fs/ocfs2/quota_global.c

   1 /*
   2  *  Implementation of operations over global quota file
   3  */
   4 #include <linux/spinlock.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/quota.h>
   8 #include <linux/quotaops.h>
   9 #include <linux/dqblk_qtree.h>
  10 #include <linux/jiffies.h>
  11 #include <linux/writeback.h>
  12 #include <linux/workqueue.h>
  13
  14 #include <cluster/masklog.h>
  15
  16 #include "ocfs2_fs.h"
  17 #include "ocfs2.h"
  18 #include "alloc.h"
  19 #include "blockcheck.h"
  20 #include "inode.h"
  21 #include "journal.h"
  22 #include "file.h"
  23 #include "sysfile.h"
  24 #include "dlmglue.h"
  25 #include "uptodate.h"
  26 #include "super.h"
  27 #include "buffer_head_io.h"
  28 #include "quota.h"
  29 #include "ocfs2_trace.h"
  30
  31 /*
  32  * Locking of quotas with OCFS2 is rather complex. Here are rules that
  33  * should be obeyed by all the functions:
  34  * - any write of quota structure (either to local or global file) is protected
  35  *   by dqio_mutex or dquot->dq_lock.
  36  * - any modification of global quota file holds inode cluster lock, i_mutex,
  37  *   and ip_alloc_sem of the global quota file (achieved by
  38  *   ocfs2_lock_global_qf). It also has to hold qinfo_lock.
  39  * - an allocation of new blocks for local quota file is protected by
  40  *   its ip_alloc_sem
  41  *
  42  * A rough sketch of locking dependencies (lf = local file, gf = global file):
  43  * Normal filesystem operation:
  44  *   start_trans -> dqio_mutex -> write to lf
  45  * Syncing of local and global file:
  46  *   ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
  47  *     write to gf
  48  *                                                     -> write to lf
  49  * Acquire dquot for the first time:
  50  *   dq_lock -> ocfs2_lock_global_qf -> qinfo_lock -> read from gf
  51  *                                   -> alloc space for gf
  52  *                                   -> start_trans -> qinfo_lock -> write to gf
  53  *           -> ip_alloc_sem of lf -> alloc space for lf
  54  *           -> write to lf
  55  * Release last reference to dquot:
  56  *   dq_lock -> ocfs2_lock_global_qf -> start_trans -> qinfo_lock -> write to gf
  57  *           -> write to lf
  58  * Note that all the above operations also hold the inode cluster lock of lf.
  59  * Recovery:
  60  *   inode cluster lock of recovered lf
  61  *     -> read bitmaps -> ip_alloc_sem of lf
  62  *     -> ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
  63  *        write to gf
  64  */
  65
  66 static void qsync_work_fn(struct work_struct *work);
  67
  68 static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
  69 {
  70         struct ocfs2_global_disk_dqblk *d = dp;
  71         struct mem_dqblk *m = &dquot->dq_dqb;
  72
  73         /* Update from disk only entries not set by the admin */
  74         if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
  75                 m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
  76                 m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
  77         }
  78         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
  79                 m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
  80         if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
  81                 m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
  82                 m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
  83         }
  84         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
  85                 m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
  86         if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
  87                 m->dqb_btime = le64_to_cpu(d->dqb_btime);
  88         if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
  89                 m->dqb_itime = le64_to_cpu(d->dqb_itime);
  90         OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
  91 }
  92
  93 static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
  94 {
  95         struct ocfs2_global_disk_dqblk *d = dp;
  96         struct mem_dqblk *m = &dquot->dq_dqb;
  97
  98         d->dqb_id = cpu_to_le32(dquot->dq_id);
  99         d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
 100         d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
 101         d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
 102         d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
 103         d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
 104         d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
 105         d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
 106         d->dqb_btime = cpu_to_le64(m->dqb_btime);
 107         d->dqb_itime = cpu_to_le64(m->dqb_itime);
 108         d->dqb_pad1 = d->dqb_pad2 = 0;
 109 }
 110
 111 static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
 112 {
 113         struct ocfs2_global_disk_dqblk *d = dp;
 114         struct ocfs2_mem_dqinfo *oinfo =
 115                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
 116
 117         if (qtree_entry_unused(&oinfo->dqi_gi, dp))
 118                 return 0;
 119         return le32_to_cpu(d->dqb_id) == dquot->dq_id;
 120 }
 121
 122 struct qtree_fmt_operations ocfs2_global_ops = {
 123         .mem2disk_dqblk = ocfs2_global_mem2diskdqb,
 124         .disk2mem_dqblk = ocfs2_global_disk2memdqb,
 125         .is_id = ocfs2_global_is_id,
 126 };
 127
 128 int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
 129 {
 130         struct ocfs2_disk_dqtrailer *dqt =
 131                 ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
 132
 133         trace_ocfs2_validate_quota_block((unsigned long long)bh->b_blocknr);
 134
 135         BUG_ON(!buffer_uptodate(bh));
 136
 137         /*
 138          * If the ecc fails, we return the error but otherwise
 139          * leave the filesystem running.  We know any error is
 140          * local to this block.
