Linux Audio

Check our new training course

Loading...
v4.17
 
  1/*
  2 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
  3 * Copyright (c) 2014 Christoph Hellwig.
  4 * All Rights Reserved.
  5 *
  6 * This program is free software; you can redistribute it and/or
  7 * modify it under the terms of the GNU General Public License as
  8 * published by the Free Software Foundation.
  9 *
 10 * This program is distributed in the hope that it would be useful,
 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 13 * GNU General Public License for more details.
 14 *
 15 * You should have received a copy of the GNU General Public License
 16 * along with this program; if not, write the Free Software Foundation,
 17 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 18 */
 19#include "xfs.h"
 
 20#include "xfs_format.h"
 21#include "xfs_log_format.h"
 22#include "xfs_trans_resv.h"
 23#include "xfs_sb.h"
 24#include "xfs_mount.h"
 25#include "xfs_defer.h"
 26#include "xfs_inode.h"
 27#include "xfs_bmap.h"
 28#include "xfs_bmap_util.h"
 29#include "xfs_alloc.h"
 30#include "xfs_mru_cache.h"
 31#include "xfs_filestream.h"
 32#include "xfs_trace.h"
 
 33#include "xfs_ag_resv.h"
 
 
 34
 35struct xfs_fstrm_item {
 36	struct xfs_mru_cache_elem	mru;
 37	xfs_agnumber_t			ag; /* AG in use for this directory */
 38};
 39
 40enum xfs_fstrm_alloc {
 41	XFS_PICK_USERDATA = 1,
 42	XFS_PICK_LOWSPACE = 2,
 43};
 44
 45/*
 46 * Allocation group filestream associations are tracked with per-ag atomic
 47 * counters.  These counters allow xfs_filestream_pick_ag() to tell whether a
 48 * particular AG already has active filestreams associated with it. The mount
 49 * point's m_peraglock is used to protect these counters from per-ag array
 50 * re-allocation during a growfs operation.  When xfs_growfs_data_private() is
 51 * about to reallocate the array, it calls xfs_filestream_flush() with the
 52 * m_peraglock held in write mode.
 53 *
 54 * Since xfs_mru_cache_flush() guarantees that all the free functions for all
 55 * the cache elements have finished executing before it returns, it's safe for
 56 * the free functions to use the atomic counters without m_peraglock protection.
 57 * This allows the implementation of xfs_fstrm_free_func() to be agnostic about
 58 * whether it was called with the m_peraglock held in read mode, write mode or
 59 * not held at all.  The race condition this addresses is the following:
 60 *
 61 *  - The work queue scheduler fires and pulls a filestream directory cache
 62 *    element off the LRU end of the cache for deletion, then gets pre-empted.
 63 *  - A growfs operation grabs the m_peraglock in write mode, flushes all the
 64 *    remaining items from the cache and reallocates the mount point's per-ag
 65 *    array, resetting all the counters to zero.
 66 *  - The work queue thread resumes and calls the free function for the element
 67 *    it started cleaning up earlier.  In the process it decrements the
 68 *    filestreams counter for an AG that now has no references.
 69 *
 70 * With a shrinkfs feature, the above scenario could panic the system.
 71 *
 72 * All other uses of the following macros should be protected by either the
 73 * m_peraglock held in read mode, or the cache's internal locking exposed by the
 74 * interval between a call to xfs_mru_cache_lookup() and a call to
 75 * xfs_mru_cache_done().  In addition, the m_peraglock must be held in read mode
 76 * when new elements are added to the cache.
 77 *
 78 * Combined, these locking rules ensure that no associations will ever exist in
 79 * the cache that reference per-ag array elements that have since been
 80 * reallocated.
 81 */
 82int
 83xfs_filestream_peek_ag(
 84	xfs_mount_t	*mp,
 85	xfs_agnumber_t	agno)
 86{
 87	struct xfs_perag *pag;
 88	int		ret;
 89
 90	pag = xfs_perag_get(mp, agno);
 91	ret = atomic_read(&pag->pagf_fstrms);
 92	xfs_perag_put(pag);
 93	return ret;
 94}
 95
 96static int
 97xfs_filestream_get_ag(
 98	xfs_mount_t	*mp,
 99	xfs_agnumber_t	agno)
100{
101	struct xfs_perag *pag;
102	int		ret;
103
104	pag = xfs_perag_get(mp, agno);
105	ret = atomic_inc_return(&pag->pagf_fstrms);
106	xfs_perag_put(pag);
107	return ret;
108}
109
110static void
111xfs_filestream_put_ag(
112	xfs_mount_t	*mp,
113	xfs_agnumber_t	agno)
114{
115	struct xfs_perag *pag;
116
117	pag = xfs_perag_get(mp, agno);
118	atomic_dec(&pag->pagf_fstrms);
119	xfs_perag_put(pag);
120}
121
122static void
123xfs_fstrm_free_func(
124	void			*data,
125	struct xfs_mru_cache_elem *mru)
126{
127	struct xfs_mount	*mp = data;
128	struct xfs_fstrm_item	*item =
129		container_of(mru, struct xfs_fstrm_item, mru);
 
