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_V1] = LUSTRE_INITQVERSIONS,
  66         [LUSTRE_QUOTA_V2] = LUSTRE_INITQVERSIONS_V2
  67 };
  68
  69 static const int lustre_dqstrinblk[] = {
  70         [LUSTRE_QUOTA_V1] = LUSTRE_DQSTRINBLK,
  71         [LUSTRE_QUOTA_V2] = LUSTRE_DQSTRINBLK_V2
  72 };
  73
  74 static const int lustre_disk_dqblk_sz[] = {
  75         [LUSTRE_QUOTA_V1] = sizeof(struct lustre_disk_dqblk),
  76         [LUSTRE_QUOTA_V2] = sizeof(struct lustre_disk_dqblk_v2)
  77 };
  78
  79 int check_quota_file(struct file *f, struct inode *inode, int type,
  80                      lustre_quota_version_t version)
  81 {
  82         struct lustre_disk_dqheader dqhead;
  83         mm_segment_t fs;
  84         ssize_t size;
  85         loff_t offset = 0;
  86         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
  87         const uint *quota_versions = lustre_initqversions[version];
  88
  89         if (f) {
  90                 fs = get_fs();
  91                 set_fs(KERNEL_DS);
  92                 size = f->f_op->read(f, (char *)&dqhead,
  93                                      sizeof(struct lustre_disk_dqheader),
  94                                      &offset);
  95                 set_fs(fs);
  96         } else {
  97 #ifndef KERNEL_SUPPORTS_QUOTA_READ
  98                 size = 0;
  99 #else
 100                 struct super_block *sb = inode->i_sb;
 101                 size = sb->s_op->quota_read(sb, type, (char *)&dqhead,
 102                                             sizeof(struct lustre_disk_dqheader), 0);
 103 #endif
 104         }
 105         if (size != sizeof(struct lustre_disk_dqheader))
 106                 return -EINVAL;
 107         if (le32_to_cpu(dqhead.dqh_magic) != quota_magics[type] ||
 108             le32_to_cpu(dqhead.dqh_version) != quota_versions[type])
 109                 return -EINVAL;
 110         return 0;
 111 }
 112
 113 /* Check whether given file is really lustre admin quotafile */
 114 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type)
 115 {
 116         struct file *f = lqi->qi_files[type];
 117         return check_quota_file(f, NULL, type, lqi->qi_version);
 118 }
 119
 120 int lustre_read_quota_file_info(struct file* f, struct lustre_mem_dqinfo* info)
 121 {
 122         mm_segment_t fs;
 123         struct lustre_disk_dqinfo dinfo;
 124         ssize_t size;
 125         loff_t offset = LUSTRE_DQINFOOFF;
 126
 127         fs = get_fs();
 128         set_fs(KERNEL_DS);
 129         size = f->f_op->read(f, (char *)&dinfo,
 130                              sizeof(struct lustre_disk_dqinfo), &offset);
 131         set_fs(fs);
 132         if (size != sizeof(struct lustre_disk_dqinfo)) {
 133                 CERROR("Can't read info structure on device %s.\n",
 134                        f->f_vfsmnt->mnt_sb->s_id);
 135                 return -EINVAL;
 136         }
 137         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 138         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 139         info->dqi_flags = le32_to_cpu(dinfo.dqi_flags);
 140         info->dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 141         info->dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 142         info->dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 143         return 0;
 144 }
 145
 146 /* Read information header from quota file */
 147 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type)
 148 {
 149         return lustre_read_quota_file_info(lqi->qi_files[type], &lqi->qi_info[type]);
 150 }
 151
 152 /* Write information header to quota file */
 153 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type)
 154 {
 155         mm_segment_t fs;
 156         struct lustre_disk_dqinfo dinfo;
 157         struct lustre_mem_dqinfo *info = &lqi->qi_info[type];
 158         struct file *f = lqi->qi_files[type];
 159         ssize_t size;
 160         loff_t offset = LUSTRE_DQINFOOFF;
 161
 162         info->dqi_flags &= ~DQF_INFO_DIRTY;
 163         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 164         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 165         dinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK);
 166         dinfo.dqi_blocks = cpu_to_le32(info->dqi_blocks);
 167         dinfo.dqi_free_blk = cpu_to_le32(info->dqi_free_blk);
 168         dinfo.dqi_free_entry = cpu_to_le32(info->dqi_free_entry);
 169         fs = get_fs();
 170         set_fs(KERNEL_DS);
 171         size = f->f_op->write(f, (char *)&dinfo,
 172                               sizeof(struct lustre_disk_dqinfo), &offset);
 173         set_fs(fs);
 174         if (size != sizeof(struct lustre_disk_dqinfo)) {
 175                 CWARN("Can't write info structure on device %s.\n",
 176                       f->f_vfsmnt->mnt_sb->s_id);
 177                 return -1;
 178         }
 179         return 0;
 180 }
 181
 182 #define DQ2MQ(v) ((sizeof(v) == sizeof(__u64)) ? \
 183                 le64_to_cpu(v) : le32_to_cpu(v))
 184
 185 #define MQ2DQ(v,newv) ((sizeof(v) == sizeof(__u64)) ? \
 186                 (v = cpu_to_le64((__u64)newv)) : (v = cpu_to_le32((__u32)newv)))
 187
 188 #define DQF_GET(var,ver,field) ((ver == LUSTRE_QUOTA_V1)?\
 189                 DQ2MQ(((struct lustre_disk_dqblk*)(var))->field):\