 141          */
 142         return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
 143 }
 144
 145 int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
 146                                 struct buffer_head **bhp)
 147 {
 148         int rc;
 149
 150         *bhp = NULL;
 151         rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
 152                                ocfs2_validate_quota_block);
 153         if (rc)
 154                 mlog_errno(rc);
 155         return rc;
 156 }
 157
 158 /* Read data from global quotafile - avoid pagecache and such because we cannot
 159  * afford acquiring the locks... We use quota cluster lock to serialize
 160  * operations. Caller is responsible for acquiring it. */
 161 ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
 162                          size_t len, loff_t off)
 163 {
 164         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 165         struct inode *gqinode = oinfo->dqi_gqinode;
 166         loff_t i_size = i_size_read(gqinode);
 167         int offset = off & (sb->s_blocksize - 1);
 168         sector_t blk = off >> sb->s_blocksize_bits;
 169         int err = 0;
 170         struct buffer_head *bh;
 171         size_t toread, tocopy;
 172         u64 pblock = 0, pcount = 0;
 173
 174         if (off > i_size)
 175                 return 0;
 176         if (off + len > i_size)
 177                 len = i_size - off;
 178         toread = len;
 179         while (toread > 0) {
 180                 tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
 181                 if (!pcount) {
 182                         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
 183                                                           &pcount, NULL);
 184                         if (err) {
 185                                 mlog_errno(err);
 186                                 return err;
 187                         }
 188                 } else {
 189                         pcount--;
 190                         pblock++;
 191                 }
 192                 bh = NULL;
 193                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 194                 if (err) {
 195                         mlog_errno(err);
 196                         return err;
 197                 }
 198                 memcpy(data, bh->b_data + offset, tocopy);
 199                 brelse(bh);
 200                 offset = 0;
 201                 toread -= tocopy;
 202                 data += tocopy;
 203                 blk++;
 204         }
 205         return len;
 206 }
 207
 208 /* Write to quotafile (we know the transaction is already started and has
 209  * enough credits) */
 210 ssize_t ocfs2_quota_write(struct super_block *sb, int type,
 211                           const char *data, size_t len, loff_t off)
 212 {
 213         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 214         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 215         struct inode *gqinode = oinfo->dqi_gqinode;
 216         int offset = off & (sb->s_blocksize - 1);
 217         sector_t blk = off >> sb->s_blocksize_bits;
 218         int err = 0, new = 0, ja_type;
 219         struct buffer_head *bh = NULL;
 220         handle_t *handle = journal_current_handle();
 221         u64 pblock, pcount;
 222
 223         if (!handle) {
 224                 mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
 225                      "because transaction was not started.\n",
 226                      (unsigned long long)off, (unsigned long long)len);
 227                 return -EIO;
 228         }
 229         if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
 230                 WARN_ON(1);
 231                 len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
 232         }
 233
 234         if (gqinode->i_size < off + len) {
 235                 loff_t rounded_end =
 236                                 ocfs2_align_bytes_to_blocks(sb, off + len);
 237
 238                 /* Space is already allocated in ocfs2_acquire_dquot() */
 239                 err = ocfs2_simple_size_update(gqinode,
 240                                                oinfo->dqi_gqi_bh,
 241                                                rounded_end);
 242                 if (err < 0)
 243                         goto out;
 244                 new = 1;
 245         }
 246         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
 247         if (err) {
 248                 mlog_errno(err);
 249                 goto out;
 250         }
 251         /* Not rewriting whole block? */
 252         if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
 253             !new) {
 254                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 255                 ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
 256         } else {
 257                 bh = sb_getblk(sb, pblock);
 258                 if (!bh)
 259                         err = -ENOMEM;
 260                 ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
 261         }
 262         if (err) {
 263                 mlog_errno(err);
 264                 goto out;
 265         }
 266         lock_buffer(bh);
 267         if (new)
 268                 memset(bh->b_data, 0, sb->s_blocksize);
 269         memcpy(bh->b_data + offset, data, len);
 270         flush_dcache_page(bh->b_page);
 271         set_buffer_uptodate(bh);
