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 #include <linux/quotaio_v1.h>
  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 int check_quota_file(struct file *f, struct inode *inode, int type,
  77                      lustre_quota_version_t version)
  78 {
  79         struct lustre_disk_dqheader dqhead;
  80         mm_segment_t fs;
  81         ssize_t size;
  82         loff_t offset = 0;
  83         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
  84         const uint *quota_versions = lustre_initqversions[version];
  85
  86         if (f) {
  87                 fs = get_fs();
  88                 set_fs(KERNEL_DS);
  89                 size = f->f_op->read(f, (char *)&dqhead,
  90                                      sizeof(struct lustre_disk_dqheader),
  91                                      &offset);
  92                 set_fs(fs);
  93         } else {
  94 #ifndef KERNEL_SUPPORTS_QUOTA_READ
  95                 size = 0;
  96 #else
  97                 struct super_block *sb = inode->i_sb;
  98                 size = sb->s_op->quota_read(sb, type, (char *)&dqhead,
  99                                             sizeof(struct lustre_disk_dqheader), 0);
 100 #endif
 101         }
 102         if (size != sizeof(struct lustre_disk_dqheader))
 103                 return -EINVAL;
 104         if (le32_to_cpu(dqhead.dqh_magic) != quota_magics[type] ||
 105             le32_to_cpu(dqhead.dqh_version) != quota_versions[type])
 106                 return -EINVAL;
 107         return 0;
 108 }
 109
 110 /**
 111  * Check whether given file is really lustre admin quotafile
 112  */
 113 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type)
 114 {
 115         struct file *f = lqi->qi_files[type];
 116         return check_quota_file(f, NULL, type, lqi->qi_version);
 117 }
 118
 119 int lustre_read_quota_file_info(struct file* f, struct lustre_mem_dqinfo* info)
 120 {
 121         mm_segment_t fs;
 122         struct lustre_disk_dqinfo dinfo;
 123         ssize_t size;
 124         loff_t offset = LUSTRE_DQINFOOFF;
 125
 126         fs = get_fs();
 127         set_fs(KERNEL_DS);
 128         size = f->f_op->read(f, (char *)&dinfo,
 129                              sizeof(struct lustre_disk_dqinfo), &offset);
 130         set_fs(fs);
 131         if (size != sizeof(struct lustre_disk_dqinfo)) {
 132                 CDEBUG(D_ERROR, "Can't read info structure on device %s.\n",
 133                        f->f_vfsmnt->mnt_sb->s_id);
 134                 return -EINVAL;
 135         }
 136         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 137         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 138         info->dqi_flags = le32_to_cpu(dinfo.dqi_flags);
 139         info->dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 140         info->dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 141         info->dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 142         return 0;
 143 }
 144
 145 /**
 146  * Read information header from quota file
 147  */
 148 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type)
 149 {
 150         return lustre_read_quota_file_info(lqi->qi_files[type], &lqi->qi_info[type]);
 151 }
 152
 153 /**
 154  * Write information header to quota file
 155  */
 156 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type)
 157 {
 158         mm_segment_t fs;
 159         struct lustre_disk_dqinfo dinfo;
 160         struct lustre_mem_dqinfo *info = &lqi->qi_info[type];
 161         struct file *f = lqi->qi_files[type];
 162         ssize_t size;
 163         loff_t offset = LUSTRE_DQINFOOFF;
 164
 165         info->dqi_flags &= ~DQF_INFO_DIRTY;
 166         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 167         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 168         dinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK);
 169         dinfo.dqi_blocks = cpu_to_le32(info->dqi_blocks);
 170         dinfo.dqi_free_blk = cpu_to_le32(info->dqi_free_blk);
 171         dinfo.dqi_free_entry = cpu_to_le32(info->dqi_free_entry);
 172         fs = get_fs();
 173         set_fs(KERNEL_DS);
 174         size = f->f_op->write(f, (char *)&dinfo,
 175                               sizeof(struct lustre_disk_dqinfo), &offset);
 176         set_fs(fs);
 177         if (size != sizeof(struct lustre_disk_dqinfo)) {
 178                 CDEBUG(D_WARNING,
 179                        "Can't write info structure on device %s.\n",
 180                        f->f_vfsmnt->mnt_sb->s_id);
 181                 return -1;
 182         }
 183         return 0;
 184 }
 185
 186 void disk2memdqb(struct lustre_mem_dqblk *m, void *d,
 187                  lustre_quota_version_t version)
 188 {
 189         struct lustre_disk_dqblk_v2 *dqblk = (struct lustre_disk_dqblk_v2 *)d;
 190
 191         LASSERT(version == LUSTRE_QUOTA_V2);
 192
 193         m->dqb_ihardlimit = le64_to_cpu(dqblk->dqb_ihardlimit);
 194         m->dqb_isoftlimit = le64_to_cpu(dqblk->dqb_isoftlimit);
 195         m->dqb_curinodes = le64_to_cpu(dqblk->dqb_curinodes);
 196         m->dqb_itime = le64_to_cpu(dqblk->dqb_itime);
 197         m->dqb_bhardlimit = le64_to_cpu(dqblk->dqb_bhardlimit);
 198         m->dqb_bsoftlimit = le64_to_cpu(dqblk->dqb_bsoftlimit);
 199         m->dqb_curspace = le64_to_cpu(dqblk->dqb_curspace);
