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
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_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);
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	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	ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
361	oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
362	oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
363	oinfo->dqi_gqi_bh = NULL;
364	oinfo->dqi_gqi_count = 0;
365	oinfo->dqi_gqinode = gqinode;
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(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
407out_err:
408	return status;
409out_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 exlusive lock on quota
416 * file inode and quota info */
417static 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
445int 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
451	down_write(&dqopt->dqio_sem);
452	err = ocfs2_qinfo_lock(info, 1);
453	if (err < 0)
454		goto out_sem;
455	err = __ocfs2_global_write_info(sb, type);
456	ocfs2_qinfo_unlock(info, 1);
457out_sem:
458	up_write(&dqopt->dqio_sem);
459	return err;
460}
461
462static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
463{
464	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
465
466	/*
467	 * We may need to allocate tree blocks and a leaf block but not the
468	 * root block
469	 */
470	return oinfo->dqi_gi.dqi_qtree_depth;
471}
472
473static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
474{
475	/* We modify all the allocated blocks, tree root, info block and
476	 * the inode */
477	return (ocfs2_global_qinit_alloc(sb, type) + 2) *
478			OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
479}
480
481/* Sync local information about quota modifications with global quota file.
482 * Caller must have started the transaction and obtained exclusive lock for
483 * global quota file inode */
484int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
485{
486	int err, err2;
487	struct super_block *sb = dquot->dq_sb;
488	int type = dquot->dq_id.type;
489	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
490	struct ocfs2_global_disk_dqblk dqblk;
491	s64 spacechange, inodechange;
492	time64_t olditime, oldbtime;
493
494	err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
495				   sizeof(struct ocfs2_global_disk_dqblk),
496				   dquot->dq_off);
497	if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
498		if (err >= 0) {
499			mlog(ML_ERROR, "Short read from global quota file "
500				       "(%u read)\n", err);
501			err = -EIO;
502		}
503		goto out;
504	}
505
506	/* Update space and inode usage. Get also other information from
507	 * global quota file so that we don't overwrite any changes there.
508	 * We are */
509	spin_lock(&dquot->dq_dqb_lock);
510	spacechange = dquot->dq_dqb.dqb_curspace -
511					OCFS2_DQUOT(dquot)->dq_origspace;
512	inodechange = dquot->dq_dqb.dqb_curinodes -
513					OCFS2_DQUOT(dquot)->dq_originodes;
514	olditime = dquot->dq_dqb.dqb_itime;
515	oldbtime = dquot->dq_dqb.dqb_btime;
516	ocfs2_global_disk2memdqb(dquot, &dqblk);
517	trace_ocfs2_sync_dquot(from_kqid(&init_user_ns, dquot->dq_id),
518			       dquot->dq_dqb.dqb_curspace,
519			       (long long)spacechange,
520			       dquot->dq_dqb.dqb_curinodes,
521			       (long long)inodechange);
522	if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
523		dquot->dq_dqb.dqb_curspace += spacechange;
524	if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
525		dquot->dq_dqb.dqb_curinodes += inodechange;
526	/* Set properly space grace time... */
527	if (dquot->dq_dqb.dqb_bsoftlimit &&
528	    dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
529		if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
530		    oldbtime > 0) {
531			if (dquot->dq_dqb.dqb_btime > 0)
