lustre/lvfs/lustre_quota_fmt.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/lustre_quota_fmt.c
  37  *
  38  * Lustre administrative quota format.
  39  * from linux/fs/quota_v2.c
  40  */
  41
  42 #ifndef EXPORT_SYMTAB
  43 # define EXPORT_SYMTAB
  44 #endif
  45
  46 #include <linux/errno.h>
  47 #include <linux/fs.h>
  48 #include <linux/mount.h>
  49 #include <linux/kernel.h>
  50 #include <linux/init.h>
  51 #include <linux/module.h>
  52 #include <linux/slab.h>
  53 #ifdef HAVE_QUOTAIO_V1_H
  54 # include <linux/quotaio_v1.h>
  55 #endif
  56
  57 #include <asm/byteorder.h>
  58 #include <asm/uaccess.h>
  59
  60 #include <lustre_quota.h>
  61 #include <obd_support.h>
  62 #include "lustre_quota_fmt.h"
  63
  64 #ifdef HAVE_QUOTA_SUPPORT
  65
  66 static const uint lustre_initqversions[][MAXQUOTAS] = {
  67         [LUSTRE_QUOTA_V1] = LUSTRE_INITQVERSIONS_V1,
  68         [LUSTRE_QUOTA_V2] = LUSTRE_INITQVERSIONS_V2
  69 };
  70
  71 static const int lustre_dqstrinblk[] = {
  72         [LUSTRE_QUOTA_V1] = LUSTRE_DQSTRINBLK,
  73         [LUSTRE_QUOTA_V2] = LUSTRE_DQSTRINBLK_V2
  74 };
  75
  76 static const int lustre_disk_dqblk_sz[] = {
  77         [LUSTRE_QUOTA_V1] = sizeof(struct lustre_disk_dqblk),
  78         [LUSTRE_QUOTA_V2] = sizeof(struct lustre_disk_dqblk_v2)
  79 };
  80
  81 static const union
  82 {
  83         struct lustre_disk_dqblk    r0;
  84         struct lustre_disk_dqblk_v2 r1;
  85 } fakedquot[] = {
  86         [LUSTRE_QUOTA_V1] = {.r0 = {.dqb_itime = __constant_cpu_to_le64(1LLU)}},
  87         [LUSTRE_QUOTA_V2] = {.r1 = {.dqb_itime = __constant_cpu_to_le64(1LLU)}}
  88 };
  89
  90 static const union
  91 {
  92         struct lustre_disk_dqblk    r0;
  93         struct lustre_disk_dqblk_v2 r1;
  94 } emptydquot[] = {
  95         [LUSTRE_QUOTA_V1] = {.r0 = { 0 }},
  96         [LUSTRE_QUOTA_V2] = {.r1 = { 0 }}
  97 };
  98
  99 extern void *lustre_quota_journal_start(struct inode *inode, int delete);
 100 extern void lustre_quota_journal_stop(void *handle);
 101 extern ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
 102                                  char *buf, int count, loff_t pos);
 103 extern ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos);
 104
 105 int check_quota_file(struct file *f, struct inode *inode, int type,
 106                      lustre_quota_version_t version)
 107 {
 108         struct lustre_disk_dqheader dqhead;
 109         ssize_t size;
 110         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 111         const uint *quota_versions = lustre_initqversions[version];
 112
 113         if (!f && !inode) {
 114                 CERROR("check_quota_file failed!\n");
 115                 libcfs_debug_dumpstack(NULL);
 116                 return -EINVAL;
 117         }
 118
 119         size = lustre_read_quota(f, inode, type, (char *)&dqhead,
 120                                  sizeof(struct lustre_disk_dqheader), 0);
 121
 122         if (size != sizeof(struct lustre_disk_dqheader))
 123                 return -EINVAL;
 124         if (le32_to_cpu(dqhead.dqh_magic) != quota_magics[type] ||
 125             le32_to_cpu(dqhead.dqh_version) != quota_versions[type])
 126                 return -EINVAL;
 127         return 0;
 128 }
 129
 130 /* Check whether given file is really lustre admin quotafile */
 131 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type)
 132 {
 133         struct file *f = lqi->qi_files[type];
 134         return check_quota_file(f, NULL, type, lqi->qi_version);
 135 }
 136
 137 int lustre_read_quota_file_info(struct file* f, struct lustre_mem_dqinfo* info)
 138 {
 139         struct lustre_disk_dqinfo dinfo;
 140         ssize_t size;
 141
 142         size = lustre_read_quota(f, NULL, 0, (char *)&dinfo,
 143                                  sizeof(struct lustre_disk_dqinfo),
 144                                  LUSTRE_DQINFOOFF);
 145
 146         if (size != sizeof(struct lustre_disk_dqinfo)) {
 147                 CERROR("Can't read info structure on device %s.\n",
 148                        f->f_vfsmnt->mnt_sb->s_id);
 149                 return -EINVAL;
 150         }
 151         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 152         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 153         info->dqi_flags = le32_to_cpu(dinfo.dqi_flags);
 154         info->dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 155         info->dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 156         info->dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 157         return 0;
 158 }
 159
 160 /* Read information header from quota file */
 161 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type)
 162 {
 163         return lustre_read_quota_file_info(lqi->qi_files[type], &lqi->qi_info[type]);
 164 }
 165
 166 /* Write information header to quota file */
 167 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type)
 168 {
 169         struct lustre_disk_dqinfo dinfo;
 170         struct lustre_mem_dqinfo *info = &lqi->qi_info[type];
 171         struct file *f = lqi->qi_files[type];
 172         ssize_t size;
 173
 174         info->dqi_flags &= ~DQF_INFO_DIRTY;
