).\n",
 528                                ref[GETIDINDEX(dquot->dq_id, depth)]);
 529                         ret = -EIO;
 530                         goto out_buf;
 531                 }
 532
 533                 newblk = find_free_dqentry(dquot, &ret, version);
 534         } else
 535                 ret = do_insert_tree(dquot, &newblk, depth + 1, version);
 536         if (newson && ret >= 0) {
 537                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(newblk);
 538                 ret = write_blk(filp, *treeblk, buf);
 539         } else if (newact && ret < 0)
 540                 put_free_dqblk(filp, info, buf, *treeblk);
 541 out_buf:
 542         freedqbuf(buf);
 543         return ret;
 544 }
 545
 546 /**
 547  * Wrapper for inserting quota structure into tree
 548  */
 549 static inline int dq_insert_tree(struct lustre_dquot *dquot,
 550                                  lustre_quota_version_t version)
 551 {
 552         int tmp = LUSTRE_DQTREEOFF;
 553         return do_insert_tree(dquot, &tmp, 0, version);
 554 }
 555
 556 /**
 557  * We don't have to be afraid of deadlocks as we never have quotas on
 558  * quota files...
 559  */
 560 static int lustre_write_dquot(struct lustre_dquot *dquot,
 561                               lustre_quota_version_t version)
 562 {
 563         int type = dquot->dq_type;
 564         struct file *filp;
 565         loff_t offset;
 566         ssize_t ret;
 567         int dqblk_sz = lustre_disk_dqblk_sz[version];
 568         struct lustre_disk_dqblk_v2 ddquot;
 569
 570         ret = mem2diskdqb(&ddquot, &dquot->dq_dqb, dquot->dq_id, version);
 571         if (ret < 0)
 572                 return ret;
 573
 574         if (!dquot->dq_off)
 575                 if ((ret = dq_insert_tree(dquot, version)) < 0) {
 576                         CDEBUG(D_ERROR,
 577                                "VFS: Error %Zd occurred while creating "
 578                                "quota.\n", ret);
 579                         return ret;
 580                 }
 581         filp = dquot->dq_info->qi_files[type];
 582         offset = dquot->dq_off;
 583         /* Argh... We may need to write structure full of zeroes but that would
 584          * be treated as an empty place by the rest of the code. Format change
 585          * would be definitely cleaner but the problems probably are not worth
 586          * it */
 587         if (!memcmp((char *)&emptydquot[version], (char *)&ddquot, dqblk_sz))
 588                 ddquot.dqb_itime = cpu_to_le64(1);
 589
 590         ret = lustre_write_quota(filp, (char *)&ddquot, dqblk_sz, offset);
 591         if (ret != dqblk_sz) {
 592                 CDEBUG(D_WARNING, "VFS: dquota write failed on dev %s\n",
 593                        filp->f_dentry->d_sb->s_id);
 594                 if (ret >= 0)
 595                         ret = -ENOSPC;
 596         } else
 597                 ret = 0;
 598
 599         return ret;
 600 }
 601
 602 /**
 603  * Free dquot entry in data block
 604  */
 605 static int free_dqentry(struct lustre_dquot *dquot, uint blk,
 606                         lustre_quota_version_t version)
 607 {
 608         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 609         struct lustre_mem_dqinfo *info =
 610             &dquot->dq_info->qi_info[dquot->dq_type];
 611         struct lustre_disk_dqdbheader *dh;
 612         dqbuf_t buf = getdqbuf();
 613         int dqstrinblk = lustre_dqstrinblk[version];
 614         int ret = 0;
 615
 616         if (!buf)
 617                 return -ENOMEM;
 618         if (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS != blk) {
 619                 CDEBUG(D_ERROR,
 620                        "VFS: Quota structure has offset to other block (%u) "
 621                        "than it should (%u).\n",
 622                        blk, (uint) (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS));
 623                 goto out_buf;
 624         }
 625         if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0) {
 626                 CDEBUG(D_ERROR, "VFS: Can't read quota data block %u\n", blk);
