blob: 7def4c6993433452c6a1591083f8f98b73d8b0b7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott4ce31212005-11-02 14:59:41 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott4ce31212005-11-02 14:59:41 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott4ce31212005-11-02 14:59:41 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott4ce31212005-11-02 14:59:41 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
19#include "xfs_fs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_clnt.h"
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dir2.h"
28#include "xfs_alloc.h"
29#include "xfs_dmapi.h"
30#include "xfs_quota.h"
31#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110036#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_dinode.h"
38#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110039#include "xfs_btree.h"
40#include "xfs_ialloc.h"
41#include "xfs_itable.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_rtalloc.h"
43#include "xfs_error.h"
Nathan Scotta844f452005-11-02 14:38:42 +110044#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include "xfs_rw.h"
46#include "xfs_acl.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_attr.h"
48#include "xfs_buf_item.h"
49#include "xfs_trans_space.h"
50#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include "xfs_qm.h"
52
53/*
54 * The global quota manager. There is only one of these for the entire
55 * system, _not_ one per file system. XQM keeps track of the overall
56 * quota functionality, including maintaining the freelist and hash
57 * tables of dquots.
58 */
Nathan Scott6b3f6b52005-11-02 15:08:25 +110059mutex_t xfs_Gqm_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060struct xfs_qm *xfs_Gqm;
Nathan Scott6b3f6b52005-11-02 15:08:25 +110061uint ndquot;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63kmem_zone_t *qm_dqzone;
64kmem_zone_t *qm_dqtrxzone;
Andrew Morton3260f782007-05-24 15:25:42 +100065static struct shrinker *xfs_qm_shaker;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
David Chinner7989cb82007-02-10 18:34:56 +110067static cred_t xfs_zerocr;
Nathan Scottdae81d42006-02-28 12:30:13 +110068
Linus Torvalds1da177e2005-04-16 15:20:36 -070069STATIC void xfs_qm_list_init(xfs_dqlist_t *, char *, int);
70STATIC void xfs_qm_list_destroy(xfs_dqlist_t *);
71
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100072STATIC void xfs_qm_freelist_init(xfs_frlist_t *);
73STATIC void xfs_qm_freelist_destroy(xfs_frlist_t *);
74STATIC int xfs_qm_mplist_nowait(xfs_mount_t *);
75STATIC int xfs_qm_dqhashlock_nowait(xfs_dquot_t *);
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077STATIC int xfs_qm_init_quotainos(xfs_mount_t *);
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100078STATIC int xfs_qm_init_quotainfo(xfs_mount_t *);
Al Viro51bfb752005-12-15 09:17:24 +000079STATIC int xfs_qm_shake(int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81#ifdef DEBUG
82extern mutex_t qcheck_lock;
83#endif
84
85#ifdef QUOTADEBUG
86#define XQM_LIST_PRINT(l, NXT, title) \
87{ \
88 xfs_dquot_t *dqp; int i = 0; \
89 cmn_err(CE_DEBUG, "%s (#%d)", title, (int) (l)->qh_nelems); \
90 for (dqp = (l)->qh_next; dqp != NULL; dqp = dqp->NXT) { \
91 cmn_err(CE_DEBUG, " %d. \"%d (%s)\" " \
92 "bcnt = %d, icnt = %d, refs = %d", \
Christoph Hellwig1149d962005-11-02 15:01:12 +110093 ++i, (int) be32_to_cpu(dqp->q_core.d_id), \
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 DQFLAGTO_TYPESTR(dqp), \
Christoph Hellwig1149d962005-11-02 15:01:12 +110095 (int) be64_to_cpu(dqp->q_core.d_bcount), \
96 (int) be64_to_cpu(dqp->q_core.d_icount), \
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 (int) dqp->q_nrefs); } \
98}
99#else
100#define XQM_LIST_PRINT(l, NXT, title) do { } while (0)
101#endif
102
103/*
104 * Initialize the XQM structure.
105 * Note that there is not one quota manager per file system.
106 */
107STATIC struct xfs_qm *
108xfs_Gqm_init(void)
109{
Nathan Scott6b3f6b52005-11-02 15:08:25 +1100110 xfs_dqhash_t *udqhash, *gdqhash;
111 xfs_qm_t *xqm;
Nathan Scott215101c2006-09-28 11:04:43 +1000112 size_t hsize;
113 uint i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
115 /*
116 * Initialize the dquot hash tables.
117 */
Nathan Scott77e46352006-09-28 11:03:27 +1000118 udqhash = kmem_zalloc_greedy(&hsize,
119 XFS_QM_HASHSIZE_LOW, XFS_QM_HASHSIZE_HIGH,
120 KM_SLEEP | KM_MAYFAIL | KM_LARGE);
121 gdqhash = kmem_zalloc(hsize, KM_SLEEP | KM_LARGE);
122 hsize /= sizeof(xfs_dqhash_t);
Nathan Scott6b3f6b52005-11-02 15:08:25 +1100123 ndquot = hsize << 8;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Nathan Scott6b3f6b52005-11-02 15:08:25 +1100125 xqm = kmem_zalloc(sizeof(xfs_qm_t), KM_SLEEP);
126 xqm->qm_dqhashmask = hsize - 1;
127 xqm->qm_usr_dqhtable = udqhash;
128 xqm->qm_grp_dqhtable = gdqhash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 ASSERT(xqm->qm_usr_dqhtable != NULL);
130 ASSERT(xqm->qm_grp_dqhtable != NULL);
131
132 for (i = 0; i < hsize; i++) {
133 xfs_qm_list_init(&(xqm->qm_usr_dqhtable[i]), "uxdqh", i);
134 xfs_qm_list_init(&(xqm->qm_grp_dqhtable[i]), "gxdqh", i);
135 }
136
137 /*
138 * Freelist of all dquots of all file systems
139 */
140 xfs_qm_freelist_init(&(xqm->qm_dqfreelist));
141
142 /*
143 * dquot zone. we register our own low-memory callback.
144 */
145 if (!qm_dqzone) {
146 xqm->qm_dqzone = kmem_zone_init(sizeof(xfs_dquot_t),
147 "xfs_dquots");
148 qm_dqzone = xqm->qm_dqzone;
149 } else
150 xqm->qm_dqzone = qm_dqzone;
151
Andrew Morton3260f782007-05-24 15:25:42 +1000152 xfs_qm_shaker = set_shrinker(DEFAULT_SEEKS, xfs_qm_shake);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154 /*
155 * The t_dqinfo portion of transactions.
156 */
157 if (!qm_dqtrxzone) {
158 xqm->qm_dqtrxzone = kmem_zone_init(sizeof(xfs_dquot_acct_t),
159 "xfs_dqtrx");
160 qm_dqtrxzone = xqm->qm_dqtrxzone;
161 } else
162 xqm->qm_dqtrxzone = qm_dqtrxzone;
163
164 atomic_set(&xqm->qm_totaldquots, 0);
165 xqm->qm_dqfree_ratio = XFS_QM_DQFREE_RATIO;
166 xqm->qm_nrefs = 0;
167#ifdef DEBUG
Nathan Scottc2e81432006-01-16 16:21:34 +1100168 mutex_init(&qcheck_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169#endif
170 return xqm;
171}
172
173/*
174 * Destroy the global quota manager when its reference count goes to zero.
175 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000176STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177xfs_qm_destroy(
178 struct xfs_qm *xqm)
179{
180 int hsize, i;
181
182 ASSERT(xqm != NULL);
183 ASSERT(xqm->qm_nrefs == 0);
Andrew Morton3260f782007-05-24 15:25:42 +1000184 remove_shrinker(xfs_qm_shaker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 hsize = xqm->qm_dqhashmask + 1;
186 for (i = 0; i < hsize; i++) {
187 xfs_qm_list_destroy(&(xqm->qm_usr_dqhtable[i]));
188 xfs_qm_list_destroy(&(xqm->qm_grp_dqhtable[i]));
189 }
190 kmem_free(xqm->qm_usr_dqhtable, hsize * sizeof(xfs_dqhash_t));
191 kmem_free(xqm->qm_grp_dqhtable, hsize * sizeof(xfs_dqhash_t));
192 xqm->qm_usr_dqhtable = NULL;
193 xqm->qm_grp_dqhtable = NULL;
194 xqm->qm_dqhashmask = 0;
195 xfs_qm_freelist_destroy(&(xqm->qm_dqfreelist));
196#ifdef DEBUG
197 mutex_destroy(&qcheck_lock);
198#endif
199 kmem_free(xqm, sizeof(xfs_qm_t));
200}
201
202/*
203 * Called at mount time to let XQM know that another file system is
204 * starting quotas. This isn't crucial information as the individual mount
205 * structures are pretty independent, but it helps the XQM keep a
206 * global view of what's going on.
207 */
208/* ARGSUSED */
209STATIC int
210xfs_qm_hold_quotafs_ref(
211 struct xfs_mount *mp)
212{
213 /*
214 * Need to lock the xfs_Gqm structure for things like this. For example,
215 * the structure could disappear between the entry to this routine and
216 * a HOLD operation if not locked.
217 */
218 XFS_QM_LOCK(xfs_Gqm);
219
220 if (xfs_Gqm == NULL)
221 xfs_Gqm = xfs_Gqm_init();
222 /*
223 * We can keep a list of all filesystems with quotas mounted for
224 * debugging and statistical purposes, but ...
225 * Just take a reference and get out.
226 */
227 XFS_QM_HOLD(xfs_Gqm);
228 XFS_QM_UNLOCK(xfs_Gqm);
229
230 return 0;
231}
232
233
234/*
235 * Release the reference that a filesystem took at mount time,
236 * so that we know when we need to destroy the entire quota manager.
237 */
238/* ARGSUSED */
239STATIC void
240xfs_qm_rele_quotafs_ref(
241 struct xfs_mount *mp)
242{
243 xfs_dquot_t *dqp, *nextdqp;
244
245 ASSERT(xfs_Gqm);
246 ASSERT(xfs_Gqm->qm_nrefs > 0);
247
248 /*
249 * Go thru the freelist and destroy all inactive dquots.
250 */
251 xfs_qm_freelist_lock(xfs_Gqm);
252
253 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
254 dqp != (xfs_dquot_t *)&(xfs_Gqm->qm_dqfreelist); ) {
255 xfs_dqlock(dqp);
256 nextdqp = dqp->dq_flnext;
257 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
258 ASSERT(dqp->q_mount == NULL);
259 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
260 ASSERT(dqp->HL_PREVP == NULL);
261 ASSERT(dqp->MPL_PREVP == NULL);
262 XQM_FREELIST_REMOVE(dqp);
263 xfs_dqunlock(dqp);
264 xfs_qm_dqdestroy(dqp);
265 } else {
266 xfs_dqunlock(dqp);
267 }
268 dqp = nextdqp;
269 }
270 xfs_qm_freelist_unlock(xfs_Gqm);
271
272 /*
273 * Destroy the entire XQM. If somebody mounts with quotaon, this'll
274 * be restarted.
275 */
276 XFS_QM_LOCK(xfs_Gqm);
277 XFS_QM_RELE(xfs_Gqm);
278 if (xfs_Gqm->qm_nrefs == 0) {
279 xfs_qm_destroy(xfs_Gqm);
280 xfs_Gqm = NULL;
281 }
282 XFS_QM_UNLOCK(xfs_Gqm);
283}
284
285/*
286 * This is called at mount time from xfs_mountfs to initialize the quotainfo
Nathan Scottc41564b2006-03-29 08:55:14 +1000287 * structure and start the global quota manager (xfs_Gqm) if it hasn't done
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 * so already. Note that the superblock has not been read in yet.
289 */
290void
291xfs_qm_mount_quotainit(
292 xfs_mount_t *mp,
293 uint flags)
294{
295 /*
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000296 * User, projects or group quotas has to be on.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000298 ASSERT(flags & (XFSMNT_UQUOTA | XFSMNT_PQUOTA | XFSMNT_GQUOTA));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300 /*
301 * Initialize the flags in the mount structure. From this point
302 * onwards we look at m_qflags to figure out if quotas's ON/OFF, etc.
303 * Note that we enforce nothing if accounting is off.
304 * ie. XFSMNT_*QUOTA must be ON for XFSMNT_*QUOTAENF.
305 * It isn't necessary to take the quotaoff lock to do this; this is
306 * called from mount.
307 */
308 if (flags & XFSMNT_UQUOTA) {
309 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
310 if (flags & XFSMNT_UQUOTAENF)
311 mp->m_qflags |= XFS_UQUOTA_ENFD;
312 }
313 if (flags & XFSMNT_GQUOTA) {
314 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
315 if (flags & XFSMNT_GQUOTAENF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000316 mp->m_qflags |= XFS_OQUOTA_ENFD;
317 } else if (flags & XFSMNT_PQUOTA) {
318 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
319 if (flags & XFSMNT_PQUOTAENF)
320 mp->m_qflags |= XFS_OQUOTA_ENFD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 }
322}
323
324/*
325 * Just destroy the quotainfo structure.
