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_mutex,
  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         struct inode *gqinode = NULL;
 341         unsigned int ino[OCFS2_MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
 342                                               GROUP_QUOTA_SYSTEM_INODE };
 343         struct ocfs2_global_disk_dqinfo dinfo;
 344         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 345         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 346         u64 pcount;
 347         int status;
 348
 349         /* Read global header */
 350         gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
 351                         OCFS2_INVALID_SLOT);
 352         if (!gqinode) {
 353                 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
 354                         type);
 355                 status = -EINVAL;
 356                 goto out_err;
 357         }
 358         oinfo->dqi_gi.dqi_sb = sb;
 359         oinfo->dqi_gi.dqi_type = type;
 360         oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
 361         oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
 362         oinfo->dqi_gqi_bh = NULL;
 363         oinfo->dqi_gqi_count = 0;
 364         oinfo->dqi_gqinode = gqinode;
 365         status = ocfs2_lock_global_qf(oinfo, 0);
 366         if (status < 0) {
 367                 mlog_errno(status);
 368                 goto out_err;
 369         }
 370
 371         status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
 372                                              &pcount, NULL);
 373         if (status < 0)
 374                 goto out_unlock;
 375
 376         status = ocfs2_qinfo_lock(oinfo, 0);
 377         if (status < 0)
 378                 goto out_unlock;
 379         status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
 380                                       sizeof(struct ocfs2_global_disk_dqinfo),
 381                                       OCFS2_GLOBAL_INFO_OFF);
 382         ocfs2_qinfo_unlock(oinfo, 0);
 383         ocfs2_unlock_global_qf(oinfo, 0);
 384         if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
 385                 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
 386                      status);
 387                 if (status >= 0)
 388                         status = -EIO;
 389                 mlog_errno(status);
 390                 goto out_err;
 391         }
 392         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 393         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 394         oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
 395         oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 396         oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 397         oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 398         oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
 399         oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
 400                                                 OCFS2_QBLK_RESERVED_SPACE;
 401         oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
 402         INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
 403         schedule_delayed_work(&oinfo->dqi_sync_work,
 404                               msecs_to_jiffies(oinfo->dqi_syncms));
 405
 406 out_err:
 407         return status;
 408 out_unlock:
 409         ocfs2_unlock_global_qf(oinfo, 0);
 410         mlog_errno(status);
 411         goto out_err;
 412 }
 413
 414 /* Write information to global quota file. Expects exlusive lock on quota
 415  * file inode and quota info */
 416 static int __ocfs2_global_write_info(struct super_block *sb, int type)
 417 {
 418         struct mem_dqinfo *info = sb_dqinfo(sb, type);
 419         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 420         struct ocfs2_global_disk_dqinfo dinfo;
 421         ssize_t size;
 422
 423         spin_lock(&dq_data_lock);
 424         info->dqi_flags &= ~DQF_INFO_DIRTY;
 425         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 426         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 427         spin_unlock(&dq_data_lock);
 428         dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
 429         dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
 430         dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
 431         dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
 432         size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
 433                                      sizeof(struct ocfs2_global_disk_dqinfo),
 434                                      OCFS2_GLOBAL_INFO_OFF);
 435         if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
 436                 mlog(ML_ERROR, "Cannot write global quota info structure\n");
 437                 if (size >= 0)
 438                         size = -EIO;
 439                 return size;
 440         }
 441         return 0;
 442 }
 443
 444 int ocfs2_global_write_info(struct super_block *sb, int type)
 445 {
 446         int err;
 447         struct quota_info *dqopt = sb_dqopt(sb);
 448         struct ocfs2_mem_dqinfo *info = dqopt->info[type].dqi_priv;
 449
 450         down_write(&dqopt->dqio_sem);
 451         err = ocfs2_qinfo_lock(info, 1);
 452         if (err < 0)
 453                 goto out_sem;
 454         err = __ocfs2_global_write_info(sb, type);
 455         ocfs2_qinfo_unlock(info, 1);
 456 out_sem:
 457         up_write(&dqopt->dqio_sem);
 458         return err;
 459 }
 460
 461 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
 462 {
 463         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 464
 465         /*
 466          * We may need to allocate tree blocks and a leaf block but not the
 467          * root block
 468          */
 469         return oinfo->dqi_gi.dqi_qtree_depth;
 470 }
 471
 472 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
 473 {
 474         /* We modify all the allocated blocks, tree root, info block and
 475          * the inode */
 476         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
 477                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
 478 }
 479
 480 /* Sync local information about quota modifications with global quota file.