 627                 goto out_buf;
 628         }
 629         dh = (struct lustre_disk_dqdbheader *)buf;
 630         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) - 1);
 631         if (!le16_to_cpu(dh->dqdh_entries)) {   /* Block got free? */
 632                 if ((ret = remove_free_dqentry(filp, info, buf, blk)) < 0 ||
 633                     (ret = put_free_dqblk(filp, info, buf, blk)) < 0) {
 634                         CDEBUG(D_ERROR,
 635                                "VFS: Can't move quota data block (%u) to free "
 636                                "list.\n", blk);
 637                         goto out_buf;
 638                 }
 639         } else {
 640                 memset(buf + (dquot->dq_off & ((1<<LUSTRE_DQBLKSIZE_BITS) - 1)),
 641                        0, lustre_disk_dqblk_sz[version]);
 642                 if (le16_to_cpu(dh->dqdh_entries) == dqstrinblk - 1) {
 643                         /* Insert will write block itself */
 644                         if ((ret =
 645                              insert_free_dqentry(filp, info, buf, blk)) < 0) {
 646                                 CDEBUG(D_ERROR,
 647                                        "VFS: Can't insert quota data block "
 648                                        "(%u) to free entry list.\n", blk);
 649                                 goto out_buf;
 650                         }
 651                 } else if ((ret = write_blk(filp, blk, buf)) < 0) {
 652                         CDEBUG(D_ERROR,
 653                                "VFS: Can't write quota data block %u\n", blk);
 654                         goto out_buf;
 655                 }
 656         }
 657         dquot->dq_off = 0;      /* Quota is now unattached */
 658 out_buf:
 659         freedqbuf(buf);
 660         return ret;
 661 }
 662
 663 /**
 664  * Remove reference to dquot from tree
 665  */
 666 static int remove_tree(struct lustre_dquot *dquot, uint * blk, int depth,
 667                        lustre_quota_version_t version)
 668 {
 669         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 670         struct lustre_mem_dqinfo *info =
 671             &dquot->dq_info->qi_info[dquot->dq_type];
 672         dqbuf_t buf = getdqbuf();
 673         int ret = 0;
 674         uint newblk;
 675         u32 *ref = (u32 *) buf;
 676
 677         if (!buf)
 678                 return -ENOMEM;
 679         if ((ret = read_blk(filp, NULL, 0, *blk, buf)) < 0) {
 680                 CERROR("VFS: Can't read quota data block %u\n", *blk);
 681                 goto out_buf;
 682         }
 683         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 684         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 685                 ret = free_dqentry(dquot, newblk, version);
 686                 newblk = 0;
 687         } else
 688                 ret = remove_tree(dquot, &newblk, depth + 1, version);
 689         if (ret >= 0 && !newblk) {
 690                 int i;
 691                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(0);
 692                 for (i = 0; i < LUSTRE_DQBLKSIZE && !buf[i]; i++)
 693                         /* Block got empty? */ ;
 694                 /* don't put the root block into free blk list! */
 695                 if (i == LUSTRE_DQBLKSIZE && *blk != LUSTRE_DQTREEOFF) {
 696                         put_free_dqblk(filp, info, buf, *blk);
 697                         *blk = 0;
 698                 } else if ((ret = write_blk(filp, *blk, buf)) < 0)
 699                         CDEBUG(D_ERROR,
 700                                "VFS: Can't write quota tree block %u.\n", *blk);
 701         }
 702 out_buf:
 703         freedqbuf(buf);
 704         return ret;
 705 }
 706
 707 /**
 708  * Delete dquot from tree
 709  */
 710 static int lustre_delete_dquot(struct lustre_dquot *dquot,
 711                                 lustre_quota_version_t version)
 712 {
 713         uint tmp = LUSTRE_DQTREEOFF;
 714
 715         if (!dquot->dq_off)     /* Even not allocated? */
 716                 return 0;
 717         return remove_tree(dquot, &tmp, 0, version);
