Blame view
fs/xfs/xfs_filestream.c
11.2 KB
2a82b8be8 [XFS] Concurrent ... |
1 2 |
/* * Copyright (c) 2006-2007 Silicon Graphics, Inc. |
2cd2ef6a3 xfs: rewrite the ... |
3 |
* Copyright (c) 2014 Christoph Hellwig. |
2a82b8be8 [XFS] Concurrent ... |
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 |
* 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" |
a4fbe6ab1 xfs: decouple ino... |
20 |
#include "xfs_format.h" |
239880ef6 xfs: decouple log... |
21 22 |
#include "xfs_log_format.h" #include "xfs_trans_resv.h" |
239880ef6 xfs: decouple log... |
23 24 |
#include "xfs_sb.h" #include "xfs_mount.h" |
3ab78df2a xfs: rework xfs_b... |
25 |
#include "xfs_defer.h" |
2a82b8be8 [XFS] Concurrent ... |
26 |
#include "xfs_inode.h" |
2a82b8be8 [XFS] Concurrent ... |
27 |
#include "xfs_bmap.h" |
689881145 xfs: create xfs_b... |
28 |
#include "xfs_bmap_util.h" |
2a82b8be8 [XFS] Concurrent ... |
29 |
#include "xfs_alloc.h" |
2a82b8be8 [XFS] Concurrent ... |
30 31 |
#include "xfs_mru_cache.h" #include "xfs_filestream.h" |
0b1b213fc xfs: event tracin... |
32 |
#include "xfs_trace.h" |
3fd129b63 xfs: set up per-A... |
33 |
#include "xfs_ag_resv.h" |
2a82b8be8 [XFS] Concurrent ... |
34 |
|
2cd2ef6a3 xfs: rewrite the ... |
35 36 37 38 39 40 41 42 43 44 |
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, }; |
2a82b8be8 [XFS] Concurrent ... |
45 |
|
0664ce8d0 xfs: clean up fil... |
46 47 |
/* * Allocation group filestream associations are tracked with per-ag atomic |
2cd2ef6a3 xfs: rewrite the ... |
48 |
* counters. These counters allow xfs_filestream_pick_ag() to tell whether a |
0664ce8d0 xfs: clean up fil... |
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 |
* 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. */ |
b94acd478 xfs: add filestre... |
83 |
int |
0664ce8d0 xfs: clean up fil... |
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 |
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); } |
2a82b8be8 [XFS] Concurrent ... |
122 |
|
2cd2ef6a3 xfs: rewrite the ... |
123 124 125 126 127 128 129 130 |
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); |
b94acd478 xfs: add filestre... |
131 |
trace_xfs_filestream_free(item->ip, item->ag); |
2cd2ef6a3 xfs: rewrite the ... |
132 |
|
1919adda0 xfs: don't create... |
133 |
kmem_free(item); |
2cd2ef6a3 xfs: rewrite the ... |
134 |
} |
2a82b8be8 [XFS] Concurrent ... |
135 136 137 138 139 |
/* * Scan the AGs starting at startag looking for an AG that isn't in use and has * at least minlen blocks free. */ static int |
2cd2ef6a3 xfs: rewrite the ... |
140 141 142 143 144 145 |
xfs_filestream_pick_ag( struct xfs_inode *ip, xfs_agnumber_t startag, xfs_agnumber_t *agp, int flags, xfs_extlen_t minlen) |
2a82b8be8 [XFS] Concurrent ... |
146 |
{ |
2cd2ef6a3 xfs: rewrite the ... |
147 148 149 |
struct xfs_mount *mp = ip->i_mount; struct xfs_fstrm_item *item; struct xfs_perag *pag; |
b94acd478 xfs: add filestre... |
150 |
xfs_extlen_t longest, free = 0, minfree, maxfree = 0; |
2cd2ef6a3 xfs: rewrite the ... |
151 |
xfs_agnumber_t ag, max_ag = NULLAGNUMBER; |