326 */
327void
328xfs_qm_unmount_quotadestroy(
329 xfs_mount_t *mp)
330{
331 if (mp->m_quotainfo)
332 xfs_qm_destroy_quotainfo(mp);
333}
334
335
336/*
337 * This is called from xfs_mountfs to start quotas and initialize all
338 * necessary data structures like quotainfo. This is also responsible for
339 * running a quotacheck as necessary. We are guaranteed that the superblock
340 * is consistently read in at this point.
341 */
342int
343xfs_qm_mount_quotas(
344 xfs_mount_t *mp,
345 int mfsi_flags)
346{
347 unsigned long s;
348 int error = 0;
349 uint sbf;
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352 /*
353 * If quotas on realtime volumes is not supported, we disable
354 * quotas immediately.
355 */
356 if (mp->m_sb.sb_rextents) {
357 cmn_err(CE_NOTE,
358 "Cannot turn on quotas for realtime filesystem %s",
359 mp->m_fsname);
360 mp->m_qflags = 0;
361 goto write_changes;
362 }
363
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
Nathan Scott155ffd02005-09-02 16:43:48 +1000365
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 /*
367 * Allocate the quotainfo structure inside the mount struct, and
368 * create quotainode(s), and change/rev superblock if necessary.
369 */
370 if ((error = xfs_qm_init_quotainfo(mp))) {
371 /*
372 * We must turn off quotas.
373 */
374 ASSERT(mp->m_quotainfo == NULL);
375 mp->m_qflags = 0;
376 goto write_changes;
377 }
378 /*
379 * If any of the quotas are not consistent, do a quotacheck.
380 */
381 if (XFS_QM_NEED_QUOTACHECK(mp) &&
382 !(mfsi_flags & XFS_MFSI_NO_QUOTACHECK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 if ((error = xfs_qm_quotacheck(mp))) {
384 /* Quotacheck has failed and quotas have
385 * been disabled.
386 */
387 return XFS_ERROR(error);
388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 }
Donald Douwsma646d5bd2007-05-08 13:49:09 +1000390 /*
391 * If one type of quotas is off, then it will lose its
392 * quotachecked status, since we won't be doing accounting for
393 * that type anymore.
394 */
395 if (!XFS_IS_UQUOTA_ON(mp)) {
396 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
397 }
398 if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp))) {
399 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
400 }
Nathan Scott155ffd02005-09-02 16:43:48 +1000401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 write_changes:
403 /*
404 * We actually don't have to acquire the SB_LOCK at all.
405 * This can only be called from mount, and that's single threaded. XXX
406 */
407 s = XFS_SB_LOCK(mp);
408 sbf = mp->m_sb.sb_qflags;
409 mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
410 XFS_SB_UNLOCK(mp, s);
411
412 if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
413 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
414 /*
415 * We could only have been turning quotas off.
416 * We aren't in very good shape actually because
417 * the incore structures are convinced that quotas are
418 * off, but the on disk superblock doesn't know that !
419 */
420 ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
421 xfs_fs_cmn_err(CE_ALERT, mp,
422 "XFS mount_quotas: Superblock update failed!");
423 }
424 }
425
426 if (error) {
427 xfs_fs_cmn_err(CE_WARN, mp,
428 "Failed to initialize disk quotas.");
429 }
430 return XFS_ERROR(error);
431}
432
433/*
434 * Called from the vfsops layer.
435 */
436int
437xfs_qm_unmount_quotas(
438 xfs_mount_t *mp)
439{
440 xfs_inode_t *uqp, *gqp;
441 int error = 0;
442
443 /*
444 * Release the dquots that root inode, et al might be holding,
445 * before we flush quotas and blow away the quotainfo structure.
446 */
447 ASSERT(mp->m_rootip);
448 xfs_qm_dqdetach(mp->m_rootip);
449 if (mp->m_rbmip)
450 xfs_qm_dqdetach(mp->m_rbmip);
451 if (mp->m_rsumip)
452 xfs_qm_dqdetach(mp->m_rsumip);
453
454 /*
455 * Flush out the quota inodes.
456 */
457 uqp = gqp = NULL;
458 if (mp->m_quotainfo) {
459 if ((uqp = mp->m_quotainfo->qi_uquotaip) != NULL) {
460 xfs_ilock(uqp, XFS_ILOCK_EXCL);
461 xfs_iflock(uqp);
462 error = xfs_iflush(uqp, XFS_IFLUSH_SYNC);
463 xfs_iunlock(uqp, XFS_ILOCK_EXCL);
464 if (unlikely(error == EFSCORRUPTED)) {
465 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(1)",
466 XFS_ERRLEVEL_LOW, mp);
467 goto out;
468 }
469 }
470 if ((gqp = mp->m_quotainfo->qi_gquotaip) != NULL) {
471 xfs_ilock(gqp, XFS_ILOCK_EXCL);
472 xfs_iflock(gqp);
473 error = xfs_iflush(gqp, XFS_IFLUSH_SYNC);
474 xfs_iunlock(gqp, XFS_ILOCK_EXCL);
475 if (unlikely(error == EFSCORRUPTED)) {
476 XFS_ERROR_REPORT("xfs_qm_unmount_quotas(2)",
477 XFS_ERRLEVEL_LOW, mp);
478 goto out;
479 }
480 }
481 }
482 if (uqp) {
483 XFS_PURGE_INODE(uqp);
484 mp->m_quotainfo->qi_uquotaip = NULL;
485 }
486 if (gqp) {
487 XFS_PURGE_INODE(gqp);
488 mp->m_quotainfo->qi_gquotaip = NULL;
489 }
490out:
491 return XFS_ERROR(error);
492}
493
494/*
495 * Flush all dquots of the given file system to disk. The dquots are
496 * _not_ purged from memory here, just their data written to disk.
497 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000498STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499xfs_qm_dqflush_all(
500 xfs_mount_t *mp,
501 int flags)
502{
503 int recl;
504 xfs_dquot_t *dqp;
505 int niters;
506 int error;
507
508 if (mp->m_quotainfo == NULL)
Jesper Juhl014c2542006-01-15 02:37:08 +0100509 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 niters = 0;
511again:
512 xfs_qm_mplist_lock(mp);
513 FOREACH_DQUOT_IN_MP(dqp, mp) {
514 xfs_dqlock(dqp);
515 if (! XFS_DQ_IS_DIRTY(dqp)) {
516 xfs_dqunlock(dqp);
517 continue;
518 }
519 xfs_dqtrace_entry(dqp, "FLUSHALL: DQDIRTY");
520 /* XXX a sentinel would be better */
521 recl = XFS_QI_MPLRECLAIMS(mp);
522 if (! xfs_qm_dqflock_nowait(dqp)) {
523 /*
524 * If we can't grab the flush lock then check
525 * to see if the dquot has been flushed delayed
526 * write. If so, grab its buffer and send it
527 * out immediately. We'll be able to acquire
528 * the flush lock when the I/O completes.
529 */
530 xfs_qm_dqflock_pushbuf_wait(dqp);
531 }
532 /*
533 * Let go of the mplist lock. We don't want to hold it
534 * across a disk write.
535 */
536 xfs_qm_mplist_unlock(mp);
537 error = xfs_qm_dqflush(dqp, flags);
538 xfs_dqunlock(dqp);
539 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100540 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 xfs_qm_mplist_lock(mp);
543 if (recl != XFS_QI_MPLRECLAIMS(mp)) {
544 xfs_qm_mplist_unlock(mp);
545 /* XXX restart limit */
546 goto again;
547 }
548 }
549
550 xfs_qm_mplist_unlock(mp);
551 /* return ! busy */
Jesper Juhl014c2542006-01-15 02:37:08 +0100552 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553}
554/*
555 * Release the group dquot pointers the user dquots may be
556 * carrying around as a hint. mplist is locked on entry and exit.
557 */
558STATIC void
559xfs_qm_detach_gdquots(
560 xfs_mount_t *mp)
561{
562 xfs_dquot_t *dqp, *gdqp;
563 int nrecl;
564
565 again:
566 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
567 dqp = XFS_QI_MPLNEXT(mp);
568 while (dqp) {
569 xfs_dqlock(dqp);
570 if ((gdqp = dqp->q_gdquot)) {
571 xfs_dqlock(gdqp);
572 dqp->q_gdquot = NULL;
573 }
574 xfs_dqunlock(dqp);
575
576 if (gdqp) {
577 /*
578 * Can't hold the mplist lock across a dqput.
579 * XXXmust convert to marker based iterations here.
580 */
581 nrecl = XFS_QI_MPLRECLAIMS(mp);
582 xfs_qm_mplist_unlock(mp);
583 xfs_qm_dqput(gdqp);
584
585 xfs_qm_mplist_lock(mp);
586 if (nrecl != XFS_QI_MPLRECLAIMS(mp))
587 goto again;
588 }
589 dqp = dqp->MPL_NEXT;
590 }
591}
592
593/*
594 * Go through all the incore dquots of this file system and take them
595 * off the mplist and hashlist, if the dquot type matches the dqtype
596 * parameter. This is used when turning off quota accounting for
597 * users and/or groups, as well as when the filesystem is unmounting.
598 */
599STATIC int
600xfs_qm_dqpurge_int(
601 xfs_mount_t *mp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000602 uint flags) /* QUOTAOFF/UMOUNTING/UQUOTA/PQUOTA/GQUOTA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
604 xfs_dquot_t *dqp;
605 uint dqtype;
606 int nrecl;
607 xfs_dquot_t *nextdqp;
608 int nmisses;
609
610 if (mp->m_quotainfo == NULL)
Jesper Juhl014c2542006-01-15 02:37:08 +0100611 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
613 dqtype = (flags & XFS_QMOPT_UQUOTA) ? XFS_DQ_USER : 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000614 dqtype |= (flags & XFS_QMOPT_PQUOTA) ? XFS_DQ_PROJ : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 dqtype |= (flags & XFS_QMOPT_GQUOTA) ? XFS_DQ_GROUP : 0;
616
617 xfs_qm_mplist_lock(mp);
618
619 /*
620 * In the first pass through all incore dquots of this filesystem,
621 * we release the group dquot pointers the user dquots may be
622 * carrying around as a hint. We need to do this irrespective of
623 * what's being turned off.
624 */
625 xfs_qm_detach_gdquots(mp);
626
627 again:
628 nmisses = 0;
629 ASSERT(XFS_QM_IS_MPLIST_LOCKED(mp));
630 /*
631 * Try to get rid of all of the unwanted dquots. The idea is to
632 * get them off mplist and hashlist, but leave them on freelist.
633 */
634 dqp = XFS_QI_MPLNEXT(mp);
635 while (dqp) {
636 /*
637 * It's OK to look at the type without taking dqlock here.
638 * We're holding the mplist lock here, and that's needed for
639 * a dqreclaim.
640 */
641 if ((dqp->dq_flags & dqtype) == 0) {
642 dqp = dqp->MPL_NEXT;
643 continue;
644 }
645
646 if (! xfs_qm_dqhashlock_nowait(dqp)) {
647 nrecl = XFS_QI_MPLRECLAIMS(mp);
648 xfs_qm_mplist_unlock(mp);
649 XFS_DQ_HASH_LOCK(dqp->q_hash);
650 xfs_qm_mplist_lock(mp);
651
652 /*
653 * XXXTheoretically, we can get into a very long
654 * ping pong game here.
655 * No one can be adding dquots to the mplist at
656 * this point, but somebody might be taking things off.
657 */
658 if (nrecl != XFS_QI_MPLRECLAIMS(mp)) {
659 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
660 goto again;
661 }
662 }
663
664 /*
665 * Take the dquot off the mplist and hashlist. It may remain on
666 * freelist in INACTIVE state.
667 */
668 nextdqp = dqp->MPL_NEXT;
669 nmisses += xfs_qm_dqpurge(dqp, flags);
670 dqp = nextdqp;
671 }
672 xfs_qm_mplist_unlock(mp);
673 return nmisses;
674}
675
676int
677xfs_qm_dqpurge_all(
678 xfs_mount_t *mp,
679 uint flags)
680{
681 int ndquots;
682
683 /*
684 * Purge the dquot cache.
685 * None of the dquots should really be busy at this point.
686 */
687 if (mp->m_quotainfo) {
688 while ((ndquots = xfs_qm_dqpurge_int(mp, flags))) {
689 delay(ndquots * 10);
690 }
691 }
692 return 0;
693}
694
695STATIC int
696xfs_qm_dqattach_one(
697 xfs_inode_t *ip,
698 xfs_dqid_t id,
699 uint type,
700 uint doalloc,
701 uint dolock,
702 xfs_dquot_t *udqhint, /* hint */
703 xfs_dquot_t **IO_idqpp)
704{
705 xfs_dquot_t *dqp;
706 int error;
707
708 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
709 error = 0;
710 /*
711 * See if we already have it in the inode itself. IO_idqpp is
712 * &i_udquot or &i_gdquot. This made the code look weird, but
713 * made the logic a lot simpler.
714 */
715 if ((dqp = *IO_idqpp)) {
716 if (dolock)
717 xfs_dqlock(dqp);
718 xfs_dqtrace_entry(dqp, "DQATTACH: found in ip");
719 goto done;
720 }
721
722 /*
723 * udqhint is the i_udquot field in inode, and is non-NULL only
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000724 * when the type arg is group/project. Its purpose is to save a
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
726 * the user dquot.
