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  * Lustre administrative quota format.
   5  *
   6  *  from
   7  *  linux/fs/quota_v2.c
   8  */
   9
  10 #ifndef EXPORT_SYMTAB
  11 # define EXPORT_SYMTAB
  12 #endif
  13
  14 #include <linux/errno.h>
  15 #include <linux/fs.h>
  16 #include <linux/mount.h>
  17 #include <linux/kernel.h>
  18 #include <linux/init.h>
  19 #include <linux/module.h>
  20 #include <linux/slab.h>
  21 #include <linux/quotaio_v1.h>
  22
  23 #include <asm/byteorder.h>
  24 #include <asm/uaccess.h>
  25
  26 #include <lustre_quota.h>
  27 #include <obd_support.h>
  28 #include "lustre_quota_fmt.h"
  29
  30 static const uint lustre_initqversions[][MAXQUOTAS] = {
  31         [LUSTRE_QUOTA_V1] = LUSTRE_INITQVERSIONS,
  32         [LUSTRE_QUOTA_V2] = LUSTRE_INITQVERSIONS_V2
  33 };
  34
  35 static const int lustre_dqstrinblk[] = {
  36         [LUSTRE_QUOTA_V1] = LUSTRE_DQSTRINBLK,
  37         [LUSTRE_QUOTA_V2] = LUSTRE_DQSTRINBLK_V2
  38 };
  39
  40 static const int lustre_disk_dqblk_sz[] = {
  41         [LUSTRE_QUOTA_V1] = sizeof(struct lustre_disk_dqblk),
  42         [LUSTRE_QUOTA_V2] = sizeof(struct lustre_disk_dqblk_v2)
  43 };
  44
  45 int check_quota_file(struct file *f, struct inode *inode, int type,
  46                      lustre_quota_version_t version)
  47 {
  48         struct lustre_disk_dqheader dqhead;
  49         mm_segment_t fs;
  50         ssize_t size;
  51         loff_t offset = 0;
  52         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
  53         const uint *quota_versions = lustre_initqversions[version];
  54
  55         if (f) {
  56                 fs = get_fs();
  57                 set_fs(KERNEL_DS);
  58                 size = f->f_op->read(f, (char *)&dqhead,
  59                                      sizeof(struct lustre_disk_dqheader),
  60                                      &offset);
  61                 set_fs(fs);
  62         } else {
  63 #ifndef KERNEL_SUPPORTS_QUOTA_READ
  64                 size = 0;
  65 #else
  66                 struct super_block *sb = inode->i_sb;
  67                 size = sb->s_op->quota_read(sb, type, (char *)&dqhead,
  68                                             sizeof(struct lustre_disk_dqheader), 0);
  69 #endif
  70         }
  71         if (size != sizeof(struct lustre_disk_dqheader))
  72                 return -EINVAL;
  73         if (le32_to_cpu(dqhead.dqh_magic) != quota_magics[type] ||
  74             le32_to_cpu(dqhead.dqh_version) != quota_versions[type])
  75                 return -EINVAL;
  76         return 0;
  77 }
  78
  79 /* Check whether given file is really lustre admin quotafile */
  80 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type)
  81 {
  82         struct file *f = lqi->qi_files[type];
  83         return check_quota_file(f, NULL, type, lqi->qi_version);
  84 }
  85
  86 int lustre_read_quota_file_info(struct file* f, struct lustre_mem_dqinfo* info)
  87 {
  88         mm_segment_t fs;
  89         struct lustre_disk_dqinfo dinfo;
  90         ssize_t size;
  91         loff_t offset = LUSTRE_DQINFOOFF;
  92
  93         fs = get_fs();
  94         set_fs(KERNEL_DS);
  95         size = f->f_op->read(f, (char *)&dinfo,
  96                              sizeof(struct lustre_disk_dqinfo), &offset);
  97         set_fs(fs);
  98         if (size != sizeof(struct lustre_disk_dqinfo)) {
  99                 CERROR("Can't read info structure on device %s.\n",
 100                        f->f_vfsmnt->mnt_sb->s_id);
 101                 return -EINVAL;
 102         }
 103         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 104         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 105         info->dqi_flags = le32_to_cpu(dinfo.dqi_flags);
 106         info->dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 107         info->dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 108         info->dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 109         return 0;
 110 }
 111
 112 /* Read information header from quota file */
 113 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type)
 114 {
 115         return lustre_read_quota_file_info(lqi->qi_files[type], &lqi->qi_info[type]);
 116 }
 117
 118 /* Write information header to quota file */
 119 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type)
 120 {
 121         mm_segment_t fs;
 122         struct lustre_disk_dqinfo dinfo;
 123         struct lustre_mem_dqinfo *info = &lqi->qi_info[type];
 124         struct file *f = lqi->qi_files[type];
 125         ssize_t size;
 126         loff_t offset = LUSTRE_DQINFOOFF;
 127
 128         info->dqi_flags &= ~DQF_INFO_DIRTY;
 129         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 130         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 131         dinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK);
 132         dinfo.dqi_blocks = cpu_to_le32(info->dqi_blocks);
 133         dinfo.dqi_free_blk = cpu_to_le32(info->dqi_free_blk);
 134         dinfo.dqi_free_entry = cpu_to_le32(info->dqi_free_entry);
 135         fs = get_fs();
 136         set_fs(KERNEL_DS);
 137         size = f->f_op->write(f, (char *)&dinfo,
