1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
4 * Copyright (c) 2014 Christoph Hellwig.
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"
15 #include "xfs_bmap_util.h"
16 #include "xfs_alloc.h"
17 #include "xfs_mru_cache.h"
18 #include "xfs_trace.h"
20 #include "xfs_ag_resv.h"
21 #include "xfs_trans.h"
22 #include "xfs_filestream.h"
24 struct xfs_fstrm_item {
25 struct xfs_mru_cache_elem mru;
26 struct xfs_perag *pag; /* AG in use for this directory */
29 enum xfs_fstrm_alloc {
30 XFS_PICK_USERDATA = 1,
31 XFS_PICK_LOWSPACE = 2,
37 struct xfs_mru_cache_elem *mru)
39 struct xfs_fstrm_item *item =
40 container_of(mru, struct xfs_fstrm_item, mru);
41 struct xfs_perag *pag = item->pag;
43 trace_xfs_filestream_free(pag, mru->key);
44 atomic_dec(&pag->pagf_fstrms);
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.
56 xfs_filestream_pick_ag(
57 struct xfs_alloc_arg *args,
59 xfs_agnumber_t start_agno,
61 xfs_extlen_t *longest)
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 free = 0, minfree, maxfree = 0;
69 bool first_pass = true;
72 /* 2% of an AG's blocks must be free for it to be chosen. */
73 minfree = mp->m_sb.sb_agblocks / 50;
76 for_each_perag_wrap(mp, start_agno, agno, pag) {
77 trace_xfs_filestream_scan(pag, pino);
79 err = xfs_bmap_longest_free_extent(pag, NULL, longest);
83 /* Couldn't lock the AGF, skip this AG. */
88 /* Keep track of the AG with the most free blocks. */
89 if (pag->pagf_freeblks > maxfree) {
90 maxfree = pag->pagf_freeblks;
92 xfs_perag_rele(max_pag);
93 atomic_inc(&pag->pag_active_ref);
98 * The AG reference count does two things: it enforces mutual
99 * exclusion when examining the suitability of an AG in this
100 * loop, and it guards against two filestreams being established
101 * in the same AG as each other.
103 if (atomic_inc_return(&pag->pagf_fstrms) <= 1) {
104 if (((minlen && *longest >= minlen) ||
105 (!minlen && pag->pagf_freeblks >= minfree)) &&
106 (!xfs_perag_prefers_metadata(pag) ||
107 !(flags & XFS_PICK_USERDATA) ||
108 (flags & XFS_PICK_LOWSPACE))) {
109 /* Break out, retaining the reference on the AG. */
110 free = pag->pagf_freeblks;
115 /* Drop the reference on this AG, it's not usable. */
116 atomic_dec(&pag->pagf_fstrms);
122 xfs_perag_rele(max_pag);
128 * Allow a second pass to give xfs_bmap_longest_free_extent()
129 * another attempt at locking AGFs that it might have skipped
130 * over before we fail.
138 * We must be low on data space, so run a final lowspace
139 * optimised selection pass if we haven't already.
141 if (!(flags & XFS_PICK_LOWSPACE)) {
142 flags |= XFS_PICK_LOWSPACE;
147 * No unassociated AGs are available, so select the AG with the
148 * most free space, regardless of whether it's already in use by
149 * another filestream. It none suit, just use whatever AG we can
153 for_each_perag_wrap(args->mp, 0, start_agno, args->pag)
155 atomic_inc(&args->pag->pagf_fstrms);
160 atomic_inc(&pag->pagf_fstrms);
162 } else if (max_pag) {
163 xfs_perag_rele(max_pag);
166 trace_xfs_filestream_pick(pag, pino, free);
172 static struct xfs_inode *
173 xfs_filestream_get_parent(
174 struct xfs_inode *ip)
176 struct inode *inode = VFS_I(ip), *dir = NULL;
177 struct dentry *dentry, *parent;
179 dentry = d_find_alias(inode);
183 parent = dget_parent(dentry);
187 dir = igrab(d_inode(parent));
193 return dir ? XFS_I(dir) : NULL;
197 * Lookup the mru cache for an existing association. If one exists and we can
198 * use it, return with an active perag reference indicating that the allocation
199 * will proceed with that association.
201 * If we have no association, or we cannot use the current one and have to
202 * destroy it, return with longest = 0 to tell the caller to create a new
206 xfs_filestream_lookup_association(
207 struct xfs_bmalloca *ap,
208 struct xfs_alloc_arg *args,
210 xfs_extlen_t *longest)
212 struct xfs_mount *mp = args->mp;
213 struct xfs_perag *pag;
214 struct xfs_mru_cache_elem *mru;
218 mru = xfs_mru_cache_lookup(mp->m_filestream, pino);
222 * Grab the pag and take an extra active reference for the caller whilst
223 * the mru item cannot go away. This means we'll pin the perag with
224 * the reference we get here even if the filestreams association is torn
225 * down immediately after we mark the lookup as done.
