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
  69static void qsync_work_fn(struct work_struct *work);
  70
  71static 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
  96static 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
 114static 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
 128const 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
 134int 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
 151int 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. */
 167ssize_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) */
 216ssize_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_folio(bh->b_folio);
 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);
 288out:
 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
 298int 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
 321void 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 */
 338int 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		mlog_errno(status);
 376		goto out_unlock;
 377	}
 378
 379	status = ocfs2_qinfo_lock(oinfo, 0);
 380	if (status < 0) {
 381		mlog_errno(status);
 382		goto out_unlock;
 383	}
 384	status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
 385				      sizeof(struct ocfs2_global_disk_dqinfo),
 386				      OCFS2_GLOBAL_INFO_OFF);
 387	ocfs2_qinfo_unlock(oinfo, 0);
 388	ocfs2_unlock_global_qf(oinfo, 0);
 389	if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
 390		mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
 391		     status);
 392		if (status >= 0)
 393			status = -EIO;
 394		mlog_errno(status);
 395		goto out_err;
 396	}
 397	info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
 398	info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
 399	oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
 400	oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
 401	oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
 402	oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
 403	oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
 404	oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
 405						OCFS2_QBLK_RESERVED_SPACE;
 406	oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
 407	INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
 408	schedule_delayed_work(&oinfo->dqi_sync_work,
 409			      msecs_to_jiffies(oinfo->dqi_syncms));
 410
 411	return 0;
 412out_unlock:
 413	ocfs2_unlock_global_qf(oinfo, 0);
 414out_err:
 415	return status;
 416}
 417
 418/* Write information to global quota file. Expects exclusive lock on quota
 419 * file inode and quota info */
 420static int __ocfs2_global_write_info(struct super_block *sb, int type)
 421{
 422	struct mem_dqinfo *info = sb_dqinfo(sb, type);
 423	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
 424	struct ocfs2_global_disk_dqinfo dinfo;
 425	ssize_t size;
 426
 427	spin_lock(&dq_data_lock);
 428	info->dqi_flags &= ~DQF_INFO_DIRTY;
 429	dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
 430	dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
 431	spin_unlock(&dq_data_lock);
 432	dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
 433	dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
 434	dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
 435	dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
 436	size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
 437				     sizeof(struct ocfs2_global_disk_dqinfo),
 438				     OCFS2_GLOBAL_INFO_OFF);
 439	if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
 440		mlog(ML_ERROR, "Cannot write global quota info structure\n");
 441		if (size >= 0)
 442			size = -EIO;
 443		return size;
 444	}
 445	return 0;
 446}
 447
 448int ocfs2_global_write_info(struct super_block *sb, int type)
 449{
 450	int err;
 451	struct quota_info *dqopt = sb_dqopt(sb);
 452	struct ocfs2_mem_dqinfo *info = dqopt->info[type].dqi_priv;
 453	unsigned int memalloc;
 454
 455	down_write(&dqopt->dqio_sem);
 456	memalloc = memalloc_nofs_save();
 457	err = ocfs2_qinfo_lock(info, 1);
 458	if (err < 0)
 459		goto out_sem;
 460	err = __ocfs2_global_write_info(sb, type);
 461	ocfs2_qinfo_unlock(info, 1);
 462out_sem:
 463	memalloc_nofs_restore(memalloc);
 464	up_write(&dqopt->dqio_sem);
 465	return err;
 466}
 467
 468static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
 469{
 470	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 471
 472	/*
 473	 * We may need to allocate tree blocks and a leaf block but not the
 474	 * root block
 475	 */
 476	return oinfo->dqi_gi.dqi_qtree_depth;
 477}
 478
 479static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
 480{
 481	/* We modify all the allocated blocks, tree root, info block and
 482	 * the inode */
 483	return (ocfs2_global_qinit_alloc(sb, type) + 2) *
 484			OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
 485}
 486
 487/* Sync local information about quota modifications with global quota file.
