fs/ocfs2/quota_global.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  Implementation of operations over global quota file
   4  */
   5 #include <linux/spinlock.h>
   6 #include <linux/fs.h>
   7 #include <linux/slab.h>
   8 #include <linux/quota.h>
   9 #include <linux/quotaops.h>
  10 #include <linux/dqblk_qtree.h>
  11 #include <linux/jiffies.h>
  12 #include <linux/writeback.h>
  13 #include <linux/workqueue.h>
  14 #include <linux/llist.h>
  15 #include <linux/iversion.h>
  16
  17 #include <cluster/masklog.h>
  18
  19 #include "ocfs2_fs.h"
  20 #include "ocfs2.h"
  21 #include "alloc.h"
  22 #include "blockcheck.h"
  23 #include "inode.h"
  24 #include "journal.h"
  25 #include "file.h"
  26 #include "sysfile.h"
  27 #include "dlmglue.h"
  28 #include "uptodate.h"
  29 #include "super.h"
  30 #include "buffer_head_io.h"
  31 #include "quota.h"
  32 #include "ocfs2_trace.h"
  33
  34 /*
  35  * Locking of quotas with OCFS2 is rather complex. Here are rules that
  36  * should be obeyed by all the functions:
  37  * - any write of quota structure (either to local or global file) is protected
  38  *   by dqio_sem or dquot->dq_lock.
  39  * - any modification of global quota file holds inode cluster lock, i_rwsem,
  40  *   and ip_alloc_sem of the global quota file (achieved by
  41  *   ocfs2_lock_global_qf). It also has to hold qinfo_lock.
  42  * - an allocation of new blocks for local quota file is protected by
  43  *   its ip_alloc_sem
  44  *
  45  * A rough sketch of locking dependencies (lf = local file, gf = global file):
  46  * Normal filesystem operation:
  47  *   start_trans -> dqio_sem -> write to lf
  48  * Syncing of local and global file:
  49  *   ocfs2_lock_global_qf -> start_trans -> dqio_sem -> qinfo_lock ->
  50  *     write to gf
  51  *                                                     -> write to lf
  52  * Acquire dquot for the first time:
  53  *   dq_lock -> ocfs2_lock_global_qf -> qinfo_lock -> read from gf
  54  *                                   -> alloc space for gf
  55  *                                   -> start_trans -> qinfo_lock -> write to gf
  56  *           -> ip_alloc_sem of lf -> alloc space for lf
  57  *           -> write to lf
  58  * Release last reference to dquot:
  59  *   dq_lock -> ocfs2_lock_global_qf -> start_trans -> qinfo_lock -> write to gf
  60  *           -> write to lf
  61  * Note that all the above operations also hold the inode cluster lock of lf.
  62  * Recovery:
  63  *   inode cluster lock of recovered lf
  64  *     -> read bitmaps -> ip_alloc_sem of lf
  65  *     -> ocfs2_lock_global_qf -> start_trans -> dqio_sem -> qinfo_lock ->
  66  *        write to gf
  67  */
  68
  69 static void qsync_work_fn(struct work_struct *work);
  70
  71 static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
  72 {
  73         struct ocfs2_global_disk_dqblk *d = dp;
  74         struct mem_dqblk *m = &dquot->dq_dqb;
  75
  76         /* Update from disk only entries not set by the admin */
  77         if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
  78                 m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
  79                 m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
  80         }
  81         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
  82                 m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
  83         if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
  84                 m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
  85                 m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
  86         }
  87         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
  88                 m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
  89         if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
  90                 m->dqb_btime = le64_to_cpu(d->dqb_btime);
  91         if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
  92                 m->dqb_itime = le64_to_cpu(d->dqb_itime);
  93         OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
  94 }
  95
  96 static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
  97 {
  98         struct ocfs2_global_disk_dqblk *d = dp;
  99         struct mem_dqblk *m = &dquot->dq_dqb;
 100
 101         d->dqb_id = cpu_to_le32(from_kqid(&init_user_ns, dquot->dq_id));
 102         d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
 103         d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
 104         d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
 105         d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
 106         d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
 107         d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
 108         d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
 109         d->dqb_btime = cpu_to_le64(m->dqb_btime);
 110         d->dqb_itime = cpu_to_le64(m->dqb_itime);
 111         d->dqb_pad1 = d->dqb_pad2 = 0;
 112 }
 113
 114 static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
 115 {
 116         struct ocfs2_global_disk_dqblk *d = dp;
 117         struct ocfs2_mem_dqinfo *oinfo =
 118                         sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
 119
 120         if (qtree_entry_unused(&oinfo->dqi_gi, dp))
 121                 return 0;
 122
 123         return qid_eq(make_kqid(&init_user_ns, dquot->dq_id.type,
 124                                 le32_to_cpu(d->dqb_id)),
 125                       dquot->dq_id);
 126 }
 127
 128 const struct qtree_fmt_operations ocfs2_global_ops = {
 129         .mem2disk_dqblk = ocfs2_global_mem2diskdqb,
 130         .disk2mem_dqblk = ocfs2_global_disk2memdqb,
 131         .is_id = ocfs2_global_is_id,
 132 };
 133
 134 int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
 135 {
 136         struct ocfs2_disk_dqtrailer *dqt =
 137                 ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
 138
 139         trace_ocfs2_validate_quota_block((unsigned long long)bh->b_blocknr);
 140
 141         BUG_ON(!buffer_uptodate(bh));
 142
 143         /*
 144          * If the ecc fails, we return the error but otherwise
 145          * leave the filesystem running.  We know any error is
 146          * local to this block.