727 */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000728 ASSERT(!udqhint || type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 if (udqhint && !dolock)
730 xfs_dqlock(udqhint);
731
732 /*
733 * No need to take dqlock to look at the id.
734 * The ID can't change until it gets reclaimed, and it won't
735 * be reclaimed as long as we have a ref from inode and we hold
736 * the ilock.
737 */
738 if (udqhint &&
739 (dqp = udqhint->q_gdquot) &&
Christoph Hellwig1149d962005-11-02 15:01:12 +1100740 (be32_to_cpu(dqp->q_core.d_id) == id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 ASSERT(XFS_DQ_IS_LOCKED(udqhint));
742 xfs_dqlock(dqp);
743 XFS_DQHOLD(dqp);
744 ASSERT(*IO_idqpp == NULL);
745 *IO_idqpp = dqp;
746 if (!dolock) {
747 xfs_dqunlock(dqp);
748 xfs_dqunlock(udqhint);
749 }
750 goto done;
751 }
752 /*
753 * We can't hold a dquot lock when we call the dqget code.
754 * We'll deadlock in no time, because of (not conforming to)
755 * lock ordering - the inodelock comes before any dquot lock,
756 * and we may drop and reacquire the ilock in xfs_qm_dqget().
757 */
758 if (udqhint)
759 xfs_dqunlock(udqhint);
760 /*
761 * Find the dquot from somewhere. This bumps the
762 * reference count of dquot and returns it locked.
763 * This can return ENOENT if dquot didn't exist on
764 * disk and we didn't ask it to allocate;
765 * ESRCH if quotas got turned off suddenly.
766 */
767 if ((error = xfs_qm_dqget(ip->i_mount, ip, id, type,
768 doalloc|XFS_QMOPT_DOWARN, &dqp))) {
769 if (udqhint && dolock)
770 xfs_dqlock(udqhint);
771 goto done;
772 }
773
774 xfs_dqtrace_entry(dqp, "DQATTACH: found by dqget");
775 /*
776 * dqget may have dropped and re-acquired the ilock, but it guarantees
777 * that the dquot returned is the one that should go in the inode.
778 */
779 *IO_idqpp = dqp;
780 ASSERT(dqp);
781 ASSERT(XFS_DQ_IS_LOCKED(dqp));
782 if (! dolock) {
783 xfs_dqunlock(dqp);
784 goto done;
785 }
786 if (! udqhint)
787 goto done;
788
789 ASSERT(udqhint);
790 ASSERT(dolock);
791 ASSERT(XFS_DQ_IS_LOCKED(dqp));
792 if (! xfs_qm_dqlock_nowait(udqhint)) {
793 xfs_dqunlock(dqp);
794 xfs_dqlock(udqhint);
795 xfs_dqlock(dqp);
796 }
797 done:
798#ifdef QUOTADEBUG
799 if (udqhint) {
800 if (dolock)
801 ASSERT(XFS_DQ_IS_LOCKED(udqhint));
802 }
803 if (! error) {
804 if (dolock)
805 ASSERT(XFS_DQ_IS_LOCKED(dqp));
806 }
807#endif
Jesper Juhl014c2542006-01-15 02:37:08 +0100808 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809}
810
811
812/*
813 * Given a udquot and gdquot, attach a ptr to the group dquot in the
814 * udquot as a hint for future lookups. The idea sounds simple, but the
815 * execution isn't, because the udquot might have a group dquot attached
Nathan Scottc41564b2006-03-29 08:55:14 +1000816 * already and getting rid of that gets us into lock ordering constraints.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 * The process is complicated more by the fact that the dquots may or may not
818 * be locked on entry.
819 */
820STATIC void
821xfs_qm_dqattach_grouphint(
822 xfs_dquot_t *udq,
823 xfs_dquot_t *gdq,
824 uint locked)
825{
826 xfs_dquot_t *tmp;
827
828#ifdef QUOTADEBUG
829 if (locked) {
830 ASSERT(XFS_DQ_IS_LOCKED(udq));
831 ASSERT(XFS_DQ_IS_LOCKED(gdq));
832 }
833#endif
834 if (! locked)
835 xfs_dqlock(udq);
836
837 if ((tmp = udq->q_gdquot)) {
838 if (tmp == gdq) {
839 if (! locked)
840 xfs_dqunlock(udq);
841 return;
842 }
843
844 udq->q_gdquot = NULL;
845 /*
846 * We can't keep any dqlocks when calling dqrele,
847 * because the freelist lock comes before dqlocks.
848 */
849 xfs_dqunlock(udq);
850 if (locked)
851 xfs_dqunlock(gdq);
852 /*
853 * we took a hard reference once upon a time in dqget,
854 * so give it back when the udquot no longer points at it
855 * dqput() does the unlocking of the dquot.
856 */
857 xfs_qm_dqrele(tmp);
858
859 xfs_dqlock(udq);
860 xfs_dqlock(gdq);
861
862 } else {
863 ASSERT(XFS_DQ_IS_LOCKED(udq));
864 if (! locked) {
865 xfs_dqlock(gdq);
866 }
867 }
868
869 ASSERT(XFS_DQ_IS_LOCKED(udq));
870 ASSERT(XFS_DQ_IS_LOCKED(gdq));
871 /*
872 * Somebody could have attached a gdquot here,
873 * when we dropped the uqlock. If so, just do nothing.
874 */
875 if (udq->q_gdquot == NULL) {
876 XFS_DQHOLD(gdq);
877 udq->q_gdquot = gdq;
878 }
879 if (! locked) {
880 xfs_dqunlock(gdq);
881 xfs_dqunlock(udq);
882 }
883}
884
885
886/*
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000887 * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
888 * into account.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
890 * If XFS_QMOPT_DQLOCK, the dquot(s) will be returned locked. This option pretty
891 * much made this code a complete mess, but it has been pretty useful.
892 * If XFS_QMOPT_ILOCKED, then inode sent is already locked EXCL.
893 * Inode may get unlocked and relocked in here, and the caller must deal with
894 * the consequences.
895 */
896int
897xfs_qm_dqattach(
898 xfs_inode_t *ip,
899 uint flags)
900{
901 xfs_mount_t *mp = ip->i_mount;
902 uint nquotas = 0;
903 int error = 0;
904
905 if ((! XFS_IS_QUOTA_ON(mp)) ||
906 (! XFS_NOT_DQATTACHED(mp, ip)) ||
907 (ip->i_ino == mp->m_sb.sb_uquotino) ||
908 (ip->i_ino == mp->m_sb.sb_gquotino))
Jesper Juhl014c2542006-01-15 02:37:08 +0100909 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
911 ASSERT((flags & XFS_QMOPT_ILOCKED) == 0 ||
912 XFS_ISLOCKED_INODE_EXCL(ip));
913
914 if (! (flags & XFS_QMOPT_ILOCKED))
915 xfs_ilock(ip, XFS_ILOCK_EXCL);
916
917 if (XFS_IS_UQUOTA_ON(mp)) {
918 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
919 flags & XFS_QMOPT_DQALLOC,
920 flags & XFS_QMOPT_DQLOCK,
921 NULL, &ip->i_udquot);
922 if (error)
923 goto done;
924 nquotas++;
925 }
926 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000927 if (XFS_IS_OQUOTA_ON(mp)) {
928 error = XFS_IS_GQUOTA_ON(mp) ?
929 xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
930 flags & XFS_QMOPT_DQALLOC,
931 flags & XFS_QMOPT_DQLOCK,
932 ip->i_udquot, &ip->i_gdquot) :
933 xfs_qm_dqattach_one(ip, ip->i_d.di_projid, XFS_DQ_PROJ,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 flags & XFS_QMOPT_DQALLOC,
935 flags & XFS_QMOPT_DQLOCK,
936 ip->i_udquot, &ip->i_gdquot);
937 /*
938 * Don't worry about the udquot that we may have
939 * attached above. It'll get detached, if not already.
940 */
941 if (error)
942 goto done;
943 nquotas++;
944 }
945
946 /*
947 * Attach this group quota to the user quota as a hint.
948 * This WON'T, in general, result in a thrash.
949 */
950 if (nquotas == 2) {
951 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
952 ASSERT(ip->i_udquot);
953 ASSERT(ip->i_gdquot);
954
955 /*
956 * We may or may not have the i_udquot locked at this point,
957 * but this check is OK since we don't depend on the i_gdquot to
958 * be accurate 100% all the time. It is just a hint, and this
959 * will succeed in general.
960 */
961 if (ip->i_udquot->q_gdquot == ip->i_gdquot)
962 goto done;
963 /*
964 * Attach i_gdquot to the gdquot hint inside the i_udquot.
965 */
966 xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot,
967 flags & XFS_QMOPT_DQLOCK);
968 }
969
970 done:
971
972#ifdef QUOTADEBUG
973 if (! error) {
974 if (ip->i_udquot) {
975 if (flags & XFS_QMOPT_DQLOCK)
976 ASSERT(XFS_DQ_IS_LOCKED(ip->i_udquot));
977 }
978 if (ip->i_gdquot) {
979 if (flags & XFS_QMOPT_DQLOCK)
980 ASSERT(XFS_DQ_IS_LOCKED(ip->i_gdquot));
981 }
982 if (XFS_IS_UQUOTA_ON(mp))
983 ASSERT(ip->i_udquot);
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000984 if (XFS_IS_OQUOTA_ON(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 ASSERT(ip->i_gdquot);
986 }
987#endif
988
989 if (! (flags & XFS_QMOPT_ILOCKED))
990 xfs_iunlock(ip, XFS_ILOCK_EXCL);
991
992#ifdef QUOTADEBUG
993 else
994 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
995#endif
Jesper Juhl014c2542006-01-15 02:37:08 +0100996 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997}
998
999/*
1000 * Release dquots (and their references) if any.
1001 * The inode should be locked EXCL except when this's called by
1002 * xfs_ireclaim.
1003 */
1004void
1005xfs_qm_dqdetach(
1006 xfs_inode_t *ip)
1007{
1008 if (!(ip->i_udquot || ip->i_gdquot))
1009 return;
1010
1011 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
1012 ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (ip->i_udquot) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001014 xfs_dqtrace_entry_ino(ip->i_udquot, "DQDETTACH", ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 xfs_qm_dqrele(ip->i_udquot);
1016 ip->i_udquot = NULL;
1017 }
1018 if (ip->i_gdquot) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001019 xfs_dqtrace_entry_ino(ip->i_gdquot, "DQDETTACH", ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 xfs_qm_dqrele(ip->i_gdquot);
1021 ip->i_gdquot = NULL;
1022 }
1023}
1024
1025/*
1026 * This is called by VFS_SYNC and flags arg determines the caller,
1027 * and its motives, as done in xfs_sync.
1028 *
1029 * vfs_sync: SYNC_FSDATA|SYNC_ATTR|SYNC_BDFLUSH 0x31
1030 * syscall sync: SYNC_FSDATA|SYNC_ATTR|SYNC_DELWRI 0x25
1031 * umountroot : SYNC_WAIT | SYNC_CLOSE | SYNC_ATTR | SYNC_FSDATA
1032 */
1033
1034int
1035xfs_qm_sync(
1036 xfs_mount_t *mp,
1037 short flags)
1038{
1039 int recl, restarts;
1040 xfs_dquot_t *dqp;
1041 uint flush_flags;
1042 boolean_t nowait;
1043 int error;
1044
1045 restarts = 0;
1046 /*
1047 * We won't block unless we are asked to.
1048 */
1049 nowait = (boolean_t)(flags & SYNC_BDFLUSH || (flags & SYNC_WAIT) == 0);
1050
1051 again:
1052 xfs_qm_mplist_lock(mp);
1053 /*
1054 * dqpurge_all() also takes the mplist lock and iterate thru all dquots
1055 * in quotaoff. However, if the QUOTA_ACTIVE bits are not cleared
1056 * when we have the mplist lock, we know that dquots will be consistent
1057 * as long as we have it locked.
1058 */
1059 if (! XFS_IS_QUOTA_ON(mp)) {
1060 xfs_qm_mplist_unlock(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001061 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 }
1063 FOREACH_DQUOT_IN_MP(dqp, mp) {
1064 /*
1065 * If this is vfs_sync calling, then skip the dquots that
1066 * don't 'seem' to be dirty. ie. don't acquire dqlock.
1067 * This is very similar to what xfs_sync does with inodes.
1068 */
1069 if (flags & SYNC_BDFLUSH) {
1070 if (! XFS_DQ_IS_DIRTY(dqp))
1071 continue;
1072 }
1073
1074 if (nowait) {
1075 /*
1076 * Try to acquire the dquot lock. We are NOT out of
1077 * lock order, but we just don't want to wait for this
1078 * lock, unless somebody wanted us to.
1079 */
1080 if (! xfs_qm_dqlock_nowait(dqp))
1081 continue;
1082 } else {
1083 xfs_dqlock(dqp);
1084 }
1085
1086 /*
1087 * Now, find out for sure if this dquot is dirty or not.