 718 }
 719
 720 /**
 721  * Find entry in block
 722  */
 723 static loff_t find_block_dqentry(struct lustre_dquot *dquot, uint blk,
 724                                  lustre_quota_version_t version)
 725 {
 726         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 727         dqbuf_t buf = getdqbuf();
 728         loff_t ret = 0;
 729         int i;
 730         struct lustre_disk_dqblk_v2 *ddquot =
 731                 (struct lustre_disk_dqblk_v2 *)GETENTRIES(buf, version);
 732         int dqblk_sz = lustre_disk_dqblk_sz[version];
 733         int dqstrinblk = lustre_dqstrinblk[version];
 734
 735         LASSERT(version == LUSTRE_QUOTA_V2);
 736
 737         if (!buf)
 738                 return -ENOMEM;
 739         if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0) {
 740                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 741                 goto out_buf;
 742         }
 743         if (dquot->dq_id)
 744                 for (i = 0; i < dqstrinblk &&
 745                      le32_to_cpu(ddquot[i].dqb_id) != dquot->dq_id;
 746                      i++) ;
 747         else {                  /* ID 0 as a bit more complicated searching... */
 748                 for (i = 0; i < dqstrinblk; i++)
 749                         if (!le32_to_cpu(ddquot[i].dqb_id)
 750                             && memcmp((char *)&emptydquot[version],
 751                                       (char *)&ddquot[i], dqblk_sz))
 752                                 break;
 753         }
 754         if (i == dqstrinblk) {
 755                 CDEBUG(D_ERROR,
 756                        "VFS: Quota for id %u referenced but not present.\n",
 757                        dquot->dq_id);
 758                 ret = -EIO;
 759                 goto out_buf;
 760         } else
 761                 ret =
 762                     (blk << LUSTRE_DQBLKSIZE_BITS) +
 763                     sizeof(struct lustre_disk_dqdbheader) +
 764                     i * dqblk_sz;
 765 out_buf:
 766         freedqbuf(buf);
 767         return ret;
 768 }
 769
 770 /**
 771  * Find entry for given id in the tree
 772  */
 773 static loff_t find_tree_dqentry(struct lustre_dquot *dquot, uint blk, int depth,
 774                                 lustre_quota_version_t version)
 775 {
 776         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 777         dqbuf_t buf = getdqbuf();
 778         loff_t ret = 0;
 779         u32 *ref = (u32 *) buf;
 780
 781         if (!buf)
 782                 return -ENOMEM;
 783         if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0) {
 784                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 785                 goto out_buf;
 786         }
 787         ret = 0;
 788         blk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 789         if (!blk)               /* No reference? */
 790                 goto out_buf;
 791         if (depth < LUSTRE_DQTREEDEPTH - 1)
 792                 ret = find_tree_dqentry(dquot, blk, depth + 1, version);
 793         else
 794                 ret = find_block_dqentry(dquot, blk, version);
 795 out_buf:
 796         freedqbuf(buf);
 797         return ret;
 798 }
 799
 800 /**
 801  * Find entry for given id in the tree - wrapper function
 802  */
 803 static inline loff_t find_dqentry(struct lustre_dquot *dquot,
 804                                   lustre_quota_version_t version)
 805 {
 806         return find_tree_dqentry(dquot, LUSTRE_DQTREEOFF, 0, version);
 807 }
 808
 809 int lustre_read_dquot(struct lustre_dquot *dquot)
 810 {
 811         int type = dquot->dq_type;
 812         struct file *filp;
 813         loff_t offset;
 814         int ret = 0, dqblk_sz;
 815         lustre_quota_version_t version;
 816
 817         /* Invalidated quota? */
 818         if (!dquot->dq_info || !(filp = dquot->dq_info->qi_files[type])) {
 819                 CDEBUG(D_ERROR, "VFS: Quota invalidated while reading!\n");
 820                 return -ESRCH;
 821         }
 822
 823         version = dquot->dq_info->qi_version;
 824         LASSERT(version == LUSTRE_QUOTA_V2);
