1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
4 * Copyright (c) 2014 Christoph Hellwig.
5 * All Rights Reserved.
6 */
7#include "xfs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_sb.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_bmap.h"
16#include "xfs_alloc.h"
17#include "xfs_mru_cache.h"
18#include "xfs_trace.h"
19#include "xfs_ag_resv.h"
20#include "xfs_trans.h"
21#include "xfs_filestream.h"
22
23struct xfs_fstrm_item {
24	struct xfs_mru_cache_elem	mru;
25	xfs_agnumber_t			ag; /* AG in use for this directory */
26};
27
28enum xfs_fstrm_alloc {
29	XFS_PICK_USERDATA = 1,
30	XFS_PICK_LOWSPACE = 2,
31};
32
33/*
34 * Allocation group filestream associations are tracked with per-ag atomic
35 * counters.  These counters allow xfs_filestream_pick_ag() to tell whether a
36 * particular AG already has active filestreams associated with it.
37 */
38int
39xfs_filestream_peek_ag(
40	xfs_mount_t	*mp,
41	xfs_agnumber_t	agno)
42{
43	struct xfs_perag *pag;
44	int		ret;
45
46	pag = xfs_perag_get(mp, agno);
47	ret = atomic_read(&pag->pagf_fstrms);
48	xfs_perag_put(pag);
49	return ret;
50}
51
52static int
53xfs_filestream_get_ag(
54	xfs_mount_t	*mp,
55	xfs_agnumber_t	agno)
56{
57	struct xfs_perag *pag;
58	int		ret;
59
60	pag = xfs_perag_get(mp, agno);
61	ret = atomic_inc_return(&pag->pagf_fstrms);
62	xfs_perag_put(pag);
63	return ret;
64}
65
66static void
67xfs_filestream_put_ag(
68	xfs_mount_t	*mp,
69	xfs_agnumber_t	agno)
70{
71	struct xfs_perag *pag;
72
73	pag = xfs_perag_get(mp, agno);
74	atomic_dec(&pag->pagf_fstrms);
75	xfs_perag_put(pag);
76}
77
78static void
79xfs_fstrm_free_func(
80	void			*data,
81	struct xfs_mru_cache_elem *mru)
82{
83	struct xfs_mount	*mp = data;
84	struct xfs_fstrm_item	*item =
85		container_of(mru, struct xfs_fstrm_item, mru);
86
87	xfs_filestream_put_ag(mp, item->ag);
88	trace_xfs_filestream_free(mp, mru->key, item->ag);
89
90	kmem_free(item);
91}
92
93/*
94 * Scan the AGs starting at startag looking for an AG that isn't in use and has
95 * at least minlen blocks free.
96 */
97static int
98xfs_filestream_pick_ag(
99	struct xfs_inode	*ip,
100	xfs_agnumber_t		startag,
101	xfs_agnumber_t		*agp,
102	int			flags,
103	xfs_extlen_t		minlen)
104{
105	struct xfs_mount	*mp = ip->i_mount;
106	struct xfs_fstrm_item	*item;
107	struct xfs_perag	*pag;
108	xfs_extlen_t		longest, free = 0, minfree, maxfree = 0;
109	xfs_agnumber_t		ag, max_ag = NULLAGNUMBER;
110	int			err, trylock, nscan;
111
112	ASSERT(S_ISDIR(VFS_I(ip)->i_mode));
113
114	/* 2% of an AG's blocks must be free for it to be chosen. */
115	minfree = mp->m_sb.sb_agblocks / 50;
116
117	ag = startag;
118	*agp = NULLAGNUMBER;
119
120	/* For the first pass, don't sleep trying to init the per-AG. */
121	trylock = XFS_ALLOC_FLAG_TRYLOCK;
122
123	for (nscan = 0; 1; nscan++) {
124		trace_xfs_filestream_scan(mp, ip->i_ino, ag);
125
126		pag = xfs_perag_get(mp, ag);
127
128		if (!pag->pagf_init) {
129			err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
130			if (err) {
131				if (err != -EAGAIN) {
132					xfs_perag_put(pag);
133					return err;
134				}
135				/* Couldn't lock the AGF, skip this AG. */
136				goto next_ag;
137			}
138		}
139
140		/* Keep track of the AG with the most free blocks. */
141		if (pag->pagf_freeblks > maxfree) {
142			maxfree = pag->pagf_freeblks;
143			max_ag = ag;
144		}
145
146		/*
147		 * The AG reference count does two things: it enforces mutual
148		 * exclusion when examining the suitability of an AG in this
149		 * loop, and it guards against two filestreams being established
150		 * in the same AG as each other.