 200         m->dqb_btime = le64_to_cpu(dqblk->dqb_btime);
 201 }
 202
 203 static int mem2diskdqb(void *d, struct lustre_mem_dqblk *m,
 204                        qid_t id, lustre_quota_version_t version)
 205 {
 206         struct lustre_disk_dqblk_v2 *dqblk = (struct lustre_disk_dqblk_v2 *)d;
 207
 208         LASSERT(version == LUSTRE_QUOTA_V2);
 209
 210         dqblk->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
 211         dqblk->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
 212         dqblk->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
 213         dqblk->dqb_itime = cpu_to_le64(m->dqb_itime);
 214         dqblk->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
 215         dqblk->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
 216         dqblk->dqb_curspace = cpu_to_le64(m->dqb_curspace);
 217         dqblk->dqb_btime = cpu_to_le64(m->dqb_btime);
 218         dqblk->dqb_id = cpu_to_le32(id);
 219
 220         return 0;
 221 }
 222
 223 dqbuf_t getdqbuf(void)
 224 {
 225         dqbuf_t buf = kmalloc(LUSTRE_DQBLKSIZE, GFP_NOFS);
 226         if (!buf)
 227                 CDEBUG(D_WARNING,
 228                        "VFS: Not enough memory for quota buffers.\n");
 229         return buf;
 230 }
 231
 232 void freedqbuf(dqbuf_t buf)
 233 {
 234         kfree(buf);
 235 }
 236
 237 ssize_t read_blk(struct file *filp, uint blk, dqbuf_t buf)
 238 {
 239         mm_segment_t fs;
 240         ssize_t ret;
 241         loff_t offset = blk << LUSTRE_DQBLKSIZE_BITS;
 242
 243         memset(buf, 0, LUSTRE_DQBLKSIZE);
 244         fs = get_fs();
 245         set_fs(KERNEL_DS);
 246         ret = filp->f_op->read(filp, (char *)buf, LUSTRE_DQBLKSIZE, &offset);
 247         set_fs(fs);
 248         return ret;
 249 }
 250
 251 ssize_t write_blk(struct file *filp, uint blk, dqbuf_t buf)
 252 {
 253         mm_segment_t fs;
 254         ssize_t ret;
 255         loff_t offset = blk << LUSTRE_DQBLKSIZE_BITS;
 256
 257         fs = get_fs();
 258         set_fs(KERNEL_DS);
 259         ret = filp->f_op->write(filp, (char *)buf, LUSTRE_DQBLKSIZE, &offset);
 260         set_fs(fs);
 261         return ret;
 262 }
 263
 264 void lustre_mark_info_dirty(struct lustre_mem_dqinfo *info)
 265 {
 266         set_bit(DQF_INFO_DIRTY_B, &info->dqi_flags);
 267 }
 268
 269 /**
 270  * Remove empty block from list and return it
 271  */
 272 int get_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info)
 273 {
 274         dqbuf_t buf = getdqbuf();
 275         struct lustre_disk_dqdbheader *dh =
 276             (struct lustre_disk_dqdbheader *)buf;
 277         int ret, blk;
 278
 279         if (!buf)
 280                 return -ENOMEM;
 281         if (info->dqi_free_blk) {
 282                 blk = info->dqi_free_blk;
 283                 if ((ret = read_blk(filp, blk, buf)) < 0)
 284                         goto out_buf;
 285                 info->dqi_free_blk = le32_to_cpu(dh->dqdh_next_free);
 286         } else {
 287                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 288                 /* Assure block allocation... */
 289                 if ((ret = write_blk(filp, info->dqi_blocks, buf)) < 0)
 290                         goto out_buf;
 291                 blk = info->dqi_blocks++;
 292         }
 293         lustre_mark_info_dirty(info);
 294         ret = blk;
 295 out_buf:
 296         freedqbuf(buf);
 297         return ret;
 298 }
 299
 300 /**
 301  * Insert empty block to the list
 302  */
 303 int put_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info,
 304                    dqbuf_t buf, uint blk)
 305 {
 306         struct lustre_disk_dqdbheader *dh =
 307             (struct lustre_disk_dqdbheader *)buf;
 308         int err;
 309
 310         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_blk);
 311         dh->dqdh_prev_free = cpu_to_le32(0);
 312         dh->dqdh_entries = cpu_to_le16(0);
 313         info->dqi_free_blk = blk;
 314         lustre_mark_info_dirty(info);
 315         if ((err = write_blk(filp, blk, buf)) < 0)
 316                 /* Some strange block. We had better leave it... */
 317                 return err;
 318         return 0;
 319 }
 320
 321 /**
 322  * Remove given block from the list of blocks with free entries
 323  */
 324 int remove_free_dqentry(struct file *filp,
 325                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 326                         uint blk)
 327 {
 328         dqbuf_t tmpbuf = getdqbuf();
 329         struct lustre_disk_dqdbheader *dh =
 330             (struct lustre_disk_dqdbheader *)buf;
 331         uint nextblk = le32_to_cpu(dh->dqdh_next_free), prevblk =
 332             le32_to_cpu(dh->dqdh_prev_free);
 333         int err;
 334
 335         if (!tmpbuf)
 336                 return -ENOMEM;
 337         if (nextblk) {
 338                 if ((err = read_blk(filp, nextblk, tmpbuf)) < 0)
 339                         goto out_buf;
 340                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 341                     dh->dqdh_prev_free;
 342                 if ((err = write_blk(filp, nextblk, tmpbuf)) < 0)
 343                         goto out_buf;
 344         }
 345         if (prevblk) {
 346                 if ((err = read_blk(filp, prevblk, tmpbuf)) < 0)