 272         unlock_buffer(bh);
 273         ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
 274         err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
 275                                       ja_type);
 276         if (err < 0) {
 277                 brelse(bh);
 278                 goto out;
 279         }
 280         ocfs2_journal_dirty(handle, bh);
 281         brelse(bh);
 282 out:
 283         if (err) {
 284                 mlog_errno(err);
 285                 return err;
 286         }
 287         gqinode->i_version++;
 288         ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
 289         return len;
 290 }
 291
 292 int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 293 {
 294         int status;
 295         struct buffer_head *bh = NULL;
 296
 297         status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
 298         if (status < 0)
 299                 return status;
 300         spin_lock(&dq_data_lock);
 301         if (!oinfo->dqi_gqi_count++)
 302                 oinfo->dqi_gqi_bh = bh;
 303         else
 304                 WARN_ON(bh != oinfo->dqi_gqi_bh);
 305         spin_unlock(&dq_data_lock);
 306         if (ex) {
 307                 mutex_lock(&oinfo->dqi_gqinode->i_mutex);
 308                 down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 309         } else {
 310                 down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 311         }
 312         return 0;
 313 }
 314
 315 void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 316 {
 317         if (ex) {
 318                 up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 319                 mutex_unlock(&oinfo->dqi_gqinode->i_mutex);
 320         } else {
 321                 up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 322         }
 323         ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
 324         brelse(oinfo->dqi_gqi_bh);
 325         spin_lock(&dq_data_lock);
 326         if (!--oinfo->dqi_gqi_count)
 327                 oinfo->dqi_gqi_bh = NULL;
 328         spin_unlock(&dq_data_lock);
 329 }
 330
 331 /* Read information header from global quota file */
 332 int ocfs2_global_read_info(struct super_block *sb, int type)
 333 {
 334         struct inode *gqinode = NULL;
 335         unsigned int ino[MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
 336                                         GROUP_QUOTA_SYSTEM_INODE };
 337         struct ocfs2_global_disk_dqinfo dinfo;
 338         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 339         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 340         u64 pcount;
 341         int status;
 342
 343         /* Read global header */
 344         gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
 345                         OCFS2_INVALID_SLOT);
 346         if (!gqinode) {
 347                 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
 348                         type);
 349                 status = -EINVAL;
 350                 goto out_err;
 351         }
 352         oinfo->dqi_gi.dqi_sb = sb;
 353         oinfo->dqi_gi.dqi_type = type;
 354         ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
 355         oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
 356         oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
 357         oinfo->dqi_gqi_bh = NULL;
 358         oinfo->dqi_gqi_count = 0;
 359         oinfo->dqi_gqinode = gqinode;
 360         status = ocfs2_lock_global_qf(oinfo, 0);
 361         if (status < 0) {
 362                 mlog_errno(status);
 363                 goto out_err;
 364         }
 365
 366         status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
 367                                              &pcount, NULL);
 368         if (status < 0)
 369                 goto out_unlock;
 370
 371         status = ocfs2_qinfo_lock(oinfo, 0);
 372         if (status < 0)
 373                 goto out_unlock;
 374         status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
 375                                       sizeof(struct ocfs2_global_disk_dqinfo),
 376                                       OCFS2_GLOBAL_INFO_OFF);
 377         ocfs2_qinfo_unlock(oinfo, 0);
 378         ocfs2_unlock_global_qf(oinfo, 0);
 379         if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
 380                 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
 381                      status);
 382                 if (status >= 0)
 383                         status = -EIO;
 384                 mlog_errno(status);
 385                 goto out_err;
 386         }
 387         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 388         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 389         oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
 390         oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 391         oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 392         oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 393         oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
 394         oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
 395                                                 OCFS2_QBLK_RESERVED_SPACE;
 396         oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
 397         INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
 398         schedule_delayed_work(&oinfo->dqi_sync_work,