 147          */
 148         return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
 149 }
 150
 151 int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
 152                                 struct buffer_head **bhp)
 153 {
 154         int rc;
 155
 156         *bhp = NULL;
 157         rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
 158                                ocfs2_validate_quota_block);
 159         if (rc)
 160                 mlog_errno(rc);
 161         return rc;
 162 }
 163
 164 /* Read data from global quotafile - avoid pagecache and such because we cannot
 165  * afford acquiring the locks... We use quota cluster lock to serialize
 166  * operations. Caller is responsible for acquiring it. */
 167 ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
 168                          size_t len, loff_t off)
 169 {
 170         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 171         struct inode *gqinode = oinfo->dqi_gqinode;
 172         loff_t i_size = i_size_read(gqinode);
 173         int offset = off & (sb->s_blocksize - 1);
 174         sector_t blk = off >> sb->s_blocksize_bits;
 175         int err = 0;
 176         struct buffer_head *bh;
 177         size_t toread, tocopy;
 178         u64 pblock = 0, pcount = 0;
 179
 180         if (off > i_size)
 181                 return 0;
 182         if (off + len > i_size)
 183                 len = i_size - off;
 184         toread = len;
 185         while (toread > 0) {
 186                 tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
 187                 if (!pcount) {
 188                         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
 189                                                           &pcount, NULL);
 190                         if (err) {
 191                                 mlog_errno(err);
 192                                 return err;
 193                         }
 194                 } else {
 195                         pcount--;
 196                         pblock++;
 197                 }
 198                 bh = NULL;
 199                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 200                 if (err) {
 201                         mlog_errno(err);
 202                         return err;
 203                 }
 204                 memcpy(data, bh->b_data + offset, tocopy);
 205                 brelse(bh);
 206                 offset = 0;
 207                 toread -= tocopy;
 208                 data += tocopy;
 209                 blk++;
 210         }
 211         return len;
 212 }
 213
 214 /* Write to quotafile (we know the transaction is already started and has
 215  * enough credits) */
 216 ssize_t ocfs2_quota_write(struct super_block *sb, int type,
 217                           const char *data, size_t len, loff_t off)
 218 {
 219         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 220         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 221         struct inode *gqinode = oinfo->dqi_gqinode;
 222         int offset = off & (sb->s_blocksize - 1);
 223         sector_t blk = off >> sb->s_blocksize_bits;
 224         int err = 0, new = 0, ja_type;
 225         struct buffer_head *bh = NULL;
 226         handle_t *handle = journal_current_handle();
 227         u64 pblock, pcount;
 228
 229         if (!handle) {
 230                 mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
 231                      "because transaction was not started.\n",
 232                      (unsigned long long)off, (unsigned long long)len);
 233                 return -EIO;
 234         }
 235         if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
 236                 WARN_ON(1);
 237                 len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
 238         }
 239
 240         if (i_size_read(gqinode) < off + len) {
 241                 loff_t rounded_end =
 242                                 ocfs2_align_bytes_to_blocks(sb, off + len);
 243
 244                 /* Space is already allocated in ocfs2_acquire_dquot() */
 245                 err = ocfs2_simple_size_update(gqinode,
 246                                                oinfo->dqi_gqi_bh,
 247                                                rounded_end);
 248                 if (err < 0)
 249                         goto out;
 250                 new = 1;
 251         }
 252         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
 253         if (err) {
 254                 mlog_errno(err);
 255                 goto out;
 256         }
 257         /* Not rewriting whole block? */
 258         if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
 259             !new) {
 260                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
 261                 ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
 262         } else {
 263                 bh = sb_getblk(sb, pblock);
 264                 if (!bh)
 265                         err = -ENOMEM;
 266                 ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
 267         }