 190                 DQ2MQ(((struct lustre_disk_dqblk_v2*)(var))->field))
 191
 192 #define DQF_PUT(var,ver,field,val) ((ver == LUSTRE_QUOTA_V1)?\
 193                 MQ2DQ(((struct lustre_disk_dqblk*)(var))->field, val):\
 194                 MQ2DQ(((struct lustre_disk_dqblk_v2*)(var))->field, val))
 195
 196 void disk2memdqb(struct lustre_mem_dqblk *m, void *d,
 197                  lustre_quota_version_t version)
 198 {
 199         m->dqb_ihardlimit = DQF_GET(d, version, dqb_ihardlimit);
 200         m->dqb_isoftlimit = DQF_GET(d, version, dqb_isoftlimit);
 201         m->dqb_curinodes = DQF_GET(d, version, dqb_curinodes);
 202         m->dqb_itime = DQF_GET(d, version, dqb_itime);
 203         m->dqb_bhardlimit = DQF_GET(d, version, dqb_bhardlimit);
 204         m->dqb_bsoftlimit = DQF_GET(d, version, dqb_bsoftlimit);
 205         m->dqb_curspace = DQF_GET(d, version, dqb_curspace);
 206         m->dqb_btime = DQF_GET(d, version, dqb_btime);
 207 }
 208
 209 static int check_quota_bounds(struct lustre_mem_dqblk *m,
 210                               lustre_quota_version_t version)
 211 {
 212         return (version == LUSTRE_QUOTA_V1  &&
 213                 m->dqb_ihardlimit <= MAX_UL &&
 214                 m->dqb_isoftlimit <= MAX_UL &&
 215                 m->dqb_curinodes <= MAX_UL  &&
 216                 m->dqb_bhardlimit <= MAX_UL &&
 217                 m->dqb_bsoftlimit <= MAX_UL) ||
 218                 version != LUSTRE_QUOTA_V1;
 219 }
 220
 221 static int mem2diskdqb(void *d, struct lustre_mem_dqblk *m,
 222                        qid_t id, lustre_quota_version_t version)
 223 {
 224         if (!check_quota_bounds(m, version))
 225                 return -EINVAL;
 226
 227         DQF_PUT(d, version, dqb_ihardlimit, m->dqb_ihardlimit);
 228         DQF_PUT(d, version, dqb_isoftlimit, m->dqb_isoftlimit);
 229         DQF_PUT(d, version, dqb_curinodes, m->dqb_curinodes);
 230         DQF_PUT(d, version, dqb_itime, m->dqb_itime);
 231         DQF_PUT(d, version, dqb_bhardlimit, m->dqb_bhardlimit);
 232         DQF_PUT(d, version, dqb_bsoftlimit, m->dqb_bsoftlimit);
 233         DQF_PUT(d, version, dqb_curspace, m->dqb_curspace);
 234         DQF_PUT(d, version, dqb_btime, m->dqb_btime);
 235         DQF_PUT(d, version, dqb_id, id);
 236
 237         return 0;
 238 }
 239
 240 dqbuf_t getdqbuf(void)
 241 {
 242         dqbuf_t buf = kmalloc(LUSTRE_DQBLKSIZE, GFP_NOFS);
 243         if (!buf)
 244                 CWARN("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 /* Remove empty block from list and return it */
 286 int get_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info)
 287 {
 288         dqbuf_t buf = getdqbuf();
 289         struct lustre_disk_dqdbheader *dh =
 290             (struct lustre_disk_dqdbheader *)buf;
 291         int ret, blk;
 292
 293         if (!buf)
 294                 return -ENOMEM;
 295         if (info->dqi_free_blk) {
 296                 blk = info->dqi_free_blk;
 297                 if ((ret = read_blk(filp, blk, buf)) < 0)
 298                         goto out_buf;
 299                 info->dqi_free_blk = le32_to_cpu(dh->dqdh_next_free);
 300         } else {
 301                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 302                 /* Assure block allocation... */
 303                 if ((ret = write_blk(filp, info->dqi_blocks, buf)) < 0)
 304                         goto out_buf;
 305                 blk = info->dqi_blocks++;
 306         }
 307         lustre_mark_info_dirty(info);
 308         ret = blk;
 309 out_buf:
 310         freedqbuf(buf);
 311         return ret;
 312 }
 313
 314 /* Insert empty block to the list */
 315 int put_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info,
 316                    dqbuf_t buf, uint blk)
 317 {
 318         struct lustre_disk_dqdbheader *dh =
 319             (struct lustre_disk_dqdbheader *)buf;
 320         int err;
 321
 322         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_blk);
 323         dh->dqdh_prev_free = cpu_to_le32(0);
 324         dh->dqdh_entries = cpu_to_le16(0);
 325         info->dqi_free_blk = blk;
 326         lustre_mark_info_dirty(info);
 327         if ((err = write_blk(filp, blk, buf)) < 0)
 328                 /* Some strange block. We had better leave it... */
 329                 return err;
 330         return 0;
 331 }
 332
 333 /* Remove given block from the list of blocks with free entries */
 334 int remove_free_dqentry(struct file *filp,
 335                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 336                         uint blk)
 337 {
 338         dqbuf_t tmpbuf = getdqbuf();
 339         struct lustre_disk_dqdbheader *dh =
 340             (struct lustre_disk_dqdbheader *)buf;
 341         uint nextblk = le32_to_cpu(dh->dqdh_next_free), prevblk =
 342             le32_to_cpu(dh->dqdh_prev_free);
 343         int err;
 344
 345         if (!tmpbuf)
 346                 return -ENOMEM;
 347         if (nextblk) {
 348                 if ((err = read_blk(filp, nextblk, tmpbuf)) < 0)
 349                         goto out_buf;
 350                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 351                     dh->dqdh_prev_free;
 352                 if ((err = write_blk(filp, nextblk, tmpbuf)) < 0)
 353                         goto out_buf;