 481  * Caller must have started the transaction and obtained exclusive lock for
 482  * global quota file inode */
 483 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
 484 {
 485         int err, err2;
 486         struct super_block *sb = dquot->dq_sb;
 487         int type = dquot->dq_id.type;
 488         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 489         struct ocfs2_global_disk_dqblk dqblk;
 490         s64 spacechange, inodechange;
 491         time64_t olditime, oldbtime;
 492
 493         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
 494                                    sizeof(struct ocfs2_global_disk_dqblk),
 495                                    dquot->dq_off);
 496         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
 497                 if (err >= 0) {
 498                         mlog(ML_ERROR, "Short read from global quota file "
 499                                        "(%u read)\n", err);
 500                         err = -EIO;
 501                 }
 502                 goto out;
 503         }
 504
 505         /* Update space and inode usage. Get also other information from
 506          * global quota file so that we don't overwrite any changes there.
 507          * We are */
 508         spin_lock(&dquot->dq_dqb_lock);
 509         spacechange = dquot->dq_dqb.dqb_curspace -
 510                                         OCFS2_DQUOT(dquot)->dq_origspace;
 511         inodechange = dquot->dq_dqb.dqb_curinodes -
 512                                         OCFS2_DQUOT(dquot)->dq_originodes;
 513         olditime = dquot->dq_dqb.dqb_itime;
 514         oldbtime = dquot->dq_dqb.dqb_btime;
 515         ocfs2_global_disk2memdqb(dquot, &dqblk);
 516         trace_ocfs2_sync_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 517                                dquot->dq_dqb.dqb_curspace,
 518                                (long long)spacechange,
 519                                dquot->dq_dqb.dqb_curinodes,
 520                                (long long)inodechange);
 521         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
 522                 dquot->dq_dqb.dqb_curspace += spacechange;
 523         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
 524                 dquot->dq_dqb.dqb_curinodes += inodechange;
 525         /* Set properly space grace time... */
 526         if (dquot->dq_dqb.dqb_bsoftlimit &&
 527             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
 528                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
 529                     oldbtime > 0) {
 530                         if (dquot->dq_dqb.dqb_btime > 0)
 531                                 dquot->dq_dqb.dqb_btime =
 532                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
 533                         else
 534                                 dquot->dq_dqb.dqb_btime = oldbtime;
 535                 }
 536         } else {
 537                 dquot->dq_dqb.dqb_btime = 0;
 538                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 539         }
 540         /* Set properly inode grace time... */
 541         if (dquot->dq_dqb.dqb_isoftlimit &&
 542             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
 543                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
 544                     olditime > 0) {
 545                         if (dquot->dq_dqb.dqb_itime > 0)
 546                                 dquot->dq_dqb.dqb_itime =
 547                                         min(dquot->dq_dqb.dqb_itime, olditime);
 548                         else
 549                                 dquot->dq_dqb.dqb_itime = olditime;
 550                 }
 551         } else {
 552                 dquot->dq_dqb.dqb_itime = 0;
 553                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
 554         }
 555         /* All information is properly updated, clear the flags */
 556         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
 557         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
 558         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
 559         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
 560         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
 561         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
 562         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 563         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 564         spin_unlock(&dquot->dq_dqb_lock);
 565         err = ocfs2_qinfo_lock(info, freeing);
 566         if (err < 0) {
 567                 mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
 568                                " (type=%d, id=%u)\n", dquot->dq_id.type,
 569                                (unsigned)from_kqid(&init_user_ns, dquot->dq_id));
 570                 goto out;
 571         }
 572         if (freeing)
 573                 OCFS2_DQUOT(dquot)->dq_use_count--;
 574         err = qtree_write_dquot(&info->dqi_gi, dquot);
 575         if (err < 0)
 576                 goto out_qlock;
 577         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
 578                 err = qtree_release_dquot(&info->dqi_gi, dquot);
 579                 if (info_dirty(sb_dqinfo(sb, type))) {
 580                         err2 = __ocfs2_global_write_info(sb, type);
