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 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 int check_quota_file(struct file *f, struct inode *inode, int type,
  91                      lustre_quota_version_t version)
  92 {
  93         struct lustre_disk_dqheader dqhead;
  94         mm_segment_t fs;
  95         ssize_t size;
  96         loff_t offset = 0;
  97         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
  98         const uint *quota_versions = lustre_initqversions[version];
  99
 100         if (f) {
 101                 fs = get_fs();
 102                 set_fs(KERNEL_DS);
 103                 size = f->f_op->read(f, (char *)&dqhead,
 104                                      sizeof(struct lustre_disk_dqheader),
 105                                      &offset);
 106                 set_fs(fs);
 107         } else {
 108 #ifndef KERNEL_SUPPORTS_QUOTA_READ
 109                 size = 0;
 110 #else
 111                 struct super_block *sb = inode->i_sb;
 112                 size = sb->s_op->quota_read(sb, type, (char *)&dqhead,
 113                                             sizeof(struct lustre_disk_dqheader),
 114                                             0);
 115 #endif
 116         }
 117         if (size != sizeof(struct lustre_disk_dqheader))
 118                 return -EINVAL;
 119         if (le32_to_cpu(dqhead.dqh_magic) != quota_magics[type] ||
 120             le32_to_cpu(dqhead.dqh_version) != quota_versions[type])
 121                 return -EINVAL;
 122         return 0;
 123 }
 124
 125 /**
 126  * Check whether given file is really lustre admin quotafile
 127  */
 128 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type)
 129 {
 130         struct file *f = lqi->qi_files[type];
 131         return check_quota_file(f, NULL, type, lqi->qi_version);
 132 }
 133
 134 int lustre_read_quota_file_info(struct file* f, struct lustre_mem_dqinfo* info)
 135 {
 136         mm_segment_t fs;
 137         struct lustre_disk_dqinfo dinfo;
 138         ssize_t size;
 139         loff_t offset = LUSTRE_DQINFOOFF;
 140
 141         fs = get_fs();
 142         set_fs(KERNEL_DS);
 143         size = f->f_op->read(f, (char *)&dinfo,
 144                              sizeof(struct lustre_disk_dqinfo), &offset);
 145         set_fs(fs);
 146         if (size != sizeof(struct lustre_disk_dqinfo)) {
 147                 CDEBUG(D_ERROR, "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 /**
 161  * Read information header from quota file
 162  */
 163 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type)
 164 {
 165         return lustre_read_quota_file_info(lqi->qi_files[type],
 166                                            &lqi->qi_info[type]);
 167 }
 168
 169 /**
 170  * Write information header to quota file
 171  */
 172 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type)
 173 {
 174         mm_segment_t fs;
 175         struct lustre_disk_dqinfo dinfo;
 176         struct lustre_mem_dqinfo *info = &lqi->qi_info[type];
 177         struct file *f = lqi->qi_files[type];
 178         ssize_t size;
 179         loff_t offset = LUSTRE_DQINFOOFF;
 180
 181         info->dqi_flags &= ~DQF_INFO_DIRTY;
 182         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 183         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 184         dinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK);
 185         dinfo.dqi_blocks = cpu_to_le32(info->dqi_blocks);
 186         dinfo.dqi_free_blk = cpu_to_le32(info->dqi_free_blk);
 187         dinfo.dqi_free_entry = cpu_to_le32(info->dqi_free_entry);
 188         fs = get_fs();
 189         set_fs(KERNEL_DS);
 190         size = f->f_op->write(f, (char *)&dinfo,
 191                               sizeof(struct lustre_disk_dqinfo), &offset);
 192         set_fs(fs);
 193         if (size != sizeof(struct lustre_disk_dqinfo)) {
 194                 CDEBUG(D_WARNING,
 195                        "Can't write info structure on device %s.\n",
 196                        f->f_vfsmnt->mnt_sb->s_id);
 197                 return -1;
 198         }
 199         return 0;
 200 }
 201
 202 void disk2memdqb(struct lustre_mem_dqblk *m, void *d,
 203                  lustre_quota_version_t version)
 204 {
 205         struct lustre_disk_dqblk_v2 *dqblk = (struct lustre_disk_dqblk_v2 *)d;
 206
 207         LASSERT(version == LUSTRE_QUOTA_V2);
 208
 209         m->dqb_ihardlimit = le64_to_cpu(dqblk->dqb_ihardlimit);
 210         m->dqb_isoftlimit = le64_to_cpu(dqblk->dqb_isoftlimit);
 211         m->dqb_curinodes = le64_to_cpu(dqblk->dqb_curinodes);
 212         m->dqb_itime = le64_to_cpu(dqblk->dqb_itime);
 213         m->dqb_bhardlimit = le64_to_cpu(dqblk->dqb_bhardlimit);
 214         m->dqb_bsoftlimit = le64_to_cpu(dqblk->dqb_bsoftlimit);
 215         m->dqb_curspace = le64_to_cpu(dqblk->dqb_curspace);
 216         m->dqb_btime = le64_to_cpu(dqblk->dqb_btime);
 217 }
 218
 219 static int mem2diskdqb(void *d, struct lustre_mem_dqblk *m,
 220                        qid_t id, lustre_quota_version_t version)
 221 {
 222         struct lustre_disk_dqblk_v2 *dqblk = (struct lustre_disk_dqblk_v2 *)d;
 223
 224         LASSERT(version == LUSTRE_QUOTA_V2);
 225
 226         dqblk->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