 268         if (err) {
 269                 mlog_errno(err);
 270                 goto out;
 271         }
 272         lock_buffer(bh);
 273         if (new)
 274                 memset(bh->b_data, 0, sb->s_blocksize);
 275         memcpy(bh->b_data + offset, data, len);
 276         flush_dcache_page(bh->b_page);
 277         set_buffer_uptodate(bh);
 278         unlock_buffer(bh);
 279         ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
 280         err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
 281                                       ja_type);
 282         if (err < 0) {
 283                 brelse(bh);
 284                 goto out;
 285         }
 286         ocfs2_journal_dirty(handle, bh);
 287         brelse(bh);
 288 out:
 289         if (err) {
 290                 mlog_errno(err);
 291                 return err;
 292         }
 293         inode_inc_iversion(gqinode);
 294         ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
 295         return len;
 296 }
 297
 298 int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 299 {
 300         int status;
 301         struct buffer_head *bh = NULL;
 302
 303         status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
 304         if (status < 0)
 305                 return status;
 306         spin_lock(&dq_data_lock);
 307         if (!oinfo->dqi_gqi_count++)
 308                 oinfo->dqi_gqi_bh = bh;
 309         else
 310                 WARN_ON(bh != oinfo->dqi_gqi_bh);
 311         spin_unlock(&dq_data_lock);
 312         if (ex) {
 313                 inode_lock(oinfo->dqi_gqinode);
 314                 down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 315         } else {
 316                 down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 317         }
 318         return 0;
 319 }
 320
 321 void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
 322 {
 323         if (ex) {
 324                 up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 325                 inode_unlock(oinfo->dqi_gqinode);
 326         } else {
 327                 up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
 328         }
 329         ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
 330         brelse(oinfo->dqi_gqi_bh);
 331         spin_lock(&dq_data_lock);
 332         if (!--oinfo->dqi_gqi_count)
 333                 oinfo->dqi_gqi_bh = NULL;
 334         spin_unlock(&dq_data_lock);
 335 }
 336
 337 /* Read information header from global quota file */
 338 int ocfs2_global_read_info(struct super_block *sb, int type)
 339 {
 340         unsigned int ino[OCFS2_MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
 341                                               GROUP_QUOTA_SYSTEM_INODE };
 342         struct ocfs2_global_disk_dqinfo dinfo;
 343         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 344         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 345         u64 pcount;
 346         int status;
 347
 348         oinfo->dqi_gi.dqi_sb = sb;
 349         oinfo->dqi_gi.dqi_type = type;
 350         ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
 351         oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
 352         oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
 353         oinfo->dqi_gqi_bh = NULL;
 354         oinfo->dqi_gqi_count = 0;
 355
 356         /* Read global header */
 357         oinfo->dqi_gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
 358                         OCFS2_INVALID_SLOT);
 359         if (!oinfo->dqi_gqinode) {
 360                 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
 361                         type);
 362                 status = -EINVAL;
 363                 goto out_err;
 364         }
 365
 366         status = ocfs2_lock_global_qf(oinfo, 0);
 367         if (status < 0) {
 368                 mlog_errno(status);
 369                 goto out_err;
 370         }
 371
 372         status = ocfs2_extent_map_get_blocks(oinfo->dqi_gqinode, 0, &oinfo->dqi_giblk,
 373                                              &pcount, NULL);
 374         if (status < 0)
 375                 goto out_unlock;
 376
 377         status = ocfs2_qinfo_lock(oinfo, 0);
 378         if (status < 0)
 379                 goto out_unlock;
 380         status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
 381                                       sizeof(struct ocfs2_global_disk_dqinfo),
 382                                       OCFS2_GLOBAL_INFO_OFF);
 383         ocfs2_qinfo_unlock(oinfo, 0);
 384         ocfs2_unlock_global_qf(oinfo, 0);
 385         if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
 386                 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
 387                      status);
 388                 if (status >= 0)
 389                         status = -EIO;
 390                 mlog_errno(status);
 391                 goto out_err;
 392         }
 393         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 394         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 395         oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