130
131	xfs_filestream_put_ag(mp, item->ag);
132	trace_xfs_filestream_free(mp, mru->key, item->ag);
 
133
134	kmem_free(item);
135}
136
137/*
138 * Scan the AGs starting at startag looking for an AG that isn't in use and has
139 * at least minlen blocks free.
 
140 */
141static int
142xfs_filestream_pick_ag(
143	struct xfs_inode	*ip,
144	xfs_agnumber_t		startag,
145	xfs_agnumber_t		*agp,
146	int			flags,
147	xfs_extlen_t		minlen)
148{
149	struct xfs_mount	*mp = ip->i_mount;
150	struct xfs_fstrm_item	*item;
151	struct xfs_perag	*pag;
152	xfs_extlen_t		longest, free = 0, minfree, maxfree = 0;
153	xfs_agnumber_t		ag, max_ag = NULLAGNUMBER;
154	int			err, trylock, nscan;
155
156	ASSERT(S_ISDIR(VFS_I(ip)->i_mode));
157
158	/* 2% of an AG's blocks must be free for it to be chosen. */
159	minfree = mp->m_sb.sb_agblocks / 50;
160
161	ag = startag;
162	*agp = NULLAGNUMBER;
163
164	/* For the first pass, don't sleep trying to init the per-AG. */
165	trylock = XFS_ALLOC_FLAG_TRYLOCK;
166
167	for (nscan = 0; 1; nscan++) {
168		trace_xfs_filestream_scan(mp, ip->i_ino, ag);
169
170		pag = xfs_perag_get(mp, ag);
171
172		if (!pag->pagf_init) {
173			err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
174			if (err && !trylock) {
175				xfs_perag_put(pag);
176				return err;
177			}
 
 
 
 
178		}
179
180		/* Might fail sometimes during the 1st pass with trylock set. */
181		if (!pag->pagf_init)
182			goto next_ag;
183
184		/* Keep track of the AG with the most free blocks. */
185		if (pag->pagf_freeblks > maxfree) {
186			maxfree = pag->pagf_freeblks;
187			max_ag = ag;
 
 
 