 354         }
 355         if (prevblk) {
 356                 if ((err = read_blk(filp, prevblk, tmpbuf)) < 0)
 357                         goto out_buf;
 358                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_next_free =
 359                     dh->dqdh_next_free;
 360                 if ((err = write_blk(filp, prevblk, tmpbuf)) < 0)
 361                         goto out_buf;
 362         } else {
 363                 info->dqi_free_entry = nextblk;
 364                 lustre_mark_info_dirty(info);
 365         }
 366         freedqbuf(tmpbuf);
 367         dh->dqdh_next_free = dh->dqdh_prev_free = cpu_to_le32(0);
 368         err = write_blk(filp, blk, buf);
 369         if (err < 0)      /* No matter whether write succeeds block is out of list */
 370                 CERROR("VFS: Can't write block (%u) with "
 371                        "free entries (rc=%d).\n", blk, err);
 372         return 0;
 373 out_buf:
 374         freedqbuf(tmpbuf);
 375         return err;
 376 }
 377
 378 /* Insert given block to the beginning of list with free entries */
 379 int insert_free_dqentry(struct file *filp,
 380                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 381                         uint blk)
 382 {
 383         dqbuf_t tmpbuf = getdqbuf();
 384         struct lustre_disk_dqdbheader *dh =
 385             (struct lustre_disk_dqdbheader *)buf;
 386         int err;
 387
 388         if (!tmpbuf)
 389                 return -ENOMEM;
 390         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_entry);
 391         dh->dqdh_prev_free = cpu_to_le32(0);
 392         if ((err = write_blk(filp, blk, buf)) < 0)
 393                 goto out_buf;
 394         if (info->dqi_free_entry) {
 395                 if ((err = read_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 396                         goto out_buf;
 397                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 398                     cpu_to_le32(blk);
 399                 if ((err = write_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 400                         goto out_buf;
 401         }
 402         freedqbuf(tmpbuf);
 403         info->dqi_free_entry = blk;
 404         lustre_mark_info_dirty(info);
 405         return 0;
 406 out_buf:
 407         freedqbuf(tmpbuf);
 408         return err;
 409 }
 410
 411
 412
 413 /* Find space for dquot */
 414 static uint find_free_dqentry(struct lustre_dquot *dquot, int *err,
 415                               lustre_quota_version_t version)
 416 {
 417         struct lustre_quota_info *lqi = dquot->dq_info;
 418         struct file *filp = lqi->qi_files[dquot->dq_type];
 419         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 420         uint blk, i;
 421         struct lustre_disk_dqdbheader *dh;
 422         void *ddquot;
 423         int dqblk_sz = lustre_disk_dqblk_sz[version];
 424         int dqstrinblk = lustre_dqstrinblk[version];
 425         char fakedquot[dqblk_sz];
 426         dqbuf_t buf;
 427
 428         *err = 0;
 429         if (!(buf = getdqbuf())) {
 430                 *err = -ENOMEM;
 431                 return 0;
 432         }
 433         dh = (struct lustre_disk_dqdbheader *)buf;
 434         ddquot = GETENTRIES(buf, version);
 435         if (info->dqi_free_entry) {
 436                 blk = info->dqi_free_entry;
 437                 if ((*err = read_blk(filp, blk, buf)) < 0)
 438                         goto out_buf;
 439         } else {
 440                 blk = get_free_dqblk(filp, info);
 441                 if ((int)blk < 0) {
 442                         *err = blk;
 443                         freedqbuf(buf);
 444                         return 0;
 445                 }
 446                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 447                 info->dqi_free_entry = blk; /* This is enough as block is
 448                                                already zeroed and entry list
 449                                                is empty... */
 450                 lustre_mark_info_dirty(info);
 451         }
 452
 453         /* Will block be full */
 454         if (le16_to_cpu(dh->dqdh_entries) + 1 >= dqstrinblk)
 455                 if ((*err = remove_free_dqentry(filp, info, buf, blk)) < 0) {
 456                         CERROR("VFS: Can't remove block %u"
 457                                " from entry free list.\n", blk);
 458                         goto out_buf;
 459                 }
 460         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) + 1);
 461         memset(fakedquot, 0, dqblk_sz);
 462         /* Find free structure in block */
 463         for (i = 0; i < dqstrinblk &&
 464              memcmp(fakedquot, (char*)ddquot + i * dqblk_sz,
 465                     sizeof(fakedquot)); i++);
 466
 467         if (i == dqstrinblk) {
 468                 CERROR("VFS: Data block full but it shouldn't.\n");
 469                 *err = -EIO;
 470                 goto out_buf;
 471         }
 472
 473         if ((*err = write_blk(filp, blk, buf)) < 0) {
 474                 CERROR("VFS: Can't write quota data block %u.\n", blk);
 475                 goto out_buf;
 476         }
 477         dquot->dq_off =
 478             (blk << LUSTRE_DQBLKSIZE_BITS) +