 175         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 176         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 177         dinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK);
 178         dinfo.dqi_blocks = cpu_to_le32(info->dqi_blocks);
 179         dinfo.dqi_free_blk = cpu_to_le32(info->dqi_free_blk);
 180         dinfo.dqi_free_entry = cpu_to_le32(info->dqi_free_entry);
 181
 182         size = lustre_write_quota(f, (char *)&dinfo,
 183                                   sizeof(struct lustre_disk_dqinfo),
 184                                   LUSTRE_DQINFOOFF);
 185
 186         if (size != sizeof(struct lustre_disk_dqinfo)) {
 187                 CWARN("Can't write info structure on device %s.\n",
 188                       f->f_vfsmnt->mnt_sb->s_id);
 189                 return -1;
 190         }
 191         return 0;
 192 }
 193
 194 #define DQ2MQ(v) ((sizeof(v) == sizeof(__u64)) ? \
 195                 le64_to_cpu(v) : le32_to_cpu(v))
 196
 197 #define MQ2DQ(v,newv) ((sizeof(v) == sizeof(__u64)) ? \
 198                 (v = cpu_to_le64((__u64)newv)) : (v = cpu_to_le32((__u32)newv)))
 199
 200 #define DQF_GET(var,ver,field) ((ver == LUSTRE_QUOTA_V1)?\
 201                 DQ2MQ(((struct lustre_disk_dqblk*)(var))->field):\
 202                 DQ2MQ(((struct lustre_disk_dqblk_v2*)(var))->field))
 203
 204 #define DQF_PUT(var,ver,field,val) ((ver == LUSTRE_QUOTA_V1)?\
 205                 MQ2DQ(((struct lustre_disk_dqblk*)(var))->field, val):\
 206                 MQ2DQ(((struct lustre_disk_dqblk_v2*)(var))->field, val))
 207
 208 void disk2memdqb(struct lustre_mem_dqblk *m, void *d,
 209                  lustre_quota_version_t version)
 210 {
 211         m->dqb_ihardlimit = DQF_GET(d, version, dqb_ihardlimit);
 212         m->dqb_isoftlimit = DQF_GET(d, version, dqb_isoftlimit);
 213         m->dqb_curinodes = DQF_GET(d, version, dqb_curinodes);
 214         m->dqb_itime = DQF_GET(d, version, dqb_itime);
 215         m->dqb_bhardlimit = DQF_GET(d, version, dqb_bhardlimit);
 216         m->dqb_bsoftlimit = DQF_GET(d, version, dqb_bsoftlimit);
 217         m->dqb_curspace = DQF_GET(d, version, dqb_curspace);
 218         m->dqb_btime = DQF_GET(d, version, dqb_btime);
 219 }
 220
 221 static int check_quota_bounds(struct lustre_mem_dqblk *m,
 222                               lustre_quota_version_t version)
 223 {
 224         return (version == LUSTRE_QUOTA_V1  &&
 225                 m->dqb_ihardlimit <= MAX_UL &&
 226                 m->dqb_isoftlimit <= MAX_UL &&
 227                 m->dqb_curinodes <= MAX_UL  &&
 228                 m->dqb_bhardlimit <= MAX_UL &&
 229                 m->dqb_bsoftlimit <= MAX_UL) ||
 230                 version != LUSTRE_QUOTA_V1;
 231 }
 232
 233 static int mem2diskdqb(void *d, struct lustre_mem_dqblk *m,
 234                        qid_t id, lustre_quota_version_t version)
 235 {
 236         if (!check_quota_bounds(m, version))
 237                 return -EINVAL;
 238
 239         DQF_PUT(d, version, dqb_ihardlimit, m->dqb_ihardlimit);
 240         DQF_PUT(d, version, dqb_isoftlimit, m->dqb_isoftlimit);
 241         DQF_PUT(d, version, dqb_curinodes, m->dqb_curinodes);
 242         DQF_PUT(d, version, dqb_itime, m->dqb_itime);
 243         DQF_PUT(d, version, dqb_bhardlimit, m->dqb_bhardlimit);
 244         DQF_PUT(d, version, dqb_bsoftlimit, m->dqb_bsoftlimit);
 245         DQF_PUT(d, version, dqb_curspace, m->dqb_curspace);
 246         DQF_PUT(d, version, dqb_btime, m->dqb_btime);
 247         DQF_PUT(d, version, dqb_id, id);
 248
 249         return 0;
 250 }
 251
 252 dqbuf_t getdqbuf(void)
 253 {
 254         dqbuf_t buf = kmalloc(LUSTRE_DQBLKSIZE, GFP_NOFS);
 255         if (!buf)
 256                 CWARN("VFS: Not enough memory for quota buffers.\n");
 257         return buf;
 258 }
 259
 260 void freedqbuf(dqbuf_t buf)
 261 {
 262         kfree(buf);
 263 }
 264
 265 ssize_t read_blk(struct file *filp, struct inode *inode, int type,
 266                  uint blk, dqbuf_t buf)
 267 {
 268         ssize_t ret;
 269
 270         memset(buf, 0, LUSTRE_DQBLKSIZE);
 271         ret = lustre_read_quota(filp, inode, type, (char *)buf, LUSTRE_DQBLKSIZE,
 272                                 blk << LUSTRE_DQBLKSIZE_BITS);
 273
 274         /* Reading past EOF just returns a block of zeros */
 275         if (ret == -EBADR)
 276                 ret = 0;
 277
 278         return ret;
 279 }
 280
 281 ssize_t write_blk(struct file *filp, uint blk, dqbuf_t buf)
 282 {
 283         ssize_t ret;
 284
 285         ret = lustre_write_quota(filp, (char *)buf, LUSTRE_DQBLKSIZE,
 286                                  blk << LUSTRE_DQBLKSIZE_BITS);
 287
 288         return ret;
 289 }
 290
 291 void lustre_mark_info_dirty(struct lustre_mem_dqinfo *info)
 292 {
 293         set_bit(DQF_INFO_DIRTY_B, &info->dqi_flags);
 294 }
 295
 296 /* Remove empty block from list and return it */
 297 int get_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info)
 298 {
 299         dqbuf_t buf = getdqbuf();
 300         struct lustre_disk_dqdbheader *dh =
 301             (struct lustre_disk_dqdbheader *)buf;
 302         int ret, blk;
 303
 304         if (!buf)
 305                 return -ENOMEM;
 306         if (info->dqi_free_blk) {
 307                 blk = info->dqi_free_blk;
 308                 if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0)