 347                         goto out_buf;
 348                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_next_free =
 349                     dh->dqdh_next_free;
 350                 if ((err = write_blk(filp, prevblk, tmpbuf)) < 0)
 351                         goto out_buf;
 352         } else {
 353                 info->dqi_free_entry = nextblk;
 354                 lustre_mark_info_dirty(info);
 355         }
 356         freedqbuf(tmpbuf);
 357         dh->dqdh_next_free = dh->dqdh_prev_free = cpu_to_le32(0);
 358         if (write_blk(filp, blk, buf) < 0)
 359                 /* No matter whether write succeeds block is out of list */
 360                 CDEBUG(D_ERROR,
 361                        "VFS: Can't write block (%u) with free entries.\n", blk);
 362         return 0;
 363 out_buf:
 364         freedqbuf(tmpbuf);
 365         return err;
 366 }
 367
 368 /**
 369  * Insert given block to the beginning of list with free entries
 370  */
 371 int insert_free_dqentry(struct file *filp,
 372                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 373                         uint blk)
 374 {
 375         dqbuf_t tmpbuf = getdqbuf();
 376         struct lustre_disk_dqdbheader *dh =
 377             (struct lustre_disk_dqdbheader *)buf;
 378         int err;
 379
 380         if (!tmpbuf)
 381                 return -ENOMEM;
 382         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_entry);
 383         dh->dqdh_prev_free = cpu_to_le32(0);
 384         if ((err = write_blk(filp, blk, buf)) < 0)
 385                 goto out_buf;
 386         if (info->dqi_free_entry) {
 387                 if ((err = read_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 388                         goto out_buf;
 389                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 390                     cpu_to_le32(blk);
 391                 if ((err = write_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 392                         goto out_buf;
 393         }
 394         freedqbuf(tmpbuf);
 395         info->dqi_free_entry = blk;
 396         lustre_mark_info_dirty(info);
 397         return 0;
 398 out_buf:
 399         freedqbuf(tmpbuf);
 400         return err;
 401 }
 402
 403
 404
 405 /**
 406  * Find space for dquot
 407  */
 408 static uint find_free_dqentry(struct lustre_dquot *dquot, int *err,
 409                               lustre_quota_version_t version)
 410 {
 411         struct lustre_quota_info *lqi = dquot->dq_info;
 412         struct file *filp = lqi->qi_files[dquot->dq_type];
 413         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 414         uint blk, i;
 415         struct lustre_disk_dqdbheader *dh;
 416         void *ddquot;
 417         int dqblk_sz = lustre_disk_dqblk_sz[version];
 418         int dqstrinblk = lustre_dqstrinblk[version];
 419         char fakedquot[dqblk_sz];
 420         dqbuf_t buf;
 421
 422         *err = 0;
 423         if (!(buf = getdqbuf())) {
 424                 *err = -ENOMEM;
 425                 return 0;
 426         }
 427         dh = (struct lustre_disk_dqdbheader *)buf;
 428         ddquot = GETENTRIES(buf, version);
 429         if (info->dqi_free_entry) {
 430                 blk = info->dqi_free_entry;
 431                 if ((*err = read_blk(filp, blk, buf)) < 0)
 432                         goto out_buf;
 433         } else {
 434                 blk = get_free_dqblk(filp, info);
 435                 if ((int)blk < 0) {
 436                         *err = blk;
 437                         freedqbuf(buf);
 438                         return 0;
 439                 }
 440                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 441                 info->dqi_free_entry = blk; /* This is enough as block is
 442                                                already zeroed and entry list
 443                                                is empty... */
 444                 lustre_mark_info_dirty(info);
 445         }
 446
 447         /* Will block be full */
 448         if (le16_to_cpu(dh->dqdh_entries) + 1 >= dqstrinblk)
 449                 if ((*err = remove_free_dqentry(filp, info, buf, blk)) < 0) {
 450                         CDEBUG(D_ERROR,
 451                                "VFS: find_free_dqentry(): Can't remove block (%u) from entry free list.\n",
 452                                blk);
 453                         goto out_buf;
 454                 }
 455         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) + 1);
 456         memset(fakedquot, 0, dqblk_sz);
 457         /* Find free structure in block */
 458         for (i = 0; i < dqstrinblk &&
 459              memcmp(fakedquot, (char*)ddquot + i * dqblk_sz,
 460                     sizeof(fakedquot)); i++);
 461
 462         if (i == dqstrinblk) {
 463                 CDEBUG(D_ERROR,
 464                        "VFS: find_free_dqentry(): Data block full but it shouldn't.\n");
 465                 *err = -EIO;
 466                 goto out_buf;
 467         }
 468
 469         if ((*err = write_blk(filp, blk, buf)) < 0) {
 470                 CDEBUG(D_ERROR,
 471                        "VFS: find_free_dqentry(): Can't write quota data block %u.\n",
 472                        blk);
 473                 goto out_buf;
 474         }