532				dquot->dq_dqb.dqb_btime =
533					min(dquot->dq_dqb.dqb_btime, oldbtime);
534			else
535				dquot->dq_dqb.dqb_btime = oldbtime;
536		}
537	} else {
538		dquot->dq_dqb.dqb_btime = 0;
539		clear_bit(DQ_BLKS_B, &dquot->dq_flags);
540	}
541	/* Set properly inode grace time... */
542	if (dquot->dq_dqb.dqb_isoftlimit &&
543	    dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
544		if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
545		    olditime > 0) {
546			if (dquot->dq_dqb.dqb_itime > 0)
547				dquot->dq_dqb.dqb_itime =
548					min(dquot->dq_dqb.dqb_itime, olditime);
549			else
550				dquot->dq_dqb.dqb_itime = olditime;
551		}
552	} else {
553		dquot->dq_dqb.dqb_itime = 0;
554		clear_bit(DQ_INODES_B, &dquot->dq_flags);
555	}
556	/* All information is properly updated, clear the flags */
557	__clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
558	__clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
559	__clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
560	__clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
561	__clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
562	__clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
563	OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
564	OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
565	spin_unlock(&dquot->dq_dqb_lock);
566	err = ocfs2_qinfo_lock(info, freeing);
567	if (err < 0) {
568		mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
569			       " (type=%d, id=%u)\n", dquot->dq_id.type,
570			       (unsigned)from_kqid(&init_user_ns, dquot->dq_id));
571		goto out;
572	}
573	if (freeing)
574		OCFS2_DQUOT(dquot)->dq_use_count--;
575	err = qtree_write_dquot(&info->dqi_gi, dquot);
576	if (err < 0)
577		goto out_qlock;
578	if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
579		err = qtree_release_dquot(&info->dqi_gi, dquot);
580		if (info_dirty(sb_dqinfo(sb, type))) {
581			err2 = __ocfs2_global_write_info(sb, type);
582			if (!err)
583				err = err2;
584		}
585	}
586out_qlock:
587	ocfs2_qinfo_unlock(info, freeing);
588out:
589	if (err < 0)
590		mlog_errno(err);
591	return err;
592}
593
594/*
595 *  Functions for periodic syncing of dquots with global file
596 */
597static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
598{
599	handle_t *handle;
600	struct super_block *sb = dquot->dq_sb;
601	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
602	struct ocfs2_super *osb = OCFS2_SB(sb);
603	int status = 0;
604
605	trace_ocfs2_sync_dquot_helper(from_kqid(&init_user_ns, dquot->dq_id),
606				      dquot->dq_id.type,
607				      type, sb->s_id);
608	if (type != dquot->dq_id.type)
609		goto out;
610	status = ocfs2_lock_global_qf(oinfo, 1);
611	if (status < 0)
612		goto out;
613
614	handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
615	if (IS_ERR(handle)) {
616		status = PTR_ERR(handle);
617		mlog_errno(status);
618		goto out_ilock;
619	}
620	down_write(&sb_dqopt(sb)->dqio_sem);
621	status = ocfs2_sync_dquot(dquot);
622	if (status < 0)
623		mlog_errno(status);
624	/* We have to write local structure as well... */
625	status = ocfs2_local_write_dquot(dquot);
626	if (status < 0)
627		mlog_errno(status);
628	up_write(&sb_dqopt(sb)->dqio_sem);
629	ocfs2_commit_trans(osb, handle);
630out_ilock:
631	ocfs2_unlock_global_qf(oinfo, 1);
632out:
633	return status;
634}
635
636static void qsync_work_fn(struct work_struct *work)
637{
638	struct ocfs2_mem_dqinfo *oinfo = container_of(work,
639						      struct ocfs2_mem_dqinfo,
640						      dqi_sync_work.work);
641	struct super_block *sb = oinfo->dqi_gqinode->i_sb;
642
643	/*
644	 * We have to be careful here not to deadlock on s_umount as umount
645	 * disabling quotas may be in progress and it waits for this work to
646	 * complete. If trylock fails, we'll do the sync next time...
