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 struct workqueue_struct *ocfs2_quota_wq = NULL;
  67
  68 static void qsync_work_fn(struct work_struct *work);
  69
  70 static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
  71 {
  72         struct ocfs2_global_disk_dqblk *d = dp;
  73         struct mem_dqblk *m = &dquot->dq_dqb;
  74
  75         /* Update from disk only entries not set by the admin */
  76         if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
  77                 m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
  78                 m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
  79         }
  80         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
  81                 m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
  82         if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
  83                 m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
  84                 m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
  85         }
  86         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
  87                 m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
  88         if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
  89                 m->dqb_btime = le64_to_cpu(d->dqb_btime);
  90         if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
  91                 m->dqb_itime = le64_to_cpu(d->dqb_itime);
  92         OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
  93 }
  94
  95 static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
  96 {
  97         struct ocfs2_global_disk_dqblk *d = dp;
  98         struct mem_dqblk *m = &dquot->dq_dqb;
  99
 100         d->dqb_id = cpu_to_le32(dquot->dq_id);
 101         d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
 102         d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
 103         d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
 104         d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
 105         d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
 106         d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
 107         d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
 108         d->dqb_btime = cpu_to_le64(m->dqb_btime);
 109         d->dqb_itime = cpu_to_le64(m->dqb_itime);
 110         d->dqb_pad1 = d->dqb_pad2 = 0;
 111 }
 112
 113 static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
 114 {
 115         struct ocfs2_global_disk_dqblk *d = dp;
 116         struct ocfs2_mem_dqinfo *oinfo =
 117                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
 118
 119         if (qtree_entry_unused(&oinfo->dqi_gi, dp))
 120                 return 0;
 121         return le32_to_cpu(d->dqb_id) == dquot->dq_id;
 122 }
 123
 124 struct qtree_fmt_operations ocfs2_global_ops = {
 125         .mem2disk_dqblk = ocfs2_global_mem2diskdqb,
 126         .disk2mem_dqblk = ocfs2_global_disk2memdqb,
 127         .is_id = ocfs2_global_is_id,
 128 };
 129
 130 int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
 131 {
 132         struct ocfs2_disk_dqtrailer *dqt =
 133                 ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
 134
 135         trace_ocfs2_validate_quota_block((unsigned long long)bh->b_blocknr);
 136
 137         BUG_ON(!buffer_uptodate(bh));
 138
 139         /*
 140          * If the ecc fails, we return the error but otherwise
 141          * leave the filesystem running.  We know any error is
 142          * local to this block.
 143          */
 144         return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
 145 }
 146
 147 int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
 148                                 struct buffer_head **bhp)
 149 {
 150         int rc;
 151
 152         *bhp = NULL;
 153         rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
 154                                ocfs2_validate_quota_block);
 155         if (rc)
 156                 mlog_errno(rc);
 157         return rc;
 158 }
 159
 160 /* Read data from global quotafile - avoid pagecache and such because we cannot
 161  * afford acquiring the locks... We use quota cluster lock to serialize
 162  * operations. Caller is responsible for acquiring it. */
 163 ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
 164                          size_t len, loff_t off)
 165 {
 166         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 167         struct inode *gqinode = oinfo->dqi_gqinode;
 168         loff_t i_size = i_size_read(gqinode);
 169         int offset = off & (sb->s_blocksize - 1);
 170         sector_t blk = off >> sb->s_blocksize_bits;
 171         int err = 0;
 172         struct buffer_head *bh;
 173         size_t toread, tocopy;
 174         u64 pblock = 0, pcount = 0;
 175
 176         if (off > i_size)
 177                 return 0;
 178         if (off + len > i_size)
 179                 len = i_size - off;
 180         toread = len;
 181         while (toread > 0) {
 182                 tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
 183                 if (!pcount) {
 184                         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
 185                                                           &pcount, NULL);
 186                         if (err) {
 187                                 mlog_errno(err);
 188                                 return err;
 189                         }
 190                 } else {
 191                         pcount--;
 192                         pblock++;
 193                 }
 194                 bh = NULL;
 195                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 196                 if (err) {
 197                         mlog_errno(err);