 581                         if (!err)
 582                                 err = err2;
 583                 }
 584         }
 585 out_qlock:
 586         ocfs2_qinfo_unlock(info, freeing);
 587 out:
 588         if (err < 0)
 589                 mlog_errno(err);
 590         return err;
 591 }
 592
 593 /*
 594  *  Functions for periodic syncing of dquots with global file
 595  */
 596 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
 597 {
 598         handle_t *handle;
 599         struct super_block *sb = dquot->dq_sb;
 600         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 601         struct ocfs2_super *osb = OCFS2_SB(sb);
 602         int status = 0;
 603
 604         trace_ocfs2_sync_dquot_helper(from_kqid(&init_user_ns, dquot->dq_id),
 605                                       dquot->dq_id.type,
 606                                       type, sb->s_id);
 607         if (type != dquot->dq_id.type)
 608                 goto out;
 609         status = ocfs2_lock_global_qf(oinfo, 1);
 610         if (status < 0)
 611                 goto out;
 612
 613         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 614         if (IS_ERR(handle)) {
 615                 status = PTR_ERR(handle);
 616                 mlog_errno(status);
 617                 goto out_ilock;
 618         }
 619         down_write(&sb_dqopt(sb)->dqio_sem);
 620         status = ocfs2_sync_dquot(dquot);
 621         if (status < 0)
 622                 mlog_errno(status);
 623         /* We have to write local structure as well... */
 624         status = ocfs2_local_write_dquot(dquot);
 625         if (status < 0)
 626                 mlog_errno(status);
 627         up_write(&sb_dqopt(sb)->dqio_sem);
 628         ocfs2_commit_trans(osb, handle);
 629 out_ilock:
 630         ocfs2_unlock_global_qf(oinfo, 1);
 631 out:
 632         return status;
 633 }
 634
 635 static void qsync_work_fn(struct work_struct *work)
 636 {
 637         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
 638                                                       struct ocfs2_mem_dqinfo,
 639                                                       dqi_sync_work.work);
 640         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
 641
 642         /*
 643          * We have to be careful here not to deadlock on s_umount as umount
 644          * disabling quotas may be in progress and it waits for this work to
 645          * complete. If trylock fails, we'll do the sync next time...
 646          */
 647         if (down_read_trylock(&sb->s_umount)) {
 648                 dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
 649                 up_read(&sb->s_umount);
 650         }
 651         schedule_delayed_work(&oinfo->dqi_sync_work,
 652                               msecs_to_jiffies(oinfo->dqi_syncms));
 653 }
 654
 655 /*
 656  *  Wrappers for generic quota functions
 657  */
 658
 659 static int ocfs2_write_dquot(struct dquot *dquot)
 660 {
 661         handle_t *handle;
 662         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 663         int status = 0;
 664
 665         trace_ocfs2_write_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 666                                 dquot->dq_id.type);
 667
 668         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
 669         if (IS_ERR(handle)) {
 670                 status = PTR_ERR(handle);
 671                 mlog_errno(status);
 672                 goto out;
 673         }
 674         down_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 675         status = ocfs2_local_write_dquot(dquot);
 676         up_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 677         ocfs2_commit_trans(osb, handle);
 678 out:
 679         return status;
 680 }
 681
 682 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
 683 {
 684         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 685         /*
 686          * We modify tree, leaf block, global info, local chunk header,
 687          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
 688          * accounts for inode update
 689          */
 690         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
 691                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
 692                OCFS2_QINFO_WRITE_CREDITS +
 693                OCFS2_INODE_UPDATE_CREDITS;
 694 }
 695
 696 void ocfs2_drop_dquot_refs(struct work_struct *work)
 697 {
 698         struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
 699                                                dquot_drop_work);
 700         struct llist_node *list;
 701         struct ocfs2_dquot *odquot, *next_odquot;
 702
 703         list = llist_del_all(&osb->dquot_drop_list);
 704         llist_for_each_entry_safe(odquot, next_odquot, list, list) {
 705                 /* Drop the reference we acquired in ocfs2_dquot_release() */
 706                 dqput(&odquot->dq_dquot);
 707         }
 708 }
 709
 710 /*
 711  * Called when the last reference to dquot is dropped. If we are called from
 712  * downconvert thread, we cannot do all the handling here because grabbing
 713  * quota lock could deadlock (the node holding the quota lock could need some
 714  * other cluster lock to proceed but with blocked downconvert thread we cannot
 715  * release any lock).
