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 #include <linux/llist.h>
  14
  15 #include <cluster/masklog.h>
  16
  17 #include "ocfs2_fs.h"
  18 #include "ocfs2.h"
  19 #include "alloc.h"
  20 #include "blockcheck.h"
  21 #include "inode.h"
  22 #include "journal.h"
  23 #include "file.h"
  24 #include "sysfile.h"
  25 #include "dlmglue.h"
  26 #include "uptodate.h"
  27 #include "super.h"
  28 #include "buffer_head_io.h"
  29 #include "quota.h"
  30 #include "ocfs2_trace.h"
  31
  32 /*
  33  * Locking of quotas with OCFS2 is rather complex. Here are rules that
  34  * should be obeyed by all the functions:
  35  * - any write of quota structure (either to local or global file) is protected
  36  *   by dqio_sem or dquot->dq_lock.
  37  * - any modification of global quota file holds inode cluster lock, i_mutex,
  38  *   and ip_alloc_sem of the global quota file (achieved by
  39  *   ocfs2_lock_global_qf). It also has to hold qinfo_lock.
  40  * - an allocation of new blocks for local quota file is protected by
  41  *   its ip_alloc_sem
  42  *
  43  * A rough sketch of locking dependencies (lf = local file, gf = global file):
  44  * Normal filesystem operation:
  45  *   start_trans -> dqio_sem -> write to lf
  46  * Syncing of local and global file:
  47  *   ocfs2_lock_global_qf -> start_trans -> dqio_sem -> qinfo_lock ->
  48  *     write to gf
  49  *                                                     -> write to lf
  50  * Acquire dquot for the first time:
  51  *   dq_lock -> ocfs2_lock_global_qf -> qinfo_lock -> read from gf
  52  *                                   -> alloc space for gf
  53  *                                   -> start_trans -> qinfo_lock -> write to gf
  54  *           -> ip_alloc_sem of lf -> alloc space for lf
  55  *           -> write to lf
  56  * Release last reference to dquot:
  57  *   dq_lock -> ocfs2_lock_global_qf -> start_trans -> qinfo_lock -> write to gf
  58  *           -> write to lf
  59  * Note that all the above operations also hold the inode cluster lock of lf.
  60  * Recovery:
  61  *   inode cluster lock of recovered lf
  62  *     -> read bitmaps -> ip_alloc_sem of lf
  63  *     -> ocfs2_lock_global_qf -> start_trans -> dqio_sem -> qinfo_lock ->
  64  *        write to gf
  65  */
  66
  67 static void qsync_work_fn(struct work_struct *work);
  68
  69 static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
  70 {
  71         struct ocfs2_global_disk_dqblk *d = dp;
  72         struct mem_dqblk *m = &dquot->dq_dqb;
  73
  74         /* Update from disk only entries not set by the admin */
  75         if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
  76                 m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
  77                 m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
  78         }
  79         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
  80                 m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
  81         if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
  82                 m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
  83                 m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
  84         }
  85         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
  86                 m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
  87         if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
  88                 m->dqb_btime = le64_to_cpu(d->dqb_btime);
  89         if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
  90                 m->dqb_itime = le64_to_cpu(d->dqb_itime);
  91         OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
  92 }
  93
  94 static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
  95 {
  96         struct ocfs2_global_disk_dqblk *d = dp;
  97         struct mem_dqblk *m = &dquot->dq_dqb;
  98
  99         d->dqb_id = cpu_to_le32(from_kqid(&init_user_ns, dquot->dq_id));
 100         d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
 101         d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
 102         d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
 103         d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
 104         d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
 105         d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
 106         d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
 107         d->dqb_btime = cpu_to_le64(m->dqb_btime);
 108         d->dqb_itime = cpu_to_le64(m->dqb_itime);
 109         d->dqb_pad1 = d->dqb_pad2 = 0;
 110 }
 111
 112 static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
 113 {
 114         struct ocfs2_global_disk_dqblk *d = dp;
 115         struct ocfs2_mem_dqinfo *oinfo =
 116                         sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
 117
 118         if (qtree_entry_unused(&oinfo->dqi_gi, dp))
 119                 return 0;
 120
 121         return qid_eq(make_kqid(&init_user_ns, dquot->dq_id.type,
 122                                 le32_to_cpu(d->dqb_id)),
 123                       dquot->dq_id);
 124 }
 125
 126 const struct qtree_fmt_operations ocfs2_global_ops = {
 127         .mem2disk_dqblk = ocfs2_global_mem2diskdqb,
 128         .disk2mem_dqblk = ocfs2_global_disk2memdqb,
 129         .is_id = ocfs2_global_is_id,
 130 };
 131
 132 int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
 133 {
 134         struct ocfs2_disk_dqtrailer *dqt =
 135                 ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
 136
 137         trace_ocfs2_validate_quota_block((unsigned long long)bh->b_blocknr);
 138
 139         BUG_ON(!buffer_uptodate(bh));
 140
 141         /*
 142          * If the ecc fails, we return the error but otherwise
 143          * leave the filesystem running.  We know any error is
 144          * local to this block.