 309                         goto out_buf;
 310                 info->dqi_free_blk = le32_to_cpu(dh->dqdh_next_free);
 311         } else {
 312                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 313                 /* Assure block allocation... */
 314                 if ((ret = write_blk(filp, info->dqi_blocks, buf)) < 0)
 315                         goto out_buf;
 316                 blk = info->dqi_blocks++;
 317         }
 318         lustre_mark_info_dirty(info);
 319         ret = blk;
 320 out_buf:
 321         freedqbuf(buf);
 322         return ret;
 323 }
 324
 325 /* Insert empty block to the list */
 326 int put_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info,
 327                    dqbuf_t buf, uint blk)
 328 {
 329         struct lustre_disk_dqdbheader *dh =
 330             (struct lustre_disk_dqdbheader *)buf;
 331         int err;
 332
 333         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_blk);
 334         dh->dqdh_prev_free = cpu_to_le32(0);
 335         dh->dqdh_entries = cpu_to_le16(0);
 336         info->dqi_free_blk = blk;
 337         lustre_mark_info_dirty(info);
 338         if ((err = write_blk(filp, blk, buf)) < 0)
 339                 /* Some strange block. We had better leave it... */
 340                 return err;
 341         return 0;
 342 }
 343
 344 /* Remove given block from the list of blocks with free entries */
 345 int remove_free_dqentry(struct file *filp,
 346                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 347                         uint blk)
 348 {
 349         dqbuf_t tmpbuf = getdqbuf();
 350         struct lustre_disk_dqdbheader *dh =
 351             (struct lustre_disk_dqdbheader *)buf;
 352         uint nextblk = le32_to_cpu(dh->dqdh_next_free), prevblk =
 353             le32_to_cpu(dh->dqdh_prev_free);
 354         int err;
 355
 356         if (!tmpbuf)
 357                 return -ENOMEM;
 358         if (nextblk) {
 359                 if ((err = read_blk(filp, NULL, 0, nextblk, tmpbuf)) < 0)
 360                         goto out_buf;
 361                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 362                     dh->dqdh_prev_free;
 363                 if ((err = write_blk(filp, nextblk, tmpbuf)) < 0)
 364                         goto out_buf;
 365         }
 366         if (prevblk) {
 367                 if ((err = read_blk(filp, NULL, 0, prevblk, tmpbuf)) < 0)
 368                         goto out_buf;
 369                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_next_free =
 370                     dh->dqdh_next_free;
 371                 if ((err = write_blk(filp, prevblk, tmpbuf)) < 0)
 372                         goto out_buf;
 373         } else {
 374                 info->dqi_free_entry = nextblk;
 375                 lustre_mark_info_dirty(info);
 376         }
 377         freedqbuf(tmpbuf);
 378         dh->dqdh_next_free = dh->dqdh_prev_free = cpu_to_le32(0);
 379         err = write_blk(filp, blk, buf);
 380         if (err < 0)      /* No matter whether write succeeds block is out of list */
 381                 CERROR("VFS: Can't write block (%u) with "
 382                        "free entries (rc=%d).\n", blk, err);
 383         return 0;
 384 out_buf:
 385         freedqbuf(tmpbuf);
 386         return err;
 387 }
 388
 389 /* Insert given block to the beginning of list with free entries */
 390 int insert_free_dqentry(struct file *filp,
 391                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 392                         uint blk)
 393 {
 394         dqbuf_t tmpbuf = getdqbuf();
 395         struct lustre_disk_dqdbheader *dh =
 396             (struct lustre_disk_dqdbheader *)buf;
 397         int err;
 398
 399         if (!tmpbuf)
 400                 return -ENOMEM;
 401         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_entry);
 402         dh->dqdh_prev_free = cpu_to_le32(0);
 403         if ((err = write_blk(filp, blk, buf)) < 0)
 404                 goto out_buf;
 405         if (info->dqi_free_entry) {
 406                 if ((err = read_blk(filp, NULL, 0, info->dqi_free_entry, tmpbuf)) < 0)
 407                         goto out_buf;
 408                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 409                     cpu_to_le32(blk);
 410                 if ((err = write_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 411                         goto out_buf;
 412         }
 413         freedqbuf(tmpbuf);
 414         info->dqi_free_entry = blk;
 415         lustre_mark_info_dirty(info);
 416         return 0;
 417 out_buf:
 418         freedqbuf(tmpbuf);
 419         return err;
 420 }
 421
 422
 423
 424 /* Find space for dquot */
 425 static uint find_free_dqentry(struct lustre_dquot *dquot, int *err,
 426                               lustre_quota_version_t version)
 427 {
 428         struct lustre_quota_info *lqi = dquot->dq_info;
 429         struct file *filp = lqi->qi_files[dquot->dq_type];
 430         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 431         uint blk, i;
 432         struct lustre_disk_dqdbheader *dh;
 433         void *ddquot;
 434         int dqblk_sz = lustre_disk_dqblk_sz[version];
 435         int dqstrinblk = lustre_dqstrinblk[version];