 475         dquot->dq_off =
 476             (blk << LUSTRE_DQBLKSIZE_BITS) +
 477             sizeof(struct lustre_disk_dqdbheader) +
 478             i * dqblk_sz;
 479         freedqbuf(buf);
 480         return blk;
 481 out_buf:
 482         freedqbuf(buf);
 483         return 0;
 484 }
 485
 486 /**
 487  * Insert reference to structure into the trie
 488  */
 489 static int do_insert_tree(struct lustre_dquot *dquot, uint * treeblk, int depth,
 490                           lustre_quota_version_t version)
 491 {
 492         struct lustre_quota_info *lqi = dquot->dq_info;
 493         struct file *filp = lqi->qi_files[dquot->dq_type];
 494         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 495         dqbuf_t buf;
 496         int ret = 0, newson = 0, newact = 0;
 497         u32 *ref;
 498         uint newblk;
 499
 500         if (!(buf = getdqbuf()))
 501                 return -ENOMEM;
 502         if (!*treeblk) {
 503                 ret = get_free_dqblk(filp, info);
 504                 if (ret < 0)
 505                         goto out_buf;
 506                 *treeblk = ret;
 507                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 508                 newact = 1;
 509         } else {
 510                 if ((ret = read_blk(filp, *treeblk, buf)) < 0) {
 511                         CDEBUG(D_ERROR,
 512                                "VFS: Can't read tree quota block %u.\n",
 513                                *treeblk);
 514                         goto out_buf;
 515                 }
 516         }
 517         ref = (u32 *) buf;
 518         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 519         if (!newblk)
 520                 newson = 1;
 521         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 522
 523                 if (newblk) {
 524                         CDEBUG(D_ERROR,
 525                                "VFS: Inserting already present quota entry (block %u).\n",
 526                                ref[GETIDINDEX(dquot->dq_id, depth)]);
 527                         ret = -EIO;
 528                         goto out_buf;
 529                 }
 530
 531                 newblk = find_free_dqentry(dquot, &ret, version);
 532         } else
 533                 ret = do_insert_tree(dquot, &newblk, depth + 1, version);
 534         if (newson && ret >= 0) {
 535                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(newblk);
 536                 ret = write_blk(filp, *treeblk, buf);
 537         } else if (newact && ret < 0)
 538                 put_free_dqblk(filp, info, buf, *treeblk);
 539 out_buf:
 540         freedqbuf(buf);
 541         return ret;
 542 }
 543
 544 /**
 545  * Wrapper for inserting quota structure into tree
 546  */
 547 static inline int dq_insert_tree(struct lustre_dquot *dquot,
 548                                  lustre_quota_version_t version)
 549 {
 550         int tmp = LUSTRE_DQTREEOFF;
 551         return do_insert_tree(dquot, &tmp, 0, version);
 552 }
 553
 554 /**
 555  * We don't have to be afraid of deadlocks as we never have quotas on
 556  * quota files...
 557  */
 558 static int lustre_write_dquot(struct lustre_dquot *dquot,
 559                               lustre_quota_version_t version)
 560 {
 561         int type = dquot->dq_type;
 562         struct file *filp;
 563         mm_segment_t fs;
 564         loff_t offset;
 565         ssize_t ret;
 566         int dqblk_sz = lustre_disk_dqblk_sz[version];
 567         char ddquot[dqblk_sz], empty[dqblk_sz];
 568
 569         ret = mem2diskdqb(ddquot, &dquot->dq_dqb, dquot->dq_id, version);
 570         if (ret < 0)
 571                 return ret;
 572
 573         if (!dquot->dq_off)
 574                 if ((ret = dq_insert_tree(dquot, version)) < 0) {
 575                         CDEBUG(D_ERROR,
 576                                "VFS: Error %Zd occurred while creating quota.\n",
 577                                ret);
 578                         return ret;
 579                 }
 580         filp = dquot->dq_info->qi_files[type];
 581         offset = dquot->dq_off;
 582         /* Argh... We may need to write structure full of zeroes but that would be
 583          * treated as an empty place by the rest of the code. Format change would
 584          * be definitely cleaner but the problems probably are not worth it */
 585         memset(empty, 0, dqblk_sz);
 586         if (!memcmp(empty, ddquot, dqblk_sz))
 587                 ((struct lustre_disk_dqblk_v2 *)ddquot)->dqb_itime = cpu_to_le64(1);
 588         fs = get_fs();
 589         set_fs(KERNEL_DS);
 590         ret = filp->f_op->write(filp, ddquot,
 591                                 dqblk_sz, &offset);
 592         set_fs(fs);
 593         if (ret != dqblk_sz) {
 594                 CDEBUG(D_WARNING, "VFS: dquota write failed on dev %s\n",
 595                        filp->f_dentry->d_sb->s_id);
 596                 if (ret >= 0)
 597                         ret = -ENOSPC;
 598         } else
 599                 ret = 0;
 600
 601         return ret;
 602 }
 603
 604 /**
 605  * Free dquot entry in data block
 606  */
 607 static int free_dqentry(struct lustre_dquot *dquot, uint blk,
 608                         lustre_quota_version_t version)
 609 {