1088 */
1089 if (! XFS_DQ_IS_DIRTY(dqp)) {
1090 xfs_dqunlock(dqp);
1091 continue;
1092 }
1093
1094 /* XXX a sentinel would be better */
1095 recl = XFS_QI_MPLRECLAIMS(mp);
1096 if (! xfs_qm_dqflock_nowait(dqp)) {
1097 if (nowait) {
1098 xfs_dqunlock(dqp);
1099 continue;
1100 }
1101 /*
1102 * If we can't grab the flush lock then if the caller
Nathan Scottc41564b2006-03-29 08:55:14 +10001103 * really wanted us to give this our best shot, so
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 * see if we can give a push to the buffer before we wait
1105 * on the flush lock. At this point, we know that
Nathan Scottc41564b2006-03-29 08:55:14 +10001106 * even though the dquot is being flushed,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 * it has (new) dirty data.
1108 */
1109 xfs_qm_dqflock_pushbuf_wait(dqp);
1110 }
1111 /*
1112 * Let go of the mplist lock. We don't want to hold it
1113 * across a disk write
1114 */
1115 flush_flags = (nowait) ? XFS_QMOPT_DELWRI : XFS_QMOPT_SYNC;
1116 xfs_qm_mplist_unlock(mp);
1117 xfs_dqtrace_entry(dqp, "XQM_SYNC: DQFLUSH");
1118 error = xfs_qm_dqflush(dqp, flush_flags);
1119 xfs_dqunlock(dqp);
1120 if (error && XFS_FORCED_SHUTDOWN(mp))
Jesper Juhl014c2542006-01-15 02:37:08 +01001121 return 0; /* Need to prevent umount failure */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 else if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001123 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 xfs_qm_mplist_lock(mp);
1126 if (recl != XFS_QI_MPLRECLAIMS(mp)) {
1127 if (++restarts >= XFS_QM_SYNC_MAX_RESTARTS)
1128 break;
1129
1130 xfs_qm_mplist_unlock(mp);
1131 goto again;
1132 }
1133 }
1134
1135 xfs_qm_mplist_unlock(mp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001136 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139
1140/*
1141 * This initializes all the quota information that's kept in the
1142 * mount structure
1143 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001144STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145xfs_qm_init_quotainfo(
1146 xfs_mount_t *mp)
1147{
1148 xfs_quotainfo_t *qinf;
1149 int error;
1150 xfs_dquot_t *dqp;
1151
1152 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1153
1154 /*
1155 * Tell XQM that we exist as soon as possible.
1156 */
1157 if ((error = xfs_qm_hold_quotafs_ref(mp))) {
Jesper Juhl014c2542006-01-15 02:37:08 +01001158 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
1160
1161 qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
1162
1163 /*
1164 * See if quotainodes are setup, and if not, allocate them,
1165 * and change the superblock accordingly.
1166 */
1167 if ((error = xfs_qm_init_quotainos(mp))) {
1168 kmem_free(qinf, sizeof(xfs_quotainfo_t));
1169 mp->m_quotainfo = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +01001170 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 }
1172
1173 spinlock_init(&qinf->qi_pinlock, "xfs_qinf_pin");
1174 xfs_qm_list_init(&qinf->qi_dqlist, "mpdqlist", 0);
1175 qinf->qi_dqreclaims = 0;
1176
1177 /* mutex used to serialize quotaoffs */
Jes Sorensen794ee1b2006-01-09 15:59:21 -08001178 mutex_init(&qinf->qi_quotaofflock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 /* Precalc some constants */
1181 qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1182 ASSERT(qinf->qi_dqchunklen);
1183 qinf->qi_dqperchunk = BBTOB(qinf->qi_dqchunklen);
1184 do_div(qinf->qi_dqperchunk, sizeof(xfs_dqblk_t));
1185
1186 mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
1187
1188 /*
1189 * We try to get the limits from the superuser's limits fields.
1190 * This is quite hacky, but it is standard quota practice.
1191 * We look at the USR dquot with id == 0 first, but if user quotas
1192 * are not enabled we goto the GRP dquot with id == 0.
1193 * We don't really care to keep separate default limits for user
1194 * and group quotas, at least not at this point.
1195 */
1196 error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)0,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001197 XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
1198 (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
1199 XFS_DQ_PROJ),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 XFS_QMOPT_DQSUSER|XFS_QMOPT_DOWARN,
1201 &dqp);
1202 if (! error) {
1203 xfs_disk_dquot_t *ddqp = &dqp->q_core;
1204
1205 /*
1206 * The warnings and timers set the grace period given to
1207 * a user or group before he or she can not perform any
1208 * more writing. If it is zero, a default is used.
1209 */
Christoph Hellwig1149d962005-11-02 15:01:12 +11001210 qinf->qi_btimelimit = ddqp->d_btimer ?
1211 be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
1212 qinf->qi_itimelimit = ddqp->d_itimer ?
1213 be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
1214 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
1215 be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
1216 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
1217 be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
1218 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
1219 be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
1220 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
1221 be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
1222 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
1223 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
1224 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
1225 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
1226 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
1227 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
1229 /*
1230 * We sent the XFS_QMOPT_DQSUSER flag to dqget because
1231 * we don't want this dquot cached. We haven't done a
1232 * quotacheck yet, and quotacheck doesn't like incore dquots.
1233 */
1234 xfs_qm_dqdestroy(dqp);
1235 } else {
1236 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
1237 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
1238 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
1239 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
1240 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
Nathan Scott06d10dd2005-06-21 15:48:47 +10001241 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 }
1243
Jesper Juhl014c2542006-01-15 02:37:08 +01001244 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245}
1246
1247
1248/*
1249 * Gets called when unmounting a filesystem or when all quotas get
1250 * turned off.
1251 * This purges the quota inodes, destroys locks and frees itself.
1252 */
1253void
1254xfs_qm_destroy_quotainfo(
1255 xfs_mount_t *mp)
1256{
1257 xfs_quotainfo_t *qi;
1258
1259 qi = mp->m_quotainfo;
1260 ASSERT(qi != NULL);
1261 ASSERT(xfs_Gqm != NULL);
1262
1263 /*
1264 * Release the reference that XQM kept, so that we know
1265 * when the XQM structure should be freed. We cannot assume
1266 * that xfs_Gqm is non-null after this point.
1267 */
1268 xfs_qm_rele_quotafs_ref(mp);
1269
1270 spinlock_destroy(&qi->qi_pinlock);
1271 xfs_qm_list_destroy(&qi->qi_dqlist);
1272
1273 if (qi->qi_uquotaip) {
1274 XFS_PURGE_INODE(qi->qi_uquotaip);
1275 qi->qi_uquotaip = NULL; /* paranoia */
1276 }
1277 if (qi->qi_gquotaip) {
1278 XFS_PURGE_INODE(qi->qi_gquotaip);
1279 qi->qi_gquotaip = NULL;
1280 }
1281 mutex_destroy(&qi->qi_quotaofflock);
1282 kmem_free(qi, sizeof(xfs_quotainfo_t));
1283 mp->m_quotainfo = NULL;
1284}
1285
1286
1287
1288/* ------------------- PRIVATE STATIC FUNCTIONS ----------------------- */
1289
1290/* ARGSUSED */
1291STATIC void
1292xfs_qm_list_init(
1293 xfs_dqlist_t *list,
1294 char *str,
1295 int n)
1296{
Jes Sorensen794ee1b2006-01-09 15:59:21 -08001297 mutex_init(&list->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 list->qh_next = NULL;
1299 list->qh_version = 0;
1300 list->qh_nelems = 0;
1301}
1302
1303STATIC void
1304xfs_qm_list_destroy(
1305 xfs_dqlist_t *list)
1306{
1307 mutex_destroy(&(list->qh_lock));
1308}
1309
1310
1311/*
1312 * Stripped down version of dqattach. This doesn't attach, or even look at the
1313 * dquots attached to the inode. The rationale is that there won't be any
1314 * attached at the time this is called from quotacheck.
1315 */
1316STATIC int
1317xfs_qm_dqget_noattach(
1318 xfs_inode_t *ip,
1319 xfs_dquot_t **O_udqpp,
1320 xfs_dquot_t **O_gdqpp)
1321{
1322 int error;
1323 xfs_mount_t *mp;
1324 xfs_dquot_t *udqp, *gdqp;
1325
1326 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
1327 mp = ip->i_mount;
1328 udqp = NULL;
1329 gdqp = NULL;
1330
1331 if (XFS_IS_UQUOTA_ON(mp)) {
1332 ASSERT(ip->i_udquot == NULL);
1333 /*
1334 * We want the dquot allocated if it doesn't exist.
1335 */
1336 if ((error = xfs_qm_dqget(mp, ip, ip->i_d.di_uid, XFS_DQ_USER,
1337 XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN,
1338 &udqp))) {
1339 /*
1340 * Shouldn't be able to turn off quotas here.
1341 */
1342 ASSERT(error != ESRCH);
1343 ASSERT(error != ENOENT);
Jesper Juhl014c2542006-01-15 02:37:08 +01001344 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 }
1346 ASSERT(udqp);
1347 }
1348
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001349 if (XFS_IS_OQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 ASSERT(ip->i_gdquot == NULL);
1351 if (udqp)
1352 xfs_dqunlock(udqp);
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001353 error = XFS_IS_GQUOTA_ON(mp) ?
1354 xfs_qm_dqget(mp, ip,
1355 ip->i_d.di_gid, XFS_DQ_GROUP,
1356 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1357 &gdqp) :
1358 xfs_qm_dqget(mp, ip,
1359 ip->i_d.di_projid, XFS_DQ_PROJ,
1360 XFS_QMOPT_DQALLOC|XFS_QMOPT_DOWARN,
1361 &gdqp);
1362 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 if (udqp)
1364 xfs_qm_dqrele(udqp);
1365 ASSERT(error != ESRCH);
1366 ASSERT(error != ENOENT);
Jesper Juhl014c2542006-01-15 02:37:08 +01001367 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
1369 ASSERT(gdqp);
1370
1371 /* Reacquire the locks in the right order */
1372 if (udqp) {
1373 if (! xfs_qm_dqlock_nowait(udqp)) {
1374 xfs_dqunlock(gdqp);
1375 xfs_dqlock(udqp);
1376 xfs_dqlock(gdqp);
1377 }
1378 }
1379 }
1380
1381 *O_udqpp = udqp;
1382 *O_gdqpp = gdqp;
1383
1384#ifdef QUOTADEBUG
1385 if (udqp) ASSERT(XFS_DQ_IS_LOCKED(udqp));
1386 if (gdqp) ASSERT(XFS_DQ_IS_LOCKED(gdqp));
1387#endif
Jesper Juhl014c2542006-01-15 02:37:08 +01001388 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
1390
1391/*
1392 * Create an inode and return with a reference already taken, but unlocked
1393 * This is how we create quota inodes
1394 */
1395STATIC int
1396xfs_qm_qino_alloc(
1397 xfs_mount_t *mp,
1398 xfs_inode_t **ip,
1399 __int64_t sbfields,
1400 uint flags)
1401{
1402 xfs_trans_t *tp;
1403 int error;
Nathan Scott061f7202006-01-11 15:27:50 +11001404 unsigned long s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 int committed;
1406
Nathan Scott061f7202006-01-11 15:27:50 +11001407 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 if ((error = xfs_trans_reserve(tp,
1409 XFS_QM_QINOCREATE_SPACE_RES(mp),
1410 XFS_CREATE_LOG_RES(mp), 0,
1411 XFS_TRANS_PERM_LOG_RES,
1412 XFS_CREATE_LOG_COUNT))) {
1413 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001414 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
David Chinnerb11f94d2007-07-11 11:09:33 +10001417 if ((error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0,
Nathan Scottdae81d42006-02-28 12:30:13 +11001418 &xfs_zerocr, 0, 1, ip, &committed))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1420 XFS_TRANS_ABORT);
Jesper Juhl014c2542006-01-15 02:37:08 +01001421 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423
1424 /*
1425 * Keep an extra reference to this quota inode. This inode is
1426 * locked exclusively and joined to the transaction already.
1427 */
1428 ASSERT(XFS_ISLOCKED_INODE_EXCL(*ip));
1429 VN_HOLD(XFS_ITOV((*ip)));
1430
1431 /*
1432 * Make the changes in the superblock, and log those too.
1433 * sbfields arg may contain fields other than *QUOTINO;
1434 * VERSIONNUM for example.