 436         dqbuf_t buf;
 437
 438         *err = 0;
 439         if (!(buf = getdqbuf())) {
 440                 *err = -ENOMEM;
 441                 return 0;
 442         }
 443         dh = (struct lustre_disk_dqdbheader *)buf;
 444         ddquot = GETENTRIES(buf, version);
 445         if (info->dqi_free_entry) {
 446                 blk = info->dqi_free_entry;
 447                 if ((*err = read_blk(filp, NULL, 0, blk, buf)) < 0)
 448                         goto out_buf;
 449         } else {
 450                 blk = get_free_dqblk(filp, info);
 451                 if ((int)blk < 0) {
 452                         *err = blk;
 453                         freedqbuf(buf);
 454                         return 0;
 455                 }
 456                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 457                 info->dqi_free_entry = blk; /* This is enough as block is
 458                                                already zeroed and entry list
 459                                                is empty... */
 460                 lustre_mark_info_dirty(info);
 461         }
 462
 463         /* Will block be full */
 464         if (le16_to_cpu(dh->dqdh_entries) + 1 >= dqstrinblk)
 465                 if ((*err = remove_free_dqentry(filp, info, buf, blk)) < 0) {
 466                         CERROR("VFS: Can't remove block %u"
 467                                " from entry free list.\n", blk);
 468                         goto out_buf;
 469                 }
 470         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) + 1);
 471         /* Find free structure in block */
 472         for (i = 0; i < dqstrinblk &&
 473              memcmp((char *)&emptydquot[version],
 474                     (char*)ddquot + i * dqblk_sz,
 475                     dqblk_sz); i++);
 476
 477         if (i == dqstrinblk) {
 478                 CERROR("VFS: Data block full but it shouldn't.\n");
 479                 *err = -EIO;
 480                 goto out_buf;
 481         }
 482
 483         if ((*err = write_blk(filp, blk, buf)) < 0) {
 484                 CERROR("VFS: Can't write quota data block %u.\n", blk);
 485                 goto out_buf;
 486         }
 487         dquot->dq_off =
 488             (blk << LUSTRE_DQBLKSIZE_BITS) +
 489             sizeof(struct lustre_disk_dqdbheader) +
 490             i * dqblk_sz;
 491         freedqbuf(buf);
 492         return blk;
 493 out_buf:
 494         freedqbuf(buf);
 495         return 0;
 496 }
 497
 498 /* Insert reference to structure into the trie */
 499 static int do_insert_tree(struct lustre_dquot *dquot, uint * treeblk, int depth,
 500                           lustre_quota_version_t version)
 501 {
 502         struct lustre_quota_info *lqi = dquot->dq_info;
 503         struct file *filp = lqi->qi_files[dquot->dq_type];
 504         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 505         dqbuf_t buf;
 506         int ret = 0, newson = 0, newact = 0;
 507         u32 *ref;
 508         uint newblk;
 509
 510         if (!(buf = getdqbuf()))
 511                 return -ENOMEM;
 512         if (!*treeblk) {
 513                 ret = get_free_dqblk(filp, info);
 514                 if (ret < 0)
 515                         goto out_buf;
 516                 *treeblk = ret;
 517                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 518                 newact = 1;
 519         } else {
 520                 if ((ret = read_blk(filp, NULL, 0, *treeblk, buf)) < 0) {
 521                         CERROR("VFS: Can't read tree quota block %u.\n",
 522                                *treeblk);
 523                         goto out_buf;
 524                 }
 525         }
 526         ref = (u32 *) buf;
 527         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 528         if (!newblk)
 529                 newson = 1;
 530         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 531
 532                 if (newblk) {
 533                         CERROR("VFS: Inserting already present quota entry "
 534                                "(block %u).\n",
 535                                ref[GETIDINDEX(dquot->dq_id, depth)]);
 536                         ret = -EIO;
 537                         goto out_buf;
 538                 }
 539
 540                 newblk = find_free_dqentry(dquot, &ret, version);
 541         } else
 542                 ret = do_insert_tree(dquot, &newblk, depth + 1, version);
 543         if (newson && ret >= 0) {
 544                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(newblk);
 545                 ret = write_blk(filp, *treeblk, buf);
 546         } else if (newact && ret < 0)
 547                 put_free_dqblk(filp, info, buf, *treeblk);
 548 out_buf:
 549         freedqbuf(buf);
 550         return ret;
 551 }
 552
 553 /* Wrapper for inserting quota structure into tree */
 554 static inline int dq_insert_tree(struct lustre_dquot *dquot,
 555                                  lustre_quota_version_t version)
 556 {
 557         int tmp = LUSTRE_DQTREEOFF;
 558         return do_insert_tree(dquot, &tmp, 0, version);
 559 }
 560
 561 /*
 562  *  We don't have to be afraid of deadlocks as we never have quotas on quota files...