 825         dqblk_sz = lustre_disk_dqblk_sz[version];
 826
 827         offset = find_dqentry(dquot, version);
 828         if (offset <= 0) {      /* Entry not present? */
 829                 if (offset < 0)
 830                         CDEBUG(D_ERROR,
 831                                "VFS: Can't read quota structure for id %u.\n",
 832                                dquot->dq_id);
 833                 dquot->dq_off = 0;
 834                 cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
 835                 memset(&dquot->dq_dqb, 0, sizeof(struct lustre_mem_dqblk));
 836                 ret = offset;
 837         } else {
 838                 struct lustre_disk_dqblk_v2 ddquot;
 839
 840                 dquot->dq_off = offset;
 841                 if ((ret = lustre_read_quota(filp, NULL, type, (char *)&ddquot,
 842                                              dqblk_sz, offset)) != dqblk_sz) {
 843                         if (ret >= 0)
 844                                 ret = -EIO;
 845                         CDEBUG(D_ERROR,
 846                                "VFS: Error while reading quota structure for id "
 847                                "%u.\n", dquot->dq_id);
 848                         memset((char *)&ddquot, 0, dqblk_sz);
 849                 } else {
 850                         ret = 0;
 851                         /* We need to escape back all-zero structure */
 852                         if (!memcmp((char *)&fakedquot[version],
 853                                     (char *)&ddquot, dqblk_sz))
 854                                 ddquot.dqb_itime = cpu_to_le64(0);
 855                 }
 856                 disk2memdqb(&dquot->dq_dqb, &ddquot, version);
 857         }
 858
 859         return ret;
 860 }
 861 EXPORT_SYMBOL(lustre_read_dquot);
 862
 863 /**
 864  * Commit changes of dquot to disk - it might also mean deleting
 865  * it when quota became fake.
 866  */
 867 int lustre_commit_dquot(struct lustre_dquot *dquot)
 868 {
 869         int rc = 0;
 870         lustre_quota_version_t version = dquot->dq_info->qi_version;
 871         void *handle;
 872         struct inode *inode = dquot->dq_info->qi_files[dquot->dq_type]->f_dentry->d_inode;
 873         int delete = 0;
 874
 875         /* always clear the flag so we don't loop on an IO error... */
 876         cfs_clear_bit(DQ_MOD_B, &dquot->dq_flags);
 877
 878         /* The block/inode usage in admin quotafile isn't the real usage
 879          * over all cluster, so keep the fake dquot entry on disk is
 880          * meaningless, just remove it */
 881         if (cfs_test_bit(DQ_FAKE_B, &dquot->dq_flags))
 882                 delete = 1;
 883         handle = lustre_quota_journal_start(inode, delete);
 884         if (unlikely(IS_ERR(handle))) {
 885                 rc = PTR_ERR(handle);
 886                 CERROR("fail to lustre_quota_journal_start: rc = %d\n", rc);
 887                 return rc;
 888         }
 889
 890         if (delete)
 891                 rc = lustre_delete_dquot(dquot, version);
 892         else
 893                 rc = lustre_write_dquot(dquot, version);
 894         lustre_quota_journal_stop(handle);
 895
 896         if (rc < 0)
 897                 return rc;
 898
 899         if (lustre_info_dirty(&dquot->dq_info->qi_info[dquot->dq_type]))
 900                 rc = lustre_write_quota_info(dquot->dq_info, dquot->dq_type);
 901
 902         return rc;
 903 }
 904 EXPORT_SYMBOL(lustre_commit_dquot);
 905
 906 int lustre_init_quota_header(struct lustre_quota_info *lqi, int type,
 907                              int fakemagics)
 908 {
 909         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 910         static const uint fake_magics[] = LUSTRE_BADQMAGICS;
 911         const uint* quota_versions = lustre_initqversions[lqi->qi_version];
 912         struct lustre_disk_dqheader dqhead;
 913         ssize_t size;
 914         struct file *fp = lqi->qi_files[type];
 915         int rc = 0;
 916
 917         /* write quotafile header */
 918         dqhead.dqh_magic = cpu_to_le32(fakemagics ?