 227         dqblk->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
 228         dqblk->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
 229         dqblk->dqb_itime = cpu_to_le64(m->dqb_itime);
 230         dqblk->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
 231         dqblk->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
 232         dqblk->dqb_curspace = cpu_to_le64(m->dqb_curspace);
 233         dqblk->dqb_btime = cpu_to_le64(m->dqb_btime);
 234         dqblk->dqb_id = cpu_to_le32(id);
 235
 236         return 0;
 237 }
 238
 239 dqbuf_t getdqbuf(void)
 240 {
 241         dqbuf_t buf = kmalloc(LUSTRE_DQBLKSIZE, GFP_NOFS);
 242         if (!buf)
 243                 CDEBUG(D_WARNING,
 244                        "VFS: Not enough memory for quota buffers.\n");
 245         return buf;
 246 }
 247
 248 void freedqbuf(dqbuf_t buf)
 249 {
 250         kfree(buf);
 251 }
 252
 253 ssize_t read_blk(struct file *filp, uint blk, dqbuf_t buf)
 254 {
 255         mm_segment_t fs;
 256         ssize_t ret;
 257         loff_t offset = blk << LUSTRE_DQBLKSIZE_BITS;
 258
 259         memset(buf, 0, LUSTRE_DQBLKSIZE);
 260         fs = get_fs();
 261         set_fs(KERNEL_DS);
 262         ret = filp->f_op->read(filp, (char *)buf, LUSTRE_DQBLKSIZE, &offset);
 263         set_fs(fs);
 264         return ret;
 265 }
 266
 267 ssize_t write_blk(struct file *filp, uint blk, dqbuf_t buf)
 268 {
 269         mm_segment_t fs;
 270         ssize_t ret;
 271         loff_t offset = blk << LUSTRE_DQBLKSIZE_BITS;
 272
 273         fs = get_fs();
 274         set_fs(KERNEL_DS);
 275         ret = filp->f_op->write(filp, (char *)buf, LUSTRE_DQBLKSIZE, &offset);
 276         set_fs(fs);
 277         return ret;
 278 }
 279
 280 void lustre_mark_info_dirty(struct lustre_mem_dqinfo *info)
 281 {
 282         set_bit(DQF_INFO_DIRTY_B, &info->dqi_flags);
 283 }
 284
 285 /**
 286  * Remove empty block from list and return it
 287  */
 288 int get_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info)
 289 {
 290         dqbuf_t buf = getdqbuf();
 291         struct lustre_disk_dqdbheader *dh =
 292             (struct lustre_disk_dqdbheader *)buf;
 293         int ret, blk;
 294
 295         if (!buf)
 296                 return -ENOMEM;
 297         if (info->dqi_free_blk) {
 298                 blk = info->dqi_free_blk;
 299                 if ((ret = read_blk(filp, blk, buf)) < 0)
 300                         goto out_buf;
 301                 info->dqi_free_blk = le32_to_cpu(dh->dqdh_next_free);
 302         } else {
 303                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 304                 /* Assure block allocation... */
 305                 if ((ret = write_blk(filp, info->dqi_blocks, buf)) < 0)
 306                         goto out_buf;
 307                 blk = info->dqi_blocks++;
 308         }
 309         lustre_mark_info_dirty(info);
 310         ret = blk;
 311 out_buf:
 312         freedqbuf(buf);
 313         return ret;
 314 }
 315
 316 /**
 317  * Insert empty block to the list
 318  */
 319 int put_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info,
 320                    dqbuf_t buf, uint blk)
 321 {
 322         struct lustre_disk_dqdbheader *dh =
 323             (struct lustre_disk_dqdbheader *)buf;
 324         int err;
 325
 326         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_blk);
 327         dh->dqdh_prev_free = cpu_to_le32(0);
 328         dh->dqdh_entries = cpu_to_le16(0);
 329         info->dqi_free_blk = blk;
 330         lustre_mark_info_dirty(info);
 331         if ((err = write_blk(filp, blk, buf)) < 0)
 332                 /* Some strange block. We had better leave it... */
 333                 return err;
 334         return 0;
 335 }
 336
 337 /**
 338  * Remove given block from the list of blocks with free entries
 339  */
 340 int remove_free_dqentry(struct file *filp,
 341                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 342                         uint blk)
 343 {
 344         dqbuf_t tmpbuf = getdqbuf();
 345         struct lustre_disk_dqdbheader *dh =
 346             (struct lustre_disk_dqdbheader *)buf;
 347         uint nextblk = le32_to_cpu(dh->dqdh_next_free), prevblk =
 348             le32_to_cpu(dh->dqdh_prev_free);
 349         int err;
 350
 351         if (!tmpbuf)
 352                 return -ENOMEM;
 353         if (nextblk) {
 354                 if ((err = read_blk(filp, nextblk, tmpbuf)) < 0)
 355                         goto out_buf;
 356                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 357                     dh->dqdh_prev_free;
 358                 if ((err = write_blk(filp, nextblk, tmpbuf)) < 0)
 359                         goto out_buf;
 360         }
 361         if (prevblk) {
 362                 if ((err = read_blk(filp, prevblk, tmpbuf)) < 0)
 363                         goto out_buf;
 364                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_next_free =
 365                     dh->dqdh_next_free;
 366                 if ((err = write_blk(filp, prevblk, tmpbuf)) < 0)
 367                         goto out_buf;
 368         } else {
 369                 info->dqi_free_entry = nextblk;
 370                 lustre_mark_info_dirty(info);