 488 * Caller must have started the transaction and obtained exclusive lock for
 489 * global quota file inode */
 490int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
 491{
 492	int err, err2;
 493	struct super_block *sb = dquot->dq_sb;
 494	int type = dquot->dq_id.type;
 495	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 496	struct ocfs2_global_disk_dqblk dqblk;
 497	s64 spacechange, inodechange;
 498	time64_t olditime, oldbtime;
 499
 500	err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
 501				   sizeof(struct ocfs2_global_disk_dqblk),
 502				   dquot->dq_off);
 503	if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
 504		if (err >= 0) {
 505			mlog(ML_ERROR, "Short read from global quota file "
 506				       "(%u read)\n", err);
 507			err = -EIO;
 508		}
 509		goto out;
 510	}
 511
 512	/* Update space and inode usage. Get also other information from
 513	 * global quota file so that we don't overwrite any changes there.
 514	 * We are */
 515	spin_lock(&dquot->dq_dqb_lock);
 516	spacechange = dquot->dq_dqb.dqb_curspace -
 517					OCFS2_DQUOT(dquot)->dq_origspace;
 518	inodechange = dquot->dq_dqb.dqb_curinodes -
 519					OCFS2_DQUOT(dquot)->dq_originodes;
 520	olditime = dquot->dq_dqb.dqb_itime;
 521	oldbtime = dquot->dq_dqb.dqb_btime;
 522	ocfs2_global_disk2memdqb(dquot, &dqblk);
 523	trace_ocfs2_sync_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 524			       dquot->dq_dqb.dqb_curspace,
 525			       (long long)spacechange,
 526			       dquot->dq_dqb.dqb_curinodes,
 527			       (long long)inodechange);
 528	if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
 529		dquot->dq_dqb.dqb_curspace += spacechange;
 530	if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
 531		dquot->dq_dqb.dqb_curinodes += inodechange;
 532	/* Set properly space grace time... */
 533	if (dquot->dq_dqb.dqb_bsoftlimit &&
 534	    dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
 535		if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
 536		    oldbtime > 0) {
 537			if (dquot->dq_dqb.dqb_btime > 0)
 538				dquot->dq_dqb.dqb_btime =
 539					min(dquot->dq_dqb.dqb_btime, oldbtime);
 540			else
 541				dquot->dq_dqb.dqb_btime = oldbtime;
 542		}
 543	} else {
 544		dquot->dq_dqb.dqb_btime = 0;
 545		clear_bit(DQ_BLKS_B, &dquot->dq_flags);
 546	}
 547	/* Set properly inode grace time... */
 548	if (dquot->dq_dqb.dqb_isoftlimit &&
 549	    dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
 550		if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
 551		    olditime > 0) {
 552			if (dquot->dq_dqb.dqb_itime > 0)
 553				dquot->dq_dqb.dqb_itime =
 554					min(dquot->dq_dqb.dqb_itime, olditime);
 555			else
 556				dquot->dq_dqb.dqb_itime = olditime;
 557		}
 558	} else {
 559		dquot->dq_dqb.dqb_itime = 0;
 560		clear_bit(DQ_INODES_B, &dquot->dq_flags);
 561	}
 562	/* All information is properly updated, clear the flags */
 563	__clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
 564	__clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
 565	__clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
 566	__clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
 567	__clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
 568	__clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
 569	OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 570	OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 571	spin_unlock(&dquot->dq_dqb_lock);
 572	err = ocfs2_qinfo_lock(info, freeing);
 573	if (err < 0) {
 574		mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
 575			       " (type=%d, id=%u)\n", dquot->dq_id.type,
 576			       (unsigned)from_kqid(&init_user_ns, dquot->dq_id));
 577		goto out;
 578	}
 579	if (freeing)
 580		OCFS2_DQUOT(dquot)->dq_use_count--;
 581	err = qtree_write_dquot(&info->dqi_gi, dquot);
 582	if (err < 0)
 583		goto out_qlock;
 584	if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
 585		err = qtree_release_dquot(&info->dqi_gi, dquot);
 586		if (info_dirty(sb_dqinfo(sb, type))) {
 587			err2 = __ocfs2_global_write_info(sb, type);
 588			if (!err)
 589				err = err2;
 590		}
 591	}
 592out_qlock:
 593	ocfs2_qinfo_unlock(info, freeing);
 594out:
 595	if (err < 0)
 596		mlog_errno(err);
 597	return err;
 598}
 599
 600/*