1435 */
1436 s = XFS_SB_LOCK(mp);
1437 if (flags & XFS_QMOPT_SBVERSION) {
1438#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1439 unsigned oldv = mp->m_sb.sb_versionnum;
1440#endif
1441 ASSERT(!XFS_SB_VERSION_HASQUOTA(&mp->m_sb));
1442 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1443 XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
1444 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1445 XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
1446
1447 XFS_SB_VERSION_ADDQUOTA(&mp->m_sb);
1448 mp->m_sb.sb_uquotino = NULLFSINO;
1449 mp->m_sb.sb_gquotino = NULLFSINO;
1450
1451 /* qflags will get updated _after_ quotacheck */
1452 mp->m_sb.sb_qflags = 0;
1453#if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
1454 cmn_err(CE_NOTE,
1455 "Old superblock version %x, converting to %x.",
1456 oldv, mp->m_sb.sb_versionnum);
1457#endif
1458 }
1459 if (flags & XFS_QMOPT_UQUOTA)
1460 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1461 else
1462 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1463 XFS_SB_UNLOCK(mp, s);
1464 xfs_mod_sb(tp, sbfields);
1465
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001466 if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 xfs_fs_cmn_err(CE_ALERT, mp, "XFS qino_alloc failed!");
Jesper Juhl014c2542006-01-15 02:37:08 +01001468 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001470 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471}
1472
1473
1474STATIC int
1475xfs_qm_reset_dqcounts(
1476 xfs_mount_t *mp,
1477 xfs_buf_t *bp,
1478 xfs_dqid_t id,
1479 uint type)
1480{
1481 xfs_disk_dquot_t *ddq;
1482 int j;
1483
1484 xfs_buftrace("RESET DQUOTS", bp);
1485 /*
1486 * Reset all counters and timers. They'll be
1487 * started afresh by xfs_qm_quotacheck.
1488 */
1489#ifdef DEBUG
1490 j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1491 do_div(j, sizeof(xfs_dqblk_t));
1492 ASSERT(XFS_QM_DQPERBLK(mp) == j);
1493#endif
1494 ddq = (xfs_disk_dquot_t *)XFS_BUF_PTR(bp);
1495 for (j = 0; j < XFS_QM_DQPERBLK(mp); j++) {
1496 /*
1497 * Do a sanity check, and if needed, repair the dqblk. Don't
1498 * output any warnings because it's perfectly possible to
Nathan Scottc41564b2006-03-29 08:55:14 +10001499 * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 */
1501 (void) xfs_qm_dqcheck(ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1502 "xfs_quotacheck");
Christoph Hellwig1149d962005-11-02 15:01:12 +11001503 ddq->d_bcount = 0;
1504 ddq->d_icount = 0;
1505 ddq->d_rtbcount = 0;
1506 ddq->d_btimer = 0;
1507 ddq->d_itimer = 0;
1508 ddq->d_rtbtimer = 0;
1509 ddq->d_bwarns = 0;
1510 ddq->d_iwarns = 0;
1511 ddq->d_rtbwarns = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
1513 }
1514
Jesper Juhl014c2542006-01-15 02:37:08 +01001515 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516}
1517
1518STATIC int
1519xfs_qm_dqiter_bufs(
1520 xfs_mount_t *mp,
1521 xfs_dqid_t firstid,
1522 xfs_fsblock_t bno,
1523 xfs_filblks_t blkcnt,
1524 uint flags)
1525{
1526 xfs_buf_t *bp;
1527 int error;
1528 int notcommitted;
1529 int incr;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001530 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
1532 ASSERT(blkcnt > 0);
1533 notcommitted = 0;
1534 incr = (blkcnt > XFS_QM_MAX_DQCLUSTER_LOGSZ) ?
1535 XFS_QM_MAX_DQCLUSTER_LOGSZ : blkcnt;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001536 type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1537 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 error = 0;
1539
1540 /*
1541 * Blkcnt arg can be a very big number, and might even be
1542 * larger than the log itself. So, we have to break it up into
1543 * manageable-sized transactions.
1544 * Note that we don't start a permanent transaction here; we might
1545 * not be able to get a log reservation for the whole thing up front,
1546 * and we don't really care to either, because we just discard
1547 * everything if we were to crash in the middle of this loop.
1548 */
1549 while (blkcnt--) {
1550 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1551 XFS_FSB_TO_DADDR(mp, bno),
1552 (int)XFS_QI_DQCHUNKLEN(mp), 0, &bp);
1553 if (error)
1554 break;
1555
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001556 (void) xfs_qm_reset_dqcounts(mp, bp, firstid, type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 xfs_bdwrite(mp, bp);
1558 /*
1559 * goto the next block.
1560 */
1561 bno++;
1562 firstid += XFS_QM_DQPERBLK(mp);
1563 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001564 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565}
1566
1567/*
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001568 * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 * caller supplied function for every chunk of dquots that we find.
1570 */
1571STATIC int
1572xfs_qm_dqiterate(
1573 xfs_mount_t *mp,
1574 xfs_inode_t *qip,
1575 uint flags)
1576{
1577 xfs_bmbt_irec_t *map;
1578 int i, nmaps; /* number of map entries */
1579 int error; /* return value */
1580 xfs_fileoff_t lblkno;
1581 xfs_filblks_t maxlblkcnt;
1582 xfs_dqid_t firstid;
1583 xfs_fsblock_t rablkno;
1584 xfs_filblks_t rablkcnt;
1585
1586 error = 0;
1587 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001588 * This looks racy, but we can't keep an inode lock across a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 * trans_reserve. But, this gets called during quotacheck, and that
1590 * happens only at mount time which is single threaded.
1591 */
1592 if (qip->i_d.di_nblocks == 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01001593 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595 map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1596
1597 lblkno = 0;
1598 maxlblkcnt = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1599 do {
1600 nmaps = XFS_DQITER_MAP_SIZE;
1601 /*
1602 * We aren't changing the inode itself. Just changing
1603 * some of its data. No new blocks are added here, and
1604 * the inode is never added to the transaction.
1605 */
1606 xfs_ilock(qip, XFS_ILOCK_SHARED);
1607 error = xfs_bmapi(NULL, qip, lblkno,
1608 maxlblkcnt - lblkno,
1609 XFS_BMAPI_METADATA,
1610 NULL,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001611 0, map, &nmaps, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1613 if (error)
1614 break;
1615
1616 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1617 for (i = 0; i < nmaps; i++) {
1618 ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1619 ASSERT(map[i].br_blockcount);
1620
1621
1622 lblkno += map[i].br_blockcount;
1623
1624 if (map[i].br_startblock == HOLESTARTBLOCK)
1625 continue;
1626
1627 firstid = (xfs_dqid_t) map[i].br_startoff *
1628 XFS_QM_DQPERBLK(mp);
1629 /*
1630 * Do a read-ahead on the next extent.
1631 */
1632 if ((i+1 < nmaps) &&
1633 (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1634 rablkcnt = map[i+1].br_blockcount;
1635 rablkno = map[i+1].br_startblock;
1636 while (rablkcnt--) {
1637 xfs_baread(mp->m_ddev_targp,
1638 XFS_FSB_TO_DADDR(mp, rablkno),
1639 (int)XFS_QI_DQCHUNKLEN(mp));
1640 rablkno++;
1641 }
1642 }
1643 /*
1644 * Iterate thru all the blks in the extent and
1645 * reset the counters of all the dquots inside them.
1646 */
1647 if ((error = xfs_qm_dqiter_bufs(mp,
1648 firstid,
1649 map[i].br_startblock,
1650 map[i].br_blockcount,
1651 flags))) {
1652 break;
1653 }
1654 }
1655
1656 if (error)
1657 break;
1658 } while (nmaps > 0);
1659
1660 kmem_free(map, XFS_DQITER_MAP_SIZE * sizeof(*map));
1661
Jesper Juhl014c2542006-01-15 02:37:08 +01001662 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663}
1664
1665/*
1666 * Called by dqusage_adjust in doing a quotacheck.
1667 * Given the inode, and a dquot (either USR or GRP, doesn't matter),
1668 * this updates its incore copy as well as the buffer copy. This is
1669 * so that once the quotacheck is done, we can just log all the buffers,
1670 * as opposed to logging numerous updates to individual dquots.
1671 */
1672STATIC void
1673xfs_qm_quotacheck_dqadjust(
1674 xfs_dquot_t *dqp,
1675 xfs_qcnt_t nblks,
1676 xfs_qcnt_t rtblks)
1677{
1678 ASSERT(XFS_DQ_IS_LOCKED(dqp));
1679 xfs_dqtrace_entry(dqp, "QCHECK DQADJUST");
1680 /*
1681 * Adjust the inode count and the block count to reflect this inode's
1682 * resource usage.
1683 */
Christoph Hellwig1149d962005-11-02 15:01:12 +11001684 be64_add(&dqp->q_core.d_icount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 dqp->q_res_icount++;
1686 if (nblks) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11001687 be64_add(&dqp->q_core.d_bcount, nblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 dqp->q_res_bcount += nblks;
1689 }
1690 if (rtblks) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11001691 be64_add(&dqp->q_core.d_rtbcount, rtblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 dqp->q_res_rtbcount += rtblks;
1693 }
1694
1695 /*
1696 * Set default limits, adjust timers (since we changed usages)
1697 */
1698 if (! XFS_IS_SUSER_DQUOT(dqp)) {
1699 xfs_qm_adjust_dqlimits(dqp->q_mount, &dqp->q_core);
1700 xfs_qm_adjust_dqtimers(dqp->q_mount, &dqp->q_core);
1701 }
1702
1703 dqp->dq_flags |= XFS_DQ_DIRTY;
1704}
1705
1706STATIC int
1707xfs_qm_get_rtblks(
1708 xfs_inode_t *ip,
1709 xfs_qcnt_t *O_rtblks)
1710{
1711 xfs_filblks_t rtblks; /* total rt blks */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001712 xfs_extnum_t idx; /* extent record index */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 xfs_ifork_t *ifp; /* inode fork pointer */
1714 xfs_extnum_t nextents; /* number of extent entries */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 xfs_bmbt_rec_t *ep; /* pointer to an extent entry */
1716 int error;
1717
1718 ASSERT(XFS_IS_REALTIME_INODE(ip));
1719 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1720 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1721 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001722 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 }
1724 rtblks = 0;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001725 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1726 for (idx = 0; idx < nextents; idx++) {
1727 ep = xfs_iext_get_ext(ifp, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 rtblks += xfs_bmbt_get_blockcount(ep);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 *O_rtblks = (xfs_qcnt_t)rtblks;
Jesper Juhl014c2542006-01-15 02:37:08 +01001731 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732}
1733
1734/*
1735 * callback routine supplied to bulkstat(). Given an inumber, find its
1736 * dquots and update them to account for resources taken by that inode.
1737 */
1738/* ARGSUSED */
1739STATIC int
1740xfs_qm_dqusage_adjust(
1741 xfs_mount_t *mp, /* mount point for filesystem */
1742 xfs_ino_t ino, /* inode number to get data for */
1743 void __user *buffer, /* not used */
1744 int ubsize, /* not used */
1745 void *private_data, /* not used */
1746 xfs_daddr_t bno, /* starting block of inode cluster */
1747 int *ubused, /* not used */
1748 void *dip, /* on-disk inode pointer (not used) */
1749 int *res) /* result code value */
1750{
1751 xfs_inode_t *ip;
1752 xfs_dquot_t *udqp, *gdqp;
1753 xfs_qcnt_t nblks, rtblks;
1754 int error;
1755
1756 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1757
1758 /*
1759 * rootino must have its resources accounted for, not so with the quota
1760 * inodes.
1761 */
1762 if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1763 *res = BULKSTAT_RV_NOTHING;
1764 return XFS_ERROR(EINVAL);
1765 }
1766
1767 /*
1768 * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1769 * interface expects the inode to be exclusively locked because that's
1770 * the case in all other instances. It's OK that we do this because
1771 * quotacheck is done only at mount time.
1772 */
1773 if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip, bno))) {
1774 *res = BULKSTAT_RV_NOTHING;
Jesper Juhl014c2542006-01-15 02:37:08 +01001775 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 }
1777
1778 if (ip->i_d.di_mode == 0) {
1779 xfs_iput_new(ip, XFS_ILOCK_EXCL);
1780 *res = BULKSTAT_RV_NOTHING;
1781 return XFS_ERROR(ENOENT);
1782 }
1783
1784 /*
1785 * Obtain the locked dquots. In case of an error (eg. allocation
1786 * fails for ENOSPC), we return the negative of the error number
1787 * to bulkstat, so that it can get propagated to quotacheck() and
1788 * making us disable quotas for the file system.
1789 */
1790 if ((error = xfs_qm_dqget_noattach(ip, &udqp, &gdqp))) {
1791 xfs_iput(ip, XFS_ILOCK_EXCL);
1792 *res = BULKSTAT_RV_GIVEUP;
Jesper Juhl014c2542006-01-15 02:37:08 +01001793 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 }
1795
1796 rtblks = 0;
1797 if (! XFS_IS_REALTIME_INODE(ip)) {
1798 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks;
1799 } else {
1800 /*
1801 * Walk thru the extent list and count the realtime blocks.
1802 */
1803 if ((error = xfs_qm_get_rtblks(ip, &rtblks))) {
1804 xfs_iput(ip, XFS_ILOCK_EXCL);
1805 if (udqp)
1806 xfs_qm_dqput(udqp);
1807 if (gdqp)
1808 xfs_qm_dqput(gdqp);
1809 *res = BULKSTAT_RV_GIVEUP;
Jesper Juhl014c2542006-01-15 02:37:08 +01001810 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 }
1812 nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1813 }
1814 ASSERT(ip->i_delayed_blks == 0);
1815
1816 /*
1817 * We can't release the inode while holding its dquot locks.