 138                               sizeof(struct lustre_disk_dqinfo), &offset);
 139         set_fs(fs);
 140         if (size != sizeof(struct lustre_disk_dqinfo)) {
 141                 printk(KERN_WARNING
 142                        "Can't write info structure on device %s.\n",
 143                        f->f_vfsmnt->mnt_sb->s_id);
 144                 return -1;
 145         }
 146         return 0;
 147 }
 148
 149 #define DQ2MQ(v) ((sizeof(v) == sizeof(__u64)) ? \
 150                 le64_to_cpu(v) : le32_to_cpu(v))
 151
 152 #define MQ2DQ(v,newv) ((sizeof(v) == sizeof(__u64)) ? \
 153                 (v = cpu_to_le64((__u64)newv)) : (v = cpu_to_le32((__u32)newv)))
 154
 155 #define DQF_GET(var,ver,field) ((ver == LUSTRE_QUOTA_V1)?\
 156                 DQ2MQ(((struct lustre_disk_dqblk*)(var))->field):\
 157                 DQ2MQ(((struct lustre_disk_dqblk_v2*)(var))->field))
 158
 159 #define DQF_PUT(var,ver,field,val) ((ver == LUSTRE_QUOTA_V1)?\
 160                 MQ2DQ(((struct lustre_disk_dqblk*)(var))->field, val):\
 161                 MQ2DQ(((struct lustre_disk_dqblk_v2*)(var))->field, val))
 162
 163 void disk2memdqb(struct lustre_mem_dqblk *m, void *d,
 164                  lustre_quota_version_t version)
 165 {
 166         m->dqb_ihardlimit = DQF_GET(d, version, dqb_ihardlimit);
 167         m->dqb_isoftlimit = DQF_GET(d, version, dqb_isoftlimit);
 168         m->dqb_curinodes = DQF_GET(d, version, dqb_curinodes);
 169         m->dqb_itime = DQF_GET(d, version, dqb_itime);
 170         m->dqb_bhardlimit = DQF_GET(d, version, dqb_bhardlimit);
 171         m->dqb_bsoftlimit = DQF_GET(d, version, dqb_bsoftlimit);
 172         m->dqb_curspace = DQF_GET(d, version, dqb_curspace);
 173         m->dqb_btime = DQF_GET(d, version, dqb_btime);
 174 }
 175
 176 static int check_quota_bounds(struct lustre_mem_dqblk *m,
 177                               lustre_quota_version_t version)
 178 {
 179         return (version == LUSTRE_QUOTA_V1  &&
 180                 m->dqb_ihardlimit <= MAX_UL &&
 181                 m->dqb_isoftlimit <= MAX_UL &&
 182                 m->dqb_curinodes <= MAX_UL  &&
 183                 m->dqb_bhardlimit <= MAX_UL &&
 184                 m->dqb_bsoftlimit <= MAX_UL) ||
 185                 version != LUSTRE_QUOTA_V1;
 186 }
 187
 188 static int mem2diskdqb(void *d, struct lustre_mem_dqblk *m,
 189                        qid_t id, lustre_quota_version_t version)
 190 {
 191         if (!check_quota_bounds(m, version))
 192                 return -EINVAL;
 193
 194         DQF_PUT(d, version, dqb_ihardlimit, m->dqb_ihardlimit);
 195         DQF_PUT(d, version, dqb_isoftlimit, m->dqb_isoftlimit);
 196         DQF_PUT(d, version, dqb_curinodes, m->dqb_curinodes);
 197         DQF_PUT(d, version, dqb_itime, m->dqb_itime);
 198         DQF_PUT(d, version, dqb_bhardlimit, m->dqb_bhardlimit);
 199         DQF_PUT(d, version, dqb_bsoftlimit, m->dqb_bsoftlimit);
 200         DQF_PUT(d, version, dqb_curspace, m->dqb_curspace);
 201         DQF_PUT(d, version, dqb_btime, m->dqb_btime);
 202         DQF_PUT(d, version, dqb_id, id);
 203
 204         return 0;
 205 }
 206
 207 dqbuf_t getdqbuf(void)
 208 {
 209         dqbuf_t buf = kmalloc(LUSTRE_DQBLKSIZE, GFP_NOFS);
 210         if (!buf)
 211                 printk(KERN_WARNING
 212                        "VFS: Not enough memory for quota buffers.\n");
 213         return buf;
 214 }
 215
 216 void freedqbuf(dqbuf_t buf)
 217 {
 218         kfree(buf);
 219 }
 220
 221 ssize_t read_blk(struct file *filp, uint blk, dqbuf_t buf)
 222 {
 223         mm_segment_t fs;
 224         ssize_t ret;
 225         loff_t offset = blk << LUSTRE_DQBLKSIZE_BITS;
 226
 227         memset(buf, 0, LUSTRE_DQBLKSIZE);
 228         fs = get_fs();
 229         set_fs(KERNEL_DS);
 230         ret = filp->f_op->read(filp, (char *)buf, LUSTRE_DQBLKSIZE, &offset);
 231         set_fs(fs);
 232         return ret;
 233 }
 234
 235 ssize_t write_blk(struct file *filp, uint blk, dqbuf_t buf)
 236 {
 237         mm_segment_t fs;
 238         ssize_t ret;
 239         loff_t offset = blk << LUSTRE_DQBLKSIZE_BITS;
 240
 241         fs = get_fs();
 242         set_fs(KERNEL_DS);
 243         ret = filp->f_op->write(filp, (char *)buf, LUSTRE_DQBLKSIZE, &offset);
 244         set_fs(fs);
 245         return ret;
 246 }
 247
 248 void lustre_mark_info_dirty(struct lustre_mem_dqinfo *info)
 249 {
 250         set_bit(DQF_INFO_DIRTY_B, &info->dqi_flags);
 251 }
 252
 253 /* Remove empty block from list and return it */
 254 int get_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info)
 255 {
 256         dqbuf_t buf = getdqbuf();
 257         struct lustre_disk_dqdbheader *dh =
 258             (struct lustre_disk_dqdbheader *)buf;
 259         int ret, blk;
 260
 261         if (!buf)
 262                 return -ENOMEM;
 263         if (info->dqi_free_blk) {
 264                 blk = info->dqi_free_blk;
 265                 if ((ret = read_blk(filp, blk, buf)) < 0)
 266                         goto out_buf;
 267                 info->dqi_free_blk = le32_to_cpu(dh->dqdh_next_free);
 268         } else {