 371         }
 372         freedqbuf(tmpbuf);
 373         dh->dqdh_next_free = dh->dqdh_prev_free = cpu_to_le32(0);
 374         if (write_blk(filp, blk, buf) < 0)
 375                 /* No matter whether write succeeds block is out of list */
 376                 CDEBUG(D_ERROR,
 377                        "VFS: Can't write block (%u) with free entries.\n", blk);
 378         return 0;
 379 out_buf:
 380         freedqbuf(tmpbuf);
 381         return err;
 382 }
 383
 384 /**
 385  * Insert given block to the beginning of list with free entries
 386  */
 387 int insert_free_dqentry(struct file *filp,
 388                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 389                         uint blk)
 390 {
 391         dqbuf_t tmpbuf = getdqbuf();
 392         struct lustre_disk_dqdbheader *dh =
 393             (struct lustre_disk_dqdbheader *)buf;
 394         int err;
 395
 396         if (!tmpbuf)
 397                 return -ENOMEM;
 398         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_entry);
 399         dh->dqdh_prev_free = cpu_to_le32(0);
 400         if ((err = write_blk(filp, blk, buf)) < 0)
 401                 goto out_buf;
 402         if (info->dqi_free_entry) {
 403                 if ((err = read_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 404                         goto out_buf;
 405                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 406                     cpu_to_le32(blk);
 407                 if ((err = write_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 408                         goto out_buf;
 409         }
 410         freedqbuf(tmpbuf);
 411         info->dqi_free_entry = blk;
 412         lustre_mark_info_dirty(info);
 413         return 0;
 414 out_buf:
 415         freedqbuf(tmpbuf);
 416         return err;
 417 }
 418
 419
 420
 421 /**
 422  * Find space for dquot
 423  */
 424 static uint find_free_dqentry(struct lustre_dquot *dquot, int *err,
 425                               lustre_quota_version_t version)
 426 {
 427         struct lustre_quota_info *lqi = dquot->dq_info;
 428         struct file *filp = lqi->qi_files[dquot->dq_type];
 429         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 430         uint blk, i;
 431         struct lustre_disk_dqdbheader *dh;
 432         void *ddquot;
 433         int dqblk_sz = lustre_disk_dqblk_sz[version];
 434         int dqstrinblk = lustre_dqstrinblk[version];
 435         dqbuf_t buf;
 436
 437         *err = 0;
 438         if (!(buf = getdqbuf())) {
 439                 *err = -ENOMEM;
 440                 return 0;
 441         }
 442         dh = (struct lustre_disk_dqdbheader *)buf;
 443         ddquot = GETENTRIES(buf, version);
 444         if (info->dqi_free_entry) {
 445                 blk = info->dqi_free_entry;
 446                 if ((*err = read_blk(filp, blk, buf)) < 0)
 447                         goto out_buf;
 448         } else {
 449                 blk = get_free_dqblk(filp, info);
 450                 if ((int)blk < 0) {
 451                         *err = blk;
 452                         freedqbuf(buf);
 453                         return 0;
 454                 }
 455                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 456                 info->dqi_free_entry = blk; /* This is enough as block is
 457                                                already zeroed and entry list
 458                                                is empty... */
 459                 lustre_mark_info_dirty(info);
 460         }
 461
 462         /* Will block be full */
 463         if (le16_to_cpu(dh->dqdh_entries) + 1 >= dqstrinblk)
 464                 if ((*err = remove_free_dqentry(filp, info, buf, blk)) < 0) {
 465                         CDEBUG(D_ERROR,
 466                                "VFS: find_free_dqentry(): Can't remove block "
 467                                "(%u) 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, dqblk_sz);
 475              i++);
 476
 477         if (i == dqstrinblk) {
 478                 CDEBUG(D_ERROR,
 479                        "VFS: find_free_dqentry(): Data block full but it "
 480                        "shouldn't.\n");
 481                 *err = -EIO;
 482                 goto out_buf;
 483         }
 484
 485         if ((*err = write_blk(filp, blk, buf)) < 0) {
 486                 CDEBUG(D_ERROR,
 487                        "VFS: find_free_dqentry(): Can't write quota data "
 488                        "block %u.\n", blk);
 489                 goto out_buf;
 490         }
 491         dquot->dq_off =
 492             (blk << LUSTRE_DQBLKSIZE_BITS) +
 493             sizeof(struct lustre_disk_dqdbheader) +
 494             i * dqblk_sz;
 495         freedqbuf(buf);
 496         return blk;
 497 out_buf:
 498         freedqbuf(buf);
 499         return 0;
 500 }
 501
 502 /**
 503  * Insert reference to structure into the trie
 504  */
 505 static int do_insert_tree(struct lustre_dquot *dquot, uint * treeblk, int depth,
 506                           lustre_quota_version_t version)
 507 {
 508         struct lustre_quota_info *lqi = dquot->dq_info;
 509         struct file *filp = lqi->qi_files[dquot->dq_type];
 510         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 511         dqbuf_t buf;