 919                                        fake_magics[type] : quota_magics[type]);
 920         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
 921         size = lustre_write_quota(fp, (char *)&dqhead,
 922                                   sizeof(struct lustre_disk_dqheader), 0);
 923
 924         if (size != sizeof(struct lustre_disk_dqheader)) {
 925                 CDEBUG(D_ERROR, "error writing quoafile header (rc:%d)\n", rc);
 926                 rc = size;
 927         }
 928
 929         return rc;
 930 }
 931
 932 /**
 933  * We need to export this function to initialize quotafile, because we haven't
 934  * user level check utility
 935  */
 936 int lustre_init_quota_info_generic(struct lustre_quota_info *lqi, int type,
 937                                    int fakemagics)
 938 {
 939         struct lustre_mem_dqinfo *dqinfo = &lqi->qi_info[type];
 940         int rc;
 941
 942         rc = lustre_init_quota_header(lqi, type, fakemagics);
 943         if (rc)
 944                 return rc;
 945
 946         /* write init quota info */
 947         memset(dqinfo, 0, sizeof(*dqinfo));
 948         dqinfo->dqi_bgrace = MAX_DQ_TIME;
 949         dqinfo->dqi_igrace = MAX_IQ_TIME;
 950         dqinfo->dqi_blocks = LUSTRE_DQTREEOFF + 1;
 951
 952         return lustre_write_quota_info(lqi, type);
 953 }
 954
 955 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type)
 956 {
 957         return lustre_init_quota_info_generic(lqi, type, 0);
 958 }
 959 EXPORT_SYMBOL(lustre_init_quota_info);
 960
 961 static int walk_block_dqentry(struct file *filp, struct inode *inode, int type,
 962                               uint blk, cfs_list_t *list)
 963 {
 964         dqbuf_t buf = getdqbuf();
 965         loff_t ret = 0;
 966         struct lustre_disk_dqdbheader *dqhead =
 967             (struct lustre_disk_dqdbheader *)buf;
 968         struct dqblk *blk_item;
 969         struct dqblk *pos;
 970         cfs_list_t *tmp;
 971
 972         if (!buf)
 973                 return -ENOMEM;
 974         if ((ret = read_blk(filp, inode, type, blk, buf)) < 0) {
 975                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 976                 goto out_buf;
 977         }
 978         ret = 0;
 979
 980         if (!le32_to_cpu(dqhead->dqdh_entries))
 981                 goto out_buf;
 982
 983         if (cfs_list_empty(list)) {
 984                 tmp = list;
 985                 goto done;
 986         }
 987
 988         cfs_list_for_each_entry(pos, list, link) {
 989                 if (blk == pos->blk)    /* we got this blk already */
 990                         goto out_buf;
 991                 if (blk > pos->blk)
 992                         continue;
 993                 break;
 994         }
 995         tmp = &pos->link;
 996 done:
 997         blk_item = kmalloc(sizeof(*blk_item), GFP_NOFS);
 998         if (!blk_item) {
 999                 ret = -ENOMEM;
1000                 goto out_buf;
1001         }
1002         blk_item->blk = blk;
1003         CFS_INIT_LIST_HEAD(&blk_item->link);
1004
1005         cfs_list_add_tail(&blk_item->link, tmp);
1006
1007 out_buf:
1008         freedqbuf(buf);
1009         return ret;
1010 }
1011
1012 int walk_tree_dqentry(struct file *filp, struct inode *inode, int type,
1013                       uint blk, int depth, cfs_list_t *list)
1014 {
1015         dqbuf_t buf = getdqbuf();
1016         loff_t ret = 0;
1017         int index;
1018         u32 *ref = (u32 *) buf;
1019
1020         if (!buf)
1021                 return -ENOMEM;
1022         if ((ret = read_blk(filp, inode, type, blk, buf)) < 0) {
1023                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
1024                 goto out_buf;
1025         }
1026         ret = 0;
1027
1028         for (index = 0; index <= 0xff && !ret; index++) {
1029                 blk = le32_to_cpu(ref[index]);
1030                 if (!blk)       /* No reference */