 145          */
 146         return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
 147 }
 148
 149 int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
 150                                 struct buffer_head **bhp)
 151 {
 152         int rc;
 153
 154         *bhp = NULL;
 155         rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
 156                                ocfs2_validate_quota_block);
 157         if (rc)
 158                 mlog_errno(rc);
 159         return rc;
 160 }
 161
 162 /* Read data from global quotafile - avoid pagecache and such because we cannot
 163  * afford acquiring the locks... We use quota cluster lock to serialize
 164  * operations. Caller is responsible for acquiring it. */
 165 ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
 166                          size_t len, loff_t off)
 167 {
 168         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 169         struct inode *gqinode = oinfo->dqi_gqinode;
 170         loff_t i_size = i_size_read(gqinode);
 171         int offset = off & (sb->s_blocksize - 1);
 172         sector_t blk = off >> sb->s_blocksize_bits;
 173         int err = 0;
 174         struct buffer_head *bh;
 175         size_t toread, tocopy;
 176         u64 pblock = 0, pcount = 0;
 177
 178         if (off > i_size)
 179                 return 0;
 180         if (off + len > i_size)
 181                 len = i_size - off;
 182         toread = len;
 183         while (toread > 0) {
 184                 tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
 185                 if (!pcount) {
 186                         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
 187                                                           &pcount, NULL);
 188                         if (err) {
 189                                 mlog_errno(err);
 190                                 return err;
 191                         }
 192                 } else {
 193                         pcount--;
 194                         pblock++;
 195                 }
 196                 bh = NULL;
 197                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 198                 if (err) {
 199                         mlog_errno(err);
 200                         return err;
 201                 }
 202                 memcpy(data, bh->b_data + offset, tocopy);
 203                 brelse(bh);
 204                 offset = 0;
 205                 toread -= tocopy;
 206                 data += tocopy;
 207                 blk++;
 208         }
 209         return len;
 210 }
 211
 212 /* Write to quotafile (we know the transaction is already started and has
 213  * enough credits) */
 214 ssize_t ocfs2_quota_write(struct super_block *sb, int type,
 215                           const char *data, size_t len, loff_t off)
 216 {
 217         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 218         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 219         struct inode *gqinode = oinfo->dqi_gqinode;
 220         int offset = off & (sb->s_blocksize - 1);
 221         sector_t blk = off >> sb->s_blocksize_bits;
 222         int err = 0, new = 0, ja_type;
 223         struct buffer_head *bh = NULL;
 224         handle_t *handle = journal_current_handle();
 225         u64 pblock, pcount;
 226
 227         if (!handle) {
 228                 mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
 229                      "because transaction was not started.\n",
 230                      (unsigned long long)off, (unsigned long long)len);
 231                 return -EIO;
 232         }
 233         if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
 234                 WARN_ON(1);
 235                 len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
 236         }
 237
 238         if (i_size_read(gqinode) < off + len) {
 239                 loff_t rounded_end =
 240                                 ocfs2_align_bytes_to_blocks(sb, off + len);
 241
 242                 /* Space is already allocated in ocfs2_acquire_dquot() */
 243                 err = ocfs2_simple_size_update(gqinode,
 244                                                oinfo->dqi_gqi_bh,
 245                                                rounded_end);
 246                 if (err < 0)
 247                         goto out;
 248                 new = 1;
 249         }
 250         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
 251         if (err) {
 252                 mlog_errno(err);
 253                 goto out;
 254         }
 255         /* Not rewriting whole block? */
 256         if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
 257             !new) {
 258                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 259                 ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
 260         } else {
 261                 bh = sb_getblk(sb, pblock);
 262                 if (!bh)
 263                         err = -ENOMEM;