 399                               msecs_to_jiffies(oinfo->dqi_syncms));
 400
 401 out_err:
 402         return status;
 403 out_unlock:
 404         ocfs2_unlock_global_qf(oinfo, 0);
 405         mlog_errno(status);
 406         goto out_err;
 407 }
 408
 409 /* Write information to global quota file. Expects exlusive lock on quota
 410  * file inode and quota info */
 411 static int __ocfs2_global_write_info(struct super_block *sb, int type)
 412 {
 413         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 414         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 415         struct ocfs2_global_disk_dqinfo dinfo;
 416         ssize_t size;
 417
 418         spin_lock(&dq_data_lock);
 419         info->dqi_flags &= ~DQF_INFO_DIRTY;
 420         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 421         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 422         spin_unlock(&dq_data_lock);
 423         dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
 424         dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
 425         dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
 426         dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
 427         size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
 428                                      sizeof(struct ocfs2_global_disk_dqinfo),
 429                                      OCFS2_GLOBAL_INFO_OFF);
 430         if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
 431                 mlog(ML_ERROR, "Cannot write global quota info structure\n");
 432                 if (size >= 0)
 433                         size = -EIO;
 434                 return size;
 435         }
 436         return 0;
 437 }
 438
 439 int ocfs2_global_write_info(struct super_block *sb, int type)
 440 {
 441         int err;
 442         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 443
 444         err = ocfs2_qinfo_lock(info, 1);
 445         if (err < 0)
 446                 return err;
 447         err = __ocfs2_global_write_info(sb, type);
 448         ocfs2_qinfo_unlock(info, 1);
 449         return err;
 450 }
 451
 452 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
 453 {
 454         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 455
 456         /*
 457          * We may need to allocate tree blocks and a leaf block but not the
 458          * root block
 459          */
 460         return oinfo->dqi_gi.dqi_qtree_depth;
 461 }
 462
 463 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
 464 {
 465         /* We modify all the allocated blocks, tree root, info block and
 466          * the inode */
 467         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
 468                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
 469 }
 470
 471 /* Sync local information about quota modifications with global quota file.
 472  * Caller must have started the transaction and obtained exclusive lock for
 473  * global quota file inode */
 474 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
 475 {
 476         int err, err2;
 477         struct super_block *sb = dquot->dq_sb;
 478         int type = dquot->dq_type;
 479         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 480         struct ocfs2_global_disk_dqblk dqblk;
 481         s64 spacechange, inodechange;
 482         time_t olditime, oldbtime;
 483
 484         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
 485                                    sizeof(struct ocfs2_global_disk_dqblk),
 486                                    dquot->dq_off);
 487         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
 488                 if (err >= 0) {
 489                         mlog(ML_ERROR, "Short read from global quota file "
 490                                        "(%u read)\n", err);
 491                         err = -EIO;
 492                 }
 493                 goto out;
 494         }
 495
 496         /* Update space and inode usage. Get also other information from
 497          * global quota file so that we don't overwrite any changes there.
 498          * We are */
 499         spin_lock(&dq_data_lock);
 500         spacechange = dquot->dq_dqb.dqb_curspace -
 501                                         OCFS2_DQUOT(dquot)->dq_origspace;
 502         inodechange = dquot->dq_dqb.dqb_curinodes -
 503                                         OCFS2_DQUOT(dquot)->dq_originodes;
 504         olditime = dquot->dq_dqb.dqb_itime;
 505         oldbtime = dquot->dq_dqb.dqb_btime;
 506         ocfs2_global_disk2memdqb(dquot, &dqblk);
 507         trace_ocfs2_sync_dquot(dquot->dq_id, dquot->dq_dqb.dqb_curspace,
 508                                (long long)spacechange,
 509                                dquot->dq_dqb.dqb_curinodes,
 510                                (long long)inodechange);
 511         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
 512                 dquot->dq_dqb.dqb_curspace += spacechange;
 513         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
 514                 dquot->dq_dqb.dqb_curinodes += inodechange;
 515         /* Set properly space grace time... */
 516         if (dquot->dq_dqb.dqb_bsoftlimit &&
 517             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
 518                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
 519                     oldbtime > 0) {