 563  */
 564 static int lustre_write_dquot(struct lustre_dquot *dquot,
 565                               lustre_quota_version_t version)
 566 {
 567         int type = dquot->dq_type;
 568         struct file *filp;
 569         loff_t offset;
 570         ssize_t ret;
 571         int dqblk_sz = lustre_disk_dqblk_sz[version];
 572         char ddquot[sizeof(union lustre_disk_dqblk_un)];
 573
 574         ret = mem2diskdqb(ddquot, &dquot->dq_dqb, dquot->dq_id, version);
 575         if (ret < 0)
 576                 return ret;
 577
 578         if (!dquot->dq_off)
 579                 if ((ret = dq_insert_tree(dquot, version)) < 0) {
 580                         CERROR("VFS: Error %Zd occurred while creating quota.\n",
 581                                ret);
 582                         return ret;
 583                 }
 584         filp = dquot->dq_info->qi_files[type];
 585         offset = dquot->dq_off;
 586         /* Argh... We may need to write structure full of zeroes but that would be
 587          * treated as an empty place by the rest of the code. Format change would
 588          * be definitely cleaner but the problems probably are not worth it */
 589         if (!memcmp((char *)&emptydquot[version], ddquot, dqblk_sz))
 590                 DQF_PUT(ddquot, version, dqb_itime, 1);
 591
 592         ret = lustre_write_quota(filp, ddquot,
 593                                  dqblk_sz, offset);
 594         if (ret != dqblk_sz) {
 595                 CWARN("VFS: dquota write failed on dev %s\n",
 596                       filp->f_dentry->d_sb->s_id);
 597                 if (ret >= 0)
 598                         ret = -ENOSPC;
 599         } else
 600                 ret = 0;
 601
 602         return ret;
 603 }
 604
 605 /* Free dquot entry in data block */
 606 static int free_dqentry(struct lustre_dquot *dquot, uint blk,
 607                         lustre_quota_version_t version)
 608 {
 609         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 610         struct lustre_mem_dqinfo *info =
 611             &dquot->dq_info->qi_info[dquot->dq_type];
 612         struct lustre_disk_dqdbheader *dh;
 613         dqbuf_t buf = getdqbuf();
 614         int dqstrinblk = lustre_dqstrinblk[version];
 615         int ret = 0;
 616
 617         if (!buf)
 618                 return -ENOMEM;
 619         if (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS != blk) {
 620                 CERROR("VFS: Quota structure has offset to other block (%u) "
 621                        "than it should (%u).\n", blk,
 622                        (uint)(dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS));
 623                 goto out_buf;
 624         }
 625         if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0) {
 626                 CERROR("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                         CERROR("VFS: Can't move quota data block (%u) "
 635                                "to free list.\n", blk);
 636                         goto out_buf;
 637                 }
 638         } else {
 639                 memset(buf + (dquot->dq_off & ((1<<LUSTRE_DQBLKSIZE_BITS) - 1)),
 640                        0, lustre_disk_dqblk_sz[version]);
 641                 if (le16_to_cpu(dh->dqdh_entries) == dqstrinblk - 1) {
 642                         /* Insert will write block itself */
 643                         if ((ret =
 644                              insert_free_dqentry(filp, info, buf, blk)) < 0) {
 645                                 CERROR("VFS: Can't insert quota data block (%u) "
 646                                        "to free entry list.\n", blk);
 647                                 goto out_buf;
 648                         }
 649                 } else if ((ret = write_blk(filp, blk, buf)) < 0) {
 650                         CERROR("VFS: Can't write quota data block %u\n", blk);
 651                         goto out_buf;
 652                 }
 653         }
 654         dquot->dq_off = 0;      /* Quota is now unattached */
 655 out_buf:
 656         freedqbuf(buf);
 657         return ret;
 658 }
 659
 660 /* Remove reference to dquot from tree */
 661 static int remove_tree(struct lustre_dquot *dquot, uint * blk, int depth,
 662                        lustre_quota_version_t version)
 663 {
 664         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 665         struct lustre_mem_dqinfo *info =
 666             &dquot->dq_info->qi_info[dquot->dq_type];
 667         dqbuf_t buf = getdqbuf();
 668         int ret = 0;
 669         uint newblk;
 670         u32 *ref = (u32 *) buf;
 671
 672         if (!buf)
 673                 return -ENOMEM;
 674         if ((ret = read_blk(filp, NULL, 0, *blk, buf)) < 0) {
 675                 CERROR("VFS: Can't read quota data block %u\n", *blk);
 676                 goto out_buf;
 677         }
 678         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 679         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 680                 ret = free_dqentry(dquot, newblk, version);
 681                 newblk = 0;
 682         } else
 683                 ret = remove_tree(dquot, &newblk, depth + 1, version);
 684         if (ret >= 0 && !newblk) {
 685                 int i;
 686                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(0);
 687                 for (i = 0; i < LUSTRE_DQBLKSIZE && !buf[i]; i++)
 688                         /* Block got empty? */ ;
 689                 /* don't put the root block into free blk list! */