 716  */
 717 static int ocfs2_release_dquot(struct dquot *dquot)
 718 {
 719         handle_t *handle;
 720         struct ocfs2_mem_dqinfo *oinfo =
 721                         sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
 722         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 723         int status = 0;
 724
 725         trace_ocfs2_release_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 726                                   dquot->dq_id.type);
 727
 728         mutex_lock(&dquot->dq_lock);
 729         /* Check whether we are not racing with some other dqget() */
 730         if (dquot_is_busy(dquot))
 731                 goto out;
 732         /* Running from downconvert thread? Postpone quota processing to wq */
 733         if (current == osb->dc_task) {
 734                 /*
 735                  * Grab our own reference to dquot and queue it for delayed
 736                  * dropping.  Quota code rechecks after calling
 737                  * ->release_dquot() and won't free dquot structure.
 738                  */
 739                 dqgrab(dquot);
 740                 /* First entry on list -> queue work */
 741                 if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list))
 742                         queue_work(osb->ocfs2_wq, &osb->dquot_drop_work);
 743                 goto out;
 744         }
 745         status = ocfs2_lock_global_qf(oinfo, 1);
 746         if (status < 0)
 747                 goto out;
 748         handle = ocfs2_start_trans(osb,
 749                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_id.type));
 750         if (IS_ERR(handle)) {
 751                 status = PTR_ERR(handle);
 752                 mlog_errno(status);
 753                 goto out_ilock;
 754         }
 755
 756         status = ocfs2_global_release_dquot(dquot);
 757         if (status < 0) {
 758                 mlog_errno(status);
 759                 goto out_trans;
 760         }
 761         status = ocfs2_local_release_dquot(handle, dquot);
 762         /*
 763          * If we fail here, we cannot do much as global structure is
 764          * already released. So just complain...
 765          */
 766         if (status < 0)
 767                 mlog_errno(status);
 768         /*
 769          * Clear dq_off so that we search for the structure in quota file next
 770          * time we acquire it. The structure might be deleted and reallocated
 771          * elsewhere by another node while our dquot structure is on freelist.
 772          */
 773         dquot->dq_off = 0;
 774         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 775 out_trans:
 776         ocfs2_commit_trans(osb, handle);
 777 out_ilock:
 778         ocfs2_unlock_global_qf(oinfo, 1);
 779 out:
 780         mutex_unlock(&dquot->dq_lock);
 781         if (status)
 782                 mlog_errno(status);
 783         return status;
 784 }
 785
 786 /*
 787  * Read global dquot structure from disk or create it if it does
 788  * not exist. Also update use count of the global structure and
 789  * create structure in node-local quota file.
