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