647	 */
648	if (down_read_trylock(&sb->s_umount)) {
649		dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
650		up_read(&sb->s_umount);
651	}
652	schedule_delayed_work(&oinfo->dqi_sync_work,
653			      msecs_to_jiffies(oinfo->dqi_syncms));
654}
655
656/*
657 *  Wrappers for generic quota functions
658 */
659
660static int ocfs2_write_dquot(struct dquot *dquot)
661{
662	handle_t *handle;
663	struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
664	int status = 0;
665
666	trace_ocfs2_write_dquot(from_kqid(&init_user_ns, dquot->dq_id),
667				dquot->dq_id.type);
668
669	handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
670	if (IS_ERR(handle)) {
671		status = PTR_ERR(handle);
672		mlog_errno(status);
673		goto out;
674	}
675	down_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
676	status = ocfs2_local_write_dquot(dquot);
677	up_write(&sb_dqopt(dquot->dq_sb)->dqio_sem);
678	ocfs2_commit_trans(osb, handle);
679out:
680	return status;
681}
682
683static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
684{
685	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
686	/*
687	 * We modify tree, leaf block, global info, local chunk header,
688	 * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
689	 * accounts for inode update
690	 */
691	return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
692	       OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
693	       OCFS2_QINFO_WRITE_CREDITS +
694	       OCFS2_INODE_UPDATE_CREDITS;
695}
696
697void ocfs2_drop_dquot_refs(struct work_struct *work)
698{
699	struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
700					       dquot_drop_work);
701	struct llist_node *list;
702	struct ocfs2_dquot *odquot, *next_odquot;
703
704	list = llist_del_all(&osb->dquot_drop_list);
705	llist_for_each_entry_safe(odquot, next_odquot, list, list) {
706		/* Drop the reference we acquired in ocfs2_dquot_release() */
707		dqput(&odquot->dq_dquot);
708	}
709}
710
711/*
712 * Called when the last reference to dquot is dropped. If we are called from
713 * downconvert thread, we cannot do all the handling here because grabbing
714 * quota lock could deadlock (the node holding the quota lock could need some
715 * other cluster lock to proceed but with blocked downconvert thread we cannot
716 * release any lock).
717 */
718static int ocfs2_release_dquot(struct dquot *dquot)
719{
720	handle_t *handle;
721	struct ocfs2_mem_dqinfo *oinfo =
722			sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
723	struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
724	int status = 0;
725
726	trace_ocfs2_release_dquot(from_kqid(&init_user_ns, dquot->dq_id),
727				  dquot->dq_id.type);
728
729	mutex_lock(&dquot->dq_lock);
730	/* Check whether we are not racing with some other dqget() */
731	if (dquot_is_busy(dquot))
732		goto out;
733	/* Running from downconvert thread? Postpone quota processing to wq */
734	if (current == osb->dc_task) {
735		/*
736		 * Grab our own reference to dquot and queue it for delayed
737		 * dropping.  Quota code rechecks after calling
738		 * ->release_dquot() and won't free dquot structure.
739		 */
740		dqgrab(dquot);
741		/* First entry on list -> queue work */
742		if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list))
743			queue_work(osb->ocfs2_wq, &osb->dquot_drop_work);
744		goto out;
745	}
746	status = ocfs2_lock_global_qf(oinfo, 1);
747	if (status < 0)
748		goto out;
749	handle = ocfs2_start_trans(osb,
750		ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_id.type));
751	if (IS_ERR(handle)) {
752		status = PTR_ERR(handle);
753		mlog_errno(status);
754		goto out_ilock;
755	}
756
757	status = ocfs2_global_release_dquot(dquot);
758	if (status < 0) {
759		mlog_errno(status);
760		goto out_trans;
761	}
762	status = ocfs2_local_release_dquot(handle, dquot);
763	/*
764	 * If we fail here, we cannot do much as global structure is
765	 * already released. So just complain...
766	 */
767	if (status < 0)
768		mlog_errno(status);
769	/*
770	 * Clear dq_off so that we search for the structure in quota file next
771	 * time we acquire it. The structure might be deleted and reallocated
772	 * elsewhere by another node while our dquot structure is on freelist.
773	 */
774	dquot->dq_off = 0;
775	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
776out_trans:
777	ocfs2_commit_trans(osb, handle);
778out_ilock:
779	ocfs2_unlock_global_qf(oinfo, 1);
780out:
781	mutex_unlock(&dquot->dq_lock);
782	if (status)
783		mlog_errno(status);
784	return status;
785}
786
787/*
788 * Read global dquot structure from disk or create it if it does
789 * not exist. Also update use count of the global structure and
790 * create structure in node-local quota file.