151		 */
152		if (xfs_filestream_get_ag(mp, ag) > 1) {
153			xfs_filestream_put_ag(mp, ag);
154			goto next_ag;
155		}
156
157		longest = xfs_alloc_longest_free_extent(pag,
158				xfs_alloc_min_freelist(mp, pag),
159				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
160		if (((minlen && longest >= minlen) ||
161		     (!minlen && pag->pagf_freeblks >= minfree)) &&
162		    (!pag->pagf_metadata || !(flags & XFS_PICK_USERDATA) ||
163		     (flags & XFS_PICK_LOWSPACE))) {
164
165			/* Break out, retaining the reference on the AG. */
166			free = pag->pagf_freeblks;
167			xfs_perag_put(pag);
168			*agp = ag;
169			break;
170		}
171
172		/* Drop the reference on this AG, it's not usable. */
173		xfs_filestream_put_ag(mp, ag);
174next_ag:
175		xfs_perag_put(pag);
176		/* Move to the next AG, wrapping to AG 0 if necessary. */
177		if (++ag >= mp->m_sb.sb_agcount)
178			ag = 0;
179
180		/* If a full pass of the AGs hasn't been done yet, continue. */
181		if (ag != startag)
182			continue;
183
184		/* Allow sleeping in xfs_alloc_pagf_init() on the 2nd pass. */
185		if (trylock != 0) {
186			trylock = 0;
187			continue;
188		}
189
190		/* Finally, if lowspace wasn't set, set it for the 3rd pass. */
191		if (!(flags & XFS_PICK_LOWSPACE)) {
192			flags |= XFS_PICK_LOWSPACE;
193			continue;
194		}
195
196		/*
197		 * Take the AG with the most free space, regardless of whether
198		 * it's already in use by another filestream.
199		 */
200		if (max_ag != NULLAGNUMBER) {
201			xfs_filestream_get_ag(mp, max_ag);
202			free = maxfree;
203			*agp = max_ag;
204			break;
205		}
206
207		/* take AG 0 if none matched */
208		trace_xfs_filestream_pick(ip, *agp, free, nscan);
209		*agp = 0;
210		return 0;
211	}
212
213	trace_xfs_filestream_pick(ip, *agp, free, nscan);
214
215	if (*agp == NULLAGNUMBER)
216		return 0;
217
218	err = -ENOMEM;
219	item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
220	if (!item)
221		goto out_put_ag;
222
223	item->ag = *agp;
224
225	err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);
226	if (err) {
227		if (err == -EEXIST)
228			err = 0;
229		goto out_free_item;
230	}
231
232	return 0;
233
234out_free_item:
235	kmem_free(item);
236out_put_ag:
237	xfs_filestream_put_ag(mp, *agp);
238	return err;
239}
240
241static struct xfs_inode *
242xfs_filestream_get_parent(
243	struct xfs_inode	*ip)
244{
245	struct inode		*inode = VFS_I(ip), *dir = NULL;
246	struct dentry		*dentry, *parent;
247
248	dentry = d_find_alias(inode);
249	if (!dentry)
250		goto out;
251
252	parent = dget_parent(dentry);
253	if (!parent)
254		goto out_dput;
255
256	dir = igrab(d_inode(parent));
257	dput(parent);
258
259out_dput:
260	dput(dentry);
261out:
262	return dir ? XFS_I(dir) : NULL;
263}
264
265/*
266 * Find the right allocation group for a file, either by finding an
267 * existing file stream or creating a new one.
268 *
269 * Returns NULLAGNUMBER in case of an error.