 512         int ret = 0, newson = 0, newact = 0;
 513         u32 *ref;
 514         uint newblk;
 515
 516         if (!(buf = getdqbuf()))
 517                 return -ENOMEM;
 518         if (!*treeblk) {
 519                 ret = get_free_dqblk(filp, info);
 520                 if (ret < 0)
 521                         goto out_buf;
 522                 *treeblk = ret;
 523                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 524                 newact = 1;
 525         } else {
 526                 if ((ret = read_blk(filp, *treeblk, buf)) < 0) {
 527                         CDEBUG(D_ERROR,
 528                                "VFS: Can't read tree quota block %u.\n",
 529                                *treeblk);
 530                         goto out_buf;
 531                 }
 532         }
 533         ref = (u32 *) buf;
 534         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 535         if (!newblk)
 536                 newson = 1;
 537         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 538
 539                 if (newblk) {
 540                         CDEBUG(D_ERROR,
 541                                "VFS: Inserting already present quota entry "
 542                                "(block %u).\n",
 543                                ref[GETIDINDEX(dquot->dq_id, depth)]);
 544                         ret = -EIO;
 545                         goto out_buf;
 546                 }
 547
 548                 newblk = find_free_dqentry(dquot, &ret, version);
 549         } else
 550                 ret = do_insert_tree(dquot, &newblk, depth + 1, version);
 551         if (newson && ret >= 0) {
 552                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(newblk);
 553                 ret = write_blk(filp, *treeblk, buf);
 554         } else if (newact && ret < 0)
 555                 put_free_dqblk(filp, info, buf, *treeblk);
 556 out_buf:
 557         freedqbuf(buf);
 558         return ret;
 559 }
 560
 561 /**
 562  * Wrapper for inserting quota structure into tree
 563  */
 564 static inline int dq_insert_tree(struct lustre_dquot *dquot,
 565                                  lustre_quota_version_t version)
 566 {
 567         int tmp = LUSTRE_DQTREEOFF;
 568         return do_insert_tree(dquot, &tmp, 0, version);
 569 }
 570
 571 /**
 572  * We don't have to be afraid of deadlocks as we never have quotas on
 573  * quota files...
 574  */
 575 static int lustre_write_dquot(struct lustre_dquot *dquot,
 576                               lustre_quota_version_t version)
 577 {
 578         int type = dquot->dq_type;
 579         struct file *filp;
 580         mm_segment_t fs;
 581         loff_t offset;
 582         ssize_t ret;
 583         int dqblk_sz = lustre_disk_dqblk_sz[version];
 584         char ddquot[dqblk_sz];
 585
 586         ret = mem2diskdqb(ddquot, &dquot->dq_dqb, dquot->dq_id, version);
 587         if (ret < 0)
 588                 return ret;
 589
 590         if (!dquot->dq_off)
 591                 if ((ret = dq_insert_tree(dquot, version)) < 0) {
 592                         CDEBUG(D_ERROR,
 593                                "VFS: Error %Zd occurred while creating "
 594                                "quota.\n", ret);
 595                         return ret;
 596                 }
 597         filp = dquot->dq_info->qi_files[type];
 598         offset = dquot->dq_off;
 599         /* Argh... We may need to write structure full of zeroes but that would
 600          * be treated as an empty place by the rest of the code. Format change
 601          * would be definitely cleaner but the problems probably are not worth
 602          * it */
 603         if (!memcmp((char *)&emptydquot[version], ddquot, dqblk_sz))
 604                 ((struct lustre_disk_dqblk_v2 *)ddquot)->dqb_itime =
 605                                                                 cpu_to_le64(1);
 606         fs = get_fs();
 607         set_fs(KERNEL_DS);
 608         ret = filp->f_op->write(filp, ddquot,
 609                                 dqblk_sz, &offset);
 610         set_fs(fs);
 611         if (ret != dqblk_sz) {
 612                 CDEBUG(D_WARNING, "VFS: dquota write failed on dev %s\n",
 613                        filp->f_dentry->d_sb->s_id);
 614                 if (ret >= 0)
 615                         ret = -ENOSPC;
 616         } else
 617                 ret = 0;
 618
 619         return ret;
 620 }
 621
 622 /**
 623  * Free dquot entry in data block
 624  */
 625 static int free_dqentry(struct lustre_dquot *dquot, uint blk,
 626                         lustre_quota_version_t version)
 627 {
 628         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 629         struct lustre_mem_dqinfo *info =
 630             &dquot->dq_info->qi_info[dquot->dq_type];
 631         struct lustre_disk_dqdbheader *dh;
 632         dqbuf_t buf = getdqbuf();
 633         int dqstrinblk = lustre_dqstrinblk[version];
 634         int ret = 0;
 635
 636         if (!buf)
 637                 return -ENOMEM;
 638         if (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS != blk) {
 639                 CDEBUG(D_ERROR,
 640                        "VFS: Quota structure has offset to other block (%u) "
 641                        "than it should (%u).\n",