 269                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 270                 /* Assure block allocation... */
 271                 if ((ret = write_blk(filp, info->dqi_blocks, buf)) < 0)
 272                         goto out_buf;
 273                 blk = info->dqi_blocks++;
 274         }
 275         lustre_mark_info_dirty(info);
 276         ret = blk;
 277 out_buf:
 278         freedqbuf(buf);
 279         return ret;
 280 }
 281
 282 /* Insert empty block to the list */
 283 int put_free_dqblk(struct file *filp, struct lustre_mem_dqinfo *info,
 284                    dqbuf_t buf, uint blk)
 285 {
 286         struct lustre_disk_dqdbheader *dh =
 287             (struct lustre_disk_dqdbheader *)buf;
 288         int err;
 289
 290         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_blk);
 291         dh->dqdh_prev_free = cpu_to_le32(0);
 292         dh->dqdh_entries = cpu_to_le16(0);
 293         info->dqi_free_blk = blk;
 294         lustre_mark_info_dirty(info);
 295         if ((err = write_blk(filp, blk, buf)) < 0)
 296                 /* Some strange block. We had better leave it... */
 297                 return err;
 298         return 0;
 299 }
 300
 301 /* Remove given block from the list of blocks with free entries */
 302 int remove_free_dqentry(struct file *filp,
 303                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 304                         uint blk)
 305 {
 306         dqbuf_t tmpbuf = getdqbuf();
 307         struct lustre_disk_dqdbheader *dh =
 308             (struct lustre_disk_dqdbheader *)buf;
 309         uint nextblk = le32_to_cpu(dh->dqdh_next_free), prevblk =
 310             le32_to_cpu(dh->dqdh_prev_free);
 311         int err;
 312
 313         if (!tmpbuf)
 314                 return -ENOMEM;
 315         if (nextblk) {
 316                 if ((err = read_blk(filp, nextblk, tmpbuf)) < 0)
 317                         goto out_buf;
 318                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 319                     dh->dqdh_prev_free;
 320                 if ((err = write_blk(filp, nextblk, tmpbuf)) < 0)
 321                         goto out_buf;
 322         }
 323         if (prevblk) {
 324                 if ((err = read_blk(filp, prevblk, tmpbuf)) < 0)
 325                         goto out_buf;
 326                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_next_free =
 327                     dh->dqdh_next_free;
 328                 if ((err = write_blk(filp, prevblk, tmpbuf)) < 0)
 329                         goto out_buf;
 330         } else {
 331                 info->dqi_free_entry = nextblk;
 332                 lustre_mark_info_dirty(info);
 333         }
 334         freedqbuf(tmpbuf);
 335         dh->dqdh_next_free = dh->dqdh_prev_free = cpu_to_le32(0);
 336         err = write_blk(filp, blk, buf);
 337         if (err < 0)      /* No matter whether write succeeds block is out of list */
 338                 CERROR("VFS: Can't write block (%u) with "
 339                        "free entries (rc=%d).\n", blk, err);
 340         return 0;
 341 out_buf:
 342         freedqbuf(tmpbuf);
 343         return err;
 344 }
 345
 346 /* Insert given block to the beginning of list with free entries */
 347 int insert_free_dqentry(struct file *filp,
 348                         struct lustre_mem_dqinfo *info, dqbuf_t buf,
 349                         uint blk)
 350 {
 351         dqbuf_t tmpbuf = getdqbuf();
 352         struct lustre_disk_dqdbheader *dh =
 353             (struct lustre_disk_dqdbheader *)buf;
 354         int err;
 355
 356         if (!tmpbuf)
 357                 return -ENOMEM;
 358         dh->dqdh_next_free = cpu_to_le32(info->dqi_free_entry);
 359         dh->dqdh_prev_free = cpu_to_le32(0);
 360         if ((err = write_blk(filp, blk, buf)) < 0)
 361                 goto out_buf;
 362         if (info->dqi_free_entry) {
 363                 if ((err = read_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 364                         goto out_buf;
 365                 ((struct lustre_disk_dqdbheader *)tmpbuf)->dqdh_prev_free =
 366                     cpu_to_le32(blk);
 367                 if ((err = write_blk(filp, info->dqi_free_entry, tmpbuf)) < 0)
 368                         goto out_buf;
 369         }
 370         freedqbuf(tmpbuf);
 371         info->dqi_free_entry = blk;
 372         lustre_mark_info_dirty(info);
 373         return 0;
 374 out_buf:
 375         freedqbuf(tmpbuf);
 376         return err;
 377 }
 378
 379
 380
 381 /* Find space for dquot */
 382 static uint find_free_dqentry(struct lustre_dquot *dquot, int *err,
 383                               lustre_quota_version_t version)
 384 {
 385         struct lustre_quota_info *lqi = dquot->dq_info;
 386         struct file *filp = lqi->qi_files[dquot->dq_type];
 387         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 388         uint blk, i;
 389         struct lustre_disk_dqdbheader *dh;
 390         void *ddquot;
 391         int dqblk_sz = lustre_disk_dqblk_sz[version];
 392         int dqstrinblk = lustre_dqstrinblk[version];
 393         char fakedquot[dqblk_sz];
 394         dqbuf_t buf;
 395
 396         *err = 0;