 790  */
 791 static int ocfs2_acquire_dquot(struct dquot *dquot)
 792 {
 793         int status = 0, err;
 794         int ex = 0;
 795         struct super_block *sb = dquot->dq_sb;
 796         struct ocfs2_super *osb = OCFS2_SB(sb);
 797         int type = dquot->dq_id.type;
 798         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 799         struct inode *gqinode = info->dqi_gqinode;
 800         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
 801         handle_t *handle;
 802
 803         trace_ocfs2_acquire_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 804                                   type);
 805         mutex_lock(&dquot->dq_lock);
 806         /*
 807          * We need an exclusive lock, because we're going to update use count
 808          * and instantiate possibly new dquot structure
 809          */
 810         status = ocfs2_lock_global_qf(info, 1);
 811         if (status < 0)
 812                 goto out;
 813         status = ocfs2_qinfo_lock(info, 0);
 814         if (status < 0)
 815                 goto out_dq;
 816         /*
 817          * We always want to read dquot structure from disk because we don't
 818          * know what happened with it while it was on freelist.
 819          */
 820         status = qtree_read_dquot(&info->dqi_gi, dquot);
 821         ocfs2_qinfo_unlock(info, 0);
 822         if (status < 0)
 823                 goto out_dq;
 824
 825         OCFS2_DQUOT(dquot)->dq_use_count++;
 826         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 827         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 828         if (!dquot->dq_off) {   /* No real quota entry? */
 829                 ex = 1;
 830                 /*
 831                  * Add blocks to quota file before we start a transaction since
 832                  * locking allocators ranks above a transaction start
 833                  */
 834                 WARN_ON(journal_current_handle());
 835                 status = ocfs2_extend_no_holes(gqinode, NULL,
 836                         i_size_read(gqinode) + (need_alloc << sb->s_blocksize_bits),
 837                         i_size_read(gqinode));
 838                 if (status < 0)
 839                         goto out_dq;
 840         }
 841
 842         handle = ocfs2_start_trans(osb,
 843                                    ocfs2_calc_global_qinit_credits(sb, type));
 844         if (IS_ERR(handle)) {
 845                 status = PTR_ERR(handle);
 846                 goto out_dq;
 847         }
 848         status = ocfs2_qinfo_lock(info, ex);
 849         if (status < 0)
 850                 goto out_trans;
 851         status = qtree_write_dquot(&info->dqi_gi, dquot);
 852         if (ex && info_dirty(sb_dqinfo(sb, type))) {
 853                 err = __ocfs2_global_write_info(sb, type);
 854                 if (!status)
 855                         status = err;
 856         }
 857         ocfs2_qinfo_unlock(info, ex);
 858 out_trans:
 859         ocfs2_commit_trans(osb, handle);
 860 out_dq:
 861         ocfs2_unlock_global_qf(info, 1);
 862         if (status < 0)
 863                 goto out;
 864
 865         status = ocfs2_create_local_dquot(dquot);
 866         if (status < 0)
 867                 goto out;
 868         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 869 out:
 870         mutex_unlock(&dquot->dq_lock);
 871         if (status)
 872                 mlog_errno(status);
 873         return status;
 874 }
 875
 876 static int ocfs2_get_next_id(struct super_block *sb, struct kqid *qid)
 877 {
 878         int type = qid->type;
 879         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 880         int status = 0;
 881
 882         trace_ocfs2_get_next_id(from_kqid(&init_user_ns, *qid), type);
 883         if (!sb_has_quota_loaded(sb, type)) {
 884                 status = -ESRCH;
 885                 goto out;
 886         }
 887         status = ocfs2_lock_global_qf(info, 0);
 888         if (status < 0)
 889                 goto out;
 890         status = ocfs2_qinfo_lock(info, 0);
 891         if (status < 0)
 892                 goto out_global;
 893         status = qtree_get_next_id(&info->dqi_gi, qid);
 894         ocfs2_qinfo_unlock(info, 0);
 895 out_global:
 896         ocfs2_unlock_global_qf(info, 0);
 897 out:
 898         /*
 899          * Avoid logging ENOENT since it just means there isn't next ID and
 900          * ESRCH which means quota isn't enabled for the filesystem.