 642                        blk, (uint) (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS));
 643                 goto out_buf;
 644         }
 645         if ((ret = read_blk(filp, blk, buf)) < 0) {
 646                 CDEBUG(D_ERROR, "VFS: Can't read quota data block %u\n", blk);
 647                 goto out_buf;
 648         }
 649         dh = (struct lustre_disk_dqdbheader *)buf;
 650         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) - 1);
 651         if (!le16_to_cpu(dh->dqdh_entries)) {   /* Block got free? */
 652                 if ((ret = remove_free_dqentry(filp, info, buf, blk)) < 0 ||
 653                     (ret = put_free_dqblk(filp, info, buf, blk)) < 0) {
 654                         CDEBUG(D_ERROR,
 655                                "VFS: Can't move quota data block (%u) to free "
 656                                "list.\n", blk);
 657                         goto out_buf;
 658                 }
 659         } else {
 660                 memset(buf + (dquot->dq_off & ((1<<LUSTRE_DQBLKSIZE_BITS) - 1)),
 661                        0, lustre_disk_dqblk_sz[version]);
 662                 if (le16_to_cpu(dh->dqdh_entries) == dqstrinblk - 1) {
 663                         /* Insert will write block itself */
 664                         if ((ret =
 665                              insert_free_dqentry(filp, info, buf, blk)) < 0) {
 666                                 CDEBUG(D_ERROR,
 667                                        "VFS: Can't insert quota data block "
 668                                        "(%u) to free entry list.\n", blk);
 669                                 goto out_buf;
 670                         }
 671                 } else if ((ret = write_blk(filp, blk, buf)) < 0) {
 672                         CDEBUG(D_ERROR,
 673                                "VFS: Can't write quota data block %u\n", blk);
 674                         goto out_buf;
 675                 }
 676         }
 677         dquot->dq_off = 0;      /* Quota is now unattached */
 678 out_buf:
 679         freedqbuf(buf);
 680         return ret;
 681 }
 682
 683 /**
 684  * Remove reference to dquot from tree
 685  */
 686 static int remove_tree(struct lustre_dquot *dquot, uint * blk, int depth,
 687                        lustre_quota_version_t version)
 688 {
 689         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 690         struct lustre_mem_dqinfo *info =
 691             &dquot->dq_info->qi_info[dquot->dq_type];
 692         dqbuf_t buf = getdqbuf();
 693         int ret = 0;
 694         uint newblk;
 695         u32 *ref = (u32 *) buf;
 696
 697         if (!buf)
 698                 return -ENOMEM;
 699         if ((ret = read_blk(filp, *blk, buf)) < 0) {
 700                 CDEBUG(D_ERROR, "VFS: Can't read quota data block %u\n", *blk);
 701                 goto out_buf;
 702         }
 703         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 704         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 705                 ret = free_dqentry(dquot, newblk, version);
 706                 newblk = 0;
 707         } else
 708                 ret = remove_tree(dquot, &newblk, depth + 1, version);
 709         if (ret >= 0 && !newblk) {
 710                 int i;
 711                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(0);
 712                 for (i = 0; i < LUSTRE_DQBLKSIZE && !buf[i]; i++)
 713                         /* Block got empty? */ ;
 714                 /* don't put the root block into free blk list! */
 715                 if (i == LUSTRE_DQBLKSIZE && *blk != LUSTRE_DQTREEOFF) {
 716                         put_free_dqblk(filp, info, buf, *blk);
 717                         *blk = 0;
 718                 } else if ((ret = write_blk(filp, *blk, buf)) < 0)
 719                         CDEBUG(D_ERROR,
 720                                "VFS: Can't write quota tree block %u.\n", *blk);
 721         }
 722 out_buf:
 723         freedqbuf(buf);
 724         return ret;
 725 }
 726
 727 /**
 728  * Delete dquot from tree
 729  */
 730 static int lustre_delete_dquot(struct lustre_dquot *dquot,
 731                                 lustre_quota_version_t version)
 732 {
 733         uint tmp = LUSTRE_DQTREEOFF;
 734
 735         if (!dquot->dq_off)     /* Even not allocated? */
 736                 return 0;
 737         return remove_tree(dquot, &tmp, 0, version);
 738 }
 739
 740 /**
 741  * Find entry in block
 742  */
 743 static loff_t find_block_dqentry(struct lustre_dquot *dquot, uint blk,
 744                                  lustre_quota_version_t version)
 745 {
 746         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 747         dqbuf_t buf = getdqbuf();
 748         loff_t ret = 0;
 749         int i;
 750         struct lustre_disk_dqblk_v2 *ddquot =
 751                 (struct lustre_disk_dqblk_v2 *)GETENTRIES(buf, version);
 752         int dqblk_sz = lustre_disk_dqblk_sz[version];
 753         int dqstrinblk = lustre_dqstrinblk[version];
 754
 755         LASSERT(version == LUSTRE_QUOTA_V2);
 756
 757         if (!buf)
 758                 return -ENOMEM;
 759         if ((ret = read_blk(filp, blk, buf)) < 0) {
 760                 CDEBUG(D_ERROR, "VFS: Can't read quota tree block %u.\n", blk);