 198                         return err;
 199                 }
 200                 memcpy(data, bh->b_data + offset, tocopy);
 201                 brelse(bh);
 202                 offset = 0;
 203                 toread -= tocopy;
 204                 data += tocopy;
 205                 blk++;
 206         }
 207         return len;
 208 }
 209
 210 /* Write to quotafile (we know the transaction is already started and has
 211  * enough credits) */
 212 ssize_t ocfs2_quota_write(struct super_block *sb, int type,
 213                           const char *data, size_t len, loff_t off)
 214 {
 215         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 216         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 217         struct inode *gqinode = oinfo->dqi_gqinode;
 218         int offset = off & (sb->s_blocksize - 1);
 219         sector_t blk = off >> sb->s_blocksize_bits;
 220         int err = 0, new = 0, ja_type;
 221         struct buffer_head *bh = NULL;
 222         handle_t *handle = journal_current_handle();
 223         u64 pblock, pcount;
 224
 225         if (!handle) {
 226                 mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
 227                      "because transaction was not started.\n",
 228                      (unsigned long long)off, (unsigned long long)len);
 229                 return -EIO;
 230         }
 231         if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
 232                 WARN_ON(1);
 233                 len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
 234         }
 235
 236         if (gqinode->i_size < off + len) {
 237                 loff_t rounded_end =
 238                                 ocfs2_align_bytes_to_blocks(sb, off + len);
 239
 240                 /* Space is already allocated in ocfs2_acquire_dquot() */
 241                 err = ocfs2_simple_size_update(gqinode,
 242                                                oinfo->dqi_gqi_bh,
 243                                                rounded_end);
 244                 if (err < 0)
 245                         goto out;
 246                 new = 1;
 247         }
 248         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
 249         if (err) {
 250                 mlog_errno(err);
 251                 goto out;
 252         }
 253         /* Not rewriting whole block? */
 254         if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
 255             !new) {
 256                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 257                 ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
 258         } else {
 259                 bh = sb_getblk(sb, pblock);
 260                 if (!bh)
 261                         err = -ENOMEM;
 262                 ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
 263         }
 264         if (err) {
 265                 mlog_errno(err);
 266                 goto out;
 267         }
 268         lock_buffer(bh);
 269         if (new)
 270                 memset(bh->b_data, 0, sb->s_blocksize);
 271         memcpy(bh->b_data + offset, data, len);
 272         flush_dcache_page(bh->b_page);
 273         set_buffer_uptodate(bh);
 274         unlock_buffer(bh);
 275         ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
 276         err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
 277                                       ja_type);
 278         if (err < 0) {
 279                 brelse(bh);
 280                 goto out;
 281         }
 282         ocfs2_journal_dirty(handle, bh);
 283         brelse(bh);
 284 out:
 285         if (err) {
 286                 mlog_errno(err);
 287                 return err;
 288         }
 289         gqinode->i_version++;
 290         ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
 291         return len;
 292 }
 293
 294 int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 295 {
 296         int status;
 297         struct buffer_head *bh = NULL;
 298
 299         status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
 300         if (status < 0)
 301                 return status;
 302         spin_lock(&dq_data_lock);
 303         if (!oinfo->dqi_gqi_count++)
 304                 oinfo->dqi_gqi_bh = bh;
 305         else
 306                 WARN_ON(bh != oinfo->dqi_gqi_bh);
 307         spin_unlock(&dq_data_lock);
 308         if (ex) {
 309                 mutex_lock(&oinfo->dqi_gqinode->i_mutex);
 310                 down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 311         } else {
 312                 down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 313         }
 314         return 0;
 315 }
 316
 317 void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 318 {
 319         if (ex) {
 320                 up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 321                 mutex_unlock(&oinfo->dqi_gqinode->i_mutex);
 322         } else {
 323                 up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 324         }
 325         ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
 326         brelse(oinfo->dqi_gqi_bh);
 327         spin_lock(&dq_data_lock);
 328         if (!--oinfo->dqi_gqi_count)
 329                 oinfo->dqi_gqi_bh = NULL;
 330         spin_unlock(&dq_data_lock);
 331 }
 332
 333 /* Read information header from global quota file */
 334 int ocfs2_global_read_info(struct super_block *sb, int type)
 335 {
 336         struct inode *gqinode = NULL;
 337         unsigned int ino[MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
 338                                         GROUP_QUOTA_SYSTEM_INODE };
 339         struct ocfs2_global_disk_dqinfo dinfo;
 340         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 341         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 342         u64 pcount;