791 */
792static int ocfs2_acquire_dquot(struct dquot *dquot)
793{
794	int status = 0, err;
795	int ex = 0;
796	struct super_block *sb = dquot->dq_sb;
797	struct ocfs2_super *osb = OCFS2_SB(sb);
798	int type = dquot->dq_id.type;
799	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
800	struct inode *gqinode = info->dqi_gqinode;
801	int need_alloc = ocfs2_global_qinit_alloc(sb, type);
802	handle_t *handle;
803
804	trace_ocfs2_acquire_dquot(from_kqid(&init_user_ns, dquot->dq_id),
805				  type);
806	mutex_lock(&dquot->dq_lock);
807	/*
808	 * We need an exclusive lock, because we're going to update use count
809	 * and instantiate possibly new dquot structure
810	 */
811	status = ocfs2_lock_global_qf(info, 1);
812	if (status < 0)
813		goto out;
814	status = ocfs2_qinfo_lock(info, 0);
815	if (status < 0)
816		goto out_dq;
817	/*
818	 * We always want to read dquot structure from disk because we don't
819	 * know what happened with it while it was on freelist.
820	 */
821	status = qtree_read_dquot(&info->dqi_gi, dquot);
822	ocfs2_qinfo_unlock(info, 0);
823	if (status < 0)
824		goto out_dq;
825
826	OCFS2_DQUOT(dquot)->dq_use_count++;
827	OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
828	OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
829	if (!dquot->dq_off) {	/* No real quota entry? */
830		ex = 1;
831		/*
832		 * Add blocks to quota file before we start a transaction since
833		 * locking allocators ranks above a transaction start
834		 */
835		WARN_ON(journal_current_handle());
836		status = ocfs2_extend_no_holes(gqinode, NULL,
837			i_size_read(gqinode) + (need_alloc << sb->s_blocksize_bits),
838			i_size_read(gqinode));
839		if (status < 0)
840			goto out_dq;
841	}
842
843	handle = ocfs2_start_trans(osb,
844				   ocfs2_calc_global_qinit_credits(sb, type));
845	if (IS_ERR(handle)) {
846		status = PTR_ERR(handle);
847		goto out_dq;
848	}
849	status = ocfs2_qinfo_lock(info, ex);
850	if (status < 0)
851		goto out_trans;
852	status = qtree_write_dquot(&info->dqi_gi, dquot);
853	if (ex && info_dirty(sb_dqinfo(sb, type))) {
854		err = __ocfs2_global_write_info(sb, type);
855		if (!status)
856			status = err;
857	}
858	ocfs2_qinfo_unlock(info, ex);
859out_trans:
860	ocfs2_commit_trans(osb, handle);
861out_dq:
862	ocfs2_unlock_global_qf(info, 1);
863	if (status < 0)
864		goto out;
865
866	status = ocfs2_create_local_dquot(dquot);
867	if (status < 0)
868		goto out;
869	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
870out:
871	mutex_unlock(&dquot->dq_lock);
872	if (status)
873		mlog_errno(status);
874	return status;
875}
876
877static int ocfs2_get_next_id(struct super_block *sb, struct kqid *qid)
878{
879	int type = qid->type;
880	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
881	int status = 0;
882
883	trace_ocfs2_get_next_id(from_kqid(&init_user_ns, *qid), type);
884	if (!sb_has_quota_loaded(sb, type)) {
885		status = -ESRCH;
886		goto out;
887	}
888	status = ocfs2_lock_global_qf(info, 0);
889	if (status < 0)
890		goto out;
891	status = ocfs2_qinfo_lock(info, 0);
892	if (status < 0)
893		goto out_global;
894	status = qtree_get_next_id(&info->dqi_gi, qid);
895	ocfs2_qinfo_unlock(info, 0);
896out_global:
897	ocfs2_unlock_global_qf(info, 0);
898out:
899	/*
900	 * Avoid logging ENOENT since it just means there isn't next ID and
901	 * ESRCH which means quota isn't enabled for the filesystem.