 396         oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 397         oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 398         oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 399         oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
 400         oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
 401                                                 OCFS2_QBLK_RESERVED_SPACE;
 402         oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
 403         INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
 404         schedule_delayed_work(&oinfo->dqi_sync_work,
 405                               msecs_to_jiffies(oinfo->dqi_syncms));
 406
 407 out_err:
 408         return status;
 409 out_unlock:
 410         ocfs2_unlock_global_qf(oinfo, 0);
 411         mlog_errno(status);
 412         goto out_err;
 413 }
 414
 415 /* Write information to global quota file. Expects exclusive lock on quota
 416  * file inode and quota info */
 417 static int __ocfs2_global_write_info(struct super_block *sb, int type)
 418 {
 419         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 420         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 421         struct ocfs2_global_disk_dqinfo dinfo;
 422         ssize_t size;
 423
 424         spin_lock(&dq_data_lock);
 425         info->dqi_flags &= ~DQF_INFO_DIRTY;
 426         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 427         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 428         spin_unlock(&dq_data_lock);
 429         dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
 430         dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
 431         dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
 432         dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
 433         size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
 434                                      sizeof(struct ocfs2_global_disk_dqinfo),
 435                                      OCFS2_GLOBAL_INFO_OFF);
 436         if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
 437                 mlog(ML_ERROR, "Cannot write global quota info structure\n");
 438                 if (size >= 0)
 439                         size = -EIO;
 440                 return size;
 441         }
 442         return 0;
 443 }
 444
 445 int ocfs2_global_write_info(struct super_block *sb, int type)
 446 {
 447         int err;
 448         struct quota_info *dqopt = sb_dqopt(sb);
 449         struct ocfs2_mem_dqinfo *info = dqopt->info[type].dqi_priv;
 450         unsigned int memalloc;
 451
 452         down_write(&dqopt->dqio_sem);
 453         memalloc = memalloc_nofs_save();
 454         err = ocfs2_qinfo_lock(info, 1);
 455         if (err < 0)
 456                 goto out_sem;
 457         err = __ocfs2_global_write_info(sb, type);
 458         ocfs2_qinfo_unlock(info, 1);
 459 out_sem:
 460         memalloc_nofs_restore(memalloc);
 461         up_write(&dqopt->dqio_sem);
 462         return err;
 463 }
 464
 465 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
 466 {
 467         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 468
 469         /*
 470          * We may need to allocate tree blocks and a leaf block but not the
 471          * root block
 472          */
 473         return oinfo->dqi_gi.dqi_qtree_depth;
 474 }
 475
 476 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
 477 {
 478         /* We modify all the allocated blocks, tree root, info block and
 479          * the inode */
 480         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
 481                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
 482 }
 483
 484 /* Sync local information about quota modifications with global quota file.
 485  * Caller must have started the transaction and obtained exclusive lock for
 486  * global quota file inode */
 487 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
 488 {
 489         int err, err2;
 490         struct super_block *sb = dquot->dq_sb;
 491         int type = dquot->dq_id.type;
 492         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 493         struct ocfs2_global_disk_dqblk dqblk;
 494         s64 spacechange, inodechange;
 495         time64_t olditime, oldbtime;
 496
 497         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
 498                                    sizeof(struct ocfs2_global_disk_dqblk),
 499                                    dquot->dq_off);
 500         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
 501                 if (err >= 0) {
 502                         mlog(ML_ERROR, "Short read from global quota file "
 503                                        "(%u read)\n", err);
 504                         err = -EIO;
 505                 }
 506                 goto out;
 507         }
 508
 509         /* Update space and inode usage. Get also other information from
 510          * global quota file so that we don't overwrite any changes there.