188		}
189
190		/*
191		 * The AG reference count does two things: it enforces mutual
192		 * exclusion when examining the suitability of an AG in this
193		 * loop, and it guards against two filestreams being established
194		 * in the same AG as each other.
195		 */
196		if (xfs_filestream_get_ag(mp, ag) > 1) {
197			xfs_filestream_put_ag(mp, ag);
198			goto next_ag;
199		}
200
201		longest = xfs_alloc_longest_free_extent(pag,
202				xfs_alloc_min_freelist(mp, pag),
203				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
204		if (((minlen && longest >= minlen) ||
205		     (!minlen && pag->pagf_freeblks >= minfree)) &&
206		    (!pag->pagf_metadata || !(flags & XFS_PICK_USERDATA) ||
207		     (flags & XFS_PICK_LOWSPACE))) {
208
209			/* Break out, retaining the reference on the AG. */
210			free = pag->pagf_freeblks;
211			xfs_perag_put(pag);
212			*agp = ag;
213			break;
214		}
215
216		/* Drop the reference on this AG, it's not usable. */
217		xfs_filestream_put_ag(mp, ag);
218next_ag:
219		xfs_perag_put(pag);
220		/* Move to the next AG, wrapping to AG 0 if necessary. */
221		if (++ag >= mp->m_sb.sb_agcount)
222			ag = 0;
223
224		/* If a full pass of the AGs hasn't been done yet, continue. */
225		if (ag != startag)
226			continue;
227
228		/* Allow sleeping in xfs_alloc_pagf_init() on the 2nd pass. */
229		if (trylock != 0) {
230			trylock = 0;
231			continue;
232		}
233
234		/* Finally, if lowspace wasn't set, set it for the 3rd pass. */
235		if (!(flags & XFS_PICK_LOWSPACE)) {
236			flags |= XFS_PICK_LOWSPACE;
237			continue;
238		}
239
240		/*
241		 * Take the AG with the most free space, regardless of whether
242		 * it's already in use by another filestream.
243		 */
244		if (max_ag != NULLAGNUMBER) {
245			xfs_filestream_get_ag(mp, max_ag);
246			free = maxfree;
247			*agp = max_ag;
248			break;
249		}
250
251		/* take AG 0 if none matched */
252		trace_xfs_filestream_pick(ip, *agp, free, nscan);
253		*agp = 0;
254		return 0;
255	}
256
257	trace_xfs_filestream_pick(ip, *agp, free, nscan);
258
259	if (*agp == NULLAGNUMBER)
260		return 0;
 
 
 
 
 
261
262	err = -ENOMEM;
263	item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
264	if (!item)
265		goto out_put_ag;
 
 
 
 
266
267	item->ag = *agp;
 
 
 
 
 
 
 
 
 
268
269	err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);
270	if (err) {
271		if (err == -EEXIST)
272			err = 0;
273		goto out_free_item;
274	}
275
 
 
 
 
 
276	return 0;
277
278out_free_item:
279	kmem_free(item);
280out_put_ag:
281	xfs_filestream_put_ag(mp, *agp);
282	return err;
283}
284
285static struct xfs_inode *
286xfs_filestream_get_parent(
287	struct xfs_inode	*ip)
288{
289	struct inode		*inode = VFS_I(ip), *dir = NULL;
290	struct dentry		*dentry, *parent;
291
292	dentry = d_find_alias(inode);
293	if (!dentry)
294		goto out;
295
296	parent = dget_parent(dentry);
297	if (!parent)
298		goto out_dput;
299
300	dir = igrab(d_inode(parent));
301	dput(parent);
302
303out_dput:
304	dput(dentry);
305out:
306	return dir ? XFS_I(dir) : NULL;
307}
308
309/*
310 * Find the right allocation group for a file, either by finding an
311 * existing file stream or creating a new one.
 
312 *
313 * Returns NULLAGNUMBER in case of an error.
 
 
314 */
315xfs_agnumber_t
316xfs_filestream_lookup_ag(
317	struct xfs_inode	*ip)
 
 
 
318{
319	struct xfs_mount	*mp = ip->i_mount;
320	struct xfs_inode	*pip = NULL;
321	xfs_agnumber_t		startag, ag = NULLAGNUMBER;
322	struct xfs_mru_cache_elem *mru;
 
323
324	ASSERT(S_ISREG(VFS_I(ip)->i_mode));
325
326	pip = xfs_filestream_get_parent(ip);
327	if (!pip)
328		return NULLAGNUMBER;
 
 
 
 
 
 
 
 
329
330	mru = xfs_mru_cache_lookup(mp->m_filestream, pip->i_ino);
331	if (mru) {
332		ag = container_of(mru, struct xfs_fstrm_item, mru)->ag;
333		xfs_mru_cache_done(mp->m_filestream);
334
335		trace_xfs_filestream_lookup(mp, ip->i_ino, ag);
336		goto out;
337	}
338
339	/*
340	 * Set the starting AG using the rotor for inode32, otherwise
341	 * use the directory inode's AG.
 