 397         if (!(buf = getdqbuf())) {
 398                 *err = -ENOMEM;
 399                 return 0;
 400         }
 401         dh = (struct lustre_disk_dqdbheader *)buf;
 402         ddquot = GETENTRIES(buf, version);
 403         if (info->dqi_free_entry) {
 404                 blk = info->dqi_free_entry;
 405                 if ((*err = read_blk(filp, blk, buf)) < 0)
 406                         goto out_buf;
 407         } else {
 408                 blk = get_free_dqblk(filp, info);
 409                 if ((int)blk < 0) {
 410                         *err = blk;
 411                         freedqbuf(buf);
 412                         return 0;
 413                 }
 414                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 415                 info->dqi_free_entry = blk; /* This is enough as block is
 416                                                already zeroed and entry list
 417                                                is empty... */
 418                 lustre_mark_info_dirty(info);
 419         }
 420
 421         /* Will block be full */
 422         if (le16_to_cpu(dh->dqdh_entries) + 1 >= dqstrinblk)
 423                 if ((*err = remove_free_dqentry(filp, info, buf, blk)) < 0) {
 424                         CERROR("VFS: Can't remove block %u"
 425                                " from entry free list.\n", blk);
 426                         goto out_buf;
 427                 }
 428         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) + 1);
 429         memset(fakedquot, 0, dqblk_sz);
 430         /* Find free structure in block */
 431         for (i = 0; i < dqstrinblk &&
 432              memcmp(fakedquot, (char*)ddquot + i * dqblk_sz,
 433                     sizeof(fakedquot)); i++);
 434
 435         if (i == dqstrinblk) {
 436                 CERROR("VFS: Data block full but it shouldn't.\n");
 437                 *err = -EIO;
 438                 goto out_buf;
 439         }
 440
 441         if ((*err = write_blk(filp, blk, buf)) < 0) {
 442                 CERROR("VFS: Can't write quota data block %u.\n", blk);
 443                 goto out_buf;
 444         }
 445         dquot->dq_off =
 446             (blk << LUSTRE_DQBLKSIZE_BITS) +
 447             sizeof(struct lustre_disk_dqdbheader) +
 448             i * dqblk_sz;
 449         freedqbuf(buf);
 450         return blk;
 451 out_buf:
 452         freedqbuf(buf);
 453         return 0;
 454 }
 455
 456 /* Insert reference to structure into the trie */
 457 static int do_insert_tree(struct lustre_dquot *dquot, uint * treeblk, int depth,
 458                           lustre_quota_version_t version)
 459 {
 460         struct lustre_quota_info *lqi = dquot->dq_info;
 461         struct file *filp = lqi->qi_files[dquot->dq_type];
 462         struct lustre_mem_dqinfo *info = &lqi->qi_info[dquot->dq_type];
 463         dqbuf_t buf;
 464         int ret = 0, newson = 0, newact = 0;
 465         u32 *ref;
 466         uint newblk;
 467
 468         if (!(buf = getdqbuf()))
 469                 return -ENOMEM;
 470         if (!*treeblk) {
 471                 ret = get_free_dqblk(filp, info);
 472                 if (ret < 0)
 473                         goto out_buf;
 474                 *treeblk = ret;
 475                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 476                 newact = 1;
 477         } else {
 478                 if ((ret = read_blk(filp, *treeblk, buf)) < 0) {
 479                         CERROR("VFS: Can't read tree quota block %u.\n",
 480                                *treeblk);
 481                         goto out_buf;
 482                 }
 483         }
 484         ref = (u32 *) buf;
 485         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 486         if (!newblk)
 487                 newson = 1;
 488         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 489
 490                 if (newblk) {
 491                         CERROR("VFS: Inserting already present quota entry "
 492                                "(block %u).\n",
 493                                ref[GETIDINDEX(dquot->dq_id, depth)]);
 494                         ret = -EIO;
 495                         goto out_buf;
 496                 }
 497
 498                 newblk = find_free_dqentry(dquot, &ret, version);
 499         } else
 500                 ret = do_insert_tree(dquot, &newblk, depth + 1, version);
 501         if (newson && ret >= 0) {
 502                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(newblk);
 503                 ret = write_blk(filp, *treeblk, buf);
 504         } else if (newact && ret < 0)
 505                 put_free_dqblk(filp, info, buf, *treeblk);
 506 out_buf:
 507         freedqbuf(buf);
 508         return ret;
 509 }
 510
 511 /* Wrapper for inserting quota structure into tree */
 512 static inline int dq_insert_tree(struct lustre_dquot *dquot,
 513                                  lustre_quota_version_t version)
 514 {
 515         int tmp = LUSTRE_DQTREEOFF;
 516         return do_insert_tree(dquot, &tmp, 0, version);
 517 }
 518
 519 /*
 520  *  We don't have to be afraid of deadlocks as we never have quotas on quota files...