 343         int status;
 344
 345         /* Read global header */
 346         gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
 347                         OCFS2_INVALID_SLOT);
 348         if (!gqinode) {
 349                 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
 350                         type);
 351                 status = -EINVAL;
 352                 goto out_err;
 353         }
 354         oinfo->dqi_gi.dqi_sb = sb;
 355         oinfo->dqi_gi.dqi_type = type;
 356         ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
 357         oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
 358         oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
 359         oinfo->dqi_gqi_bh = NULL;
 360         oinfo->dqi_gqi_count = 0;
 361         oinfo->dqi_gqinode = gqinode;
 362         status = ocfs2_lock_global_qf(oinfo, 0);
 363         if (status < 0) {
 364                 mlog_errno(status);
 365                 goto out_err;
 366         }
 367
 368         status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
 369                                              &pcount, NULL);
 370         if (status < 0)
 371                 goto out_unlock;
 372
 373         status = ocfs2_qinfo_lock(oinfo, 0);
 374         if (status < 0)
 375                 goto out_unlock;
 376         status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
 377                                       sizeof(struct ocfs2_global_disk_dqinfo),
 378                                       OCFS2_GLOBAL_INFO_OFF);
 379         ocfs2_qinfo_unlock(oinfo, 0);
 380         ocfs2_unlock_global_qf(oinfo, 0);
 381         if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
 382                 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
 383                      status);
 384                 if (status >= 0)
 385                         status = -EIO;
 386                 mlog_errno(status);
 387                 goto out_err;
 388         }
 389         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 390         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 391         oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
 392         oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 393         oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 394         oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 395         oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
 396         oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
 397                                                 OCFS2_QBLK_RESERVED_SPACE;
 398         oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
 399         INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
 400         queue_delayed_work(ocfs2_quota_wq, &oinfo->dqi_sync_work,
 401                            msecs_to_jiffies(oinfo->dqi_syncms));
 402
 403 out_err:
 404         if (status)
 405                 mlog_errno(status);
 406         return status;
 407 out_unlock:
 408         ocfs2_unlock_global_qf(oinfo, 0);
 409         mlog_errno(status);
 410         goto out_err;
 411 }
 412
 413 /* Write information to global quota file. Expects exlusive lock on quota
 414  * file inode and quota info */
 415 static int __ocfs2_global_write_info(struct super_block *sb, int type)
 416 {
 417         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 418         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 419         struct ocfs2_global_disk_dqinfo dinfo;
 420         ssize_t size;
 421
 422         spin_lock(&dq_data_lock);
 423         info->dqi_flags &= ~DQF_INFO_DIRTY;
 424         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 425         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 426         spin_unlock(&dq_data_lock);
 427         dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
 428         dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
 429         dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
 430         dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
 431         size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
 432                                      sizeof(struct ocfs2_global_disk_dqinfo),
 433                                      OCFS2_GLOBAL_INFO_OFF);
 434         if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
 435                 mlog(ML_ERROR, "Cannot write global quota info structure\n");
 436                 if (size >= 0)
 437                         size = -EIO;
 438                 return size;
 439         }
 440         return 0;
 441 }
 442
 443 int ocfs2_global_write_info(struct super_block *sb, int type)
 444 {
 445         int err;
 446         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 447
 448         err = ocfs2_qinfo_lock(info, 1);
 449         if (err < 0)
 450                 return err;
 451         err = __ocfs2_global_write_info(sb, type);
 452         ocfs2_qinfo_unlock(info, 1);
 453         return err;
 454 }
 455
 456 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
 457 {
 458         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 459
 460         /*
 461          * We may need to allocate tree blocks and a leaf block but not the
 462          * root block
 463          */
 464         return oinfo->dqi_gi.dqi_qtree_depth;
 465 }
 466
 467 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
 468 {
 469         /* We modify all the allocated blocks, tree root, info block and
 470          * the inode */
 471         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
 472                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
 473 }
 474
 475 /* Sync local information about quota modifications with global quota file.
