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         if (status)
 403                 mlog_errno(status);
 404         return status;
 405 out_unlock:
 406         ocfs2_unlock_global_qf(oinfo, 0);
 407         mlog_errno(status);
 408         goto out_err;
 409 }
 410
 411 /* Write information to global quota file. Expects exlusive lock on quota
 412  * file inode and quota info */
 413 static int __ocfs2_global_write_info(struct super_block *sb, int type)
 414 {
 415         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 416         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 417         struct ocfs2_global_disk_dqinfo dinfo;
 418         ssize_t size;
 419
 420         spin_lock(&dq_data_lock);
 421         info->dqi_flags &= ~DQF_INFO_DIRTY;
 422         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 423         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 424         spin_unlock(&dq_data_lock);
 425         dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
 426         dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
 427         dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
 428         dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
 429         size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
 430                                      sizeof(struct ocfs2_global_disk_dqinfo),
 431                                      OCFS2_GLOBAL_INFO_OFF);
 432         if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
 433                 mlog(ML_ERROR, "Cannot write global quota info structure\n");
 434                 if (size >= 0)
 435                         size = -EIO;
 436                 return size;
 437         }
 438         return 0;
 439 }
 440
 441 int ocfs2_global_write_info(struct super_block *sb, int type)
 442 {
 443         int err;
 444         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 445
 446         err = ocfs2_qinfo_lock(info, 1);
 447         if (err < 0)
 448                 return err;
 449         err = __ocfs2_global_write_info(sb, type);
 450         ocfs2_qinfo_unlock(info, 1);
 451         return err;
 452 }
 453
 454 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
 455 {
 456         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 457
 458         /*
 459          * We may need to allocate tree blocks and a leaf block but not the
 460          * root block
 461          */
 462         return oinfo->dqi_gi.dqi_qtree_depth;
 463 }
 464
 465 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
 466 {
 467         /* We modify all the allocated blocks, tree root, info block and
 468          * the inode */
 469         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
 470                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
 471 }
 472
 473 /* Sync local information about quota modifications with global quota file.
 474  * Caller must have started the transaction and obtained exclusive lock for
 475  * global quota file inode */
 476 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
 477 {
 478         int err, err2;
 479         struct super_block *sb = dquot->dq_sb;
 480         int type = dquot->dq_type;
 481         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 482         struct ocfs2_global_disk_dqblk dqblk;
 483         s64 spacechange, inodechange;
 484         time_t olditime, oldbtime;
 485
 486         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
 487                                    sizeof(struct ocfs2_global_disk_dqblk),
 488                                    dquot->dq_off);
 489         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
 490                 if (err >= 0) {
 491                         mlog(ML_ERROR, "Short read from global quota file "
 492                                        "(%u read)\n", err);
 493                         err = -EIO;
 494                 }
 495                 goto out;
 496         }
 497
 498         /* Update space and inode usage. Get also other information from
 499          * global quota file so that we don't overwrite any changes there.
 500          * We are */
 501         spin_lock(&dq_data_lock);
 502         spacechange = dquot->dq_dqb.dqb_curspace -
 503                                         OCFS2_DQUOT(dquot)->dq_origspace;
 504         inodechange = dquot->dq_dqb.dqb_curinodes -
 505                                         OCFS2_DQUOT(dquot)->dq_originodes;
 506         olditime = dquot->dq_dqb.dqb_itime;
 507         oldbtime = dquot->dq_dqb.dqb_btime;
 508         ocfs2_global_disk2memdqb(dquot, &dqblk);
 509         trace_ocfs2_sync_dquot(dquot->dq_id, dquot->dq_dqb.dqb_curspace,
 510                                (long long)spacechange,
 511                                dquot->dq_dqb.dqb_curinodes,
 512                                (long long)inodechange);
 513         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
 514                 dquot->dq_dqb.dqb_curspace += spacechange;
 515         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
 516                 dquot->dq_dqb.dqb_curinodes += inodechange;
 517         /* Set properly space grace time... */
 518         if (dquot->dq_dqb.dqb_bsoftlimit &&
 519             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
 520                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
 521                     oldbtime > 0) {
 522                         if (dquot->dq_dqb.dqb_btime > 0)
 523                                 dquot->dq_dqb.dqb_btime =
 524                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
 525                         else
 526                                 dquot->dq_dqb.dqb_btime = oldbtime;
 527                 }
 528         } else {
 529                 dquot->dq_dqb.dqb_btime = 0;
 530                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 531         }
 532         /* Set properly inode grace time... */
 533         if (dquot->dq_dqb.dqb_isoftlimit &&
 534             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
 535                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
 536                     olditime > 0) {
 537                         if (dquot->dq_dqb.dqb_itime > 0)
 538                                 dquot->dq_dqb.dqb_itime =
 539                                         min(dquot->dq_dqb.dqb_itime, olditime);
 540                         else
 541                                 dquot->dq_dqb.dqb_itime = olditime;
 542                 }
 543         } else {
 544                 dquot->dq_dqb.dqb_itime = 0;
 545                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
 546         }