2cd2ef6a3 xfs: rewrite the ... |
152 |
int err, trylock, nscan; |
c19b3b05a xfs: mode di_mode... |
153 |
ASSERT(S_ISDIR(VFS_I(ip)->i_mode)); |
2a82b8be8 [XFS] Concurrent ... |
154 155 156 157 158 159 160 161 162 163 164 |
/* 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++) { |
b94acd478 xfs: add filestre... |
165 |
trace_xfs_filestream_scan(ip, ag); |
4196ac08c xfs: Convert file... |
166 |
pag = xfs_perag_get(mp, ag); |
2a82b8be8 [XFS] Concurrent ... |
167 168 169 |
if (!pag->pagf_init) { err = xfs_alloc_pagf_init(mp, NULL, ag, trylock); |
4196ac08c xfs: Convert file... |
170 171 |
if (err && !trylock) { xfs_perag_put(pag); |
2a82b8be8 [XFS] Concurrent ... |
172 |
return err; |
4196ac08c xfs: Convert file... |
173 |
} |
2a82b8be8 [XFS] Concurrent ... |
174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 |
} /* 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; } |
50adbcb4c xfs: xfs_alloc_fi... |
196 |
longest = xfs_alloc_longest_free_extent(mp, pag, |
3fd129b63 xfs: set up per-A... |
197 198 |
xfs_alloc_min_freelist(mp, pag), xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); |
2a82b8be8 [XFS] Concurrent ... |
199 200 201 202 203 204 205 |
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; |
4196ac08c xfs: Convert file... |
206 |
xfs_perag_put(pag); |
2a82b8be8 [XFS] Concurrent ... |
207 208 209 210 211 212 213 |
*agp = ag; break; } /* Drop the reference on this AG, it's not usable. */ xfs_filestream_put_ag(mp, ag); next_ag: |
4196ac08c xfs: Convert file... |
214 |
xfs_perag_put(pag); |
2a82b8be8 [XFS] Concurrent ... |
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 |
/* 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); |
2a82b8be8 [XFS] Concurrent ... |
241 242 243 244 245 246 |
free = maxfree; *agp = max_ag; break; } /* take AG 0 if none matched */ |
b94acd478 xfs: add filestre... |
247 |
trace_xfs_filestream_pick(ip, *agp, free, nscan); |
2a82b8be8 [XFS] Concurrent ... |
248 249 250 |
*agp = 0; return 0; } |
b94acd478 xfs: add filestre... |
251 |
trace_xfs_filestream_pick(ip, *agp, free, nscan); |
2a82b8be8 [XFS] Concurrent ... |
252 |
|
2cd2ef6a3 xfs: rewrite the ... |
253 |
if (*agp == NULLAGNUMBER) |
2a82b8be8 [XFS] Concurrent ... |
254 |
return 0; |
2a82b8be8 [XFS] Concurrent ... |
255 |
|
2451337dd xfs: global error... |
256 |
err = -ENOMEM; |
1919adda0 xfs: don't create... |
257 |
item = kmem_alloc(sizeof(*item), KM_MAYFAIL); |
2a82b8be8 [XFS] Concurrent ... |
258 |
if (!item) |
2cd2ef6a3 xfs: rewrite the ... |
259 |
goto out_put_ag; |
2a82b8be8 [XFS] Concurrent ... |
260 |
|
2cd2ef6a3 xfs: rewrite the ... |
261 |
item->ag = *agp; |
2a82b8be8 [XFS] Concurrent ... |
262 |
item->ip = ip; |
2a82b8be8 [XFS] Concurrent ... |
263 |
|
22328d712 xfs: embedd mru_e... |
264 |
err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru); |
2a82b8be8 [XFS] Concurrent ... |
265 |
if (err) { |
2451337dd xfs: global error... |
266 |
if (err == -EEXIST) |
2cd2ef6a3 xfs: rewrite the ... |
267 268 |
err = 0; goto out_free_item; |