 476  * Caller must have started the transaction and obtained exclusive lock for
 477  * global quota file inode */
 478 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
 479 {
 480         int err, err2;
 481         struct super_block *sb = dquot->dq_sb;
 482         int type = dquot->dq_type;
 483         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 484         struct ocfs2_global_disk_dqblk dqblk;
 485         s64 spacechange, inodechange;
 486         time_t olditime, oldbtime;
 487
 488         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
 489                                    sizeof(struct ocfs2_global_disk_dqblk),
 490                                    dquot->dq_off);
 491         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
 492                 if (err >= 0) {
 493                         mlog(ML_ERROR, "Short read from global quota file "
 494                                        "(%u read)\n", err);
 495                         err = -EIO;
 496                 }
 497                 goto out;
 498         }
 499
 500         /* Update space and inode usage. Get also other information from
 501          * global quota file so that we don't overwrite any changes there.
 502          * We are */
 503         spin_lock(&dq_data_lock);
 504         spacechange = dquot->dq_dqb.dqb_curspace -
 505                                         OCFS2_DQUOT(dquot)->dq_origspace;
 506         inodechange = dquot->dq_dqb.dqb_curinodes -
 507                                         OCFS2_DQUOT(dquot)->dq_originodes;
 508         olditime = dquot->dq_dqb.dqb_itime;
 509         oldbtime = dquot->dq_dqb.dqb_btime;
 510         ocfs2_global_disk2memdqb(dquot, &dqblk);
 511         trace_ocfs2_sync_dquot(dquot->dq_id, dquot->dq_dqb.dqb_curspace,
 512                                (long long)spacechange,
 513                                dquot->dq_dqb.dqb_curinodes,
 514                                (long long)inodechange);
 515         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
 516                 dquot->dq_dqb.dqb_curspace += spacechange;
 517         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
 518                 dquot->dq_dqb.dqb_curinodes += inodechange;
 519         /* Set properly space grace time... */
 520         if (dquot->dq_dqb.dqb_bsoftlimit &&
 521             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
 522                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
 523                     oldbtime > 0) {
 524                         if (dquot->dq_dqb.dqb_btime > 0)
 525                                 dquot->dq_dqb.dqb_btime =
 526                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
 527                         else
 528                                 dquot->dq_dqb.dqb_btime = oldbtime;
 529                 }
 530         } else {
 531                 dquot->dq_dqb.dqb_btime = 0;
 532                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 533         }
 534         /* Set properly inode grace time... */
 535         if (dquot->dq_dqb.dqb_isoftlimit &&
 536             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
 537                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
 538                     olditime > 0) {
 539                         if (dquot->dq_dqb.dqb_itime > 0)
 540                                 dquot->dq_dqb.dqb_itime =
 541                                         min(dquot->dq_dqb.dqb_itime, olditime);
 542                         else
 543                                 dquot->dq_dqb.dqb_itime = olditime;
 544                 }
 545         } else {
 546                 dquot->dq_dqb.dqb_itime = 0;
 547                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
 548         }
 549         /* All information is properly updated, clear the flags */
 550         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
 551         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
 552         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
 553         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
 554         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
 555         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
 556         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 557         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 558         spin_unlock(&dq_data_lock);
 559         err = ocfs2_qinfo_lock(info, freeing);
 560         if (err < 0) {
 561                 mlog(ML_ERROR, "Failed to lock quota info, loosing quota write"
 562                                " (type=%d, id=%u)\n", dquot->dq_type,
 563                                (unsigned)dquot->dq_id);
 564                 goto out;
 565         }
 566         if (freeing)
 567                 OCFS2_DQUOT(dquot)->dq_use_count--;
 568         err = qtree_write_dquot(&info->dqi_gi, dquot);
 569         if (err < 0)
 570                 goto out_qlock;
 571         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
 572                 err = qtree_release_dquot(&info->dqi_gi, dquot);
 573                 if (info_dirty(sb_dqinfo(sb, type))) {
 574                         err2 = __ocfs2_global_write_info(sb, type);