 520                         if (dquot->dq_dqb.dqb_btime > 0)
 521                                 dquot->dq_dqb.dqb_btime =
 522                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
 523                         else
 524                                 dquot->dq_dqb.dqb_btime = oldbtime;
 525                 }
 526         } else {
 527                 dquot->dq_dqb.dqb_btime = 0;
 528                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 529         }
 530         /* Set properly inode grace time... */
 531         if (dquot->dq_dqb.dqb_isoftlimit &&
 532             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
 533                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
 534                     olditime > 0) {
 535                         if (dquot->dq_dqb.dqb_itime > 0)
 536                                 dquot->dq_dqb.dqb_itime =
 537                                         min(dquot->dq_dqb.dqb_itime, olditime);
 538                         else
 539                                 dquot->dq_dqb.dqb_itime = olditime;
 540                 }
 541         } else {
 542                 dquot->dq_dqb.dqb_itime = 0;
 543                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
 544         }
 545         /* All information is properly updated, clear the flags */
 546         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
 547         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
 548         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
 549         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
 550         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
 551         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
 552         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 553         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 554         spin_unlock(&dq_data_lock);
 555         err = ocfs2_qinfo_lock(info, freeing);
 556         if (err < 0) {
 557                 mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
 558                                " (type=%d, id=%u)\n", dquot->dq_type,
 559                                (unsigned)dquot->dq_id);
 560                 goto out;
 561         }
 562         if (freeing)
 563                 OCFS2_DQUOT(dquot)->dq_use_count--;
 564         err = qtree_write_dquot(&info->dqi_gi, dquot);
 565         if (err < 0)
 566                 goto out_qlock;
 567         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
 568                 err = qtree_release_dquot(&info->dqi_gi, dquot);
 569                 if (info_dirty(sb_dqinfo(sb, type))) {
 570                         err2 = __ocfs2_global_write_info(sb, type);
 571                         if (!err)
 572                                 err = err2;
 573                 }
 574         }
 575 out_qlock:
 576         ocfs2_qinfo_unlock(info, freeing);
 577 out:
 578         if (err < 0)
 579                 mlog_errno(err);
 580         return err;
 581 }
 582
 583 /*
 584  *  Functions for periodic syncing of dquots with global file
 585  */
 586 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
 587 {
 588         handle_t *handle;
 589         struct super_block *sb = dquot->dq_sb;
 590         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 591         struct ocfs2_super *osb = OCFS2_SB(sb);
 592         int status = 0;
 593
 594         trace_ocfs2_sync_dquot_helper(dquot->dq_id, dquot->dq_type,
 595                                       type, sb->s_id);
 596         if (type != dquot->dq_type)
 597                 goto out;
 598         status = ocfs2_lock_global_qf(oinfo, 1);
 599         if (status < 0)
 600                 goto out;
 601
 602         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 603         if (IS_ERR(handle)) {
 604                 status = PTR_ERR(handle);
 605                 mlog_errno(status);
 606                 goto out_ilock;
 607         }
 608         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
 609         status = ocfs2_sync_dquot(dquot);
 610         if (status < 0)
 611                 mlog_errno(status);
 612         /* We have to write local structure as well... */
 613         status = ocfs2_local_write_dquot(dquot);
 614         if (status < 0)
 615                 mlog_errno(status);
 616         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
 617         ocfs2_commit_trans(osb, handle);
 618 out_ilock:
 619         ocfs2_unlock_global_qf(oinfo, 1);
 620 out:
 621         return status;
 622 }
 623
 624 static void qsync_work_fn(struct work_struct *work)
 625 {
 626         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
 627                                                       struct ocfs2_mem_dqinfo,
 628                                                       dqi_sync_work.work);
 629         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
 630
 631         dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
 632         schedule_delayed_work(&oinfo->dqi_sync_work,
 633                               msecs_to_jiffies(oinfo->dqi_syncms));
 634 }
 635
 636 /*
 637  *  Wrappers for generic quota functions
 638  */
 639
 640 static int ocfs2_write_dquot(struct dquot *dquot)
 641 {
 642         handle_t *handle;
 643         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 644         int status = 0;
 645
 646         trace_ocfs2_write_dquot(dquot->dq_id, dquot->dq_type);