 547         /* All information is properly updated, clear the flags */
 548         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
 549         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
 550         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
 551         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
 552         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
 553         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
 554         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 555         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 556         spin_unlock(&dq_data_lock);
 557         err = ocfs2_qinfo_lock(info, freeing);
 558         if (err < 0) {
 559                 mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
 560                                " (type=%d, id=%u)\n", dquot->dq_type,
 561                                (unsigned)dquot->dq_id);
 562                 goto out;
 563         }
 564         if (freeing)
 565                 OCFS2_DQUOT(dquot)->dq_use_count--;
 566         err = qtree_write_dquot(&info->dqi_gi, dquot);
 567         if (err < 0)
 568                 goto out_qlock;
 569         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
 570                 err = qtree_release_dquot(&info->dqi_gi, dquot);
 571                 if (info_dirty(sb_dqinfo(sb, type))) {
 572                         err2 = __ocfs2_global_write_info(sb, type);
 573                         if (!err)
 574                                 err = err2;
 575                 }
 576         }
 577 out_qlock:
 578         ocfs2_qinfo_unlock(info, freeing);
 579 out:
 580         if (err < 0)
 581                 mlog_errno(err);
 582         return err;
 583 }
 584
 585 /*
 586  *  Functions for periodic syncing of dquots with global file
 587  */
 588 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
 589 {
 590         handle_t *handle;
 591         struct super_block *sb = dquot->dq_sb;
 592         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 593         struct ocfs2_super *osb = OCFS2_SB(sb);
 594         int status = 0;
 595
 596         trace_ocfs2_sync_dquot_helper(dquot->dq_id, dquot->dq_type,
 597                                       type, sb->s_id);
 598         if (type != dquot->dq_type)
 599                 goto out;
 600         status = ocfs2_lock_global_qf(oinfo, 1);
 601         if (status < 0)
 602                 goto out;
 603
 604         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 605         if (IS_ERR(handle)) {
 606                 status = PTR_ERR(handle);
 607                 mlog_errno(status);
 608                 goto out_ilock;
 609         }
 610         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
 611         status = ocfs2_sync_dquot(dquot);
 612         if (status < 0)
 613                 mlog_errno(status);
 614         /* We have to write local structure as well... */
 615         status = ocfs2_local_write_dquot(dquot);
 616         if (status < 0)
 617                 mlog_errno(status);
 618         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
 619         ocfs2_commit_trans(osb, handle);
 620 out_ilock:
 621         ocfs2_unlock_global_qf(oinfo, 1);
 622 out:
 623         return status;
 624 }
 625
 626 static void qsync_work_fn(struct work_struct *work)
 627 {
 628         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
 629                                                       struct ocfs2_mem_dqinfo,
 630                                                       dqi_sync_work.work);
 631         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
 632
 633         dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
 634         schedule_delayed_work(&oinfo->dqi_sync_work,
 635                               msecs_to_jiffies(oinfo->dqi_syncms));
 636 }
 637
 638 /*
 639  *  Wrappers for generic quota functions
 640  */
 641
 642 static int ocfs2_write_dquot(struct dquot *dquot)
 643 {
 644         handle_t *handle;
 645         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 646         int status = 0;
 647
 648         trace_ocfs2_write_dquot(dquot->dq_id, dquot->dq_type);
 649
 650         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
 651         if (IS_ERR(handle)) {
 652                 status = PTR_ERR(handle);
 653                 mlog_errno(status);
 654                 goto out;
 655         }
 656         mutex_lock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
 657         status = ocfs2_local_write_dquot(dquot);
 658         mutex_unlock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
 659         ocfs2_commit_trans(osb, handle);
 660 out:
 661         return status;
 662 }
 663
 664 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
 665 {
 666         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 667         /*
 668          * We modify tree, leaf block, global info, local chunk header,
 669          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
 670          * accounts for inode update
 671          */
 672         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
 673                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
 674                OCFS2_QINFO_WRITE_CREDITS +
 675                OCFS2_INODE_UPDATE_CREDITS;
 676 }
 677
 678 static int ocfs2_release_dquot(struct dquot *dquot)
 679 {
 680         handle_t *handle;
 681         struct ocfs2_mem_dqinfo *oinfo =
 682                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
 683         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 684         int status = 0;
 685
 686         trace_ocfs2_release_dquot(dquot->dq_id, dquot->dq_type);
 687
 688         mutex_lock(&dquot->dq_lock);
 689         /* Check whether we are not racing with some other dqget() */
 690         if (atomic_read(&dquot->dq_count) > 1)
 691                 goto out;
 692         status = ocfs2_lock_global_qf(oinfo, 1);
 693         if (status < 0)
 694                 goto out;
 695         handle = ocfs2_start_trans(osb,
 696                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_type));
 697         if (IS_ERR(handle)) {
 698                 status = PTR_ERR(handle);
 699                 mlog_errno(status);
 700                 goto out_ilock;
 701         }
 702
 703         status = ocfs2_global_release_dquot(dquot);
 704         if (status < 0) {
 705                 mlog_errno(status);
 706                 goto out_trans;
 707         }
 708         status = ocfs2_local_release_dquot(handle, dquot);
 709         /*
 710          * If we fail here, we cannot do much as global structure is
 711          * already released. So just complain...