 601 *  Functions for periodic syncing of dquots with global file
 602 */
 603static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
 604{
 605	handle_t *handle;
 606	struct super_block *sb = dquot->dq_sb;
 607	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 608	struct ocfs2_super *osb = OCFS2_SB(sb);
 609	int status = 0;
 610	unsigned int memalloc;
 611
 612	trace_ocfs2_sync_dquot_helper(from_kqid(&init_user_ns, dquot->dq_id),
 613				      dquot->dq_id.type,
 614				      type, sb->s_id);
 615	if (type != dquot->dq_id.type)
 616		goto out;
 617	status = ocfs2_lock_global_qf(oinfo, 1);
 618	if (status < 0)
 619		goto out;
 620
 621	handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 622	if (IS_ERR(handle)) {
 623		status = PTR_ERR(handle);
 624		mlog_errno(status);
 625		goto out_ilock;
 626	}
 627	down_write(&sb_dqopt(sb)->dqio_sem);
 628	memalloc = memalloc_nofs_save();
 629	status = ocfs2_sync_dquot(dquot);
 630	if (status < 0)
 631		mlog_errno(status);
 632	/* We have to write local structure as well... */
 633	status = ocfs2_local_write_dquot(dquot);
 634	if (status < 0)
 635		mlog_errno(status);
 636	memalloc_nofs_restore(memalloc);
 637	up_write(&sb_dqopt(sb)->dqio_sem);
 638	ocfs2_commit_trans(osb, handle);
 639out_ilock:
 640	ocfs2_unlock_global_qf(oinfo, 1);
 641out:
 642	return status;
 643}
 644
 645static void qsync_work_fn(struct work_struct *work)
 646{
 647	struct ocfs2_mem_dqinfo *oinfo = container_of(work,
 648						      struct ocfs2_mem_dqinfo,
 649						      dqi_sync_work.work);
 650	struct super_block *sb = oinfo->dqi_gqinode->i_sb;
 651
 652	/*
 653	 * We have to be careful here not to deadlock on s_umount as umount
 654	 * disabling quotas may be in progress and it waits for this work to
 655	 * complete. If trylock fails, we'll do the sync next time...
 656	 */
 657	if (down_read_trylock(&sb->s_umount)) {
 658		dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
 659		up_read(&sb->s_umount);
 660	}
 661	schedule_delayed_work(&oinfo->dqi_sync_work,
 662			      msecs_to_jiffies(oinfo->dqi_syncms));
 663}
 664
 665/*
 666 *  Wrappers for generic quota functions
 667 */
 668
 669static int ocfs2_write_dquot(struct dquot *dquot)
 670{
 671	handle_t *handle;
 672	struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 673	int status = 0;
 674	unsigned int memalloc;
 675
 676	trace_ocfs2_write_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 677				dquot->dq_id.type);
 678
 679	handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
 680	if (IS_ERR(handle)) {
 681		status = PTR_ERR(handle);
 682		mlog_errno(status);
 683		goto out;
 684	}
 685	down_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 686	memalloc = memalloc_nofs_save();
 687	status = ocfs2_local_write_dquot(dquot);
 688	memalloc_nofs_restore(memalloc);
 689	up_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
 690	ocfs2_commit_trans(osb, handle);
 691out:
 692	return status;
 693}
 694
 695static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
 696{
 697	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 698	/*
 699	 * We modify tree, leaf block, global info, local chunk header,
 700	 * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
 701	 * accounts for inode update
 702	 */
 703	return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
 704	       OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
 705	       OCFS2_QINFO_WRITE_CREDITS +
 706	       OCFS2_INODE_UPDATE_CREDITS;
 707}
 708
 709void ocfs2_drop_dquot_refs(struct work_struct *work)
 710{
 711	struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
 712					       dquot_drop_work);
 713	struct llist_node *list;
 714	struct ocfs2_dquot *odquot, *next_odquot;
 715
 716	list = llist_del_all(&osb->dquot_drop_list);
 717	llist_for_each_entry_safe(odquot, next_odquot, list, list) {
 718		/* Drop the reference we acquired in ocfs2_dquot_release() */
 719		dqput(&odquot->dq_dquot);
 720	}
 721}
 722
 723/*
 724 * Called when the last reference to dquot is dropped. If we are called from
 725 * downconvert thread, we cannot do all the handling here because grabbing
 726 * quota lock could deadlock (the node holding the quota lock could need some
 727 * other cluster lock to proceed but with blocked downconvert thread we cannot
 728 * release any lock).