 647
 648         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
 649         if (IS_ERR(handle)) {
 650                 status = PTR_ERR(handle);
 651                 mlog_errno(status);
 652                 goto out;
 653         }
 654         mutex_lock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
 655         status = ocfs2_local_write_dquot(dquot);
 656         mutex_unlock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
 657         ocfs2_commit_trans(osb, handle);
 658 out:
 659         return status;
 660 }
 661
 662 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
 663 {
 664         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 665         /*
 666          * We modify tree, leaf block, global info, local chunk header,
 667          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
 668          * accounts for inode update
 669          */
 670         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
 671                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
 672                OCFS2_QINFO_WRITE_CREDITS +
 673                OCFS2_INODE_UPDATE_CREDITS;
 674 }
 675
 676 static int ocfs2_release_dquot(struct dquot *dquot)
 677 {
 678         handle_t *handle;
 679         struct ocfs2_mem_dqinfo *oinfo =
 680                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
 681         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 682         int status = 0;
 683
 684         trace_ocfs2_release_dquot(dquot->dq_id, dquot->dq_type);
 685
 686         mutex_lock(&dquot->dq_lock);
 687         /* Check whether we are not racing with some other dqget() */
 688         if (atomic_read(&dquot->dq_count) > 1)
 689                 goto out;
 690         status = ocfs2_lock_global_qf(oinfo, 1);
 691         if (status < 0)
 692                 goto out;
 693         handle = ocfs2_start_trans(osb,
 694                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_type));
 695         if (IS_ERR(handle)) {
 696                 status = PTR_ERR(handle);
 697                 mlog_errno(status);
 698                 goto out_ilock;
 699         }
 700
 701         status = ocfs2_global_release_dquot(dquot);
 702         if (status < 0) {
 703                 mlog_errno(status);
 704                 goto out_trans;
 705         }
 706         status = ocfs2_local_release_dquot(handle, dquot);
 707         /*
 708          * If we fail here, we cannot do much as global structure is
 709          * already released. So just complain...
 710          */
 711         if (status < 0)
 712                 mlog_errno(status);
 713         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 714 out_trans:
 715         ocfs2_commit_trans(osb, handle);
 716 out_ilock:
 717         ocfs2_unlock_global_qf(oinfo, 1);
 718 out:
 719         mutex_unlock(&dquot->dq_lock);
 720         if (status)
 721                 mlog_errno(status);
 722         return status;
 723 }
 724
 725 /*
 726  * Read global dquot structure from disk or create it if it does
 727  * not exist. Also update use count of the global structure and
 728  * create structure in node-local quota file.
 729  */
 730 static int ocfs2_acquire_dquot(struct dquot *dquot)
 731 {
 732         int status = 0, err;
 733         int ex = 0;
 734         struct super_block *sb = dquot->dq_sb;
 735         struct ocfs2_super *osb = OCFS2_SB(sb);
 736         int type = dquot->dq_type;
 737         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 738         struct inode *gqinode = info->dqi_gqinode;
 739         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
 740         handle_t *handle;
 741
 742         trace_ocfs2_acquire_dquot(dquot->dq_id, type);
 743         mutex_lock(&dquot->dq_lock);
 744         /*
 745          * We need an exclusive lock, because we're going to update use count
 746          * and instantiate possibly new dquot structure
 747          */
 748         status = ocfs2_lock_global_qf(info, 1);
 749         if (status < 0)
 750                 goto out;
 751         if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
 752                 status = ocfs2_qinfo_lock(info, 0);
 753                 if (status < 0)
 754                         goto out_dq;
 755                 status = qtree_read_dquot(&info->dqi_gi, dquot);
 756                 ocfs2_qinfo_unlock(info, 0);
 757                 if (status < 0)
 758                         goto out_dq;
 759         }
 760         set_bit(DQ_READ_B, &dquot->dq_flags);
 761
 762         OCFS2_DQUOT(dquot)->dq_use_count++;
 763         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 764         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 765         if (!dquot->dq_off) {   /* No real quota entry? */
 766                 ex = 1;
 767                 /*
 768                  * Add blocks to quota file before we start a transaction since
 769                  * locking allocators ranks above a transaction start
 770                  */
 771                 WARN_ON(journal_current_handle());
 772                 status = ocfs2_extend_no_holes(gqinode, NULL,
 773                         gqinode->i_size + (need_alloc << sb->s_blocksize_bits),
 774                         gqinode->i_size);
 775                 if (status < 0)
 776                         goto out_dq;
 777         }
 778
 779         handle = ocfs2_start_trans(osb,