 712          */
 713         if (status < 0)
 714                 mlog_errno(status);
 715         /*
 716          * Clear dq_off so that we search for the structure in quota file next
 717          * time we acquire it. The structure might be deleted and reallocated
 718          * elsewhere by another node while our dquot structure is on freelist.
 719          */
 720         dquot->dq_off = 0;
 721         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 722 out_trans:
 723         ocfs2_commit_trans(osb, handle);
 724 out_ilock:
 725         ocfs2_unlock_global_qf(oinfo, 1);
 726 out:
 727         mutex_unlock(&dquot->dq_lock);
 728         if (status)
 729                 mlog_errno(status);
 730         return status;
 731 }
 732
 733 /*
 734  * Read global dquot structure from disk or create it if it does
 735  * not exist. Also update use count of the global structure and
 736  * create structure in node-local quota file.
 737  */
 738 static int ocfs2_acquire_dquot(struct dquot *dquot)
 739 {
 740         int status = 0, err;
 741         int ex = 0;
 742         struct super_block *sb = dquot->dq_sb;
 743         struct ocfs2_super *osb = OCFS2_SB(sb);
 744         int type = dquot->dq_type;
 745         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 746         struct inode *gqinode = info->dqi_gqinode;
 747         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
 748         handle_t *handle;
 749
 750         trace_ocfs2_acquire_dquot(dquot->dq_id, type);
 751         mutex_lock(&dquot->dq_lock);
 752         /*
 753          * We need an exclusive lock, because we're going to update use count
 754          * and instantiate possibly new dquot structure
 755          */
 756         status = ocfs2_lock_global_qf(info, 1);
 757         if (status < 0)
 758                 goto out;
 759         status = ocfs2_qinfo_lock(info, 0);
 760         if (status < 0)
 761                 goto out_dq;
 762         /*
 763          * We always want to read dquot structure from disk because we don't
 764          * know what happened with it while it was on freelist.