 521  */
 522 static int lustre_write_dquot(struct lustre_dquot *dquot,
 523                               lustre_quota_version_t version)
 524 {
 525         int type = dquot->dq_type;
 526         struct file *filp;
 527         mm_segment_t fs;
 528         loff_t offset;
 529         ssize_t ret;
 530         int dqblk_sz = lustre_disk_dqblk_sz[version];
 531         char ddquot[dqblk_sz], empty[dqblk_sz];
 532
 533         ret = mem2diskdqb(ddquot, &dquot->dq_dqb, dquot->dq_id, version);
 534         if (ret < 0)
 535                 return ret;
 536
 537         if (!dquot->dq_off)
 538                 if ((ret = dq_insert_tree(dquot, version)) < 0) {
 539                         CERROR("VFS: Error %Zd occurred while creating quota.\n",
 540                                ret);
 541                         return ret;
 542                 }
 543         filp = dquot->dq_info->qi_files[type];
 544         offset = dquot->dq_off;
 545         /* Argh... We may need to write structure full of zeroes but that would be
 546          * treated as an empty place by the rest of the code. Format change would
 547          * be definitely cleaner but the problems probably are not worth it */
 548         memset(empty, 0, dqblk_sz);
 549         if (!memcmp(empty, ddquot, dqblk_sz))
 550                 DQF_PUT(ddquot, version, dqb_itime, 1);
 551         fs = get_fs();
 552         set_fs(KERNEL_DS);
 553         ret = filp->f_op->write(filp, ddquot,
 554                                 dqblk_sz, &offset);
 555         set_fs(fs);
 556         if (ret != dqblk_sz) {
 557                 printk(KERN_WARNING "VFS: dquota write failed on dev %s\n",
 558                        filp->f_dentry->d_sb->s_id);
 559                 if (ret >= 0)
 560                         ret = -ENOSPC;
 561         } else
 562                 ret = 0;
 563
 564         return ret;
 565 }
 566
 567 /* Free dquot entry in data block */
 568 static int free_dqentry(struct lustre_dquot *dquot, uint blk,
 569                         lustre_quota_version_t version)
 570 {
 571         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 572         struct lustre_mem_dqinfo *info =
 573             &dquot->dq_info->qi_info[dquot->dq_type];
 574         struct lustre_disk_dqdbheader *dh;
 575         dqbuf_t buf = getdqbuf();
 576         int dqstrinblk = lustre_dqstrinblk[version];
 577         int ret = 0;
 578
 579         if (!buf)
 580                 return -ENOMEM;
 581         if (dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS != blk) {
 582                 CERROR("VFS: Quota structure has offset to other block (%u) "
 583                        "than it should (%u).\n", blk,
 584                        (uint)(dquot->dq_off >> LUSTRE_DQBLKSIZE_BITS));
 585                 goto out_buf;
 586         }
 587         if ((ret = read_blk(filp, blk, buf)) < 0) {
 588                 CERROR("VFS: Can't read quota data block %u\n", blk);
 589                 goto out_buf;
 590         }
 591         dh = (struct lustre_disk_dqdbheader *)buf;
 592         dh->dqdh_entries = cpu_to_le16(le16_to_cpu(dh->dqdh_entries) - 1);
 593         if (!le16_to_cpu(dh->dqdh_entries)) {   /* Block got free? */
 594                 if ((ret = remove_free_dqentry(filp, info, buf, blk)) < 0 ||
 595                     (ret = put_free_dqblk(filp, info, buf, blk)) < 0) {
 596                         CERROR("VFS: Can't move quota data block (%u) "
 597                                "to free list.\n", blk);
 598                         goto out_buf;
 599                 }
 600         } else {
 601                 memset(buf + (dquot->dq_off & ((1<<LUSTRE_DQBLKSIZE_BITS) - 1)),
 602                        0, lustre_disk_dqblk_sz[version]);
 603                 if (le16_to_cpu(dh->dqdh_entries) == dqstrinblk - 1) {
 604                         /* Insert will write block itself */
 605                         if ((ret =
 606                              insert_free_dqentry(filp, info, buf, blk)) < 0) {
 607                                 CERROR("VFS: Can't insert quota data block (%u) "
 608                                        "to free entry list.\n", blk);
 609                                 goto out_buf;
 610                         }
 611                 } else if ((ret = write_blk(filp, blk, buf)) < 0) {
 612                         CERROR("VFS: Can't write quota data block %u\n", blk);
 613                         goto out_buf;
 614                 }
 615         }
 616         dquot->dq_off = 0;      /* Quota is now unattached */
 617 out_buf:
 618         freedqbuf(buf);
 619         return ret;
 620 }
 621
 622 /* Remove reference to dquot from tree */
 623 static int remove_tree(struct lustre_dquot *dquot, uint * blk, int depth,
 624                        lustre_quota_version_t version)
 625 {
 626         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 627         struct lustre_mem_dqinfo *info =
 628             &dquot->dq_info->qi_info[dquot->dq_type];
 629         dqbuf_t buf = getdqbuf();
 630         int ret = 0;
 631         uint newblk;
 632         u32 *ref = (u32 *) buf;
 633
 634         if (!buf)
 635                 return -ENOMEM;
 636         if ((ret = read_blk(filp, *blk, buf)) < 0) {
 637                 CERROR("VFS: Can't read quota data block %u\n", *blk);
 638                 goto out_buf;
 639         }
 640         newblk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 641         if (depth == LUSTRE_DQTREEDEPTH - 1) {
 642                 ret = free_dqentry(dquot, newblk, version);
 643                 newblk = 0;
 644         } else
 645                 ret = remove_tree(dquot, &newblk, depth + 1, version);