227 pag = container_of(mru, struct xfs_fstrm_item, mru)->pag;
228 atomic_inc(&pag->pag_active_ref);
229 xfs_mru_cache_done(mp->m_filestream);
231 trace_xfs_filestream_lookup(pag, ap->ip->i_ino);
233 ap->blkno = XFS_AGB_TO_FSB(args->mp, pag->pag_agno, 0);
234 xfs_bmap_adjacent(ap);
237 * If there is very little free space before we start a filestreams
238 * allocation, we're almost guaranteed to fail to find a large enough
239 * free space available so just use the cached AG.
241 if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
246 error = xfs_bmap_longest_free_extent(pag, args->tp, longest);
247 if (error == -EAGAIN)
249 if (error || *longest < args->maxlen) {
250 /* We aren't going to use this perag */
262 xfs_filestream_create_association(
263 struct xfs_bmalloca *ap,
264 struct xfs_alloc_arg *args,
266 xfs_extlen_t *longest)
268 struct xfs_mount *mp = args->mp;
269 struct xfs_mru_cache_elem *mru;
270 struct xfs_fstrm_item *item;
271 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, pino);
275 /* Changing parent AG association now, so remove the existing one. */
276 mru = xfs_mru_cache_remove(mp->m_filestream, pino);
278 struct xfs_fstrm_item *item =
279 container_of(mru, struct xfs_fstrm_item, mru);
281 agno = (item->pag->pag_agno + 1) % mp->m_sb.sb_agcount;
282 xfs_fstrm_free_func(mp, mru);
283 } else if (xfs_is_inode32(mp)) {
284 xfs_agnumber_t rotorstep = xfs_rotorstep;
286 agno = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
287 mp->m_agfrotor = (mp->m_agfrotor + 1) %
288 (mp->m_sb.sb_agcount * rotorstep);
291 ap->blkno = XFS_AGB_TO_FSB(args->mp, agno, 0);
292 xfs_bmap_adjacent(ap);
294 if (ap->datatype & XFS_ALLOC_USERDATA)
295 flags |= XFS_PICK_USERDATA;
296 if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
297 flags |= XFS_PICK_LOWSPACE;
299 *longest = ap->length;
300 error = xfs_filestream_pick_ag(args, pino, agno, flags, longest);
305 * We are going to use this perag now, so create an assoication for it.
306 * xfs_filestream_pick_ag() has already bumped the perag fstrms counter
307 * for us, so all we need to do here is take another active reference to
308 * the perag for the cached association.
310 * If we fail to store the association, we need to drop the fstrms
311 * counter as well as drop the perag reference we take here for the
312 * item. We do not need to return an error for this failure - as long as
313 * we return a referenced AG, the allocation can still go ahead just
316 item = kmem_alloc(sizeof(*item), KM_MAYFAIL);
320 atomic_inc(&args->pag->pag_active_ref);
321 item->pag = args->pag;
322 error = xfs_mru_cache_insert(mp->m_filestream, pino, &item->mru);
328 xfs_perag_rele(item->pag);
331 atomic_dec(&args->pag->pagf_fstrms);
336 * Search for an allocation group with a single extent large enough for
337 * the request. First we look for an existing association and use that if it
338 * is found. Otherwise, we create a new association by selecting an AG that fits
339 * the allocation criteria.
341 * We return with a referenced perag in args->pag to indicate which AG we are
342 * allocating into or an error with no references held.
345 xfs_filestream_select_ag(
346 struct xfs_bmalloca *ap,
347 struct xfs_alloc_arg *args,
348 xfs_extlen_t *longest)
350 struct xfs_mount *mp = args->mp;
351 struct xfs_inode *pip;
356 args->total = ap->total;
357 pip = xfs_filestream_get_parent(ap->ip);
360 error = xfs_filestream_lookup_association(ap, args, ino,
365 if (*longest >= args->maxlen)
367 if (ap->tp->t_flags & XFS_TRANS_LOWMODE)
371 error = xfs_filestream_create_association(ap, args, ino, longest);
376 ap->blkno = XFS_AGB_TO_FSB(mp, args->pag->pag_agno, 0);
381 xfs_filestream_deassociate(
382 struct xfs_inode *ip)
384 xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino);
388 xfs_filestream_mount(
392 * The filestream timer tunable is currently fixed within the range of
393 * one second to four minutes, with five seconds being the default. The
394 * group count is somewhat arbitrary, but it'd be nice to adhere to the
395 * timer tunable to within about 10 percent. This requires at least 10
398 return xfs_mru_cache_create(&mp->m_filestream, mp,
399 xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);
403 xfs_filestream_unmount(
406 xfs_mru_cache_destroy(mp->m_filestream);