342	 */
343	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
344		xfs_agnumber_t	 rotorstep = xfs_rotorstep;
345		startag = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
346		mp->m_agfrotor = (mp->m_agfrotor + 1) %
347		                 (mp->m_sb.sb_agcount * rotorstep);
348	} else
349		startag = XFS_INO_TO_AGNO(mp, pip->i_ino);
350
351	if (xfs_filestream_pick_ag(pip, startag, &ag, 0, 0))
352		ag = NULLAGNUMBER;
353out:
354	IRELE(pip);
355	return ag;
 
 
 
 
 
 
 
 
356}
357
358/*
359 * Pick a new allocation group for the current file and its file stream.
360 *
361 * This is called when the allocator can't find a suitable extent in the
362 * current AG, and we have to move the stream into a new AG with more space.
363 */
364int
365xfs_filestream_new_ag(
366	struct xfs_bmalloca	*ap,
367	xfs_agnumber_t		*agp)
 
 
368{
369	struct xfs_inode	*ip = ap->ip, *pip;
370	struct xfs_mount	*mp = ip->i_mount;
371	xfs_extlen_t		minlen = ap->length;
372	xfs_agnumber_t		startag = 0;
373	int			flags = 0;
374	int			err = 0;
375	struct xfs_mru_cache_elem *mru;
 
 
 
 
376
377	*agp = NULLAGNUMBER;
378
379	pip = xfs_filestream_get_parent(ip);
380	if (!pip)
381		goto exit;
382
383	mru = xfs_mru_cache_remove(mp->m_filestream, pip->i_ino);
384	if (mru) {
385		struct xfs_fstrm_item *item =
386			container_of(mru, struct xfs_fstrm_item, mru);
387		startag = (item->ag + 1) % mp->m_sb.sb_agcount;
 
 
 
 
 
 
 
 
388	}
389
390	if (xfs_alloc_is_userdata(ap->datatype))
 
 
 
391		flags |= XFS_PICK_USERDATA;
392	if (ap->dfops->dop_low)
393		flags |= XFS_PICK_LOWSPACE;
394
395	err = xfs_filestream_pick_ag(pip, startag, agp, flags, minlen);
 
 
 
396
397	/*
398	 * Only free the item here so we skip over the old AG earlier.
 
 
 
 
 
 
 
 
 
399	 */
400	if (mru)
401		xfs_fstrm_free_func(mp, mru);
 
 
 
 
 
 
 
 
402
403	IRELE(pip);
404exit:
405	if (*agp == NULLAGNUMBER)
406		*agp = 0;
407	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
408}
409
410void
411xfs_filestream_deassociate(
412	struct xfs_inode	*ip)
413{
414	xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino);
415}
416
417int
418xfs_filestream_mount(
419	xfs_mount_t	*mp)
420{
421	/*
422	 * The filestream timer tunable is currently fixed within the range of
423	 * one second to four minutes, with five seconds being the default.  The
424	 * group count is somewhat arbitrary, but it'd be nice to adhere to the
425	 * timer tunable to within about 10 percent.  This requires at least 10
426	 * groups.
427	 */
428	return xfs_mru_cache_create(&mp->m_filestream, mp,
429			xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);
430}
431
432void
433xfs_filestream_unmount(
434	xfs_mount_t	*mp)
435{
436	xfs_mru_cache_destroy(mp->m_filestream);
437}
v6.13.7
  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_mount.h"
 
 13#include "xfs_inode.h"
 14#include "xfs_bmap.h"
 15#include "xfs_bmap_util.h"
 16#include "xfs_alloc.h"
 17#include "xfs_mru_cache.h"
 
 18#include "xfs_trace.h"
 19#include "xfs_ag.h"
 20#include "xfs_ag_resv.h"
 21#include "xfs_trans.h"
 22#include "xfs_filestream.h"
 23
 24struct xfs_fstrm_item {
 25	struct xfs_mru_cache_elem	mru;
 26	struct xfs_perag		*pag; /* AG in use for this directory */
 27};
 28
 29enum xfs_fstrm_alloc {
 30	XFS_PICK_USERDATA = 1,
 31	XFS_PICK_LOWSPACE = 2,
 32};
 33
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 34static void
 35xfs_fstrm_free_func(
 36	void			*data,
 37	struct xfs_mru_cache_elem *mru)
 38{
 