 761                 goto out_buf;
 762         }
 763         if (dquot->dq_id)
 764                 for (i = 0; i < dqstrinblk &&
 765                      le32_to_cpu(ddquot[i].dqb_id) != dquot->dq_id;
 766                      i++) ;
 767         else {                  /* ID 0 as a bit more complicated searching... */
 768                 for (i = 0; i < dqstrinblk; i++)
 769                         if (!le32_to_cpu(ddquot[i].dqb_id)
 770                             && memcmp((char *)&emptydquot[version],
 771                                       (char *)&ddquot[i], dqblk_sz))
 772                                 break;
 773         }
 774         if (i == dqstrinblk) {
 775                 CDEBUG(D_ERROR,
 776                        "VFS: Quota for id %u referenced but not present.\n",
 777                        dquot->dq_id);
 778                 ret = -EIO;
 779                 goto out_buf;
 780         } else
 781                 ret =
 782                     (blk << LUSTRE_DQBLKSIZE_BITS) +
 783                     sizeof(struct lustre_disk_dqdbheader) +
 784                     i * dqblk_sz;
 785 out_buf:
 786         freedqbuf(buf);
 787         return ret;
 788 }
 789
 790 /**
 791  * Find entry for given id in the tree
 792  */
 793 static loff_t find_tree_dqentry(struct lustre_dquot *dquot, uint blk, int depth,
 794                                 lustre_quota_version_t version)
 795 {
 796         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 797         dqbuf_t buf = getdqbuf();
 798         loff_t ret = 0;
 799         u32 *ref = (u32 *) buf;
 800
 801         if (!buf)
 802                 return -ENOMEM;
 803         if ((ret = read_blk(filp, blk, buf)) < 0) {
 804                 CDEBUG(D_ERROR, "VFS: Can't read quota tree block %u.\n", blk);
 805                 goto out_buf;
 806         }
 807         ret = 0;
 808         blk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 809         if (!blk)               /* No reference? */
 810                 goto out_buf;
 811         if (depth < LUSTRE_DQTREEDEPTH - 1)
 812                 ret = find_tree_dqentry(dquot, blk, depth + 1, version);
 813         else
 814                 ret = find_block_dqentry(dquot, blk, version);
 815 out_buf:
 816         freedqbuf(buf);
 817         return ret;
 818 }
 819
 820 /**
 821  * Find entry for given id in the tree - wrapper function
 822  */
 823 static inline loff_t find_dqentry(struct lustre_dquot *dquot,
 824                                   lustre_quota_version_t version)
 825 {
 826         return find_tree_dqentry(dquot, LUSTRE_DQTREEOFF, 0, version);
 827 }
 828
 829 int lustre_read_dquot(struct lustre_dquot *dquot)
 830 {
 831         int type = dquot->dq_type;
 832         struct file *filp;
 833         mm_segment_t fs;
 834         loff_t offset;
 835         int ret = 0, dqblk_sz;
 836         lustre_quota_version_t version;
 837
 838         /* Invalidated quota? */
 839         if (!dquot->dq_info || !(filp = dquot->dq_info->qi_files[type])) {
 840                 CDEBUG(D_ERROR, "VFS: Quota invalidated while reading!\n");
 841                 return -EIO;
 842         }
 843
 844         version = dquot->dq_info->qi_version;
 845         LASSERT(version == LUSTRE_QUOTA_V2);
 846         dqblk_sz = lustre_disk_dqblk_sz[version];
 847
 848         offset = find_dqentry(dquot, version);
 849         if (offset <= 0) {      /* Entry not present? */
 850                 if (offset < 0)
 851                         CDEBUG(D_ERROR,
 852                                "VFS: Can't read quota structure for id %u.\n",
 853                                dquot->dq_id);
 854                 dquot->dq_off = 0;
 855                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
 856                 memset(&dquot->dq_dqb, 0, sizeof(struct lustre_mem_dqblk));
 857                 ret = offset;
 858         } else {
 859                 char ddquot[dqblk_sz];
 860
 861                 dquot->dq_off = offset;
 862                 fs = get_fs();
 863                 set_fs(KERNEL_DS);
 864                 if ((ret = filp->f_op->read(filp, ddquot, dqblk_sz, &offset)) !=
 865                     dqblk_sz) {
 866                         if (ret >= 0)
 867                                 ret = -EIO;
 868                         CDEBUG(D_ERROR,
 869                                "VFS: Error while reading quota structure for id "
 870                                "%u.\n", dquot->dq_id);
 871                         memset(ddquot, 0, dqblk_sz);
 872                 } else {
 873                         ret = 0;
 874                         /* We need to escape back all-zero structure */
 875                         if (!memcmp((char *)&fakedquot[version],
 876                                     ddquot, dqblk_sz))
 877                            ((struct lustre_disk_dqblk_v2 *)ddquot)->dqb_itime =
 878                                                                 cpu_to_le64(0);
 879                 }
 880                 set_fs(fs);
 881                 disk2memdqb(&dquot->dq_dqb, ddquot, version);
 882         }
 883
 884         return ret;
 885 }
 886
 887 /**
 888  * Commit changes of dquot to disk - it might also mean deleting
 889  * it when quota became fake.