 511          * We are */
 512         spin_lock(&dquot->dq_dqb_lock);
 513         spacechange = dquot->dq_dqb.dqb_curspace -
 514                                         OCFS2_DQUOT(dquot)->dq_origspace;
 515         inodechange = dquot->dq_dqb.dqb_curinodes -
 516                                         OCFS2_DQUOT(dquot)->dq_originodes;
 517         olditime = dquot->dq_dqb.dqb_itime;
 518         oldbtime = dquot->dq_dqb.dqb_btime;
 519         ocfs2_global_disk2memdqb(dquot, &dqblk);
 520         trace_ocfs2_sync_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 521                                dquot->dq_dqb.dqb_curspace,
 522                                (long long)spacechange,
 523                                dquot->dq_dqb.dqb_curinodes,
 524                                (long long)inodechange);
 525         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
 526                 dquot->dq_dqb.dqb_curspace += spacechange;
 527         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
 528                 dquot->dq_dqb.dqb_curinodes += inodechange;
 529         /* Set properly space grace time... */
 530         if (dquot->dq_dqb.dqb_bsoftlimit &&
 531             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
 532                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
 533                     oldbtime > 0) {
 534                         if (dquot->dq_dqb.dqb_btime > 0)
 535                                 dquot->dq_dqb.dqb_btime =
 536                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
 537                         else
 538                                 dquot->dq_dqb.dqb_btime = oldbtime;
 539                 }
 540         } else {
 541                 dquot->dq_dqb.dqb_btime = 0;
 542                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 543         }
 544         /* Set properly inode grace time... */
 545         if (dquot->dq_dqb.dqb_isoftlimit &&
 546             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
 547                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
 548                     olditime > 0) {
 549                         if (dquot->dq_dqb.dqb_itime > 0)
 550                                 dquot->dq_dqb.dqb_itime =
 551                                         min(dquot->dq_dqb.dqb_itime, olditime);
 552                         else
 553                                 dquot->dq_dqb.dqb_itime = olditime;
 554                 }
 555         } else {
 556                 dquot->dq_dqb.dqb_itime = 0;
 557                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
 558         }
 559         /* All information is properly updated, clear the flags */
 560         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
 561         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
 562         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
 563         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
 564         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
 565         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
 566         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 567         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 568         spin_unlock(&dquot->dq_dqb_lock);
 569         err = ocfs2_qinfo_lock(info, freeing);
 570         if (err < 0) {
 571                 mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
 572                                " (type=%d, id=%u)\n", dquot->dq_id.type,
 573                                (unsigned)from_kqid(&init_user_ns, dquot->dq_id));
 574                 goto out;
 575         }
 576         if (freeing)
 577                 OCFS2_DQUOT(dquot)->dq_use_count--;
 578         err = qtree_write_dquot(&info->dqi_gi, dquot);
 579         if (err < 0)
 580                 goto out_qlock;
 581         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
 582                 err = qtree_release_dquot(&info->dqi_gi, dquot);
 583                 if (info_dirty(sb_dqinfo(sb, type))) {
 584                         err2 = __ocfs2_global_write_info(sb, type);
 585                         if (!err)
 586                                 err = err2;
 587                 }
 588         }
 589 out_qlock:
 590         ocfs2_qinfo_unlock(info, freeing);
 591 out:
 592         if (err < 0)
 593                 mlog_errno(err);
 594         return err;
 595 }
 596
 597 /*
 598  *  Functions for periodic syncing of dquots with global file
 599  */
 600 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
 601 {
 602         handle_t *handle;
 603         struct super_block *sb = dquot->dq_sb;
 604         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 605         struct ocfs2_super *osb = OCFS2_SB(sb);
 606         int status = 0;
 607         unsigned int memalloc;
 608
 609         trace_ocfs2_sync_dquot_helper(from_kqid(&init_user_ns, dquot->dq_id),
 610                                       dquot->dq_id.type,
 611                                       type, sb->s_id);
 612         if (type != dquot->dq_id.type)
 613                 goto out;
 614         status = ocfs2_lock_global_qf(oinfo, 1);
 615         if (status < 0)
 616                 goto out;
 617
 618         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 619         if (IS_ERR(handle)) {
 620                 status = PTR_ERR(handle);
 621                 mlog_errno(status);
 622                 goto out_ilock;
 623         }
 624         down_write(&sb_dqopt(sb)->dqio_sem);
 625         memalloc = memalloc_nofs_save();
 626         status = ocfs2_sync_dquot(dquot);
 627         if (status < 0)
 628                 mlog_errno(status);
 629         /* We have to write local structure as well... */
 630         status = ocfs2_local_write_dquot(dquot);
 631         if (status < 0)
 632                 mlog_errno(status);
 633         memalloc_nofs_restore(memalloc);
 634         up_write(&sb_dqopt(sb)->dqio_sem);
 635         ocfs2_commit_trans(osb, handle);
 636 out_ilock:
 637         ocfs2_unlock_global_qf(oinfo, 1);
 638 out:
 639         return status;
 640 }
 641
 642 static void qsync_work_fn(struct work_struct *work)
 643 {
 644         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
 645                                                       struct ocfs2_mem_dqinfo,
 646                                                       dqi_sync_work.work);
 647         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
 648
 649         /*
 650          * We have to be careful here not to deadlock on s_umount as umount
 651          * disabling quotas may be in progress and it waits for this work to
 652          * complete. If trylock fails, we'll do the sync next time...