 264                 ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
 265         }
 266         if (err) {
 267                 mlog_errno(err);
 268                 goto out;
 269         }
 270         lock_buffer(bh);
 271         if (new)
 272                 memset(bh->b_data, 0, sb->s_blocksize);
 273         memcpy(bh->b_data + offset, data, len);
 274         flush_dcache_page(bh->b_page);
 275         set_buffer_uptodate(bh);
 276         unlock_buffer(bh);
 277         ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
 278         err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
 279                                       ja_type);
 280         if (err < 0) {
 281                 brelse(bh);
 282                 goto out;
 283         }
 284         ocfs2_journal_dirty(handle, bh);
 285         brelse(bh);
 286 out:
 287         if (err) {
 288                 mlog_errno(err);
 289                 return err;
 290         }
 291         gqinode->i_version++;
 292         ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
 293         return len;
 294 }
 295
 296 int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 297 {
 298         int status;
 299         struct buffer_head *bh = NULL;
 300
 301         status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
 302         if (status < 0)
 303                 return status;
 304         spin_lock(&dq_data_lock);
 305         if (!oinfo->dqi_gqi_count++)
 306                 oinfo->dqi_gqi_bh = bh;
 307         else
 308                 WARN_ON(bh != oinfo->dqi_gqi_bh);
 309         spin_unlock(&dq_data_lock);
 310         if (ex) {
 311                 inode_lock(oinfo->dqi_gqinode);
 312                 down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 313         } else {
 314                 down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 315         }
 316         return 0;
 317 }
 318
 319 void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 320 {
 321         if (ex) {
 322                 up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 323                 inode_unlock(oinfo->dqi_gqinode);
 324         } else {
 325                 up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 326         }
 327         ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
 328         brelse(oinfo->dqi_gqi_bh);
 329         spin_lock(&dq_data_lock);
 330         if (!--oinfo->dqi_gqi_count)
 331                 oinfo->dqi_gqi_bh = NULL;
 332         spin_unlock(&dq_data_lock);
 333 }
 334
 335 /* Read information header from global quota file */
 336 int ocfs2_global_read_info(struct super_block *sb, int type)
 337 {
 338         struct inode *gqinode = NULL;
 339         unsigned int ino[OCFS2_MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
 340                                               GROUP_QUOTA_SYSTEM_INODE };
 341         struct ocfs2_global_disk_dqinfo dinfo;
 342         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 343         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 344         u64 pcount;
 345         int status;
 346
 347         /* Read global header */
 348         gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
 349                         OCFS2_INVALID_SLOT);
 350         if (!gqinode) {
 351                 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
 352                         type);
 353                 status = -EINVAL;
 354                 goto out_err;
 355         }
 356         oinfo->dqi_gi.dqi_sb = sb;
 357         oinfo->dqi_gi.dqi_type = type;
 358         ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
 359         oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
 360         oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
 361         oinfo->dqi_gqi_bh = NULL;
 362         oinfo->dqi_gqi_count = 0;
 363         oinfo->dqi_gqinode = gqinode;
 364         status = ocfs2_lock_global_qf(oinfo, 0);
 365         if (status < 0) {
 366                 mlog_errno(status);
 367                 goto out_err;
 368         }
 369
 370         status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
 371                                              &pcount, NULL);
 372         if (status < 0)
 373                 goto out_unlock;
 374
 375         status = ocfs2_qinfo_lock(oinfo, 0);
 376         if (status < 0)
 377                 goto out_unlock;
 378         status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
 379                                       sizeof(struct ocfs2_global_disk_dqinfo),
 380                                       OCFS2_GLOBAL_INFO_OFF);
 381         ocfs2_qinfo_unlock(oinfo, 0);
 382         ocfs2_unlock_global_qf(oinfo, 0);
 383         if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
 384                 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
 385                      status);
 386                 if (status >= 0)
 387                         status = -EIO;
 388                 mlog_errno(status);