 901          */
 902         if (status && status != -ENOENT && status != -ESRCH)
 903                 mlog_errno(status);
 904         return status;
 905 }
 906
 907 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
 908 {
 909         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
 910                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
 911                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
 912                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
 913                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
 914                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
 915         int sync = 0;
 916         int status;
 917         struct super_block *sb = dquot->dq_sb;
 918         int type = dquot->dq_id.type;
 919         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 920         handle_t *handle;
 921         struct ocfs2_super *osb = OCFS2_SB(sb);
 922
 923         trace_ocfs2_mark_dquot_dirty(from_kqid(&init_user_ns, dquot->dq_id),
 924                                      type);
 925
 926         /* In case user set some limits, sync dquot immediately to global
 927          * quota file so that information propagates quicker */
 928         spin_lock(&dquot->dq_dqb_lock);
 929         if (dquot->dq_flags & mask)
 930                 sync = 1;
 931         spin_unlock(&dquot->dq_dqb_lock);
 932         /* This is a slight hack but we can't afford getting global quota
 933          * lock if we already have a transaction started. */
 934         if (!sync || journal_current_handle()) {
 935                 status = ocfs2_write_dquot(dquot);
 936                 goto out;
 937         }
 938         status = ocfs2_lock_global_qf(oinfo, 1);
 939         if (status < 0)
 940                 goto out;
 941         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 942         if (IS_ERR(handle)) {
 943                 status = PTR_ERR(handle);
 944                 mlog_errno(status);
 945                 goto out_ilock;
 946         }
 947         down_write(&sb_dqopt(sb)->dqio_sem);
 948         status = ocfs2_sync_dquot(dquot);
 949         if (status < 0) {
 950                 mlog_errno(status);
 951                 goto out_dlock;
 952         }
 953         /* Now write updated local dquot structure */
 954         status = ocfs2_local_write_dquot(dquot);
 955 out_dlock:
 956         up_write(&sb_dqopt(sb)->dqio_sem);
 957         ocfs2_commit_trans(osb, handle);
 958 out_ilock:
 959         ocfs2_unlock_global_qf(oinfo, 1);
 960 out:
 961         if (status)
 962                 mlog_errno(status);
 963         return status;
 964 }
 965
 966 /* This should happen only after set_dqinfo(). */
 967 static int ocfs2_write_info(struct super_block *sb, int type)
 968 {
 969         handle_t *handle;
 970         int status = 0;
 971         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 972
 973         status = ocfs2_lock_global_qf(oinfo, 1);
 974         if (status < 0)
 975                 goto out;
 976         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
 977         if (IS_ERR(handle)) {
 978                 status = PTR_ERR(handle);
 979                 mlog_errno(status);
 980                 goto out_ilock;
 981         }
 982         status = dquot_commit_info(sb, type);
 983         ocfs2_commit_trans(OCFS2_SB(sb), handle);
 984 out_ilock:
 985         ocfs2_unlock_global_qf(oinfo, 1);
 986 out:
 987         if (status)
 988                 mlog_errno(status);
 989         return status;
 990 }
 991
 992 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
 993 {
 994         struct ocfs2_dquot *dquot =
 995                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
 996
 997         if (!dquot)
 998                 return NULL;
 999         return &dquot->dq_dquot;
1000 }
1001
1002 static void ocfs2_destroy_dquot(struct dquot *dquot)
1003 {
1004         kmem_cache_free(ocfs2_dquot_cachep, dquot);
1005 }
1006
1007 const struct dquot_operations ocfs2_quota_operations = {
1008         /* We never make dquot dirty so .write_dquot is never called */
1009         .acquire_dquot  = ocfs2_acquire_dquot,
1010         .release_dquot  = ocfs2_release_dquot,
1011         .mark_dirty     = ocfs2_mark_dquot_dirty,
1012         .write_info     = ocfs2_write_info,
1013         .alloc_dquot    = ocfs2_alloc_dquot,
1014         .destroy_dquot  = ocfs2_destroy_dquot,
1015         .get_next_id    = ocfs2_get_next_id,
1016 };