 646         if (ret >= 0 && !newblk) {
 647                 int i;
 648                 ref[GETIDINDEX(dquot->dq_id, depth)] = cpu_to_le32(0);
 649                 for (i = 0; i < LUSTRE_DQBLKSIZE && !buf[i]; i++)
 650                         /* Block got empty? */ ;
 651                 /* don't put the root block into free blk list! */
 652                 if (i == LUSTRE_DQBLKSIZE && *blk != LUSTRE_DQTREEOFF) {
 653                         put_free_dqblk(filp, info, buf, *blk);
 654                         *blk = 0;
 655                 } else if ((ret = write_blk(filp, *blk, buf)) < 0)
 656                         CERROR("VFS: Can't write quota tree block %u.\n", *blk);
 657         }
 658 out_buf:
 659         freedqbuf(buf);
 660         return ret;
 661 }
 662
 663 /* Delete dquot from tree */
 664 static int lustre_delete_dquot(struct lustre_dquot *dquot,
 665                                 lustre_quota_version_t version)
 666 {
 667         uint tmp = LUSTRE_DQTREEOFF;
 668
 669         if (!dquot->dq_off)     /* Even not allocated? */
 670                 return 0;
 671         return remove_tree(dquot, &tmp, 0, version);
 672 }
 673
 674 /* Find entry in block */
 675 static loff_t find_block_dqentry(struct lustre_dquot *dquot, uint blk,
 676                                  lustre_quota_version_t version)
 677 {
 678         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 679         dqbuf_t buf = getdqbuf();
 680         loff_t ret = 0;
 681         int i;
 682         char *ddquot = GETENTRIES(buf, version);
 683         int dqblk_sz = lustre_disk_dqblk_sz[version];
 684         int dqstrinblk = lustre_dqstrinblk[version];
 685
 686         if (!buf)
 687                 return -ENOMEM;
 688         if ((ret = read_blk(filp, blk, buf)) < 0) {
 689                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 690                 goto out_buf;
 691         }
 692         if (dquot->dq_id)
 693                 for (i = 0; i < dqstrinblk &&
 694                      DQF_GET(ddquot+i*dqblk_sz, version, dqb_id) != dquot->dq_id;
 695                      i++) ;
 696         else {                  /* ID 0 as a bit more complicated searching... */
 697                 char fakedquot[dqblk_sz];
 698
 699                 memset(fakedquot, 0, sizeof(fakedquot));
 700                 for (i = 0; i < dqstrinblk; i++)
 701                         if (!DQF_GET(ddquot + i*dqblk_sz, version, dqb_id)
 702                             && memcmp(fakedquot, ddquot + i*dqblk_sz,
 703                                       dqblk_sz))
 704                                 break;
 705         }
 706         if (i == dqstrinblk) {
 707                 CERROR("VFS: Quota for id %u referenced but not present.\n",
 708                        dquot->dq_id);
 709                 ret = -EIO;
 710                 goto out_buf;
 711         } else
 712                 ret =
 713                     (blk << LUSTRE_DQBLKSIZE_BITS) +
 714                     sizeof(struct lustre_disk_dqdbheader) +
 715                     i * dqblk_sz;
 716 out_buf:
 717         freedqbuf(buf);
 718         return ret;
 719 }
 720
 721 /* Find entry for given id in the tree */
 722 static loff_t find_tree_dqentry(struct lustre_dquot *dquot, uint blk, int depth,
 723                                 lustre_quota_version_t version)
 724 {
 725         struct file *filp = dquot->dq_info->qi_files[dquot->dq_type];
 726         dqbuf_t buf = getdqbuf();
 727         loff_t ret = 0;
 728         u32 *ref = (u32 *) buf;
 729
 730         if (!buf)
 731                 return -ENOMEM;
 732         if ((ret = read_blk(filp, blk, buf)) < 0) {
 733                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 734                 goto out_buf;
 735         }
 736         ret = 0;
 737         blk = le32_to_cpu(ref[GETIDINDEX(dquot->dq_id, depth)]);
 738         if (!blk)               /* No reference? */
 739                 goto out_buf;
 740         if (depth < LUSTRE_DQTREEDEPTH - 1)
 741                 ret = find_tree_dqentry(dquot, blk, depth + 1, version);
 742         else
 743                 ret = find_block_dqentry(dquot, blk, version);
 744 out_buf:
 745         freedqbuf(buf);
 746         return ret;
 747 }
 748
 749 /* Find entry for given id in the tree - wrapper function */
 750 static inline loff_t find_dqentry(struct lustre_dquot *dquot,
 751                                   lustre_quota_version_t version)
 752 {
 753         return find_tree_dqentry(dquot, LUSTRE_DQTREEOFF, 0, version);
 754 }
 755
 756 int lustre_read_dquot(struct lustre_dquot *dquot)
 757 {
 758         int type = dquot->dq_type;
 759         struct file *filp;
 760         mm_segment_t fs;
 761         loff_t offset;
 762         int ret = 0, dqblk_sz;
 763         lustre_quota_version_t version;
 764
 765         /* Invalidated quota? */
 766         if (!dquot->dq_info || !(filp = dquot->dq_info->qi_files[type])) {
 767                 CERROR("VFS: Quota invalidated while reading!\n");
 768                 return -EIO;
 769         }
 770
 771         version = dquot->dq_info->qi_version;
 772         dqblk_sz = lustre_disk_dqblk_sz[version];
 773
 774         offset = find_dqentry(dquot, version);
 775         if (offset <= 0) {      /* Entry not present? */
 776                 if (offset < 0)
 777                         CERROR("VFS: Can't read quota structure for id %u.\n",
 778                                dquot->dq_id);
 779                 dquot->dq_off = 0;
 780                 set_bit(DQ_FAKE_B, &dquot->dq_flags);
 781                 memset(&dquot->dq_dqb, 0, sizeof(struct lustre_mem_dqblk));
 782                 ret = offset;