 690                 if (i == LUSTRE_DQBLKSIZE && *blk != LUSTRE_DQTREEOFF) {
 691                         put_free_dqblk(filp, info, buf, *blk);
 692                         *blk = 0;
 693                 } else if ((ret = write_blk(filp, *blk, buf)) < 0)
 694                         CERROR("VFS: Can't write quota tree block %u.\n", *blk);
 695         }
 696 out_buf:
 697         freedqbuf(buf);
 698         return ret;
 699 }
 700
 701 /* Delete dquot from tree */
 702 static int lustre_delete_dquot(struct lustre_dquot *dquot,
 703                                 lustre_quota_version_t version)
 704 {
 705         uint tmp = LUSTRE_DQTREEOFF;
 706
 707         if (!dquot->dq_off)     /* Even not allocated? */
 708                 return 0;
 709         return remove_tree(dquot, &tmp, 0, version);
 710 }
 711
 712 /* Find entry in block */
 713 static loff_t find_block_dqentry(struct lustre_dquot *dquot, uint blk,
 714                                  lustre_quota_version_t version)
 715 {
 716         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 717         dqbuf_t buf = getdqbuf();
 718         loff_t ret = 0;
 719         int i;
 720         char *ddquot = GETENTRIES(buf, version);
 721         int dqblk_sz = lustre_disk_dqblk_sz[version];
 722         int dqstrinblk = lustre_dqstrinblk[version];
 723
 724         if (!buf)
 725                 return -ENOMEM;
 726         if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0) {
 727                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 728                 goto out_buf;
 729         }
 730         if (dquot->dq_id)
 731                 for (i = 0; i < dqstrinblk &&
 732                      DQF_GET(ddquot+i*dqblk_sz, version, dqb_id) != dquot->dq_id;
 733                      i++) ;
 734         else {                  /* ID 0 as a bit more complicated searching... */
 735                 for (i = 0; i < dqstrinblk; i++)
 736                         if (!DQF_GET(ddquot + i*dqblk_sz, version, dqb_id)
 737                             && memcmp((char *)&emptydquot[version],
 738                                       ddquot + i*dqblk_sz,
 739                                       dqblk_sz))
 740                                 break;
 741         }
 742         if (i == dqstrinblk) {
 743                 CERROR("VFS: Quota for id %u referenced but not present.\n",
 744                        dquot->dq_id);
 745                 ret = -EIO;
 746                 goto out_buf;
 747         } else
 748                 ret =
 749                     (blk << LUSTRE_DQBLKSIZE_BITS) +
 750                     sizeof(struct lustre_disk_dqdbheader) +
 751                     i * dqblk_sz;
 752 out_buf:
 753         freedqbuf(buf);
 754         return ret;
 755 }
 756
 757 /* Find entry for given id in the tree */
 758 static loff_t find_tree_dqentry(struct lustre_dquot *dquot, uint blk, int depth,
 759                                 lustre_quota_version_t version)
 760 {
 761         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 762         dqbuf_t buf = getdqbuf();
 763         loff_t ret = 0;
 764         u32 *ref = (u32 *) buf;
 765
 766         if (!buf)
 767                 return -ENOMEM;
 768         if ((ret = read_blk(filp, NULL, 0, blk, buf)) < 0) {
 769                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 770                 goto out_buf;
 771         }
 772         ret = 0;
 773         blk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 774         if (!blk)               /* No reference? */
 775                 goto out_buf;
 776         if (depth < LUSTRE_DQTREEDEPTH - 1)
 777                 ret = find_tree_dqentry(dquot, blk, depth + 1, version);
 778         else
 779                 ret = find_block_dqentry(dquot, blk, version);
 780 out_buf:
 781         freedqbuf(buf);
 782         return ret;
 783 }
 784
 785 /* Find entry for given id in the tree - wrapper function */
 786 static inline loff_t find_dqentry(struct lustre_dquot *dquot,
 787                                   lustre_quota_version_t version)
 788 {
 789         return find_tree_dqentry(dquot, LUSTRE_DQTREEOFF, 0, version);
 790 }
 791
 792
 793 int lustre_read_dquot(struct lustre_dquot *dquot)
 794 {
 795         int type = dquot->dq_type;
 796         struct file *filp;
 797         loff_t offset;
 798         int ret = 0, dqblk_sz;
 799         lustre_quota_version_t version;
 800
 801         /* Invalidated quota? */
 802         if (!dquot->dq_info || !(filp = dquot->dq_info->qi_files[type])) {
 803                 CERROR("VFS: Quota invalidated while reading!\n");
 804                 return -EIO;
 805         }
 806
 807         version = dquot->dq_info->qi_version;
 808         dqblk_sz = lustre_disk_dqblk_sz[version];
 809
 810         offset = find_dqentry(dquot, version);
 811         if (offset <= 0) {      /* Entry not present? */
 812                 if (offset < 0)
 813                         CERROR("VFS: Can't read quota structure for id %u.\n",
 814                                dquot->dq_id);
 815                 dquot->dq_off = 0;
 816                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
 817                 memset(&dquot->dq_dqb, 0, sizeof(struct lustre_mem_dqblk));
 818                 ret = offset;
 819         } else {
 820                 char ddquot[sizeof(union lustre_disk_dqblk_un)];
 821
 822                 dquot->dq_off = offset;
 823                 if ((ret = lustre_read_quota(filp, NULL, type, ddquot,