 479             sizeof(struct lustre_disk_dqdbheader) +
 480             i * dqblk_sz;
 481         freedqbuf(buf);
 482         return blk;
 483 out_buf:
 484         freedqbuf(buf);
 485         return 0;
 486 }
 487
 488 /* Insert reference to structure into the trie */
 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                         CERROR("VFS: Can't read tree quota block %u.\n",
 512                                *treeblk);
 513                         goto out_buf;
 514                 }
 515         }
 516         ref = (u32 *) buf;
 517         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 518         if (!newblk)
 519                 newson = 1;
 520         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 521
 522                 if (newblk) {
 523                         CERROR("VFS: Inserting already present quota entry "
 524                                "(block %u).\n",
 525                                ref[GETIDINDEX(dquot->dq_id, depth)]);
 526                         ret = -EIO;
 527                         goto out_buf;
 528                 }
 529
 530                 newblk = find_free_dqentry(dquot, &ret, version);
 531         } else
 532                 ret = do_insert_tree(dquot, &newblk, depth + 1, version);
 533         if (newson && ret >= 0) {
 534                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(newblk);
 535                 ret = write_blk(filp, *treeblk, buf);
 536         } else if (newact && ret < 0)
 537                 put_free_dqblk(filp, info, buf, *treeblk);
 538 out_buf:
 539         freedqbuf(buf);
 540         return ret;
 541 }
 542
 543 /* Wrapper for inserting quota structure into tree */
 544 static inline int dq_insert_tree(struct lustre_dquot *dquot,
 545                                  lustre_quota_version_t version)
 546 {
 547         int tmp = LUSTRE_DQTREEOFF;
 548         return do_insert_tree(dquot, &tmp, 0, version);
 549 }
 550
 551 /*
 552  *  We don't have to be afraid of deadlocks as we never have quotas on quota files...
 553  */
 554 static int lustre_write_dquot(struct lustre_dquot *dquot,
 555                               lustre_quota_version_t version)
 556 {
 557         int type = dquot->dq_type;
 558         struct file *filp;
 559         mm_segment_t fs;
 560         loff_t offset;
 561         ssize_t ret;
 562         int dqblk_sz = lustre_disk_dqblk_sz[version];
 563         char ddquot[dqblk_sz], empty[dqblk_sz];
 564
 565         ret = mem2diskdqb(ddquot, &dquot->dq_dqb, dquot->dq_id, version);
 566         if (ret < 0)
 567                 return ret;
 568
 569         if (!dquot->dq_off)
 570                 if ((ret = dq_insert_tree(dquot, version)) < 0) {
 571                         CERROR("VFS: Error %Zd occurred while creating quota.\n",
 572                                ret);
 573                         return ret;
 574                 }
 575         filp = dquot->dq_info->qi_files[type];
 576         offset = dquot->dq_off;
 577         /* Argh... We may need to write structure full of zeroes but that would be
 578          * treated as an empty place by the rest of the code. Format change would
 579          * be definitely cleaner but the problems probably are not worth it */
 580         memset(empty, 0, dqblk_sz);
 581         if (!memcmp(empty, ddquot, dqblk_sz))
 582                 DQF_PUT(ddquot, version, dqb_itime, 1);
 583         fs = get_fs();
 584         set_fs(KERNEL_DS);
 585         ret = filp->f_op->write(filp, ddquot,
 586                                 dqblk_sz, &offset);
 587         set_fs(fs);
 588         if (ret != dqblk_sz) {
 589                 CWARN("VFS: dquota write failed on dev %s\n",
 590                       filp->f_dentry->d_sb->s_id);
 591                 if (ret >= 0)
 592                         ret = -ENOSPC;
 593         } else
 594                 ret = 0;
 595
 596         return ret;
 597 }
 598
 599 /* Free dquot entry in data block */
 600 static int free_dqentry(struct lustre_dquot *dquot, uint blk,
 601                         lustre_quota_version_t version)
 602 {
 603         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 604         struct lustre_mem_dqinfo *info =
 605             &dquot->dq_info->qi_info[dquot->dq_type];
 606         struct lustre_disk_dqdbheader *dh;
 607         dqbuf_t buf = getdqbuf();
 608         int dqstrinblk = lustre_dqstrinblk[version];
 609         int ret = 0;
 610
 611         if (!buf)
 612                 return -ENOMEM;
 613         if (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS != blk) {
 614                 CERROR("VFS: Quota structure has offset to other block (%u) "
 615                        "than it should (%u).\n", blk,
 616                        (uint)(dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS));
 617                 goto out_buf;
 618         }
 619         if ((ret = read_blk(filp, blk, buf)) < 0) {
 620                 CERROR("VFS: Can't read quota data block %u\n", blk);
 621                 goto out_buf;
 622         }
 623         dh = (struct lustre_disk_dqdbheader *)buf;
 624         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) - 1);
 625         if (!le16_to_cpu(dh->dqdh_entries)) {   /* Block got free? */
 626                 if ((ret = remove_free_dqentry(filp, info, buf, blk)) < 0 ||