2a82b8be8 [XFS] Concurrent ... |
269 |
} |
2a82b8be8 [XFS] Concurrent ... |
270 |
return 0; |
2a82b8be8 [XFS] Concurrent ... |
271 |
|
2cd2ef6a3 xfs: rewrite the ... |
272 |
out_free_item: |
1919adda0 xfs: don't create... |
273 |
kmem_free(item); |
2cd2ef6a3 xfs: rewrite the ... |
274 275 276 |
out_put_ag: xfs_filestream_put_ag(mp, *agp); return err; |
2a82b8be8 [XFS] Concurrent ... |
277 |
} |
2cd2ef6a3 xfs: rewrite the ... |
278 279 280 |
static struct xfs_inode * xfs_filestream_get_parent( struct xfs_inode *ip) |
2a82b8be8 [XFS] Concurrent ... |
281 |
{ |
2cd2ef6a3 xfs: rewrite the ... |
282 283 |
struct inode *inode = VFS_I(ip), *dir = NULL; struct dentry *dentry, *parent; |
2a82b8be8 [XFS] Concurrent ... |
284 |
|
2cd2ef6a3 xfs: rewrite the ... |
285 286 287 |
dentry = d_find_alias(inode); if (!dentry) goto out; |
2a82b8be8 [XFS] Concurrent ... |
288 |
|
2cd2ef6a3 xfs: rewrite the ... |
289 290 291 |
parent = dget_parent(dentry); if (!parent) goto out_dput; |
2a82b8be8 [XFS] Concurrent ... |
292 |
|
2b0143b5c VFS: normal files... |
293 |
dir = igrab(d_inode(parent)); |
2cd2ef6a3 xfs: rewrite the ... |
294 |
dput(parent); |
2a82b8be8 [XFS] Concurrent ... |
295 |
|
2cd2ef6a3 xfs: rewrite the ... |
296 297 298 299 |
out_dput: dput(dentry); out: return dir ? XFS_I(dir) : NULL; |
2a82b8be8 [XFS] Concurrent ... |
300 301 302 |
} /* |
3b8d90766 xfs: remove xfs_f... |
303 304 305 306 |
* 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. |
2a82b8be8 [XFS] Concurrent ... |
307 308 309 |
*/ xfs_agnumber_t xfs_filestream_lookup_ag( |
2cd2ef6a3 xfs: rewrite the ... |
310 |
struct xfs_inode *ip) |
2a82b8be8 [XFS] Concurrent ... |
311 |
{ |
2cd2ef6a3 xfs: rewrite the ... |
312 |
struct xfs_mount *mp = ip->i_mount; |
2cd2ef6a3 xfs: rewrite the ... |
313 |
struct xfs_inode *pip = NULL; |
3b8d90766 xfs: remove xfs_f... |
314 |
xfs_agnumber_t startag, ag = NULLAGNUMBER; |
22328d712 xfs: embedd mru_e... |
315 |
struct xfs_mru_cache_elem *mru; |
2a82b8be8 [XFS] Concurrent ... |
316 |
|
c19b3b05a xfs: mode di_mode... |
317 |
ASSERT(S_ISREG(VFS_I(ip)->i_mode)); |
2a82b8be8 [XFS] Concurrent ... |
318 |
|
2cd2ef6a3 xfs: rewrite the ... |
319 320 |
pip = xfs_filestream_get_parent(ip); if (!pip) |
b26384dc5 xfs: fix NULL poi... |
321 |
return NULLAGNUMBER; |
2cd2ef6a3 xfs: rewrite the ... |
322 323 |
mru = xfs_mru_cache_lookup(mp->m_filestream, pip->i_ino); |
22328d712 xfs: embedd mru_e... |
324 |
if (mru) { |
3b8d90766 xfs: remove xfs_f... |
325 |
ag = container_of(mru, struct xfs_fstrm_item, mru)->ag; |
22328d712 xfs: embedd mru_e... |
326 |
xfs_mru_cache_done(mp->m_filestream); |
3b8d90766 xfs: remove xfs_f... |
327 |
|
b94acd478 xfs: add filestre... |
328 |
trace_xfs_filestream_lookup(ip, ag); |
3b8d90766 xfs: remove xfs_f... |
329 |
goto out; |
2a82b8be8 [XFS] Concurrent ... |
330 331 332 333 334 335 336 |
} /* * Set the starting AG using the rotor for inode32, otherwise * use the directory inode's AG. */ if (mp->m_flags & XFS_MOUNT_32BITINODES) { |
2cd2ef6a3 xfs: rewrite the ... |
337 |
xfs_agnumber_t rotorstep = xfs_rotorstep; |