270 */
271xfs_agnumber_t
272xfs_filestream_lookup_ag(
273	struct xfs_inode	*ip)
274{
275	struct xfs_mount	*mp = ip->i_mount;
276	struct xfs_inode	*pip = NULL;
277	xfs_agnumber_t		startag, ag = NULLAGNUMBER;
278	struct xfs_mru_cache_elem *mru;
279
280	ASSERT(S_ISREG(VFS_I(ip)->i_mode));
281
282	pip = xfs_filestream_get_parent(ip);
283	if (!pip)
284		return NULLAGNUMBER;
285
286	mru = xfs_mru_cache_lookup(mp->m_filestream, pip->i_ino);
287	if (mru) {
288		ag = container_of(mru, struct xfs_fstrm_item, mru)->ag;
289		xfs_mru_cache_done(mp->m_filestream);
290
291		trace_xfs_filestream_lookup(mp, ip->i_ino, ag);
292		goto out;
293	}
294
295	/*
296	 * Set the starting AG using the rotor for inode32, otherwise
297	 * use the directory inode's AG.
298	 */
299	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
300		xfs_agnumber_t	 rotorstep = xfs_rotorstep;
301		startag = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
302		mp->m_agfrotor = (mp->m_agfrotor + 1) %
303		                 (mp->m_sb.sb_agcount * rotorstep);
304	} else
305		startag = XFS_INO_TO_AGNO(mp, pip->i_ino);
306
307	if (xfs_filestream_pick_ag(pip, startag, &ag, 0, 0))
308		ag = NULLAGNUMBER;
309out:
310	xfs_irele(pip);
311	return ag;
312}
313
314/*
315 * Pick a new allocation group for the current file and its file stream.
316 *
317 * This is called when the allocator can't find a suitable extent in the
318 * current AG, and we have to move the stream into a new AG with more space.
319 */
320int
321xfs_filestream_new_ag(
322	struct xfs_bmalloca	*ap,
323	xfs_agnumber_t		*agp)
324{
325	struct xfs_inode	*ip = ap->ip, *pip;
326	struct xfs_mount	*mp = ip->i_mount;
327	xfs_extlen_t		minlen = ap->length;
328	xfs_agnumber_t		startag = 0;
329	int			flags = 0;
330	int			err = 0;
331	struct xfs_mru_cache_elem *mru;
332
333	*agp = NULLAGNUMBER;
334
335	pip = xfs_filestream_get_parent(ip);
336	if (!pip)
337		goto exit;
338
339	mru = xfs_mru_cache_remove(mp->m_filestream, pip->i_ino);
340	if (mru) {
341		struct xfs_fstrm_item *item =
342			container_of(mru, struct xfs_fstrm_item, mru);
343		startag = (item->ag + 1) % mp->m_sb.sb_agcount;
344	}
345
346	if (ap->datatype & XFS_ALLOC_USERDATA)
347		flags |= XFS_PICK_USERDATA;
348	if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
349		flags |= XFS_PICK_LOWSPACE;
350
351	err = xfs_filestream_pick_ag(pip, startag, agp, flags, minlen);
352
353	/*
354	 * Only free the item here so we skip over the old AG earlier.
355	 */
356	if (mru)
357		xfs_fstrm_free_func(mp, mru);
358
359	xfs_irele(pip);
360exit:
361	if (*agp == NULLAGNUMBER)
362		*agp = 0;
363	return err;
364}
365
366void
367xfs_filestream_deassociate(
368	struct xfs_inode	*ip)
369{
370	xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino);
371}
372
373int
374xfs_filestream_mount(
375	xfs_mount_t	*mp)
376{
377	/*
378	 * The filestream timer tunable is currently fixed within the range of
379	 * one second to four minutes, with five seconds being the default.  The
380	 * group count is somewhat arbitrary, but it'd be nice to adhere to the
381	 * timer tunable to within about 10 percent.  This requires at least 10
382	 * groups.
383	 */
384	return xfs_mru_cache_create(&mp->m_filestream, mp,
385			xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);
386}
387
388void
389xfs_filestream_unmount(
390	xfs_mount_t	*mp)
391{
392	xfs_mru_cache_destroy(mp->m_filestream);
393}
394