summaryrefslogtreecommitdiffstats
path: root/fs/xfs/xfs_filestream.c
blob: 8ec81bed7992149420efd5781cfb12c9596633f5 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
/*
 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
 * Copyright (c) 2014 Christoph Hellwig.
 * All Rights Reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */
#include "xfs.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_ag.h"
#include "xfs_sb.h"
#include "xfs_mount.h"
#include "xfs_inum.h"
#include "xfs_inode.h"
#include "xfs_bmap.h"
#include "xfs_bmap_util.h"
#include "xfs_alloc.h"
#include "xfs_mru_cache.h"
#include "xfs_dinode.h"
#include "xfs_filestream.h"
#include "xfs_trace.h"

struct xfs_fstrm_item {
	struct xfs_mru_cache_elem	mru;
	struct xfs_inode		*ip;
	xfs_agnumber_t			ag; /* AG in use for this directory */
};

enum xfs_fstrm_alloc {
	XFS_PICK_USERDATA = 1,
	XFS_PICK_LOWSPACE = 2,
};

/*
 * Allocation group filestream associations are tracked with per-ag atomic
 * counters.  These counters allow xfs_filestream_pick_ag() to tell whether a
 * particular AG already has active filestreams associated with it. The mount
 * point's m_peraglock is used to protect these counters from per-ag array
 * re-allocation during a growfs operation.  When xfs_growfs_data_private() is
 * about to reallocate the array, it calls xfs_filestream_flush() with the
 * m_peraglock held in write mode.
 *
 * Since xfs_mru_cache_flush() guarantees that all the free functions for all
 * the cache elements have finished executing before it returns, it's safe for
 * the free functions to use the atomic counters without m_peraglock protection.
 * This allows the implementation of xfs_fstrm_free_func() to be agnostic about
 * whether it was called with the m_peraglock held in read mode, write mode or
 * not held at all.  The race condition this addresses is the following:
 *
 *  - The work queue scheduler fires and pulls a filestream directory cache
 *    element off the LRU end of the cache for deletion, then gets pre-empted.
 *  - A growfs operation grabs the m_peraglock in write mode, flushes all the
 *    remaining items from the cache and reallocates the mount point's per-ag
 *    array, resetting all the counters to zero.
 *  - The work queue thread resumes and calls the free function for the element
 *    it started cleaning up earlier.  In the process it decrements the
 *    filestreams counter for an AG that now has no references.
 *
 * With a shrinkfs feature, the above scenario could panic the system.
 *
 * All other uses of the following macros should be protected by either the
 * m_peraglock held in read mode, or the cache's internal locking exposed by the
 * interval between a call to xfs_mru_cache_lookup() and a call to
 * xfs_mru_cache_done().  In addition, the m_peraglock must be held in read mode
 * when new elements are added to the cache.
 *
 * Combined, these locking rules ensure that no associations will ever exist in
 * the cache that reference per-ag array elements that have since been
 * reallocated.
 */
int
xfs_filestream_peek_ag(
	xfs_mount_t	*mp,
	xfs_agnumber_t	agno)
{
	struct xfs_perag *pag;
	int		ret;

	pag = xfs_perag_get(mp, agno);
	ret = atomic_read(&pag->pagf_fstrms);
	xfs_perag_put(pag);
	return ret;
}

static int
xfs_filestream_get_ag(
	xfs_mount_t	*mp,
	xfs_agnumber_t	agno)
{
	struct xfs_perag *pag;
	int		ret;

	pag = xfs_perag_get(mp, agno);
	ret = atomic_inc_return(&pag->pagf_fstrms);
	xfs_perag_put(pag);
	return ret;
}

static void
xfs_filestream_put_ag(
	xfs_mount_t	*mp,
	xfs_agnumber_t	agno)
{
	struct xfs_perag *pag;

	pag = xfs_perag_get(mp, agno);
	atomic_dec(&pag->pagf_fstrms);
	xfs_perag_put(pag);
}

static void
xfs_fstrm_free_func(
	struct xfs_mru_cache_elem *mru)
{
	struct xfs_fstrm_item	*item =
		container_of(mru, struct xfs_fstrm_item, mru);

	xfs_filestream_put_ag(item->ip->i_mount, item->ag);

	trace_xfs_filestream_free(item->ip, item->ag);

	kmem_free(item);
}

/*
 * Scan the AGs starting at startag looking for an AG that isn't in use and has
 * at least minlen blocks free.
 */
static int
xfs_filestream_pick_ag(
	struct xfs_inode	*ip,
	xfs_agnumber_t		startag,
	xfs_agnumber_t		*agp,
	int			flags,
	xfs_extlen_t		minlen)
{
	struct xfs_mount	*mp = ip->i_mount;
	struct xfs_fstrm_item	*item;
	struct xfs_perag	*pag;
	xfs_extlen_t		longest, free = 0, minfree, maxfree = 0;
	xfs_agnumber_t		ag, max_ag = NULLAGNUMBER;
	int			err, trylock, nscan;