1818 * The inode can go into inactive and might try to acquire the dquotlocks.
1819 * So, just unlock here and do a vn_rele at the end.
1820 */
1821 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1822
1823 /*
1824 * Add the (disk blocks and inode) resources occupied by this
1825 * inode to its dquots. We do this adjustment in the incore dquot,
1826 * and also copy the changes to its buffer.
1827 * We don't care about putting these changes in a transaction
1828 * envelope because if we crash in the middle of a 'quotacheck'
1829 * we have to start from the beginning anyway.
1830 * Once we're done, we'll log all the dquot bufs.
1831 *
Nathan Scottc41564b2006-03-29 08:55:14 +10001832 * The *QUOTA_ON checks below may look pretty racy, but quotachecks
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 * and quotaoffs don't race. (Quotachecks happen at mount time only).
1834 */
1835 if (XFS_IS_UQUOTA_ON(mp)) {
1836 ASSERT(udqp);
1837 xfs_qm_quotacheck_dqadjust(udqp, nblks, rtblks);
1838 xfs_qm_dqput(udqp);
1839 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001840 if (XFS_IS_OQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 ASSERT(gdqp);
1842 xfs_qm_quotacheck_dqadjust(gdqp, nblks, rtblks);
1843 xfs_qm_dqput(gdqp);
1844 }
1845 /*
1846 * Now release the inode. This will send it to 'inactive', and
1847 * possibly even free blocks.
1848 */
1849 VN_RELE(XFS_ITOV(ip));
1850
1851 /*
1852 * Goto next inode.
1853 */
1854 *res = BULKSTAT_RV_DIDONE;
Jesper Juhl014c2542006-01-15 02:37:08 +01001855 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856}
1857
1858/*
1859 * Walk thru all the filesystem inodes and construct a consistent view
1860 * of the disk quota world. If the quotacheck fails, disable quotas.
1861 */
1862int
1863xfs_qm_quotacheck(
1864 xfs_mount_t *mp)
1865{
1866 int done, count, error;
1867 xfs_ino_t lastino;
1868 size_t structsz;
1869 xfs_inode_t *uip, *gip;
1870 uint flags;
1871
1872 count = INT_MAX;
1873 structsz = 1;
1874 lastino = 0;
1875 flags = 0;
1876
1877 ASSERT(XFS_QI_UQIP(mp) || XFS_QI_GQIP(mp));
1878 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1879
1880 /*
1881 * There should be no cached dquots. The (simplistic) quotacheck
1882 * algorithm doesn't like that.
1883 */
1884 ASSERT(XFS_QI_MPLNDQUOTS(mp) == 0);
1885
1886 cmn_err(CE_NOTE, "XFS quotacheck %s: Please wait.", mp->m_fsname);
1887
1888 /*
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001889 * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 * their counters to zero. We need a clean slate.
1891 * We don't log our changes till later.
1892 */
1893 if ((uip = XFS_QI_UQIP(mp))) {
1894 if ((error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA)))
1895 goto error_return;
1896 flags |= XFS_UQUOTA_CHKD;
1897 }
1898
1899 if ((gip = XFS_QI_GQIP(mp))) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001900 if ((error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1901 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 goto error_return;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001903 flags |= XFS_OQUOTA_CHKD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
1905
1906 do {
1907 /*
1908 * Iterate thru all the inodes in the file system,
1909 * adjusting the corresponding dquot counters in core.
1910 */
1911 if ((error = xfs_bulkstat(mp, &lastino, &count,
1912 xfs_qm_dqusage_adjust, NULL,
Nathan Scott9c488762006-06-09 15:29:22 +10001913 structsz, NULL, BULKSTAT_FG_IGET, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 break;
1915
1916 } while (! done);
1917
1918 /*
1919 * We can get this error if we couldn't do a dquot allocation inside
1920 * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1921 * dirty dquots that might be cached, we just want to get rid of them
1922 * and turn quotaoff. The dquots won't be attached to any of the inodes
1923 * at this point (because we intentionally didn't in dqget_noattach).
1924 */
1925 if (error) {
Nathan Scottee2a4f72006-01-11 15:33:36 +11001926 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_QUOTAOFF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 goto error_return;
1928 }
1929 /*
1930 * We've made all the changes that we need to make incore.
1931 * Now flush_them down to disk buffers.
1932 */
1933 xfs_qm_dqflush_all(mp, XFS_QMOPT_DELWRI);
1934
1935 /*
1936 * We didn't log anything, because if we crashed, we'll have to
1937 * start the quotacheck from scratch anyway. However, we must make
1938 * sure that our dquot changes are secure before we put the
1939 * quotacheck'd stamp on the superblock. So, here we do a synchronous
1940 * flush.
1941 */
1942 XFS_bflush(mp->m_ddev_targp);
1943
1944 /*
1945 * If one type of quotas is off, then it will lose its
1946 * quotachecked status, since we won't be doing accounting for
1947 * that type anymore.
1948 */
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001949 mp->m_qflags &= ~(XFS_OQUOTA_CHKD | XFS_UQUOTA_CHKD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 mp->m_qflags |= flags;
1951
1952 XQM_LIST_PRINT(&(XFS_QI_MPL_LIST(mp)), MPL_NEXT, "++++ Mp list +++");
1953
1954 error_return:
1955 if (error) {
1956 cmn_err(CE_WARN, "XFS quotacheck %s: Unsuccessful (Error %d): "
1957 "Disabling quotas.",
1958 mp->m_fsname, error);
1959 /*
1960 * We must turn off quotas.
1961 */
1962 ASSERT(mp->m_quotainfo != NULL);
1963 ASSERT(xfs_Gqm != NULL);
1964 xfs_qm_destroy_quotainfo(mp);
Tim Shimmin4cd4a032005-09-05 08:24:10 +10001965 (void)xfs_mount_reset_sbqflags(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 } else {
1967 cmn_err(CE_NOTE, "XFS quotacheck %s: Done.", mp->m_fsname);
1968 }
1969 return (error);
1970}
1971
1972/*
1973 * This is called after the superblock has been read in and we're ready to
1974 * iget the quota inodes.
1975 */
1976STATIC int
1977xfs_qm_init_quotainos(
1978 xfs_mount_t *mp)
1979{
1980 xfs_inode_t *uip, *gip;
1981 int error;
1982 __int64_t sbflags;
1983 uint flags;
1984
1985 ASSERT(mp->m_quotainfo);
1986 uip = gip = NULL;
1987 sbflags = 0;
1988 flags = 0;
1989
1990 /*
1991 * Get the uquota and gquota inodes
1992 */
1993 if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb)) {
1994 if (XFS_IS_UQUOTA_ON(mp) &&
1995 mp->m_sb.sb_uquotino != NULLFSINO) {
1996 ASSERT(mp->m_sb.sb_uquotino > 0);
1997 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1998 0, 0, &uip, 0)))
1999 return XFS_ERROR(error);
2000 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002001 if (XFS_IS_OQUOTA_ON(mp) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 mp->m_sb.sb_gquotino != NULLFSINO) {
2003 ASSERT(mp->m_sb.sb_gquotino > 0);
2004 if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
2005 0, 0, &gip, 0))) {
2006 if (uip)
2007 VN_RELE(XFS_ITOV(uip));
2008 return XFS_ERROR(error);
2009 }
2010 }
2011 } else {
2012 flags |= XFS_QMOPT_SBVERSION;
2013 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
2014 XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
2015 }
2016
2017 /*
2018 * Create the two inodes, if they don't exist already. The changes
2019 * made above will get added to a transaction and logged in one of
2020 * the qino_alloc calls below. If the device is readonly,
2021 * temporarily switch to read-write to do this.
2022 */
2023 if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
2024 if ((error = xfs_qm_qino_alloc(mp, &uip,
2025 sbflags | XFS_SB_UQUOTINO,
2026 flags | XFS_QMOPT_UQUOTA)))
2027 return XFS_ERROR(error);
2028
2029 flags &= ~XFS_QMOPT_SBVERSION;
2030 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002031 if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
2032 flags |= (XFS_IS_GQUOTA_ON(mp) ?
2033 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
2034 error = xfs_qm_qino_alloc(mp, &gip,
2035 sbflags | XFS_SB_GQUOTINO, flags);
2036 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 if (uip)
2038 VN_RELE(XFS_ITOV(uip));
2039
2040 return XFS_ERROR(error);
2041 }
2042 }
2043
2044 XFS_QI_UQIP(mp) = uip;
2045 XFS_QI_GQIP(mp) = gip;
2046
Jesper Juhl014c2542006-01-15 02:37:08 +01002047 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048}
2049
2050
2051/*
2052 * Traverse the freelist of dquots and attempt to reclaim a maximum of
2053 * 'howmany' dquots. This operation races with dqlookup(), and attempts to
2054 * favor the lookup function ...
2055 * XXXsup merge this with qm_reclaim_one().
2056 */
2057STATIC int
2058xfs_qm_shake_freelist(
2059 int howmany)
2060{
2061 int nreclaimed;
2062 xfs_dqhash_t *hash;
2063 xfs_dquot_t *dqp, *nextdqp;
2064 int restarts;
2065 int nflushes;
2066
2067 if (howmany <= 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01002068 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
2070 nreclaimed = 0;
2071 restarts = 0;
2072 nflushes = 0;
2073
2074#ifdef QUOTADEBUG
2075 cmn_err(CE_DEBUG, "Shake free 0x%x", howmany);
2076#endif
2077 /* lock order is : hashchainlock, freelistlock, mplistlock */
2078 tryagain:
2079 xfs_qm_freelist_lock(xfs_Gqm);
2080
2081 for (dqp = xfs_Gqm->qm_dqfreelist.qh_next;
2082 ((dqp != (xfs_dquot_t *) &xfs_Gqm->qm_dqfreelist) &&
2083 nreclaimed < howmany); ) {
2084 xfs_dqlock(dqp);
2085
2086 /*
2087 * We are racing with dqlookup here. Naturally we don't
2088 * want to reclaim a dquot that lookup wants.
2089 */
2090 if (dqp->dq_flags & XFS_DQ_WANT) {
2091 xfs_dqunlock(dqp);
2092 xfs_qm_freelist_unlock(xfs_Gqm);
2093 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
Jesper Juhl014c2542006-01-15 02:37:08 +01002094 return nreclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
2096 goto tryagain;
2097 }
2098
2099 /*
2100 * If the dquot is inactive, we are assured that it is
2101 * not on the mplist or the hashlist, and that makes our
2102 * life easier.
2103 */
2104 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2105 ASSERT(dqp->q_mount == NULL);
2106 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2107 ASSERT(dqp->HL_PREVP == NULL);
2108 ASSERT(dqp->MPL_PREVP == NULL);
2109 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2110 nextdqp = dqp->dq_flnext;
2111 goto off_freelist;
2112 }
2113
2114 ASSERT(dqp->MPL_PREVP);
2115 /*
2116 * Try to grab the flush lock. If this dquot is in the process of
2117 * getting flushed to disk, we don't want to reclaim it.
2118 */
2119 if (! xfs_qm_dqflock_nowait(dqp)) {
2120 xfs_dqunlock(dqp);
2121 dqp = dqp->dq_flnext;
2122 continue;
2123 }
2124
2125 /*
2126 * We have the flush lock so we know that this is not in the
2127 * process of being flushed. So, if this is dirty, flush it
2128 * DELWRI so that we don't get a freelist infested with
2129 * dirty dquots.
2130 */
2131 if (XFS_DQ_IS_DIRTY(dqp)) {
2132 xfs_dqtrace_entry(dqp, "DQSHAKE: DQDIRTY");
2133 /*
2134 * We flush it delayed write, so don't bother
2135 * releasing the mplock.
2136 */
2137 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2138 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2139 dqp = dqp->dq_flnext;
2140 continue;
2141 }
2142 /*
2143 * We're trying to get the hashlock out of order. This races
2144 * with dqlookup; so, we giveup and goto the next dquot if
2145 * we couldn't get the hashlock. This way, we won't starve
2146 * a dqlookup process that holds the hashlock that is
2147 * waiting for the freelist lock.
2148 */
2149 if (! xfs_qm_dqhashlock_nowait(dqp)) {
2150 xfs_dqfunlock(dqp);
2151 xfs_dqunlock(dqp);
2152 dqp = dqp->dq_flnext;
2153 continue;
2154 }
2155 /*
2156 * This races with dquot allocation code as well as dqflush_all
2157 * and reclaim code. So, if we failed to grab the mplist lock,
2158 * giveup everything and start over.