 729 */
 730static int ocfs2_release_dquot(struct dquot *dquot)
 731{
 732	handle_t *handle;
 733	struct ocfs2_mem_dqinfo *oinfo =
 734			sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
 735	struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
 736	int status = 0;
 737
 738	trace_ocfs2_release_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 739				  dquot->dq_id.type);
 740
 741	mutex_lock(&dquot->dq_lock);
 742	/* Check whether we are not racing with some other dqget() */
 743	if (dquot_is_busy(dquot))
 744		goto out;
 745	/* Running from downconvert thread? Postpone quota processing to wq */
 746	if (current == osb->dc_task) {
 747		/*
 748		 * Grab our own reference to dquot and queue it for delayed
 749		 * dropping.  Quota code rechecks after calling
 750		 * ->release_dquot() and won't free dquot structure.
 751		 */
 752		dqgrab(dquot);
 753		/* First entry on list -> queue work */
 754		if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list))
 755			queue_work(osb->ocfs2_wq, &osb->dquot_drop_work);
 756		goto out;
 757	}
 758	status = ocfs2_lock_global_qf(oinfo, 1);
 759	if (status < 0)
 760		goto out;
 761	handle = ocfs2_start_trans(osb,
 762		ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_id.type));
 763	if (IS_ERR(handle)) {
 764		/*
 765		 * Mark dquot as inactive to avoid endless cycle in
 766		 * quota_release_workfn().
 767		 */
 768		clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 769		status = PTR_ERR(handle);
 770		mlog_errno(status);
 771		goto out_ilock;
 772	}
 773
 774	status = ocfs2_global_release_dquot(dquot);
 775	if (status < 0) {
 776		mlog_errno(status);
 777		goto out_trans;
 778	}
 779	status = ocfs2_local_release_dquot(handle, dquot);
 780	/*
 781	 * If we fail here, we cannot do much as global structure is
 782	 * already released. So just complain...
 783	 */
 784	if (status < 0)
 785		mlog_errno(status);
 786	/*
 787	 * Clear dq_off so that we search for the structure in quota file next
 788	 * time we acquire it. The structure might be deleted and reallocated
 789	 * elsewhere by another node while our dquot structure is on freelist.
 790	 */
 791	dquot->dq_off = 0;
 792	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 793out_trans:
 794	ocfs2_commit_trans(osb, handle);
 795out_ilock:
 796	ocfs2_unlock_global_qf(oinfo, 1);
 797out:
 798	mutex_unlock(&dquot->dq_lock);
 799	if (status)
 800		mlog_errno(status);
 801	return status;
 802}
 803
 804/*
 805 * Read global dquot structure from disk or create it if it does
 806 * not exist. Also update use count of the global structure and
 807 * create structure in node-local quota file.