 765          */
 766         status = qtree_read_dquot(&info->dqi_gi, dquot);
 767         ocfs2_qinfo_unlock(info, 0);
 768         if (status < 0)
 769                 goto out_dq;
 770
 771         OCFS2_DQUOT(dquot)->dq_use_count++;
 772         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 773         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 774         if (!dquot->dq_off) {   /* No real quota entry? */
 775                 ex = 1;
 776                 /*
 777                  * Add blocks to quota file before we start a transaction since
 778                  * locking allocators ranks above a transaction start
 779                  */
 780                 WARN_ON(journal_current_handle());
 781                 status = ocfs2_extend_no_holes(gqinode, NULL,
 782                         gqinode->i_size + (need_alloc << sb->s_blocksize_bits),
 783                         gqinode->i_size);
 784                 if (status < 0)
 785                         goto out_dq;
 786         }
 787
 788         handle = ocfs2_start_trans(osb,
 789                                    ocfs2_calc_global_qinit_credits(sb, type));
 790         if (IS_ERR(handle)) {
 791                 status = PTR_ERR(handle);
 792                 goto out_dq;
 793         }
 794         status = ocfs2_qinfo_lock(info, ex);
 795         if (status < 0)
 796                 goto out_trans;
 797         status = qtree_write_dquot(&info->dqi_gi, dquot);
 798         if (ex && info_dirty(sb_dqinfo(sb, type))) {
 799                 err = __ocfs2_global_write_info(sb, type);
 800                 if (!status)
 801                         status = err;
 802         }
 803         ocfs2_qinfo_unlock(info, ex);
 804 out_trans:
 805         ocfs2_commit_trans(osb, handle);
 806 out_dq:
 807         ocfs2_unlock_global_qf(info, 1);
 808         if (status < 0)
 809                 goto out;
 810
 811         status = ocfs2_create_local_dquot(dquot);
 812         if (status < 0)
 813                 goto out;
 814         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 815 out:
 816         mutex_unlock(&dquot->dq_lock);
 817         if (status)
 818                 mlog_errno(status);
 819         return status;
 820 }
 821
 822 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
 823 {
 824         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
 825                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
 826                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
 827                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
 828                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
 829                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
 830         int sync = 0;
 831         int status;
 832         struct super_block *sb = dquot->dq_sb;
 833         int type = dquot->dq_type;
 834         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 835         handle_t *handle;
 836         struct ocfs2_super *osb = OCFS2_SB(sb);
 837
 838         trace_ocfs2_mark_dquot_dirty(dquot->dq_id, type);
 839
 840         /* In case user set some limits, sync dquot immediately to global
 841          * quota file so that information propagates quicker */
 842         spin_lock(&dq_data_lock);
 843         if (dquot->dq_flags & mask)
 844                 sync = 1;
 845         spin_unlock(&dq_data_lock);
 846         /* This is a slight hack but we can't afford getting global quota
 847          * lock if we already have a transaction started. */
 848         if (!sync || journal_current_handle()) {
 849                 status = ocfs2_write_dquot(dquot);
 850                 goto out;
 851         }
 852         status = ocfs2_lock_global_qf(oinfo, 1);
 853         if (status < 0)
 854                 goto out;
 855         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 856         if (IS_ERR(handle)) {
 857                 status = PTR_ERR(handle);
 858                 mlog_errno(status);
 859                 goto out_ilock;
 860         }
 861         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
 862         status = ocfs2_sync_dquot(dquot);
 863         if (status < 0) {
 864                 mlog_errno(status);
 865                 goto out_dlock;
 866         }
 867         /* Now write updated local dquot structure */
 868         status = ocfs2_local_write_dquot(dquot);
 869 out_dlock:
 870         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
 871         ocfs2_commit_trans(osb, handle);
 872 out_ilock:
 873         ocfs2_unlock_global_qf(oinfo, 1);
 874 out:
 875         if (status)
 876                 mlog_errno(status);
 877         return status;
 878 }
 879
 880 /* This should happen only after set_dqinfo(). */
 881 static int ocfs2_write_info(struct super_block *sb, int type)
 882 {
 883         handle_t *handle;
 884         int status = 0;
 885         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 886
 887         status = ocfs2_lock_global_qf(oinfo, 1);
 888         if (status < 0)
 889                 goto out;
 890         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
 891         if (IS_ERR(handle)) {
 892                 status = PTR_ERR(handle);
 893                 mlog_errno(status);
 894                 goto out_ilock;
 895         }
 896         status = dquot_commit_info(sb, type);
 897         ocfs2_commit_trans(OCFS2_SB(sb), handle);
 898 out_ilock:
 899         ocfs2_unlock_global_qf(oinfo, 1);
 900 out:
 901         if (status)
 902                 mlog_errno(status);
 903         return status;
 904 }
 905
 906 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
 907 {
 908         struct ocfs2_dquot *dquot =
 909                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
 910
 911         if (!dquot)
 912                 return NULL;
 913         return &dquot->dq_dquot;
 914 }
 915
 916 static void ocfs2_destroy_dquot(struct dquot *dquot)
 917 {
 918         kmem_cache_free(ocfs2_dquot_cachep, dquot);
 919 }
 920
 921 const struct dquot_operations ocfs2_quota_operations = {
 922         /* We never make dquot dirty so .write_dquot is never called */
 923         .acquire_dquot  = ocfs2_acquire_dquot,
 924         .release_dquot  = ocfs2_release_dquot,
 925         .mark_dirty     = ocfs2_mark_dquot_dirty,
 926         .write_info     = ocfs2_write_info,
 927         .alloc_dquot    = ocfs2_alloc_dquot,
 928         .destroy_dquot  = ocfs2_destroy_dquot,
 929 };