 575                         if (!err)
 576                                 err = err2;
 577                 }
 578         }
 579 out_qlock:
 580         ocfs2_qinfo_unlock(info, freeing);
 581 out:
 582         if (err < 0)
 583                 mlog_errno(err);
 584         return err;
 585 }
 586
 587 /*
 588  *  Functions for periodic syncing of dquots with global file
 589  */
 590 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
 591 {
 592         handle_t *handle;
 593         struct super_block *sb = dquot->dq_sb;
 594         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 595         struct ocfs2_super *osb = OCFS2_SB(sb);
 596         int status = 0;
 597
 598         trace_ocfs2_sync_dquot_helper(dquot->dq_id, dquot->dq_type,
 599                                       type, sb->s_id);
 600         if (type != dquot->dq_type)
 601                 goto out;
 602         status = ocfs2_lock_global_qf(oinfo, 1);
 603         if (status < 0)
 604                 goto out;
 605
 606         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 607         if (IS_ERR(handle)) {
 608                 status = PTR_ERR(handle);
 609                 mlog_errno(status);
 610                 goto out_ilock;
 611         }
 612         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
 613         status = ocfs2_sync_dquot(dquot);
 614         if (status < 0)
 615                 mlog_errno(status);
 616         /* We have to write local structure as well... */
 617         status = ocfs2_local_write_dquot(dquot);
 618         if (status < 0)
 619                 mlog_errno(status);
 620         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
 621         ocfs2_commit_trans(osb, handle);
 622 out_ilock:
 623         ocfs2_unlock_global_qf(oinfo, 1);
 624 out:
 625         return status;
 626 }
 627
 628 static void qsync_work_fn(struct work_struct *work)
 629 {
 630         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
 631                                                       struct ocfs2_mem_dqinfo,
 632                                                       dqi_sync_work.work);
 633         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
 634
 635         dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
 636         queue_delayed_work(ocfs2_quota_wq, &oinfo->dqi_sync_work,
 637                            msecs_to_jiffies(oinfo->dqi_syncms));
 638 }
 639
 640 /*
 641  *  Wrappers for generic quota functions
 642  */
 643
 644 static int ocfs2_write_dquot(struct dquot *dquot)
 645 {
 646         handle_t *handle;
 647         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 648         int status = 0;
 649
 650         trace_ocfs2_write_dquot(dquot->dq_id, dquot->dq_type);
 651
 652         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
 653         if (IS_ERR(handle)) {
 654                 status = PTR_ERR(handle);
 655                 mlog_errno(status);
 656                 goto out;
 657         }
 658         mutex_lock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
 659         status = ocfs2_local_write_dquot(dquot);
 660         mutex_unlock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
 661         ocfs2_commit_trans(osb, handle);
 662 out:
 663         return status;
 664 }
 665
 666 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
 667 {
 668         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 669         /*
 670          * We modify tree, leaf block, global info, local chunk header,
 671          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
 672          * accounts for inode update
 673          */
 674         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
 675                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
 676                OCFS2_QINFO_WRITE_CREDITS +
 677                OCFS2_INODE_UPDATE_CREDITS;
 678 }
 679
 680 static int ocfs2_release_dquot(struct dquot *dquot)
 681 {
 682         handle_t *handle;
 683         struct ocfs2_mem_dqinfo *oinfo =
 684                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
 685         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 686         int status = 0;
 687
 688         trace_ocfs2_release_dquot(dquot->dq_id, dquot->dq_type);
 689
 690         mutex_lock(&dquot->dq_lock);
 691         /* Check whether we are not racing with some other dqget() */
 692         if (atomic_read(&dquot->dq_count) > 1)
 693                 goto out;
 694         status = ocfs2_lock_global_qf(oinfo, 1);
 695         if (status < 0)
 696                 goto out;
 697         handle = ocfs2_start_trans(osb,
 698                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_type));
 699         if (IS_ERR(handle)) {
 700                 status = PTR_ERR(handle);
 701                 mlog_errno(status);
 702                 goto out_ilock;
 703         }
 704
 705         status = ocfs2_global_release_dquot(dquot);
 706         if (status < 0) {
 707                 mlog_errno(status);
 708                 goto out_trans;
 709         }
 710         status = ocfs2_local_release_dquot(handle, dquot);
 711         /*
 712          * If we fail here, we cannot do much as global structure is
 713          * already released. So just complain...