 389                 goto out_err;
 390         }
 391         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 392         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 393         oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
 394         oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 395         oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 396         oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 397         oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
 398         oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
 399                                                 OCFS2_QBLK_RESERVED_SPACE;
 400         oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
 401         INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
 402         schedule_delayed_work(&oinfo->dqi_sync_work,
 403                               msecs_to_jiffies(oinfo->dqi_syncms));
 404
 405 out_err:
 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 quota_info *dqopt = sb_dqopt(sb);
 447         struct ocfs2_mem_dqinfo *info = dqopt->info[type].dqi_priv;
 448
 449         down_write(&dqopt->dqio_sem);
 450         err = ocfs2_qinfo_lock(info, 1);
 451         if (err < 0)
 452                 goto out_sem;
 453         err = __ocfs2_global_write_info(sb, type);
 454         ocfs2_qinfo_unlock(info, 1);
 455 out_sem:
 456         up_write(&dqopt->dqio_sem);
 457         return err;
 458 }
 459
 460 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
 461 {
 462         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 463
 464         /*
 465          * We may need to allocate tree blocks and a leaf block but not the
 466          * root block
 467          */
 468         return oinfo->dqi_gi.dqi_qtree_depth;
 469 }
 470
 471 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
 472 {
 473         /* We modify all the allocated blocks, tree root, info block and
 474          * the inode */
 475         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
 476                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
 477 }
 478
 479 /* Sync local information about quota modifications with global quota file.
 480  * Caller must have started the transaction and obtained exclusive lock for
 481  * global quota file inode */
 482 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
 483 {
 484         int err, err2;
 485         struct super_block *sb = dquot->dq_sb;
 486         int type = dquot->dq_id.type;
 487         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 488         struct ocfs2_global_disk_dqblk dqblk;
 489         s64 spacechange, inodechange;
 490         time64_t olditime, oldbtime;
 491
 492         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
 493                                    sizeof(struct ocfs2_global_disk_dqblk),
 494                                    dquot->dq_off);
 495         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
 496                 if (err >= 0) {
 497                         mlog(ML_ERROR, "Short read from global quota file "
 498                                        "(%u read)\n", err);
 499                         err = -EIO;
 500                 }
 501                 goto out;
 502         }
 503
 504         /* Update space and inode usage. Get also other information from
 505          * global quota file so that we don't overwrite any changes there.
 506          * We are */
 507         spin_lock(&dquot->dq_dqb_lock);
 508         spacechange = dquot->dq_dqb.dqb_curspace -
 509                                         OCFS2_DQUOT(dquot)->dq_origspace;
 510         inodechange = dquot->dq_dqb.dqb_curinodes -
 511                                         OCFS2_DQUOT(dquot)->dq_originodes;
 512         olditime = dquot->dq_dqb.dqb_itime;
 513         oldbtime = dquot->dq_dqb.dqb_btime;
 514         ocfs2_global_disk2memdqb(dquot, &dqblk);
 515         trace_ocfs2_sync_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 516                                dquot->dq_dqb.dqb_curspace,
 517                                (long long)spacechange,
 518                                dquot->dq_dqb.dqb_curinodes,
 519                                (long long)inodechange);
 520         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
 521                 dquot->dq_dqb.dqb_curspace += spacechange;
 522         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
 523                 dquot->dq_dqb.dqb_curinodes += inodechange;
 524         /* Set properly space grace time... */
 525         if (dquot->dq_dqb.dqb_bsoftlimit &&
 526             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
 527                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
 528                     oldbtime > 0) {
 529                         if (dquot->dq_dqb.dqb_btime > 0)
 530                                 dquot->dq_dqb.dqb_btime =
 531                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
 532                         else