 653          */
 654         if (down_read_trylock(&sb->s_umount)) {
 655                 dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
 656                 up_read(&sb->s_umount);
 657         }
 658         schedule_delayed_work(&oinfo->dqi_sync_work,
 659                               msecs_to_jiffies(oinfo->dqi_syncms));
 660 }
 661
 662 /*
 663  *  Wrappers for generic quota functions
 664  */
 665
 666 static int ocfs2_write_dquot(struct dquot *dquot)
 667 {
 668         handle_t *handle;
 669         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 670         int status = 0;
 671         unsigned int memalloc;
 672
 673         trace_ocfs2_write_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 674                                 dquot->dq_id.type);
 675
 676         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
 677         if (IS_ERR(handle)) {
 678                 status = PTR_ERR(handle);
 679                 mlog_errno(status);
 680                 goto out;
 681         }
 682         down_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 683         memalloc = memalloc_nofs_save();
 684         status = ocfs2_local_write_dquot(dquot);
 685         memalloc_nofs_restore(memalloc);
 686         up_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 687         ocfs2_commit_trans(osb, handle);
 688 out:
 689         return status;
 690 }
 691
 692 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
 693 {
 694         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 695         /*
 696          * We modify tree, leaf block, global info, local chunk header,
 697          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
 698          * accounts for inode update
 699          */
 700         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
 701                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
 702                OCFS2_QINFO_WRITE_CREDITS +
 703                OCFS2_INODE_UPDATE_CREDITS;
 704 }
 705
 706 void ocfs2_drop_dquot_refs(struct work_struct *work)
 707 {
 708         struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
 709                                                dquot_drop_work);
 710         struct llist_node *list;
 711         struct ocfs2_dquot *odquot, *next_odquot;
 712
 713         list = llist_del_all(&osb->dquot_drop_list);
 714         llist_for_each_entry_safe(odquot, next_odquot, list, list) {
 715                 /* Drop the reference we acquired in ocfs2_dquot_release() */
 716                 dqput(&odquot->dq_dquot);
 717         }
 718 }
 719
 720 /*
 721  * Called when the last reference to dquot is dropped. If we are called from
 722  * downconvert thread, we cannot do all the handling here because grabbing
 723  * quota lock could deadlock (the node holding the quota lock could need some
 724  * other cluster lock to proceed but with blocked downconvert thread we cannot
 725  * release any lock).
 726  */
 727 static int ocfs2_release_dquot(struct dquot *dquot)
 728 {
 729         handle_t *handle;
 730         struct ocfs2_mem_dqinfo *oinfo =
 731                         sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
 732         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 733         int status = 0;
 734
 735         trace_ocfs2_release_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 736                                   dquot->dq_id.type);
 737
 738         mutex_lock(&dquot->dq_lock);
 739         /* Check whether we are not racing with some other dqget() */
 740         if (dquot_is_busy(dquot))
 741                 goto out;
 742         /* Running from downconvert thread? Postpone quota processing to wq */
 743         if (current == osb->dc_task) {
 744                 /*
 745                  * Grab our own reference to dquot and queue it for delayed
 746                  * dropping.  Quota code rechecks after calling
 747                  * ->release_dquot() and won't free dquot structure.
 748                  */
 749                 dqgrab(dquot);
 750                 /* First entry on list -> queue work */
 751                 if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list))
 752                         queue_work(osb->ocfs2_wq, &osb->dquot_drop_work);
 753                 goto out;
 754         }
 755         status = ocfs2_lock_global_qf(oinfo, 1);
 756         if (status < 0)
 757                 goto out;
 758         handle = ocfs2_start_trans(osb,
 759                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_id.type));
 760         if (IS_ERR(handle)) {
 761                 status = PTR_ERR(handle);
 762                 mlog_errno(status);
 763                 goto out_ilock;
 764         }
 765
 766         status = ocfs2_global_release_dquot(dquot);
 767         if (status < 0) {
 768                 mlog_errno(status);
 769                 goto out_trans;
 770         }
 771         status = ocfs2_local_release_dquot(handle, dquot);
 772         /*
 773          * If we fail here, we cannot do much as global structure is
 774          * already released. So just complain...
 775          */
 776         if (status < 0)
 777                 mlog_errno(status);
 778         /*
 779          * Clear dq_off so that we search for the structure in quota file next
 780          * time we acquire it. The structure might be deleted and reallocated
 781          * elsewhere by another node while our dquot structure is on freelist.