902	 */
903	if (status && status != -ENOENT && status != -ESRCH)
904		mlog_errno(status);
905	return status;
906}
907
908static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
909{
910	unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
911			     (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
912			     (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
913			     (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
914			     (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
915			     (1 << (DQ_LASTSET_B + QIF_ITIME_B));
916	int sync = 0;
917	int status;
918	struct super_block *sb = dquot->dq_sb;
919	int type = dquot->dq_id.type;
920	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
921	handle_t *handle;
922	struct ocfs2_super *osb = OCFS2_SB(sb);
923
924	trace_ocfs2_mark_dquot_dirty(from_kqid(&init_user_ns, dquot->dq_id),
925				     type);
926
927	/* In case user set some limits, sync dquot immediately to global
928	 * quota file so that information propagates quicker */
929	spin_lock(&dquot->dq_dqb_lock);
930	if (dquot->dq_flags & mask)
931		sync = 1;
932	spin_unlock(&dquot->dq_dqb_lock);
933	/* This is a slight hack but we can't afford getting global quota
934	 * lock if we already have a transaction started. */
935	if (!sync || journal_current_handle()) {
936		status = ocfs2_write_dquot(dquot);
937		goto out;
938	}
939	status = ocfs2_lock_global_qf(oinfo, 1);
940	if (status < 0)
941		goto out;
942	handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
943	if (IS_ERR(handle)) {
944		status = PTR_ERR(handle);
945		mlog_errno(status);
946		goto out_ilock;
947	}
948	down_write(&sb_dqopt(sb)->dqio_sem);
949	status = ocfs2_sync_dquot(dquot);
950	if (status < 0) {
951		mlog_errno(status);
952		goto out_dlock;
953	}
954	/* Now write updated local dquot structure */
955	status = ocfs2_local_write_dquot(dquot);
956out_dlock:
957	up_write(&sb_dqopt(sb)->dqio_sem);
958	ocfs2_commit_trans(osb, handle);
959out_ilock:
960	ocfs2_unlock_global_qf(oinfo, 1);
961out:
962	if (status)
963		mlog_errno(status);
964	return status;
965}
966
967/* This should happen only after set_dqinfo(). */
968static int ocfs2_write_info(struct super_block *sb, int type)
969{
970	handle_t *handle;
971	int status = 0;
972	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
973
974	status = ocfs2_lock_global_qf(oinfo, 1);
975	if (status < 0)
976		goto out;
977	handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
978	if (IS_ERR(handle)) {
979		status = PTR_ERR(handle);
980		mlog_errno(status);
981		goto out_ilock;
982	}
983	status = dquot_commit_info(sb, type);
984	ocfs2_commit_trans(OCFS2_SB(sb), handle);
985out_ilock:
986	ocfs2_unlock_global_qf(oinfo, 1);
987out:
988	if (status)
989		mlog_errno(status);
990	return status;
991}
992
993static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
994{
995	struct ocfs2_dquot *dquot =
996				kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
997
998	if (!dquot)
999		return NULL;
1000	return &dquot->dq_dquot;
1001}
1002
1003static void ocfs2_destroy_dquot(struct dquot *dquot)
1004{
1005	kmem_cache_free(ocfs2_dquot_cachep, dquot);
1006}
1007
1008const struct dquot_operations ocfs2_quota_operations = {
1009	/* We never make dquot dirty so .write_dquot is never called */
1010	.acquire_dquot	= ocfs2_acquire_dquot,
1011	.release_dquot	= ocfs2_release_dquot,
1012	.mark_dirty	= ocfs2_mark_dquot_dirty,
1013	.write_info	= ocfs2_write_info,
1014	.alloc_dquot	= ocfs2_alloc_dquot,
1015	.destroy_dquot	= ocfs2_destroy_dquot,
1016	.get_next_id	= ocfs2_get_next_id,
1017};
1018