 533                                 dquot->dq_dqb.dqb_btime = oldbtime;
 534                 }
 535         } else {
 536                 dquot->dq_dqb.dqb_btime = 0;
 537                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 538         }
 539         /* Set properly inode grace time... */
 540         if (dquot->dq_dqb.dqb_isoftlimit &&
 541             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
 542                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
 543                     olditime > 0) {
 544                         if (dquot->dq_dqb.dqb_itime > 0)
 545                                 dquot->dq_dqb.dqb_itime =
 546                                         min(dquot->dq_dqb.dqb_itime, olditime);
 547                         else
 548                                 dquot->dq_dqb.dqb_itime = olditime;
 549                 }
 550         } else {
 551                 dquot->dq_dqb.dqb_itime = 0;
 552                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
 553         }
 554         /* All information is properly updated, clear the flags */
 555         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
 556         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
 557         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
 558         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
 559         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
 560         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
 561         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 562         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 563         spin_unlock(&dquot->dq_dqb_lock);
 564         err = ocfs2_qinfo_lock(info, freeing);
 565         if (err < 0) {
 566                 mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
 567                                " (type=%d, id=%u)\n", dquot->dq_id.type,
 568                                (unsigned)from_kqid(&init_user_ns, dquot->dq_id));
 569                 goto out;
 570         }
 571         if (freeing)
 572                 OCFS2_DQUOT(dquot)->dq_use_count--;
 573         err = qtree_write_dquot(&info->dqi_gi, dquot);
 574         if (err < 0)
 575                 goto out_qlock;
 576         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
 577                 err = qtree_release_dquot(&info->dqi_gi, dquot);
 578                 if (info_dirty(sb_dqinfo(sb, type))) {
 579                         err2 = __ocfs2_global_write_info(sb, type);
 580                         if (!err)
 581                                 err = err2;
 582                 }
 583         }
 584 out_qlock:
 585         ocfs2_qinfo_unlock(info, freeing);
 586 out:
 587         if (err < 0)
 588                 mlog_errno(err);
 589         return err;
 590 }
 591
 592 /*
 593  *  Functions for periodic syncing of dquots with global file
 594  */
 595 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
 596 {
 597         handle_t *handle;
 598         struct super_block *sb = dquot->dq_sb;
 599         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 600         struct ocfs2_super *osb = OCFS2_SB(sb);
 601         int status = 0;
 602
 603         trace_ocfs2_sync_dquot_helper(from_kqid(&init_user_ns, dquot->dq_id),
 604                                       dquot->dq_id.type,
 605                                       type, sb->s_id);
 606         if (type != dquot->dq_id.type)
 607                 goto out;
 608         status = ocfs2_lock_global_qf(oinfo, 1);
 609         if (status < 0)
 610                 goto out;
 611
 612         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 613         if (IS_ERR(handle)) {
 614                 status = PTR_ERR(handle);
 615                 mlog_errno(status);
 616                 goto out_ilock;
 617         }
 618         down_write(&sb_dqopt(sb)->dqio_sem);
 619         status = ocfs2_sync_dquot(dquot);
 620         if (status < 0)
 621                 mlog_errno(status);
 622         /* We have to write local structure as well... */
 623         status = ocfs2_local_write_dquot(dquot);
 624         if (status < 0)
 625                 mlog_errno(status);
 626         up_write(&sb_dqopt(sb)->dqio_sem);
 627         ocfs2_commit_trans(osb, handle);
 628 out_ilock:
 629         ocfs2_unlock_global_qf(oinfo, 1);
 630 out:
 631         return status;
 632 }
 633
 634 static void qsync_work_fn(struct work_struct *work)
 635 {
 636         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
 637                                                       struct ocfs2_mem_dqinfo,
 638                                                       dqi_sync_work.work);
 639         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
 640
 641         /*
 642          * We have to be careful here not to deadlock on s_umount as umount
 643          * disabling quotas may be in progress and it waits for this work to
 644          * complete. If trylock fails, we'll do the sync next time...