 783         } else {
 784                 char ddquot[dqblk_sz], empty[dqblk_sz];
 785
 786                 dquot->dq_off = offset;
 787                 fs = get_fs();
 788                 set_fs(KERNEL_DS);
 789                 if ((ret = filp->f_op->read(filp, ddquot, dqblk_sz, &offset)) !=
 790                     dqblk_sz) {
 791                         if (ret >= 0)
 792                                 ret = -EIO;
 793                         CERROR("VFS: Error while reading quota structure "
 794                                "for id %u.\n", dquot->dq_id);
 795                         memset(ddquot, 0, dqblk_sz);
 796                 } else {
 797                         ret = 0;
 798                         /* We need to escape back all-zero structure */
 799                         memset(empty, 0, dqblk_sz);
 800                         DQF_PUT(empty, version, dqb_itime, 1);
 801                         if (!memcmp(empty, ddquot, dqblk_sz))
 802                                 DQF_PUT(ddquot, version, dqb_itime, 0);
 803                 }
 804                 set_fs(fs);
 805                 disk2memdqb(&dquot->dq_dqb, ddquot, version);
 806         }
 807
 808         return ret;
 809 }
 810
 811 /* Commit changes of dquot to disk - it might also mean deleting it when quota became fake */
 812 int lustre_commit_dquot(struct lustre_dquot *dquot)
 813 {
 814         int rc = 0;
 815         lustre_quota_version_t version = dquot->dq_info->qi_version;
 816
 817         /* always clear the flag so we don't loop on an IO error... */
 818         clear_bit(DQ_MOD_B, &dquot->dq_flags);
 819
 820         /* The block/inode usage in admin quotafile isn't the real usage
 821          * over all cluster, so keep the fake dquot entry on disk is
 822          * meaningless, just remove it */
 823         if (test_bit(DQ_FAKE_B, &dquot->dq_flags))
 824                 rc = lustre_delete_dquot(dquot, version);
 825         else
 826                 rc = lustre_write_dquot(dquot, version);
 827
 828         if (rc < 0)
 829                 return rc;
 830
 831         if (lustre_info_dirty(&dquot->dq_info->qi_info[dquot->dq_type]))
 832                 rc = lustre_write_quota_info(dquot->dq_info, dquot->dq_type);
 833
 834         return rc;
 835 }
 836
 837 int lustre_init_quota_header(struct lustre_quota_info *lqi, int type, int fakemagics)
 838 {
 839         static const uint quota_magics[] = LUSTRE_INITQMAGICS;
 840         static const uint fake_magics[] = LUSTRE_BADQMAGICS;
 841         const uint* quota_versions = lustre_initqversions[lqi->qi_version];
 842         struct lustre_disk_dqheader dqhead;
 843         ssize_t size;
 844         loff_t offset = 0;
 845         struct file *fp = lqi->qi_files[type];
 846         int rc = 0;
 847
 848         /* write quotafile header */
 849         dqhead.dqh_magic = cpu_to_le32(fakemagics ?
 850                                        fake_magics[type] : quota_magics[type]);
 851         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
 852         size = fp->f_op->write(fp, (char *)&dqhead,
 853                                sizeof(struct lustre_disk_dqheader), &offset);
 854
 855         if (size != sizeof(struct lustre_disk_dqheader)) {
 856                 CERROR("error writing quotafile header (rc:%d)\n", rc);
 857                 rc = size;
 858         }
 859
 860         return rc;
 861 }
 862
 863 /* We need to export this function to initialize quotafile, because we haven't
 864  * user level check utility */
 865 int lustre_init_quota_info_generic(struct lustre_quota_info *lqi, int type,
 866                                    int fakemagics)
 867 {
 868         struct lustre_mem_dqinfo *dqinfo = &lqi->qi_info[type];
 869         int rc;
 870
 871         rc = lustre_init_quota_header(lqi, type, fakemagics);
 872         if (rc)
 873                 return rc;
 874
 875         /* write init quota info */
 876         memset(dqinfo, 0, sizeof(*dqinfo));
 877         dqinfo->dqi_bgrace = MAX_DQ_TIME;
 878         dqinfo->dqi_igrace = MAX_IQ_TIME;
 879         dqinfo->dqi_blocks = LUSTRE_DQTREEOFF + 1;
 880
 881         return lustre_write_quota_info(lqi, type);
 882 }
 883
 884 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type)
 885 {
 886         return lustre_init_quota_info_generic(lqi, type, 0);
 887 }
 888
 889 ssize_t quota_read(struct file *file, struct inode *inode, int type,
 890                    uint blk, dqbuf_t buf)
 891 {
 892         if (file) {
 893                 return read_blk(file, blk, buf);
 894         } else {
 895 #ifndef KERNEL_SUPPORTS_QUOTA_READ
 896                 return -ENOTSUPP;
 897 #else
 898                 struct super_block *sb = inode->i_sb;
 899                 memset(buf, 0, LUSTRE_DQBLKSIZE);
 900                 return sb->s_op->quota_read(sb, type, (char *)buf,
 901                                             LUSTRE_DQBLKSIZE,
 902                                             blk << LUSTRE_DQBLKSIZE_BITS);
 903 #endif
 904         }
 905 }
 906
 907 static int walk_block_dqentry(struct file *filp, struct inode *inode, int type,
 908                               uint blk, struct list_head *list)
 909 {
 910         dqbuf_t buf = getdqbuf();
 911         loff_t ret = 0;
 912         struct lustre_disk_dqdbheader *dqhead =
 913             (struct lustre_disk_dqdbheader *)buf;
 914         struct dqblk *blk_item;
 915         struct dqblk *pos;
 916         struct list_head *tmp;
 917
 918         if (!buf)
 919                 return -ENOMEM;
 920         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
 921                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 922                 goto out_buf;