 627                     (ret = put_free_dqblk(filp, info, buf, blk)) < 0) {
 628                         CERROR("VFS: Can't move quota data block (%u) "
 629                                "to free list.\n", blk);
 630                         goto out_buf;
 631                 }
 632         } else {
 633                 memset(buf + (dquot->dq_off & ((1<<LUSTRE_DQBLKSIZE_BITS) - 1)),
 634                        0, lustre_disk_dqblk_sz[version]);
 635                 if (le16_to_cpu(dh->dqdh_entries) == dqstrinblk - 1) {
 636                         /* Insert will write block itself */
 637                         if ((ret =
 638                              insert_free_dqentry(filp, info, buf, blk)) < 0) {
 639                                 CERROR("VFS: Can't insert quota data block (%u) "
 640                                        "to free entry list.\n", blk);
 641                                 goto out_buf;
 642                         }
 643                 } else if ((ret = write_blk(filp, blk, buf)) < 0) {
 644                         CERROR("VFS: Can't write quota data block %u\n", blk);
 645                         goto out_buf;
 646                 }
 647         }
 648         dquot->dq_off = 0;      /* Quota is now unattached */
 649 out_buf:
 650         freedqbuf(buf);
 651         return ret;
 652 }
 653
 654 /* Remove reference to dquot from tree */
 655 static int remove_tree(struct lustre_dquot *dquot, uint * blk, int depth,
 656                        lustre_quota_version_t version)
 657 {
 658         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 659         struct lustre_mem_dqinfo *info =
 660             &dquot->dq_info->qi_info[dquot->dq_type];
 661         dqbuf_t buf = getdqbuf();
 662         int ret = 0;
 663         uint newblk;
 664         u32 *ref = (u32 *) buf;
 665
 666         if (!buf)
 667                 return -ENOMEM;
 668         if ((ret = read_blk(filp, *blk, buf)) < 0) {
 669                 CERROR("VFS: Can't read quota data block %u\n", *blk);
 670                 goto out_buf;
 671         }
 672         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 673         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 674                 ret = free_dqentry(dquot, newblk, version);
 675                 newblk = 0;
 676         } else
 677                 ret = remove_tree(dquot, &newblk, depth + 1, version);
 678         if (ret >= 0 && !newblk) {
 679                 int i;
 680                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(0);
 681                 for (i = 0; i < LUSTRE_DQBLKSIZE && !buf[i]; i++)
 682                         /* Block got empty? */ ;
 683                 /* don't put the root block into free blk list! */
 684                 if (i == LUSTRE_DQBLKSIZE && *blk != LUSTRE_DQTREEOFF) {
 685                         put_free_dqblk(filp, info, buf, *blk);
 686                         *blk = 0;
 687                 } else if ((ret = write_blk(filp, *blk, buf)) < 0)
 688                         CERROR("VFS: Can't write quota tree block %u.\n", *blk);
 689         }
 690 out_buf:
 691         freedqbuf(buf);
 692         return ret;
 693 }
 694
 695 /* Delete dquot from tree */
 696 static int lustre_delete_dquot(struct lustre_dquot *dquot,
 697                                 lustre_quota_version_t version)
 698 {
 699         uint tmp = LUSTRE_DQTREEOFF;
 700
 701         if (!dquot->dq_off)     /* Even not allocated? */
 702                 return 0;
 703         return remove_tree(dquot, &tmp, 0, version);
 704 }
 705
 706 /* Find entry in block */
 707 static loff_t find_block_dqentry(struct lustre_dquot *dquot, uint blk,
 708                                  lustre_quota_version_t version)
 709 {
 710         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 711         dqbuf_t buf = getdqbuf();
 712         loff_t ret = 0;
 713         int i;
 714         char *ddquot = GETENTRIES(buf, version);
 715         int dqblk_sz = lustre_disk_dqblk_sz[version];
 716         int dqstrinblk = lustre_dqstrinblk[version];
 717
 718         if (!buf)
 719                 return -ENOMEM;
 720         if ((ret = read_blk(filp, blk, buf)) < 0) {
 721                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 722                 goto out_buf;
 723         }
 724         if (dquot->dq_id)
 725                 for (i = 0; i < dqstrinblk &&
 726                      DQF_GET(ddquot+i*dqblk_sz, version, dqb_id) != dquot->dq_id;
 727                      i++) ;
 728         else {                  /* ID 0 as a bit more complicated searching... */
 729                 char fakedquot[dqblk_sz];
 730
 731                 memset(fakedquot, 0, sizeof(fakedquot));
 732                 for (i = 0; i < dqstrinblk; i++)
 733                         if (!DQF_GET(ddquot + i*dqblk_sz, version, dqb_id)
 734                             && memcmp(fakedquot, ddquot + i*dqblk_sz,
 735                                       dqblk_sz))
 736                                 break;
 737         }
 738         if (i == dqstrinblk) {
 739                 CERROR("VFS: Quota for id %u referenced but not present.\n",
 740                        dquot->dq_id);
 741                 ret = -EIO;
 742                 goto out_buf;
 743         } else
 744                 ret =