 39	struct xfs_fstrm_item	*item =
 40		container_of(mru, struct xfs_fstrm_item, mru);
 41	struct xfs_perag	*pag = item->pag;
 42
 43	trace_xfs_filestream_free(pag, mru->key);
 44	atomic_dec(&pag->pagf_fstrms);
 45	xfs_perag_rele(pag);
 46
 47	kfree(item);
 48}
 49
 50/*
 51 * Scan the AGs starting at start_agno looking for an AG that isn't in use and
 52 * has at least minlen blocks free. If no AG is found to match the allocation
 53 * requirements, pick the AG with the most free space in it.
 54 */
 55static int
 56xfs_filestream_pick_ag(
 57	struct xfs_alloc_arg	*args,
 58	xfs_ino_t		pino,
 59	xfs_agnumber_t		start_agno,
 60	int			flags,
 61	xfs_extlen_t		*longest)
 62{
 63	struct xfs_mount	*mp = args->mp;
 
 64	struct xfs_perag	*pag;
 65	struct xfs_perag	*max_pag = NULL;
 66	xfs_extlen_t		minlen = *longest;
 67	xfs_extlen_t		minfree, maxfree = 0;
 68	xfs_agnumber_t		agno;
 69	bool			first_pass = true;
 70
 71	/* 2% of an AG's blocks must be free for it to be chosen. */
 72	minfree = mp->m_sb.sb_agblocks / 50;
 73
 74restart:
 75	for_each_perag_wrap(mp, start_agno, agno, pag) {
 76		int		err;
 77
 78		trace_xfs_filestream_scan(pag, pino);
 79
 80		*longest = 0;
 81		err = xfs_bmap_longest_free_extent(pag, NULL, longest);
 82		if (err) {
 83			if (err == -EAGAIN) {
 84				/* Couldn't lock the AGF, skip this AG. */
 85				err = 0;
 86				continue;
 
 
 
 87			}
 88			xfs_perag_rele(pag);
 89			if (max_pag)
 90				xfs_perag_rele(max_pag);
 91			return err;
 92		}
 93
 
 
 
 
 94		/* Keep track of the AG with the most free blocks. */
 95		if (pag->pagf_freeblks > maxfree) {
 96			maxfree = pag->pagf_freeblks;
 97			if (max_pag)
 98				xfs_perag_rele(max_pag);
 99			atomic_inc(&pag_group(pag)->xg_active_ref);
100			max_pag = pag;
101		}
102
103		/*
104		 * The AG reference count does two things: it enforces mutual
105		 * exclusion when examining the suitability of an AG in this
106		 * loop, and it guards against two filestreams being established
107		 * in the same AG as each other.
108		 */
109		if (atomic_inc_return(&pag->pagf_fstrms) <= 1) {
110			if (((minlen && *longest >= minlen) ||
111			     (!minlen && pag->pagf_freeblks >= minfree)) &&
112			    (!xfs_perag_prefers_metadata(pag) ||
113			     !(flags & XFS_PICK_USERDATA) ||
114			     (flags & XFS_PICK_LOWSPACE))) {
115				/* Break out, retaining the reference on the AG. */
116				if (max_pag)
117					xfs_perag_rele(max_pag);
118				goto done;
119			}
 
 
 
 
 
 
 
120		}
121
122		/* Drop the reference on this AG, it's not usable. */
123		atomic_dec(&pag->pagf_fstrms);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
124	}
125
126	/*
127	 * Allow a second pass to give xfs_bmap_longest_free_extent() another
128	 * attempt at locking AGFs that it might have skipped over before we
129	 * fail.
130	 */
131	if (first_pass) {
132		first_pass = false;
133		goto restart;
134	}
135
136	/*
137	 * We must be low on data space, so run a final lowspace optimised
138	 * selection pass if we haven't already.
139	 */
140	if (!(flags & XFS_PICK_LOWSPACE)) {
141		flags |= XFS_PICK_LOWSPACE;
142		goto restart;
143	}
144
145	/*
146	 * No unassociated AGs are available, so select the AG with the most
147	 * free space, regardless of whether it's already in use by another
148	 * filestream. It none suit, just use whatever AG we can grab.
149	 */
150	if (!max_pag) {
151		for_each_perag_wrap(args->mp, 0, start_agno, pag) {
152			max_pag = pag;
153			break;
154		}
155
156		/* Bail if there are no AGs at all to select from. */
157		if (!max_pag)
158			return -ENOSPC;
 