 824                                              dqblk_sz, offset)) != dqblk_sz) {
 825                         if (ret >= 0)
 826                                 ret = -EIO;
 827                         CERROR("VFS: Error while reading quota structure "
 828                                "for id %u.\n", dquot->dq_id);
 829                         memset(ddquot, 0, dqblk_sz);
 830                 } else {
 831                         ret = 0;
 832                         /* We need to escape back all-zero structure */
 833                         if (!memcmp((char *)&fakedquot[version],
 834                                     ddquot, dqblk_sz))
 835                                 DQF_PUT(ddquot, version, dqb_itime, 0);
 836                 }
 837                 disk2memdqb(&dquot->dq_dqb, ddquot, version);
 838         }
 839
 840         return ret;
 841 }
 842
 843 /* Commit changes of dquot to disk - it might also mean deleting it when quota became fake */
 844 int lustre_commit_dquot(struct lustre_dquot *dquot)
 845 {
 846         int rc = 0;
 847         lustre_quota_version_t version = dquot->dq_info->qi_version;
 848         void *handle;
 849         struct inode *inode = dquot->dq_info->qi_files[dquot->dq_type]->f_dentry->d_inode;
 850
 851         /* always clear the flag so we don't loop on an IO error... */
 852         clear_bit(DQ_MOD_B, &dquot->dq_flags);
 853
 854         /* The block/inode usage in admin quotafile isn't the real usage
 855          * over all cluster, so keep the fake dquot entry on disk is
 856          * meaningless, just remove it */
 857         if (test_bit(DQ_FAKE_B, &dquot->dq_flags)) {
 858                 handle = lustre_quota_journal_start(inode, 1);
 859                 rc = lustre_delete_dquot(dquot, version);
 860                 lustre_quota_journal_stop(handle);
 861         } else {
 862                 handle = lustre_quota_journal_start(inode, 0);
 863                 rc = lustre_write_dquot(dquot, version);
 864                 lustre_quota_journal_stop(handle);
 865         }
 866
 867         if (rc < 0)
 868                 return rc;
 869
 870         if (lustre_info_dirty(&dquot->dq_info->qi_info[dquot->dq_type]))
 871                 rc = lustre_write_quota_info(dquot->dq_info, dquot->dq_type);
 872
 873         return rc;
 874 }
 875
 876 int lustre_init_quota_header(struct lustre_quota_info *lqi, int type, int fakemagics)
 877 {
 878         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 879         static const uint fake_magics[] = LUSTRE_BADQMAGICS;
 880         const uint* quota_versions = lustre_initqversions[lqi->qi_version];
 881         struct lustre_disk_dqheader dqhead;
 882         ssize_t size;
 883         struct file *fp = lqi->qi_files[type];
 884         int rc = 0;
 885
 886         /* write quotafile header */
 887         dqhead.dqh_magic = cpu_to_le32(fakemagics ?
 888                                        fake_magics[type] : quota_magics[type]);
 889         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
 890         size = lustre_write_quota(fp, (char *)&dqhead,
 891                                   sizeof(struct lustre_disk_dqheader), 0);
 892
 893         if (size != sizeof(struct lustre_disk_dqheader)) {
 894                 CERROR("error writing quotafile header (rc:%d)\n", rc);
 895                 rc = size;
 896         }
 897
 898         return rc;
 899 }
 900
 901 /* We need to export this function to initialize quotafile, because we haven't
 902  * user level check utility */
 903 int lustre_init_quota_info_generic(struct lustre_quota_info *lqi, int type,
 904                                    int fakemagics)
 905 {
 906         struct lustre_mem_dqinfo *dqinfo = &lqi->qi_info[type];
 907         int rc;
 908
 909         rc = lustre_init_quota_header(lqi, type, fakemagics);
 910         if (rc)
 911                 return rc;
 912
 913         /* write init quota info */
 914         memset(dqinfo, 0, sizeof(*dqinfo));
 915         dqinfo->dqi_bgrace = MAX_DQ_TIME;
 916         dqinfo->dqi_igrace = MAX_IQ_TIME;
 917         dqinfo->dqi_blocks = LUSTRE_DQTREEOFF + 1;
 918
 919         return lustre_write_quota_info(lqi, type);
 920 }
 921
 922 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type)
 923 {
 924         return lustre_init_quota_info_generic(lqi, type, 0);
 925 }
 926
 927 static int walk_block_dqentry(struct file *filp, struct inode *inode, int type,
 928                               uint blk, struct list_head *list)
 929 {
 930         dqbuf_t buf = getdqbuf();
 931         loff_t ret = 0;
 932         struct lustre_disk_dqdbheader *dqhead =
 933             (struct lustre_disk_dqdbheader *)buf;
 934         struct dqblk *blk_item;
 935         struct dqblk *pos;
 936         struct list_head *tmp;
 937
 938         if (!buf)
 939                 return -ENOMEM;
 940         if ((ret = read_blk(filp, inode, type, blk, buf)) < 0) {
 941                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 942                 goto out_buf;
 943         }
 944         ret = 0;
 945
 946         if (!le32_to_cpu(dqhead->dqdh_entries))
 947                 goto out_buf;
 948
 949         if (list_empty(list)) {
 950                 tmp = list;
 951                 goto done;
 952         }
 953
 954         list_for_each_entry(pos, list, link) {
 955                 if (blk == pos->blk)    /* we got this blk already */
 956                         goto out_buf;
 957                 if (blk > pos->blk)
 958                         continue;