 782          */
 783         dquot->dq_off = 0;
 784         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 785 out_trans:
 786         ocfs2_commit_trans(osb, handle);
 787 out_ilock:
 788         ocfs2_unlock_global_qf(oinfo, 1);
 789 out:
 790         mutex_unlock(&dquot->dq_lock);
 791         if (status)
 792                 mlog_errno(status);
 793         return status;
 794 }
 795
 796 /*
 797  * Read global dquot structure from disk or create it if it does
 798  * not exist. Also update use count of the global structure and
 799  * create structure in node-local quota file.
 800  */
 801 static int ocfs2_acquire_dquot(struct dquot *dquot)
 802 {
 803         int status = 0, err;
 804         int ex = 0;
 805         struct super_block *sb = dquot->dq_sb;
 806         struct ocfs2_super *osb = OCFS2_SB(sb);
 807         int type = dquot->dq_id.type;
 808         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 809         struct inode *gqinode = info->dqi_gqinode;
 810         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
 811         handle_t *handle;
 812
 813         trace_ocfs2_acquire_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 814                                   type);
 815         mutex_lock(&dquot->dq_lock);
 816         /*
 817          * We need an exclusive lock, because we're going to update use count
 818          * and instantiate possibly new dquot structure
 819          */
 820         status = ocfs2_lock_global_qf(info, 1);
 821         if (status < 0)
 822                 goto out;
 823         status = ocfs2_qinfo_lock(info, 0);
 824         if (status < 0)
 825                 goto out_dq;
 826         /*
 827          * We always want to read dquot structure from disk because we don't
 828          * know what happened with it while it was on freelist.
 829          */
 830         status = qtree_read_dquot(&info->dqi_gi, dquot);
 831         ocfs2_qinfo_unlock(info, 0);
 832         if (status < 0)
 833                 goto out_dq;
 834
 835         OCFS2_DQUOT(dquot)->dq_use_count++;
 836         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 837         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 838         if (!dquot->dq_off) {   /* No real quota entry? */
 839                 ex = 1;
 840                 /*
 841                  * Add blocks to quota file before we start a transaction since
 842                  * locking allocators ranks above a transaction start
 843                  */
 844                 WARN_ON(journal_current_handle());
 845                 status = ocfs2_extend_no_holes(gqinode, NULL,
 846                         i_size_read(gqinode) + (need_alloc << sb->s_blocksize_bits),
 847                         i_size_read(gqinode));
 848                 if (status < 0)
 849                         goto out_dq;
 850         }
 851
 852         handle = ocfs2_start_trans(osb,
 853                                    ocfs2_calc_global_qinit_credits(sb, type));
 854         if (IS_ERR(handle)) {
 855                 status = PTR_ERR(handle);
 856                 goto out_dq;
 857         }
 858         status = ocfs2_qinfo_lock(info, ex);
 859         if (status < 0)
 860                 goto out_trans;
 861         status = qtree_write_dquot(&info->dqi_gi, dquot);
 862         if (ex && info_dirty(sb_dqinfo(sb, type))) {
 863                 err = __ocfs2_global_write_info(sb, type);
 864                 if (!status)
 865                         status = err;
 866         }
 867         ocfs2_qinfo_unlock(info, ex);
 868 out_trans:
 869         ocfs2_commit_trans(osb, handle);
 870 out_dq:
 871         ocfs2_unlock_global_qf(info, 1);
 872         if (status < 0)
 873                 goto out;
 874
 875         status = ocfs2_create_local_dquot(dquot);
 876         if (status < 0)
 877                 goto out;
 878         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 879 out:
 880         mutex_unlock(&dquot->dq_lock);
 881         if (status)
 882                 mlog_errno(status);
 883         return status;
 884 }
 885
 886 static int ocfs2_get_next_id(struct super_block *sb, struct kqid *qid)
 887 {
 888         int type = qid->type;
 889         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 890         int status = 0;
 891
 892         trace_ocfs2_get_next_id(from_kqid(&init_user_ns, *qid), type);
 893         if (!sb_has_quota_loaded(sb, type)) {
 894                 status = -ESRCH;
 895                 goto out;
 896         }
 897         status = ocfs2_lock_global_qf(info, 0);
 898         if (status < 0)
 899                 goto out;
 900         status = ocfs2_qinfo_lock(info, 0);
 901         if (status < 0)
 902                 goto out_global;
 903         status = qtree_get_next_id(&info->dqi_gi, qid);
 904         ocfs2_qinfo_unlock(info, 0);
 905 out_global:
 906         ocfs2_unlock_global_qf(info, 0);
 907 out:
 908         /*
 909          * Avoid logging ENOENT since it just means there isn't next ID and
 910          * ESRCH which means quota isn't enabled for the filesystem.