 808 */
 809static int ocfs2_acquire_dquot(struct dquot *dquot)
 810{
 811	int status = 0, err;
 812	int ex = 0;
 813	struct super_block *sb = dquot->dq_sb;
 814	struct ocfs2_super *osb = OCFS2_SB(sb);
 815	int type = dquot->dq_id.type;
 816	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 817	struct inode *gqinode = info->dqi_gqinode;
 818	int need_alloc = ocfs2_global_qinit_alloc(sb, type);
 819	handle_t *handle;
 820
 821	trace_ocfs2_acquire_dquot(from_kqid(&init_user_ns, dquot->dq_id),
 822				  type);
 823	mutex_lock(&dquot->dq_lock);
 824	/*
 825	 * We need an exclusive lock, because we're going to update use count
 826	 * and instantiate possibly new dquot structure
 827	 */
 828	status = ocfs2_lock_global_qf(info, 1);
 829	if (status < 0)
 830		goto out;
 831	status = ocfs2_qinfo_lock(info, 0);
 832	if (status < 0)
 833		goto out_dq;
 834	/*
 835	 * We always want to read dquot structure from disk because we don't
 836	 * know what happened with it while it was on freelist.
 837	 */
 838	status = qtree_read_dquot(&info->dqi_gi, dquot);
 839	ocfs2_qinfo_unlock(info, 0);
 840	if (status < 0)
 841		goto out_dq;
 842
 843	OCFS2_DQUOT(dquot)->dq_use_count++;
 844	OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
 845	OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
 846	if (!dquot->dq_off) {	/* No real quota entry? */
 847		ex = 1;
 848		/*
 849		 * Add blocks to quota file before we start a transaction since
 850		 * locking allocators ranks above a transaction start
 851		 */
 852		WARN_ON(journal_current_handle());
 853		status = ocfs2_extend_no_holes(gqinode, NULL,
 854			i_size_read(gqinode) + (need_alloc << sb->s_blocksize_bits),
 855			i_size_read(gqinode));
 856		if (status < 0)
 857			goto out_dq;
 858	}
 859
 860	handle = ocfs2_start_trans(osb,
 861				   ocfs2_calc_global_qinit_credits(sb, type));
 862	if (IS_ERR(handle)) {
 863		status = PTR_ERR(handle);
 864		goto out_dq;
 865	}
 866	status = ocfs2_qinfo_lock(info, ex);
 867	if (status < 0)
 868		goto out_trans;
 869	status = qtree_write_dquot(&info->dqi_gi, dquot);
 870	if (ex && info_dirty(sb_dqinfo(sb, type))) {
 871		err = __ocfs2_global_write_info(sb, type);
 872		if (!status)
 873			status = err;
 874	}
 875	ocfs2_qinfo_unlock(info, ex);
 876out_trans:
 877	ocfs2_commit_trans(osb, handle);
 878out_dq:
 879	ocfs2_unlock_global_qf(info, 1);
 880	if (status < 0)
 881		goto out;
 882
 883	status = ocfs2_create_local_dquot(dquot);
 884	if (status < 0)
 885		goto out;
 886	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
 887out:
 888	mutex_unlock(&dquot->dq_lock);
 889	if (status)
 890		mlog_errno(status);
 891	return status;
 892}
 893
 894static int ocfs2_get_next_id(struct super_block *sb, struct kqid *qid)
 895{
 896	int type = qid->type;
 897	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
 898	int status = 0;
 899
 900	trace_ocfs2_get_next_id(from_kqid(&init_user_ns, *qid), type);
 901	if (!sb_has_quota_active(sb, type)) {
 902		status = -ESRCH;
 903		goto out;
 904	}
 905	status = ocfs2_lock_global_qf(info, 0);
 906	if (status < 0)
 907		goto out;
 908	status = ocfs2_qinfo_lock(info, 0);
 909	if (status < 0)
 910		goto out_global;
 911	status = qtree_get_next_id(&info->dqi_gi, qid);
 912	ocfs2_qinfo_unlock(info, 0);
 913out_global:
 914	ocfs2_unlock_global_qf(info, 0);
 915out:
 916	/*
 917	 * Avoid logging ENOENT since it just means there isn't next ID and
 918	 * ESRCH which means quota isn't enabled for the filesystem.