	ASSERT(S_ISDIR(ip->i_d.di_mode));

	/* 2% of an AG's blocks must be free for it to be chosen. */
	minfree = mp->m_sb.sb_agblocks / 50;

	ag = startag;
	*agp = NULLAGNUMBER;

	/* For the first pass, don't sleep trying to init the per-AG. */
	trylock = XFS_ALLOC_FLAG_TRYLOCK;

	for (nscan = 0; 1; nscan++) {
		trace_xfs_filestream_scan(ip, ag);

		pag = xfs_perag_get(mp, ag);

		if (!pag->pagf_init) {
			err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
			if (err && !trylock) {
				xfs_perag_put(pag);
				return err;
			}
		}

		/* Might fail sometimes during the 1st pass with trylock set. */
		if (!pag->pagf_init)
			goto next_ag;

		/* Keep track of the AG with the most free blocks. */
		if (pag->pagf_freeblks > maxfree) {
			maxfree = pag->pagf_freeblks;
			max_ag = ag;
		}

		/*
		 * The AG reference count does two things: it enforces mutual
		 * exclusion when examining the suitability of an AG in this
		 * loop, and it guards against two filestreams being established
		 * in the same AG as each other.
		 */
		if (xfs_filestream_get_ag(mp, ag) > 1) {
			xfs_filestream_put_ag(mp, ag);
			goto next_ag;
		}

		longest = xfs_alloc_longest_free_extent(mp, pag);
		if (((minlen && longest >= minlen) ||
		     (!minlen && pag->pagf_freeblks >= minfree)) &&
		    (!pag->pagf_metadata || !(flags & XFS_PICK_USERDATA) ||
		     (flags & XFS_PICK_LOWSPACE))) {

			/* Break out, retaining the reference on the AG. */
			free = pag->pagf_freeblks;
			xfs_perag_put(pag);
			*agp = ag;
			break;
		}

		/* Drop the reference on this AG, it's not usable. */
		xfs_filestream_put_ag(mp, ag);
next_ag:
		xfs_perag_put(pag);
		/* Move to the next AG, wrapping to AG 0 if necessary. */
		if (++ag >= mp->m_sb.sb_agcount)
			ag = 0;

		/* If a full pass of the AGs hasn't been done yet, continue. */
		if (ag != startag)
			continue;

		/* Allow sleeping in xfs_alloc_pagf_init() on the 2nd pass. */
		if (trylock != 0) {
			trylock = 0;
			continue;
		}

		/* Finally, if lowspace wasn't set, set it for the 3rd pass. */
		if (!(flags & XFS_PICK_LOWSPACE)) {
			flags |= XFS_PICK_LOWSPACE;
			continue;
		}

		/*
		 * Take the AG with the most free space, regardless of whether
		 * it's already in use by another filestream.
		 */
		if (max_ag != NULLAGNUMBER) {
			xfs_filestream_get_ag(mp, max_ag);
			free = maxfree;
			*agp = max_ag;
			break;
		}

		/* take AG 0 if none matched */
		trace_xfs_filestream_pick(ip, *agp, free, nscan);
		*agp = 0;
		return 0;
	}

	trace_xfs_filestream_pick(ip, *agp, free, nscan);

	if (*agp == NULLAGNUMBER)
		return 0;

	err = ENOMEM;
	item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
	if (!item)
		goto out_put_ag;

	item->ag = *agp;
	item->ip = ip;

	err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);
	if (err) {
		if (err == EEXIST)
			err = 0;
		goto out_free_item;
	}

	return 0;

out_free_item:
	kmem_free(item);
out_put_ag:
	xfs_filestream_put_ag(mp, *agp);
	return err;
}

static struct xfs_inode *
xfs_filestream_get_parent(
	struct xfs_inode	*ip)
{
	struct inode		*inode = VFS_I(ip), *dir = NULL;
	struct dentry		*dentry, *parent;