 890  */
 891 int lustre_commit_dquot(struct lustre_dquot *dquot)
 892 {
 893         int rc = 0;
 894         lustre_quota_version_t version = dquot->dq_info->qi_version;
 895
 896         /* always clear the flag so we don't loop on an IO error... */
 897         clear_bit(DQ_MOD_B, &dquot->dq_flags);
 898
 899         /* The block/inode usage in admin quotafile isn't the real usage
 900          * over all cluster, so keep the fake dquot entry on disk is
 901          * meaningless, just remove it */
 902         if (test_bit(DQ_FAKE_B, &dquot->dq_flags))
 903                 rc = lustre_delete_dquot(dquot, version);
 904         else
 905                 rc = lustre_write_dquot(dquot, version);
 906
 907         if (rc < 0)
 908                 return rc;
 909
 910         if (lustre_info_dirty(&dquot->dq_info->qi_info[dquot->dq_type]))
 911                 rc = lustre_write_quota_info(dquot->dq_info, dquot->dq_type);
 912
 913         return rc;
 914 }
 915
 916 int lustre_init_quota_header(struct lustre_quota_info *lqi, int type,
 917                              int fakemagics)
 918 {
 919         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 920         static const uint fake_magics[] = LUSTRE_BADQMAGICS;
 921         const uint* quota_versions = lustre_initqversions[lqi->qi_version];
 922         struct lustre_disk_dqheader dqhead;
 923         ssize_t size;
 924         loff_t offset = 0;
 925         struct file *fp = lqi->qi_files[type];
 926         int rc = 0;
 927
 928         /* write quotafile header */
 929         dqhead.dqh_magic = cpu_to_le32(fakemagics ?
 930                                        fake_magics[type] : quota_magics[type]);
 931         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
 932         size = fp->f_op->write(fp, (char *)&dqhead,
 933                                sizeof(struct lustre_disk_dqheader), &offset);
 934
 935         if (size != sizeof(struct lustre_disk_dqheader)) {
 936                 CDEBUG(D_ERROR, "error writing quoafile header (rc:%d)\n", rc);
 937                 rc = size;
 938         }
 939
 940         return rc;
 941 }
 942
 943 /**
 944  * We need to export this function to initialize quotafile, because we haven't
 945  * user level check utility
 946  */
 947 int lustre_init_quota_info_generic(struct lustre_quota_info *lqi, int type,
 948                                    int fakemagics)
 949 {
 950         struct lustre_mem_dqinfo *dqinfo = &lqi->qi_info[type];
 951         int rc;
 952
 953         rc = lustre_init_quota_header(lqi, type, fakemagics);
 954         if (rc)
 955                 return rc;
 956
 957         /* write init quota info */
 958         memset(dqinfo, 0, sizeof(*dqinfo));
 959         dqinfo->dqi_bgrace = MAX_DQ_TIME;
 960         dqinfo->dqi_igrace = MAX_IQ_TIME;
 961         dqinfo->dqi_blocks = LUSTRE_DQTREEOFF + 1;
 962
 963         return lustre_write_quota_info(lqi, type);
 964 }
 965
 966 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type)
 967 {
 968         return lustre_init_quota_info_generic(lqi, type, 0);
 969 }
 970
 971 ssize_t quota_read(struct file *file, struct inode *inode, int type,
 972                    uint blk, dqbuf_t buf)
 973 {
 974         if (file) {
 975                 return read_blk(file, blk, buf);
 976         } else {
 977 #ifndef KERNEL_SUPPORTS_QUOTA_READ
 978                 return -ENOTSUPP;
 979 #else
 980                 struct super_block *sb = inode->i_sb;
 981                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 982                 return sb->s_op->quota_read(sb, type, (char *)buf,
 983                                             LUSTRE_DQBLKSIZE,
 984                                             blk << LUSTRE_DQBLKSIZE_BITS);
 985 #endif
 986         }
 987 }
 988
 989 static int walk_block_dqentry(struct file *filp, struct inode *inode, int type,
 990                               uint blk, struct list_head *list)
 991 {
 992         dqbuf_t buf = getdqbuf();
 993         loff_t ret = 0;
 994         struct lustre_disk_dqdbheader *dqhead =
 995             (struct lustre_disk_dqdbheader *)buf;
 996         struct dqblk *blk_item;
 997         struct dqblk *pos;
 998         struct list_head *tmp;
 999
1000         if (!buf)
1001                 return -ENOMEM;
1002         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
1003                 CDEBUG(D_ERROR, "VFS: Can't read quota tree block %u.\n", blk);
1004                 goto out_buf;
1005         }
1006         ret = 0;
1007
1008         if (!le32_to_cpu(dqhead->dqdh_entries))
1009                 goto out_buf;
1010
1011         if (list_empty(list)) {
1012                 tmp = list;
1013                 goto done;
1014         }
1015
1016         list_for_each_entry(pos, list, link) {
1017                 if (blk == pos->blk)    /* we got this blk already */
1018                         goto out_buf;
1019                 if (blk > pos->blk)
1020                         continue;
1021                 break;
1022         }