 645          */
 646         if (down_read_trylock(&sb->s_umount)) {
 647                 dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
 648                 up_read(&sb->s_umount);
 649         }
 650         schedule_delayed_work(&oinfo->dqi_sync_work,
 651                               msecs_to_jiffies(oinfo->dqi_syncms));
 652 }
 653
 654 /*
 655  *  Wrappers for generic quota functions
 656  */
 657
 658 static int ocfs2_write_dquot(struct dquot *dquot)
 659 {
 660         handle_t *handle;
 661         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 662         int status = 0;
 663
 664         trace_ocfs2_write_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 665                                 dquot->dq_id.type);
 666
 667         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
 668         if (IS_ERR(handle)) {
 669                 status = PTR_ERR(handle);
 670                 mlog_errno(status);
 671                 goto out;
 672         }
 673         down_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 674         status = ocfs2_local_write_dquot(dquot);
 675         up_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 676         ocfs2_commit_trans(osb, handle);
 677 out:
 678         return status;
 679 }
 680
 681 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
 682 {
 683         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 684         /*
 685          * We modify tree, leaf block, global info, local chunk header,
 686          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
 687          * accounts for inode update
 688          */
 689         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
 690                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
 691                OCFS2_QINFO_WRITE_CREDITS +
 692                OCFS2_INODE_UPDATE_CREDITS;
 693 }
 694
 695 void ocfs2_drop_dquot_refs(struct work_struct *work)
 696 {
 697         struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
 698                                                dquot_drop_work);
 699         struct llist_node *list;
 700         struct ocfs2_dquot *odquot, *next_odquot;
 701
 702         list = llist_del_all(&osb->dquot_drop_list);
 703         llist_for_each_entry_safe(odquot, next_odquot, list, list) {
 704                 /* Drop the reference we acquired in ocfs2_dquot_release() */
 705                 dqput(&odquot->dq_dquot);
 706         }
 707 }
 708
 709 /*
 710  * Called when the last reference to dquot is dropped. If we are called from
 711  * downconvert thread, we cannot do all the handling here because grabbing
 712  * quota lock could deadlock (the node holding the quota lock could need some
 713  * other cluster lock to proceed but with blocked downconvert thread we cannot
 714  * release any lock).
 715  */
 716 static int ocfs2_release_dquot(struct dquot *dquot)
 717 {
 718         handle_t *handle;
 719         struct ocfs2_mem_dqinfo *oinfo =
 720                         sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
 721         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 722         int status = 0;
 723
 724         trace_ocfs2_release_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 725                                   dquot->dq_id.type);
 726
 727         mutex_lock(&dquot->dq_lock);
 728         /* Check whether we are not racing with some other dqget() */
 729         if (atomic_read(&dquot->dq_count) > 1)
 730                 goto out;
 731         /* Running from downconvert thread? Postpone quota processing to wq */
 732         if (current == osb->dc_task) {
 733                 /*
 734                  * Grab our own reference to dquot and queue it for delayed
 735                  * dropping.  Quota code rechecks after calling
 736                  * ->release_dquot() and won't free dquot structure.
 737                  */
 738                 dqgrab(dquot);
 739                 /* First entry on list -> queue work */
 740                 if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list))
 741                         queue_work(osb->ocfs2_wq, &osb->dquot_drop_work);
 742                 goto out;
 743         }
 744         status = ocfs2_lock_global_qf(oinfo, 1);
 745         if (status < 0)
 746                 goto out;
 747         handle = ocfs2_start_trans(osb,
 748                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_id.type));
 749         if (IS_ERR(handle)) {
 750                 status = PTR_ERR(handle);
 751                 mlog_errno(status);
 752                 goto out_ilock;
 753         }
 754
 755         status = ocfs2_global_release_dquot(dquot);
 756         if (status < 0) {
 757                 mlog_errno(status);
 758                 goto out_trans;
 759         }
 760         status = ocfs2_local_release_dquot(handle, dquot);
 761         /*
 762          * If we fail here, we cannot do much as global structure is
 763          * already released. So just complain...
 764          */
 765         if (status < 0)
 766                 mlog_errno(status);
 767         /*
 768          * Clear dq_off so that we search for the structure in quota file next
 769          * time we acquire it. The structure might be deleted and reallocated
 770          * elsewhere by another node while our dquot structure is on freelist.
 771          */
 772         dquot->dq_off = 0;
 773         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 774 out_trans:
 775         ocfs2_commit_trans(osb, handle);
 776 out_ilock:
 777         ocfs2_unlock_global_qf(oinfo, 1);
 778 out:
 779         mutex_unlock(&dquot->dq_lock);
 780         if (status)
 781                 mlog_errno(status);
 782         return status;
 783 }
 784
 785 /*
 786  * Read global dquot structure from disk or create it if it does
 787  * not exist. Also update use count of the global structure and
 788  * create structure in node-local quota file.