 745                     (blk << LUSTRE_DQBLKSIZE_BITS) +
 746                     sizeof(struct lustre_disk_dqdbheader) +
 747                     i * dqblk_sz;
 748 out_buf:
 749         freedqbuf(buf);
 750         return ret;
 751 }
 752
 753 /* Find entry for given id in the tree */
 754 static loff_t find_tree_dqentry(struct lustre_dquot *dquot, uint blk, int depth,
 755                                 lustre_quota_version_t version)
 756 {
 757         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 758         dqbuf_t buf = getdqbuf();
 759         loff_t ret = 0;
 760         u32 *ref = (u32 *) buf;
 761
 762         if (!buf)
 763                 return -ENOMEM;
 764         if ((ret = read_blk(filp, blk, buf)) < 0) {
 765                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 766                 goto out_buf;
 767         }
 768         ret = 0;
 769         blk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 770         if (!blk)               /* No reference? */
 771                 goto out_buf;
 772         if (depth < LUSTRE_DQTREEDEPTH - 1)
 773                 ret = find_tree_dqentry(dquot, blk, depth + 1, version);
 774         else
 775                 ret = find_block_dqentry(dquot, blk, version);
 776 out_buf:
 777         freedqbuf(buf);
 778         return ret;
 779 }
 780
 781 /* Find entry for given id in the tree - wrapper function */
 782 static inline loff_t find_dqentry(struct lustre_dquot *dquot,
 783                                   lustre_quota_version_t version)
 784 {
 785         return find_tree_dqentry(dquot, LUSTRE_DQTREEOFF, 0, version);
 786 }
 787
 788 int lustre_read_dquot(struct lustre_dquot *dquot)
 789 {
 790         int type = dquot->dq_type;
 791         struct file *filp;
 792         mm_segment_t fs;
 793         loff_t offset;
 794         int ret = 0, dqblk_sz;
 795         lustre_quota_version_t version;
 796
 797         /* Invalidated quota? */
 798         if (!dquot->dq_info || !(filp = dquot->dq_info->qi_files[type])) {
 799                 CERROR("VFS: Quota invalidated while reading!\n");
 800                 return -EIO;
 801         }
 802
 803         version = dquot->dq_info->qi_version;
 804         dqblk_sz = lustre_disk_dqblk_sz[version];
 805
 806         offset = find_dqentry(dquot, version);
 807         if (offset <= 0) {      /* Entry not present? */
 808                 if (offset < 0)
 809                         CERROR("VFS: Can't read quota structure for id %u.\n",
 810                                dquot->dq_id);
 811                 dquot->dq_off = 0;
 812                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
 813                 memset(&dquot->dq_dqb, 0, sizeof(struct lustre_mem_dqblk));
 814                 ret = offset;
 815         } else {
 816                 char ddquot[dqblk_sz], empty[dqblk_sz];
 817
 818                 dquot->dq_off = offset;
 819                 fs = get_fs();
 820                 set_fs(KERNEL_DS);
 821                 if ((ret = filp->f_op->read(filp, ddquot, dqblk_sz, &offset)) !=
 822                     dqblk_sz) {
 823                         if (ret >= 0)
 824                                 ret = -EIO;
 825                         CERROR("VFS: Error while reading quota structure "
 826                                "for id %u.\n", dquot->dq_id);
 827                         memset(ddquot, 0, dqblk_sz);
 828                 } else {
 829                         ret = 0;
 830                         /* We need to escape back all-zero structure */
 831                         memset(empty, 0, dqblk_sz);
 832                         DQF_PUT(empty, version, dqb_itime, 1);
 833                         if (!memcmp(empty, ddquot, dqblk_sz))
 834                                 DQF_PUT(ddquot, version, dqb_itime, 0);
 835                 }
 836                 set_fs(fs);
 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
 849         /* always clear the flag so we don't loop on an IO error... */
 850         clear_bit(DQ_MOD_B, &dquot->dq_flags);
 851
 852         /* The block/inode usage in admin quotafile isn't the real usage
 853          * over all cluster, so keep the fake dquot entry on disk is
 854          * meaningless, just remove it */
 855         if (test_bit(DQ_FAKE_B, &dquot->dq_flags))
 856                 rc = lustre_delete_dquot(dquot, version);
 857         else
 858                 rc = lustre_write_dquot(dquot, version);
 859
 860         if (rc < 0)
 861                 return rc;
 862
 863         if (lustre_info_dirty(&dquot->dq_info->qi_info[dquot->dq_type]))
 864                 rc = lustre_write_quota_info(dquot->dq_info, dquot->dq_type);
 865
 866         return rc;
 867 }
 868
 869 int lustre_init_quota_header(struct lustre_quota_info *lqi, int type, int fakemagics)
 870 {
 871         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 872         static const uint fake_magics[] = LUSTRE_BADQMAGICS;
 873         const uint* quota_versions = lustre_initqversions[lqi->qi_version];
 874         struct lustre_disk_dqheader dqhead;
 875         ssize_t size;
 876         loff_t offset = 0;
 877         struct file *fp = lqi->qi_files[type];
 878         int rc = 0;
 879
 880         /* write quotafile header */
 881         dqhead.dqh_magic = cpu_to_le32(fakemagics ?