 610         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 611         struct lustre_mem_dqinfo *info =
 612             &dquot->dq_info->qi_info[dquot->dq_type];
 613         struct lustre_disk_dqdbheader *dh;
 614         dqbuf_t buf = getdqbuf();
 615         int dqstrinblk = lustre_dqstrinblk[version];
 616         int ret = 0;
 617
 618         if (!buf)
 619                 return -ENOMEM;
 620         if (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS != blk) {
 621                 CDEBUG(D_ERROR,
 622                        "VFS: Quota structure has offset to other block (%u) than it should (%u).\n",
 623                        blk, (uint) (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS));
 624                 goto out_buf;
 625         }
 626         if ((ret = read_blk(filp, blk, buf)) < 0) {
 627                 CDEBUG(D_ERROR, "VFS: Can't read quota data block %u\n", blk);
 628                 goto out_buf;
 629         }
 630         dh = (struct lustre_disk_dqdbheader *)buf;
 631         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) - 1);
 632         if (!le16_to_cpu(dh->dqdh_entries)) {   /* Block got free? */
 633                 if ((ret = remove_free_dqentry(filp, info, buf, blk)) < 0 ||
 634                     (ret = put_free_dqblk(filp, info, buf, blk)) < 0) {
 635                         CDEBUG(D_ERROR,
 636                                "VFS: Can't move quota data block (%u) to free list.\n",
 637                                blk);
 638                         goto out_buf;
 639                 }
 640         } else {
 641                 memset(buf + (dquot->dq_off & ((1<<LUSTRE_DQBLKSIZE_BITS) - 1)),
 642                        0, lustre_disk_dqblk_sz[version]);
 643                 if (le16_to_cpu(dh->dqdh_entries) == dqstrinblk - 1) {
 644                         /* Insert will write block itself */
 645                         if ((ret =
 646                              insert_free_dqentry(filp, info, buf, blk)) < 0) {
 647                                 CDEBUG(D_ERROR,
 648                                        "VFS: Can't insert quota data block (%u) to free entry list.\n",
 649                                        blk);
 650                                 goto out_buf;
 651                         }
 652                 } else if ((ret = write_blk(filp, blk, buf)) < 0) {
 653                         CDEBUG(D_ERROR,
 654                                "VFS: Can't write quota data block %u\n", blk);
 655                         goto out_buf;
 656                 }
 657         }
 658         dquot->dq_off = 0;      /* Quota is now unattached */
 659 out_buf:
 660         freedqbuf(buf);
 661         return ret;
 662 }
 663
 664 /**
 665  * Remove reference to dquot from tree
 666  */
 667 static int remove_tree(struct lustre_dquot *dquot, uint * blk, int depth,
 668                        lustre_quota_version_t version)
 669 {
 670         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 671         struct lustre_mem_dqinfo *info =
 672             &dquot->dq_info->qi_info[dquot->dq_type];
 673         dqbuf_t buf = getdqbuf();
 674         int ret = 0;
 675         uint newblk;
 676         u32 *ref = (u32 *) buf;
 677
 678         if (!buf)
 679                 return -ENOMEM;
 680         if ((ret = read_blk(filp, *blk, buf)) < 0) {
 681                 CDEBUG(D_ERROR, "VFS: Can't read quota data block %u\n", *blk);
 682                 goto out_buf;
 683         }
 684         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 685         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 686                 ret = free_dqentry(dquot, newblk, version);
 687                 newblk = 0;
 688         } else
 689                 ret = remove_tree(dquot, &newblk, depth + 1, version);
 690         if (ret >= 0 && !newblk) {
 691                 int i;
 692                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(0);
 693                 for (i = 0; i < LUSTRE_DQBLKSIZE && !buf[i]; i++)
 694                         /* Block got empty? */ ;
 695                 /* don't put the root block into free blk list! */
 696                 if (i == LUSTRE_DQBLKSIZE && *blk != LUSTRE_DQTREEOFF) {
 697                         put_free_dqblk(filp, info, buf, *blk);
 698                         *blk = 0;
 699                 } else if ((ret = write_blk(filp, *blk, buf)) < 0)
 700                         CDEBUG(D_ERROR,
 701                                "VFS: Can't write quota tree block %u.\n", *blk);
 702         }
 703 out_buf:
 704         freedqbuf(buf);
 705         return ret;
 706 }
 707
 708 /**
 709  * Delete dquot from tree
 710  */
 711 static int lustre_delete_dquot(struct lustre_dquot *dquot,
 712                                 lustre_quota_version_t version)
 713 {
 714         uint tmp = LUSTRE_DQTREEOFF;
 715
 716         if (!dquot->dq_off)     /* Even not allocated? */
 717                 return 0;
 718         return remove_tree(dquot, &tmp, 0, version);
 719 }
 720
 721 /**
 722  * Find entry in block
 723  */
 724 static loff_t find_block_dqentry(struct lustre_dquot *dquot, uint blk,
 725                                  lustre_quota_version_t version)
 726 {
 727         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 728         dqbuf_t buf = getdqbuf();
 729         loff_t ret = 0;