2159 */
2160 hash = dqp->q_hash;
2161 ASSERT(hash);
2162 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2163 /* XXX put a sentinel so that we can come back here */
2164 xfs_dqfunlock(dqp);
2165 xfs_dqunlock(dqp);
2166 XFS_DQ_HASH_UNLOCK(hash);
2167 xfs_qm_freelist_unlock(xfs_Gqm);
2168 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
Jesper Juhl014c2542006-01-15 02:37:08 +01002169 return nreclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 goto tryagain;
2171 }
2172 xfs_dqtrace_entry(dqp, "DQSHAKE: UNLINKING");
2173#ifdef QUOTADEBUG
2174 cmn_err(CE_DEBUG, "Shake 0x%p, ID 0x%x\n",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002175 dqp, be32_to_cpu(dqp->q_core.d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176#endif
2177 ASSERT(dqp->q_nrefs == 0);
2178 nextdqp = dqp->dq_flnext;
2179 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2180 XQM_HASHLIST_REMOVE(hash, dqp);
2181 xfs_dqfunlock(dqp);
2182 xfs_qm_mplist_unlock(dqp->q_mount);
2183 XFS_DQ_HASH_UNLOCK(hash);
2184
2185 off_freelist:
2186 XQM_FREELIST_REMOVE(dqp);
2187 xfs_dqunlock(dqp);
2188 nreclaimed++;
2189 XQM_STATS_INC(xqmstats.xs_qm_dqshake_reclaims);
2190 xfs_qm_dqdestroy(dqp);
2191 dqp = nextdqp;
2192 }
2193 xfs_qm_freelist_unlock(xfs_Gqm);
Jesper Juhl014c2542006-01-15 02:37:08 +01002194 return nreclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195}
2196
2197
2198/*
2199 * The kmem_shake interface is invoked when memory is running low.
2200 */
2201/* ARGSUSED */
2202STATIC int
Al Viro51bfb752005-12-15 09:17:24 +00002203xfs_qm_shake(int nr_to_scan, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204{
2205 int ndqused, nfree, n;
2206
2207 if (!kmem_shake_allow(gfp_mask))
Jesper Juhl014c2542006-01-15 02:37:08 +01002208 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 if (!xfs_Gqm)
Jesper Juhl014c2542006-01-15 02:37:08 +01002210 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211
2212 nfree = xfs_Gqm->qm_dqfreelist.qh_nelems; /* free dquots */
2213 /* incore dquots in all f/s's */
2214 ndqused = atomic_read(&xfs_Gqm->qm_totaldquots) - nfree;
2215
2216 ASSERT(ndqused >= 0);
2217
2218 if (nfree <= ndqused && nfree < ndquot)
Jesper Juhl014c2542006-01-15 02:37:08 +01002219 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
2221 ndqused *= xfs_Gqm->qm_dqfree_ratio; /* target # of free dquots */
2222 n = nfree - ndqused - ndquot; /* # over target */
2223
2224 return xfs_qm_shake_freelist(MAX(nfree, n));
2225}
2226
2227
2228/*
2229 * Just pop the least recently used dquot off the freelist and
2230 * recycle it. The returned dquot is locked.
2231 */
2232STATIC xfs_dquot_t *
2233xfs_qm_dqreclaim_one(void)
2234{
2235 xfs_dquot_t *dqpout;
2236 xfs_dquot_t *dqp;
2237 int restarts;
2238 int nflushes;
2239
2240 restarts = 0;
2241 dqpout = NULL;
2242 nflushes = 0;
2243
2244 /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */
2245 startagain:
2246 xfs_qm_freelist_lock(xfs_Gqm);
2247
2248 FOREACH_DQUOT_IN_FREELIST(dqp, &(xfs_Gqm->qm_dqfreelist)) {
2249 xfs_dqlock(dqp);
2250
2251 /*
2252 * We are racing with dqlookup here. Naturally we don't
2253 * want to reclaim a dquot that lookup wants. We release the
2254 * freelist lock and start over, so that lookup will grab
2255 * both the dquot and the freelistlock.
2256 */
2257 if (dqp->dq_flags & XFS_DQ_WANT) {
2258 ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE));
2259 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQWANT");
2260 xfs_dqunlock(dqp);
2261 xfs_qm_freelist_unlock(xfs_Gqm);
2262 if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS)
Jesper Juhl014c2542006-01-15 02:37:08 +01002263 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 XQM_STATS_INC(xqmstats.xs_qm_dqwants);
2265 goto startagain;
2266 }
2267
2268 /*
2269 * If the dquot is inactive, we are assured that it is
2270 * not on the mplist or the hashlist, and that makes our
2271 * life easier.
2272 */
2273 if (dqp->dq_flags & XFS_DQ_INACTIVE) {
2274 ASSERT(dqp->q_mount == NULL);
2275 ASSERT(! XFS_DQ_IS_DIRTY(dqp));
2276 ASSERT(dqp->HL_PREVP == NULL);
2277 ASSERT(dqp->MPL_PREVP == NULL);
2278 XQM_FREELIST_REMOVE(dqp);
2279 xfs_dqunlock(dqp);
2280 dqpout = dqp;
2281 XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims);
2282 break;
2283 }
2284
2285 ASSERT(dqp->q_hash);
2286 ASSERT(dqp->MPL_PREVP);
2287
2288 /*
2289 * Try to grab the flush lock. If this dquot is in the process of
2290 * getting flushed to disk, we don't want to reclaim it.
2291 */
2292 if (! xfs_qm_dqflock_nowait(dqp)) {
2293 xfs_dqunlock(dqp);
2294 continue;
2295 }
2296
2297 /*
2298 * We have the flush lock so we know that this is not in the
2299 * process of being flushed. So, if this is dirty, flush it
2300 * DELWRI so that we don't get a freelist infested with
2301 * dirty dquots.
2302 */
2303 if (XFS_DQ_IS_DIRTY(dqp)) {
2304 xfs_dqtrace_entry(dqp, "DQRECLAIM: DQDIRTY");
2305 /*
2306 * We flush it delayed write, so don't bother
2307 * releasing the freelist lock.
2308 */
2309 (void) xfs_qm_dqflush(dqp, XFS_QMOPT_DELWRI);
2310 xfs_dqunlock(dqp); /* dqflush unlocks dqflock */
2311 continue;
2312 }
2313
2314 if (! xfs_qm_mplist_nowait(dqp->q_mount)) {
2315 xfs_dqfunlock(dqp);
2316 xfs_dqunlock(dqp);
2317 continue;
2318 }
2319
2320 if (! xfs_qm_dqhashlock_nowait(dqp))
2321 goto mplistunlock;
2322
2323 ASSERT(dqp->q_nrefs == 0);
2324 xfs_dqtrace_entry(dqp, "DQRECLAIM: UNLINKING");
2325 XQM_MPLIST_REMOVE(&(XFS_QI_MPL_LIST(dqp->q_mount)), dqp);
2326 XQM_HASHLIST_REMOVE(dqp->q_hash, dqp);
2327 XQM_FREELIST_REMOVE(dqp);
2328 dqpout = dqp;
2329 XFS_DQ_HASH_UNLOCK(dqp->q_hash);
2330 mplistunlock:
2331 xfs_qm_mplist_unlock(dqp->q_mount);
2332 xfs_dqfunlock(dqp);
2333 xfs_dqunlock(dqp);
2334 if (dqpout)
2335 break;
2336 }
2337
2338 xfs_qm_freelist_unlock(xfs_Gqm);
Jesper Juhl014c2542006-01-15 02:37:08 +01002339 return dqpout;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340}
2341
2342
2343/*------------------------------------------------------------------*/
2344
2345/*
2346 * Return a new incore dquot. Depending on the number of
2347 * dquots in the system, we either allocate a new one on the kernel heap,
2348 * or reclaim a free one.
2349 * Return value is B_TRUE if we allocated a new dquot, B_FALSE if we managed
2350 * to reclaim an existing one from the freelist.
2351 */
2352boolean_t
2353xfs_qm_dqalloc_incore(
2354 xfs_dquot_t **O_dqpp)
2355{
2356 xfs_dquot_t *dqp;
2357
2358 /*
2359 * Check against high water mark to see if we want to pop
2360 * a nincompoop dquot off the freelist.
2361 */
2362 if (atomic_read(&xfs_Gqm->qm_totaldquots) >= ndquot) {
2363 /*
2364 * Try to recycle a dquot from the freelist.
2365 */
2366 if ((dqp = xfs_qm_dqreclaim_one())) {
2367 XQM_STATS_INC(xqmstats.xs_qm_dqreclaims);
2368 /*
2369 * Just zero the core here. The rest will get
2370 * reinitialized by caller. XXX we shouldn't even
2371 * do this zero ...
2372 */
2373 memset(&dqp->q_core, 0, sizeof(dqp->q_core));
2374 *O_dqpp = dqp;
Jesper Juhl014c2542006-01-15 02:37:08 +01002375 return B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 }
2377 XQM_STATS_INC(xqmstats.xs_qm_dqreclaim_misses);
2378 }
2379
2380 /*
2381 * Allocate a brand new dquot on the kernel heap and return it
2382 * to the caller to initialize.
2383 */
2384 ASSERT(xfs_Gqm->qm_dqzone != NULL);
2385 *O_dqpp = kmem_zone_zalloc(xfs_Gqm->qm_dqzone, KM_SLEEP);
2386 atomic_inc(&xfs_Gqm->qm_totaldquots);
2387
Jesper Juhl014c2542006-01-15 02:37:08 +01002388 return B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389}
2390
2391
2392/*
2393 * Start a transaction and write the incore superblock changes to
2394 * disk. flags parameter indicates which fields have changed.
2395 */
2396int
2397xfs_qm_write_sb_changes(
2398 xfs_mount_t *mp,
2399 __int64_t flags)
2400{
2401 xfs_trans_t *tp;
2402 int error;
2403
2404#ifdef QUOTADEBUG
2405 cmn_err(CE_NOTE, "Writing superblock quota changes :%s", mp->m_fsname);
2406#endif
2407 tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
2408 if ((error = xfs_trans_reserve(tp, 0,
2409 mp->m_sb.sb_sectsize + 128, 0,
2410 0,
2411 XFS_DEFAULT_LOG_COUNT))) {
2412 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01002413 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 }
2415
2416 xfs_mod_sb(tp, flags);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002417 (void) xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
Jesper Juhl014c2542006-01-15 02:37:08 +01002419 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420}
2421
2422
2423/* --------------- utility functions for vnodeops ---------------- */
2424
2425
2426/*
2427 * Given an inode, a uid and gid (from cred_t) make sure that we have
2428 * allocated relevant dquot(s) on disk, and that we won't exceed inode
2429 * quotas by creating this file.
2430 * This also attaches dquot(s) to the given inode after locking it,
2431 * and returns the dquots corresponding to the uid and/or gid.
2432 *
2433 * in : inode (unlocked)
2434 * out : udquot, gdquot with references taken and unlocked
2435 */
2436int
2437xfs_qm_vop_dqalloc(
2438 xfs_mount_t *mp,
2439 xfs_inode_t *ip,
2440 uid_t uid,
2441 gid_t gid,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002442 prid_t prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 uint flags,
2444 xfs_dquot_t **O_udqpp,
2445 xfs_dquot_t **O_gdqpp)
2446{
2447 int error;
2448 xfs_dquot_t *uq, *gq;
2449 uint lockflags;
2450
2451 if (!XFS_IS_QUOTA_ON(mp))
2452 return 0;
2453
2454 lockflags = XFS_ILOCK_EXCL;
2455 xfs_ilock(ip, lockflags);
2456
2457 if ((flags & XFS_QMOPT_INHERIT) &&
2458 XFS_INHERIT_GID(ip, XFS_MTOVFS(mp)))
2459 gid = ip->i_d.di_gid;
2460
2461 /*
2462 * Attach the dquot(s) to this inode, doing a dquot allocation
2463 * if necessary. The dquot(s) will not be locked.
2464 */
2465 if (XFS_NOT_DQATTACHED(mp, ip)) {
2466 if ((error = xfs_qm_dqattach(ip, XFS_QMOPT_DQALLOC |
2467 XFS_QMOPT_ILOCKED))) {
2468 xfs_iunlock(ip, lockflags);
Jesper Juhl014c2542006-01-15 02:37:08 +01002469 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 }
2471 }
2472
2473 uq = gq = NULL;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002474 if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 if (ip->i_d.di_uid != uid) {
2476 /*
2477 * What we need is the dquot that has this uid, and
2478 * if we send the inode to dqget, the uid of the inode
2479 * takes priority over what's sent in the uid argument.
2480 * We must unlock inode here before calling dqget if
2481 * we're not sending the inode, because otherwise
2482 * we'll deadlock by doing trans_reserve while
2483 * holding ilock.
2484 */
2485 xfs_iunlock(ip, lockflags);
2486 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
2487 XFS_DQ_USER,
2488 XFS_QMOPT_DQALLOC |
2489 XFS_QMOPT_DOWARN,
2490 &uq))) {
2491 ASSERT(error != ENOENT);
Jesper Juhl014c2542006-01-15 02:37:08 +01002492 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 }
2494 /*
2495 * Get the ilock in the right order.