 882                                        fake_magics[type] : quota_magics[type]);
 883         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
 884         size = fp->f_op->write(fp, (char *)&dqhead,
 885                                sizeof(struct lustre_disk_dqheader), &offset);
 886
 887         if (size != sizeof(struct lustre_disk_dqheader)) {
 888                 CERROR("error writing quotafile header (rc:%d)\n", rc);
 889                 rc = size;
 890         }
 891
 892         return rc;
 893 }
 894
 895 /* We need to export this function to initialize quotafile, because we haven't
 896  * user level check utility */
 897 int lustre_init_quota_info_generic(struct lustre_quota_info *lqi, int type,
 898                                    int fakemagics)
 899 {
 900         struct lustre_mem_dqinfo *dqinfo = &lqi->qi_info[type];
 901         int rc;
 902
 903         rc = lustre_init_quota_header(lqi, type, fakemagics);
 904         if (rc)
 905                 return rc;
 906
 907         /* write init quota info */
 908         memset(dqinfo, 0, sizeof(*dqinfo));
 909         dqinfo->dqi_bgrace = MAX_DQ_TIME;
 910         dqinfo->dqi_igrace = MAX_IQ_TIME;
 911         dqinfo->dqi_blocks = LUSTRE_DQTREEOFF + 1;
 912
 913         return lustre_write_quota_info(lqi, type);
 914 }
 915
 916 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type)
 917 {
 918         return lustre_init_quota_info_generic(lqi, type, 0);
 919 }
 920
 921 ssize_t quota_read(struct file *file, struct inode *inode, int type,
 922                    uint blk, dqbuf_t buf)
 923 {
 924         if (file) {
 925                 return read_blk(file, blk, buf);
 926         } else {
 927 #ifndef KERNEL_SUPPORTS_QUOTA_READ
 928                 return -ENOTSUPP;
 929 #else
 930                 struct super_block *sb = inode->i_sb;
 931                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 932                 return sb->s_op->quota_read(sb, type, (char *)buf,
 933                                             LUSTRE_DQBLKSIZE,
 934                                             blk << LUSTRE_DQBLKSIZE_BITS);
 935 #endif
 936         }
 937 }
 938
 939 static int walk_block_dqentry(struct file *filp, struct inode *inode, int type,
 940                               uint blk, struct list_head *list)
 941 {
 942         dqbuf_t buf = getdqbuf();
 943         loff_t ret = 0;
 944         struct lustre_disk_dqdbheader *dqhead =
 945             (struct lustre_disk_dqdbheader *)buf;
 946         struct dqblk *blk_item;
 947         struct dqblk *pos;
 948         struct list_head *tmp;
 949
 950         if (!buf)
 951                 return -ENOMEM;
 952         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
 953                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 954                 goto out_buf;
 955         }
 956         ret = 0;
 957
 958         if (!le32_to_cpu(dqhead->dqdh_entries))
 959                 goto out_buf;
 960
 961         if (list_empty(list)) {
 962                 tmp = list;
 963                 goto done;
 964         }
 965
 966         list_for_each_entry(pos, list, link) {
 967                 if (blk == pos->blk)    /* we got this blk already */
 968                         goto out_buf;
 969                 if (blk > pos->blk)
 970                         continue;
 971                 break;
 972         }
 973         tmp = &pos->link;
 974 done:
 975         blk_item = kmalloc(sizeof(*blk_item), GFP_NOFS);
 976         if (!blk_item) {
 977                 ret = -ENOMEM;
 978                 goto out_buf;
 979         }
 980         blk_item->blk = blk;
 981         INIT_LIST_HEAD(&blk_item->link);
 982
 983         list_add_tail(&blk_item->link, tmp);
 984
 985 out_buf:
 986         freedqbuf(buf);
 987         return ret;
 988 }
 989
 990 int walk_tree_dqentry(struct file *filp, struct inode *inode, int type,
 991                       uint blk, int depth, struct list_head *list)
 992 {
 993         dqbuf_t buf = getdqbuf();
 994         loff_t ret = 0;
 995         int index;
 996         u32 *ref = (u32 *) buf;
 997
 998         if (!buf)
 999                 return -ENOMEM;
1000         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