1031                         continue;
1032
1033                 if (depth < LUSTRE_DQTREEDEPTH - 1)
1034                         ret = walk_tree_dqentry(filp, inode, type, blk,
1035                                                 depth + 1, list);
1036                 else
1037                         ret = walk_block_dqentry(filp, inode, type, blk, list);
1038         }
1039 out_buf:
1040         freedqbuf(buf);
1041         return ret;
1042 }
1043
1044 /**
1045  * Walk through the quota file (v2 format) to get all ids with quota limit
1046  */
1047 int lustre_get_qids(struct file *fp, struct inode *inode, int type,
1048                     cfs_list_t *list)
1049 {
1050         cfs_list_t blk_list;
1051         struct dqblk *blk_item, *tmp;
1052         dqbuf_t buf = NULL;
1053         struct lustre_disk_dqblk_v2 *ddquot;
1054         int rc;
1055         lustre_quota_version_t version;
1056
1057         ENTRY;
1058
1059         LASSERT(ergo(fp == NULL, inode != NULL));
1060
1061         if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V2) == 0)
1062                 version = LUSTRE_QUOTA_V2;
1063         else {
1064                 CDEBUG(D_ERROR, "unknown quota file format!\n");
1065                 RETURN(-EINVAL);
1066         }
1067
1068         if (!cfs_list_empty(list)) {
1069                 CDEBUG(D_ERROR, "not empty list\n");
1070                 RETURN(-EINVAL);
1071         }
1072
1073         CFS_INIT_LIST_HEAD(&blk_list);
1074         rc = walk_tree_dqentry(fp, inode, type, LUSTRE_DQTREEOFF, 0, &blk_list);
1075         if (rc) {
1076                 CDEBUG(D_ERROR, "walk through quota file failed!(%d)\n", rc);
1077                 GOTO(out_free, rc);
1078         }
1079         if (cfs_list_empty(&blk_list))
1080                 RETURN(0);
1081
1082         buf = getdqbuf();
1083         if (!buf)
1084                 RETURN(-ENOMEM);
1085         ddquot = (struct lustre_disk_dqblk_v2 *)GETENTRIES(buf, version);
1086
1087         cfs_list_for_each_entry(blk_item, &blk_list, link) {
1088                 loff_t ret = 0;
1089                 int i, dqblk_sz = lustre_disk_dqblk_sz[version];
1090
1091                 memset(buf, 0, LUSTRE_DQBLKSIZE);
1092                 if ((ret = read_blk(fp, inode, type, blk_item->blk, buf)) < 0) {
1093                         CERROR("VFS: Can't read quota tree block %u.\n",
1094                                blk_item->blk);
1095                         GOTO(out_free, rc = ret);
1096                 }
1097
1098                 for (i = 0; i < lustre_dqstrinblk[version]; i++) {
1099                         struct dquot_id *dqid;
1100                         /* skip empty entry */
1101                         if (!memcmp((char *)&emptydquot[version],
1102                                     (char *)&ddquot[i], dqblk_sz))
1103                                 continue;
1104
1105                         OBD_ALLOC_GFP(dqid, sizeof(*dqid), GFP_NOFS);
1106                         if (!dqid)
1107                                 GOTO(out_free, rc = -ENOMEM);
1108
1109                         dqid->di_id    = le32_to_cpu(ddquot[i].dqb_id);
1110                         dqid->di_flag  = le64_to_cpu(ddquot[i].dqb_ihardlimit) ?
1111                                          QI_SET : 0;
1112                         dqid->di_flag |= le64_to_cpu(ddquot[i].dqb_bhardlimit) ?
1113                                          QB_SET : 0;
1114
1115                         CFS_INIT_LIST_HEAD(&dqid->di_link);
1116                         cfs_list_add(&dqid->di_link, list);
1117                 }
1118         }
1119
1120 out_free:
1121         cfs_list_for_each_entry_safe(blk_item, tmp, &blk_list, link) {
1122                 cfs_list_del_init(&blk_item->link);
1123                 kfree(blk_item);
1124         }
1125         if (buf)
1126                 freedqbuf(buf);
1127
1128         RETURN(rc);
1129 }
1130 EXPORT_SYMBOL(lustre_get_qids);
1131 #endif