2496 */
2497 xfs_dqunlock(uq);
2498 lockflags = XFS_ILOCK_SHARED;
2499 xfs_ilock(ip, lockflags);
2500 } else {
2501 /*
2502 * Take an extra reference, because we'll return
2503 * this to caller
2504 */
2505 ASSERT(ip->i_udquot);
2506 uq = ip->i_udquot;
2507 xfs_dqlock(uq);
2508 XFS_DQHOLD(uq);
2509 xfs_dqunlock(uq);
2510 }
2511 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002512 if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 if (ip->i_d.di_gid != gid) {
2514 xfs_iunlock(ip, lockflags);
2515 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
2516 XFS_DQ_GROUP,
2517 XFS_QMOPT_DQALLOC |
2518 XFS_QMOPT_DOWARN,
2519 &gq))) {
2520 if (uq)
2521 xfs_qm_dqrele(uq);
2522 ASSERT(error != ENOENT);
Jesper Juhl014c2542006-01-15 02:37:08 +01002523 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 }
2525 xfs_dqunlock(gq);
2526 lockflags = XFS_ILOCK_SHARED;
2527 xfs_ilock(ip, lockflags);
2528 } else {
2529 ASSERT(ip->i_gdquot);
2530 gq = ip->i_gdquot;
2531 xfs_dqlock(gq);
2532 XFS_DQHOLD(gq);
2533 xfs_dqunlock(gq);
2534 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002535 } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
2536 if (ip->i_d.di_projid != prid) {
2537 xfs_iunlock(ip, lockflags);
2538 if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
2539 XFS_DQ_PROJ,
2540 XFS_QMOPT_DQALLOC |
2541 XFS_QMOPT_DOWARN,
2542 &gq))) {
2543 if (uq)
2544 xfs_qm_dqrele(uq);
2545 ASSERT(error != ENOENT);
2546 return (error);
2547 }
2548 xfs_dqunlock(gq);
2549 lockflags = XFS_ILOCK_SHARED;
2550 xfs_ilock(ip, lockflags);
2551 } else {
2552 ASSERT(ip->i_gdquot);
2553 gq = ip->i_gdquot;
2554 xfs_dqlock(gq);
2555 XFS_DQHOLD(gq);
2556 xfs_dqunlock(gq);
2557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 }
2559 if (uq)
2560 xfs_dqtrace_entry_ino(uq, "DQALLOC", ip);
2561
2562 xfs_iunlock(ip, lockflags);
2563 if (O_udqpp)
2564 *O_udqpp = uq;
2565 else if (uq)
2566 xfs_qm_dqrele(uq);
2567 if (O_gdqpp)
2568 *O_gdqpp = gq;
2569 else if (gq)
2570 xfs_qm_dqrele(gq);
Jesper Juhl014c2542006-01-15 02:37:08 +01002571 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572}
2573
2574/*
2575 * Actually transfer ownership, and do dquot modifications.
2576 * These were already reserved.
2577 */
2578xfs_dquot_t *
2579xfs_qm_vop_chown(
2580 xfs_trans_t *tp,
2581 xfs_inode_t *ip,
2582 xfs_dquot_t **IO_olddq,
2583 xfs_dquot_t *newdq)
2584{
2585 xfs_dquot_t *prevdq;
Nathan Scott06d10dd2005-06-21 15:48:47 +10002586 uint bfield = XFS_IS_REALTIME_INODE(ip) ?
2587 XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
2588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
2590 ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
2591
2592 /* old dquot */
2593 prevdq = *IO_olddq;
2594 ASSERT(prevdq);
2595 ASSERT(prevdq != newdq);
2596
Nathan Scott06d10dd2005-06-21 15:48:47 +10002597 xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
2598 xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599
2600 /* the sparkling new dquot */
Nathan Scott06d10dd2005-06-21 15:48:47 +10002601 xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
2602 xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
2604 /*
2605 * Take an extra reference, because the inode
2606 * is going to keep this dquot pointer even
2607 * after the trans_commit.
2608 */
2609 xfs_dqlock(newdq);
2610 XFS_DQHOLD(newdq);
2611 xfs_dqunlock(newdq);
2612 *IO_olddq = newdq;
2613
Jesper Juhl014c2542006-01-15 02:37:08 +01002614 return prevdq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615}
2616
2617/*
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002618 * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 */
2620int
2621xfs_qm_vop_chown_reserve(
2622 xfs_trans_t *tp,
2623 xfs_inode_t *ip,
2624 xfs_dquot_t *udqp,
2625 xfs_dquot_t *gdqp,
2626 uint flags)
2627{
2628 int error;
2629 xfs_mount_t *mp;
Nathan Scott9a2a7de2006-03-31 13:04:49 +10002630 uint delblks, blkflags, prjflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 xfs_dquot_t *unresudq, *unresgdq, *delblksudq, *delblksgdq;
2632
2633 ASSERT(XFS_ISLOCKED_INODE(ip));
2634 mp = ip->i_mount;
2635 ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2636
2637 delblks = ip->i_delayed_blks;
2638 delblksudq = delblksgdq = unresudq = unresgdq = NULL;
Nathan Scott06d10dd2005-06-21 15:48:47 +10002639 blkflags = XFS_IS_REALTIME_INODE(ip) ?
2640 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
2642 if (XFS_IS_UQUOTA_ON(mp) && udqp &&
Christoph Hellwig1149d962005-11-02 15:01:12 +11002643 ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 delblksudq = udqp;
2645 /*
2646 * If there are delayed allocation blocks, then we have to
2647 * unreserve those from the old dquot, and add them to the
2648 * new dquot.
2649 */
2650 if (delblks) {
2651 ASSERT(ip->i_udquot);
2652 unresudq = ip->i_udquot;
2653 }
2654 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002655 if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
Nathan Scott9a2a7de2006-03-31 13:04:49 +10002656 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
2657 ip->i_d.di_projid != be32_to_cpu(gdqp->q_core.d_id))
2658 prjflags = XFS_QMOPT_ENOSPC;
2659
2660 if (prjflags ||
2661 (XFS_IS_GQUOTA_ON(ip->i_mount) &&
2662 ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002663 delblksgdq = gdqp;
2664 if (delblks) {
2665 ASSERT(ip->i_gdquot);
2666 unresgdq = ip->i_gdquot;
2667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 }
2669 }
2670
2671 if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2672 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
Nathan Scott9a2a7de2006-03-31 13:04:49 +10002673 flags | blkflags | prjflags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return (error);
2675
2676 /*
2677 * Do the delayed blks reservations/unreservations now. Since, these
2678 * are done without the help of a transaction, if a reservation fails
2679 * its previous reservations won't be automatically undone by trans
2680 * code. So, we have to do it manually here.
2681 */
2682 if (delblks) {
2683 /*
2684 * Do the reservations first. Unreservation can't fail.
2685 */
2686 ASSERT(delblksudq || delblksgdq);
2687 ASSERT(unresudq || unresgdq);
2688 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2689 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
Nathan Scott9a2a7de2006-03-31 13:04:49 +10002690 flags | blkflags | prjflags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 return (error);
2692 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2693 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
Nathan Scott06d10dd2005-06-21 15:48:47 +10002694 blkflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 }
2696
2697 return (0);
2698}
2699
2700int
2701xfs_qm_vop_rename_dqattach(
2702 xfs_inode_t **i_tab)
2703{
2704 xfs_inode_t *ip;
2705 int i;
2706 int error;
2707
2708 ip = i_tab[0];
2709
2710 if (! XFS_IS_QUOTA_ON(ip->i_mount))
Jesper Juhl014c2542006-01-15 02:37:08 +01002711 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712
2713 if (XFS_NOT_DQATTACHED(ip->i_mount, ip)) {
2714 error = xfs_qm_dqattach(ip, 0);
2715 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002716 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 }
2718 for (i = 1; (i < 4 && i_tab[i]); i++) {
2719 /*
2720 * Watch out for duplicate entries in the table.
2721 */
2722 if ((ip = i_tab[i]) != i_tab[i-1]) {
2723 if (XFS_NOT_DQATTACHED(ip->i_mount, ip)) {
2724 error = xfs_qm_dqattach(ip, 0);
2725 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002726 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 }
2728 }
2729 }
Jesper Juhl014c2542006-01-15 02:37:08 +01002730 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731}
2732
2733void
2734xfs_qm_vop_dqattach_and_dqmod_newinode(
2735 xfs_trans_t *tp,
2736 xfs_inode_t *ip,
2737 xfs_dquot_t *udqp,
2738 xfs_dquot_t *gdqp)
2739{
2740 if (!XFS_IS_QUOTA_ON(tp->t_mountp))
2741 return;
2742
2743 ASSERT(XFS_ISLOCKED_INODE_EXCL(ip));
2744 ASSERT(XFS_IS_QUOTA_RUNNING(tp->t_mountp));
2745
2746 if (udqp) {
2747 xfs_dqlock(udqp);
2748 XFS_DQHOLD(udqp);
2749 xfs_dqunlock(udqp);
2750 ASSERT(ip->i_udquot == NULL);
2751 ip->i_udquot = udqp;
Nathan Scottee2a4f72006-01-11 15:33:36 +11002752 ASSERT(XFS_IS_UQUOTA_ON(tp->t_mountp));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002753 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2755 }
2756 if (gdqp) {
2757 xfs_dqlock(gdqp);
2758 XFS_DQHOLD(gdqp);
2759 xfs_dqunlock(gdqp);
2760 ASSERT(ip->i_gdquot == NULL);
2761 ip->i_gdquot = gdqp;
Nathan Scottee2a4f72006-01-11 15:33:36 +11002762 ASSERT(XFS_IS_OQUOTA_ON(tp->t_mountp));
2763 ASSERT((XFS_IS_GQUOTA_ON(tp->t_mountp) ?
2764 ip->i_d.di_gid : ip->i_d.di_projid) ==
2765 be32_to_cpu(gdqp->q_core.d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2767 }
2768}
2769
2770/* ------------- list stuff -----------------*/
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002771STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772xfs_qm_freelist_init(xfs_frlist_t *ql)
2773{
2774 ql->qh_next = ql->qh_prev = (xfs_dquot_t *) ql;
Jes Sorensen794ee1b2006-01-09 15:59:21 -08002775 mutex_init(&ql->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 ql->qh_version = 0;
2777 ql->qh_nelems = 0;
2778}
2779
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002780STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781xfs_qm_freelist_destroy(xfs_frlist_t *ql)
2782{
2783 xfs_dquot_t *dqp, *nextdqp;
2784
Jes Sorensen794ee1b2006-01-09 15:59:21 -08002785 mutex_lock(&ql->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 for (dqp = ql->qh_next;
2787 dqp != (xfs_dquot_t *)ql; ) {
2788 xfs_dqlock(dqp);
2789 nextdqp = dqp->dq_flnext;
2790#ifdef QUOTADEBUG
2791 cmn_err(CE_DEBUG, "FREELIST destroy 0x%p", dqp);
2792#endif
2793 XQM_FREELIST_REMOVE(dqp);
2794 xfs_dqunlock(dqp);
2795 xfs_qm_dqdestroy(dqp);
2796 dqp = nextdqp;
2797 }
Nathan Scott20722a92006-03-14 13:19:08 +11002798 mutex_unlock(&ql->qh_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 mutex_destroy(&ql->qh_lock);
2800
2801 ASSERT(ql->qh_nelems == 0);
2802}
2803
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002804STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805xfs_qm_freelist_insert(xfs_frlist_t *ql, xfs_dquot_t *dq)
2806{
2807 dq->dq_flnext = ql->qh_next;
2808 dq->dq_flprev = (xfs_dquot_t *)ql;
2809 ql->qh_next = dq;
2810 dq->dq_flnext->dq_flprev = dq;
2811 xfs_Gqm->qm_dqfreelist.qh_nelems++;
2812 xfs_Gqm->qm_dqfreelist.qh_version++;
2813}
2814
2815void
2816xfs_qm_freelist_unlink(xfs_dquot_t *dq)
2817{
2818 xfs_dquot_t *next = dq->dq_flnext;
2819 xfs_dquot_t *prev = dq->dq_flprev;
2820
2821 next->dq_flprev = prev;
2822 prev->dq_flnext = next;
2823 dq->dq_flnext = dq->dq_flprev = dq;
2824 xfs_Gqm->qm_dqfreelist.qh_nelems--;
2825 xfs_Gqm->qm_dqfreelist.qh_version++;
2826}
2827
2828void
2829xfs_qm_freelist_append(xfs_frlist_t *ql, xfs_dquot_t *dq)
2830{
2831 xfs_qm_freelist_insert((xfs_frlist_t *)ql->qh_prev, dq);
2832}
2833
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002834STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835xfs_qm_dqhashlock_nowait(
2836 xfs_dquot_t *dqp)
2837{
2838 int locked;
2839
2840 locked = mutex_trylock(&((dqp)->q_hash->qh_lock));
Jesper Juhl014c2542006-01-15 02:37:08 +01002841 return locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842}
2843
2844int
2845xfs_qm_freelist_lock_nowait(
2846 xfs_qm_t *xqm)
2847{
2848 int locked;
2849
2850 locked = mutex_trylock(&(xqm->qm_dqfreelist.qh_lock));
Jesper Juhl014c2542006-01-15 02:37:08 +01002851 return locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852}
2853
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002854STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855xfs_qm_mplist_nowait(
2856 xfs_mount_t *mp)
2857{
2858 int locked;
2859
2860 ASSERT(mp->m_quotainfo);
2861 locked = mutex_trylock(&(XFS_QI_MPLLOCK(mp)));
Jesper Juhl014c2542006-01-15 02:37:08 +01002862 return locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863}