	dentry = d_find_alias(inode);
	if (!dentry)
		goto out;

	parent = dget_parent(dentry);
	if (!parent)
		goto out_dput;

	dir = igrab(parent->d_inode);
	dput(parent);

out_dput:
	dput(dentry);
out:
	return dir ? XFS_I(dir) : NULL;
}

/*
 * Find the right allocation group for a file, either by finding an
 * existing file stream or creating a new one.
 *
 * Returns NULLAGNUMBER in case of an error.
 */
xfs_agnumber_t
xfs_filestream_lookup_ag(
	struct xfs_inode	*ip)
{
	struct xfs_mount	*mp = ip->i_mount;
	struct xfs_inode	*pip = NULL;
	xfs_agnumber_t		startag, ag = NULLAGNUMBER;
	struct xfs_mru_cache_elem *mru;

	ASSERT(S_ISREG(ip->i_d.di_mode));

	pip = xfs_filestream_get_parent(ip);
	if (!pip)
		goto out;

	mru = xfs_mru_cache_lookup(mp->m_filestream, pip->i_ino);
	if (mru) {
		ag = container_of(mru, struct xfs_fstrm_item, mru)->ag;
		xfs_mru_cache_done(mp->m_filestream);

		trace_xfs_filestream_lookup(ip, ag);
		goto out;
	}

	/*
	 * Set the starting AG using the rotor for inode32, otherwise
	 * use the directory inode's AG.
	 */
	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
		xfs_agnumber_t	 rotorstep = xfs_rotorstep;
		startag = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
		mp->m_agfrotor = (mp->m_agfrotor + 1) %
		                 (mp->m_sb.sb_agcount * rotorstep);
	} else
		startag = XFS_INO_TO_AGNO(mp, pip->i_ino);

	if (xfs_filestream_pick_ag(pip, startag, &ag, 0, 0))
		ag = NULLAGNUMBER;
out:
	IRELE(pip);
	return ag;
}

/*
 * Pick a new allocation group for the current file and its file stream.
 *
 * This is called when the allocator can't find a suitable extent in the
 * current AG, and we have to move the stream into a new AG with more space.
 */
int
xfs_filestream_new_ag(
	struct xfs_bmalloca	*ap,
	xfs_agnumber_t		*agp)
{
	struct xfs_inode	*ip = ap->ip, *pip;
	struct xfs_mount	*mp = ip->i_mount;
	xfs_extlen_t		minlen = ap->length;
	xfs_agnumber_t		startag = 0;
	int			flags, err = 0;
	struct xfs_mru_cache_elem *mru;

	*agp = NULLAGNUMBER;

	pip = xfs_filestream_get_parent(ip);
	if (!pip)
		goto exit;

	mru = xfs_mru_cache_remove(mp->m_filestream, pip->i_ino);
	if (mru) {
		struct xfs_fstrm_item *item =
			container_of(mru, struct xfs_fstrm_item, mru);
		startag = (item->ag + 1) % mp->m_sb.sb_agcount;
	}

	flags = (ap->userdata ? XFS_PICK_USERDATA : 0) |
	        (ap->flist->xbf_low ? XFS_PICK_LOWSPACE : 0);

	err = xfs_filestream_pick_ag(pip, startag, agp, flags, minlen);

	/*
	 * Only free the item here so we skip over the old AG earlier.
	 */
	if (mru)
		xfs_fstrm_free_func(mru);

	IRELE(pip);
exit:
	if (*agp == NULLAGNUMBER)
		*agp = 0;
	return err;
}

void
xfs_filestream_deassociate(
	struct xfs_inode	*ip)
{
	xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino);
}

int
xfs_filestream_mount(
	xfs_mount_t	*mp)
{
	/*
	 * The filestream timer tunable is currently fixed within the range of
	 * one second to four minutes, with five seconds being the default.  The
	 * group count is somewhat arbitrary, but it'd be nice to adhere to the
	 * timer tunable to within about 10 percent.  This requires at least 10
	 * groups.
	 */
	return xfs_mru_cache_create(&mp->m_filestream, xfs_fstrm_centisecs * 10,
				    10, xfs_fstrm_free_func);
}

void
xfs_filestream_unmount(
	xfs_mount_t	*mp)
{
	xfs_mru_cache_destroy(mp->m_filestream);
}