2a82b8be8 [XFS] Concurrent ... |
338 339 340 341 342 |
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); |
3b8d90766 xfs: remove xfs_f... |
343 344 345 346 347 |
if (xfs_filestream_pick_ag(pip, startag, &ag, 0, 0)) ag = NULLAGNUMBER; out: IRELE(pip); return ag; |
2a82b8be8 [XFS] Concurrent ... |
348 349 350 |
} /* |
2cd2ef6a3 xfs: rewrite the ... |
351 352 353 354 |
* 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. |
2a82b8be8 [XFS] Concurrent ... |
355 356 357 |
*/ int xfs_filestream_new_ag( |
689881145 xfs: create xfs_b... |
358 359 |
struct xfs_bmalloca *ap, xfs_agnumber_t *agp) |
2a82b8be8 [XFS] Concurrent ... |
360 |
{ |
2cd2ef6a3 xfs: rewrite the ... |
361 362 363 364 |
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; |
292378edc xfs: remote attri... |
365 366 |
int flags = 0; int err = 0; |
2cd2ef6a3 xfs: rewrite the ... |
367 |
struct xfs_mru_cache_elem *mru; |
2a82b8be8 [XFS] Concurrent ... |
368 |
|
2cd2ef6a3 xfs: rewrite the ... |
369 |
*agp = NULLAGNUMBER; |
2a82b8be8 [XFS] Concurrent ... |
370 |
|
2cd2ef6a3 xfs: rewrite the ... |
371 372 373 |
pip = xfs_filestream_get_parent(ip); if (!pip) goto exit; |
2a82b8be8 [XFS] Concurrent ... |
374 |
|
2cd2ef6a3 xfs: rewrite the ... |
375 376 377 378 379 |
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; |
2a82b8be8 [XFS] Concurrent ... |
380 |
} |
292378edc xfs: remote attri... |
381 382 383 384 |
if (xfs_alloc_is_userdata(ap->datatype)) flags |= XFS_PICK_USERDATA; if (ap->dfops->dop_low) flags |= XFS_PICK_LOWSPACE; |
2a82b8be8 [XFS] Concurrent ... |
385 |
|
2cd2ef6a3 xfs: rewrite the ... |
386 |
err = xfs_filestream_pick_ag(pip, startag, agp, flags, minlen); |
2a82b8be8 [XFS] Concurrent ... |
387 388 |
/* |
2cd2ef6a3 xfs: rewrite the ... |
389 |
* Only free the item here so we skip over the old AG earlier. |
2a82b8be8 [XFS] Concurrent ... |
390 |
*/ |
2cd2ef6a3 xfs: rewrite the ... |
391 392 |
if (mru) xfs_fstrm_free_func(mru); |
2a82b8be8 [XFS] Concurrent ... |
393 |
|
2cd2ef6a3 xfs: rewrite the ... |
394 395 396 397 398 399 |
IRELE(pip); exit: if (*agp == NULLAGNUMBER) *agp = 0; return err; } |
2a82b8be8 [XFS] Concurrent ... |
400 |
|
2cd2ef6a3 xfs: rewrite the ... |
401 402 403 404 405 406 |
void xfs_filestream_deassociate( struct xfs_inode *ip) { xfs_mru_cache_delete(ip->i_mount->m_filestream, ip->i_ino); } |
2a82b8be8 [XFS] Concurrent ... |
407 |
|
2cd2ef6a3 xfs: rewrite the ... |
408 409 410 411 |
int xfs_filestream_mount( xfs_mount_t *mp) { |
2a82b8be8 [XFS] Concurrent ... |
412 |
/* |
2cd2ef6a3 xfs: rewrite the ... |
413 414 415 416 417 |
* 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. |
2a82b8be8 [XFS] Concurrent ... |
418 |
*/ |
2cd2ef6a3 xfs: rewrite the ... |
419 420 421 |
return xfs_mru_cache_create(&mp->m_filestream, xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func); } |
2a82b8be8 [XFS] Concurrent ... |
422 |
|
2cd2ef6a3 xfs: rewrite the ... |
423 424 425 426 427 428 |
void xfs_filestream_unmount( xfs_mount_t *mp) { xfs_mru_cache_destroy(mp->m_filestream); } |