 911          */
 912         if (status && status != -ENOENT && status != -ESRCH)
 913                 mlog_errno(status);
 914         return status;
 915 }
 916
 917 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
 918 {
 919         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
 920                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
 921                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
 922                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
 923                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
 924                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
 925         int sync = 0;
 926         int status;
 927         struct super_block *sb = dquot->dq_sb;
 928         int type = dquot->dq_id.type;
 929         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 930         handle_t *handle;
 931         struct ocfs2_super *osb = OCFS2_SB(sb);
 932         unsigned int memalloc;
 933
 934         trace_ocfs2_mark_dquot_dirty(from_kqid(&init_user_ns, dquot->dq_id),
 935                                      type);
 936
 937         /* In case user set some limits, sync dquot immediately to global
 938          * quota file so that information propagates quicker */
 939         spin_lock(&dquot->dq_dqb_lock);
 940         if (dquot->dq_flags & mask)
 941                 sync = 1;
 942         spin_unlock(&dquot->dq_dqb_lock);
 943         /* This is a slight hack but we can't afford getting global quota
 944          * lock if we already have a transaction started. */
 945         if (!sync || journal_current_handle()) {
 946                 status = ocfs2_write_dquot(dquot);
 947                 goto out;
 948         }
 949         status = ocfs2_lock_global_qf(oinfo, 1);
 950         if (status < 0)
 951                 goto out;
 952         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 953         if (IS_ERR(handle)) {
 954                 status = PTR_ERR(handle);
 955                 mlog_errno(status);
 956                 goto out_ilock;
 957         }
 958         down_write(&sb_dqopt(sb)->dqio_sem);
 959         memalloc = memalloc_nofs_save();
 960         status = ocfs2_sync_dquot(dquot);
 961         if (status < 0) {
 962                 mlog_errno(status);
 963                 goto out_dlock;
 964         }
 965         /* Now write updated local dquot structure */
 966         status = ocfs2_local_write_dquot(dquot);
 967 out_dlock:
 968         memalloc_nofs_restore(memalloc);
 969         up_write(&sb_dqopt(sb)->dqio_sem);
 970         ocfs2_commit_trans(osb, handle);
 971 out_ilock:
 972         ocfs2_unlock_global_qf(oinfo, 1);
 973 out:
 974         if (status)
 975                 mlog_errno(status);
 976         return status;
 977 }
 978
 979 /* This should happen only after set_dqinfo(). */
 980 static int ocfs2_write_info(struct super_block *sb, int type)
 981 {
 982         handle_t *handle;
 983         int status = 0;
 984         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 985
 986         status = ocfs2_lock_global_qf(oinfo, 1);
 987         if (status < 0)
 988                 goto out;
 989         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
 990         if (IS_ERR(handle)) {
 991                 status = PTR_ERR(handle);
 992                 mlog_errno(status);
 993                 goto out_ilock;
 994         }
 995         status = dquot_commit_info(sb, type);
 996         ocfs2_commit_trans(OCFS2_SB(sb), handle);
 997 out_ilock:
 998         ocfs2_unlock_global_qf(oinfo, 1);
 999 out:
1000         if (status)
1001                 mlog_errno(status);
1002         return status;
1003 }
1004
1005 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
1006 {
1007         struct ocfs2_dquot *dquot =
1008                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
1009
1010         if (!dquot)
1011                 return NULL;
1012         return &dquot->dq_dquot;
1013 }
1014
1015 static void ocfs2_destroy_dquot(struct dquot *dquot)
1016 {
1017         kmem_cache_free(ocfs2_dquot_cachep, dquot);
1018 }
1019
1020 const struct dquot_operations ocfs2_quota_operations = {
1021         /* We never make dquot dirty so .write_dquot is never called */
1022         .acquire_dquot  = ocfs2_acquire_dquot,
1023         .release_dquot  = ocfs2_release_dquot,
1024         .mark_dirty     = ocfs2_mark_dquot_dirty,
1025         .write_info     = ocfs2_write_info,
1026         .alloc_dquot    = ocfs2_alloc_dquot,
1027         .destroy_dquot  = ocfs2_destroy_dquot,
1028         .get_next_id    = ocfs2_get_next_id,
1029 };