 789  */
 790 static int ocfs2_acquire_dquot(struct dquot *dquot)
 791 {
 792         int status = 0, err;
 793         int ex = 0;
 794         struct super_block *sb = dquot->dq_sb;
 795         struct ocfs2_super *osb = OCFS2_SB(sb);
 796         int type = dquot->dq_id.type;
 797         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 798         struct inode *gqinode = info->dqi_gqinode;
 799         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
 800         handle_t *handle;
 801
 802         trace_ocfs2_acquire_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 803                                   type);
 804         mutex_lock(&dquot->dq_lock);
 805         /*
 806          * We need an exclusive lock, because we're going to update use count
 807          * and instantiate possibly new dquot structure
 808          */
 809         status = ocfs2_lock_global_qf(info, 1);
 810         if (status < 0)
 811                 goto out;
 812         status = ocfs2_qinfo_lock(info, 0);
 813         if (status < 0)
 814                 goto out_dq;
 815         /*
 816          * We always want to read dquot structure from disk because we don't
 817          * know what happened with it while it was on freelist.
 818          */
 819         status = qtree_read_dquot(&info->dqi_gi, dquot);
 820         ocfs2_qinfo_unlock(info, 0);
 821         if (status < 0)
 822                 goto out_dq;
 823
 824         OCFS2_DQUOT(dquot)->dq_use_count++;
 825         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 826         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 827         if (!dquot->dq_off) {   /* No real quota entry? */
 828                 ex = 1;
 829                 /*
 830                  * Add blocks to quota file before we start a transaction since
 831                  * locking allocators ranks above a transaction start
 832                  */
 833                 WARN_ON(journal_current_handle());
 834                 status = ocfs2_extend_no_holes(gqinode, NULL,
 835                         i_size_read(gqinode) + (need_alloc << sb->s_blocksize_bits),
 836                         i_size_read(gqinode));
 837                 if (status < 0)
 838                         goto out_dq;
 839         }
 840
 841         handle = ocfs2_start_trans(osb,
 842                                    ocfs2_calc_global_qinit_credits(sb, type));
 843         if (IS_ERR(handle)) {
 844                 status = PTR_ERR(handle);
 845                 goto out_dq;
 846         }
 847         status = ocfs2_qinfo_lock(info, ex);
 848         if (status < 0)
 849                 goto out_trans;
 850         status = qtree_write_dquot(&info->dqi_gi, dquot);
 851         if (ex && info_dirty(sb_dqinfo(sb, type))) {
 852                 err = __ocfs2_global_write_info(sb, type);
 853                 if (!status)
 854                         status = err;
 855         }
 856         ocfs2_qinfo_unlock(info, ex);
 857 out_trans:
 858         ocfs2_commit_trans(osb, handle);
 859 out_dq:
 860         ocfs2_unlock_global_qf(info, 1);
 861         if (status < 0)
 862                 goto out;
 863
 864         status = ocfs2_create_local_dquot(dquot);
 865         if (status < 0)
 866                 goto out;
 867         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 868 out:
 869         mutex_unlock(&dquot->dq_lock);
 870         if (status)
 871                 mlog_errno(status);
 872         return status;
 873 }
 874
 875 static int ocfs2_get_next_id(struct super_block *sb, struct kqid *qid)
 876 {
 877         int type = qid->type;
 878         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 879         int status = 0;
 880
 881         trace_ocfs2_get_next_id(from_kqid(&init_user_ns, *qid), type);
 882         if (!sb_has_quota_loaded(sb, type)) {
 883                 status = -ESRCH;
 884                 goto out;
 885         }
 886         status = ocfs2_lock_global_qf(info, 0);
 887         if (status < 0)
 888                 goto out;
 889         status = ocfs2_qinfo_lock(info, 0);
 890         if (status < 0)
 891                 goto out_global;
 892         status = qtree_get_next_id(&info->dqi_gi, qid);
 893         ocfs2_qinfo_unlock(info, 0);
 894 out_global:
 895         ocfs2_unlock_global_qf(info, 0);
 896 out:
 897         /*
 898          * Avoid logging ENOENT since it just means there isn't next ID and
 899          * ESRCH which means quota isn't enabled for the filesystem.