 730         int i;
 731         struct lustre_disk_dqblk_v2 *ddquot = (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, blk, buf)) < 0) {
 740                 CDEBUG(D_ERROR, "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                 char fakedquot[dqblk_sz];
 749
 750                 memset(fakedquot, 0, sizeof(fakedquot));
 751                 for (i = 0; i < dqstrinblk; i++)
 752                         if (!le32_to_cpu(ddquot[i].dqb_id)
 753                             && memcmp(fakedquot, ddquot + i,
 754                                       dqblk_sz))
 755                                 break;
 756         }
 757         if (i == dqstrinblk) {
 758                 CDEBUG(D_ERROR,
 759                        "VFS: Quota for id %u referenced but not present.\n",
 760                        dquot->dq_id);
 761                 ret = -EIO;
 762                 goto out_buf;
 763         } else
 764                 ret =
 765                     (blk << LUSTRE_DQBLKSIZE_BITS) +
 766                     sizeof(struct lustre_disk_dqdbheader) +
 767                     i * dqblk_sz;
 768 out_buf:
 769         freedqbuf(buf);
 770         return ret;
 771 }
 772
 773 /**
 774  * Find entry for given id in the tree
 775  */
 776 static loff_t find_tree_dqentry(struct lustre_dquot *dquot, uint blk, int depth,
 777                                 lustre_quota_version_t version)
 778 {
 779         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 780         dqbuf_t buf = getdqbuf();
 781         loff_t ret = 0;
 782         u32 *ref = (u32 *) buf;
 783
 784         if (!buf)
 785                 return -ENOMEM;
 786         if ((ret = read_blk(filp, blk, buf)) < 0) {
 787                 CDEBUG(D_ERROR, "VFS: Can't read quota tree block %u.\n", blk);
 788                 goto out_buf;
 789         }
 790         ret = 0;
 791         blk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 792         if (!blk)               /* No reference? */
 793                 goto out_buf;
 794         if (depth < LUSTRE_DQTREEDEPTH - 1)
 795                 ret = find_tree_dqentry(dquot, blk, depth + 1, version);
 796         else
 797                 ret = find_block_dqentry(dquot, blk, version);
 798 out_buf:
 799         freedqbuf(buf);
 800         return ret;
 801 }
 802
 803 /**
 804  * Find entry for given id in the tree - wrapper function
 805  */
 806 static inline loff_t find_dqentry(struct lustre_dquot *dquot,
 807                                   lustre_quota_version_t version)
 808 {
 809         return find_tree_dqentry(dquot, LUSTRE_DQTREEOFF, 0, version);
 810 }
 811
 812 int lustre_read_dquot(struct lustre_dquot *dquot)
 813 {
 814         int type = dquot->dq_type;
 815         struct file *filp;
 816         mm_segment_t fs;
 817         loff_t offset;
 818         int ret = 0, dqblk_sz;
 819         lustre_quota_version_t version;
 820
 821         /* Invalidated quota? */
 822         if (!dquot->dq_info || !(filp = dquot->dq_info->qi_files[type])) {
 823                 CDEBUG(D_ERROR, "VFS: Quota invalidated while reading!\n");
 824                 return -EIO;
 825         }
 826
 827         version = dquot->dq_info->qi_version;
 828         LASSERT(version == LUSTRE_QUOTA_V2);
 829         dqblk_sz = lustre_disk_dqblk_sz[version];
 830
 831         offset = find_dqentry(dquot, version);
 832         if (offset <= 0) {      /* Entry not present? */
 833                 if (offset < 0)
 834                         CDEBUG(D_ERROR,
 835                                "VFS: Can't read quota structure for id %u.\n",
 836                                dquot->dq_id);
 837                 dquot->dq_off = 0;
 838                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
 839                 memset(&dquot->dq_dqb, 0, sizeof(struct lustre_mem_dqblk));
 840                 ret = offset;
 841         } else {
 842                 char ddquot[dqblk_sz], empty[dqblk_sz];
 843
 844                 dquot->dq_off = offset;
 845                 fs = get_fs();
 846                 set_fs(KERNEL_DS);
 847                 if ((ret = filp->f_op->read(filp, ddquot, dqblk_sz, &offset)) !=
 848                     dqblk_sz) {
 849                         if (ret >= 0)
 850                                 ret = -EIO;
 851                         CDEBUG(D_ERROR,
 852                                "VFS: Error while reading quota structure for id %u.\n",
 853                                dquot->dq_id);
 854                         memset(ddquot, 0, dqblk_sz);
 855                 } else {
 856                         ret = 0;
 857                         /* We need to escape back all-zero structure */
 858                         memset(empty, 0, dqblk_sz);
 859                         ((struct lustre_disk_dqblk_v2 *)empty)->dqb_itime = cpu_to_le64(1);
 860                         if (!memcmp(empty, ddquot, dqblk_sz))
 861                                 ((struct lustre_disk_dqblk_v2 *)empty)->dqb_itime = cpu_to_le64(0);
 862                 }
 863                 set_fs(fs);
 864                 disk2memdqb(&dquot->dq_dqb, ddquot, version);
 865         }
 866
 867         return ret;
 868 }
 869
 870 /**
 871  * Commit changes of dquot to disk - it might also mean deleting
 872  * it when quota became fake.