1023         tmp = &pos->link;
1024 done:
1025         blk_item = kmalloc(sizeof(*blk_item), GFP_NOFS);
1026         if (!blk_item) {
1027                 ret = -ENOMEM;
1028                 goto out_buf;
1029         }
1030         blk_item->blk = blk;
1031         INIT_LIST_HEAD(&blk_item->link);
1032
1033         list_add_tail(&blk_item->link, tmp);
1034
1035 out_buf:
1036         freedqbuf(buf);
1037         return ret;
1038 }
1039
1040 int walk_tree_dqentry(struct file *filp, struct inode *inode, int type,
1041                       uint blk, int depth, struct list_head *list)
1042 {
1043         dqbuf_t buf = getdqbuf();
1044         loff_t ret = 0;
1045         int index;
1046         u32 *ref = (u32 *) buf;
1047
1048         if (!buf)
1049                 return -ENOMEM;
1050         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
1051                 CDEBUG(D_ERROR, "VFS: Can't read quota tree block %u.\n", blk);
1052                 goto out_buf;
1053         }
1054         ret = 0;
1055
1056         for (index = 0; index <= 0xff && !ret; index++) {
1057                 blk = le32_to_cpu(ref[index]);
1058                 if (!blk)       /* No reference */
1059                         continue;
1060
1061                 if (depth < LUSTRE_DQTREEDEPTH - 1)
1062                         ret = walk_tree_dqentry(filp, inode, type, blk,
1063                                                 depth + 1, list);
1064                 else
1065                         ret = walk_block_dqentry(filp, inode, type, blk, list);
1066         }
1067 out_buf:
1068         freedqbuf(buf);
1069         return ret;
1070 }
1071
1072 /**
1073  * Walk through the quota file (v2 format) to get all ids with quota limit
1074  */
1075 int lustre_get_qids(struct file *fp, struct inode *inode, int type,
1076                     struct list_head *list)
1077 {
1078         struct list_head blk_list;
1079         struct dqblk *blk_item, *tmp;
1080         dqbuf_t buf = NULL;
1081         struct lustre_disk_dqblk_v2 *ddquot;
1082         int rc;
1083         lustre_quota_version_t version;
1084
1085         ENTRY;
1086
1087         if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V2) == 0)
1088                 version = LUSTRE_QUOTA_V2;
1089         else {
1090                 CDEBUG(D_ERROR, "unknown quota file format!\n");
1091                 RETURN(-EINVAL);
1092         }
1093
1094         if (!list_empty(list)) {
1095                 CDEBUG(D_ERROR, "not empty list\n");
1096                 RETURN(-EINVAL);
1097         }
1098
1099         INIT_LIST_HEAD(&blk_list);
1100         rc = walk_tree_dqentry(fp, inode, type, LUSTRE_DQTREEOFF, 0, &blk_list);
1101         if (rc) {
1102                 CDEBUG(D_ERROR, "walk through quota file failed!(%d)\n", rc);
1103                 GOTO(out_free, rc);
1104         }
1105         if (list_empty(&blk_list))
1106                 RETURN(0);
1107
1108         buf = getdqbuf();
1109         if (!buf)
1110                 RETURN(-ENOMEM);
1111         ddquot = (struct lustre_disk_dqblk_v2 *)GETENTRIES(buf, version);
1112
1113         list_for_each_entry(blk_item, &blk_list, link) {
1114                 loff_t ret = 0;
1115                 int i, dqblk_sz = lustre_disk_dqblk_sz[version];
1116
1117                 memset(buf, 0, LUSTRE_DQBLKSIZE);
1118                 if ((ret = quota_read(fp, inode, type, blk_item->blk, buf))<0) {
1119                         CDEBUG(D_ERROR,
1120                                "VFS: Can't read quota tree block %u.\n",
1121                                blk_item->blk);
1122                         GOTO(out_free, rc = ret);
1123                 }
1124
1125                 for (i = 0; i < lustre_dqstrinblk[version]; i++) {
1126                         struct dquot_id *dqid;
1127                         /* skip empty entry */
1128                         if (!memcmp((char *)&emptydquot[version],
1129                                     (char *)&ddquot[i], dqblk_sz))
1130                                 continue;
1131
1132                         dqid = kmalloc(sizeof(*dqid), GFP_NOFS);
1133                         if (!dqid)
1134                                 GOTO(out_free, rc = -ENOMEM);
1135
1136                         dqid->di_id = le32_to_cpu(ddquot[i].dqb_id);
1137                         INIT_LIST_HEAD(&dqid->di_link);
1138                         list_add(&dqid->di_link, list);
1139                 }
1140         }
1141
1142 out_free:
1143         list_for_each_entry_safe(blk_item, tmp, &blk_list, link) {
1144                 list_del_init(&blk_item->link);
1145                 kfree(blk_item);
1146         }
1147         if (buf)
1148                 freedqbuf(buf);
1149
1150         RETURN(rc);
1151 }
1152
1153
1154 EXPORT_SYMBOL(lustre_read_quota_info);
1155 EXPORT_SYMBOL(lustre_write_quota_info);
1156 EXPORT_SYMBOL(lustre_check_quota_file);
1157 EXPORT_SYMBOL(lustre_read_dquot);
1158 EXPORT_SYMBOL(lustre_commit_dquot);
1159 EXPORT_SYMBOL(lustre_init_quota_info);
1160 EXPORT_SYMBOL(lustre_get_qids);
1161 #endif