 919	 */
 920	if (status && status != -ENOENT && status != -ESRCH)
 921		mlog_errno(status);
 922	return status;
 923}
 924
 925static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
 926{
 927	unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
 928			     (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
 929			     (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
 930			     (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
 931			     (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
 932			     (1 << (DQ_LASTSET_B + QIF_ITIME_B));
 933	int sync = 0;
 934	int status;
 935	struct super_block *sb = dquot->dq_sb;
 936	int type = dquot->dq_id.type;
 937	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 938	handle_t *handle;
 939	struct ocfs2_super *osb = OCFS2_SB(sb);
 940	unsigned int memalloc;
 941
 942	trace_ocfs2_mark_dquot_dirty(from_kqid(&init_user_ns, dquot->dq_id),
 943				     type);
 944
 945	/* In case user set some limits, sync dquot immediately to global
 946	 * quota file so that information propagates quicker */
 947	spin_lock(&dquot->dq_dqb_lock);
 948	if (dquot->dq_flags & mask)
 949		sync = 1;
 950	spin_unlock(&dquot->dq_dqb_lock);
 951	/* This is a slight hack but we can't afford getting global quota
 952	 * lock if we already have a transaction started. */
 953	if (!sync || journal_current_handle()) {
 954		status = ocfs2_write_dquot(dquot);
 955		goto out;
 956	}
 957	status = ocfs2_lock_global_qf(oinfo, 1);
 958	if (status < 0)
 959		goto out;
 960	handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
 961	if (IS_ERR(handle)) {
 962		status = PTR_ERR(handle);
 963		mlog_errno(status);
 964		goto out_ilock;
 965	}
 966	down_write(&sb_dqopt(sb)->dqio_sem);
 967	memalloc = memalloc_nofs_save();
 968	status = ocfs2_sync_dquot(dquot);
 969	if (status < 0) {
 970		mlog_errno(status);
 971		goto out_dlock;
 972	}
 973	/* Now write updated local dquot structure */
 974	status = ocfs2_local_write_dquot(dquot);
 975out_dlock:
 976	memalloc_nofs_restore(memalloc);
 977	up_write(&sb_dqopt(sb)->dqio_sem);
 978	ocfs2_commit_trans(osb, handle);
 979out_ilock:
 980	ocfs2_unlock_global_qf(oinfo, 1);
 981out:
 982	if (status)
 983		mlog_errno(status);
 984	return status;
 985}
 986
 987/* This should happen only after set_dqinfo(). */
 988static int ocfs2_write_info(struct super_block *sb, int type)
 989{
 990	handle_t *handle;
 991	int status = 0;
 992	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
 993
 994	status = ocfs2_lock_global_qf(oinfo, 1);
 995	if (status < 0)
 996		goto out;
 997	handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
 998	if (IS_ERR(handle)) {
 999		status = PTR_ERR(handle);
1000		mlog_errno(status);
1001		goto out_ilock;
1002	}
1003	status = dquot_commit_info(sb, type);
1004	ocfs2_commit_trans(OCFS2_SB(sb), handle);
1005out_ilock:
1006	ocfs2_unlock_global_qf(oinfo, 1);
1007out:
1008	if (status)
1009		mlog_errno(status);
1010	return status;
1011}
1012
1013static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
1014{
1015	struct ocfs2_dquot *dquot =
1016				kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
1017
1018	if (!dquot)
1019		return NULL;
1020	return &dquot->dq_dquot;
1021}
1022
1023static void ocfs2_destroy_dquot(struct dquot *dquot)
1024{
1025	kmem_cache_free(ocfs2_dquot_cachep, dquot);
1026}
1027
1028const struct dquot_operations ocfs2_quota_operations = {
1029	/* We never make dquot dirty so .write_dquot is never called */
1030	.acquire_dquot	= ocfs2_acquire_dquot,
1031	.release_dquot	= ocfs2_release_dquot,
1032	.mark_dirty	= ocfs2_mark_dquot_dirty,
1033	.write_info	= ocfs2_write_info,
1034	.alloc_dquot	= ocfs2_alloc_dquot,
1035	.destroy_dquot	= ocfs2_destroy_dquot,
1036	.get_next_id	= ocfs2_get_next_id,
1037};