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