1001                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
1002                 goto out_buf;
1003         }
1004         ret = 0;
1005
1006         for (index = 0; index <= 0xff && !ret; index++) {
1007                 blk = le32_to_cpu(ref[index]);
1008                 if (!blk)       /* No reference */
1009                         continue;
1010
1011                 if (depth < LUSTRE_DQTREEDEPTH - 1)
1012                         ret = walk_tree_dqentry(filp, inode, type, blk,
1013                                                 depth + 1, list);
1014                 else
1015                         ret = walk_block_dqentry(filp, inode, type, blk, list);
1016         }
1017 out_buf:
1018         freedqbuf(buf);
1019         return ret;
1020 }
1021
1022 /* Walk through the quota file (v2 format) to get all ids with quota limit */
1023 int lustre_get_qids(struct file *fp, struct inode *inode, int type,
1024                     struct list_head *list)
1025 {
1026         struct list_head blk_list;
1027         struct dqblk *blk_item, *tmp;
1028         dqbuf_t buf = NULL;
1029         char *ddquot;
1030         int rc;
1031         lustre_quota_version_t version;
1032
1033         ENTRY;
1034
1035         if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V1) == 0)
1036                 version = LUSTRE_QUOTA_V1;
1037         else if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V2) == 0)
1038                 version = LUSTRE_QUOTA_V2;
1039         else {
1040                 CERROR("unknown quota file format!\n");
1041                 RETURN(-EINVAL);
1042         }
1043
1044         if (!list_empty(list)) {
1045                 CERROR("not empty list\n");
1046                 RETURN(-EINVAL);
1047         }
1048
1049         INIT_LIST_HEAD(&blk_list);
1050         rc = walk_tree_dqentry(fp, inode, type, LUSTRE_DQTREEOFF, 0, &blk_list);
1051         if (rc) {
1052                 CERROR("walk through quota file failed!(%d)\n", rc);
1053                 GOTO(out_free, rc);
1054         }
1055         if (list_empty(&blk_list))
1056                 RETURN(0);
1057
1058         buf = getdqbuf();
1059         if (!buf)
1060                 RETURN(-ENOMEM);
1061         ddquot = GETENTRIES(buf, version);
1062
1063         list_for_each_entry(blk_item, &blk_list, link) {
1064                 loff_t ret = 0;
1065                 int i, dqblk_sz = lustre_disk_dqblk_sz[version];
1066                 char fakedquot[dqblk_sz];
1067
1068                 memset(buf, 0, LUSTRE_DQBLKSIZE);
1069                 if ((ret = quota_read(fp, inode, type, blk_item->blk, buf))<0) {
1070                         CERROR("VFS: Can't read quota tree block %u.\n",
1071                                blk_item->blk);
1072                         GOTO(out_free, rc = ret);
1073                 }
1074
1075                 memset(fakedquot, 0, dqblk_sz);
1076                 for (i = 0; i < lustre_dqstrinblk[version]; i++) {
1077                         struct dquot_id *dqid;
1078                         /* skip empty entry */
1079                         if (!memcmp(fakedquot, ddquot + i*dqblk_sz, dqblk_sz))
1080                                 continue;
1081
1082                         dqid = kmalloc(sizeof(*dqid), GFP_NOFS);
1083                         if (!dqid)
1084                                 GOTO(out_free, rc = -ENOMEM);
1085
1086                         dqid->di_id = DQF_GET(ddquot + i * dqblk_sz,
1087                                               version, dqb_id);
1088                         INIT_LIST_HEAD(&dqid->di_link);
1089                         list_add(&dqid->di_link, list);
1090                 }
1091         }
1092
1093 out_free:
1094         list_for_each_entry_safe(blk_item, tmp, &blk_list, link) {
1095                 list_del_init(&blk_item->link);
1096                 kfree(blk_item);
1097         }
1098         if (buf)
1099                 freedqbuf(buf);
1100
1101         RETURN(rc);
1102 }
1103
1104
1105 EXPORT_SYMBOL(lustre_read_quota_info);
1106 EXPORT_SYMBOL(lustre_write_quota_info);
1107 EXPORT_SYMBOL(lustre_check_quota_file);
1108 EXPORT_SYMBOL(lustre_read_dquot);
1109 EXPORT_SYMBOL(lustre_commit_dquot);
1110 EXPORT_SYMBOL(lustre_init_quota_info);
1111 EXPORT_SYMBOL(lustre_get_qids);
1112 #endif