 873  */
 874 int lustre_commit_dquot(struct lustre_dquot *dquot)
 875 {
 876         int rc = 0;
 877         lustre_quota_version_t version = dquot->dq_info->qi_version;
 878
 879         /* always clear the flag so we don't loop on an IO error... */
 880         clear_bit(DQ_MOD_B, &dquot->dq_flags);
 881
 882         /* The block/inode usage in admin quotafile isn't the real usage
 883          * over all cluster, so keep the fake dquot entry on disk is
 884          * meaningless, just remove it */
 885         if (test_bit(DQ_FAKE_B, &dquot->dq_flags))
 886                 rc = lustre_delete_dquot(dquot, version);
 887         else
 888                 rc = lustre_write_dquot(dquot, version);
 889
 890         if (rc < 0)
 891                 return rc;
 892
 893         if (lustre_info_dirty(&dquot->dq_info->qi_info[dquot->dq_type]))
 894                 rc = lustre_write_quota_info(dquot->dq_info, dquot->dq_type);
 895
 896         return rc;
 897 }
 898
 899 int lustre_init_quota_header(struct lustre_quota_info *lqi, int type, int fakemagics)
 900 {
 901         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 902         static const uint fake_magics[] = LUSTRE_BADQMAGICS;
 903         const uint* quota_versions = lustre_initqversions[lqi->qi_version];
 904         struct lustre_disk_dqheader dqhead;
 905         ssize_t size;
 906         loff_t offset = 0;
 907         struct file *fp = lqi->qi_files[type];
 908         int rc = 0;
 909
 910         /* write quotafile header */
 911         dqhead.dqh_magic = cpu_to_le32(fakemagics ?
 912                                        fake_magics[type] : quota_magics[type]);
 913         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
 914         size = fp->f_op->write(fp, (char *)&dqhead,
 915                                sizeof(struct lustre_disk_dqheader), &offset);
 916
 917         if (size != sizeof(struct lustre_disk_dqheader)) {
 918                 CDEBUG(D_ERROR, "error writing quoafile header (rc:%d)\n", rc);
 919                 rc = size;
 920         }
 921
 922         return rc;
 923 }
 924
 925 /**
 926  * We need to export this function to initialize quotafile, because we haven't
 927  * user level check utility
 928  */
 929 int lustre_init_quota_info_generic(struct lustre_quota_info *lqi, int type,
 930                                    int fakemagics)
 931 {
 932         struct lustre_mem_dqinfo *dqinfo = &lqi->qi_info[type];
 933         int rc;
 934
 935         rc = lustre_init_quota_header(lqi, type, fakemagics);
 936         if (rc)
 937                 return rc;
 938
 939         /* write init quota info */
 940         memset(dqinfo, 0, sizeof(*dqinfo));
 941         dqinfo->dqi_bgrace = MAX_DQ_TIME;
 942         dqinfo->dqi_igrace = MAX_IQ_TIME;
 943         dqinfo->dqi_blocks = LUSTRE_DQTREEOFF + 1;
 944
 945         return lustre_write_quota_info(lqi, type);
 946 }
 947
 948 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type)
 949 {
 950         return lustre_init_quota_info_generic(lqi, type, 0);
 951 }
 952
 953 ssize_t quota_read(struct file *file, struct inode *inode, int type,
 954                    uint blk, dqbuf_t buf)
 955 {
 956         if (file) {
 957                 return read_blk(file, blk, buf);
 958         } else {
 959 #ifndef KERNEL_SUPPORTS_QUOTA_READ
 960                 return -ENOTSUPP;
 961 #else
 962                 struct super_block *sb = inode->i_sb;
 963                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 964                 return sb->s_op->quota_read(sb, type, (char *)buf,
 965                                             LUSTRE_DQBLKSIZE,
 966                                             blk << LUSTRE_DQBLKSIZE_BITS);
 967 #endif
 968         }
 969 }
 970
 971 static int walk_block_dqentry(struct file *filp, struct inode *inode, int type,
 972                               uint blk, struct list_head *list)
 973 {
 974         dqbuf_t buf = getdqbuf();
 975         loff_t ret = 0;
 976         struct lustre_disk_dqdbheader *dqhead =
 977             (struct lustre_disk_dqdbheader *)buf;
 978         struct dqblk *blk_item;
 979         struct dqblk *pos;
 980         struct list_head *tmp;
 981
 982         if (!buf)
 983                 return -ENOMEM;
 984         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
 985                 CDEBUG(D_ERROR, "VFS: Can't read quota tree block %u.\n", blk);
 986                 goto out_buf;
 987         }
 988         ret = 0;
 989
 990         if (!le32_to_cpu(dqhead->dqdh_entries))
 991                 goto out_buf;
 992
 993         if (list_empty(list)) {
 994                 tmp = list;
 995                 goto done;
 996         }
 997
 998         list_for_each_entry(pos, list, link) {
 999                 if (blk == pos->blk)    /* we got this blk already */
1000                         goto out_buf;
1001                 if (blk > pos->blk)
1002                         continue;
1003                 break;
1004         }
1005         tmp = &pos->link;
1006 done:
1007         blk_item = kmalloc(sizeof(*blk_item), GFP_NOFS);
1008         if (!blk_item) {
1009                 ret = -ENOMEM;
1010                 goto out_buf;
1011         }
1012         blk_item->blk = blk;
1013         INIT_LIST_HEAD(&blk_item->link);
1014
1015         list_add_tail(&blk_item->link, tmp);
1016
1017 out_buf:
1018         freedqbuf(buf);
1019         return ret;
1020 }
1021
1022 int walk_tree_dqentry(struct file *filp, struct inode *inode, int type,
1023                       uint blk, int depth, struct list_head *list)
1024 {
1025         dqbuf_t buf = getdqbuf();
1026         loff_t ret = 0;
1027         int index;
1028         u32 *ref = (u32 *) buf;
1029
1030         if (!buf)
1031                 return -ENOMEM;
1032         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
1033                 CDEBUG(D_ERROR, "VFS: Can't read quota tree block %u.\n", blk);
1034                 goto out_buf;
1035         }
1036         ret = 0;
1037
1038         for (index = 0; index <= 0xff && !ret; index++) {
1039                 blk = le32_to_cpu(ref[index]);
1040                 if (!blk)       /* No reference */
1041                         continue;
1042
1043                 if (depth < LUSTRE_DQTREEDEPTH - 1)
1044                         ret = walk_tree_dqentry(filp, inode, type, blk,
1045                                                 depth + 1, list);
1046                 else
1047                         ret = walk_block_dqentry(filp, inode, type, blk, list);
1048         }
1049 out_buf:
1050         freedqbuf(buf);
1051         return ret;
1052 }
1053
1054 /**
1055  * Walk through the quota file (v2 format) to get all ids with quota limit
1056  */
1057 int lustre_get_qids(struct file *fp, struct inode *inode, int type,
1058                     struct list_head *list)
1059 {
1060         struct list_head blk_list;
1061         struct dqblk *blk_item, *tmp;
1062         dqbuf_t buf = NULL;
1063         struct lustre_disk_dqblk_v2 *ddquot;
1064         int rc;
1065         lustre_quota_version_t version;
1066
1067         ENTRY;
1068
1069         if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V2) == 0)
1070                 version = LUSTRE_QUOTA_V2;
1071         else {
1072                 CDEBUG(D_ERROR, "unknown quota file format!\n");
1073                 RETURN(-EINVAL);
1074         }
1075
1076         if (!list_empty(list)) {
1077                 CDEBUG(D_ERROR, "not empty list\n");
1078                 RETURN(-EINVAL);
1079         }
1080
1081         INIT_LIST_HEAD(&blk_list);
1082         rc = walk_tree_dqentry(fp, inode, type, LUSTRE_DQTREEOFF, 0, &blk_list);
1083         if (rc) {
1084                 CDEBUG(D_ERROR, "walk through quota file failed!(%d)\n", rc);
1085                 GOTO(out_free, rc);
1086         }
1087         if (list_empty(&blk_list))
1088                 RETURN(0);
1089
1090         buf = getdqbuf();
1091         if (!buf)
1092                 RETURN(-ENOMEM);
1093         ddquot = (struct lustre_disk_dqblk_v2 *)GETENTRIES(buf, version);
1094
1095         list_for_each_entry(blk_item, &blk_list, link) {
1096                 loff_t ret = 0;
1097                 int i, dqblk_sz = lustre_disk_dqblk_sz[version];
1098                 char fakedquot[dqblk_sz];
1099
1100                 memset(buf, 0, LUSTRE_DQBLKSIZE);
1101                 if ((ret = quota_read(fp, inode, type, blk_item->blk, buf))<0) {
1102                         CDEBUG(D_ERROR,
1103                                "VFS: Can't read quota tree block %u.\n",
1104                                blk_item->blk);
1105                         GOTO(out_free, rc = ret);
1106                 }
1107
1108                 memset(fakedquot, 0, dqblk_sz);
1109                 for (i = 0; i < lustre_dqstrinblk[version]; i++) {
1110                         struct dquot_id *dqid;
1111                         /* skip empty entry */
1112                         if (!memcmp(fakedquot, ddquot + i, dqblk_sz))
1113                                 continue;
1114
1115                         dqid = kmalloc(sizeof(*dqid), GFP_NOFS);
1116                         if (!dqid)
1117                                 GOTO(out_free, rc = -ENOMEM);
1118
1119                         dqid->di_id = le32_to_cpu(ddquot[i].dqb_id);
1120                         INIT_LIST_HEAD(&dqid->di_link);
1121                         list_add(&dqid->di_link, list);
1122                 }
1123         }
1124
1125 out_free:
1126         list_for_each_entry_safe(blk_item, tmp, &blk_list, link) {
1127                 list_del_init(&blk_item->link);
1128                 kfree(blk_item);
1129         }
1130         if (buf)
1131                 freedqbuf(buf);
1132
1133         RETURN(rc);
1134 }
1135
1136
1137 EXPORT_SYMBOL(lustre_read_quota_info);
1138 EXPORT_SYMBOL(lustre_write_quota_info);
1139 EXPORT_SYMBOL(lustre_check_quota_file);
1140 EXPORT_SYMBOL(lustre_read_dquot);
1141 EXPORT_SYMBOL(lustre_commit_dquot);
1142 EXPORT_SYMBOL(lustre_init_quota_info);
1143 EXPORT_SYMBOL(lustre_get_qids);
1144 #endif