 780                                    ocfs2_calc_global_qinit_credits(sb, type));
 781         if (IS_ERR(handle)) {
 782                 status = PTR_ERR(handle);
 783                 goto out_dq;
 784         }
 785         status = ocfs2_qinfo_lock(info, ex);
 786         if (status < 0)
 787                 goto out_trans;
 788         status = qtree_write_dquot(&info->dqi_gi, dquot);
 789         if (ex && info_dirty(sb_dqinfo(sb, type))) {
 790                 err = __ocfs2_global_write_info(sb, type);
 791                 if (!status)
 792                         status = err;
 793         }
 794         ocfs2_qinfo_unlock(info, ex);
 795 out_trans:
 796         ocfs2_commit_trans(osb, handle);
 797 out_dq:
 798         ocfs2_unlock_global_qf(info, 1);
 799         if (status < 0)
 800                 goto out;
 801
 802         status = ocfs2_create_local_dquot(dquot);
 803         if (status < 0)
 804                 goto out;
 805         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 806 out:
 807         mutex_unlock(&dquot->dq_lock);
 808         if (status)
 809                 mlog_errno(status);
 810         return status;
 811 }
 812
 813 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
 814 {
 815         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
 816                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
 817                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
 818                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
 819                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
 820                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
 821         int sync = 0;
 822         int status;
 823         struct super_block *sb = dquot->dq_sb;
 824         int type = dquot->dq_type;
 825         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 826         handle_t *handle;
 827         struct ocfs2_super *osb = OCFS2_SB(sb);
 828
 829         trace_ocfs2_mark_dquot_dirty(dquot->dq_id, type);
 830
 831         /* In case user set some limits, sync dquot immediately to global
 832          * quota file so that information propagates quicker */
 833         spin_lock(&dq_data_lock);
 834         if (dquot->dq_flags & mask)
 835                 sync = 1;
 836         spin_unlock(&dq_data_lock);
 837         /* This is a slight hack but we can't afford getting global quota
 838          * lock if we already have a transaction started. */
 839         if (!sync || journal_current_handle()) {
 840                 status = ocfs2_write_dquot(dquot);
 841                 goto out;
 842         }
 843         status = ocfs2_lock_global_qf(oinfo, 1);
 844         if (status < 0)
 845                 goto out;
 846         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 847         if (IS_ERR(handle)) {
 848                 status = PTR_ERR(handle);
 849                 mlog_errno(status);
 850                 goto out_ilock;
 851         }
 852         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
 853         status = ocfs2_sync_dquot(dquot);
 854         if (status < 0) {
 855                 mlog_errno(status);
 856                 goto out_dlock;
 857         }
 858         /* Now write updated local dquot structure */
 859         status = ocfs2_local_write_dquot(dquot);
 860 out_dlock:
 861         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
 862         ocfs2_commit_trans(osb, handle);
 863 out_ilock:
 864         ocfs2_unlock_global_qf(oinfo, 1);
 865 out:
 866         if (status)
 867                 mlog_errno(status);
 868         return status;
 869 }
 870
 871 /* This should happen only after set_dqinfo(). */
 872 static int ocfs2_write_info(struct super_block *sb, int type)
 873 {
 874         handle_t *handle;
 875         int status = 0;
 876         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 877
 878         status = ocfs2_lock_global_qf(oinfo, 1);
 879         if (status < 0)
 880                 goto out;
 881         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
 882         if (IS_ERR(handle)) {
 883                 status = PTR_ERR(handle);
 884                 mlog_errno(status);
 885                 goto out_ilock;
 886         }
 887         status = dquot_commit_info(sb, type);
 888         ocfs2_commit_trans(OCFS2_SB(sb), handle);
 889 out_ilock:
 890         ocfs2_unlock_global_qf(oinfo, 1);
 891 out:
 892         if (status)
 893                 mlog_errno(status);
 894         return status;
 895 }
 896
 897 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
 898 {
 899         struct ocfs2_dquot *dquot =
 900                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
 901
 902         if (!dquot)
 903                 return NULL;
 904         return &dquot->dq_dquot;
 905 }
 906
 907 static void ocfs2_destroy_dquot(struct dquot *dquot)
 908 {
 909         kmem_cache_free(ocfs2_dquot_cachep, dquot);
 910 }
 911
 912 const struct dquot_operations ocfs2_quota_operations = {
 913         /* We never make dquot dirty so .write_dquot is never called */
 914         .acquire_dquot  = ocfs2_acquire_dquot,
 915         .release_dquot  = ocfs2_release_dquot,
 916         .mark_dirty     = ocfs2_mark_dquot_dirty,
 917         .write_info     = ocfs2_write_info,
 918         .alloc_dquot    = ocfs2_alloc_dquot,
 919         .destroy_dquot  = ocfs2_destroy_dquot,
 920 };