 714          */
 715         if (status < 0)
 716                 mlog_errno(status);
 717         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 718 out_trans:
 719         ocfs2_commit_trans(osb, handle);
 720 out_ilock:
 721         ocfs2_unlock_global_qf(oinfo, 1);
 722 out:
 723         mutex_unlock(&dquot->dq_lock);
 724         if (status)
 725                 mlog_errno(status);
 726         return status;
 727 }
 728
 729 /*
 730  * Read global dquot structure from disk or create it if it does
 731  * not exist. Also update use count of the global structure and
 732  * create structure in node-local quota file.
 733  */
 734 static int ocfs2_acquire_dquot(struct dquot *dquot)
 735 {
 736         int status = 0, err;
 737         int ex = 0;
 738         struct super_block *sb = dquot->dq_sb;
 739         struct ocfs2_super *osb = OCFS2_SB(sb);
 740         int type = dquot->dq_type;
 741         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 742         struct inode *gqinode = info->dqi_gqinode;
 743         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
 744         handle_t *handle;
 745
 746         trace_ocfs2_acquire_dquot(dquot->dq_id, type);
 747         mutex_lock(&dquot->dq_lock);
 748         /*
 749          * We need an exclusive lock, because we're going to update use count
 750          * and instantiate possibly new dquot structure
 751          */
 752         status = ocfs2_lock_global_qf(info, 1);
 753         if (status < 0)
 754                 goto out;
 755         if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
 756                 status = ocfs2_qinfo_lock(info, 0);
 757                 if (status < 0)
 758                         goto out_dq;
 759                 status = qtree_read_dquot(&info->dqi_gi, dquot);
 760                 ocfs2_qinfo_unlock(info, 0);
 761                 if (status < 0)
 762                         goto out_dq;
 763         }
 764         set_bit(DQ_READ_B, &dquot->dq_flags);
 765
 766         OCFS2_DQUOT(dquot)->dq_use_count++;
 767         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 768         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 769         if (!dquot->dq_off) {   /* No real quota entry? */
 770                 ex = 1;
 771                 /*
 772                  * Add blocks to quota file before we start a transaction since
 773                  * locking allocators ranks above a transaction start
 774                  */
 775                 WARN_ON(journal_current_handle());
 776                 status = ocfs2_extend_no_holes(gqinode, NULL,
 777                         gqinode->i_size + (need_alloc << sb->s_blocksize_bits),
 778                         gqinode->i_size);
 779                 if (status < 0)
 780                         goto out_dq;
 781         }
 782
 783         handle = ocfs2_start_trans(osb,
 784                                    ocfs2_calc_global_qinit_credits(sb, type));
 785         if (IS_ERR(handle)) {
 786                 status = PTR_ERR(handle);
 787                 goto out_dq;
 788         }
 789         status = ocfs2_qinfo_lock(info, ex);
 790         if (status < 0)
 791                 goto out_trans;
 792         status = qtree_write_dquot(&info->dqi_gi, dquot);
 793         if (ex && info_dirty(sb_dqinfo(sb, type))) {
 794                 err = __ocfs2_global_write_info(sb, type);
 795                 if (!status)
 796                         status = err;
 797         }
 798         ocfs2_qinfo_unlock(info, ex);
 799 out_trans:
 800         ocfs2_commit_trans(osb, handle);
 801 out_dq:
 802         ocfs2_unlock_global_qf(info, 1);
 803         if (status < 0)
 804                 goto out;
 805
 806         status = ocfs2_create_local_dquot(dquot);
 807         if (status < 0)
 808                 goto out;
 809         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 810 out:
 811         mutex_unlock(&dquot->dq_lock);
 812         if (status)
 813                 mlog_errno(status);
 814         return status;
 815 }
 816
 817 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
 818 {
 819         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
 820                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
 821                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
 822                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