 923         }
 924         ret = 0;
 925
 926         if (!le32_to_cpu(dqhead->dqdh_entries))
 927                 goto out_buf;
 928
 929         if (list_empty(list)) {
 930                 tmp = list;
 931                 goto done;
 932         }
 933
 934         list_for_each_entry(pos, list, link) {
 935                 if (blk == pos->blk)    /* we got this blk already */
 936                         goto out_buf;
 937                 if (blk > pos->blk)
 938                         continue;
 939                 break;
 940         }
 941         tmp = &pos->link;
 942 done:
 943         blk_item = kmalloc(sizeof(*blk_item), GFP_NOFS);
 944         if (!blk_item) {
 945                 ret = -ENOMEM;
 946                 goto out_buf;
 947         }
 948         blk_item->blk = blk;
 949         INIT_LIST_HEAD(&blk_item->link);
 950
 951         list_add_tail(&blk_item->link, tmp);
 952
 953 out_buf:
 954         freedqbuf(buf);
 955         return ret;
 956 }
 957
 958 int walk_tree_dqentry(struct file *filp, struct inode *inode, int type,
 959                       uint blk, int depth, struct list_head *list)
 960 {
 961         dqbuf_t buf = getdqbuf();
 962         loff_t ret = 0;
 963         int index;
 964         u32 *ref = (u32 *) buf;
 965
 966         if (!buf)
 967                 return -ENOMEM;
 968         if ((ret = quota_read(filp, inode, type, blk, buf)) < 0) {
 969                 CERROR("VFS: Can't read quota tree block %u.\n", blk);
 970                 goto out_buf;
 971         }
 972         ret = 0;
 973
 974         for (index = 0; index <= 0xff && !ret; index++) {
 975                 blk = le32_to_cpu(ref[index]);
 976                 if (!blk)       /* No reference */
 977                         continue;
 978
 979                 if (depth < LUSTRE_DQTREEDEPTH - 1)
 980                         ret = walk_tree_dqentry(filp, inode, type, blk,
 981                                                 depth + 1, list);
 982                 else
 983                         ret = walk_block_dqentry(filp, inode, type, blk, list);
 984         }
 985 out_buf:
 986         freedqbuf(buf);
 987         return ret;
 988 }
 989
 990 /* Walk through the quota file (v2 format) to get all ids with quota limit */
 991 int lustre_get_qids(struct file *fp, struct inode *inode, int type,
 992                     struct list_head *list)
 993 {
 994         struct list_head blk_list;
 995         struct dqblk *blk_item, *tmp;
 996         dqbuf_t buf = NULL;
 997         char *ddquot;
 998         int rc;
 999         lustre_quota_version_t version;
1000
1001         ENTRY;
1002
1003         if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V1) == 0)
1004                 version = LUSTRE_QUOTA_V1;
1005         else if (check_quota_file(fp, inode, type, LUSTRE_QUOTA_V2) == 0)
1006                 version = LUSTRE_QUOTA_V2;
1007         else {
1008                 CERROR("unknown quota file format!\n");
1009                 RETURN(-EINVAL);
1010         }
1011
1012         if (!list_empty(list)) {
1013                 CERROR("not empty list\n");
1014                 RETURN(-EINVAL);
1015         }
1016
1017         INIT_LIST_HEAD(&blk_list);
1018         rc = walk_tree_dqentry(fp, inode, type, LUSTRE_DQTREEOFF, 0, &blk_list);
1019         if (rc) {
1020                 CERROR("walk through quota file failed!(%d)\n", rc);
1021                 GOTO(out_free, rc);
1022         }
1023         if (list_empty(&blk_list))
1024                 RETURN(0);
1025
1026         buf = getdqbuf();
1027         if (!buf)
1028                 RETURN(-ENOMEM);
1029         ddquot = GETENTRIES(buf, version);
1030
1031         list_for_each_entry(blk_item, &blk_list, link) {
1032                 loff_t ret = 0;
1033                 int i, dqblk_sz = lustre_disk_dqblk_sz[version];
1034                 char fakedquot[dqblk_sz];
1035
1036                 memset(buf, 0, LUSTRE_DQBLKSIZE);
1037                 if ((ret = quota_read(fp, inode, type, blk_item->blk, buf))<0) {
1038                         CERROR("VFS: Can't read quota tree block %u.\n",
1039                                blk_item->blk);
1040                         GOTO(out_free, rc = ret);
1041                 }
1042
1043                 memset(fakedquot, 0, dqblk_sz);
1044                 for (i = 0; i < lustre_dqstrinblk[version]; i++) {
1045                         struct dquot_id *dqid;
1046                         /* skip empty entry */
1047                         if (!memcmp(fakedquot, ddquot + i*dqblk_sz, dqblk_sz))
1048                                 continue;
1049
1050                         dqid = kmalloc(sizeof(*dqid), GFP_NOFS);
1051                         if (!dqid)
1052                                 GOTO(out_free, rc = -ENOMEM);
1053
1054                         dqid->di_id = DQF_GET(ddquot + i * dqblk_sz,
1055                                               version, dqb_id);
1056                         INIT_LIST_HEAD(&dqid->di_link);
1057                         list_add(&dqid->di_link, list);
1058                 }
1059         }
1060
1061 out_free:
1062         list_for_each_entry_safe(blk_item, tmp, &blk_list, link) {
1063                 list_del_init(&blk_item->link);
1064                 kfree(blk_item);
1065         }
1066         if (buf)
1067                 freedqbuf(buf);
1068
1069         RETURN(rc);
1070 }
1071
1072
1073 EXPORT_SYMBOL(lustre_read_quota_info);
1074 EXPORT_SYMBOL(lustre_write_quota_info);
1075 EXPORT_SYMBOL(lustre_check_quota_file);
1076 EXPORT_SYMBOL(lustre_read_dquot);
1077 EXPORT_SYMBOL(lustre_commit_dquot);
1078 EXPORT_SYMBOL(lustre_init_quota_info);
1079 EXPORT_SYMBOL(lustre_get_qids);