 900          */
 901         if (status && status != -ENOENT && status != -ESRCH)
 902                 mlog_errno(status);
 903         return status;
 904 }
 905
 906 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
 907 {
 908         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
 909                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
 910                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
 911                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
 912                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
 913                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
 914         int sync = 0;
 915         int status;
 916         struct super_block *sb = dquot->dq_sb;
 917         int type = dquot->dq_id.type;
 918         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 919         handle_t *handle;
 920         struct ocfs2_super *osb = OCFS2_SB(sb);
 921
 922         trace_ocfs2_mark_dquot_dirty(from_kqid(&init_user_ns, dquot->dq_id),
 923                                      type);
 924
 925         /* In case user set some limits, sync dquot immediately to global
 926          * quota file so that information propagates quicker */
 927         spin_lock(&dquot->dq_dqb_lock);
 928         if (dquot->dq_flags & mask)
 929                 sync = 1;
 930         spin_unlock(&dquot->dq_dqb_lock);
 931         /* This is a slight hack but we can't afford getting global quota
 932          * lock if we already have a transaction started. */
 933         if (!sync || journal_current_handle()) {
 934                 status = ocfs2_write_dquot(dquot);
 935                 goto out;
 936         }
 937         status = ocfs2_lock_global_qf(oinfo, 1);
 938         if (status < 0)
 939                 goto out;
 940         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 941         if (IS_ERR(handle)) {
 942                 status = PTR_ERR(handle);
 943                 mlog_errno(status);
 944                 goto out_ilock;
 945         }
 946         down_write(&sb_dqopt(sb)->dqio_sem);
 947         status = ocfs2_sync_dquot(dquot);
 948         if (status < 0) {
 949                 mlog_errno(status);
 950                 goto out_dlock;
 951         }
 952         /* Now write updated local dquot structure */
 953         status = ocfs2_local_write_dquot(dquot);
 954 out_dlock:
 955         up_write(&sb_dqopt(sb)->dqio_sem);
 956         ocfs2_commit_trans(osb, handle);
 957 out_ilock:
 958         ocfs2_unlock_global_qf(oinfo, 1);
 959 out:
 960         if (status)
 961                 mlog_errno(status);
 962         return status;
 963 }
 964
 965 /* This should happen only after set_dqinfo(). */
 966 static int ocfs2_write_info(struct super_block *sb, int type)
 967 {
 968         handle_t *handle;
 969         int status = 0;
 970         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 971
 972         status = ocfs2_lock_global_qf(oinfo, 1);
 973         if (status < 0)
 974                 goto out;
 975         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
 976         if (IS_ERR(handle)) {
 977                 status = PTR_ERR(handle);
 978                 mlog_errno(status);
 979                 goto out_ilock;
 980         }
 981         status = dquot_commit_info(sb, type);
 982         ocfs2_commit_trans(OCFS2_SB(sb), handle);
 983 out_ilock:
 984         ocfs2_unlock_global_qf(oinfo, 1);
 985 out:
 986         if (status)
 987                 mlog_errno(status);
 988         return status;
 989 }
 990
 991 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
 992 {
 993         struct ocfs2_dquot *dquot =
 994                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
 995
 996         if (!dquot)
 997                 return NULL;
 998         return &dquot->dq_dquot;
 999 }
1000
1001 static void ocfs2_destroy_dquot(struct dquot *dquot)
1002 {
1003         kmem_cache_free(ocfs2_dquot_cachep, dquot);
1004 }
1005
1006 const struct dquot_operations ocfs2_quota_operations = {
1007         /* We never make dquot dirty so .write_dquot is never called */
1008         .acquire_dquot  = ocfs2_acquire_dquot,
1009         .release_dquot  = ocfs2_release_dquot,
1010         .mark_dirty     = ocfs2_mark_dquot_dirty,
1011         .write_info     = ocfs2_write_info,
1012         .alloc_dquot    = ocfs2_alloc_dquot,
1013         .destroy_dquot  = ocfs2_destroy_dquot,
1014         .get_next_id    = ocfs2_get_next_id,
1015 };