 
159	}
160
161	pag = max_pag;
162	atomic_inc(&pag->pagf_fstrms);
163done:
164	trace_xfs_filestream_pick(pag, pino);
165	args->pag = pag;
166	return 0;
 
 
 
 
 
 
167}
168
169static struct xfs_inode *
170xfs_filestream_get_parent(
171	struct xfs_inode	*ip)
172{
173	struct inode		*inode = VFS_I(ip), *dir = NULL;
174	struct dentry		*dentry, *parent;
175
176	dentry = d_find_alias(inode);
177	if (!dentry)
178		goto out;
179
180	parent = dget_parent(dentry);
181	if (!parent)
182		goto out_dput;
183
184	dir = igrab(d_inode(parent));
185	dput(parent);
186
187out_dput:
188	dput(dentry);
189out:
190	return dir ? XFS_I(dir) : NULL;
191}
192
193/*
194 * Lookup the mru cache for an existing association. If one exists and we can
195 * use it, return with an active perag reference indicating that the allocation
196 * will proceed with that association.
197 *
198 * If we have no association, or we cannot use the current one and have to
199 * destroy it, return with longest = 0 to tell the caller to create a new
200 * association.
201 */
202static int
203xfs_filestream_lookup_association(
204	struct xfs_bmalloca	*ap,
205	struct xfs_alloc_arg	*args,
206	xfs_ino_t		pino,
207	xfs_extlen_t		*longest)
208{
209	struct xfs_mount	*mp = args->mp;
210	struct xfs_perag	*pag;
 
211	struct xfs_mru_cache_elem *mru;
212	int			error = 0;
213
214	*longest = 0;
215	mru = xfs_mru_cache_lookup(mp->m_filestream, pino);
216	if (!mru)
217		return 0;
218	/*
219	 * Grab the pag and take an extra active reference for the caller whilst
220	 * the mru item cannot go away. This means we'll pin the perag with
221	 * the reference we get here even if the filestreams association is torn
222	 * down immediately after we mark the lookup as done.
223	 */
224	pag = container_of(mru, struct xfs_fstrm_item, mru)->pag;
225	atomic_inc(&pag_group(pag)->xg_active_ref);
226	xfs_mru_cache_done(mp->m_filestream);
227
228	trace_xfs_filestream_lookup(pag, ap->ip->i_ino);
 
 
 
229
230	ap->blkno = xfs_agbno_to_fsb(pag, 0);
231	xfs_bmap_adjacent(ap);
 
232
233	/*
234	 * If there is very little free space before we start a filestreams
235	 * allocation, we're almost guaranteed to fail to find a large enough
236	 * free space available so just use the cached AG.
237	 */
238	if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
239		*longest = 1;
240		goto out_done;
241	}
 
 
 
242
243	error = xfs_bmap_longest_free_extent(pag, args->tp, longest);
244	if (error == -EAGAIN)
245		error = 0;
246	if (error || *longest < args->maxlen) {
247		/* We aren't going to use this perag */
248		*longest = 0;
249		xfs_perag_rele(pag);
250		return error;
251	}
252
253out_done:
254	args->pag = pag;
255	return 0;
256}
257
258static int
259xfs_filestream_create_association(
 
 
 
 
 
 
260	struct xfs_bmalloca	*ap,
261	struct xfs_alloc_arg	*args,
262	xfs_ino_t		pino,
263	xfs_extlen_t		*longest)
264{
265	struct xfs_mount	*mp = args->mp;
 
 
 
 
 
266	struct xfs_mru_cache_elem *mru;
267	struct xfs_fstrm_item	*item;
268	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, pino);
269	int			flags = 0;
270	int			error;
271
272	/* Changing parent AG association now, so remove the existing one. */
273	mru = xfs_mru_cache_remove(mp->m_filestream, pino);
 
 
 
 
 