 959                 break;
 960         }
 961         tmp = &pos->link;
 962 done:
 963         blk_item = kmalloc(sizeof(*blk_item), GFP_NOFS);
 964         if (!blk_item) {
 965                 ret = -ENOMEM;
 966                 goto out_buf;
 967         }
 968         blk_item->blk = blk;
 969         INIT_LIST_HEAD(&blk_item->link);
 970
 971         list_add_tail(&blk_item->link, tmp);
 972
 973 out_buf:
 974         freedqbuf(buf);
 975         return ret;
 976 }
 977
 978 int walk_tree_dqentry(struct file *filp, struct inode *inode, int type,
 979                       uint blk, int depth, struct list_head *list)
 980 {
 981         dqbuf_t buf = getdqbuf();
 982         loff_t ret = 0;
 983         int index;
 984         u32 *ref = (u32 *) buf;
 985
 986         if (!buf)
 987                 return -ENOMEM;
 988         if ((ret = read_blk(filp, inode, type, blk, buf)) < 0) {
 989                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 990                 goto out_buf;
 991         }
 992         ret = 0;
 993
 994         for (index = 0; index <= 0xff && !ret; index++) {
 995                 blk = le32_to_cpu(ref[index]);
 996                 if (!blk)       /* No reference */
 997                         continue;
 998
 999                 if (depth < LUSTRE_DQTREEDEPTH - 1)
1000                         ret = walk_tree_dqentry(filp, inode, type, blk,
1001                                                 depth + 1, list);
1002                 else
1003                         ret = walk_block_dqentry(filp, inode, type, blk, list);
1004         }
1005 out_buf:
1006         freedqbuf(buf);
1007         return ret;
1008 }
1009
1010 /* Walk through the quota file (v2 format) to get all ids with quota limit */
1011 int lustre_get_qids(struct file *fp, struct inode *inode, int type,
1012                     struct list_head *list)
1013 {
1014         struct list_head blk_list;
1015         struct dqblk *blk_item, *tmp;
1016         dqbuf_t buf = NULL;
1017         char *ddquot;
1018         int rc;
1019         lustre_quota_version_t version;
1020
1021         ENTRY;
1022
1023         LASSERT(ergo(fp == NULL, inode != NULL));
1024
1025         if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V1) == 0)
1026                 version = LUSTRE_QUOTA_V1;
1027         else if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V2) == 0)
1028                 version = LUSTRE_QUOTA_V2;
1029         else {
1030                 CERROR("unknown quota file format!\n");
1031                 RETURN(-EINVAL);
1032         }
1033
1034         if (!list_empty(list)) {
1035                 CERROR("not empty list\n");
1036                 RETURN(-EINVAL);
1037         }
1038
1039         INIT_LIST_HEAD(&blk_list);
1040         rc = walk_tree_dqentry(fp, inode, type, LUSTRE_DQTREEOFF, 0, &blk_list);
1041         if (rc) {
1042                 CERROR("walk through quota file failed!(%d)\n", rc);
1043                 GOTO(out_free, rc);
1044         }
1045         if (list_empty(&blk_list))
1046                 RETURN(0);
1047
1048         buf = getdqbuf();
1049         if (!buf)
1050                 RETURN(-ENOMEM);
1051         ddquot = GETENTRIES(buf, version);
1052
1053         list_for_each_entry(blk_item, &blk_list, link) {
1054                 loff_t ret = 0;
1055                 int i, dqblk_sz = lustre_disk_dqblk_sz[version];
1056
1057                 memset(buf, 0, LUSTRE_DQBLKSIZE);
1058                 if ((ret = read_blk(fp, inode, type, blk_item->blk, buf)) < 0) {
1059                         CERROR("VFS: Can't read quota tree block %u.\n",
1060                                blk_item->blk);
1061                         GOTO(out_free, rc = ret);
1062                 }
1063
1064                 for (i = 0; i < lustre_dqstrinblk[version]; i++) {
1065                         struct dquot_id *dqid;
1066                         /* skip empty entry */
1067                         if (!memcmp((char *)&emptydquot[version],
1068                                     ddquot + i*dqblk_sz, dqblk_sz))
1069                                 continue;
1070
1071                         OBD_ALLOC_GFP(dqid, sizeof(*dqid), GFP_NOFS);
1072                         if (!dqid)
1073                                 GOTO(out_free, rc = -ENOMEM);
1074
1075                         dqid->di_id = DQF_GET(ddquot + i * dqblk_sz,
1076                                               version, dqb_id);
1077                         dqid->di_flag = DQF_GET(ddquot + i * dqblk_sz, version,
1078                                                 dqb_ihardlimit) ? QI_SET : 0;
1079                         dqid->di_flag |= DQF_GET(ddquot + i * dqblk_sz, version,
1080                                                  dqb_bhardlimit) ? QB_SET : 0;
1081                         INIT_LIST_HEAD(&dqid->di_link);
1082                         list_add(&dqid->di_link, list);
1083                 }
1084         }
1085
1086 out_free:
1087         list_for_each_entry_safe(blk_item, tmp, &blk_list, link) {
1088                 list_del_init(&blk_item->link);
1089                 kfree(blk_item);
1090         }
1091         if (buf)
1092                 freedqbuf(buf);
1093
1094         RETURN(rc);
1095 }
1096
1097
1098 EXPORT_SYMBOL(lustre_read_quota_info);
1099 EXPORT_SYMBOL(lustre_write_quota_info);
1100 EXPORT_SYMBOL(lustre_check_quota_file);
1101 EXPORT_SYMBOL(lustre_read_dquot);
1102 EXPORT_SYMBOL(lustre_commit_dquot);
1103 EXPORT_SYMBOL(lustre_init_quota_info);
1104 EXPORT_SYMBOL(lustre_get_qids);
1105 #endif