 823                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
 824                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
 825         int sync = 0;
 826         int status;
 827         struct super_block *sb = dquot->dq_sb;
 828         int type = dquot->dq_type;
 829         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 830         handle_t *handle;
 831         struct ocfs2_super *osb = OCFS2_SB(sb);
 832
 833         trace_ocfs2_mark_dquot_dirty(dquot->dq_id, type);
 834
 835         /* In case user set some limits, sync dquot immediately to global
 836          * quota file so that information propagates quicker */
 837         spin_lock(&dq_data_lock);
 838         if (dquot->dq_flags & mask)
 839                 sync = 1;
 840         spin_unlock(&dq_data_lock);
 841         /* This is a slight hack but we can't afford getting global quota
 842          * lock if we already have a transaction started. */
 843         if (!sync || journal_current_handle()) {
 844                 status = ocfs2_write_dquot(dquot);
 845                 goto out;
 846         }
 847         status = ocfs2_lock_global_qf(oinfo, 1);
 848         if (status < 0)
 849                 goto out;
 850         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 851         if (IS_ERR(handle)) {
 852                 status = PTR_ERR(handle);
 853                 mlog_errno(status);
 854                 goto out_ilock;
 855         }
 856         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
 857         status = ocfs2_sync_dquot(dquot);
 858         if (status < 0) {
 859                 mlog_errno(status);
 860                 goto out_dlock;
 861         }
 862         /* Now write updated local dquot structure */
 863         status = ocfs2_local_write_dquot(dquot);
 864 out_dlock:
 865         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
 866         ocfs2_commit_trans(osb, handle);
 867 out_ilock:
 868         ocfs2_unlock_global_qf(oinfo, 1);
 869 out:
 870         if (status)
 871                 mlog_errno(status);
 872         return status;
 873 }
 874
 875 /* This should happen only after set_dqinfo(). */
 876 static int ocfs2_write_info(struct super_block *sb, int type)
 877 {
 878         handle_t *handle;
 879         int status = 0;
 880         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 881
 882         status = ocfs2_lock_global_qf(oinfo, 1);
 883         if (status < 0)
 884                 goto out;
 885         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
 886         if (IS_ERR(handle)) {
 887                 status = PTR_ERR(handle);
 888                 mlog_errno(status);
 889                 goto out_ilock;
 890         }
 891         status = dquot_commit_info(sb, type);
 892         ocfs2_commit_trans(OCFS2_SB(sb), handle);
 893 out_ilock:
 894         ocfs2_unlock_global_qf(oinfo, 1);
 895 out:
 896         if (status)
 897                 mlog_errno(status);
 898         return status;
 899 }
 900
 901 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
 902 {
 903         struct ocfs2_dquot *dquot =
 904                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
 905
 906         if (!dquot)
 907                 return NULL;
 908         return &dquot->dq_dquot;
 909 }
 910
 911 static void ocfs2_destroy_dquot(struct dquot *dquot)
 912 {
 913         kmem_cache_free(ocfs2_dquot_cachep, dquot);
 914 }
 915
 916 const struct dquot_operations ocfs2_quota_operations = {
 917         /* We never make dquot dirty so .write_dquot is never called */
 918         .acquire_dquot  = ocfs2_acquire_dquot,
 919         .release_dquot  = ocfs2_release_dquot,
 920         .mark_dirty     = ocfs2_mark_dquot_dirty,
 921         .write_info     = ocfs2_write_info,
 922         .alloc_dquot    = ocfs2_alloc_dquot,
 923         .destroy_dquot  = ocfs2_destroy_dquot,
 924 };
 925
 926 int ocfs2_quota_setup(void)
 927 {
 928         ocfs2_quota_wq = create_workqueue("o2quot");
 929         if (!ocfs2_quota_wq)
 930                 return -ENOMEM;
 931         return 0;
 932 }
 933
 934 void ocfs2_quota_shutdown(void)
 935 {
 936         if (ocfs2_quota_wq) {
 937                 flush_workqueue(ocfs2_quota_wq);
 938                 destroy_workqueue(ocfs2_quota_wq);
 939                 ocfs2_quota_wq = NULL;
 940         }
 941 }