274	if (mru) {
275		struct xfs_fstrm_item *item =
276			container_of(mru, struct xfs_fstrm_item, mru);
277
278		agno = (pag_agno(item->pag) + 1) % mp->m_sb.sb_agcount;
279		xfs_fstrm_free_func(mp, mru);
280	} else if (xfs_is_inode32(mp)) {
281		xfs_agnumber_t	 rotorstep = xfs_rotorstep;
282
283		agno = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
284		mp->m_agfrotor = (mp->m_agfrotor + 1) %
285				 (mp->m_sb.sb_agcount * rotorstep);
286	}
287
288	ap->blkno = XFS_AGB_TO_FSB(args->mp, agno, 0);
289	xfs_bmap_adjacent(ap);
290
291	if (ap->datatype & XFS_ALLOC_USERDATA)
292		flags |= XFS_PICK_USERDATA;
293	if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
294		flags |= XFS_PICK_LOWSPACE;
295
296	*longest = ap->length;
297	error = xfs_filestream_pick_ag(args, pino, agno, flags, longest);
298	if (error)
299		return error;
300
301	/*
302	 * We are going to use this perag now, so create an assoication for it.
303	 * xfs_filestream_pick_ag() has already bumped the perag fstrms counter
304	 * for us, so all we need to do here is take another active reference to
305	 * the perag for the cached association.
306	 *
307	 * If we fail to store the association, we need to drop the fstrms
308	 * counter as well as drop the perag reference we take here for the
309	 * item. We do not need to return an error for this failure - as long as
310	 * we return a referenced AG, the allocation can still go ahead just
311	 * fine.
312	 */
313	item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
314	if (!item)
315		goto out_put_fstrms;
316
317	atomic_inc(&pag_group(args->pag)->xg_active_ref);
318	item->pag = args->pag;
319	error = xfs_mru_cache_insert(mp->m_filestream, pino, &item->mru);
320	if (error)
321		goto out_free_item;
322	return 0;
323
324out_free_item:
325	xfs_perag_rele(item->pag);
326	kfree(item);
327out_put_fstrms:
328	atomic_dec(&args->pag->pagf_fstrms);
329	return 0;
330}
331
332/*
333 * Search for an allocation group with a single extent large enough for
334 * the request. First we look for an existing association and use that if it
335 * is found. Otherwise, we create a new association by selecting an AG that fits
336 * the allocation criteria.
337 *
338 * We return with a referenced perag in args->pag to indicate which AG we are
339 * allocating into or an error with no references held.
340 */
341int
342xfs_filestream_select_ag(
343	struct xfs_bmalloca	*ap,
344	struct xfs_alloc_arg	*args,
345	xfs_extlen_t		*longest)
346{
347	struct xfs_inode	*pip;
348	xfs_ino_t		ino = 0;
349	int			error = 0;
350
351	*longest = 0;
352	args->total = ap->total;
353	pip = xfs_filestream_get_parent(ap->ip);
354	if (pip) {
355		ino = pip->i_ino;
356		error = xfs_filestream_lookup_association(ap, args, ino,
357				longest);
358		xfs_irele(pip);
359		if (error)
360			return error;
361		if (*longest >= args->maxlen)
362			goto out_select;
363		if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
364			goto out_select;
365	}
366
367	error = xfs_filestream_create_association(ap, args, ino, longest);
368	if (error)
369		return error;
370
371out_select:
372	ap->blkno = xfs_agbno_to_fsb(args->pag, 0);
373	return 0;
374}
375
376void
377xfs_filestream_deassociate(
378	struct xfs_inode	*ip)
379{
380	xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino);
381}
382
383int
384xfs_filestream_mount(
385	xfs_mount_t	*mp)
386{
387	/*
388	 * The filestream timer tunable is currently fixed within the range of
389	 * one second to four minutes, with five seconds being the default.  The
390	 * group count is somewhat arbitrary, but it'd be nice to adhere to the
391	 * timer tunable to within about 10 percent.  This requires at least 10
392	 * groups.
393	 */
394	return xfs_mru_cache_create(&mp->m_filestream, mp,
395			xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);
396}
397
398void
399xfs_filestream_unmount(
400	xfs_mount_t	*mp)
401{
402	xfs_mru_cache_destroy(mp->m_filestream);
403}